diff --git a/README.md b/README.md
index eac685d..9bb3b56 100644
--- a/README.md
+++ b/README.md
@@ -98,7 +98,13 @@ final FrequencyTrie<CompiledPatchCommand> polish =
 
 `Language.PL_PL` selects the documented default `pl-pl-unimorph`. The optional `pl-pl-polimorf` model requires its own runtime artifact and explicit selection; adding it does not change the default. See [Model Selection and Loading](docs/model-selection-and-loading.md) for complete executable examples and [Stemmer Models](docs/stemmer-models.md) for artifact concepts.
 
-`radixor-models-standard` is a POM-only runtime aggregate: it brings the 20 default model JARs transitively but publishes no empty aggregate JAR. `radixor-models-bom` is the separate POM-only Maven dependency BOM for version management; importing it alone adds no model. The root CycloneDX SBOM report is unrelated to that dependency BOM.
+`radixor-models-standard` is a POM-only runtime aggregate: it brings the
+original 20 standard model JARs transitively but publishes no empty aggregate
+JAR. The other 123 language defaults and optional PoliMorf remain individually
+installable and are deliberately outside that aggregate.
+`radixor-models-bom` is the separate POM-only Maven dependency BOM for version
+management across all 144 user-facing models; importing it alone adds no model.
+The root CycloneDX SBOM report is unrelated to that dependency BOM.
 
 ```java
 final FrequencyTrie<CompiledPatchCommand> polimorf =
@@ -141,16 +147,16 @@ Radixor performance is best read together with stemming quality. The English dic
 
 | Used rows | Actual row ratio | All exact | Changed exact | Root preserved | Speed ms/op | Error ms | ns/token |
 | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
-| 100% | 100.000% | 97.668% | 98.110% | 97.552% | 20.425 | 3.636 | 97.9 |
-| 90% | 90.000% | 97.239% | 95.821% | 97.612% | 17.779 | 1.827 | 85.3 |
-| 80% | 80.000% | 96.827% | 93.673% | 97.656% | 15.343 | 1.321 | 73.6 |
-| 70% | 70.000% | 96.392% | 91.430% | 97.695% | 16.444 | 2.027 | 78.9 |
-| 60% | 60.000% | 95.935% | 89.244% | 97.693% | 14.330 | 1.350 | 68.7 |
-| 50% | 50.000% | 95.453% | 86.979% | 97.678% | 14.953 | 2.625 | 71.7 |
-| 40% | 40.000% | 94.939% | 84.667% | 97.638% | 12.919 | 1.155 | 62.0 |
-| 30% | 30.000% | 94.398% | 82.443% | 97.538% | 12.166 | 1.305 | 58.3 |
-| 20% | 20.000% | 93.821% | 80.174% | 97.406% | 11.549 | 1.535 | 55.4 |
-| 10% | 10.000% | 93.057% | 77.327% | 97.190% | 14.360 | 3.524 | 68.9 |
+| 100% | 100.000% | 97.668% | 98.110% | 97.552% | 25.107 | 4.358 | 120.4 |
+| 90% | 90.000% | 97.239% | 95.821% | 97.612% | 24.021 | 1.865 | 115.2 |
+| 80% | 80.000% | 96.827% | 93.673% | 97.656% | 23.593 | 2.656 | 113.1 |
+| 70% | 70.000% | 96.392% | 91.430% | 97.695% | 18.965 | 1.559 | 90.9 |
+| 60% | 60.000% | 95.935% | 89.244% | 97.693% | 18.588 | 1.040 | 89.1 |
+| 50% | 50.000% | 95.453% | 86.979% | 97.678% | 18.479 | 2.304 | 88.6 |
+| 40% | 40.000% | 94.939% | 84.667% | 97.638% | 17.345 | 1.504 | 83.2 |
+| 30% | 30.000% | 94.398% | 82.443% | 97.538% | 17.164 | 1.252 | 82.3 |
+| 20% | 20.000% | 93.821% | 80.174% | 97.406% | 14.912 | 1.512 | 71.5 |
+| 10% | 10.000% | 93.057% | 77.327% | 97.190% | 14.937 | 2.320 | 71.6 |
 
 Column meanings:
 
@@ -163,23 +169,23 @@ Column meanings:
 - `Error ms` is the JMH score error converted to milliseconds.
 - `ns/token` is average nanoseconds per changed token in that operation.
 
-The contracted trie result is materially stronger than the older uncontracted profile: full English coverage reaches 97.668% all-token exactness and 98.110% changed-token exactness at 97.9 ns/token, while even a 10% deterministic dictionary slice remains at 93.057% all-token exactness and 77.327% changed-token exactness at 68.9 ns/token. This is why Radixor benchmark results are documented with both speed and quality instead of a single Porter speed badge.
+The contracted trie result is materially stronger than the older uncontracted profile: full English coverage reaches 97.668% all-token exactness and 98.110% changed-token exactness at 120.4 ns/token, while even a 10% deterministic dictionary slice remains at 93.057% all-token exactness and 77.327% changed-token exactness at 71.6 ns/token. This is why Radixor benchmark results are documented with both speed and quality instead of a single Porter speed badge.
 
 The English curve evaluates the complete dictionary, so it intentionally mixes
 trained and withheld rows. The separate
-[20-language dictionary-family generalization report](docs/benchmarks/generalization.md)
+[143-language dictionary-family generalization report](docs/benchmarks/generalization.md)
 isolates held-out rows, removes surface forms duplicated in training, and reports
-five frozen splits at every 10% coverage step. It contains 1,000 raw scenarios
+five frozen splits at every 10% coverage step. It contains 8,300 raw scenarios
 with exact model provenance and makes clear where transfer is strong—and where a
 small resource does not support a broad generalization claim.
 
 The complementary [edit-cost sensitivity experiment](docs/benchmarks/edit-cost-sensitivity.md)
 expands 16,700 physically measured exact command classes into a validated 234,000-observation
-logical matrix. It finds that suitable relative edit costs and their structural effect are
-language-dependent. Each [language benchmark page](docs/benchmarks/languages/index.md) therefore
-publishes its own 10%–90% knowledge curve, command-equivalence evidence, selected-cost effect,
-factor associations, and bounded conclusion; exploratory non-baseline settings are not presented
-as production defaults without external validation.
+logical matrix over the historical 20-language cohort. It finds that suitable relative edit costs
+and their structural effect are language-dependent. Those 20 language pages retain their own
+10%–90% knowledge curve, command-equivalence evidence, selected-cost effect, factor associations,
+and bounded conclusion; no edit-cost result is extrapolated to the new dictionaries, and
+exploratory non-baseline settings are not presented as production defaults without external validation.
 
 For benchmark scope, workload design, environment, commands, report locations, and interpretation guidance, see [Benchmarking](docs/benchmarking.md).
 
@@ -390,9 +396,10 @@ Egothor showed that stemming could be both algorithmic and compact. Stempel prov
 
 The established `org.egothor:radixor` artifact remains the algorithmic core and contains no language-model data. From version 4 onward, applications explicitly add individual `org.egothor:radixor-model-<model-id>` runtime artifacts or the optional metadata-only `org.egothor:radixor-models-standard` aggregate. Polish defaults to `pl-pl-unimorph`; `pl-pl-polimorf` is opt-in. See [Stemmer Models](docs/stemmer-models.md) and [Migration and Backward Compatibility](docs/migration-and-backward-compatibility.md).
 
-Radixor Java software remains licensed under BSD-3-Clause. UniMorph-derived model data is
-distributed under CC BY-SA 3.0, with upstream attribution, the canonical license URI, Radixor
-transformations, and Leo Galambos's limited contribution notice carried by each model artifact.
+Radixor Java software remains licensed under BSD-3-Clause. Each UniMorph-derived model artifact
+declares its audited file-applicable license: primarily CC BY-SA 3.0, with CC BY-SA 4.0, CC BY
+4.0, and LGPLLR exceptions where required upstream. Upstream attribution, the canonical license
+URI, Radixor transformations, and Leo Galambos's limited contribution notice travel with it.
 PoliMorf model data retains its separate BSD-2-Clause license. There is no project-wide CC license
 directory because the root artifact contains no model data.
 
diff --git a/build-logic/src/main/groovy/org/egothor/radixor/PrepareModelResourcesTask.groovy b/build-logic/src/main/groovy/org/egothor/radixor/PrepareModelResourcesTask.groovy
index 0fb661d..e13b5d3 100644
--- a/build-logic/src/main/groovy/org/egothor/radixor/PrepareModelResourcesTask.groovy
+++ b/build-logic/src/main/groovy/org/egothor/radixor/PrepareModelResourcesTask.groovy
@@ -56,7 +56,6 @@ abstract class PrepareModelResourcesTask extends DefaultTask {
     @InputFile @PathSensitive(PathSensitivity.RELATIVE) abstract RegularFileProperty getVersionFile()
     @Optional @InputFile @PathSensitive(PathSensitivity.RELATIVE) abstract RegularFileProperty getLicenseFile()
     @Optional @InputFile @PathSensitive(PathSensitivity.RELATIVE) abstract RegularFileProperty getNoticeFile()
-    @Input abstract Property<Boolean> getShareAlike()
     @Input abstract MapProperty<String, String> getDescriptorValues()
     @OutputDirectory abstract DirectoryProperty getGeneratedDirectory()
 
@@ -80,12 +79,13 @@ abstract class PrepareModelResourcesTask extends DefaultTask {
         Files.createDirectories(index.parent)
         Files.writeString(index, "META-INF/radixor/models/${id}.properties\n")
 
-        if (shareAlike.get()) {
+        if (noticeFile.isPresent()) {
             final Path notice = generated.resolve("META-INF/NOTICE/${id}-data.txt")
             Files.createDirectories(notice.parent)
             Files.copy(noticeFile.get().asFile.toPath(), notice, StandardCopyOption.REPLACE_EXISTING)
-        } else {
-            final Path license = generated.resolve('META-INF/LICENSES/PoliMorf-BSD-2-Clause.txt')
+        }
+        if (licenseFile.isPresent()) {
+            final Path license = generated.resolve("META-INF/LICENSES/${licenseFile.get().asFile.name}")
             Files.createDirectories(license.parent)
             Files.copy(licenseFile.get().asFile.toPath(), license, StandardCopyOption.REPLACE_EXISTING)
         }
@@ -102,7 +102,7 @@ model.default=${value['model.default']}
 model.format=radixor-dictionary-tsv-gzip
 model.formatVersion=1
 model.sha256=${checksum}
-model.rightToLeft=${['FA_IR', 'HE_IL', 'YI'].contains(value['model.language'])}
+model.rightToLeft=${value['model.rightToLeft']}
 model.caseProcessing=LOWERCASE_WITH_LOCALE_ROOT
 model.diacriticProcessing=AS_IS
 model.storeOriginal=true
diff --git a/build-logic/src/main/groovy/org/egothor/radixor/RadixorModelExtension.groovy b/build-logic/src/main/groovy/org/egothor/radixor/RadixorModelExtension.groovy
index 12e1e2f..2480ebd 100644
--- a/build-logic/src/main/groovy/org/egothor/radixor/RadixorModelExtension.groovy
+++ b/build-logic/src/main/groovy/org/egothor/radixor/RadixorModelExtension.groovy
@@ -50,6 +50,12 @@ abstract class RadixorModelExtension {
     /** Whether this is the documented default for its language. */
     abstract Property<Boolean> getDefaultModel()
 
+    /** Whether source identity and output bytes are governed by the pinned UniMorph manifest. */
+    abstract Property<Boolean> getManifestManaged()
+
+    /** Whether the model language is conventionally written right-to-left. */
+    abstract Property<Boolean> getRightToLeft()
+
     /** Source dictionary name. */
     abstract Property<String> getSourceName()
 
@@ -96,6 +102,8 @@ abstract class RadixorModelExtension {
     @Inject
     RadixorModelExtension(final ObjectFactory objects) {
         defaultModel.convention(false)
+        manifestManaged.convention(false)
+        rightToLeft.convention(false)
         sourceVersion.convention('unavailable')
         sourceLicense.convention('LicenseRef-Radixor-Stemmer-Data')
         licenseFileName.convention('LICENSE-stemmer-data.txt')
diff --git a/build-logic/src/main/groovy/org/egothor/radixor/RadixorModelPlugin.groovy b/build-logic/src/main/groovy/org/egothor/radixor/RadixorModelPlugin.groovy
index e586894..ad5a2fe 100644
--- a/build-logic/src/main/groovy/org/egothor/radixor/RadixorModelPlugin.groovy
+++ b/build-logic/src/main/groovy/org/egothor/radixor/RadixorModelPlugin.groovy
@@ -80,7 +80,8 @@ final class RadixorModelPlugin implements Plugin<Project> {
             versionFile = project.layout.projectDirectory.file('model-version.txt')
             modelId = model.modelId
             moduleName = project.name
-            shareAlike = model.sourceLicense.map { String license -> license == 'CC-BY-SA-3.0' }
+            shareAlike = model.sourceLicense.map { String license -> license != 'BSD-2-Clause' }
+            manifestManaged = model.manifestManaged
             metadata.put('source.project', model.sourceProject)
             metadata.put('source.repository', model.sourceRepository)
             metadata.put('source.dataset', model.sourceDataset)
@@ -99,12 +100,12 @@ final class RadixorModelPlugin implements Plugin<Project> {
             dependsOn(validate)
             dictionaryFile = project.layout.projectDirectory.file('src/modelInput/stemmer.gz')
             versionFile = project.layout.projectDirectory.file('model-version.txt')
-            shareAlike = model.sourceLicense.map { String license -> license == 'CC-BY-SA-3.0' }
             generatedDirectory = project.layout.buildDirectory.dir('generated/modelResources')
             descriptorValues.put('model.id', model.modelId)
             descriptorValues.put('model.language', model.language)
             descriptorValues.put('model.displayName', model.displayName)
             descriptorValues.put('model.default', model.defaultModel.map(String::valueOf))
+            descriptorValues.put('model.rightToLeft', model.rightToLeft.map(String::valueOf))
             descriptorValues.put('source.name', model.sourceName)
             descriptorValues.put('source.version', model.sourceVersion)
             descriptorValues.put('source.project', model.sourceProject)
@@ -119,19 +120,25 @@ final class RadixorModelPlugin implements Plugin<Project> {
             descriptorValues.put('transformations.summary', model.transformationsSummary)
         }
         project.afterEvaluate {
-            final boolean shareAlike = model.sourceLicense.get() == 'CC-BY-SA-3.0'
+            final boolean shareAlike = model.sourceLicense.get() != 'BSD-2-Clause'
             if (shareAlike) {
                 final def notice = project.layout.projectDirectory.file("src/modelInput/${model.noticeFileName.get()}")
                 validate.configure { noticeFile = notice }
                 prepare.configure { noticeFile = notice }
-            } else {
+            }
+            if (!shareAlike || model.sourceLicense.get() == 'LGPLLR') {
                 final def license = project.layout.projectDirectory.file("src/modelInput/${model.licenseFileName.get()}")
                 validate.configure { licenseFile = license }
                 prepare.configure { licenseFile = license }
             }
         }
         project.tasks.named('processResources', Copy).configure { dependsOn(prepare); duplicatesStrategy = DuplicatesStrategy.FAIL }
-        project.tasks.named('sourcesJar', Jar).configure { dependsOn(prepare); exclude('**/stemmer.gz') }
+        project.tasks.named('sourcesJar', Jar).configure {
+            dependsOn(prepare)
+            if (model.sourceLicense.get() != 'LGPLLR') {
+                exclude('**/stemmer.gz')
+            }
+        }
         project.tasks.named('javadocJar', Jar).configure { exclude('**/stemmer.gz') }
         project.tasks.named('jar', Jar).configure {
             archiveBaseName.set("radixor-model-${project.name}")
@@ -155,9 +162,12 @@ final class RadixorModelPlugin implements Plugin<Project> {
                 final File archive = project.tasks.named('jar', Jar).get().archiveFile.get().asFile
                 final List<String> names = []
                 final String resource = "org/egothor/stemmer/models/${model.modelId.get()}/stemmer.gz"
-                final boolean shareAlike = model.sourceLicense.get() == 'CC-BY-SA-3.0'
-                final String licenseResource = 'META-INF/LICENSES/PoliMorf-BSD-2-Clause.txt'
-                final File sourceLicense = shareAlike ? null : project.file("src/modelInput/${model.licenseFileName.get()}")
+                final boolean shareAlike = model.sourceLicense.get() != 'BSD-2-Clause'
+                final boolean requiresFullLicense = !shareAlike || model.sourceLicense.get() == 'LGPLLR'
+                final String licenseResource = requiresFullLicense
+                        ? "META-INF/LICENSES/${model.licenseFileName.get()}" : null
+                final File sourceLicense = requiresFullLicense
+                        ? project.file("src/modelInput/${model.licenseFileName.get()}") : null
                 final File sourceNotice = shareAlike
                         ? project.file("src/modelInput/${model.noticeFileName.get()}") : null
                 final String noticeResource = "META-INF/NOTICE/${model.modelId.get()}-data.txt"
@@ -170,7 +180,7 @@ final class RadixorModelPlugin implements Plugin<Project> {
                     if (entry != null) {
                         packagedChecksum = sha256(zip.getInputStream(entry).bytes)
                     }
-                    final def licenseEntry = zip.getEntry(licenseResource)
+                    final def licenseEntry = licenseResource == null ? null : zip.getEntry(licenseResource)
                     if (licenseEntry != null) {
                         packagedLicenseChecksum = sha256(zip.getInputStream(licenseEntry).bytes)
                     }
@@ -187,7 +197,10 @@ final class RadixorModelPlugin implements Plugin<Project> {
                 }
                 if (shareAlike) {
                     requireMatchingChecksum('notice', noticeResource, sha256(sourceNotice), packagedNoticeChecksum)
-                    validateUniMorphJarContents(names)
+                    if (requiresFullLicense) {
+                        requireMatchingChecksum('license', licenseResource, sha256(sourceLicense), packagedLicenseChecksum)
+                    }
+                    validateUniMorphJarContents(names, licenseResource)
                 } else {
                     requireMatchingChecksum('license', licenseResource, sha256(sourceLicense), packagedLicenseChecksum)
                     validatePoliMorfJarContents(names)
@@ -198,8 +211,21 @@ final class RadixorModelPlugin implements Plugin<Project> {
                 [project.tasks.named('sourcesJar', Jar).get(), project.tasks.named('javadocJar', Jar).get()].each { Jar task ->
                     final File documentationArchive = task.archiveFile.get().asFile
                     new java.util.zip.ZipFile(documentationArchive).withCloseable { zip ->
-                        if (zip.entries().any { entry -> entry.name.endsWith('/stemmer.gz') || entry.name == 'stemmer.gz' }) {
-                            throw new GradleException("Documentation artifact ${documentationArchive.name} must not contain a model dictionary.")
+                        final List<String> documentationNames = []
+                        zip.entries().each { entry -> documentationNames.add(entry.name) }
+                        final boolean containsDictionary = documentationNames.any {
+                            String name -> name.endsWith('/stemmer.gz') || name == 'stemmer.gz'
+                        }
+                        final boolean sourceFormRequired = model.sourceLicense.get() == 'LGPLLR'
+                                && task.name == 'sourcesJar'
+                        if (containsDictionary != sourceFormRequired) {
+                            final String requirement = sourceFormRequired
+                                    ? 'must contain the LGPLLR legible dictionary form'
+                                    : 'must not contain a model dictionary'
+                            throw new GradleException("Documentation artifact ${documentationArchive.name} ${requirement}.")
+                        }
+                        if (sourceFormRequired && !documentationNames.contains(licenseResource)) {
+                            throw new GradleException("Documentation artifact ${documentationArchive.name} must contain ${licenseResource}.")
                         }
                     }
                 }
@@ -221,7 +247,7 @@ final class RadixorModelPlugin implements Plugin<Project> {
                 publication.pom {
                     name.set("Radixor model ${project.name}")
                     description.set(model.displayName.zip(model.sourceLicense) { String displayName, String licenseId ->
-                        final String material = licenseId == 'CC-BY-SA-3.0'
+                        final String material = licenseId != 'BSD-2-Clause'
                                 ? 'See the packaged model-specific notice.'
                                 : 'See the packaged model-data license.'
                         return "${displayName}. This artifact contains Radixor-derived model data licensed under ${licenseId}; "
@@ -333,7 +359,13 @@ final class RadixorModelPlugin implements Plugin<Project> {
                 throw new GradleException("Required model metadata is missing: ${key}")
             }
         }
-        validateRevisionMetadata(model.sourceRevision.get(), model.sourceRevisionStatus.get())
+        validateManifestRevisionMetadata(model.sourceRevision.get(), model.sourceRevisionStatus.get(),
+                model.manifestManaged.get())
+        final Set<String> supportedLicenses = [
+                'CC-BY-SA-3.0', 'CC-BY-SA-4.0', 'CC-BY-4.0', 'LGPLLR', 'BSD-2-Clause'] as Set<String>
+        if (!supportedLicenses.contains(model.sourceLicense.get())) {
+            throw new GradleException("Unsupported or unknown model-data license: ${model.sourceLicense.get()}")
+        }
     }
 
     /** Accepts an exact recorded revision or the explicit legacy-import sentinel, but never an absent status. */
@@ -353,23 +385,45 @@ final class RadixorModelPlugin implements Plugin<Project> {
         }
     }
 
+    /** Rejects the legacy revision sentinel for manifest-managed imports. */
+    static void validateManifestRevisionMetadata(final String revision, final String status,
+            final boolean manifestManaged) {
+        validateRevisionMetadata(revision, status)
+        if (manifestManaged && revision == 'not-recorded-in-legacy-import') {
+            throw new GradleException('Manifest-managed UniMorph models require an exact pinned source revision.')
+        }
+    }
+
     /** Validates the model-specific attribution and ShareAlike notice. */
     static void validateShareAlikeNotice(final File notice, final RadixorModelExtension model) {
         validateShareAlikeNoticeText(notice.getText('UTF-8'), notice.toString(), model.modelId.get(),
-                model.sourceRepository.get(), model.sourceLicenseUri.get(), model.sourceRevision.get(),
+                model.sourceRepository.get(), model.sourceLicense.get(), model.sourceLicenseUri.get(), model.sourceRevision.get(),
                 model.sourceRevisionStatus.get())
     }
 
     /** Validates required content in one UniMorph model-data notice. */
     static void validateShareAlikeNoticeText(final String rawText, final String noticeName,
-            final String modelId, final String repository, final String licenseUri,
+            final String modelId, final String repository, final String licenseId, final String licenseUri,
             final String revision, final String revisionStatus) {
         final String text = rawText.replace('\r\n', '\n')
+        final Map<String, List<String>> licenseText = [
+                'CC-BY-SA-3.0': ['Creative Commons Attribution-ShareAlike 3.0 Unported',
+                        'is distributed under Creative Commons Attribution-ShareAlike 3.0'],
+                'CC-BY-SA-4.0': ['Creative Commons Attribution-ShareAlike 4.0 International',
+                        'is distributed under Creative Commons Attribution-ShareAlike 4.0'],
+                'CC-BY-4.0': ['Creative Commons Attribution 4.0 International',
+                        'is distributed under Creative Commons Attribution 4.0'],
+                'LGPLLR': ['Lesser General Public License For Linguistic Resources',
+                        'is distributed under the Lesser General Public License For Linguistic Resources']]
+        final List<String> expectedLicenseText = licenseText[licenseId]
+        if (expectedLicenseText == null) {
+            throw new GradleException("Unsupported notice-based model-data license: ${licenseId}")
+        }
         final List<String> required = [
                 "Model ID: ${modelId}",
                 "Official repository: ${repository}",
                 'Attribution:',
-                'License:\nCreative Commons Attribution-ShareAlike 3.0 Unported',
+                "License:\n${expectedLicenseText[0]}",
                 "Canonical license URI: ${licenseUri}",
                 'Radixor modifications:',
                 "Revision status: ${revisionStatus}",
@@ -378,7 +432,7 @@ final class RadixorModelPlugin implements Plugin<Project> {
                 'to the extent protected by applicable law.',
                 'The underlying morphological data remains attributed to UniMorph and',
                 "This derived model data, including Radixor's protectable contributions,",
-                'is distributed under Creative Commons Attribution-ShareAlike 3.0',
+                expectedLicenseText[1],
                 'Neither UniMorph nor any upstream contributor endorses Radixor.']
         if (revision == 'not-recorded-in-legacy-import') {
             required.add('The exact UniMorph commit used for the original Radixor import was not recorded.')
@@ -390,15 +444,26 @@ final class RadixorModelPlugin implements Plugin<Project> {
     }
 
     /** Rejects generic license files and foreign notices in a UniMorph model artifact. */
-    static void validateUniMorphJarContents(final List<String> names) {
-        if (names.any { String name -> name.startsWith('META-INF/LICENSES/') }) {
-            throw new GradleException('A UniMorph model artifact must use only its model-specific notice for data licensing.')
+    static void validateUniMorphJarContents(final List<String> names, final String permittedLicenseResource = null) {
+        final List<String> licenseResources = names.findAll {
+            String name -> name.startsWith('META-INF/LICENSES/') && !name.endsWith('/')
+        }
+        if (licenseResources != (permittedLicenseResource == null ? [] : [permittedLicenseResource])) {
+            throw new GradleException('A UniMorph model artifact contains unexpected model-data license resources.')
         }
         if (names.count { String name -> name.startsWith('META-INF/NOTICE/') && !name.endsWith('/') } != 1) {
             throw new GradleException('A UniMorph model artifact must contain exactly one model-specific notice.')
         }
     }
 
+    /** Validates the canonical SPDX LGPLLR text used by the Khaling model. */
+    static void validateLgpllrLicense(final File licenseFile) {
+        final String expected = 'e4e0f2f92769aad680aeca07f359521004dac4598d8b03def0e6fe507e871134'
+        if (sha256(licenseFile) != expected) {
+            throw new GradleException('The LGPLLR license must match the canonical SPDX text byte-for-byte.')
+        }
+    }
+
     /** Rejects UniMorph licensing material in the separately licensed PoliMorf artifact. */
     static void validatePoliMorfJarContents(final List<String> names) {
         if (names.any { String name -> name.startsWith('META-INF/NOTICE/')
@@ -438,16 +503,14 @@ final class RadixorModelPlugin implements Plugin<Project> {
                             long lineNumber = 0L
                             while ((line = reader.readLine()) != null) {
                 lineNumber++
-                final String trimmed = line.trim()
-                if (trimmed && !trimmed.startsWith('#') && !trimmed.startsWith('//')) {
-                    final String[] columns = line.split('\\t', -1)
-                    if (columns[0].isEmpty()) {
-                        throw new GradleException("Invalid Radixor dictionary row ${lineNumber} in ${file}.")
-                    }
-                    if (containsUnicodeWhitespace(columns[0])) continue
+                final String logicalLine = stripRemark(line).trim()
+                if (logicalLine) {
+                    final String[] columns = logicalLine.split('\\t', -1)
+                    final String stem = columns[0].strip()
+                    if (!stem || containsUnicodeWhitespace(stem)) continue
                     long acceptedRowForms = 1L
                     for (int index = 1; index < columns.length; index++) {
-                        final String variant = columns[index]
+                        final String variant = columns[index].strip()
                         if (variant.isEmpty()) {
                             ignoredEmptyVariants++
                         } else if (!containsUnicodeWhitespace(variant)) {
@@ -475,6 +538,21 @@ final class RadixorModelPlugin implements Plugin<Project> {
         return new DictionaryValidationResult(acceptedGroups, acceptedForms, ignoredEmptyVariants)
     }
 
+    /** Removes the earliest production-parser comment marker from one physical line. */
+    private static String stripRemark(final String line) {
+        final int hash = line.indexOf('#')
+        final int slash = line.indexOf('//')
+        final int remark
+        if (hash < 0) {
+            remark = slash
+        } else if (slash < 0) {
+            remark = hash
+        } else {
+            remark = Math.min(hash, slash)
+        }
+        return remark < 0 ? line : line.substring(0, remark)
+    }
+
     /** Detects Unicode whitespace in one bounded dictionary field. */
     private static boolean containsUnicodeWhitespace(final String value) {
         for (int index = 0; index < value.length(); index++) {
@@ -495,7 +573,7 @@ model.default=${model.defaultModel.get()}
 model.format=radixor-dictionary-tsv-gzip
 model.formatVersion=1
 model.sha256=${checksum}
-model.rightToLeft=${['FA_IR', 'HE_IL', 'YI'].contains(model.language.get())}
+model.rightToLeft=${model.rightToLeft.get()}
 model.caseProcessing=LOWERCASE_WITH_LOCALE_ROOT
 model.diacriticProcessing=AS_IS
 model.storeOriginal=true
diff --git a/build-logic/src/main/groovy/org/egothor/radixor/ValidateModelInputTask.groovy b/build-logic/src/main/groovy/org/egothor/radixor/ValidateModelInputTask.groovy
index c41534e..6888706 100644
--- a/build-logic/src/main/groovy/org/egothor/radixor/ValidateModelInputTask.groovy
+++ b/build-logic/src/main/groovy/org/egothor/radixor/ValidateModelInputTask.groovy
@@ -52,6 +52,7 @@ abstract class ValidateModelInputTask extends DefaultTask {
     @Input abstract Property<String> getModelId()
     @Input abstract Property<String> getModuleName()
     @Input abstract Property<Boolean> getShareAlike()
+    @Input abstract Property<Boolean> getManifestManaged()
     @Input abstract MapProperty<String, String> getMetadata()
 
     /** Performs deterministic metadata, licensing, and streaming dictionary validation. */
@@ -60,7 +61,7 @@ abstract class ValidateModelInputTask extends DefaultTask {
         final File dictionary = dictionaryFile.get().asFile
         final String id = modelId.get()
         final String version = versionFile.get().asFile.getText('UTF-8').trim()
-        if (id != moduleName.get() || !(id ==~ /[a-z]{2}(?:-[a-z]{2})?-[a-z0-9]+(?:-[a-z0-9]+)*/)) {
+        if (id != moduleName.get() || !(id ==~ /[a-z]{2,3}(?:-[a-z]{2})?-[a-z0-9]+(?:-[a-z0-9]+)*/)) {
             throw new GradleException("Model ID '${id}' must equal module '${moduleName.get()}' and use the safe model-ID syntax.")
         }
         if (!(version ==~ /[0-9]+\.[0-9]+\.[0-9]+(?:[-+][0-9A-Za-z.-]+)?/)) {
@@ -70,12 +71,16 @@ abstract class ValidateModelInputTask extends DefaultTask {
         values.each { String key, String value ->
             if (value == null || value.isBlank()) throw new GradleException("Required model metadata is missing: ${key}")
         }
-        RadixorModelPlugin.validateRevisionMetadata(values['source.revision'], values['source.revisionStatus'])
+        RadixorModelPlugin.validateManifestRevisionMetadata(values['source.revision'],
+                values['source.revisionStatus'], manifestManaged.get())
         if (shareAlike.get()) {
             final File notice = noticeFile.get().asFile
             RadixorModelPlugin.validateShareAlikeNoticeText(notice.getText('UTF-8'), notice.toString(), id,
-                    values['source.repository'], values['source.licenseUri'], values['source.revision'],
+                    values['source.repository'], values['source.license'], values['source.licenseUri'], values['source.revision'],
                     values['source.revisionStatus'])
+            if (values['source.license'] == 'LGPLLR') {
+                RadixorModelPlugin.validateLgpllrLicense(licenseFile.get().asFile)
+            }
         } else {
             final String text = licenseFile.get().asFile.getText('UTF-8')
             if (!text.contains('SPDX-License-Identifier: BSD-2-Clause')
diff --git a/build-logic/src/test/groovy/org/egothor/radixor/RadixorModelPluginTest.groovy b/build-logic/src/test/groovy/org/egothor/radixor/RadixorModelPluginTest.groovy
index c8220ce..cf468a7 100644
--- a/build-logic/src/test/groovy/org/egothor/radixor/RadixorModelPluginTest.groovy
+++ b/build-logic/src/test/groovy/org/egothor/radixor/RadixorModelPluginTest.groovy
@@ -32,6 +32,8 @@
 package org.egothor.radixor
 
 import org.gradle.api.GradleException
+import org.gradle.testkit.runner.BuildResult
+import org.gradle.testkit.runner.GradleRunner
 import org.junit.jupiter.api.Test
 import org.junit.jupiter.api.io.TempDir
 
@@ -39,6 +41,7 @@ import java.nio.charset.StandardCharsets
 import java.nio.file.Files
 import java.nio.file.Path
 import java.util.zip.GZIPOutputStream
+import java.util.zip.ZipFile
 
 import static org.junit.jupiter.api.Assertions.assertEquals
 import static org.junit.jupiter.api.Assertions.assertThrows
@@ -62,6 +65,17 @@ final class RadixorModelPluginTest {
                 'not-recorded-in-legacy-import', 'not-recorded-in-legacy-import')
     }
 
+    /** Preserves legacy compatibility while forbidding sentinels in manifest-managed models. */
+    @Test
+    void rejectsUnknownRevisionForManifestManagedModel() {
+        assertThrows(GradleException) {
+            RadixorModelPlugin.validateManifestRevisionMetadata(
+                    'not-recorded-in-legacy-import', 'not-recorded-in-legacy-import', true)
+        }
+        RadixorModelPlugin.validateManifestRevisionMetadata(
+                'not-recorded-in-legacy-import', 'not-recorded-in-legacy-import', false)
+    }
+
     /** Rejects a missing revision-status declaration. */
     @Test
     void rejectsMissingRevisionStatus() {
@@ -126,6 +140,126 @@ final class RadixorModelPluginTest {
         }
     }
 
+    /** Packages CC BY-SA 4.0 as ShareAlike notice material and describes that notice in the POM. */
+    @Test
+    void packagesCcBySaFourNoticeAndDescribesItInPom() {
+        final Path project = temporaryDirectory.resolve('cc4-model')
+        final Path modelInput = project.resolve('src/modelInput')
+        Files.createDirectories(modelInput)
+        Files.writeString(project.resolve('settings.gradle'), "rootProject.name = 'hsi-default'\n",
+                StandardCharsets.UTF_8)
+        Files.writeString(project.resolve('model-version.txt'), '1.0.0\n', StandardCharsets.UTF_8)
+        writeGzip(modelInput.resolve('stemmer.gz').toFile()) { BufferedWriter writer ->
+            writer.write('root\trooted\n')
+        }
+        Files.writeString(modelInput.resolve('NOTICE-model-data.txt'), validVersionFourNotice(),
+                StandardCharsets.UTF_8)
+        Files.writeString(project.resolve('build.gradle'), '''plugins {
+    id 'org.egothor.radixor.model'
+}
+radixorModel {
+    modelId = 'hsi-default'
+    language = 'HSI'
+    displayName = 'Hsilimo — UniMorph'
+    defaultModel = true
+    sourceName = 'UniMorph'
+    sourceVersion = 'revision'
+    sourceRevision = 'revision'
+    sourceProject = 'UniMorph'
+    sourceRepository = 'https://github.com/unimorph/hsi'
+    sourceDataset = 'UniMorph Hsilimo morphological dataset'
+    sourceRevisionStatus = 'recorded'
+    sourceLicense = 'CC-BY-SA-4.0'
+    sourceLicenseUri = 'https://creativecommons.org/licenses/by-sa/4.0/'
+    sourceAttribution = 'UniMorph contributors'
+    sourceVerificationDate = '2026-09-10'
+    transformationsSummary = 'Cleaning and deterministic model packaging'
+    noticeFileName = 'NOTICE-model-data.txt'
+}
+''', StandardCharsets.UTF_8)
+
+        final BuildResult result = GradleRunner.create()
+                .withProjectDir(project.toFile())
+                .withPluginClasspath()
+                .withArguments('verifyModelJar', 'generatePomFileForModelPublication', '--stacktrace')
+                .build()
+
+        assertTrue(result.output.contains('BUILD SUCCESSFUL'))
+        final Path archive = project.resolve('build/libs/radixor-model-hsi-default-1.0.0.jar')
+        new ZipFile(archive.toFile()).withCloseable { ZipFile zip ->
+            assertTrue(zip.getEntry('META-INF/NOTICE/hsi-default-data.txt') != null)
+            assertTrue(zip.getEntry('META-INF/LICENSES/PoliMorf-BSD-2-Clause.txt') == null)
+        }
+        final String pom = Files.readString(
+                project.resolve('build/publications/model/pom-default.xml'), StandardCharsets.UTF_8)
+        assertTrue(pom.contains('See the packaged model-specific notice.'))
+        assertTrue(!pom.contains('See the packaged model-data license.'))
+    }
+
+    /** Packages the canonical LGPLLR text and legible dictionary form in the Khaling source artifact. */
+    @Test
+    void packagesLgpllrLicenseAndLegibleDictionarySource() {
+        final Path project = temporaryDirectory.resolve('lgpllr-model')
+        final Path modelInput = project.resolve('src/modelInput')
+        Files.createDirectories(modelInput)
+        Files.writeString(project.resolve('settings.gradle'), "rootProject.name = 'klr-default'\n",
+                StandardCharsets.UTF_8)
+        Files.writeString(project.resolve('model-version.txt'), '1.0.0\n', StandardCharsets.UTF_8)
+        writeGzip(modelInput.resolve('stemmer.gz').toFile()) { BufferedWriter writer ->
+            writer.write('root\trooted\n')
+        }
+        Files.writeString(modelInput.resolve('NOTICE-model-data.txt'), validLgpllrNotice(),
+                StandardCharsets.UTF_8)
+        final List<Path> licenseCandidates = [
+                Path.of('models/klr-default/src/modelInput/LGPLLR.txt'),
+                Path.of('../models/klr-default/src/modelInput/LGPLLR.txt')]
+        final Path license = licenseCandidates.find { Path candidate -> Files.isRegularFile(candidate) }
+        assertTrue(license != null, 'The canonical LGPLLR license text must be checked in.')
+        Files.copy(license, modelInput.resolve('LGPLLR.txt'))
+        Files.writeString(project.resolve('build.gradle'), '''plugins {
+    id 'org.egothor.radixor.model'
+}
+radixorModel {
+    modelId = 'klr-default'
+    language = 'KLR'
+    displayName = 'Khaling — UniMorph'
+    defaultModel = true
+    sourceName = 'UniMorph'
+    sourceVersion = 'revision'
+    sourceRevision = 'revision'
+    sourceProject = 'UniMorph'
+    sourceRepository = 'https://github.com/unimorph/klr'
+    sourceDataset = 'UniMorph Khaling morphological dataset'
+    sourceRevisionStatus = 'recorded'
+    sourceLicense = 'LGPLLR'
+    sourceLicenseUri = 'https://spdx.org/licenses/LGPLLR.html'
+    sourceAttribution = 'UniMorph contributors'
+    sourceVerificationDate = '2026-09-11'
+    transformationsSummary = 'Cleaning and deterministic model packaging'
+    noticeFileName = 'NOTICE-model-data.txt'
+    licenseFileName = 'LGPLLR.txt'
+}
+''', StandardCharsets.UTF_8)
+
+        final BuildResult result = GradleRunner.create()
+                .withProjectDir(project.toFile())
+                .withPluginClasspath()
+                .withArguments('verifyModelJar', '--stacktrace')
+                .build()
+
+        assertTrue(result.output.contains('BUILD SUCCESSFUL'))
+        final Path archive = project.resolve('build/libs/radixor-model-klr-default-1.0.0.jar')
+        new ZipFile(archive.toFile()).withCloseable { ZipFile zip ->
+            assertTrue(zip.getEntry('META-INF/NOTICE/klr-default-data.txt') != null)
+            assertTrue(zip.getEntry('META-INF/LICENSES/LGPLLR.txt') != null)
+        }
+        final Path sources = project.resolve('build/libs/klr-default-1.0.0-sources.jar')
+        new ZipFile(sources.toFile()).withCloseable { ZipFile zip ->
+            assertTrue(zip.getEntry('org/egothor/stemmer/models/klr-default/stemmer.gz') != null)
+            assertTrue(zip.getEntry('META-INF/LICENSES/LGPLLR.txt') != null)
+        }
+    }
+
     /** Streams a large dictionary while retaining only aggregate counters and the current row. */
     @Test
     void validatesLargeDictionaryWithBoundedState() {
@@ -142,7 +276,7 @@ final class RadixorModelPluginTest {
 
         assertEquals(groups, result.acceptedGroupCount)
         assertEquals(groups * 2L, result.acceptedFormCount)
-        assertEquals(groups, result.ignoredEmptyVariantCount)
+        assertEquals(0L, result.ignoredEmptyVariantCount)
     }
 
     /** Rejects a source that is not a GZip stream. */
@@ -163,12 +297,15 @@ final class RadixorModelPluginTest {
         assertThrows(GradleException) { RadixorModelPlugin.validateDictionary(dictionary) }
     }
 
-    /** Rejects structurally invalid rows with an empty stem. */
+    /** Trims a physical row before splitting it exactly as the production parser does. */
     @Test
-    void rejectsInvalidRows() {
+    void trimsRowsBeforeSplittingFields() {
         final File dictionary = temporaryDirectory.resolve('invalid-row.gz').toFile()
-        writeGzip(dictionary) { BufferedWriter writer -> writer.write("\tvariant\n") }
-        assertThrows(GradleException) { RadixorModelPlugin.validateDictionary(dictionary) }
+        writeGzip(dictionary) { BufferedWriter writer -> writer.write("\tvariant\nvalid\tvariant\n") }
+        final RadixorModelPlugin.DictionaryValidationResult result =
+                RadixorModelPlugin.validateDictionary(dictionary)
+        assertEquals(2L, result.acceptedGroupCount)
+        assertEquals(3L, result.acceptedFormCount)
     }
 
     /** Preserves the production parser policy for Unicode-whitespace items. */
@@ -185,6 +322,26 @@ final class RadixorModelPluginTest {
         assertEquals(2L, result.acceptedFormCount)
     }
 
+    /** Mirrors production parsing for trimming, comments, NBSP, marks, joiners, and punctuation. */
+    @Test
+    void matchesProductionDictionaryFieldGrammar() {
+        final File dictionary = temporaryDirectory.resolve('parser-conformance.gz').toFile()
+        writeGzip(dictionary) { BufferedWriter writer ->
+            writer.write("  root  \t  variant  \n")
+            writer.write("bad stem\tvariant\n")
+            writer.write("nbsp\u00a0stem\taccepted\n")
+            writer.write("mark\u0301\tzwnj\u200c\tzwj\u200d\to'neil\tco-op\ta:b\ta/b\n")
+            writer.write("comment\tkept#discarded\n")
+            writer.write("slash\tkept//discarded\n")
+        }
+
+        final RadixorModelPlugin.DictionaryValidationResult result =
+                RadixorModelPlugin.validateDictionary(dictionary)
+
+        assertEquals(5L, result.acceptedGroupCount)
+        assertEquals(15L, result.acceptedFormCount)
+    }
+
     /** Streams the complete maintained PoliMorf model input successfully. */
     @Test
     void validatesFullPoliMorfInput() {
@@ -210,7 +367,8 @@ final class RadixorModelPluginTest {
 
     private static void validateNotice(final String text) {
         RadixorModelPlugin.validateShareAlikeNoticeText(text, 'test notice', 'test-model',
-                'https://github.com/unimorph/test', 'https://creativecommons.org/licenses/by-sa/3.0/',
+                'https://github.com/unimorph/test', 'CC-BY-SA-3.0',
+                'https://creativecommons.org/licenses/by-sa/3.0/',
                 'not-recorded-in-legacy-import', 'not-recorded-in-legacy-import')
     }
 
@@ -231,6 +389,44 @@ The underlying morphological data remains attributed to UniMorph and
 This derived model data, including Radixor's protectable contributions,
 is distributed under Creative Commons Attribution-ShareAlike 3.0
 Neither UniMorph nor any upstream contributor endorses Radixor.
+'''
+    }
+
+    private static String validVersionFourNotice() {
+        return '''Model ID: hsi-default
+Official repository: https://github.com/unimorph/hsi
+Attribution: UniMorph contributors
+License:
+Creative Commons Attribution-ShareAlike 4.0 International
+Canonical license URI: https://creativecommons.org/licenses/by-sa/4.0/
+Radixor modifications: Cleaning and deterministic model packaging.
+Revision status: recorded
+Copyright (C) 2026, Leo Galambos.
+Radixor-specific selection, verification, cleaning, normalization,
+to the extent protected by applicable law.
+The underlying morphological data remains attributed to UniMorph and
+This derived model data, including Radixor's protectable contributions,
+is distributed under Creative Commons Attribution-ShareAlike 4.0
+Neither UniMorph nor any upstream contributor endorses Radixor.
+'''
+    }
+
+    private static String validLgpllrNotice() {
+        return '''Model ID: klr-default
+Official repository: https://github.com/unimorph/klr
+Attribution: UniMorph contributors
+License:
+Lesser General Public License For Linguistic Resources
+Canonical license URI: https://spdx.org/licenses/LGPLLR.html
+Radixor modifications: Cleaning and deterministic model packaging.
+Revision status: recorded
+Copyright (C) 2026, Leo Galambos.
+Radixor-specific selection, verification, cleaning, normalization,
+to the extent protected by applicable law.
+The underlying morphological data remains attributed to UniMorph and
+This derived model data, including Radixor's protectable contributions,
+is distributed under the Lesser General Public License For Linguistic Resources
+Neither UniMorph nor any upstream contributor endorses Radixor.
 '''
     }
 }
diff --git a/build.gradle b/build.gradle
index 40ac853..b5b6d92 100644
--- a/build.gradle
+++ b/build.gradle
@@ -135,6 +135,47 @@ def modelProjects() {
     }
 }
 
+def quarantinedModelIds() {
+    Properties quarantine = new Properties()
+    rootProject.file('models/model-quarantine.properties').withInputStream { InputStream input ->
+        quarantine.load(input)
+    }
+    return quarantine.stringPropertyNames().toList().sort()
+}
+
+tasks.register('verifyModelQuarantine') {
+    group = 'verification'
+    description = 'Verifies quarantined model projects remain outside the active topology.'
+    inputs.files(rootProject.file('models/model-projects.properties'),
+            rootProject.file('models/model-quarantine.properties'))
+    doLast {
+        Properties topology = new Properties()
+        rootProject.file('models/model-projects.properties').withInputStream { InputStream input ->
+            topology.load(input)
+        }
+        Properties quarantine = new Properties()
+        rootProject.file('models/model-quarantine.properties').withInputStream { InputStream input ->
+            quarantine.load(input)
+        }
+        Set<String> overlap = topology.stringPropertyNames().intersect(
+                quarantine.stringPropertyNames())
+        if (!overlap.isEmpty()) {
+            throw new GradleException("Quarantined model IDs remain active: ${overlap.toList().sort().join(', ')}")
+        }
+        quarantine.stringPropertyNames().each { String modelId ->
+            String record = quarantine.getProperty(modelId)
+            if (!rootProject.file("models/${modelId}").isDirectory()) {
+                throw new GradleException("Quarantined model directory is missing: models/${modelId}")
+            }
+            if (!(record ==~ /https:\/\/github\.com\/unimorph\/[a-z0-9-]+\|\S.*\S/)) {
+                throw new GradleException("Invalid quarantine evidence record for ${modelId}.")
+            }
+        }
+        logger.lifecycle("Verified ${quarantine.size()} quarantined model projects outside "
+                + "${topology.size()} active model projects.")
+    }
+}
+
 def defaultModelProjects() {
     Properties topology = new Properties()
     rootProject.file('models/model-projects.properties').withInputStream { InputStream input ->
@@ -145,6 +186,26 @@ def defaultModelProjects() {
     }.sort().collect { String modelId -> project(":models:${modelId}") }
 }
 
+def languageDefaultModelProjects() {
+    Properties topology = new Properties()
+    rootProject.file('models/model-projects.properties').withInputStream { InputStream input ->
+        topology.load(input)
+    }
+    return topology.stringPropertyNames().findAll { String modelId ->
+        topology.getProperty(modelId) in ['default', 'standalone']
+    }.sort().collect { String modelId -> project(":models:${modelId}") }
+}
+
+def standaloneModelProjects() {
+    Properties topology = new Properties()
+    rootProject.file('models/model-projects.properties').withInputStream { InputStream input ->
+        topology.load(input)
+    }
+    return topology.stringPropertyNames().findAll { String modelId ->
+        topology.getProperty(modelId) == 'standalone'
+    }.sort().collect { String modelId -> project(":models:${modelId}") }
+}
+
 def alternativeModelProjects() {
     Properties topology = new Properties()
     rootProject.file('models/alternative-model-projects.properties').withInputStream { InputStream input ->
@@ -712,7 +773,7 @@ tasks.register('verifySoftwareLicenseHeaders') {
 tasks.register('verifyAllDefaultModels') {
     group = 'verification'
     description = 'Verifies that every language default model project is configured.'
-    dependsOn(defaultModelProjects().collect { Project modelProject ->
+    dependsOn(languageDefaultModelProjects().collect { Project modelProject ->
         modelProject.path + ':verifyModelDescriptor'
     })
 }
@@ -1160,7 +1221,7 @@ tasks.register('stemmingQuality', JavaExec) {
 
 tasks.register('benchmarkCorpusReport', JavaExec) {
     group = 'verification'
-    description = 'Reports corpus and preferred patch-command counts for every default model.'
+    description = 'Reports corpus, distinct usable forms, and patch-command counts for all user-facing models.'
     dependsOn(tasks.named('jmhClasses'))
     classpath = files(sourceSets.jmh.runtimeClasspath, configurations.stemmingQualityJmhRuntime)
     mainClass = 'org.egothor.stemmer.benchmark.BenchmarkCorpusReportApplication'
@@ -1190,6 +1251,68 @@ tasks.register('dictionaryGeneralization', JavaExec) {
     maxHeapSize = '8g'
 }
 
+tasks.register('writeStandaloneGeneralizationSelection') {
+    group = 'verification'
+    description = 'Writes the exact topology-derived standalone generalization selector.'
+    final Provider<RegularFile> selector = layout.buildDirectory.file('reports/generalization/standalone-default-models.txt')
+    outputs.file(selector)
+    doLast {
+        final List<String> modelIds = standaloneModelProjects().collect { Project modelProject -> modelProject.name }
+        if (modelIds.isEmpty()) {
+            throw new GradleException('The active topology contains no standalone default models.')
+        }
+        final File output = selector.get().asFile
+        output.parentFile.mkdirs()
+        output.setText(modelIds.join(System.lineSeparator()) + System.lineSeparator(), 'UTF-8')
+    }
+}
+
+tasks.register('dictionaryGeneralizationStandalone', JavaExec) {
+    group = 'verification'
+    description = 'Evaluates only the topology-derived standalone language defaults for snapshot continuation.'
+    dependsOn(tasks.named('testClasses'))
+    dependsOn(tasks.named('writeStandaloneGeneralizationSelection'))
+    classpath = sourceSets.test.runtimeClasspath
+    mainClass = 'org.egothor.stemmer.benchmark.generalization.DictionaryGeneralizationApplication'
+    outputs.file(layout.buildDirectory.file('reports/generalization/dictionary-generalization-standalone-2026-09-11.csv'))
+    final String dictionaryGeneralizationSourceState = providers.exec {
+        commandLine 'git', 'status', '--porcelain'
+    }.standardOutput.asText.get().trim().isEmpty() ? 'clean' : 'dirty'
+    args layout.buildDirectory.file('reports/generalization/dictionary-generalization-standalone-2026-09-11.csv')
+            .get().asFile.absolutePath,
+            dictionaryGeneralizationReleaseVersion.get(),
+            providers.exec { commandLine 'git', 'rev-parse', 'HEAD' }.standardOutput.asText.get().trim(),
+            dictionaryGeneralizationSourceState,
+            layout.projectDirectory.file('src/test/java/org/egothor/stemmer/benchmark/generalization/DictionaryGeneralizationApplication.java').asFile.absolutePath,
+            layout.buildDirectory.file('reports/generalization/standalone-default-models.txt').get().asFile.absolutePath
+    maxHeapSize = '8g'
+}
+
+tasks.register('dictionaryGeneralizationExactModel', JavaExec) {
+    group = 'verification'
+    description = 'Evaluates one exact authoritative default for targeted snapshot repair.'
+    dependsOn(tasks.named('testClasses'))
+    classpath = sourceSets.test.runtimeClasspath
+    mainClass = 'org.egothor.stemmer.benchmark.generalization.DictionaryGeneralizationApplication'
+    final String exactModel = providers.gradleProperty('dictionaryGeneralizationModel').getOrElse('')
+    final String dictionaryGeneralizationSourceState = providers.exec {
+        commandLine 'git', 'status', '--porcelain'
+    }.standardOutput.asText.get().trim().isEmpty() ? 'clean' : 'dirty'
+    args layout.buildDirectory.file('reports/generalization/dictionary-generalization-exact-model.csv')
+            .get().asFile.absolutePath,
+            dictionaryGeneralizationReleaseVersion.get(),
+            providers.exec { commandLine 'git', 'rev-parse', 'HEAD' }.standardOutput.asText.get().trim(),
+            dictionaryGeneralizationSourceState,
+            layout.projectDirectory.file('src/test/java/org/egothor/stemmer/benchmark/generalization/DictionaryGeneralizationApplication.java').asFile.absolutePath,
+            '--exact-model', exactModel
+    doFirst {
+        if (exactModel.isBlank()) {
+            throw new GradleException('Set -PdictionaryGeneralizationModel to one exact default model ID.')
+        }
+    }
+    maxHeapSize = '8g'
+}
+
 tasks.register('editCostSensitivity', JavaExec) {
     group = 'verification'
     description = 'Evaluates normalized edit-cost ratios across knowledge levels and deterministic dictionary splits.'
@@ -1285,18 +1408,35 @@ tasks.register('checkFilteredModels') {
 tasks.register('publishDictionaryGeneralizationDocumentation', Exec) {
     group = 'documentation'
     description = 'Publishes the validated multilingual dictionary-generalization report.'
-    dependsOn(tasks.named('dictionaryGeneralization'))
+    dependsOn('mergeDictionaryGeneralizationSnapshots')
     commandLine 'python3', layout.projectDirectory.file('tools/update-generalization-documentation.py').asFile,
-            layout.buildDirectory.file('reports/generalization/dictionary-generalization.csv').get().asFile,
-            layout.projectDirectory.dir('docs').asFile, 'update'
+            layout.projectDirectory.file('docs/benchmarks/data/dictionary-generalization-2026-09-11.csv').asFile,
+            layout.projectDirectory.dir('docs').asFile, 'update',
+            '--corpus', layout.projectDirectory.file('docs/benchmarks/data/java-benchmark-corpora-2026-09-11.csv').asFile,
+            '--snapshot-name', 'dictionary-generalization-2026-09-11.csv'
+}
+
+tasks.register('mergeDictionaryGeneralizationSnapshots', Exec) {
+    group = 'documentation'
+    description = 'Merges the immutable 20-model archive with the active standalone continuation.'
+    dependsOn(tasks.named('dictionaryGeneralizationStandalone'))
+    commandLine 'python3', layout.projectDirectory.file('tools/merge-generalization-snapshots.py').asFile,
+            '--archive', layout.projectDirectory.file('docs/benchmarks/data/dictionary-generalization.csv').asFile,
+            '--continuation', layout.buildDirectory.file('reports/generalization/dictionary-generalization-standalone-2026-09-11.csv').get().asFile,
+            '--corpus', layout.projectDirectory.file('docs/benchmarks/data/java-benchmark-corpora-2026-09-11.csv').asFile,
+            '--topology', layout.projectDirectory.file('models/model-projects.properties').asFile,
+            '--output', layout.projectDirectory.file('docs/benchmarks/data/dictionary-generalization-2026-09-11.csv').asFile,
+            '--manifest', layout.projectDirectory.file('docs/benchmarks/data/active-snapshots.properties').asFile
 }
 
 tasks.register('verifyDictionaryGeneralizationDocumentation', Exec) {
     group = 'verification'
     description = 'Verifies the rendered generalization page and checked-in raw counters.'
     commandLine 'python3', layout.projectDirectory.file('tools/update-generalization-documentation.py').asFile,
-            layout.projectDirectory.file('docs/benchmarks/data/dictionary-generalization.csv').asFile,
-            layout.projectDirectory.dir('docs').asFile, 'verify'
+            layout.projectDirectory.file('docs/benchmarks/data/dictionary-generalization-2026-09-11.csv').asFile,
+            layout.projectDirectory.dir('docs').asFile, 'verify',
+            '--corpus', layout.projectDirectory.file('docs/benchmarks/data/java-benchmark-corpora-2026-09-11.csv').asFile,
+            '--snapshot-name', 'dictionary-generalization-2026-09-11.csv'
 }
 
 tasks.register('publishEditCostSensitivityDocumentation', Exec) {
@@ -1323,20 +1463,28 @@ tasks.register('verifyPublishedPerformanceSnapshotChecksums', Exec) {
     commandLine 'sha256sum', '--check', 'performance-snapshots.sha256'
 }
 
+tasks.register('verifyActiveBenchmarkSnapshots', Exec) {
+    group = 'verification'
+    description = 'Verifies active benchmark pointers, selected checksums, and immutable archives.'
+    commandLine 'python3', layout.projectDirectory.file('tools/verify-active-benchmark-snapshots.py').asFile
+}
+
 tasks.register('verifyPublishedJavaBenchmarkDocumentation', Exec) {
     group = 'verification'
-    description = 'Verifies Java benchmark documentation against the checked-in 2026-08-25 reports.'
+    description = 'Verifies Java benchmark documentation against the active 2026-09-11 reports.'
     dependsOn(tasks.named('verifyPublishedPerformanceSnapshotChecksums'))
     commandLine 'python3', layout.projectDirectory.file('tools/update-benchmark-documentation.py').asFile,
             '--docs-root', layout.projectDirectory.dir('docs').asFile,
             '--readme', layout.projectDirectory.file('README.md').asFile,
-            '--corpus', layout.projectDirectory.file('docs/benchmarks/data/java-benchmark-corpora-2026-08-25.csv').asFile,
-            '--accuracy', layout.projectDirectory.file('docs/benchmarks/data/java-stemmer-accuracy-2026-08-25.csv').asFile,
-            '--speed', layout.projectDirectory.file('docs/benchmarks/data/java-stemmer-speed-2026-08-25.csv').asFile,
-            '--coverage-accuracy', layout.projectDirectory.file('docs/benchmarks/data/java-english-coverage-accuracy-2026-08-25.csv').asFile,
-            '--coverage-speed', layout.projectDirectory.file('docs/benchmarks/data/java-english-coverage-speed-2026-08-25.csv').asFile,
-            '--date', '2026-08-25',
-            '--release-version', '4.2.0-6-g84e57fb',
+            '--corpus', layout.projectDirectory.file('docs/benchmarks/data/java-benchmark-corpora-2026-09-11.csv').asFile,
+            '--accuracy', layout.projectDirectory.file('docs/benchmarks/data/java-stemmer-accuracy-2026-09-11.csv').asFile,
+            '--speed', layout.projectDirectory.file('docs/benchmarks/data/java-stemmer-speed-2026-09-11.csv').asFile,
+            '--coverage-accuracy', layout.projectDirectory.file('docs/benchmarks/data/java-english-coverage-accuracy-2026-09-10.csv').asFile,
+            '--coverage-speed', layout.projectDirectory.file('docs/benchmarks/data/java-english-coverage-speed-2026-09-10.csv').asFile,
+            '--quality', layout.projectDirectory.file('docs/benchmarks/data/stemming-quality-2026-09-11.csv').asFile,
+            '--snowball-catalog', layout.projectDirectory.file('docs/benchmarks/data/snowball-language-cases.csv').asFile,
+            '--date', '2026-09-11',
+            '--release-version', '4.3.0.dirty',
             '--mode', 'verify'
 }
 
@@ -1371,12 +1519,15 @@ tasks.register('writeJmhRuntimeClasspath') {
 
 tasks.register('prepareBenchmarkModelInputs', Sync) {
     group = 'verification'
-    description = 'Prepares default model inputs for JMH and quality evaluation without changing source data.'
+    description = 'Prepares all registered language-default inputs for JMH and quality evaluation without changing source data.'
     into(layout.buildDirectory.dir('generated/benchmark-model-inputs'))
-    defaultModelProjects().each { Project modelProject ->
-        String languageDirectory = modelProject.name == 'pl-pl-unimorph'
-                ? 'pl_pl'
-                : modelProject.name.replace('-default', '').replace('-', '_')
+    languageDefaultModelProjects().each { Project modelProject ->
+        final String modelBuildText = modelProject.file('build.gradle').getText('UTF-8')
+        final java.util.regex.Matcher languageMatcher = modelBuildText =~ /(?m)^\s*language\s*=\s*'([A-Z_]+)'\s*$/
+        if (!languageMatcher.find()) {
+            throw new GradleException("Model ${modelProject.path} does not declare a language.")
+        }
+        final String languageDirectory = languageMatcher.group(1).toLowerCase(java.util.Locale.ROOT)
         from(modelProject.file('src/modelInput/stemmer.gz')) {
             into(languageDirectory)
         }
@@ -1391,7 +1542,8 @@ tasks.register('publishStemmingQualityDocumentation', JavaExec) {
     mainClass = 'org.egothor.stemmer.benchmark.quality.StemmingQualityDocumentationPublisher'
     args layout.buildDirectory.file('reports/stemming-quality/stemming-quality.csv').get().asFile.absolutePath,
             layout.projectDirectory.dir('docs').asFile.absolutePath,
-            'update'
+            'update',
+            'stemming-quality-2026-09-11.csv'
     doFirst {
         if (!file("$buildDir/reports/stemming-quality/stemming-quality.csv").isFile()) {
             throw new GradleException('A complete stemming-quality CSV is required. Run stemmingQuality only when no validated complete report is available.')
@@ -1405,12 +1557,47 @@ tasks.register('verifyStemmingQualityDocumentation', JavaExec) {
     dependsOn(tasks.named('testClasses'))
     classpath = sourceSets.test.runtimeClasspath
     mainClass = 'org.egothor.stemmer.benchmark.quality.StemmingQualityDocumentationPublisher'
-    args layout.projectDirectory.file('docs/benchmarks/data/stemming-quality.csv').asFile.absolutePath,
+    args layout.projectDirectory.file('docs/benchmarks/data/stemming-quality-2026-09-11.csv').asFile.absolutePath,
             layout.projectDirectory.dir('docs').asFile.absolutePath,
+            'verify',
+            'stemming-quality-2026-09-11.csv'
+}
+
+tasks.register('publishActiveStemmingQualityDocumentation', JavaExec) {
+    group = 'documentation'
+    description = 'Publishes the checksum-bound active stemming-quality snapshot on every language page.'
+    dependsOn(tasks.named('testClasses'))
+    classpath = sourceSets.test.runtimeClasspath
+    mainClass = 'org.egothor.stemmer.benchmark.quality.StemmingQualityDocumentationPublisher'
+    args layout.projectDirectory.file('docs/benchmarks/data/stemming-quality-2026-09-11.csv').asFile.absolutePath,
+            layout.projectDirectory.dir('docs').asFile.absolutePath,
+            'update',
+            'stemming-quality-2026-09-11.csv'
+}
+
+tasks.register('publishSnowballLanguageCatalog', JavaExec) {
+    group = 'documentation'
+    description = 'Publishes the documentation Snowball catalog from the Java benchmark authority.'
+    dependsOn(tasks.named('jmhClasses'))
+    classpath = sourceSets.jmh.runtimeClasspath
+    mainClass = 'org.egothor.stemmer.benchmark.SnowballLanguageCatalogApplication'
+    args layout.projectDirectory.file('docs/benchmarks/data/snowball-language-cases.csv').asFile.absolutePath,
+            'update'
+}
+
+tasks.register('verifySnowballLanguageCatalog', JavaExec) {
+    group = 'verification'
+    description = 'Verifies the documentation Snowball catalog against the Java benchmark authority.'
+    dependsOn(tasks.named('jmhClasses'))
+    classpath = sourceSets.jmh.runtimeClasspath
+    mainClass = 'org.egothor.stemmer.benchmark.SnowballLanguageCatalogApplication'
+    args layout.projectDirectory.file('docs/benchmarks/data/snowball-language-cases.csv').asFile.absolutePath,
             'verify'
 }
 
 tasks.named('check') {
+    dependsOn(tasks.named('verifyModelQuarantine'))
+    dependsOn(tasks.named('verifyActiveBenchmarkSnapshots'))
     dependsOn(tasks.named('verifyStemmingQualityDocumentation'))
     dependsOn(tasks.named('verifyDictionaryGeneralizationDocumentation'))
     dependsOn(tasks.named('verifyEditCostSensitivityDocumentation'))
diff --git a/docs/architecture.md b/docs/architecture.md
index b1b6e27..541fc1f 100644
--- a/docs/architecture.md
+++ b/docs/architecture.md
@@ -11,7 +11,7 @@ This document explains the structural architecture of **Radixor**: what data is
 | `StemmerModelRegistry` | Deterministic index/descriptor discovery and selection by model ID or language default |
 | `StemmerModelDescriptor` | Immutable public view of validated runtime identity, format, resource, checksum, and source URL |
 | Model convention plugin | Validates inputs and generates the resource namespace, descriptor, index, license, and publication |
-| Standard aggregate | POM-only transitive runtime dependencies for one default per language |
+| Standard aggregate | POM-only transitive runtime dependencies for the original 20 standard models |
 | Verification classpaths | Direct individual-model dependencies for tests, quality evaluation, and JMH, including optional PoliMorf |
 | Models BOM | POM-only recommended individual model versions in Maven dependency management |
 | Documentation staging | Reviewed `docs/`, including the checked-in generated catalog; staging independently regenerates a byte-identical catalog under `build/mkdocs-source/` |
@@ -77,7 +77,15 @@ No equality relationship is implied between these values.
 
 ## Build topology and generated output
 
-`models/model-projects.properties` is the single Gradle-readable topology list for the 21 individual model projects and their default or optional aggregate role. Per-model build scripts and generated descriptors remain authoritative for language, resource, provenance, checksum, and model-specific metadata. `settings.gradle`, root verification classpaths, the standard POM, and BOM constraints all derive membership from the topology list.
+`models/model-projects.properties` is the single Gradle-readable topology list
+for the 144 user-facing model projects. The `default` role selects the original
+20 standard-aggregate members, `standalone` identifies 123 registered language
+defaults that remain individually publishable, and `optional` identifies
+PoliMorf. All three roles participate in settings, verification classpaths,
+JMH, the BOM, catalog generation, and individual publication; only `default`
+participates in the Java and Python standard aggregates. Per-model build
+scripts and generated descriptors remain authoritative for language, writing
+direction, resource, provenance, checksum, and model-specific metadata.
 
 Gradle implicitly creates the lifecycle parent `:models` because child paths are nested. It has no build script, applied project plugin, Maven coordinate, publication, or archive. The root CycloneDX plugin exposes direct-task instances to subprojects internally; every subproject instance is disabled, so only root `:cyclonedxDirectBom` can generate an SBOM. The ignored path `models/build/` is generated output, not a module, and the supported build does not write reports there. Root aggregate reports, including `verifyJmhModelClasspath`, belong under `build/reports/models/`; each individual model retains its own outputs under `models/<model-id>/build/`.
 
@@ -89,8 +97,9 @@ The `org.egothor.radixor.model` convention plugin treats `src/modelInput` as imm
 
 For UniMorph models, the convention validates and packages one model-specific attribution,
 licensing, provenance, and contribution notice. Source and packaged notice bytes must match. The
-notice identifies CC BY-SA 3.0 through its canonical URI; no project-wide CC license directory or
-duplicated full legal text is used. Descriptors distinguish exact revisions from the explicit
+notice identifies its audited license through the canonical URI; no project-wide model-data
+license is used. The LGPLLR Khaling artifact also packages the complete canonical license text and
+legible dictionary form. Descriptors distinguish exact revisions from the explicit
 legacy-import sentinel. UniMorph supplies morphological data; runtime patch commands and tries are
 constructed by Radixor. The Java software remains BSD-3-Clause, while PoliMorf data remains under
 its separately packaged BSD-2-Clause license.
diff --git a/docs/assets/stylesheets/landing-v2.css b/docs/assets/stylesheets/landing-v2.css
index 35cb9f8..4ba67b2 100644
--- a/docs/assets/stylesheets/landing-v2.css
+++ b/docs/assets/stylesheets/landing-v2.css
@@ -48,6 +48,32 @@
 
 .rx2-evidence-line{margin:10px 2px 0;color:#52617a;font-size:.68rem;line-height:1.5}.rx2-evidence-line strong{color:#263751;font-weight:750}.rx2-evidence-line a{color:#1769ef;font-weight:700;text-decoration:none}
 
+/* Dense benchmark summaries remain legible at narrow widths. */
+.rx2-highlight-stats{grid-template-columns:repeat(4,minmax(0,1fr))}
+.rx2-highlight-stat{min-width:0}
+.rx2-highlight-stat strong{max-width:100%;font-size:clamp(1.35rem,3.2vw,1.9rem);white-space:nowrap}
+
+/* Symmetric log-ratio chart: the center axis is equal runtime (1x). */
+.rx2-ratio-chart{margin-top:10px;padding:8px 10px 10px;border:1px solid #dce6f2;border-radius:11px;background:#fbfdff}
+.rx2-ratio-axis,.rx2-ratio-row{display:grid;grid-template-columns:minmax(92px,1.1fr) minmax(180px,4fr) 66px;gap:9px;align-items:center}
+.rx2-ratio-axis{margin-bottom:5px;color:#66748b;font-size:.55rem}
+.rx2-ratio-axis>*:nth-child(2){text-align:center}
+.rx2-ratio-axis>*:last-child{text-align:right}
+.rx2-ratio-row{min-height:21px;font-size:.57rem}
+.rx2-ratio-row>span:first-child{overflow:hidden;text-overflow:ellipsis;white-space:nowrap;color:#33415a}
+.rx2-ratio-row>b{text-align:right;color:#23385f;font-size:.58rem;font-variant-numeric:tabular-nums}
+.rx2-ratio-track{position:relative;height:11px;border-radius:5px;background:linear-gradient(90deg,#eef2ff 0 49.7%,#99a9c2 49.7% 50.3%,#edfbf8 50.3% 100%)}
+.rx2-ratio-track i{position:absolute;left:var(--left);width:max(2px,var(--width));height:100%;border-radius:4px;background:linear-gradient(90deg,#5b45ec,#08aa91)}
+
+.rx2-language-icon{display:grid;width:32px;height:32px;margin:auto;place-items:center;border-radius:50%;background:#f0f5fb;font-style:normal;font-size:1.35rem;line-height:1}
+
+@media(max-width:650px){
+  .rx2-highlight-stats{grid-template-columns:repeat(2,minmax(0,1fr))}
+  .rx2-highlight-stat strong{font-size:clamp(1.2rem,7vw,1.75rem)}
+  .rx2-ratio-chart{overflow-x:auto}
+  .rx2-ratio-axis,.rx2-ratio-row{grid-template-columns:92px minmax(150px,1fr) 58px;min-width:330px}
+}
+
 /* Benchmark-linked language badges */
 .rx2-language {
   text-decoration: none;
diff --git a/docs/benchmarks/data/performance-snapshots.sha256 b/docs/benchmarks/data/performance-snapshots.sha256
index 9a597dc..97a5ad3 100644
--- a/docs/benchmarks/data/performance-snapshots.sha256
+++ b/docs/benchmarks/data/performance-snapshots.sha256
@@ -3,5 +3,13 @@ c6ae08cc0d32b61e59276e5d6d4911f23029a5c783d3fe2bc293d15c00991dd9  java-benchmark
 1a0fefdcc773a46a5ab2155afe4c10a3043dfd56b834119a2aeed48d3c8ac1fe  java-stemmer-speed-2026-08-25.csv
 7bfa6bc150bfa626607f5529a68f5f6cd662c9d2404c0bc7ebf5be6bb3e5e1b9  java-english-coverage-accuracy-2026-08-25.csv
 e119f63793aa405f649c21e9c855837e7c3ff7e094b54a57af01424354915fe1  java-english-coverage-speed-2026-08-25.csv
+09cb331a28fbcce9d1b8c9e3e2ee85797b5ef3f1a41d06dea58d651eec0dd20e  java-benchmark-corpora-2026-09-10.csv
+886266fe981535115f92914b88fd1b862efe1ca18da5f409e10b3e67b951b0b4  java-stemmer-accuracy-2026-09-10.csv
+165b653523cdf04ebf8430044876efbade80aaddb5893edaad943b24858547c0  java-stemmer-speed-2026-09-10.csv
+1ba2e3e0473454839975e0dc89c9cdd7a9273447559df1b8a57e5a553ebee730  java-english-coverage-accuracy-2026-09-10.csv
+b45062a085af4cfb520ceea9360d7401ac45fb8d22fe91ea10cfd5b0f607e799  java-english-coverage-speed-2026-09-10.csv
+f6624fdc966fe0aa2476a8593b87ed77576b5e31c255737a2bc281485e104c4a  java-benchmark-corpora-2026-09-11.csv
+61581612523036b2cbe765c87c351909eca9aaa5c7ee4cd3389fcfd180c65456  java-stemmer-accuracy-2026-09-11.csv
+4090940bdbc7bcf89ddf24b596e2639d9d0b25c84824823612470358bdbd11ca  java-stemmer-speed-2026-09-11.csv
 f248c84075cf4b5aa917032327d33020adae68c41041e4379652b5325a9e780b  python-all-languages-batch-2026-08-25.csv
 43d0de0301d7a6267e08ba3c3608232eb56981cda2722c08c604f9734b62abec  python-all-languages-batch-2026-08-25.json
diff --git a/docs/benchmarks/generalization.md b/docs/benchmarks/generalization.md
index 983ec3f..9b951e4 100644
--- a/docs/benchmarks/generalization.md
+++ b/docs/benchmarks/generalization.md
@@ -4,7 +4,7 @@ Radixor learns transformations from lexical families rather than storing a close
 word-to-root answer list. This experiment measures how those transformations transfer
 to dictionary families that were not used to build the Java trie.
 
-For every one of the 20 default models, complete dictionary rows are placed in a
+For every one of the 143 default models, complete dictionary rows are placed in a
 frozen pseudorandom order. Exact-size, nested prefixes retain 10% through 100% of
 the rows for training. Five predeclared splits are evaluated against the complete
 dictionary; the primary `Unseen` columns exclude a held-out occurrence whenever its
@@ -17,31 +17,31 @@ out-of-distribution text. See the [methodology and limitations](reference/genera
 
 ## All-Language Summary
 
-Each cell is the language-macro mean of 20 per-language split medians, so large
+Each cell is the language-macro mean across languages with a defined denominator, so large
 dictionaries do not dominate small ones. Changed-form exactness is the most demanding
 measure because it excludes words whose expected root is already the input token.
 
 | Training rows | Unseen all exact | Unseen changed exact | Unseen root preserved |
 | ---: | ---: | ---: | ---: |
 | 100% | n/a | n/a | n/a |
-| 90% | 74.21% | 69.07% | 94.66% |
-| 80% | 73.77% | 68.54% | 94.24% |
-| 70% | 73.53% | 68.20% | 93.83% |
-| 60% | 72.87% | 67.55% | 93.50% |
-| 50% | 72.31% | 66.80% | 93.17% |
-| 40% | 71.60% | 65.97% | 92.67% |
-| 30% | 70.51% | 64.72% | 92.14% |
-| 20% | 69.19% | 63.13% | 91.55% |
-| 10% | 66.96% | 60.49% | 90.80% |
+| 90% | 52.61% | 44.54% | 95.46% |
+| 80% | 51.73% | 43.77% | 95.11% |
+| 70% | 51.33% | 43.33% | 94.83% |
+| 60% | 50.55% | 42.55% | 94.65% |
+| 50% | 50.01% | 41.99% | 94.15% |
+| 40% | 49.46% | 41.40% | 93.90% |
+| 30% | 48.48% | 40.32% | 93.64% |
+| 20% | 47.04% | 38.80% | 93.69% |
+| 10% | 44.68% | 36.22% | 93.20% |
 
 The language range is material and is therefore not replaced by the macro mean. At 10%
-knowledge, median unseen changed-form exactness ranges from **10.423%**
-for Hebrew to **90.053%** for
-Portuguese. At 90%, it ranges from **15.493%**
-for Persian to **94.412%** for
-Portuguese. The largest 10%–90% change is
-**+19.147 pp** for
-Norwegian Nynorsk. These are descriptive within-resource results, not a causal
+knowledge, median unseen changed-form exactness ranges from **0.000%**
+for Akan to **94.807%** for
+Bashkir. At 90%, it ranges from **0.000%**
+for Akan to **95.477%** for
+Crimean Tatar. The largest 10%–90% change is
+**+59.536 pp** for
+Cree. These are descriptive within-resource results, not a causal
 ranking of language, script, dictionary size, or regularity.
 Every language's evidence-derived endpoint conclusion is also published on its separate
 [language benchmark page](languages/index.md), rather than inferred from this macro mean.
@@ -53,62 +53,2502 @@ split range. `Whole all exact` retains comparability with the earlier English co
 curve but mixes trained and held-out rows; the unseen columns are the generalization
 evidence. At 100% there is no held-out set, so unseen metrics are `n/a`.
 
+<details markdown="block">
+<summary><strong>Adyghe</strong> — <code>ady-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 1,635 / 1,635 | 0 | 0 | n/a | n/a | n/a | 99.907% |
+| 90% | 1,472 / 1,635 | 2,029 | 2 | 63.708% (60.402–72.893) | 61.034% (57.760–71.222) | 90.798% (89.506–94.479) | 96.273% (95.944–97.211) |
+| 80% | 1,308 / 1,635 | 4,056 | 6 | 65.656% (64.362–68.064) | 63.475% (62.127–66.184) | 89.877% (89.571–92.025) | 93.067% (92.831–93.558) |
+| 70% | 1,145 / 1,635 | 6,088 | 7 | 65.177% (64.769–68.561) | 62.851% (62.366–66.643) | 91.376% (90.574–92.229) | 89.502% (89.404–90.523) |
+| 60% | 981 / 1,635 | 8,135 | 8 | 64.665% (62.917–66.052) | 62.266% (60.499–63.814) | 91.718% (90.015–92.649) | 85.815% (85.073–86.369) |
+| 50% | 818 / 1,635 | 10,164 | 9 | 65.683% (61.154–65.872) | 63.372% (58.603–63.687) | 91.043% (90.529–93.252) | 82.800% (80.517–82.908) |
+| 40% | 654 / 1,635 | 12,212 | 8 | 63.323% (60.215–63.904) | 60.819% (57.274–61.521) | 91.224% (90.900–93.878) | 77.934% (76.112–78.292) |
+| 30% | 491 / 1,635 | 14,235 | 7 | 61.064% (57.064–63.036) | 58.469% (53.887–60.594) | 91.068% (89.336–93.514) | 72.724% (69.940–74.109) |
+| 20% | 327 / 1,635 | 16,279 | 6 | 60.496% (57.284–63.137) | 57.881% (54.289–60.874) | 90.192% (89.059–91.584) | 68.364% (65.845–70.515) |
+| 10% | 164 / 1,635 | 18,317 | 3 | 60.912% (55.294–61.045) | 58.302% (52.020–58.764) | 90.136% (87.202–92.794) | 64.843% (59.756–64.942) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Afrikaans</strong> — <code>af-za-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 170,974 / 170,974 | 0 | 0 | n/a | n/a | n/a | 98.544% |
+| 90% | 153,877 / 170,974 | 29,397 | 783 | 91.405% (91.373–91.546) | 82.230% (81.935–82.270) | 98.382% (98.253–98.452) | 97.722% (97.710–97.741) |
+| 80% | 136,779 / 170,974 | 59,014 | 1,374 | 91.038% (90.962–91.290) | 81.473% (81.359–82.033) | 98.239% (98.177–98.301) | 96.847% (96.839–96.904) |
+| 70% | 119,682 / 170,974 | 88,657 | 1,817 | 90.796% (90.649–91.051) | 81.075% (80.734–81.540) | 98.164% (98.107–98.193) | 95.976% (95.927–96.032) |
+| 60% | 102,584 / 170,974 | 118,535 | 2,098 | 90.496% (90.414–90.729) | 80.504% (80.357–80.939) | 98.017% (98.008–98.091) | 95.049% (95.006–95.113) |
+| 50% | 85,487 / 170,974 | 148,525 | 2,175 | 90.278% (90.139–90.418) | 80.134% (79.899–80.400) | 97.915% (97.820–97.968) | 94.119% (94.044–94.170) |
+| 40% | 68,390 / 170,974 | 178,815 | 2,135 | 89.906% (89.824–90.038) | 79.704% (79.440–79.787) | 97.765% (97.634–97.786) | 93.069% (93.021–93.154) |
+| 30% | 51,292 / 170,974 | 209,118 | 1,874 | 89.518% (89.369–89.604) | 78.972% (78.656–79.104) | 97.556% (97.515–97.684) | 91.974% (91.869–92.039) |
+| 20% | 34,195 / 170,974 | 239,864 | 1,406 | 88.938% (88.796–89.043) | 77.870% (77.696–78.045) | 97.372% (97.239–97.417) | 90.681% (90.555–90.740) |
+| 10% | 17,097 / 170,974 | 270,632 | 804 | 87.881% (87.749–88.168) | 75.953% (75.644–76.671) | 96.964% (96.926–96.972) | 88.840% (88.718–89.104) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Aimele</strong> — <code>ail-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 636 / 636 | 0 | 0 | n/a | n/a | n/a | 97.702% |
+| 90% | 572 / 636 | 385 | 14 | 16.883% (13.127–25.103) | 2.020% (1.747–2.752) | 95.000% (93.103–96.774) | 87.520% (83.371–91.893) |
+| 80% | 509 / 636 | 616 | 23 | 19.934% (15.162–24.950) | 1.670% (1.412–2.605) | 94.309% (91.057–97.521) | 81.998% (75.263–85.669) |
+| 70% | 445 / 636 | 866 | 33 | 21.479% (18.200–23.916) | 1.799% (1.434–2.253) | 94.054% (92.350–95.676) | 76.253% (71.210–79.381) |
+| 60% | 382 / 636 | 1,205 | 35 | 21.245% (20.238–23.425) | 2.077% (1.680–2.519) | 95.161% (91.463–95.968) | 67.826% (65.592–73.125) |
+| 50% | 318 / 636 | 1,514 | 36 | 21.400% (20.619–24.302) | 1.931% (1.207–2.579) | 93.871% (92.258–95.793) | 60.294% (58.730–68.592) |
+| 40% | 254 / 636 | 1,783 | 32 | 21.287% (21.134–21.967) | 1.986% (1.618–2.284) | 94.906% (92.761–96.543) | 53.687% (51.420–55.474) |
+| 30% | 191 / 636 | 2,102 | 29 | 21.509% (19.918–21.690) | 1.562% (1.480–2.719) | 94.050% (93.593–94.977) | 45.547% (42.739–47.750) |
+| 20% | 127 / 636 | 2,521 | 21 | 20.447% (19.767–20.903) | 1.686% (1.415–1.809) | 94.800% (91.929–95.652) | 35.238% (33.387–38.047) |
+| 10% | 64 / 636 | 2,836 | 11 | 19.746% (18.628–21.392) | 1.598% (1.024–2.079) | 93.182% (90.893–95.431) | 26.939% (24.162–31.120) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Akan</strong> — <code>aka-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 79 / 79 | 0 | 0 | n/a | n/a | n/a | 99.684% |
+| 90% | 71 / 79 | 192 | 0 | 4.167% (3.646–4.167) | 0.000% | 100.000% (87.500–100.000) | 89.979% (89.926–89.979) |
+| 80% | 63 / 79 | 384 | 0 | 4.167% (3.906–4.233) | 0.000% | 100.000% (93.750–100.000) | 80.274% (80.222–80.591) |
+| 70% | 55 / 79 | 576 | 0 | 4.167% (4.167–4.211) | 0.000% | 100.000% | 70.570% (70.570–70.886) |
+| 60% | 47 / 79 | 768 | 0 | 4.167% (4.036–4.199) | 0.000% | 100.000% (96.875–100.000) | 61.129% (60.865–61.181) |
+| 50% | 40 / 79 | 936 | 0 | 4.167% (4.167–4.194) | 0.000% | 100.000% | 52.690% (52.373–52.690) |
+| 40% | 32 / 79 | 1,128 | 0 | 4.167% (4.100–4.167) | 0.000% | 100.000% (97.872–100.000) | 42.932% (42.669–42.985) |
+| 30% | 24 / 79 | 1,320 | 0 | 4.167% (4.091–4.186) | 0.000% | 100.000% (98.182–100.000) | 33.281% (32.911–33.281) |
+| 20% | 16 / 79 | 1,512 | 0 | 4.167% (4.101–4.183) | 0.000% | 100.000% (98.413–100.000) | 23.576% (23.523–23.576) |
+| 10% | 8 / 79 | 1,704 | 0 | 4.167% (4.108–4.167) | 0.000% | 100.000% (98.592–100.000) | 13.871% (13.819–13.871) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Albanian</strong> — <code>sq-al-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 587 / 587 | 0 | 0 | n/a | n/a | n/a | 99.552% |
+| 90% | 528 / 587 | 940 | 9 | 78.064% (76.395–84.163) | 77.573% (75.406–83.859) | 89.655% (83.051–91.525) | 97.808% (97.467–97.847) |
+| 80% | 470 / 587 | 1,985 | 10 | 76.717% (73.480–84.213) | 75.953% (72.512–84.056) | 88.034% (86.842–90.598) | 95.090% (93.901–96.444) |
+| 70% | 411 / 587 | 2,961 | 21 | 77.069% (75.762–81.269) | 76.346% (74.809–80.997) | 86.857% (85.632–91.429) | 92.936% (92.342–94.184) |
+| 60% | 352 / 587 | 4,029 | 24 | 77.039% (75.500–79.709) | 76.328% (74.854–79.250) | 87.234% (85.897–90.987) | 90.559% (90.004–91.758) |
+| 50% | 294 / 587 | 5,026 | 25 | 77.370% (75.716–78.898) | 76.705% (75.125–78.182) | 86.007% (84.880–90.722) | 88.416% (87.987–89.137) |
+| 40% | 235 / 587 | 6,104 | 19 | 75.168% (74.636–79.673) | 74.350% (73.866–79.255) | 86.610% (84.943–88.889) | 84.811% (84.616–87.646) |
+| 30% | 176 / 587 | 7,117 | 20 | 74.795% (73.907–76.690) | 74.149% (73.099–76.025) | 85.330% (84.597–87.561) | 82.424% (81.352–83.574) |
+| 20% | 117 / 587 | 8,179 | 17 | 73.560% (70.760–74.716) | 72.611% (69.902–73.998) | 88.085% (84.681–89.362) | 78.478% (76.705–79.657) |
+| 10% | 59 / 587 | 9,213 | 10 | 70.692% (66.959–71.613) | 69.493% (66.162–70.796) | 85.769% (80.114–90.720) | 73.207% (70.129–74.542) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Alsatian</strong> — <code>gsw-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 145 / 145 | 0 | 0 | n/a | n/a | n/a | 99.664% |
+| 90% | 131 / 145 | 115 | 0 | 58.261% (47.500–62.832) | 52.475% (41.509–57.576) | 100.000% (92.857–100.000) | 95.630% (94.370–96.134) |
+| 80% | 116 / 145 | 239 | 1 | 58.333% (53.975–64.035) | 52.607% (48.095–58.794) | 100.000% (93.103–100.000) | 91.261% (90.420–92.773) |
+| 70% | 102 / 145 | 351 | 1 | 57.434% (48.876–60.114) | 52.000% (43.131–54.545) | 97.674% (90.698–100.000) | 87.395% (84.370–87.899) |
+| 60% | 87 / 145 | 474 | 2 | 53.376% (51.483–58.656) | 47.356% (45.652–53.349) | 98.276% (93.103–98.276) | 81.092% (80.084–82.941) |
+| 50% | 73 / 145 | 583 | 2 | 55.326% (50.257–58.895) | 49.216% (44.031–53.254) | 98.611% (94.444–98.611) | 77.815% (74.454–79.748) |
+| 40% | 58 / 145 | 704 | 2 | 54.298% (50.427–58.635) | 47.954% (44.065–53.485) | 95.402% (90.805–98.851) | 72.857% (70.504–75.210) |
+| 30% | 44 / 145 | 820 | 2 | 52.683% (46.437–55.286) | 46.314% (39.832–49.584) | 96.040% (93.069–99.010) | 67.059% (63.109–68.824) |
+| 20% | 29 / 145 | 947 | 1 | 51.760% (46.845–54.136) | 45.294% (40.171–47.795) | 99.138% (93.966–100.000) | 60.840% (57.983–62.941) |
+| 10% | 15 / 145 | 1,063 | 2 | 45.908% (38.140–52.004) | 38.371% (29.762–45.387) | 99.231% (94.615–100.000) | 51.429% (44.958–56.555) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Amharic</strong> — <code>am-et-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 2,461 / 2,461 | 0 | 0 | n/a | n/a | n/a | 97.282% |
+| 90% | 2,215 / 2,461 | 4,069 | 202 | 46.550% (39.916–47.751) | 43.331% (36.518–44.590) | 97.531% (96.680–98.354) | 92.023% (91.199–92.360) |
+| 80% | 1,969 / 2,461 | 8,235 | 333 | 43.519% (38.970–45.328) | 40.003% (35.355–42.111) | 97.101% (95.868–97.938) | 86.729% (85.352–86.988) |
+| 70% | 1,723 / 2,461 | 12,121 | 377 | 41.850% (39.000–44.269) | 38.493% (35.534–40.924) | 97.383% (95.879–97.514) | 80.307% (79.306–81.765) |
+| 60% | 1,477 / 2,461 | 16,542 | 525 | 40.444% (38.470–42.177) | 36.906% (34.866–38.795) | 96.498% (96.300–96.798) | 74.547% (73.555–75.206) |
+| 50% | 1,231 / 2,461 | 20,765 | 580 | 39.639% (38.697–40.861) | 36.084% (35.202–37.461) | 96.623% (95.738–96.880) | 68.664% (67.180–69.069) |
+| 40% | 984 / 2,461 | 25,111 | 594 | 39.166% (36.597–39.775) | 35.718% (32.959–36.175) | 96.575% (94.956–96.995) | 62.103% (59.969–63.249) |
+| 30% | 738 / 2,461 | 29,207 | 581 | 36.289% (32.838–37.946) | 32.766% (28.996–34.293) | 96.021% (95.047–96.096) | 54.439% (51.594–55.963) |
+| 20% | 492 / 2,461 | 33,549 | 530 | 34.913% (32.261–35.773) | 31.169% (28.327–32.005) | 95.621% (95.262–96.478) | 47.254% (45.012–48.465) |
+| 10% | 246 / 2,461 | 38,138 | 231 | 31.003% (28.964–31.314) | 27.034% (24.945–27.301) | 95.436% (94.024–95.654) | 37.408% (35.239–38.481) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Amol Azerbaijani</strong> — <code>azg-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 258 / 258 | 0 | 0 | n/a | n/a | n/a | 99.259% |
+| 90% | 232 / 258 | 917 | 27 | 2.729% (2.658–2.873) | 0.000% | 100.000% | 89.821% (89.175–89.948) |
+| 80% | 206 / 258 | 1,885 | 31 | 2.748% (2.665–2.761) | 0.000% | 100.000% (98.077–100.000) | 79.843% (79.144–80.256) |
+| 70% | 181 / 258 | 2,780 | 23 | 2.734% (2.664–2.777) | 0.000% | 100.000% (98.701–100.000) | 70.618% (69.908–71.189) |
+| 60% | 155 / 258 | 3,742 | 23 | 2.753% (2.688–2.803) | 0.000% | 100.000% | 61.042% (60.216–61.646) |
+| 50% | 129 / 258 | 4,690 | 26 | 2.751% (2.722–2.764) | 0.000% | 100.000% | 51.223% (50.630–51.584) |
+| 40% | 103 / 258 | 5,626 | 19 | 2.755% (2.697–2.759) | 0.000% | 100.000% (98.710–100.000) | 41.574% (41.002–42.029) |
+| 30% | 77 / 258 | 6,604 | 27 | 2.737% (2.722–2.741) | 0.000% | 100.000% (98.895–100.000) | 31.745% (31.448–31.840) |
+| 20% | 52 / 258 | 7,528 | 19 | 2.729% (2.720–2.736) | 0.000% | 99.515% (99.512–100.000) | 22.243% (22.148–22.413) |
+| 10% | 26 / 258 | 8,490 | 14 | 2.728% (2.689–2.738) | 0.000% | 100.000% (98.276–100.000) | 12.403% (12.266–12.467) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Ancient Greek</strong> — <code>grc-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 2,426 / 2,426 | 0 | 0 | n/a | n/a | n/a | 99.266% |
+| 90% | 2,183 / 2,426 | 2,639 | 23 | 27.536% (24.744–31.815) | 20.639% (17.723–25.641) | 94.606% (94.239–95.062) | 92.118% (91.806–92.367) |
+| 80% | 1,941 / 2,426 | 5,280 | 61 | 27.968% (27.018–33.015) | 21.224% (20.075–26.816) | 95.445% (94.617–96.480) | 85.027% (84.745–85.858) |
+| 70% | 1,698 / 2,426 | 7,921 | 77 | 30.008% (27.640–31.131) | 23.347% (20.789–24.700) | 95.455% (94.759–96.000) | 78.451% (77.781–78.835) |
+| 60% | 1,456 / 2,426 | 10,608 | 87 | 29.025% (27.219–30.210) | 22.443% (20.416–23.788) | 94.720% (94.324–95.553) | 71.126% (70.517–71.536) |
+| 50% | 1,213 / 2,426 | 13,278 | 87 | 28.586% (27.607–29.318) | 21.972% (20.951–22.739) | 94.967% (93.476–95.782) | 63.820% (63.557–64.189) |
+| 40% | 970 / 2,426 | 15,909 | 77 | 27.728% (27.441–29.312) | 20.891% (20.812–22.704) | 95.314% (93.466–96.143) | 56.484% (56.078–57.120) |
+| 30% | 728 / 2,426 | 18,561 | 60 | 27.328% (25.502–28.164) | 20.377% (18.528–21.459) | 94.864% (94.752–95.868) | 49.016% (47.654–49.562) |
+| 20% | 485 / 2,426 | 21,180 | 48 | 25.386% (23.521–25.890) | 18.309% (16.276–18.991) | 95.513% (94.487–96.233) | 40.529% (38.742–40.567) |
+| 10% | 243 / 2,426 | 23,821 | 33 | 22.164% (20.788–24.125) | 14.963% (13.151–17.164) | 93.764% (91.976–96.564) | 29.984% (28.769–31.955) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Anglo-Norman</strong> — <code>xno-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 5 / 5 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 90% | 5 / 5 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 80% | 4 / 5 | 35 | 0 | 8.571% (2.703–100.000) | 5.882% (0.000–100.000) | 100.000% | 82.796% (80.645–100.000) |
+| 70% | 4 / 5 | 35 | 0 | 8.571% (2.703–100.000) | 5.882% (0.000–100.000) | 100.000% | 82.796% (80.645–100.000) |
+| 60% | 3 / 5 | 72 | 0 | 8.571% (2.469–50.000) | 5.882% (0.000–48.571) | 100.000% (50.000–100.000) | 65.591% (57.527–80.645) |
+| 50% | 3 / 5 | 72 | 0 | 8.571% (2.469–50.000) | 5.882% (0.000–48.571) | 100.000% (50.000–100.000) | 65.591% (57.527–80.645) |
+| 40% | 2 / 5 | 116 | 0 | 23.364% (21.552–44.860) | 21.154% (19.469–43.269) | 100.000% | 55.914% (51.075–68.280) |
+| 30% | 2 / 5 | 116 | 0 | 23.364% (21.552–44.860) | 21.154% (19.469–43.269) | 100.000% | 55.914% (51.075–68.280) |
+| 20% | 1 / 5 | 151 | 0 | 39.735% (0.704–39.735) | 38.095% (0.000–38.095) | 100.000% (25.000–100.000) | 51.075% (24.194–51.075) |
+| 10% | 1 / 5 | 151 | 0 | 39.735% (0.704–39.735) | 38.095% (0.000–38.095) | 100.000% (25.000–100.000) | 51.075% (24.194–51.075) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Arabic</strong> — <code>ar-default</code> 1.0.1</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 12,815 / 12,815 | 0 | 0 | n/a | n/a | n/a | 98.399% |
+| 90% | 11,534 / 12,815 | 44,486 | 1,336 | 14.855% (14.616–15.068) | 12.470% (12.255–12.672) | 98.164% (97.828–99.127) | 90.200% (89.946–90.503) |
+| 80% | 10,252 / 12,815 | 89,671 | 2,338 | 14.884% (14.592–15.182) | 12.439% (12.177–12.775) | 98.439% (98.031–98.806) | 81.877% (81.758–82.370) |
+| 70% | 8,971 / 12,815 | 134,443 | 3,017 | 14.652% (14.379–14.913) | 12.232% (11.959–12.480) | 98.596% (98.002–98.618) | 73.591% (73.377–73.871) |
+| 60% | 7,689 / 12,815 | 179,895 | 3,586 | 14.457% (14.354–14.575) | 12.045% (11.929–12.162) | 98.456% (97.892–98.610) | 65.184% (65.124–65.502) |
+| 50% | 6,408 / 12,815 | 225,546 | 3,714 | 14.114% (13.896–14.429) | 11.685% (11.460–12.000) | 98.274% (98.022–98.483) | 56.669% (56.546–57.051) |
+| 40% | 5,126 / 12,815 | 272,168 | 3,486 | 13.736% (13.635–13.847) | 11.282% (11.201–11.417) | 98.499% (98.223–98.516) | 48.011% (47.747–48.469) |
+| 30% | 3,845 / 12,815 | 318,095 | 3,060 | 13.313% (13.200–13.416) | 10.851% (10.738–10.967) | 98.538% (98.324–98.605) | 39.342% (38.968–39.697) |
+| 20% | 2,563 / 12,815 | 364,452 | 2,421 | 12.828% (12.761–12.975) | 10.371% (10.296–10.509) | 98.618% (98.431–98.763) | 30.424% (30.082–30.582) |
+| 10% | 1,282 / 12,815 | 412,269 | 1,410 | 11.991% (11.707–12.516) | 9.492% (9.200–10.055) | 98.800% (98.339–99.113) | 21.070% (20.598–21.276) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Armenian</strong> — <code>hy-am-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 6,990 / 6,990 | 0 | 0 | n/a | n/a | n/a | 99.505% |
+| 90% | 6,291 / 6,990 | 24,487 | 235 | 69.740% (68.616–72.259) | 69.080% (67.922–71.648) | 93.835% (93.383–95.152) | 96.521% (96.381–96.805) |
+| 80% | 5,592 / 6,990 | 49,373 | 396 | 69.925% (68.793–71.043) | 69.267% (68.105–70.382) | 93.568% (93.328–94.654) | 93.530% (93.317–93.827) |
+| 70% | 4,893 / 6,990 | 74,063 | 525 | 69.445% (69.174–70.424) | 68.763% (68.515–69.745) | 93.373% (92.537–94.431) | 90.446% (90.341–90.825) |
+| 60% | 4,194 / 6,990 | 98,745 | 578 | 69.346% (68.769–70.333) | 68.695% (68.106–69.680) | 93.306% (92.451–93.456) | 87.365% (87.146–87.771) |
+| 50% | 3,495 / 6,990 | 123,633 | 623 | 68.911% (68.483–69.122) | 68.226% (67.807–68.446) | 92.769% (92.473–93.117) | 84.082% (83.901–84.304) |
+| 40% | 2,796 / 6,990 | 148,341 | 573 | 68.280% (67.572–68.505) | 67.598% (66.889–67.827) | 92.302% (91.640–92.490) | 80.664% (80.255–80.854) |
+| 30% | 2,097 / 6,990 | 173,169 | 494 | 67.599% (66.815–68.266) | 66.900% (66.110–67.588) | 91.549% (90.862–92.046) | 77.114% (76.494–77.586) |
+| 20% | 1,398 / 6,990 | 198,108 | 392 | 66.540% (65.927–68.287) | 65.815% (65.216–67.626) | 91.304% (90.282–91.762) | 73.068% (72.578–74.480) |
+| 10% | 699 / 6,990 | 223,028 | 216 | 65.016% (64.402–66.160) | 64.272% (63.678–65.429) | 90.256% (89.494–91.529) | 68.418% (67.847–69.416) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Ashaninka</strong> — <code>cni-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 407 / 407 | 0 | 0 | n/a | n/a | n/a | 99.664% |
+| 90% | 366 / 407 | 1,019 | 1 | 9.582% (7.150–9.715) | 5.816% (3.425–5.930) | 100.000% (97.500–100.000) | 90.552% (89.920–90.907) |
+| 80% | 326 / 407 | 2,009 | 6 | 9.388% (8.211–10.147) | 5.519% (4.478–6.296) | 100.000% (98.750–100.000) | 82.032% (81.183–82.417) |
+| 70% | 285 / 407 | 3,016 | 21 | 9.798% (9.052–10.326) | 5.933% (5.285–6.489) | 99.174% (99.174–100.000) | 73.058% (72.643–73.670) |
+| 60% | 244 / 407 | 4,031 | 22 | 9.642% (8.805–10.083) | 5.899% (5.024–6.250) | 99.383% (98.773–100.000) | 63.827% (63.323–64.893) |
+| 50% | 204 / 407 | 5,010 | 21 | 9.014% (8.565–9.610) | 5.251% (4.798–5.746) | 99.505% (99.010–100.000) | 54.843% (54.072–55.929) |
+| 40% | 163 / 407 | 6,048 | 8 | 9.031% (8.399–9.753) | 5.224% (4.597–5.909) | 99.590% (99.180–100.000) | 45.572% (44.851–46.688) |
+| 30% | 122 / 407 | 7,103 | 8 | 8.971% (8.053–9.661) | 5.149% (4.259–5.834) | 99.649% (99.298–99.649) | 36.094% (34.949–37.457) |
+| 20% | 81 / 407 | 8,134 | 7 | 8.982% (8.608–9.487) | 5.238% (4.815–5.663) | 99.693% (99.080–100.000) | 26.893% (26.419–27.900) |
+| 10% | 41 / 407 | 9,122 | 2 | 8.724% (8.653–9.172) | 4.872% (4.857–5.393) | 99.726% (99.180–100.000) | 17.889% (17.603–18.393) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Assamese</strong> — <code>as-in-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 1,466 / 1,466 | 0 | 0 | n/a | n/a | n/a | 97.206% |
+| 90% | 1,319 / 1,466 | 6,589 | 327 | 69.829% (63.969–70.714) | 69.204% (63.366–70.198) | 96.183% (93.846–100.000) | 94.366% (93.693–94.436) |
+| 80% | 1,173 / 1,466 | 13,249 | 759 | 68.346% (65.575–71.057) | 67.743% (64.988–70.545) | 95.817% (94.340–97.753) | 91.168% (90.844–91.824) |
+| 70% | 1,026 / 1,466 | 20,148 | 888 | 67.749% (64.788–69.603) | 67.167% (64.183–69.069) | 95.309% (94.103–96.049) | 88.187% (87.500–88.532) |
+| 60% | 880 / 1,466 | 26,947 | 893 | 66.512% (65.657–67.539) | 65.905% (65.067–66.965) | 95.082% (93.885–95.795) | 84.602% (84.515–85.039) |
+| 50% | 733 / 1,466 | 33,854 | 944 | 66.293% (64.847–66.976) | 65.713% (64.238–66.403) | 94.614% (93.974–95.708) | 81.583% (80.610–81.770) |
+| 40% | 586 / 1,466 | 40,824 | 907 | 65.769% (65.293–66.521) | 65.139% (64.673–65.913) | 94.793% (93.972–95.611) | 78.074% (77.537–78.416) |
+| 30% | 440 / 1,466 | 47,838 | 910 | 65.115% (64.689–65.986) | 64.484% (64.040–65.363) | 95.110% (93.496–95.190) | 74.658% (74.261–75.198) |
+| 20% | 293 / 1,466 | 55,106 | 681 | 64.005% (63.915–64.449) | 63.345% (63.273–63.816) | 94.741% (93.945–96.007) | 70.591% (70.189–70.875) |
+| 10% | 147 / 1,466 | 62,369 | 382 | 61.905% (61.513–62.974) | 61.204% (60.788–62.265) | 95.336% (94.576–96.296) | 65.405% (64.991–66.293) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Asturian</strong> — <code>ast-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 436 / 436 | 0 | 0 | n/a | n/a | n/a | 98.896% |
+| 90% | 392 / 436 | 2,196 | 110 | 89.355% (87.773–91.678) | 89.149% (87.524–91.505) | 100.000% | 97.792% (97.691–98.076) |
+| 80% | 349 / 436 | 4,327 | 114 | 89.210% (86.525–89.996) | 88.989% (86.269–89.803) | 100.000% (97.701–100.000) | 96.775% (96.285–96.825) |
+| 70% | 305 / 436 | 6,479 | 114 | 88.082% (85.725–89.040) | 87.874% (85.445–88.812) | 100.000% (98.473–100.000) | 95.263% (95.066–95.694) |
+| 60% | 262 / 436 | 8,585 | 114 | 87.325% (85.906–88.106) | 87.063% (85.626–87.887) | 99.422% (98.844–100.000) | 93.967% (93.376–94.187) |
+| 50% | 218 / 436 | 10,893 | 116 | 86.606% (85.490–86.964) | 86.354% (85.200–86.708) | 99.078% (98.157–99.541) | 92.332% (91.796–92.877) |
+| 40% | 174 / 436 | 13,032 | 128 | 85.228% (84.695–87.630) | 84.937% (84.391–87.384) | 99.615% (98.467–99.618) | 90.325% (89.867–92.020) |
+| 30% | 131 / 436 | 15,205 | 198 | 85.962% (84.967–86.223) | 85.695% (84.675–85.950) | 99.340% (97.690–99.672) | 89.432% (88.621–89.689) |
+| 20% | 87 / 436 | 17,403 | 106 | 82.158% (81.228–85.516) | 81.830% (80.868–85.228) | 99.425% (98.271–99.713) | 84.851% (84.572–88.351) |
+| 10% | 44 / 436 | 19,563 | 17 | 83.849% (83.082–86.007) | 83.534% (82.762–85.733) | 99.233% (98.462–99.745) | 85.140% (84.723–87.352) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Aymara</strong> — <code>aym-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 3,319 / 3,319 | 0 | 0 | n/a | n/a | n/a | 99.580% |
+| 90% | 2,987 / 3,319 | 30,890 | 268 | 90.939% (90.434–91.492) | 90.955% (90.456–91.509) | 89.557% (88.141–90.675) | 98.703% (98.567–98.735) |
+| 80% | 2,655 / 3,319 | 62,588 | 500 | 90.889% (90.131–91.661) | 90.916% (90.151–91.689) | 88.942% (88.133–89.748) | 97.765% (97.610–97.972) |
+| 70% | 2,323 / 3,319 | 92,944 | 571 | 90.699% (90.397–91.066) | 90.732% (90.415–91.094) | 88.279% (87.461–88.620) | 96.848% (96.722–96.934) |
+| 60% | 1,991 / 3,319 | 122,831 | 610 | 90.413% (90.137–91.038) | 90.448% (90.153–91.073) | 87.744% (86.776–88.629) | 95.866% (95.747–96.134) |
+| 50% | 1,660 / 3,319 | 153,956 | 656 | 90.316% (89.707–91.066) | 90.341% (89.726–91.109) | 86.994% (85.688–87.973) | 94.871% (94.570–95.293) |
+| 40% | 1,328 / 3,319 | 184,980 | 689 | 90.329% (89.458–90.748) | 90.383% (89.499–90.789) | 86.193% (85.134–87.203) | 93.932% (93.437–94.259) |
+| 30% | 996 / 3,319 | 215,696 | 511 | 90.244% (89.589–90.929) | 90.292% (89.642–90.982) | 85.983% (84.720–86.951) | 92.879% (92.582–93.498) |
+| 20% | 664 / 3,319 | 247,764 | 396 | 90.311% (89.061–90.407) | 90.364% (89.116–90.463) | 85.322% (83.935–86.779) | 92.083% (91.174–92.186) |
+| 10% | 332 / 3,319 | 278,289 | 219 | 89.745% (89.150–90.373) | 89.796% (89.196–90.418) | 85.017% (83.036–86.137) | 90.738% (90.186–91.195) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Azerbaijani</strong> — <code>az-az-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 340 / 340 | 0 | 0 | n/a | n/a | n/a | 99.970% |
+| 90% | 306 / 340 | 683 | 0 | 80.137% (72.506–83.980) | 79.741% (71.703–83.721) | 82.353% (81.818–88.235) | 97.796% (97.587–98.546) |
+| 80% | 272 / 340 | 1,436 | 0 | 79.817% (74.875–83.503) | 79.561% (74.425–83.274) | 85.294% (82.353–88.235) | 95.758% (95.353–96.327) |
+| 70% | 238 / 340 | 2,086 | 1 | 79.703% (74.703–80.968) | 79.426% (73.985–80.645) | 85.294% (82.353–87.255) | 93.404% (92.490–94.019) |
+| 60% | 204 / 340 | 2,737 | 1 | 77.766% (75.462–80.097) | 77.422% (74.834–79.622) | 86.567% (83.088–88.971) | 90.736% (90.571–91.980) |
+| 50% | 170 / 340 | 3,370 | 1 | 76.795% (73.211–78.911) | 76.406% (72.477–78.466) | 87.059% (84.118–87.647) | 88.248% (86.164–89.582) |
+| 40% | 136 / 340 | 4,078 | 1 | 74.129% (70.682–77.229) | 73.696% (69.828–76.734) | 87.129% (82.353–91.176) | 84.185% (81.982–86.359) |
+| 30% | 102 / 340 | 4,643 | 1 | 74.275% (70.993–74.646) | 73.472% (69.996–74.134) | 90.254% (84.034–90.756) | 82.162% (79.134–82.536) |
+| 20% | 68 / 340 | 5,282 | 1 | 71.379% (67.200–72.903) | 70.619% (65.970–71.929) | 86.029% (85.294–90.809) | 76.555% (74.202–78.579) |
+| 10% | 34 / 340 | 5,924 | 0 | 67.201% (66.164–70.809) | 65.931% (64.925–69.903) | 88.562% (86.601–91.830) | 70.874% (70.394–73.857) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Bashkir</strong> — <code>bak-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 1,059 / 1,059 | 0 | 0 | n/a | n/a | n/a | 99.823% |
+| 90% | 953 / 1,059 | 1,077 | 1 | 95.079% (94.013–95.810) | 95.060% (93.880–96.281) | 95.283% (91.509–95.283) | 99.330% (99.209–99.460) |
+| 80% | 847 / 1,059 | 2,157 | 7 | 95.178% (94.265–96.219) | 95.123% (94.204–96.272) | 95.283% (91.509–95.714) | 98.855% (98.706–99.050) |
+| 70% | 741 / 1,059 | 3,263 | 10 | 95.221% (94.637–96.636) | 95.152% (94.565–96.681) | 95.283% (92.722–96.215) | 98.352% (98.250–98.799) |
+| 60% | 635 / 1,059 | 4,295 | 14 | 95.074% (94.656–96.497) | 95.082% (94.615–96.554) | 95.035% (93.128–95.972) | 97.859% (97.701–98.436) |
+| 50% | 530 / 1,059 | 5,393 | 8 | 95.179% (93.232–96.115) | 95.307% (93.053–96.272) | 94.697% (93.182–96.183) | 97.468% (96.444–97.980) |
+| 40% | 424 / 1,059 | 6,475 | 8 | 95.232% (94.707–95.814) | 95.196% (94.646–95.984) | 95.118% (93.533–95.563) | 97.002% (96.760–97.440) |
+| 30% | 318 / 1,059 | 7,506 | 3 | 94.745% (93.808–96.363) | 94.850% (93.898–96.497) | 93.919% (92.973–95.135) | 96.258% (95.606–97.431) |
+| 20% | 212 / 1,059 | 8,600 | 2 | 94.695% (93.731–95.450) | 94.976% (93.679–95.559) | 94.194% (92.090–94.444) | 95.708% (94.992–96.323) |
+| 10% | 106 / 1,059 | 9,669 | 1 | 94.564% (92.675–95.406) | 94.807% (92.643–95.764) | 92.130% (91.396–92.962) | 95.085% (93.418–95.848) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Belarusian</strong> — <code>be-by-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 1,020 / 1,020 | 0 | 0 | n/a | n/a | n/a | 99.248% |
+| 90% | 918 / 1,020 | 1,951 | 15 | 47.307% (46.233–48.319) | 44.700% (43.591–45.827) | 94.000% (89.216–96.000) | 93.977% (93.916–94.170) |
+| 80% | 816 / 1,020 | 3,857 | 29 | 45.673% (43.494–49.619) | 42.969% (40.655–47.024) | 93.564% (89.055–97.525) | 88.808% (88.224–89.189) |
+| 70% | 714 / 1,020 | 5,851 | 69 | 46.233% (43.360–48.096) | 43.642% (40.653–45.407) | 92.763% (89.735–97.360) | 83.405% (82.440–83.842) |
+| 60% | 612 / 1,020 | 7,832 | 77 | 44.005% (43.644–47.319) | 41.432% (40.979–44.796) | 92.611% (90.842–94.321) | 77.149% (76.900–78.424) |
+| 50% | 510 / 1,020 | 9,726 | 96 | 42.937% (40.520–44.971) | 40.215% (37.605–42.359) | 93.124% (92.141–94.280) | 71.349% (69.779–72.183) |
+| 40% | 408 / 1,020 | 11,732 | 96 | 40.639% (38.066–42.009) | 37.807% (34.990–39.373) | 92.799% (90.296–95.074) | 64.025% (62.684–65.027) |
+| 30% | 306 / 1,020 | 13,708 | 63 | 37.083% (36.525–37.543) | 34.021% (33.443–34.603) | 92.847% (91.268–94.803) | 55.578% (55.375–56.066) |
+| 20% | 204 / 1,020 | 15,681 | 49 | 32.223% (31.519–33.429) | 29.004% (28.180–30.250) | 91.656% (90.909–92.638) | 45.586% (45.057–46.521) |
+| 10% | 102 / 1,020 | 17,677 | 28 | 25.195% (22.885–26.106) | 21.488% (19.003–22.659) | 92.810% (89.107–94.111) | 32.778% (30.287–33.504) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Bengali</strong> — <code>bn-bd-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 111 / 111 | 0 | 0 | n/a | n/a | n/a | 99.720% |
+| 90% | 100 / 111 | 247 | 0 | 58.039% (36.025–78.261) | 56.557% (33.762–77.333) | 90.909% (90.909–100.000) | 95.972% (92.504–98.494) |
+| 80% | 89 / 111 | 534 | 5 | 48.339% (35.768–60.263) | 46.538% (33.203–58.101) | 95.455% (90.909–100.000) | 89.912% (85.884–94.431) |
+| 70% | 78 / 111 | 730 | 5 | 41.404% (37.500–56.516) | 38.647% (34.820–54.902) | 96.970% (93.939–100.000) | 85.394% (78.599–88.231) |
+| 60% | 67 / 111 | 1,049 | 5 | 50.334% (45.455–57.261) | 48.259% (43.157–55.651) | 97.727% (95.455–100.000) | 81.471% (75.061–83.433) |
+| 50% | 56 / 111 | 1,300 | 8 | 51.530% (34.463–54.996) | 49.452% (32.185–53.043) | 96.364% (96.364–98.182) | 78.459% (63.152–79.685) |
+| 40% | 44 / 111 | 1,611 | 8 | 44.726% (22.373–51.928) | 42.462% (19.623–49.756) | 98.507% (97.015–98.507) | 68.476% (47.461–74.396) |
+| 30% | 33 / 111 | 1,961 | 8 | 38.869% (25.565–49.720) | 36.416% (22.716–47.485) | 98.718% (97.436–98.718) | 57.548% (45.464–68.231) |
+| 20% | 22 / 111 | 2,249 | 5 | 35.403% (27.069–42.139) | 32.905% (24.441–39.865) | 96.629% (92.135–100.000) | 51.979% (39.229–53.485) |
+| 10% | 11 / 111 | 2,592 | 0 | 33.924% (25.077–39.100) | 31.435% (22.472–36.825) | 96.000% (87.000–100.000) | 39.755% (31.699–44.518) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Bininj Kun-wok</strong> — <code>gup-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 73 / 73 | 0 | 0 | n/a | n/a | n/a | 99.730% |
+| 90% | 66 / 73 | 31 | 0 | 22.581% (13.208–44.444) | 0.000% (0.000–9.091) | 100.000% (85.714–100.000) | 93.261% (87.332–97.035) |
+| 80% | 58 / 73 | 83 | 0 | 18.421% (14.286–24.590) | 0.000% (0.000–1.639) | 100.000% (86.667–100.000) | 81.671% (77.089–87.332) |
+| 70% | 51 / 73 | 127 | 0 | 17.969% (14.894–19.444) | 0.000% (0.000–1.163) | 95.455% (90.909–100.000) | 71.429% (67.385–76.280) |
+| 60% | 44 / 73 | 162 | 0 | 17.442% (17.262–19.580) | 0.833% (0.000–1.754) | 96.552% (89.655–96.552) | 63.612% (61.456–68.733) |
+| 50% | 37 / 73 | 192 | 0 | 18.325% (16.908–18.919) | 0.585% (0.000–1.342) | 97.222% (91.667–97.222) | 57.682% (53.369–59.299) |
+| 40% | 29 / 73 | 231 | 1 | 19.556% (17.928–19.913) | 0.552% (0.483–1.604) | 97.727% (95.455–100.000) | 49.865% (44.205–50.943) |
+| 30% | 22 / 73 | 269 | 1 | 19.847% (17.568–21.687) | 0.847% (0.408–2.525) | 100.000% (96.078–100.000) | 42.049% (33.962–47.170) |
+| 20% | 15 / 73 | 303 | 1 | 19.865% (18.692–21.107) | 0.797% (0.418–2.165) | 100.000% (96.552–100.000) | 34.771% (29.380–38.275) |
+| 10% | 7 / 73 | 333 | 0 | 20.000% (19.130–21.530) | 0.000% (0.000–3.833) | 100.000% (98.485–100.000) | 28.032% (24.798–29.380) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Braj</strong> — <code>bra-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 1,246 / 1,246 | 0 | 0 | n/a | n/a | n/a | 99.656% |
+| 90% | 1,121 / 1,246 | 197 | 1 | 65.641% (59.494–73.171) | 16.964% (7.500–18.947) | 94.355% (92.800–97.600) | 96.364% (94.939–97.494) |
+| 80% | 997 / 1,246 | 398 | 2 | 64.500% (58.730–68.320) | 15.328% (9.896–16.556) | 93.574% (91.935–96.386) | 92.727% (90.713–94.005) |
+| 70% | 872 / 1,246 | 633 | 3 | 61.538% (60.436–63.714) | 13.060% (11.583–16.667) | 94.385% (91.935–95.989) | 87.862% (87.273–89.189) |
+| 60% | 748 / 1,246 | 842 | 2 | 61.447% (60.570–64.554) | 12.319% (11.628–18.611) | 93.763% (92.540–95.582) | 83.489% (83.440–86.192) |
+| 50% | 623 / 1,246 | 1,031 | 2 | 61.463% (60.674–65.661) | 12.010% (10.918–16.972) | 93.730% (92.765–95.666) | 80.197% (79.214–83.440) |
+| 40% | 498 / 1,246 | 1,220 | 4 | 62.479% (60.841–64.370) | 12.288% (10.474–14.110) | 94.779% (91.432–95.588) | 77.641% (75.971–78.821) |
+| 30% | 374 / 1,246 | 1,426 | 3 | 62.412% (61.544–64.414) | 11.372% (9.009–14.359) | 95.183% (94.839–95.757) | 73.415% (72.875–75.332) |
+| 20% | 249 / 1,246 | 1,614 | 3 | 63.139% (60.509–65.045) | 11.437% (5.352–12.898) | 95.683% (94.679–96.690) | 70.074% (67.912–72.826) |
+| 10% | 125 / 1,246 | 1,846 | 1 | 61.038% (59.839–62.638) | 6.586% (5.660–14.141) | 95.183% (90.812–97.324) | 64.963% (63.145–66.634) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Breton</strong> — <code>bre-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 44 / 44 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 90% | 40 / 44 | 174 | 0 | 13.793% (2.963–26.286) | 11.765% (0.000–24.561) | 100.000% | 92.268% (92.216–93.351) |
+| 80% | 35 / 44 | 394 | 0 | 18.020% (2.247–28.571) | 16.104% (0.000–27.154) | 88.889% (88.889–100.000) | 83.351% (79.897–85.567) |
+| 70% | 31 / 44 | 575 | 0 | 17.419% (8.958–23.986) | 15.815% (6.988–22.383) | 92.308% (92.308–100.000) | 75.258% (71.186–77.784) |
+| 60% | 26 / 44 | 829 | 0 | 11.947% (7.212–17.884) | 10.256% (5.160–16.108) | 94.444% (88.889–100.000) | 62.010% (60.206–66.392) |
+| 50% | 22 / 44 | 1,003 | 0 | 10.267% (6.250–19.298) | 8.476% (4.260–17.529) | 90.909% (86.364–95.455) | 54.330% (51.289–59.691) |
+| 40% | 18 / 44 | 1,144 | 0 | 6.854% (5.748–20.367) | 5.036% (3.630–18.605) | 92.308% (84.615–100.000) | 45.361% (42.526–53.041) |
+| 30% | 13 / 44 | 1,361 | 0 | 10.914% (4.864–17.414) | 9.130% (3.156–15.639) | 93.548% (74.194–100.000) | 35.206% (33.454–42.062) |
+| 20% | 9 / 44 | 1,546 | 0 | 7.633% (7.246–18.499) | 5.745% (5.412–16.744) | 94.286% (80.000–97.143) | 26.907% (24.124–35.052) |
+| 10% | 4 / 44 | 1,765 | 0 | 7.025% (4.703–11.841) | 4.986% (2.493–9.855) | 97.500% (95.000–100.000) | 15.412% (13.299–19.794) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Bulgarian</strong> — <code>bg-bg-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 2,413 / 2,413 | 0 | 0 | n/a | n/a | n/a | 99.732% |
+| 90% | 2,172 / 2,413 | 4,485 | 21 | 62.842% (55.624–66.042) | 61.341% (53.532–64.641) | 91.489% (90.833–94.142) | 96.096% (95.480–96.416) |
+| 80% | 1,930 / 2,413 | 9,005 | 41 | 61.838% (60.104–62.754) | 60.148% (58.375–61.028) | 92.034% (91.476–93.347) | 92.260% (91.737–92.445) |
+| 70% | 1,689 / 2,413 | 13,589 | 51 | 62.033% (59.657–64.736) | 60.374% (57.731–63.175) | 92.190% (90.972–93.084) | 88.232% (88.116–89.318) |
+| 60% | 1,448 / 2,413 | 18,284 | 67 | 60.960% (60.091–63.742) | 59.094% (58.404–62.132) | 91.946% (90.084–93.965) | 84.528% (83.873–85.432) |
+| 50% | 1,207 / 2,413 | 22,788 | 62 | 60.919% (58.660–62.113) | 59.080% (56.877–60.518) | 91.319% (90.341–93.917) | 80.289% (79.501–80.705) |
+| 40% | 965 / 2,413 | 27,478 | 56 | 60.884% (59.487–61.497) | 59.175% (57.747–59.793) | 91.023% (90.235–92.169) | 76.396% (75.952–76.866) |
+| 30% | 724 / 2,413 | 31,900 | 48 | 59.890% (57.711–60.748) | 58.132% (55.892–59.057) | 90.309% (89.667–91.400) | 72.011% (70.742–72.403) |
+| 20% | 483 / 2,413 | 36,589 | 34 | 57.780% (54.358–58.649) | 55.927% (52.405–56.854) | 90.177% (89.195–91.178) | 66.193% (63.561–67.149) |
+| 10% | 241 / 2,413 | 41,101 | 20 | 52.580% (50.622–54.376) | 50.586% (48.563–52.412) | 89.590% (87.471–90.355) | 57.661% (55.856–59.126) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Catalan</strong> — <code>ca-es-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 15,176 / 15,176 | 0 | 0 | n/a | n/a | n/a | 95.955% |
+| 90% | 13,658 / 15,176 | 13,444 | 981 | 80.281% (77.943–81.908) | 78.945% (76.244–80.465) | 92.780% (91.913–93.682) | 94.061% (93.820–94.320) |
+| 80% | 12,141 / 15,176 | 26,286 | 1,724 | 79.707% (78.367–81.535) | 78.102% (76.515–80.354) | 92.462% (91.957–92.760) | 92.202% (92.111–92.358) |
+| 70% | 10,623 / 15,176 | 38,922 | 2,238 | 79.078% (78.033–80.075) | 77.418% (76.351–78.715) | 91.822% (91.232–92.151) | 90.342% (90.158–90.551) |
+| 60% | 9,106 / 15,176 | 51,320 | 2,797 | 78.592% (78.180–79.596) | 77.046% (76.614–78.158) | 91.324% (90.954–91.578) | 88.564% (88.039–88.906) |
+| 50% | 7,588 / 15,176 | 65,160 | 2,723 | 78.596% (77.366–78.924) | 77.137% (75.791–77.459) | 90.746% (90.203–90.998) | 86.576% (85.991–86.916) |
+| 40% | 6,070 / 15,176 | 78,802 | 2,632 | 77.563% (76.767–78.089) | 75.963% (75.151–76.617) | 89.934% (89.512–90.348) | 84.453% (83.976–84.707) |
+| 30% | 4,553 / 15,176 | 92,174 | 2,318 | 76.912% (75.581–77.281) | 75.415% (73.827–75.819) | 89.040% (88.848–89.559) | 82.231% (81.131–82.607) |
+| 20% | 3,035 / 15,176 | 106,349 | 1,781 | 75.570% (74.288–77.080) | 73.997% (72.516–75.654) | 88.558% (88.098–88.762) | 79.258% (78.195–80.619) |
+| 10% | 1,518 / 15,176 | 121,030 | 1,030 | 73.460% (72.114–76.283) | 71.772% (70.147–74.931) | 87.789% (86.869–88.353) | 75.544% (74.342–78.068) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Cebuano</strong> — <code>ceb-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 95 / 95 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 90% | 86 / 95 | 39 | 0 | 23.077% (20.930–24.324) | 0.000% | 100.000% | 93.023% (92.093–93.488) |
+| 80% | 76 / 95 | 85 | 0 | 22.353% (21.111–23.171) | 0.000% | 100.000% | 84.651% (83.488–85.349) |
+| 70% | 67 / 95 | 127 | 0 | 22.047% (21.875–22.400) | 0.000% | 100.000% | 76.977% (76.744–77.442) |
+| 60% | 57 / 95 | 172 | 0 | 22.093% (21.469–22.093) | 0.000% | 100.000% | 68.837% (67.674–68.837) |
+| 50% | 48 / 95 | 212 | 0 | 22.170% (21.461–22.816) | 0.000% | 100.000% | 61.628% (60.000–63.023) |
+| 40% | 38 / 95 | 258 | 0 | 22.093% (21.839–22.441) | 0.000% | 100.000% | 53.256% (52.558–54.186) |
+| 30% | 29 / 95 | 300 | 0 | 22.000% (21.523–22.297) | 0.000% | 100.000% (98.485–100.000) | 45.581% (44.419–46.512) |
+| 20% | 19 / 95 | 344 | 0 | 22.093% (21.714–22.485) | 0.000% | 100.000% | 37.674% (36.279–39.070) |
+| 10% | 10 / 95 | 384 | 0 | 21.990% (21.851–22.310) | 0.000% | 100.000% (98.824–100.000) | 30.465% (29.302–31.163) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Chichewa</strong> — <code>ny-mw-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 227 / 227 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 90% | 204 / 227 | 322 | 0 | 7.143% (6.832–7.143) | 0.000% | 100.000% (95.652–100.000) | 90.592% (90.560–90.592) |
+| 80% | 182 / 227 | 630 | 0 | 7.143% (6.984–7.143) | 0.000% | 100.000% (97.778–100.000) | 81.592% (81.561–81.592) |
+| 70% | 159 / 227 | 952 | 0 | 7.143% (6.933–7.143) | 0.000% | 100.000% (97.059–100.000) | 72.184% (72.121–72.184) |
+| 60% | 136 / 227 | 1,274 | 0 | 7.143% (6.986–7.143) | 0.000% | 100.000% (97.802–100.000) | 62.775% (62.712–62.775) |
+| 50% | 114 / 227 | 1,582 | 0 | 7.143% (7.080–7.143) | 0.000% | 100.000% (99.115–100.000) | 53.776% (53.744–53.776) |
+| 40% | 91 / 227 | 1,904 | 0 | 7.143% (7.090–7.143) | 0.000% | 100.000% (99.265–100.000) | 44.368% (44.336–44.368) |
+| 30% | 68 / 227 | 2,226 | 0 | 7.143% (7.098–7.143) | 0.000% | 100.000% (99.371–100.000) | 34.959% (34.928–34.959) |
+| 20% | 45 / 227 | 2,548 | 0 | 7.143% (7.104–7.143) | 0.000% | 100.000% (99.451–100.000) | 25.551% (25.519–25.551) |
+| 10% | 23 / 227 | 2,856 | 0 | 7.108% (7.108–7.143) | 0.000% | 99.510% (99.510–100.000) | 16.520% (16.520–16.551) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Chukchi</strong> — <code>ckt-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 191 / 191 | 0 | 0 | n/a | n/a | n/a | 99.342% |
+| 90% | 172 / 191 | 27 | 0 | 69.231% (63.333–76.923) | 8.333% (0.000–14.286) | 100.000% (94.737–100.000) | 96.711% (95.724–97.368) |
+| 80% | 153 / 191 | 60 | 1 | 63.333% (57.812–73.077) | 4.545% (0.000–14.286) | 97.368% (91.667–100.000) | 92.105% (90.461–94.737) |
+| 70% | 134 / 191 | 89 | 1 | 61.798% (58.163–70.732) | 6.061% (3.125–8.000) | 98.214% (94.545–100.000) | 88.158% (85.855–91.447) |
+| 60% | 115 / 191 | 118 | 1 | 61.864% (56.693–69.091) | 2.564% (1.887–5.882) | 96.000% (94.737–100.000) | 84.539% (81.250–88.158) |
+| 50% | 96 / 191 | 152 | 1 | 60.127% (58.553–69.565) | 3.125% (1.724–6.977) | 94.737% (93.617–98.936) | 78.947% (78.618–85.526) |
+| 40% | 76 / 191 | 188 | 1 | 59.574% (57.979–64.943) | 2.703% (1.351–6.667) | 95.614% (94.690–96.522) | 74.342% (73.355–79.276) |
+| 30% | 57 / 191 | 215 | 1 | 60.930% (58.636–65.854) | 2.817% (1.149–9.722) | 96.269% (94.737–97.744) | 72.039% (69.737–76.316) |
+| 20% | 38 / 191 | 248 | 1 | 61.290% (59.036–62.810) | 2.174% (1.020–6.667) | 96.053% (95.425–98.039) | 68.421% (66.118–70.066) |
+| 10% | 19 / 191 | 275 | 0 | 61.011% (60.215–63.869) | 2.830% (0.000–10.784) | 97.076% (94.767–97.674) | 64.803% (63.487–67.434) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Church Slavonic</strong> — <code>chu-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 152 / 152 | 0 | 0 | n/a | n/a | n/a | 81.379% |
+| 90% | 137 / 152 | 132 | 72 | 44.203% (15.000–58.333) | 42.857% (11.628–56.667) | 70.000% (54.545–77.778) | 78.128% (76.404–78.818) |
+| 80% | 122 / 152 | 266 | 144 | 45.161% (27.444–50.644) | 42.920% (23.457–47.222) | 69.565% (66.667–85.714) | 74.433% (72.463–75.714) |
+| 70% | 106 / 152 | 444 | 162 | 41.892% (37.721–44.632) | 39.386% (33.547–42.202) | 71.795% (59.459–85.366) | 70.443% (68.424–71.084) |
+| 60% | 91 / 152 | 659 | 162 | 41.166% (31.241–43.582) | 39.458% (27.480–40.933) | 74.074% (59.184–81.481) | 65.862% (62.266–67.389) |
+| 50% | 76 / 152 | 840 | 198 | 39.930% (33.736–43.750) | 36.849% (29.797–40.587) | 79.104% (65.079–83.582) | 62.315% (57.882–63.202) |
+| 40% | 61 / 152 | 1,075 | 162 | 38.605% (33.117–41.198) | 35.484% (29.618–38.323) | 75.904% (67.500–79.762) | 57.586% (52.857–58.818) |
+| 30% | 46 / 152 | 1,247 | 162 | 34.612% (27.987–45.389) | 31.278% (23.565–42.472) | 79.592% (66.327–86.735) | 49.852% (46.010–56.700) |
+| 20% | 30 / 152 | 1,541 | 108 | 33.899% (27.038–40.013) | 29.794% (22.222–38.257) | 81.897% (61.864–88.136) | 44.089% (39.557–49.557) |
+| 10% | 15 / 152 | 1,756 | 72 | 26.640% (23.927–40.949) | 21.680% (19.071–38.760) | 83.333% (68.148–86.567) | 33.103% (30.099–45.517) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Classical Armenian</strong> — <code>xcl-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 4,300 / 4,300 | 0 | 0 | n/a | n/a | n/a | 94.362% |
+| 90% | 3,870 / 4,300 | 5,676 | 530 | 58.951% (56.973–64.152) | 56.614% (54.823–61.939) | 91.317% (90.000–95.143) | 90.831% (90.251–91.436) |
+| 80% | 3,440 / 4,300 | 11,645 | 1,090 | 60.031% (57.825–62.927) | 57.967% (55.623–60.761) | 91.565% (91.144–93.986) | 87.255% (86.579–88.048) |
+| 70% | 3,010 / 4,300 | 17,252 | 1,495 | 58.666% (57.184–59.272) | 56.257% (54.763–56.993) | 91.584% (90.761–93.926) | 83.373% (83.282–83.626) |
+| 60% | 2,580 / 4,300 | 23,388 | 1,703 | 58.873% (57.300–59.035) | 56.638% (54.842–56.787) | 91.339% (89.882–92.536) | 79.647% (79.279–80.012) |
+| 50% | 2,150 / 4,300 | 29,673 | 1,800 | 57.626% (56.787–58.976) | 55.152% (54.461–56.709) | 91.129% (89.892–91.269) | 75.670% (74.975–76.471) |
+| 40% | 1,720 / 4,300 | 35,893 | 1,775 | 56.999% (56.075–57.373) | 54.704% (53.600–55.058) | 89.458% (88.725–90.542) | 71.695% (71.185–71.769) |
+| 30% | 1,290 / 4,300 | 42,711 | 1,436 | 56.782% (53.073–58.169) | 54.478% (50.429–56.063) | 89.114% (87.743–90.095) | 67.501% (65.194–68.717) |
+| 20% | 860 / 4,300 | 49,304 | 1,141 | 55.229% (54.188–56.999) | 52.893% (51.734–54.752) | 87.912% (86.991–88.376) | 62.877% (61.992–64.335) |
+| 10% | 430 / 4,300 | 55,842 | 592 | 54.388% (52.421–55.202) | 52.101% (50.001–52.854) | 85.680% (85.348–88.186) | 58.187% (56.640–59.114) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Classical Syriac</strong> — <code>syc-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 3,257 / 3,257 | 0 | 0 | n/a | n/a | n/a | 99.372% |
+| 90% | 2,931 / 3,257 | 2,706 | 32 | 18.071% (16.174–19.930) | 6.970% (6.250–8.713) | 98.452% (97.840–98.754) | 91.402% (90.182–92.319) |
+| 80% | 2,606 / 3,257 | 5,632 | 58 | 16.903% (16.178–18.162) | 6.553% (5.979–7.454) | 97.991% (97.531–98.302) | 82.482% (81.645–83.341) |
+| 70% | 2,280 / 3,257 | 8,401 | 76 | 17.274% (17.188–17.984) | 7.136% (6.636–7.313) | 97.844% (97.216–98.251) | 74.328% (73.353–75.411) |
+| 60% | 1,954 / 3,257 | 11,223 | 87 | 17.250% (16.964–17.541) | 6.700% (6.341–7.340) | 97.769% (96.993–98.228) | 65.950% (64.715–67.191) |
+| 50% | 1,629 / 3,257 | 13,893 | 84 | 17.282% (16.724–17.405) | 6.568% (6.212–6.930) | 97.901% (97.102–98.149) | 57.936% (56.893–59.604) |
+| 40% | 1,303 / 3,257 | 16,637 | 92 | 17.124% (16.990–17.465) | 6.498% (6.049–6.680) | 98.048% (97.228–98.821) | 49.782% (49.042–50.854) |
+| 30% | 977 / 3,257 | 19,247 | 79 | 17.482% (16.726–17.630) | 6.526% (6.381–7.017) | 97.889% (97.451–98.067) | 42.078% (39.147–43.176) |
+| 20% | 651 / 3,257 | 22,088 | 66 | 16.818% (16.312–17.279) | 6.306% (5.404–6.467) | 97.964% (96.269–98.692) | 33.560% (32.191–34.610) |
+| 10% | 326 / 3,257 | 24,625 | 37 | 16.381% (16.169–17.052) | 5.476% (5.101–6.191) | 97.439% (96.689–99.215) | 25.474% (23.546–26.257) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Congo Swahili</strong> — <code>swc-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 183 / 183 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 90% | 165 / 183 | 1,073 | 0 | 1.654% (1.573–1.798) | 0.000% | 100.000% (88.889–100.000) | 90.513% (89.992–91.161) |
+| 80% | 146 / 183 | 2,337 | 0 | 1.540% (1.443–1.713) | 0.000% | 97.297% (91.892–100.000) | 79.309% (77.889–82.457) |
+| 70% | 128 / 183 | 3,498 | 0 | 1.515% (1.483–1.570) | 0.000% | 96.364% (89.091–100.000) | 69.023% (68.303–70.740) |
+| 60% | 110 / 183 | 4,614 | 0 | 1.539% (1.472–1.659) | 0.000% | 97.260% (90.411–100.000) | 59.149% (58.862–61.631) |
+| 50% | 92 / 183 | 5,676 | 0 | 1.515% (1.371–1.565) | 0.000% | 92.308% (91.209–100.000) | 49.735% (45.661–52.477) |
+| 40% | 73 / 183 | 6,869 | 0 | 1.511% (1.482–1.548) | 0.000% | 95.455% (90.909–98.182) | 39.178% (36.696–41.660) |
+| 30% | 55 / 183 | 7,941 | 0 | 1.549% (1.516–1.575) | 0.000% | 96.094% (92.188–100.000) | 29.701% (28.082–33.378) |
+| 20% | 37 / 183 | 8,959 | 0 | 1.530% (1.457–1.608) | 0.000% | 93.836% (91.781–100.000) | 20.673% (16.689–22.687) |
+| 10% | 18 / 183 | 9,886 | 0 | 1.502% (1.406–1.671) | 0.000% | 89.091% (88.485–100.000) | 12.427% (7.302–13.929) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Copala Triqui</strong> — <code>cpa-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 555 / 555 | 0 | 0 | n/a | n/a | n/a | 88.706% |
+| 90% | 500 / 555 | 259 | 58 | 24.906% (21.933–26.606) | 5.430% (3.158–9.390) | 98.000% (97.917–100.000) | 82.621% (81.974–83.657) |
+| 80% | 444 / 555 | 507 | 99 | 25.290% (25.000–26.411) | 7.226% (6.952–9.250) | 97.959% (97.030–100.000) | 76.893% (76.149–77.896) |
+| 70% | 389 / 555 | 777 | 140 | 25.225% (24.399–25.732) | 7.414% (6.864–8.504) | 98.693% (96.026–99.333) | 71.036% (70.129–71.197) |
+| 60% | 333 / 555 | 1,047 | 176 | 25.766% (24.045–26.249) | 8.747% (6.120–9.187) | 98.500% (97.500–99.034) | 64.887% (63.657–65.469) |
+| 50% | 278 / 555 | 1,343 | 190 | 24.799% (23.973–26.585) | 8.594% (6.117–9.212) | 98.077% (97.200–99.219) | 58.220% (57.379–59.482) |
+| 40% | 222 / 555 | 1,656 | 190 | 24.494% (23.913–26.145) | 7.988% (6.543–9.201) | 98.089% (96.743–99.035) | 51.618% (51.068–52.621) |
+| 30% | 167 / 555 | 1,993 | 155 | 24.277% (23.691–24.887) | 7.396% (6.785–8.292) | 98.374% (97.581–98.919) | 44.757% (44.207–45.761) |
+| 20% | 111 / 555 | 2,315 | 135 | 24.297% (22.637–24.622) | 7.732% (5.851–8.413) | 97.163% (96.706–98.595) | 37.961% (37.314–38.641) |
+| 10% | 56 / 555 | 2,709 | 74 | 23.946% (22.148–24.859) | 7.973% (5.415–8.541) | 97.732% (95.910–99.588) | 31.100% (29.256–32.395) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Cornish</strong> — <code>cor-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 8 / 8 | 0 | 0 | n/a | n/a | n/a | 93.103% |
+| 90% | 7 / 8 | 11 | 0 | 9.091% (9.091–100.000) | 0.000% | 100.000% | 87.356% (87.356–93.103) |
+| 80% | 6 / 8 | 21 | 0 | 12.766% (4.167–18.182) | 0.000% (0.000–8.889) | 100.000% | 82.184% (66.667–87.931) |
+| 70% | 6 / 8 | 21 | 0 | 12.766% (4.167–18.182) | 0.000% (0.000–8.889) | 100.000% | 82.184% (66.667–87.931) |
+| 60% | 5 / 8 | 48 | 12 | 6.250% (2.703–14.583) | 0.000% (0.000–8.889) | 100.000% (33.333–100.000) | 69.540% (66.092–82.184) |
+| 50% | 4 / 8 | 73 | 12 | 6.154% (3.529–8.219) | 0.000% (0.000–5.797) | 75.000% (50.000–100.000) | 58.046% (45.977–61.494) |
+| 40% | 3 / 8 | 110 | 12 | 5.405% (3.922–8.264) | 0.943% (0.000–5.172) | 80.000% (80.000–100.000) | 39.655% (29.310–41.954) |
+| 30% | 2 / 8 | 132 | 12 | 7.483% (5.357–8.850) | 4.255% (0.943–4.673) | 83.333% | 22.989% (21.839–33.908) |
+| 20% | 2 / 8 | 132 | 12 | 7.483% (5.357–8.850) | 4.255% (0.943–4.673) | 83.333% | 22.989% (21.839–33.908) |
+| 10% | 1 / 8 | 157 | 0 | 7.006% (2.685–8.696) | 3.333% (0.000–3.817) | 85.714% (57.143–100.000) | 16.092% (9.770–27.586) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Cree</strong> — <code>cre-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 32 / 32 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 90% | 29 / 32 | 144 | 0 | 93.056% (77.381–97.059) | 94.118% (76.543–97.512) | 66.667% (33.333–100.000) | 99.071% (97.491–99.442) |
+| 80% | 26 / 32 | 283 | 0 | 83.772% (52.478–94.048) | 83.784% (52.522–95.062) | 83.333% (50.000–100.000) | 96.561% (84.851–99.071) |
+| 70% | 22 / 32 | 375 | 0 | 89.067% (53.061–91.538) | 89.041% (52.708–92.400) | 80.000% (70.000–100.000) | 96.004% (78.625–97.955) |
+| 60% | 19 / 32 | 459 | 0 | 79.303% (55.253–93.510) | 78.700% (54.691–94.172) | 84.615% (76.923–100.000) | 91.171% (78.625–97.955) |
+| 50% | 16 / 32 | 603 | 0 | 62.310% (38.473–81.590) | 61.838% (37.577–81.602) | 81.250% (75.000–100.000) | 76.952% (61.803–91.822) |
+| 40% | 13 / 32 | 682 | 0 | 66.499% (42.145–81.851) | 66.067% (41.379–81.768) | 84.211% (73.684–94.737) | 75.186% (56.877–90.520) |
+| 30% | 10 / 32 | 766 | 0 | 65.014% (42.978–76.780) | 64.035% (42.040–76.923) | 81.818% (72.727–95.455) | 74.628% (56.227–86.059) |
+| 20% | 6 / 32 | 913 | 0 | 41.521% (32.749–67.419) | 40.127% (30.891–67.098) | 92.308% (76.923–100.000) | 47.770% (42.937–75.836) |
+| 10% | 3 / 32 | 997 | 0 | 35.752% (8.460–41.230) | 34.581% (5.865–39.668) | 93.103% (72.414–100.000) | 43.401% (10.502–45.818) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Crimean Tatar</strong> — <code>crh-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 1,210 / 1,210 | 0 | 0 | n/a | n/a | n/a | 99.944% |
+| 90% | 1,089 / 1,210 | 717 | 1 | 95.556% (94.693–96.111) | 95.477% (93.792–96.500) | 95.041% (94.167–99.167) | 99.500% (99.417–99.555) |
+| 80% | 968 / 1,210 | 1,438 | 1 | 95.271% (94.367–96.181) | 94.987% (93.787–96.414) | 95.021% (92.562–98.340) | 99.000% (98.819–99.194) |
+| 70% | 847 / 1,210 | 2,159 | 2 | 95.370% (93.948–95.926) | 95.442% (93.169–96.105) | 95.028% (93.113–97.790) | 98.555% (98.139–98.736) |
+| 60% | 726 / 1,210 | 2,876 | 2 | 95.028% (94.286–95.859) | 95.029% (93.758–96.151) | 94.617% (93.375–96.894) | 97.972% (97.666–98.305) |
+| 50% | 605 / 1,210 | 3,593 | 2 | 95.243% (94.766–95.606) | 95.154% (94.779–95.822) | 94.702% (94.205–96.026) | 97.569% (97.347–97.764) |
+| 40% | 484 / 1,210 | 4,314 | 2 | 95.228% (94.576–95.435) | 95.210% (94.290–95.566) | 95.317% (94.069–95.994) | 97.111% (96.708–97.208) |
+| 30% | 363 / 1,210 | 5,036 | 2 | 95.092% (93.467–95.955) | 95.206% (93.009–96.282) | 95.035% (93.853–95.740) | 96.527% (95.388–97.152) |
+| 20% | 242 / 1,210 | 5,758 | 1 | 94.340% (92.480–95.084) | 94.864% (92.321–95.177) | 94.623% (91.736–95.346) | 95.458% (93.944–96.055) |
+| 10% | 121 / 1,210 | 6,479 | 1 | 94.366% (93.781–94.628) | 94.212% (93.990–94.730) | 94.123% (90.993–95.588) | 94.930% (94.388–95.166) |
+
+</div>
+</details>
+
 <details markdown="block">
 <summary><strong>Czech</strong> — <code>cs-cz-default</code> 1.0.0</summary>
 <div markdown="1">
 
 | Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
 | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
-| 100% | 5,113 / 5,113 | 0 | 0 | n/a | n/a | n/a | 99.465% |
-| 90% | 4,602 / 5,113 | 5,497 | 52 | 73.270% (71.970–75.506) | 69.900% (67.596–72.033) | 91.296% (89.606–92.449) | 96.872% (96.693–97.110) |
-| 80% | 4,090 / 5,113 | 11,295 | 108 | 74.375% (72.961–75.341) | 70.977% (68.757–71.820) | 91.822% (89.763–92.700) | 94.512% (94.070–94.620) |
-| 70% | 3,579 / 5,113 | 16,910 | 167 | 74.517% (73.186–75.890) | 70.679% (69.464–72.757) | 91.409% (90.430–92.421) | 91.850% (91.537–92.385) |
-| 60% | 3,068 / 5,113 | 22,567 | 171 | 74.064% (73.458–75.665) | 70.641% (69.532–72.243) | 91.201% (89.195–91.780) | 89.311% (89.002–89.831) |
-| 50% | 2,557 / 5,113 | 28,195 | 172 | 74.013% (73.648–74.299) | 70.612% (69.908–70.743) | 91.128% (89.182–91.481) | 86.632% (86.397–86.793) |
-| 40% | 2,045 / 5,113 | 33,944 | 151 | 74.037% (73.092–74.837) | 70.380% (69.385–71.348) | 90.721% (89.903–91.126) | 84.092% (83.521–84.645) |
-| 30% | 1,534 / 5,113 | 39,562 | 122 | 73.325% (71.387–74.104) | 69.913% (67.160–70.569) | 90.571% (89.035–90.982) | 81.062% (79.736–81.666) |
-| 20% | 1,023 / 5,113 | 45,111 | 109 | 72.003% (69.047–72.412) | 68.155% (64.204–68.726) | 89.767% (89.232–91.428) | 77.461% (75.136–77.787) |
-| 10% | 511 / 5,113 | 50,905 | 74 | 69.921% (66.235–70.712) | 65.439% (60.991–66.671) | 89.636% (88.686–90.665) | 72.859% (69.489–73.555) |
+| 100% | 5,113 / 5,113 | 0 | 0 | n/a | n/a | n/a | 99.465% |
+| 90% | 4,602 / 5,113 | 5,497 | 52 | 73.270% (71.970–75.506) | 69.900% (67.596–72.033) | 91.296% (89.606–92.449) | 96.872% (96.693–97.110) |
+| 80% | 4,090 / 5,113 | 11,295 | 108 | 74.375% (72.961–75.341) | 70.977% (68.757–71.820) | 91.822% (89.763–92.700) | 94.512% (94.070–94.620) |
+| 70% | 3,579 / 5,113 | 16,910 | 167 | 74.517% (73.186–75.890) | 70.679% (69.464–72.757) | 91.409% (90.430–92.421) | 91.850% (91.537–92.385) |
+| 60% | 3,068 / 5,113 | 22,567 | 171 | 74.064% (73.458–75.665) | 70.641% (69.532–72.243) | 91.201% (89.195–91.780) | 89.311% (89.002–89.831) |
+| 50% | 2,557 / 5,113 | 28,195 | 172 | 74.013% (73.648–74.299) | 70.612% (69.908–70.743) | 91.128% (89.182–91.481) | 86.632% (86.397–86.793) |
+| 40% | 2,045 / 5,113 | 33,944 | 151 | 74.037% (73.092–74.837) | 70.380% (69.385–71.348) | 90.721% (89.903–91.126) | 84.092% (83.521–84.645) |
+| 30% | 1,534 / 5,113 | 39,562 | 122 | 73.325% (71.387–74.104) | 69.913% (67.160–70.569) | 90.571% (89.035–90.982) | 81.062% (79.736–81.666) |
+| 20% | 1,023 / 5,113 | 45,111 | 109 | 72.003% (69.047–72.412) | 68.155% (64.204–68.726) | 89.767% (89.232–91.428) | 77.461% (75.136–77.787) |
+| 10% | 511 / 5,113 | 50,905 | 74 | 69.921% (66.235–70.712) | 65.439% (60.991–66.671) | 89.636% (88.686–90.665) | 72.859% (69.489–73.555) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Dakota</strong> — <code>dak-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 473 / 473 | 0 | 0 | n/a | n/a | n/a | 99.226% |
+| 90% | 426 / 473 | 326 | 10 | 24.375% (21.951–28.840) | 12.545% (9.609–17.216) | 95.745% (89.362–97.872) | 91.900% (91.602–92.466) |
+| 80% | 378 / 473 | 662 | 10 | 23.845% (23.688–26.300) | 12.153% (11.888–14.286) | 94.737% (94.681–98.947) | 84.336% (84.008–84.872) |
+| 70% | 331 / 473 | 992 | 13 | 24.093% (23.124–24.476) | 12.235% (11.834–12.922) | 95.070% (90.780–96.454) | 76.802% (76.712–76.891) |
+| 60% | 284 / 473 | 1,325 | 13 | 23.925% (21.278–24.851) | 12.148% (9.019–13.247) | 95.745% (92.553–96.296) | 69.267% (68.195–69.446) |
+| 50% | 237 / 473 | 1,657 | 15 | 22.269% (21.553–23.778) | 10.134% (9.397–12.197) | 94.894% (91.915–95.339) | 60.929% (60.572–61.525) |
+| 40% | 189 / 473 | 2,000 | 13 | 21.678% (20.435–24.248) | 9.366% (8.387–12.850) | 93.310% (92.580–96.466) | 52.680% (51.846–54.527) |
+| 30% | 142 / 473 | 2,333 | 14 | 21.046% (18.955–22.889) | 8.837% (6.920–10.984) | 93.636% (92.727–95.152) | 44.580% (42.942–46.010) |
+| 20% | 95 / 473 | 2,676 | 10 | 18.498% (17.779–19.933) | 5.962% (5.362–7.472) | 94.974% (92.042–96.021) | 34.693% (34.127–35.825) |
+| 10% | 47 / 473 | 3,021 | 4 | 19.185% (15.619–20.563) | 6.481% (2.696–7.977) | 94.836% (92.958–97.412) | 27.129% (24.032–28.350) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Danish</strong> — <code>da-dk-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 4,179 / 4,179 | 0 | 0 | n/a | n/a | n/a | 99.371% |
+| 90% | 3,761 / 4,179 | 3,232 | 38 | 74.192% (73.356–75.186) | 67.504% (66.790–69.750) | 92.718% (90.865–93.961) | 96.813% (96.698–96.897) |
+| 80% | 3,343 / 4,179 | 6,419 | 59 | 74.015% (73.293–74.244) | 67.477% (67.434–67.862) | 92.377% (90.132–93.494) | 94.212% (94.063–94.358) |
+| 70% | 2,925 / 4,179 | 9,622 | 72 | 72.901% (72.673–74.059) | 66.648% (65.699–67.713) | 91.066% (90.273–92.613) | 91.447% (91.276–91.726) |
+| 60% | 2,507 / 4,179 | 12,858 | 89 | 72.406% (72.124–72.935) | 65.740% (65.579–67.061) | 90.565% (89.988–91.933) | 88.498% (88.442–88.641) |
+| 50% | 2,090 / 4,179 | 16,078 | 86 | 71.433% (70.792–72.016) | 64.706% (63.871–65.934) | 90.561% (89.489–90.758) | 85.333% (85.042–85.612) |
+| 40% | 1,672 / 4,179 | 19,316 | 87 | 70.449% (69.207–70.919) | 63.495% (61.895–64.640) | 90.106% (88.956–90.560) | 81.882% (81.244–82.196) |
+| 30% | 1,254 / 4,179 | 22,543 | 70 | 68.316% (67.096–69.463) | 60.962% (59.342–62.755) | 89.335% (88.687–90.062) | 77.558% (76.727–78.444) |
+| 20% | 836 / 4,179 | 25,755 | 61 | 65.725% (64.691–66.441) | 57.555% (56.445–58.672) | 89.232% (88.279–90.082) | 72.436% (71.615–73.034) |
+| 10% | 418 / 4,179 | 29,022 | 28 | 61.104% (59.899–62.277) | 51.055% (49.995–53.218) | 88.238% (87.507–89.957) | 64.847% (63.836–65.997) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Dutch</strong> — <code>nl-nl-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 4,992 / 4,992 | 0 | 0 | n/a | n/a | n/a | 99.120% |
+| 90% | 4,493 / 4,992 | 3,085 | 57 | 74.468% (73.250–76.272) | 65.891% (63.190–67.730) | 94.512% (92.371–95.697) | 96.574% (96.437–96.777) |
+| 80% | 3,994 / 4,992 | 6,160 | 110 | 74.369% (74.091–75.340) | 64.865% (64.550–66.383) | 94.408% (92.901–95.806) | 94.083% (94.025–94.245) |
+| 70% | 3,494 / 4,992 | 9,298 | 148 | 74.137% (73.135–74.845) | 64.553% (63.215–66.338) | 94.412% (93.155–95.287) | 91.461% (91.190–91.734) |
+| 60% | 2,995 / 4,992 | 12,412 | 171 | 73.828% (73.295–73.872) | 64.368% (63.661–64.668) | 94.242% (93.471–95.070) | 88.817% (88.524–88.845) |
+| 50% | 2,496 / 4,992 | 15,581 | 174 | 73.408% (73.027–73.660) | 63.952% (63.070–64.202) | 94.186% (93.539–94.678) | 86.068% (85.791–86.153) |
+| 40% | 1,997 / 4,992 | 18,714 | 171 | 72.501% (71.803–73.110) | 62.655% (61.524–63.525) | 93.853% (93.540–94.029) | 82.969% (82.495–83.280) |
+| 30% | 1,498 / 4,992 | 21,856 | 141 | 71.097% (70.602–72.004) | 61.026% (59.937–61.731) | 93.565% (92.739–94.166) | 79.432% (79.003–79.969) |
+| 20% | 998 / 4,992 | 25,065 | 109 | 69.543% (68.997–70.471) | 58.908% (57.908–60.208) | 92.781% (92.430–93.088) | 75.447% (74.855–76.168) |
+| 10% | 499 / 4,992 | 28,267 | 62 | 67.039% (65.411–67.577) | 55.524% (53.812–55.891) | 92.616% (90.386–92.866) | 70.231% (68.760–70.692) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Eastern Chatino</strong> — <code>cly-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 185 / 185 | 0 | 0 | n/a | n/a | n/a | 90.404% |
+| 90% | 167 / 185 | 173 | 39 | 9.827% (7.692–11.047) | 1.911% (0.000–3.774) | 100.000% (87.500–100.000) | 83.474% (82.408–83.829) |
+| 80% | 148 / 185 | 381 | 70 | 9.351% (8.136–10.133) | 0.890% (0.000–2.041) | 96.875% (93.939–100.000) | 75.300% (74.767–75.700) |
+| 70% | 130 / 185 | 583 | 88 | 9.326% (8.233–9.982) | 0.752% (0.000–1.949) | 96.000% (95.833–100.000) | 67.437% (66.459–68.858) |
+| 60% | 111 / 185 | 797 | 105 | 9.206% (8.629–9.547) | 1.068% (0.417–1.872) | 95.588% (90.000–97.183) | 59.307% (58.463–59.973) |
+| 50% | 93 / 185 | 1,001 | 113 | 9.046% (8.732–9.127) | 0.849% (0.455–1.313) | 93.407% (90.805–96.386) | 52.155% (50.600–52.999) |
+| 40% | 74 / 185 | 1,226 | 120 | 8.795% (8.548–9.024) | 0.797% (0.453–1.070) | 94.340% (90.566–96.078) | 43.847% (43.003–44.114) |
+| 30% | 56 / 185 | 1,467 | 100 | 8.521% (8.328–8.725) | 0.689% (0.446–0.880) | 92.683% (91.129–96.748) | 35.318% (34.340–36.117) |
+| 20% | 37 / 185 | 1,710 | 71 | 8.568% (8.406–8.841) | 0.805% (0.640–0.831) | 93.836% (92.308–96.454) | 26.788% (24.967–27.232) |
+| 10% | 19 / 185 | 1,969 | 52 | 8.406% (8.278–8.829) | 0.769% (0.447–0.997) | 95.679% (88.415–98.160) | 17.725% (16.659–18.392) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Egyptian Arabic</strong> — <code>arz-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 6,004 / 6,004 | 0 | 0 | n/a | n/a | n/a | 97.781% |
+| 90% | 5,404 / 6,004 | 1,733 | 61 | 39.956% (39.161–40.739) | 10.760% (8.455–12.903) | 96.121% (95.110–96.965) | 92.177% (91.916–92.254) |
+| 80% | 4,803 / 6,004 | 3,518 | 124 | 39.061% (38.780–40.392) | 10.239% (9.622–11.717) | 96.293% (95.795–96.799) | 86.181% (85.995–86.475) |
+| 70% | 4,203 / 6,004 | 5,345 | 173 | 39.884% (38.874–40.123) | 11.314% (9.955–11.690) | 96.745% (95.910–96.973) | 80.462% (80.255–80.549) |
+| 60% | 3,602 / 6,004 | 7,142 | 202 | 39.386% (38.818–39.796) | 10.880% (9.841–11.227) | 96.556% (96.313–97.235) | 74.520% (74.166–74.597) |
+| 50% | 3,002 / 6,004 | 8,978 | 208 | 39.163% (38.873–40.027) | 10.907% (9.934–11.215) | 96.984% (96.480–96.986) | 68.666% (68.447–69.260) |
+| 40% | 2,402 / 6,004 | 10,747 | 182 | 39.008% (38.441–39.397) | 10.196% (9.818–10.790) | 96.596% (96.485–97.321) | 62.762% (62.124–63.171) |
+| 30% | 1,801 / 6,004 | 12,729 | 165 | 38.828% (38.659–39.011) | 10.110% (9.447–10.334) | 97.299% (96.584–97.613) | 56.509% (56.247–56.852) |
+| 20% | 1,201 / 6,004 | 14,564 | 119 | 38.382% (37.388–39.069) | 9.268% (8.010–10.134) | 97.601% (96.973–98.019) | 50.365% (49.357–50.758) |
+| 10% | 600 / 6,004 | 16,458 | 78 | 37.445% (36.685–38.165) | 7.926% (7.223–9.429) | 97.166% (96.036–97.683) | 43.813% (42.423–44.145) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>English</strong> — <code>us-uk-default</code> 1.0.1</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 396,939 / 396,939 | 0 | 0 | n/a | n/a | n/a | 97.668% |
+| 90% | 357,245 / 396,939 | 96,702 | 3,746 | 94.680% (94.609–94.847) | 78.854% (78.538–79.603) | 98.744% (98.642–98.790) | 97.241% (97.237–97.256) |
+| 80% | 317,551 / 396,939 | 193,781 | 6,659 | 94.608% (94.514–94.637) | 78.795% (78.580–79.033) | 98.592% (98.577–98.658) | 96.818% (96.816–96.831) |
+| 70% | 277,857 / 396,939 | 291,831 | 8,726 | 94.425% (94.290–94.438) | 78.505% (77.971–78.614) | 98.454% (98.436–98.500) | 96.386% (96.367–96.394) |
+| 60% | 238,163 / 396,939 | 390,750 | 10,090 | 94.178% (94.121–94.242) | 77.976% (77.824–78.311) | 98.314% (98.268–98.319) | 95.919% (95.916–95.953) |
+| 50% | 198,470 / 396,939 | 490,475 | 10,590 | 93.934% (93.888–94.011) | 77.643% (77.410–77.726) | 98.133% (98.089–98.192) | 95.434% (95.421–95.484) |
+| 40% | 158,776 / 396,939 | 591,318 | 10,198 | 93.699% (93.682–93.712) | 77.308% (77.149–77.479) | 97.940% (97.912–97.948) | 94.946% (94.916–94.960) |
+| 30% | 119,082 / 396,939 | 692,940 | 8,930 | 93.440% (93.409–93.459) | 76.944% (76.715–76.961) | 97.741% (97.718–97.755) | 94.413% (94.372–94.433) |
+| 20% | 79,388 / 396,939 | 795,287 | 6,816 | 93.155% (93.122–93.220) | 76.608% (76.306–76.669) | 97.509% (97.464–97.541) | 93.824% (93.807–93.898) |
+| 10% | 39,694 / 396,939 | 898,572 | 3,741 | 92.805% (92.757–92.843) | 76.010% (75.505–76.378) | 97.184% (97.156–97.293) | 93.157% (93.123–93.211) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Estonian</strong> — <code>et-ee-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 886 / 886 | 0 | 0 | n/a | n/a | n/a | 99.683% |
+| 90% | 797 / 886 | 2,498 | 14 | 72.543% (70.256–75.531) | 72.027% (69.473–74.979) | 91.954% (87.059–93.182) | 96.955% (96.697–97.232) |
+| 80% | 709 / 886 | 4,941 | 34 | 73.075% (70.676–74.043) | 72.300% (69.872–73.433) | 90.805% (90.173–94.286) | 94.331% (93.865–94.576) |
+| 70% | 620 / 886 | 7,416 | 41 | 73.590% (70.159–74.949) | 72.846% (69.407–74.337) | 91.154% (88.931–93.916) | 91.812% (90.796–92.246) |
+| 60% | 532 / 886 | 9,874 | 39 | 71.967% (70.482–73.764) | 71.262% (69.720–73.092) | 91.143% (89.971–92.045) | 88.594% (87.987–89.289) |
+| 50% | 443 / 886 | 12,395 | 40 | 69.786% (68.011–71.332) | 68.977% (67.150–70.599) | 91.344% (89.954–91.972) | 84.705% (83.825–85.452) |
+| 40% | 354 / 886 | 14,894 | 43 | 68.900% (67.070–70.827) | 68.110% (66.209–70.099) | 90.512% (90.341–91.255) | 81.173% (80.064–82.290) |
+| 30% | 266 / 886 | 17,352 | 34 | 67.866% (64.674–70.321) | 67.011% (63.758–69.641) | 89.739% (88.799–92.195) | 77.384% (75.159–79.168) |
+| 20% | 177 / 886 | 19,880 | 24 | 65.953% (60.416–68.265) | 64.959% (59.249–67.400) | 91.773% (88.385–93.020) | 72.680% (68.252–74.556) |
+| 10% | 89 / 886 | 22,377 | 18 | 59.670% (57.126–62.316) | 58.663% (55.871–61.203) | 91.195% (85.912–93.082) | 63.712% (61.398–66.051) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Evenki</strong> — <code>evn-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 4,493 / 4,493 | 0 | 0 | n/a | n/a | n/a | 99.224% |
+| 90% | 4,044 / 4,493 | 1,332 | 16 | 57.112% (53.841–61.939) | 40.202% (36.185–41.864) | 93.258% (91.403–94.558) | 95.042% (94.295–96.126) |
+| 80% | 3,594 / 4,493 | 2,746 | 33 | 56.535% (55.070–59.610) | 39.027% (37.948–41.680) | 93.813% (91.432–95.125) | 90.802% (89.821–91.959) |
+| 70% | 3,145 / 4,493 | 4,012 | 41 | 56.505% (55.558–57.731) | 38.088% (37.468–40.151) | 92.927% (91.811–93.967) | 86.503% (85.954–87.279) |
+| 60% | 2,696 / 4,493 | 5,325 | 46 | 55.704% (54.591–56.376) | 37.165% (36.785–38.585) | 92.772% (91.531–93.154) | 82.065% (80.703–82.827) |
+| 50% | 2,247 / 4,493 | 6,868 | 47 | 55.387% (53.368–56.451) | 36.926% (35.702–38.022) | 91.903% (91.592–92.284) | 76.975% (75.335–78.953) |
+| 40% | 1,797 / 4,493 | 8,423 | 48 | 53.520% (51.145–54.525) | 35.325% (32.679–36.057) | 92.201% (90.906–92.468) | 70.882% (69.308–72.838) |
+| 30% | 1,348 / 4,493 | 9,591 | 48 | 52.973% (51.922–53.645) | 33.658% (33.143–35.688) | 91.877% (90.215–92.231) | 66.203% (65.170–67.807) |
+| 20% | 899 / 4,493 | 11,044 | 38 | 51.920% (49.783–52.722) | 33.092% (29.788–33.853) | 91.076% (90.070–91.678) | 60.879% (58.938–62.190) |
+| 10% | 449 / 4,493 | 12,308 | 20 | 49.277% (48.384–50.374) | 29.966% (27.644–30.596) | 90.458% (88.512–91.485) | 54.390% (52.801–55.196) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Faroese</strong> — <code>fo-fo-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 3,077 / 3,077 | 0 | 0 | n/a | n/a | n/a | 94.099% |
+| 90% | 2,769 / 3,077 | 3,013 | 355 | 58.598% (55.022–60.940) | 55.616% (51.802–57.952) | 88.645% (86.447–91.513) | 90.373% (90.223–90.553) |
+| 80% | 2,462 / 3,077 | 6,049 | 627 | 57.193% (56.722–58.952) | 53.984% (53.690–55.863) | 88.889% (87.339–91.560) | 86.473% (86.260–86.740) |
+| 70% | 2,154 / 3,077 | 9,142 | 846 | 56.317% (55.675–56.778) | 53.132% (52.478–53.720) | 88.214% (87.515–90.315) | 82.564% (82.105–82.813) |
+| 60% | 1,846 / 3,077 | 12,348 | 986 | 54.858% (54.513–55.960) | 51.566% (51.238–52.872) | 87.390% (86.894–88.235) | 78.370% (77.647–78.454) |
+| 50% | 1,539 / 3,077 | 15,658 | 1,058 | 54.005% (53.129–54.560) | 50.720% (49.856–51.365) | 86.494% (85.976–86.750) | 73.594% (73.033–74.182) |
+| 40% | 1,231 / 3,077 | 18,993 | 1,004 | 51.666% (51.130–53.360) | 48.468% (47.771–50.141) | 84.696% (83.623–85.395) | 68.170% (67.801–69.556) |
+| 30% | 923 / 3,077 | 22,370 | 975 | 49.655% (48.319–51.081) | 46.296% (44.840–47.635) | 84.510% (82.821–85.068) | 62.578% (61.774–64.000) |
+| 20% | 615 / 3,077 | 25,938 | 797 | 46.439% (45.686–47.277) | 42.922% (42.262–43.516) | 81.707% (79.688–84.380) | 55.732% (54.736–56.632) |
+| 10% | 308 / 3,077 | 29,641 | 469 | 41.712% (39.458–42.240) | 37.711% (35.058–38.392) | 81.262% (79.618–83.070) | 46.740% (44.523–47.583) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Finnish</strong> — <code>fi-fi-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 57,027 / 57,027 | 0 | 0 | n/a | n/a | n/a | 98.661% |
+| 90% | 51,324 / 57,027 | 181,248 | 5,149 | 86.644% (86.280–86.823) | 86.177% (85.790–86.342) | 94.835% (94.730–95.762) | 97.363% (97.337–97.369) |
+| 80% | 45,622 / 57,027 | 363,208 | 9,479 | 86.152% (86.055–86.432) | 85.686% (85.568–85.979) | 94.265% (94.183–94.726) | 95.981% (95.972–96.026) |
+| 70% | 39,919 / 57,027 | 546,876 | 12,346 | 85.689% (85.586–85.804) | 85.219% (85.114–85.335) | 93.710% (93.575–93.808) | 94.531% (94.518–94.560) |
+| 60% | 34,216 / 57,027 | 730,977 | 14,257 | 85.172% (85.110–85.293) | 84.708% (84.639–84.829) | 93.033% (92.895–93.427) | 92.999% (92.978–93.035) |
+| 50% | 28,514 / 57,027 | 917,680 | 14,946 | 84.681% (84.520–84.723) | 84.232% (84.039–84.280) | 92.203% (92.123–92.566) | 91.382% (91.295–91.404) |
+| 40% | 22,811 / 57,027 | 1,103,628 | 14,607 | 83.991% (83.783–84.165) | 83.556% (83.313–83.715) | 91.299% (91.199–91.628) | 89.582% (89.463–89.675) |
+| 30% | 17,108 / 57,027 | 1,292,480 | 12,981 | 83.290% (82.866–83.366) | 82.848% (82.399–82.936) | 90.499% (90.312–90.537) | 87.658% (87.378–87.720) |
+| 20% | 11,405 / 57,027 | 1,481,274 | 10,279 | 81.958% (81.525–81.970) | 81.481% (81.032–81.508) | 89.459% (89.355–89.678) | 85.129% (84.780–85.155) |
+| 10% | 5,703 / 57,027 | 1,672,373 | 6,067 | 79.210% (78.547–79.594) | 78.644% (77.966–79.067) | 88.247% (87.860–88.496) | 81.064% (80.468–81.422) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>French</strong> — <code>fr-fr-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 59,240 / 59,240 | 0 | 0 | n/a | n/a | n/a | 94.831% |
+| 90% | 53,316 / 59,240 | 41,750 | 5,161 | 83.469% (82.278–83.711) | 80.031% (78.595–80.412) | 96.664% (96.351–96.813) | 93.302% (93.201–93.351) |
+| 80% | 47,392 / 59,240 | 85,315 | 9,281 | 82.721% (81.861–83.091) | 79.345% (78.141–79.693) | 96.074% (95.896–96.266) | 91.731% (91.568–91.794) |
+| 70% | 41,468 / 59,240 | 130,120 | 12,345 | 82.101% (81.643–82.329) | 78.644% (78.015–78.839) | 95.431% (95.257–95.560) | 90.086% (89.942–90.135) |
+| 60% | 35,544 / 59,240 | 175,326 | 14,017 | 81.224% (80.630–81.431) | 77.560% (76.779–77.867) | 94.813% (94.641–94.831) | 88.270% (88.228–88.450) |
+| 50% | 29,620 / 59,240 | 222,533 | 14,808 | 80.344% (79.709–80.528) | 76.569% (75.778–76.888) | 94.006% (93.785–94.099) | 86.507% (86.336–86.609) |
+| 40% | 23,696 / 59,240 | 270,536 | 14,411 | 79.255% (79.039–79.552) | 75.314% (75.088–75.702) | 93.386% (93.086–93.418) | 84.453% (84.319–84.752) |
+| 30% | 17,772 / 59,240 | 318,984 | 12,728 | 78.161% (78.141–78.274) | 74.079% (73.976–74.276) | 92.537% (92.395–92.906) | 82.316% (82.244–82.378) |
+| 20% | 11,848 / 59,240 | 369,337 | 9,783 | 76.769% (76.583–76.946) | 72.504% (72.339–72.772) | 91.463% (90.996–91.884) | 79.741% (79.646–79.873) |
+| 10% | 5,924 / 59,240 | 420,578 | 5,742 | 74.276% (73.276–74.926) | 69.627% (68.400–70.528) | 90.013% (89.987–90.359) | 75.978% (75.101–76.554) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Friulian</strong> — <code>fur-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 168 / 168 | 0 | 0 | n/a | n/a | n/a | 99.801% |
+| 90% | 151 / 168 | 505 | 2 | 78.968% (61.057–90.891) | 78.234% (59.919–90.574) | 100.000% (88.235–100.000) | 97.688% (95.834–98.884) |
+| 80% | 134 / 168 | 1,010 | 4 | 75.522% (71.767–84.378) | 74.691% (70.787–83.934) | 100.000% (94.118–100.000) | 94.897% (94.140–96.671) |
+| 70% | 118 / 168 | 1,485 | 3 | 76.142% (73.670–78.851) | 75.452% (72.911–78.182) | 96.000% (94.000–100.000) | 92.725% (92.007–93.562) |
+| 60% | 101 / 168 | 1,991 | 4 | 72.060% (70.200–73.839) | 71.258% (69.271–72.934) | 95.522% (93.939–100.000) | 88.718% (87.921–89.476) |
+| 50% | 84 / 168 | 2,499 | 6 | 70.991% (70.148–74.359) | 70.140% (69.234–73.726) | 94.048% (92.771–100.000) | 85.390% (84.991–87.084) |
+| 40% | 67 / 168 | 3,009 | 4 | 71.424% (69.042–75.573) | 70.651% (68.060–74.940) | 95.050% (94.000–97.980) | 82.759% (81.304–85.270) |
+| 30% | 50 / 168 | 3,522 | 1 | 71.275% (67.235–75.092) | 70.437% (66.186–74.404) | 96.610% (94.915–98.305) | 79.749% (76.879–82.480) |
+| 20% | 34 / 168 | 4,000 | 1 | 68.530% (67.927–71.704) | 67.493% (66.891–70.903) | 96.269% (94.776–98.507) | 74.846% (74.327–77.417) |
+| 10% | 17 / 168 | 4,507 | 1 | 63.002% (60.674–69.770) | 61.880% (59.381–68.883) | 96.026% (95.364–98.013) | 66.693% (64.640–72.793) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Ga</strong> — <code>gaa-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 94 / 94 | 0 | 0 | n/a | n/a | n/a | 99.787% |
+| 90% | 85 / 94 | 44 | 1 | 60.000% (59.091–61.364) | 50.000% (50.000–51.429) | 100.000% | 95.957% (95.957–96.170) |
+| 80% | 75 / 94 | 94 | 1 | 60.000% (58.511–60.638) | 50.000% (50.000–50.667) | 100.000% (94.444–100.000) | 91.702% (91.489–91.915) |
+| 70% | 66 / 94 | 139 | 1 | 60.000% (58.993–60.432) | 50.000% (50.000–50.450) | 100.000% (96.296–100.000) | 87.872% (87.660–88.085) |
+| 60% | 56 / 94 | 189 | 1 | 60.000% (59.259–60.317) | 50.000% (50.000–50.331) | 100.000% (97.297–100.000) | 83.830% (83.404–83.830) |
+| 50% | 47 / 94 | 234 | 1 | 60.000% (59.402–60.256) | 50.000% (50.000–50.267) | 100.000% (97.826–100.000) | 80.000% (79.574–80.000) |
+| 40% | 38 / 94 | 280 | 0 | 59.643% (57.143–60.215) | 50.000% (46.429–50.224) | 100.000% (98.182–100.000) | 75.745% (74.468–76.170) |
+| 30% | 28 / 94 | 329 | 1 | 60.000% (56.231–60.182) | 50.000% (46.970–50.190) | 100.000% (92.308–100.000) | 71.915% (69.149–71.915) |
+| 20% | 19 / 94 | 375 | 0 | 58.133% (57.067–59.358) | 50.000% (47.333–50.167) | 96.000% (90.667–100.000) | 66.596% (65.745–67.447) |
+| 10% | 9 / 94 | 425 | 0 | 56.000% (53.176–58.491) | 50.000% (42.353–50.147) | 91.765% (80.000–100.000) | 60.213% (57.660–62.340) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Galolen</strong> — <code>gal-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 486 / 486 | 0 | 0 | n/a | n/a | n/a | 97.103% |
+| 90% | 437 / 486 | 2,551 | 94 | 81.665% (80.263–89.176) | 81.307% (79.911–88.961) | 100.000% (97.959–100.000) | 95.331% (95.198–96.076) |
+| 80% | 389 / 486 | 5,001 | 227 | 81.528% (76.585–83.327) | 81.162% (76.142–83.020) | 100.000% (98.969–100.000) | 93.587% (92.888–93.957) |
+| 70% | 340 / 486 | 7,535 | 334 | 81.102% (77.226–84.013) | 80.731% (76.790–83.695) | 100.000% (99.315–100.000) | 91.739% (90.838–92.674) |
+| 60% | 292 / 486 | 10,155 | 293 | 79.048% (78.338–81.300) | 78.656% (77.914–80.946) | 99.485% (98.969–100.000) | 89.490% (89.044–90.265) |
+| 50% | 243 / 486 | 12,745 | 341 | 77.875% (76.205–80.400) | 77.476% (75.790–80.027) | 99.585% (97.531–99.588) | 86.894% (86.421–88.246) |
+| 40% | 194 / 486 | 15,410 | 319 | 77.051% (75.106–77.968) | 76.628% (74.635–77.541) | 98.969% (97.945–100.000) | 84.589% (83.669–85.020) |
+| 30% | 146 / 486 | 18,051 | 283 | 73.281% (71.502–75.971) | 72.785% (70.964–75.510) | 99.412% (98.824–100.000) | 80.130% (79.069–82.245) |
+| 20% | 97 / 486 | 20,701 | 277 | 70.301% (66.061–73.722) | 69.742% (65.425–73.231) | 99.229% (98.201–99.486) | 75.465% (71.808–77.973) |
+| 10% | 49 / 486 | 23,414 | 143 | 68.424% (62.659–71.339) | 67.824% (61.966–70.845) | 99.085% (97.254–100.000) | 71.140% (66.074–73.793) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>German</strong> — <code>de-de-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 54,092 / 54,092 | 0 | 0 | n/a | n/a | n/a | 92.725% |
+| 90% | 48,683 / 54,092 | 27,967 | 5,327 | 67.322% (66.336–67.716) | 59.108% (57.552–59.642) | 93.117% (92.600–93.345) | 89.908% (89.841–89.987) |
+| 80% | 43,274 / 54,092 | 57,109 | 9,524 | 66.531% (65.981–66.706) | 58.016% (57.177–58.316) | 92.441% (91.771–92.581) | 87.066% (86.980–87.121) |
+| 70% | 37,864 / 54,092 | 87,177 | 12,653 | 65.784% (65.442–66.138) | 56.948% (56.648–57.396) | 91.897% (91.743–91.984) | 84.105% (83.974–84.251) |
+| 60% | 32,455 / 54,092 | 118,916 | 14,631 | 65.062% (64.946–65.321) | 56.115% (55.912–56.303) | 91.372% (91.222–91.619) | 81.048% (80.993–81.213) |
+| 50% | 27,046 / 54,092 | 151,254 | 15,476 | 64.493% (64.083–64.603) | 55.371% (54.795–55.510) | 90.896% (90.661–91.008) | 78.014% (77.799–78.062) |
+| 40% | 21,637 / 54,092 | 184,273 | 15,105 | 63.564% (63.219–63.981) | 54.206% (53.646–54.739) | 90.246% (89.885–90.446) | 74.680% (74.544–74.924) |
+| 30% | 16,228 / 54,092 | 219,296 | 13,318 | 62.696% (62.282–63.101) | 53.083% (52.613–53.663) | 89.436% (88.996–89.500) | 71.309% (71.004–71.509) |
+| 20% | 10,818 / 54,092 | 255,708 | 10,520 | 61.569% (60.724–61.785) | 51.959% (50.533–52.161) | 88.244% (88.017–88.723) | 67.504% (66.844–67.696) |
+| 10% | 5,409 / 54,092 | 293,618 | 6,095 | 59.280% (58.219–59.702) | 49.432% (48.026–49.731) | 86.150% (85.836–86.733) | 62.456% (61.579–62.888) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Gothic</strong> — <code>got-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 6,227 / 6,227 | 0 | 0 | n/a | n/a | n/a | 97.403% |
+| 90% | 5,604 / 6,227 | 13,232 | 649 | 80.472% (79.383–82.707) | 80.207% (79.175–82.565) | 87.083% (84.937–87.526) | 95.574% (95.449–95.740) |
+| 80% | 4,982 / 6,227 | 26,611 | 1,193 | 80.622% (79.648–82.183) | 80.498% (79.469–82.070) | 85.100% (83.537–86.476) | 93.779% (93.537–94.088) |
+| 70% | 4,359 / 6,227 | 39,937 | 1,522 | 79.657% (79.456–80.073) | 79.510% (79.290–79.986) | 83.477% (82.269–83.808) | 91.746% (91.603–91.859) |
+| 60% | 3,736 / 6,227 | 53,676 | 1,749 | 78.335% (78.092–79.303) | 78.131% (77.945–79.165) | 82.319% (81.506–83.396) | 89.338% (89.245–89.820) |
+| 50% | 3,114 / 6,227 | 67,349 | 1,787 | 77.388% (76.558–78.588) | 77.219% (76.413–78.480) | 81.137% (79.912–81.979) | 86.951% (86.541–87.607) |
+| 40% | 2,491 / 6,227 | 81,349 | 1,720 | 75.277% (74.626–76.815) | 75.101% (74.451–76.677) | 79.386% (78.283–80.006) | 83.696% (83.412–84.687) |
+| 30% | 1,868 / 6,227 | 95,318 | 1,471 | 73.681% (71.916–74.582) | 73.521% (71.620–74.436) | 77.919% (77.080–78.837) | 80.564% (79.306–81.072) |
+| 20% | 1,245 / 6,227 | 109,649 | 1,133 | 70.595% (69.373–71.060) | 70.340% (68.955–70.794) | 77.280% (76.258–78.677) | 75.746% (74.681–76.058) |
+| 10% | 623 / 6,227 | 123,790 | 616 | 65.034% (64.610–66.241) | 64.563% (64.213–65.759) | 75.487% (73.189–76.679) | 68.243% (67.743–69.337) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Greek</strong> — <code>el-gr-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 11,793 / 11,793 | 0 | 0 | n/a | n/a | n/a | 95.703% |
+| 90% | 10,614 / 11,793 | 7,513 | 543 | 71.881% (70.673–72.434) | 67.694% (66.472–68.281) | 94.512% (93.170–95.848) | 93.126% (92.922–93.181) |
+| 80% | 9,434 / 11,793 | 15,114 | 990 | 70.762% (69.441–71.219) | 66.576% (65.106–67.130) | 94.125% (93.313–94.386) | 90.334% (90.110–90.375) |
+| 70% | 8,255 / 11,793 | 22,824 | 1,334 | 69.379% (68.405–69.738) | 64.998% (63.943–65.390) | 93.759% (93.211–94.026) | 87.210% (87.051–87.446) |
+| 60% | 7,076 / 11,793 | 30,453 | 1,688 | 68.280% (67.787–68.827) | 63.740% (63.142–64.446) | 93.424% (93.403–93.801) | 84.066% (84.041–84.361) |
+| 50% | 5,897 / 11,793 | 38,644 | 1,755 | 67.003% (66.400–67.555) | 62.382% (61.713–62.966) | 93.192% (92.651–93.592) | 80.479% (80.469–80.793) |
+| 40% | 4,717 / 11,793 | 46,615 | 1,785 | 65.678% (65.368–66.758) | 60.970% (60.594–62.151) | 92.560% (92.037–93.088) | 77.065% (76.619–77.463) |
+| 30% | 3,538 / 11,793 | 54,643 | 1,734 | 64.663% (64.102–65.469) | 59.800% (58.994–60.719) | 92.403% (92.169–93.188) | 73.205% (72.745–73.703) |
+| 20% | 2,359 / 11,793 | 62,654 | 1,562 | 62.992% (61.332–63.589) | 57.807% (55.922–58.522) | 92.323% (91.691–92.428) | 68.677% (67.388–69.178) |
+| 10% | 1,179 / 11,793 | 70,918 | 1,288 | 60.208% (59.027–61.709) | 54.741% (53.440–56.626) | 90.957% (90.152–92.617) | 63.220% (61.976–64.391) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Gulf Arabic</strong> — <code>afb-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 6,345 / 6,345 | 0 | 0 | n/a | n/a | n/a | 97.307% |
+| 90% | 5,711 / 6,345 | 2,809 | 145 | 39.840% (39.643–40.940) | 24.696% (22.936–25.303) | 97.907% (97.271–98.387) | 91.713% (91.143–92.128) |
+| 80% | 5,076 / 6,345 | 5,733 | 252 | 40.788% (39.624–41.737) | 24.656% (23.643–25.718) | 97.994% (97.753–98.320) | 86.169% (85.499–86.540) |
+| 70% | 4,442 / 6,345 | 8,691 | 323 | 39.931% (39.496–40.628) | 24.104% (23.624–24.729) | 98.134% (97.704–98.455) | 80.309% (79.659–80.757) |
+| 60% | 3,807 / 6,345 | 11,554 | 393 | 40.107% (39.628–40.305) | 23.806% (23.585–24.761) | 98.044% (97.885–98.325) | 74.721% (74.062–75.308) |
+| 50% | 3,173 / 6,345 | 14,560 | 419 | 39.533% (39.024–39.733) | 23.435% (22.678–23.971) | 98.056% (97.387–98.152) | 68.693% (68.275–69.240) |
+| 40% | 2,538 / 6,345 | 17,455 | 404 | 39.046% (38.782–39.550) | 22.895% (22.407–23.918) | 97.827% (97.510–98.198) | 62.847% (62.382–63.095) |
+| 30% | 1,904 / 6,345 | 20,720 | 366 | 38.716% (37.872–39.173) | 22.870% (21.789–23.152) | 97.508% (97.416–97.983) | 56.523% (55.770–57.272) |
+| 20% | 1,269 / 6,345 | 23,738 | 280 | 37.147% (35.732–38.147) | 20.812% (19.190–22.067) | 97.547% (97.507–97.901) | 49.605% (47.927–50.295) |
+| 10% | 635 / 6,345 | 26,926 | 162 | 35.662% (34.014–36.076) | 19.314% (16.972–19.461) | 97.754% (96.471–97.859) | 42.088% (40.264–42.446) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Haida</strong> — <code>hai-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 36 / 36 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 90% | 32 / 36 | 614 | 0 | 77.850% (52.512–81.922) | 77.705% (52.202–81.803) | 100.000% (75.000–100.000) | 97.474% (94.559–98.162) |
+| 80% | 29 / 36 | 1,076 | 0 | 57.621% (54.498–67.721) | 57.530% (54.268–67.603) | 85.714% (71.429–85.714) | 91.922% (91.012–93.556) |
+| 70% | 25 / 36 | 1,690 | 0 | 65.071% (58.885–73.609) | 64.961% (58.654–73.496) | 90.909% (81.818–90.909) | 89.025% (88.227–91.718) |
+| 60% | 22 / 36 | 2,152 | 0 | 62.207% (50.640–74.117) | 62.041% (50.347–73.994) | 85.714% (78.571–92.857) | 85.942% (81.374–89.656) |
+| 50% | 18 / 36 | 2,767 | 0 | 64.657% (47.677–70.401) | 64.499% (47.318–70.280) | 88.889% (77.778–100.000) | 81.820% (74.280–84.791) |
+| 40% | 14 / 36 | 3,261 | 0 | 61.628% (41.138–69.599) | 61.435% (40.735–69.485) | 90.909% (86.364–100.000) | 76.100% (64.661–81.578) |
+| 30% | 11 / 36 | 3,722 | 0 | 58.958% (55.611–60.183) | 58.734% (55.344–59.940) | 96.000% (92.000–100.000) | 71.625% (68.561–72.479) |
+| 20% | 7 / 36 | 4,310 | 0 | 58.320% (50.980–59.165) | 58.084% (50.673–58.888) | 96.552% (93.103–100.000) | 66.128% (60.520–67.317) |
+| 10% | 4 / 36 | 4,770 | 0 | 52.485% (50.388–61.069) | 52.164% (50.123–60.848) | 96.875% (90.625–100.000) | 57.920% (54.930–65.515) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Hebrew</strong> — <code>he-il-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 2,358 / 2,358 | 0 | 0 | n/a | n/a | n/a | 98.250% |
+| 90% | 2,122 / 2,358 | 5,914 | 196 | 21.813% (21.558–23.195) | 16.362% (15.846–17.917) | 88.785% (87.416–93.636) | 90.688% (90.578–90.888) |
+| 80% | 1,886 / 2,358 | 11,868 | 390 | 21.715% (19.936–22.602) | 16.441% (14.182–17.242) | 89.628% (87.585–91.818) | 83.007% (82.678–83.339) |
+| 70% | 1,651 / 2,358 | 17,846 | 499 | 21.488% (20.548–21.734) | 16.059% (15.072–16.256) | 88.713% (88.166–89.569) | 75.314% (75.106–75.538) |
+| 60% | 1,415 / 2,358 | 23,856 | 551 | 20.530% (20.267–21.311) | 15.043% (14.721–15.824) | 88.702% (87.417–89.456) | 67.302% (67.264–67.705) |
+| 50% | 1,179 / 2,358 | 29,985 | 576 | 20.205% (19.833–20.807) | 14.533% (14.185–15.123) | 88.771% (87.642–90.375) | 59.531% (59.185–59.763) |
+| 40% | 943 / 2,358 | 36,073 | 547 | 19.758% (19.628–20.304) | 14.068% (13.841–14.608) | 89.236% (88.497–90.062) | 51.381% (51.289–51.799) |
+| 30% | 707 / 2,358 | 42,248 | 465 | 19.125% (18.459–19.270) | 13.345% (12.730–13.462) | 88.961% (87.481–90.603) | 43.063% (42.627–43.264) |
+| 20% | 472 / 2,358 | 48,464 | 365 | 17.734% (17.573–18.304) | 11.846% (11.656–12.416) | 89.000% (88.647–89.233) | 34.046% (33.884–34.388) |
+| 10% | 236 / 2,358 | 54,723 | 206 | 16.532% (15.398–18.499) | 10.423% (9.169–12.678) | 89.756% (88.191–91.158) | 24.765% (23.666–26.661) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Hiligaynon</strong> — <code>hil-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 97 / 97 | 0 | 0 | n/a | n/a | n/a | 98.744% |
+| 90% | 87 / 97 | 38 | 1 | 23.684% (18.421–27.027) | 0.000% | 100.000% (87.500–100.000) | 91.457% (90.452–91.960) |
+| 80% | 78 / 97 | 74 | 2 | 24.658% (21.622–25.676) | 0.000% | 100.000% (94.118–100.000) | 84.925% (83.417–84.925) |
+| 70% | 68 / 97 | 114 | 4 | 24.138% (22.807–25.688) | 0.000% | 100.000% (96.296–100.000) | 76.884% (76.131–78.392) |
+| 60% | 58 / 97 | 156 | 3 | 23.780% (23.602–24.359) | 0.000% | 100.000% (97.368–100.000) | 69.095% (67.588–69.347) |
+| 50% | 49 / 97 | 198 | 3 | 24.242% (23.737–24.599) | 0.000% | 100.000% (97.872–100.000) | 62.060% (60.804–63.568) |
+| 40% | 39 / 97 | 237 | 4 | 24.348% (24.167–24.473) | 0.000% | 100.000% (98.246–100.000) | 54.271% (53.266–55.276) |
+| 30% | 29 / 97 | 277 | 3 | 24.014% (23.105–24.632) | 0.000% | 98.529% (94.118–100.000) | 45.729% (45.226–47.739) |
+| 20% | 19 / 97 | 318 | 4 | 23.975% (23.197–24.214) | 0.000% | 98.701% (94.872–98.718) | 38.442% (38.442–38.693) |
+| 10% | 10 / 97 | 356 | 0 | 24.079% (23.315–24.438) | 0.000% | 98.837% (95.402–100.000) | 31.910% (31.407–32.412) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Hsilimo</strong> — <code>hsi-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 49 / 49 | 0 | 0 | n/a | n/a | n/a | 99.371% |
+| 90% | 44 / 49 | 11 | 0 | 40.000% (25.000–45.455) | 0.000% (0.000–10.526) | 100.000% (80.000–100.000) | 95.597% (88.050–95.597) |
+| 80% | 39 / 49 | 30 | 0 | 33.333% (30.769–43.478) | 7.692% (5.000–11.538) | 90.000% (90.000–100.000) | 86.792% (82.390–91.195) |
+| 70% | 34 / 49 | 49 | 0 | 38.000% (30.952–41.667) | 11.429% (3.704–17.647) | 93.333% (80.000–100.000) | 81.132% (76.101–86.164) |
+| 60% | 29 / 49 | 60 | 0 | 37.736% (32.353–44.000) | 10.417% (9.091–13.725) | 90.000% (85.000–95.000) | 76.101% (70.440–81.761) |
+| 50% | 25 / 49 | 79 | 0 | 35.443% (32.500–40.678) | 10.256% (2.857–10.909) | 91.667% (83.333–95.833) | 67.925% (66.038–77.358) |
+| 40% | 20 / 49 | 90 | 1 | 34.043% (30.841–36.364) | 6.410% (2.500–9.836) | 86.207% (82.759–96.552) | 62.264% (53.459–71.698) |
+| 30% | 15 / 49 | 121 | 0 | 33.058% (27.344–38.095) | 6.000% (3.333–11.494) | 88.235% (85.294–97.059) | 49.057% (41.509–66.667) |
+| 20% | 10 / 49 | 132 | 0 | 32.576% (28.777–38.947) | 7.143% (1.786–12.903) | 87.179% (79.487–92.308) | 44.025% (37.736–62.893) |
+| 10% | 5 / 49 | 147 | 0 | 31.034% (19.048–35.338) | 6.731% (3.883–8.911) | 84.091% (54.545–100.000) | 37.107% (25.157–45.283) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Hungarian</strong> — <code>hu-hu-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 19,406 / 19,406 | 0 | 0 | n/a | n/a | n/a | 99.222% |
+| 90% | 17,465 / 19,406 | 92,253 | 1,360 | 92.212% (91.551–92.510) | 92.203% (91.476–92.460) | 93.446% (92.426–93.786) | 98.479% (98.407–98.493) |
+| 80% | 15,525 / 19,406 | 184,401 | 2,417 | 91.771% (91.652–92.138) | 91.756% (91.609–92.123) | 92.536% (92.147–92.726) | 97.646% (97.626–97.712) |
+| 70% | 13,584 / 19,406 | 277,343 | 3,300 | 91.756% (91.390–92.053) | 91.744% (91.350–92.058) | 91.940% (90.827–92.376) | 96.861% (96.757–96.947) |
+| 60% | 11,644 / 19,406 | 370,552 | 3,836 | 91.156% (91.058–91.366) | 91.187% (91.074–91.418) | 90.673% (90.092–91.129) | 95.888% (95.829–95.941) |
+| 50% | 9,703 / 19,406 | 463,693 | 3,952 | 90.729% (90.537–90.756) | 90.777% (90.567–90.810) | 89.530% (89.200–89.806) | 94.834% (94.749–94.863) |
+| 40% | 7,762 / 19,406 | 557,829 | 3,717 | 90.002% (89.679–90.181) | 90.053% (89.707–90.254) | 88.776% (88.318–89.016) | 93.526% (93.390–93.683) |
+| 30% | 5,822 / 19,406 | 651,802 | 3,207 | 89.047% (88.909–89.354) | 89.126% (89.006–89.437) | 87.178% (86.603–87.420) | 92.006% (91.882–92.204) |
+| 20% | 3,881 / 19,406 | 746,292 | 2,401 | 88.280% (87.847–88.842) | 88.393% (87.931–88.968) | 85.865% (85.589–85.920) | 90.395% (90.054–90.844) |
+| 10% | 1,941 / 19,406 | 840,551 | 1,399 | 86.761% (86.100–87.668) | 86.876% (86.177–87.841) | 84.086% (83.633–85.050) | 87.963% (87.354–88.786) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Icelandic</strong> — <code>is-is-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 4,772 / 4,772 | 0 | 0 | n/a | n/a | n/a | 97.769% |
+| 90% | 4,295 / 4,772 | 5,139 | 218 | 64.289% (61.499–65.188) | 61.732% (58.785–62.751) | 90.414% (89.912–90.828) | 94.390% (94.016–94.422) |
+| 80% | 3,818 / 4,772 | 10,259 | 385 | 62.943% (61.744–63.349) | 60.284% (59.010–60.723) | 90.220% (89.531–90.497) | 90.741% (90.399–90.796) |
+| 70% | 3,340 / 4,772 | 15,482 | 476 | 61.305% (60.755–62.259) | 58.577% (57.919–59.577) | 89.462% (88.501–89.753) | 86.875% (86.499–86.984) |
+| 60% | 2,863 / 4,772 | 20,771 | 546 | 60.579% (60.281–60.931) | 57.821% (57.431–58.036) | 89.411% (87.216–90.535) | 82.786% (82.637–82.899) |
+| 50% | 2,386 / 4,772 | 26,119 | 574 | 59.244% (58.720–59.451) | 56.433% (55.741–56.818) | 87.967% (86.583–89.537) | 78.407% (78.042–78.546) |
+| 40% | 1,909 / 4,772 | 31,507 | 568 | 57.580% (56.882–57.842) | 54.646% (53.900–55.045) | 87.442% (85.776–87.790) | 73.593% (73.016–73.771) |
+| 30% | 1,432 / 4,772 | 36,883 | 514 | 55.572% (54.757–56.370) | 52.696% (51.938–53.522) | 85.505% (83.693–87.120) | 68.233% (67.465–68.602) |
+| 20% | 954 / 4,772 | 42,353 | 403 | 52.643% (51.376–53.000) | 49.578% (48.304–50.022) | 83.294% (82.782–84.766) | 61.534% (60.491–61.986) |
+| 10% | 477 / 4,772 | 47,839 | 234 | 47.359% (45.806–47.853) | 43.922% (42.363–44.487) | 82.097% (80.468–82.846) | 52.246% (50.880–52.885) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Indonesian</strong> — <code>id-id-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 3,877 / 3,877 | 0 | 0 | n/a | n/a | n/a | 99.991% |
+| 90% | 3,489 / 3,877 | 2,128 | 0 | 25.191% (24.658–26.363) | 8.960% (8.724–10.172) | 97.938% (97.680–98.969) | 92.628% (92.243–92.652) |
+| 80% | 3,102 / 3,877 | 4,313 | 1 | 25.369% (24.948–26.210) | 9.489% (9.065–10.139) | 98.189% (98.065–98.450) | 84.844% (84.665–85.318) |
+| 70% | 2,714 / 3,877 | 6,413 | 1 | 25.443% (25.035–25.925) | 9.374% (9.002–9.848) | 98.194% (97.762–98.709) | 77.505% (77.129–78.181) |
+| 60% | 2,326 / 3,877 | 8,477 | 1 | 25.561% (25.455–25.857) | 9.399% (9.005–9.667) | 98.645% (97.742–98.904) | 70.387% (69.772–70.739) |
+| 50% | 1,939 / 3,877 | 10,630 | 1 | 25.680% (25.254–25.756) | 9.500% (8.838–9.904) | 98.503% (98.246–98.916) | 62.771% (62.053–63.119) |
+| 40% | 1,551 / 3,877 | 12,805 | 1 | 25.615% (25.342–25.860) | 9.478% (9.047–9.886) | 98.366% (98.108–98.710) | 55.169% (54.897–55.451) |
+| 30% | 1,163 / 3,877 | 14,933 | 0 | 25.486% (25.119–25.775) | 9.417% (8.774–9.572) | 98.563% (98.047–98.674) | 47.624% (47.432–48.051) |
+| 20% | 775 / 3,877 | 17,102 | 0 | 25.219% (24.846–25.535) | 8.976% (8.465–9.414) | 98.420% (97.840–98.775) | 39.868% (39.675–40.436) |
+| 10% | 388 / 3,877 | 19,211 | 0 | 25.084% (24.160–25.386) | 8.790% (7.665–9.036) | 98.653% (98.424–98.825) | 32.271% (31.641–32.919) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Ingrian</strong> — <code>izh-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 50 / 50 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 90% | 45 / 50 | 103 | 0 | 31.068% (4.808–38.679) | 28.571% (0.000–36.634) | 100.000% (80.000–100.000) | 93.080% (90.351–93.665) |
+| 80% | 40 / 50 | 206 | 0 | 32.500% (18.537–38.350) | 29.474% (14.359–35.204) | 100.000% (90.000–100.000) | 86.647% (83.723–87.622) |
+| 70% | 35 / 50 | 308 | 0 | 24.503% (18.065–32.692) | 21.603% (14.237–29.630) | 93.333% (80.000–100.000) | 77.778% (75.244–79.532) |
+| 60% | 30 / 50 | 408 | 0 | 22.542% (14.976–26.225) | 18.892% (10.914–23.454) | 90.000% (80.000–95.000) | 68.811% (65.692–70.663) |
+| 50% | 25 / 50 | 512 | 0 | 23.183% (15.595–26.953) | 19.421% (11.475–23.614) | 96.000% (88.000–96.000) | 61.891% (57.797–63.548) |
+| 40% | 20 / 50 | 613 | 0 | 15.447% (12.235–24.188) | 11.282% (8.576–20.819) | 90.000% (76.667–96.667) | 49.318% (47.563–54.483) |
+| 30% | 15 / 50 | 716 | 0 | 17.246% (14.385–19.720) | 13.158% (10.866–15.735) | 94.286% (74.286–97.143) | 42.008% (40.253–44.055) |
+| 20% | 10 / 50 | 819 | 0 | 14.408% (12.317–16.361) | 10.398% (8.184–12.580) | 90.000% (80.000–95.000) | 31.676% (29.825–33.236) |
+| 10% | 5 / 50 | 923 | 0 | 14.843% (8.234–16.143) | 11.060% (5.011–12.984) | 82.222% (71.111–93.333) | 23.392% (17.446–24.561) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Irish</strong> — <code>ga-ie-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 7,287 / 7,287 | 0 | 0 | n/a | n/a | n/a | 97.970% |
+| 90% | 6,558 / 7,287 | 2,441 | 97 | 62.515% (58.703–64.252) | 48.046% (45.339–50.654) | 95.286% (94.193–96.591) | 94.249% (93.625–94.628) |
+| 80% | 5,830 / 7,287 | 4,727 | 169 | 62.021% (59.558–62.780) | 47.969% (45.253–48.409) | 95.400% (94.433–95.593) | 90.494% (89.818–91.126) |
+| 70% | 5,101 / 7,287 | 7,180 | 192 | 59.959% (59.763–61.558) | 46.037% (45.003–47.023) | 94.747% (93.942–94.855) | 86.353% (85.966–87.242) |
+| 60% | 4,372 / 7,287 | 9,534 | 229 | 60.036% (58.677–61.097) | 45.376% (44.646–46.897) | 93.897% (93.196–94.281) | 82.905% (81.608–83.219) |
+| 50% | 3,644 / 7,287 | 12,029 | 243 | 58.873% (58.630–60.254) | 43.920% (43.837–46.118) | 93.630% (92.576–93.996) | 78.209% (77.981–78.967) |
+| 40% | 2,915 / 7,287 | 14,494 | 242 | 58.366% (57.960–60.425) | 43.582% (42.943–46.513) | 93.353% (92.730–93.718) | 74.161% (73.631–75.478) |
+| 30% | 2,186 / 7,287 | 16,925 | 214 | 57.339% (56.455–58.407) | 42.711% (41.117–43.849) | 92.463% (92.177–93.337) | 69.213% (68.459–70.354) |
+| 20% | 1,457 / 7,287 | 19,432 | 187 | 56.294% (55.272–56.642) | 41.141% (39.771–41.389) | 92.039% (91.943–92.453) | 64.574% (63.539–64.990) |
+| 10% | 729 / 7,287 | 21,938 | 112 | 52.789% (52.518–53.743) | 36.464% (35.603–37.611) | 91.687% (91.525–92.561) | 57.290% (56.866–58.370) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Italian</strong> — <code>it-it-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 10,009 / 10,009 | 0 | 0 | n/a | n/a | n/a | 99.056% |
+| 90% | 9,008 / 10,009 | 33,156 | 558 | 93.965% (93.205–94.474) | 93.600% (92.783–94.134) | 99.900% (99.600–100.000) | 98.473% (98.385–98.520) |
+| 80% | 8,007 / 10,009 | 66,376 | 1,008 | 93.594% (93.308–93.801) | 93.192% (92.886–93.418) | 99.900% (99.750–99.950) | 97.816% (97.782–97.848) |
+| 70% | 7,006 / 10,009 | 99,459 | 1,285 | 93.065% (92.894–93.176) | 92.631% (92.450–92.747) | 99.867% (99.800–99.900) | 97.072% (97.043–97.115) |
+| 60% | 6,005 / 10,009 | 132,730 | 1,556 | 92.510% (92.196–92.768) | 92.037% (91.712–92.312) | 99.863% (99.750–99.900) | 96.267% (96.099–96.348) |
+| 50% | 5,005 / 10,009 | 166,665 | 1,637 | 91.882% (91.717–92.283) | 91.375% (91.195–91.801) | 99.840% (99.780–99.920) | 95.296% (95.190–95.476) |
+| 40% | 4,004 / 10,009 | 199,950 | 1,479 | 91.606% (91.107–91.863) | 91.077% (90.547–91.356) | 99.842% (99.783–99.883) | 94.458% (94.135–94.521) |
+| 30% | 3,003 / 10,009 | 234,119 | 1,370 | 90.846% (90.583–91.131) | 90.272% (89.989–90.573) | 99.879% (99.800–99.929) | 93.137% (92.971–93.395) |
+| 20% | 2,002 / 10,009 | 268,626 | 1,035 | 89.611% (89.163–90.122) | 88.964% (88.479–89.500) | 99.888% (99.725–99.963) | 91.388% (91.005–91.839) |
+| 10% | 1,001 / 10,009 | 303,036 | 513 | 87.447% (87.014–87.906) | 86.665% (86.198–87.161) | 99.794% (99.556–99.917) | 88.555% (88.194–88.979) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Itelmen</strong> — <code>itl-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 1,633 / 1,633 | 0 | 0 | n/a | n/a | n/a | 99.891% |
+| 90% | 1,470 / 1,633 | 345 | 0 | 52.754% (49.322–58.095) | 14.835% (10.680–18.421) | 95.092% (93.865–98.160) | 95.441% (94.731–96.287) |
+| 80% | 1,306 / 1,633 | 699 | 0 | 52.217% (50.689–54.425) | 14.500% (12.694–15.670) | 95.719% (94.495–96.024) | 90.800% (90.145–91.455) |
+| 70% | 1,143 / 1,633 | 1,070 | 1 | 51.648% (49.228–53.571) | 14.925% (11.438–15.663) | 96.524% (95.297–96.933) | 85.804% (84.630–86.759) |
+| 60% | 980 / 1,633 | 1,423 | 2 | 51.406% (49.557–53.127) | 14.324% (12.255–16.083) | 96.319% (94.172–96.933) | 81.054% (79.716–81.682) |
+| 50% | 817 / 1,633 | 1,812 | 2 | 50.084% (49.007–50.416) | 12.755% (10.030–13.392) | 96.201% (94.969–96.687) | 74.884% (74.720–75.566) |
+| 40% | 653 / 1,633 | 2,205 | 2 | 49.162% (48.934–49.794) | 11.491% (10.440–12.893) | 96.327% (95.603–97.143) | 69.397% (69.206–69.997) |
+| 30% | 490 / 1,633 | 2,547 | 2 | 48.486% (48.284–50.640) | 10.965% (9.979–11.570) | 96.056% (95.534–97.025) | 64.182% (63.582–66.257) |
+| 20% | 327 / 1,633 | 2,914 | 1 | 48.461% (47.577–49.429) | 9.724% (9.327–10.992) | 95.865% (95.475–97.011) | 58.832% (57.439–60.060) |
+| 10% | 163 / 1,633 | 3,297 | 1 | 47.766% (47.277–49.802) | 9.390% (7.953–11.443) | 96.463% (95.034–97.211) | 53.372% (52.143–55.037) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Japanese</strong> — <code>ja-jp-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 107 / 107 | 0 | 0 | n/a | n/a | n/a | 93.582% |
+| 90% | 96 / 107 | 1,051 | 289 | 32.953% (16.580–35.775) | 32.358% (15.612–35.159) | 100.000% | 87.129% (86.680–87.759) |
+| 80% | 86 / 107 | 2,026 | 460 | 29.268% (18.252–31.935) | 28.591% (17.410–31.256) | 100.000% | 81.686% (79.055–82.057) |
+| 70% | 75 / 107 | 3,125 | 613 | 30.312% (22.755–34.724) | 29.529% (22.050–34.088) | 100.000% (93.750–100.000) | 75.224% (71.981–76.803) |
+| 60% | 64 / 107 | 4,313 | 464 | 28.292% (23.997–36.642) | 27.617% (23.279–35.994) | 100.000% (95.349–100.000) | 68.953% (65.485–70.549) |
+| 50% | 54 / 107 | 5,315 | 626 | 27.649% (23.698–35.551) | 26.984% (22.938–34.958) | 96.154% (94.231–100.000) | 61.767% (59.938–65.666) |
+| 40% | 43 / 107 | 6,522 | 403 | 29.566% (20.132–35.524) | 28.902% (19.399–34.983) | 96.825% (95.238–98.438) | 56.461% (51.259–58.756) |
+| 30% | 32 / 107 | 8,012 | 276 | 27.471% (23.357–30.781) | 26.833% (22.715–30.177) | 95.946% (93.243–98.630) | 47.886% (43.892–50.138) |
+| 20% | 21 / 107 | 9,182 | 222 | 28.009% (21.292–30.423) | 27.447% (20.600–29.827) | 95.294% (88.372–100.000) | 41.615% (35.343–42.227) |
+| 10% | 11 / 107 | 10,133 | 214 | 24.223% (19.027–29.252) | 23.578% (18.330–28.626) | 92.708% (92.632–96.875) | 32.203% (27.372–35.913) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Kabardian</strong> — <code>kbd-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 249 / 249 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 90% | 224 / 249 | 307 | 0 | 59.283% (51.792–61.489) | 56.738% (48.227–58.719) | 88.000% (80.000–96.000) | 95.907% (95.154–96.103) |
+| 80% | 199 / 249 | 612 | 0 | 57.166% (52.614–62.233) | 53.723% (49.288–59.928) | 90.000% (86.000–96.000) | 91.388% (90.504–92.469) |
+| 70% | 174 / 249 | 920 | 0 | 54.148% (51.316–56.957) | 50.654% (47.790–53.846) | 92.000% (85.333–93.333) | 86.248% (85.363–87.033) |
+| 60% | 149 / 249 | 1,222 | 0 | 54.010% (53.295–57.878) | 50.980% (50.221–54.844) | 88.000% (88.000–92.000) | 81.598% (81.205–83.104) |
+| 50% | 125 / 249 | 1,518 | 0 | 53.057% (48.219–58.432) | 49.821% (44.325–55.667) | 90.323% (89.516–91.935) | 76.621% (74.296–79.339) |
+| 40% | 100 / 249 | 1,824 | 0 | 51.371% (47.917–52.626) | 48.358% (44.119–49.792) | 89.933% (84.564–91.275) | 70.956% (68.893–71.644) |
+| 30% | 75 / 249 | 2,131 | 0 | 46.457% (43.212–50.328) | 42.514% (39.348–47.293) | 86.782% (84.483–90.805) | 62.639% (60.282–65.324) |
+| 20% | 50 / 249 | 2,441 | 0 | 46.308% (40.639–48.259) | 42.296% (36.128–44.871) | 91.457% (82.915–93.467) | 57.138% (52.554–58.644) |
+| 10% | 25 / 249 | 2,748 | 0 | 38.014% (32.860–46.308) | 32.990% (27.456–42.059) | 93.750% (80.804–94.643) | 44.204% (39.587–51.670) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Kalaallisut</strong> — <code>kl-gl-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 23 / 23 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 90% | 21 / 23 | 28 | 0 | 28.571% (7.143–75.000) | 23.077% (0.000–73.077) | 100.000% | 93.769% (91.900–97.819) |
+| 80% | 18 / 23 | 70 | 0 | 26.087% (7.246–31.429) | 20.312% (0.000–26.154) | 100.000% | 84.112% (80.062–85.047) |
+| 70% | 16 / 23 | 98 | 0 | 17.526% (11.224–28.571) | 11.111% (4.396–23.077) | 100.000% | 75.078% (72.897–78.193) |
+| 60% | 14 / 23 | 125 | 0 | 20.800% (12.800–30.952) | 14.655% (6.034–25.641) | 100.000% | 69.159% (66.044–72.897) |
+| 50% | 12 / 23 | 153 | 0 | 22.078% (14.379–29.412) | 16.084% (7.746–23.944) | 100.000% | 62.617% (59.190–66.355) |
+| 40% | 9 / 23 | 195 | 0 | 23.590% (21.026–28.718) | 17.680% (14.917–23.204) | 100.000% (92.857–100.000) | 53.583% (52.025–56.698) |
+| 30% | 7 / 23 | 223 | 0 | 26.457% (8.072–32.735) | 21.256% (0.966–27.536) | 93.750% (93.750–100.000) | 48.910% (36.137–53.271) |
+| 20% | 5 / 23 | 251 | 0 | 21.514% (8.367–33.865) | 15.451% (1.288–29.185) | 100.000% (94.444–100.000) | 38.629% (28.349–48.287) |
+| 10% | 2 / 23 | 293 | 0 | 21.502% (7.850–29.693) | 15.441% (0.735–26.103) | 100.000% (76.190–100.000) | 28.349% (15.888–35.826) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Kannada</strong> — <code>kn-in-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 159 / 159 | 0 | 0 | n/a | n/a | n/a | 91.241% |
+| 90% | 143 / 159 | 387 | 36 | 75.223% (71.008–86.996) | 74.537% (70.721–85.990) | 93.750% (75.000–100.000) | 89.177% (88.049–90.545) |
+| 80% | 127 / 159 | 779 | 77 | 78.354% (54.367–88.331) | 78.232% (52.532–87.863) | 90.625% (81.250–96.875) | 87.137% (83.969–89.513) |
+| 70% | 111 / 159 | 1,175 | 113 | 79.745% (60.090–82.735) | 79.462% (58.531–82.481) | 89.583% (85.417–95.833) | 85.529% (82.697–85.697) |
+| 60% | 95 / 159 | 1,495 | 140 | 76.067% (59.623–80.262) | 75.547% (58.065–79.783) | 89.062% (87.500–93.750) | 81.689% (78.426–84.377) |
+| 50% | 80 / 159 | 1,884 | 176 | 67.685% (63.587–83.071) | 67.094% (62.143–83.048) | 89.873% (82.278–92.405) | 76.770% (75.618–84.401) |
+| 40% | 64 / 159 | 2,257 | 221 | 71.988% (65.007–84.571) | 71.137% (63.961–84.689) | 86.316% (82.105–92.632) | 75.114% (73.938–83.633) |
+| 30% | 48 / 159 | 2,591 | 243 | 67.518% (60.948–79.644) | 66.653% (59.428–79.578) | 89.189% (81.081–96.396) | 70.050% (66.067–79.002) |
+| 20% | 32 / 159 | 3,093 | 297 | 64.719% (61.571–77.494) | 63.920% (60.302–77.249) | 84.252% (81.102–91.339) | 64.747% (62.707–76.146) |
+| 10% | 16 / 159 | 3,388 | 315 | 71.257% (46.724–75.516) | 70.520% (44.838–75.135) | 83.916% (76.224–89.510) | 67.819% (48.044–71.994) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Karelian</strong> — <code>krl-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 20 / 20 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 90% | 18 / 20 | 57 | 0 | 63.793% (3.571–91.228) | 62.500% (0.000–90.909) | 100.000% (50.000–100.000) | 96.290% (90.459–99.117) |
+| 80% | 16 / 20 | 114 | 0 | 33.913% (15.929–59.649) | 31.532% (12.844–58.182) | 100.000% (75.000–100.000) | 86.572% (83.216–91.873) |
+| 70% | 14 / 20 | 171 | 0 | 40.000% (19.883–40.351) | 38.182% (16.970–38.788) | 83.333% (83.333–100.000) | 81.979% (75.795–81.979) |
+| 60% | 12 / 20 | 227 | 0 | 36.444% (27.313–41.667) | 35.484% (24.658–40.367) | 87.500% (62.500–100.000) | 74.735% (70.848–76.678) |
+| 50% | 10 / 20 | 283 | 0 | 34.859% (13.380–43.772) | 32.847% (10.219–43.173) | 80.000% (60.000–100.000) | 67.314% (56.537–72.085) |
+| 40% | 8 / 20 | 340 | 0 | 37.941% (18.475–52.522) | 35.976% (15.502–52.000) | 83.333% (66.667–100.000) | 62.721% (50.883–71.731) |
+| 30% | 6 / 20 | 397 | 0 | 35.859% (11.055–46.348) | 34.293% (7.812–45.431) | 78.571% (71.429–100.000) | 55.124% (37.456–62.367) |
+| 20% | 4 / 20 | 453 | 0 | 23.620% (15.859–31.347) | 21.281% (13.242–29.977) | 87.500% (68.750–100.000) | 38.869% (32.509–45.053) |
+| 10% | 2 / 20 | 510 | 0 | 16.275% (13.725–24.902) | 13.415% (11.179–22.561) | 88.889% (77.778–94.444) | 24.558% (22.261–32.332) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Kashubian</strong> — <code>csb-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 37 / 37 | 0 | 0 | n/a | n/a | n/a | 98.592% |
+| 90% | 33 / 37 | 38 | 0 | 45.161% (26.667–78.947) | 37.037% (15.385–76.471) | 100.000% | 94.366% (91.549–96.338) |
+| 80% | 30 / 37 | 62 | 0 | 46.667% (22.414–88.406) | 39.623% (11.765–87.097) | 100.000% | 89.577% (87.042–96.338) |
+| 70% | 26 / 37 | 101 | 0 | 47.525% (44.444–76.852) | 41.111% (37.500–74.227) | 100.000% | 85.352% (83.662–91.549) |
+| 60% | 22 / 37 | 139 | 1 | 54.422% (43.796–70.504) | 50.000% (36.885–66.935) | 100.000% (93.333–100.000) | 80.000% (78.028–87.042) |
+| 50% | 19 / 37 | 167 | 1 | 61.932% (43.373–70.659) | 58.228% (38.255–67.333) | 94.444% (88.235–100.000) | 79.718% (73.239–84.789) |
+| 40% | 15 / 37 | 207 | 1 | 63.415% (42.927–72.947) | 59.783% (36.957–69.730) | 95.455% (95.238–100.000) | 78.028% (66.761–83.944) |
+| 30% | 11 / 37 | 247 | 1 | 61.728% (50.206–63.563) | 57.798% (44.954–59.729) | 96.000% (92.308–100.000) | 73.521% (65.634–74.085) |
+| 20% | 7 / 37 | 284 | 1 | 56.446% (45.296–61.620) | 52.140% (38.911–57.874) | 96.552% (93.333–100.000) | 63.662% (55.775–69.296) |
+| 10% | 4 / 37 | 315 | 0 | 51.746% (29.936–55.414) | 47.518% (22.420–50.355) | 93.939% (87.879–100.000) | 57.183% (38.028–60.282) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Kazakh</strong> — <code>kk-kz-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 1,744 / 1,744 | 0 | 0 | n/a | n/a | n/a | 99.870% |
+| 90% | 1,570 / 1,744 | 3,543 | 10 | 89.984% (88.675–94.239) | 89.542% (88.200–94.208) | 94.828% (93.103–98.844) | 98.835% (98.755–99.314) |
+| 80% | 1,395 / 1,744 | 7,139 | 15 | 90.499% (88.868–92.873) | 90.154% (88.588–92.829) | 94.813% (93.696–97.126) | 97.976% (97.530–98.483) |
+| 70% | 1,221 / 1,744 | 10,678 | 25 | 90.825% (89.263–91.628) | 90.536% (89.012–91.489) | 95.183% (94.242–96.346) | 97.145% (96.561–97.400) |
+| 60% | 1,046 / 1,744 | 14,436 | 21 | 90.053% (88.731–91.144) | 89.777% (88.443–90.998) | 95.396% (93.966–95.696) | 95.815% (95.205–96.365) |
+| 50% | 872 / 1,744 | 17,925 | 22 | 90.309% (89.158–91.132) | 90.042% (88.871–90.963) | 95.287% (94.374–95.502) | 95.038% (94.241–95.509) |
+| 40% | 698 / 1,744 | 21,450 | 21 | 89.753% (89.473–89.919) | 89.496% (89.200–89.648) | 94.823% (93.971–95.110) | 93.601% (93.499–93.884) |
+| 30% | 523 / 1,744 | 24,875 | 19 | 89.918% (88.571–90.303) | 89.736% (88.231–90.098) | 94.668% (93.443–95.399) | 92.844% (91.760–93.144) |
+| 20% | 349 / 1,744 | 28,152 | 22 | 89.158% (87.839–89.727) | 88.909% (87.502–89.544) | 93.978% (93.252–94.971) | 91.191% (90.056–91.769) |
+| 10% | 174 / 1,744 | 31,956 | 2 | 84.509% (81.987–87.786) | 84.045% (81.378–87.437) | 93.810% (91.013–94.837) | 86.033% (83.646–88.857) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Khakas</strong> — <code>kjh-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 74 / 74 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 90% | 67 / 74 | 110 | 0 | 89.091% (83.636–97.297) | 90.291% (83.495–97.115) | 85.714% (71.429–100.000) | 98.976% (98.464–99.744) |
+| 80% | 59 / 74 | 236 | 0 | 88.983% (86.809–89.496) | 88.739% (86.364–89.140) | 93.333% (80.000–100.000) | 97.782% (97.355–97.867) |
+| 70% | 52 / 74 | 346 | 0 | 90.517% (83.815–92.264) | 90.184% (83.642–92.049) | 95.455% (86.364–95.455) | 97.184% (95.222–97.696) |
+| 60% | 44 / 74 | 473 | 0 | 90.063% (86.469–92.437) | 90.068% (86.230–92.152) | 93.333% (90.000–96.667) | 95.990% (94.539–96.928) |
+| 50% | 37 / 74 | 585 | 0 | 87.521% (73.024–91.610) | 87.226% (71.560–91.590) | 91.892% (89.189–94.595) | 93.771% (86.604–95.819) |
+| 40% | 30 / 74 | 696 | 0 | 88.429% (75.325–90.531) | 87.957% (73.960–90.505) | 95.455% (88.636–95.455) | 93.089% (85.410–94.369) |
+| 30% | 22 / 74 | 822 | 0 | 78.710% (73.082–88.443) | 77.922% (71.521–87.953) | 96.154% (90.385–98.077) | 85.068% (81.143–91.894) |
+| 20% | 15 / 74 | 934 | 0 | 72.883% (68.128–76.231) | 71.167% (65.982–75.200) | 98.305% (91.525–100.000) | 78.413% (74.573–81.058) |
+| 10% | 7 / 74 | 1,061 | 0 | 67.295% (61.038–73.917) | 65.392% (58.812–72.663) | 95.522% (92.537–98.507) | 70.392% (64.761–76.365) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Khinalugh</strong> — <code>klr-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 591 / 591 | 0 | 0 | n/a | n/a | n/a | 91.557% |
+| 90% | 532 / 591 | 5,193 | 1,045 | 6.586% (6.293–7.570) | 5.512% (5.242–6.477) | 100.000% | 83.423% (82.881–83.851) |
+| 80% | 473 / 591 | 10,373 | 1,645 | 7.299% (6.460–7.719) | 6.228% (5.448–6.656) | 100.000% (99.153–100.000) | 75.545% (74.568–76.001) |
+| 70% | 414 / 591 | 16,046 | 2,000 | 7.716% (6.375–8.050) | 6.682% (5.331–7.017) | 100.000% (99.435–100.000) | 67.384% (66.358–67.695) |
+| 60% | 355 / 591 | 21,089 | 2,596 | 7.225% (6.959–8.578) | 6.167% (5.900–7.543) | 100.000% (99.576–100.000) | 59.711% (58.392–60.055) |
+| 50% | 296 / 591 | 27,059 | 2,482 | 7.417% (6.642–8.527) | 6.397% (5.602–7.526) | 100.000% (99.661–100.000) | 51.396% (50.589–52.169) |
+| 40% | 236 / 591 | 33,181 | 2,371 | 7.505% (6.579–8.113) | 6.504% (5.548–7.125) | 100.000% (99.437–100.000) | 43.279% (42.383–43.289) |
+| 30% | 177 / 591 | 39,347 | 2,013 | 7.712% (7.589–8.641) | 6.738% (6.606–7.664) | 100.000% (99.758–100.000) | 34.666% (33.449–35.629) |
+| 20% | 118 / 591 | 45,965 | 1,650 | 7.750% (7.367–8.809) | 6.792% (6.422–7.861) | 100.000% (99.789–100.000) | 25.996% (23.831–26.552) |
+| 10% | 59 / 591 | 52,551 | 937 | 7.120% (6.516–8.075) | 6.167% (5.560–7.133) | 100.000% (99.812–100.000) | 16.381% (14.826–17.670) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Kodi</strong> — <code>kod-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 65 / 65 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 90% | 59 / 65 | 48 | 0 | 12.500% (12.500–20.000) | 0.000% (0.000–10.204) | 100.000% | 91.985% (91.603–91.985) |
+| 80% | 52 / 65 | 104 | 0 | 15.315% (12.500–20.192) | 4.082% (0.000–9.184) | 100.000% | 83.015% (82.061–84.160) |
+| 70% | 46 / 65 | 152 | 0 | 15.789% (13.333–18.543) | 3.759% (0.763–6.818) | 100.000% | 75.191% (74.046–76.527) |
+| 60% | 39 / 65 | 213 | 0 | 14.493% (13.592–16.432) | 2.660% (1.111–4.813) | 100.000% | 66.031% (65.076–66.221) |
+| 50% | 33 / 65 | 261 | 0 | 14.504% (12.644–17.625) | 2.609% (0.437–6.114) | 100.000% | 57.252% (56.489–58.969) |
+| 40% | 26 / 65 | 317 | 0 | 15.773% (12.579–17.350) | 3.957% (0.358–5.755) | 100.000% | 49.046% (46.947–50.000) |
+| 30% | 20 / 65 | 365 | 0 | 14.835% (12.295–17.582) | 2.821% (0.000–5.956) | 100.000% | 40.840% (38.740–42.748) |
+| 20% | 13 / 65 | 420 | 0 | 17.381% (12.352–19.524) | 5.978% (0.271–8.152) | 100.000% (98.077–100.000) | 33.779% (29.580–35.496) |
+| 10% | 7 / 65 | 468 | 0 | 15.139% (13.248–16.880) | 3.171% (1.463–5.854) | 96.552% (94.828–100.000) | 24.046% (22.519–25.763) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Kongo</strong> — <code>kon-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 186 / 186 | 0 | 0 | n/a | n/a | n/a | 99.821% |
+| 90% | 167 / 186 | 57 | 0 | 64.912% (61.404–66.667) | 47.368% (42.105–50.000) | 100.000% | 96.237% (95.878–96.416) |
+| 80% | 149 / 186 | 110 | 1 | 64.865% (62.162–66.364) | 47.297% (44.595–50.000) | 100.000% (97.297–100.000) | 92.832% (92.294–93.190) |
+| 70% | 130 / 186 | 168 | 0 | 64.881% (62.500–66.467) | 48.649% (44.643–50.000) | 100.000% (94.643–100.000) | 89.427% (88.530–89.785) |
+| 60% | 112 / 186 | 222 | 0 | 65.315% (63.063–66.516) | 48.299% (45.270–50.000) | 100.000% (95.946–100.000) | 86.201% (85.125–86.559) |
+| 50% | 93 / 186 | 279 | 0 | 65.468% (64.158–66.547) | 48.649% (46.237–50.000) | 100.000% (96.774–100.000) | 82.616% (81.900–83.154) |
+| 40% | 74 / 186 | 336 | 0 | 65.774% (61.012–66.369) | 48.879% (41.518–50.000) | 100.000% (97.321–100.000) | 79.391% (76.344–79.749) |
+| 30% | 56 / 186 | 390 | 0 | 64.010% (61.282–66.410) | 45.946% (41.923–50.000) | 99.231% (97.692–100.000) | 74.731% (72.760–76.523) |
+| 20% | 37 / 186 | 447 | 0 | 66.219% (58.613–66.443) | 50.000% (38.255–50.000) | 99.329% (98.658–100.000) | 72.939% (66.667–73.118) |
+| 10% | 19 / 186 | 501 | 0 | 66.267% (62.200–66.467) | 50.000% (43.243–50.000) | 99.401% (98.802–100.000) | 69.713% (65.950–69.892) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Kyrgyz</strong> — <code>ky-kg-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 72 / 72 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 90% | 65 / 72 | 291 | 0 | 61.724% (58.419–89.078) | 60.777% (57.394–88.811) | 100.000% | 96.296% (95.963–98.932) |
+| 80% | 58 / 72 | 584 | 0 | 72.260% (67.526–76.298) | 71.579% (66.725–75.709) | 100.000% | 94.595% (93.694–95.429) |
+| 70% | 50 / 72 | 917 | 0 | 67.467% (55.761–72.618) | 66.890% (54.677–71.942) | 100.000% (90.909–100.000) | 90.057% (86.420–91.658) |
+| 60% | 43 / 72 | 1,208 | 0 | 60.348% (58.989–69.421) | 59.372% (57.980–68.840) | 100.000% (93.103–100.000) | 84.017% (83.483–87.654) |
+| 50% | 36 / 72 | 1,499 | 0 | 61.041% (56.379–67.399) | 60.219% (55.305–66.735) | 100.000% (94.444–100.000) | 80.514% (78.212–83.650) |
+| 40% | 29 / 72 | 1,791 | 0 | 57.327% (49.888–65.737) | 56.275% (48.768–65.009) | 95.349% (95.349–100.000) | 74.541% (70.103–79.513) |
+| 30% | 22 / 72 | 2,080 | 0 | 57.109% (43.861–64.856) | 56.152% (42.575–64.089) | 96.000% (96.000–100.000) | 70.204% (61.094–75.609) |
+| 20% | 14 / 72 | 2,414 | 0 | 53.499% (47.015–60.821) | 52.440% (45.794–59.941) | 96.552% (96.552–100.000) | 62.529% (57.357–68.468) |
+| 10% | 7 / 72 | 2,705 | 0 | 42.382% (31.941–45.830) | 41.038% (30.341–44.575) | 96.923% | 48.015% (38.572–51.018) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Ladin</strong> — <code>lld-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 180 / 180 | 0 | 0 | n/a | n/a | n/a | 99.279% |
+| 90% | 162 / 180 | 491 | 0 | 77.800% (60.887–79.388) | 76.956% (59.414–78.602) | 100.000% (94.444–100.000) | 97.033% (95.282–97.198) |
+| 80% | 144 / 180 | 972 | 0 | 78.969% (66.157–80.597) | 78.373% (64.868–80.066) | 97.222% (94.286–100.000) | 95.076% (92.439–95.529) |
+| 70% | 126 / 180 | 1,433 | 11 | 73.688% (70.986–76.343) | 72.820% (69.936–75.580) | 96.226% (94.444–98.148) | 91.533% (90.606–92.295) |
+| 60% | 108 / 180 | 1,911 | 11 | 76.335% (73.781–76.939) | 75.571% (72.882–76.157) | 97.183% (95.833–98.611) | 89.905% (88.752–90.214) |
+| 50% | 90 / 180 | 2,421 | 11 | 73.647% (69.540–76.177) | 72.727% (68.841–75.346) | 97.753% (87.778–98.889) | 86.341% (83.993–87.495) |
+| 40% | 72 / 180 | 2,881 | 25 | 67.750% (67.515–75.417) | 66.655% (66.546–74.576) | 97.196% (89.815–98.131) | 80.264% (79.810–84.693) |
+| 30% | 54 / 180 | 3,375 | 25 | 68.326% (64.209–74.487) | 67.251% (62.921–73.657) | 96.000% (90.476–97.600) | 77.750% (74.495–81.809) |
+| 20% | 36 / 180 | 3,865 | 25 | 71.384% (57.749–74.390) | 70.361% (56.314–73.538) | 96.503% (91.667–97.902) | 76.597% (65.843–79.151) |
+| 10% | 18 / 180 | 4,357 | 1 | 63.080% (61.212–76.946) | 62.091% (59.785–76.109) | 93.210% (86.420–98.758) | 66.399% (65.183–78.822) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Latin</strong> — <code>la-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 46,170 / 46,170 | 0 | 0 | n/a | n/a | n/a | 97.133% |
+| 90% | 41,553 / 46,170 | 47,545 | 2,582 | 78.052% (77.620–78.993) | 76.426% (76.045–77.427) | 94.736% (93.520–94.916) | 95.060% (95.027–95.163) |
+| 80% | 36,936 / 46,170 | 95,630 | 4,516 | 77.449% (77.132–78.300) | 75.824% (75.533–76.728) | 94.046% (93.465–94.163) | 92.975% (92.830–93.105) |
+| 70% | 32,319 / 46,170 | 143,770 | 5,859 | 77.125% (76.973–77.589) | 75.499% (75.342–75.988) | 93.403% (93.114–93.657) | 90.839% (90.691–90.941) |
+| 60% | 27,702 / 46,170 | 192,657 | 6,861 | 76.929% (76.487–77.265) | 75.320% (74.870–75.700) | 92.940% (92.669–93.079) | 88.703% (88.535–88.817) |
+| 50% | 23,085 / 46,170 | 242,767 | 7,221 | 76.450% (76.123–76.728) | 74.847% (74.507–75.156) | 92.520% (92.212–92.682) | 86.369% (86.304–86.552) |
+| 40% | 18,468 / 46,170 | 292,467 | 6,915 | 75.970% (75.530–76.199) | 74.377% (73.904–74.592) | 92.097% (91.700–92.251) | 84.073% (83.852–84.239) |
+| 30% | 13,851 / 46,170 | 344,053 | 6,094 | 75.177% (74.966–75.591) | 73.506% (73.313–73.994) | 91.405% (91.208–91.850) | 81.431% (81.316–81.762) |
+| 20% | 9,234 / 46,170 | 395,630 | 4,643 | 73.983% (73.755–74.765) | 72.279% (72.003–73.162) | 90.627% (90.348–91.119) | 78.388% (78.224–78.965) |
+| 10% | 4,617 / 46,170 | 447,988 | 2,618 | 72.583% (70.889–72.914) | 70.834% (68.965–71.203) | 89.804% (89.258–89.896) | 74.894% (73.426–75.227) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Latvian</strong> — <code>lv-lv-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 7,238 / 7,238 | 0 | 0 | n/a | n/a | n/a | 94.098% |
+| 90% | 6,514 / 7,238 | 7,202 | 850 | 79.567% (77.953–79.911) | 77.881% (76.181–78.385) | 96.183% (95.483–96.621) | 92.272% (92.198–92.332) |
+| 80% | 5,790 / 7,238 | 14,695 | 1,512 | 78.588% (77.419–79.735) | 76.842% (75.548–78.150) | 96.213% (95.973–96.877) | 90.436% (90.241–90.588) |
+| 70% | 5,067 / 7,238 | 22,244 | 1,986 | 77.781% (76.813–78.430) | 76.017% (74.893–76.728) | 95.775% (95.622–96.171) | 88.489% (88.253–88.615) |
+| 60% | 4,343 / 7,238 | 29,791 | 2,255 | 76.641% (76.242–77.432) | 74.797% (74.289–75.627) | 95.613% (95.404–95.948) | 86.316% (86.149–86.592) |
+| 50% | 3,619 / 7,238 | 37,686 | 2,349 | 75.857% (75.500–76.087) | 73.934% (73.553–74.183) | 95.248% (95.113–95.345) | 84.197% (83.988–84.319) |
+| 40% | 2,895 / 7,238 | 45,887 | 2,257 | 74.784% (74.270–75.101) | 72.828% (72.268–73.132) | 94.682% (94.520–95.029) | 81.728% (81.481–82.086) |
+| 30% | 2,171 / 7,238 | 54,169 | 1,991 | 73.419% (72.903–73.876) | 71.372% (70.760–71.872) | 94.029% (93.670–94.516) | 79.030% (78.728–79.433) |
+| 20% | 1,448 / 7,238 | 62,679 | 1,487 | 72.310% (71.337–72.412) | 70.176% (69.052–70.300) | 93.799% (93.797–94.400) | 76.267% (75.497–76.292) |
+| 10% | 724 / 7,238 | 71,294 | 855 | 70.293% (68.353–71.195) | 68.037% (65.975–69.051) | 92.818% (92.118–93.056) | 72.405% (70.742–73.318) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Lingala</strong> — <code>lin-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 57 / 57 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 90% | 51 / 57 | 24 | 0 | 79.167% (70.833–83.333) | 72.222% (61.111–77.778) | 100.000% | 97.826% (96.957–98.261) |
+| 80% | 46 / 57 | 44 | 0 | 79.545% (77.273–86.667) | 72.727% (69.697–82.353) | 100.000% | 96.087% (95.652–97.391) |
+| 70% | 40 / 57 | 69 | 0 | 85.507% (77.941–86.957) | 80.769% (70.588–82.692) | 100.000% | 95.652% (93.478–96.087) |
+| 60% | 34 / 57 | 93 | 0 | 84.043% (80.435–86.022) | 78.873% (73.913–81.429) | 100.000% | 93.478% (92.174–94.348) |
+| 50% | 29 / 57 | 113 | 0 | 82.301% (77.193–86.842) | 77.647% (69.767–82.558) | 100.000% (96.429–100.000) | 91.304% (88.696–93.478) |
+| 40% | 23 / 57 | 138 | 0 | 78.986% (71.014–83.212) | 72.115% (62.500–78.641) | 97.059% (97.059–100.000) | 87.391% (82.609–90.000) |
+| 30% | 17 / 57 | 162 | 0 | 74.534% (70.370–75.776) | 66.942% (60.656–68.595) | 97.500% (97.500–100.000) | 82.174% (79.130–83.043) |
+| 20% | 11 / 57 | 186 | 0 | 69.892% (68.280–71.505) | 60.714% (57.857–62.857) | 100.000% (97.826–100.000) | 75.652% (74.348–76.957) |
+| 10% | 6 / 57 | 206 | 0 | 62.621% (59.709–66.990) | 50.323% (46.452–56.129) | 100.000% (98.039–100.000) | 66.522% (63.913–70.435) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Lithuanian</strong> — <code>lt-lt-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 1,391 / 1,391 | 0 | 0 | n/a | n/a | n/a | 99.624% |
+| 90% | 1,252 / 1,391 | 3,098 | 20 | 52.408% (48.059–60.658) | 50.780% (45.635–59.213) | 92.806% (89.928–94.245) | 94.782% (94.141–95.334) |
+| 80% | 1,113 / 1,391 | 5,924 | 38 | 56.555% (52.829–58.980) | 54.788% (50.959–57.262) | 93.165% (92.446–93.885) | 90.803% (89.472–91.261) |
+| 70% | 974 / 1,391 | 8,816 | 45 | 55.172% (53.005–56.545) | 53.233% (51.076–54.632) | 93.046% (92.326–94.245) | 86.251% (85.230–87.230) |
+| 60% | 835 / 1,391 | 11,423 | 41 | 54.959% (53.373–55.165) | 53.023% (51.393–53.172) | 92.973% (92.266–93.705) | 81.954% (80.964–82.213) |
+| 50% | 696 / 1,391 | 14,406 | 42 | 54.463% (52.847–55.810) | 52.556% (50.779–53.867) | 92.374% (91.655–93.813) | 77.147% (76.819–77.926) |
+| 40% | 556 / 1,391 | 17,342 | 48 | 54.480% (50.707–55.714) | 52.544% (48.617–53.826) | 92.934% (92.695–93.533) | 72.567% (69.857–73.081) |
+| 30% | 417 / 1,391 | 20,211 | 41 | 51.961% (49.300–54.333) | 49.943% (47.092–52.387) | 92.916% (92.189–93.737) | 66.787% (64.518–67.681) |
+| 20% | 278 / 1,391 | 23,192 | 25 | 50.338% (47.680–52.416) | 48.199% (45.387–50.333) | 93.172% (91.105–93.980) | 60.139% (58.049–61.566) |
+| 10% | 139 / 1,391 | 26,154 | 15 | 41.955% (39.757–46.002) | 39.455% (37.165–43.747) | 90.895% (90.415–91.693) | 47.596% (46.017–51.131) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Livonian</strong> — <code>liv-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 201 / 201 | 0 | 0 | n/a | n/a | n/a | 99.409% |
+| 90% | 181 / 201 | 273 | 0 | 67.925% (64.945–75.824) | 65.714% (63.347–73.913) | 94.737% (85.000–100.000) | 96.456% (96.108–97.116) |
+| 80% | 161 / 201 | 562 | 0 | 72.152% (69.603–72.810) | 69.981% (67.904–70.866) | 92.500% (92.308–100.000) | 93.989% (93.294–94.406) |
+| 70% | 141 / 201 | 842 | 5 | 70.441% (65.367–73.134) | 68.590% (63.290–71.007) | 95.000% (94.915–100.000) | 90.723% (88.603–92.078) |
+| 60% | 121 / 201 | 1,128 | 6 | 66.696% (65.000–68.440) | 64.423% (62.685–66.239) | 96.250% (92.405–97.500) | 86.484% (85.302–87.526) |
+| 50% | 101 / 201 | 1,435 | 6 | 66.500% (62.604–66.690) | 64.385% (60.119–64.719) | 94.000% (92.929–96.000) | 82.835% (80.646–83.391) |
+| 40% | 80 / 201 | 1,729 | 11 | 62.558% (57.524–67.989) | 60.510% (54.592–65.926) | 94.215% (90.000–95.833) | 77.137% (74.531–80.056) |
+| 30% | 60 / 201 | 2,013 | 16 | 58.718% (51.079–66.965) | 56.487% (48.138–64.939) | 90.071% (88.571–94.326) | 70.744% (65.566–76.338) |
+| 20% | 40 / 201 | 2,295 | 11 | 57.118% (49.735–62.832) | 54.695% (46.629–60.890) | 90.062% (88.750–90.683) | 65.497% (59.868–69.805) |
+| 10% | 20 / 201 | 2,588 | 11 | 42.835% (33.321–48.370) | 38.905% (29.295–45.200) | 90.556% (87.293–95.580) | 48.158% (39.576–53.405) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Low German</strong> — <code>nds-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 325 / 325 | 0 | 0 | n/a | n/a | n/a | 98.035% |
+| 90% | 293 / 325 | 246 | 10 | 48.770% (38.846–54.545) | 41.509% (31.004–47.964) | 96.875% (93.333–100.000) | 93.202% (91.984–93.517) |
+| 80% | 260 / 325 | 501 | 20 | 49.301% (45.309–54.851) | 42.202% (37.385–48.409) | 98.438% (96.825–98.462) | 88.055% (87.466–89.077) |
+| 70% | 228 / 325 | 754 | 22 | 46.997% (42.308–53.774) | 40.179% (35.258–47.214) | 94.845% (90.625–97.917) | 82.475% (81.139–84.558) |
+| 60% | 195 / 325 | 1,016 | 27 | 44.389% (42.421–49.650) | 37.244% (35.738–43.070) | 93.023% (88.372–96.850) | 76.542% (75.324–78.232) |
+| 50% | 163 / 325 | 1,274 | 25 | 44.861% (42.048–45.997) | 37.751% (35.106–38.744) | 93.125% (88.889–96.855) | 71.434% (69.391–71.827) |
+| 40% | 130 / 325 | 1,526 | 16 | 43.559% (39.043–44.233) | 36.329% (32.077–37.698) | 88.601% (87.629–97.927) | 65.344% (62.043–65.933) |
+| 30% | 98 / 325 | 1,780 | 11 | 39.446% (37.875–42.391) | 32.172% (30.344–35.085) | 90.265% (86.726–96.000) | 56.621% (55.756–59.411) |
+| 20% | 65 / 325 | 2,032 | 11 | 33.560% (32.430–37.451) | 25.889% (23.785–30.095) | 89.535% (86.873–91.506) | 45.776% (45.540–49.470) |
+| 10% | 33 / 325 | 2,289 | 8 | 29.623% (26.394–33.814) | 20.833% (18.603–25.876) | 90.000% (80.069–92.808) | 36.582% (33.045–40.314) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Lower Sorbian</strong> — <code>dsb-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 990 / 990 | 0 | 0 | n/a | n/a | n/a | 98.713% |
+| 90% | 891 / 990 | 1,191 | 35 | 46.264% (45.200–52.122) | 42.009% (41.628–48.914) | 89.691% (82.474–94.792) | 93.572% (93.424–93.809) |
+| 80% | 792 / 990 | 2,394 | 58 | 47.460% (45.155–52.924) | 43.845% (41.610–49.707) | 88.083% (85.128–91.667) | 88.545% (87.947–89.406) |
+| 70% | 693 / 990 | 3,603 | 71 | 46.784% (43.297–49.183) | 43.096% (39.474–45.926) | 86.644% (84.983–89.726) | 83.478% (82.035–83.806) |
+| 60% | 594 / 990 | 4,825 | 83 | 44.684% (42.560–48.259) | 40.956% (38.568–44.901) | 86.957% (86.041–89.286) | 76.984% (76.845–78.460) |
+| 50% | 495 / 990 | 5,987 | 78 | 45.645% (44.147–48.204) | 42.032% (40.470–44.796) | 86.558% (84.929–89.634) | 72.352% (71.786–73.532) |
+| 40% | 396 / 990 | 7,214 | 78 | 44.627% (42.546–45.774) | 40.775% (38.793–42.062) | 87.075% (84.407–88.494) | 66.227% (65.349–67.334) |
+| 30% | 297 / 990 | 8,432 | 62 | 43.759% (42.434–44.378) | 39.792% (38.375–40.579) | 87.391% (84.761–88.646) | 60.339% (59.651–60.815) |
+| 20% | 198 / 990 | 9,687 | 47 | 42.199% (40.525–46.112) | 38.157% (36.675–42.646) | 85.298% (82.532–87.802) | 53.690% (52.329–56.609) |
+| 10% | 99 / 990 | 10,941 | 35 | 40.243% (34.758–42.127) | 36.394% (29.972–38.736) | 83.802% (80.359–88.739) | 46.163% (41.259–48.024) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Luganda</strong> — <code>lg-ug-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 87 / 87 | 0 | 0 | n/a | n/a | n/a | 98.462% |
+| 90% | 78 / 87 | 478 | 12 | 1.883% (1.863–2.079) | 0.000% | 100.000% | 88.580% (88.475–89.528) |
+| 80% | 70 / 87 | 902 | 13 | 1.885% (1.872–1.947) | 0.000% | 100.000% | 79.815% (79.688–80.426) |
+| 70% | 61 / 87 | 1,382 | 25 | 1.881% (1.854–1.929) | 0.000% | 100.000% | 69.890% (69.469–70.607) |
+| 60% | 52 / 87 | 1,846 | 73 | 1.896% (1.871–1.913) | 0.000% | 100.000% | 60.303% (59.777–60.641) |
+| 50% | 44 / 87 | 2,284 | 73 | 1.883% (1.861–1.894) | 0.000% | 100.000% | 51.370% (51.201–51.686) |
+| 40% | 35 / 87 | 2,775 | 61 | 1.874% (1.847–1.885) | 0.000% | 100.000% | 41.361% (41.256–41.445) |
+| 30% | 26 / 87 | 3,298 | 24 | 1.850% (1.835–1.869) | 0.000% | 100.000% | 31.247% (31.121–31.290) |
+| 20% | 17 / 87 | 3,815 | 0 | 1.835% (1.834–1.862) | 0.000% | 100.000% | 21.070% (20.965–21.091) |
+| 10% | 9 / 87 | 4,251 | 0 | 1.835% (1.833–1.854) | 0.000% | 100.000% | 11.989% (11.968–12.073) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Macedonian</strong> — <code>mk-mk-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 9,788 / 9,788 | 0 | 0 | n/a | n/a | n/a | 99.414% |
+| 90% | 8,809 / 9,788 | 13,423 | 164 | 80.917% (79.930–82.055) | 80.215% (79.079–81.339) | 91.014% (89.597–91.605) | 97.523% (97.448–97.623) |
+| 80% | 7,830 / 9,788 | 27,148 | 280 | 81.039% (80.669–81.481) | 80.369% (79.904–80.929) | 90.609% (88.743–91.266) | 95.655% (95.619–95.783) |
+| 70% | 6,852 / 9,788 | 40,859 | 389 | 80.732% (80.307–81.301) | 80.051% (79.540–80.702) | 89.812% (89.129–90.549) | 93.678% (93.530–93.925) |
+| 60% | 5,873 / 9,788 | 54,136 | 436 | 80.283% (79.406–80.775) | 79.588% (78.599–80.107) | 89.518% (89.353–90.080) | 91.712% (91.266–91.864) |
+| 50% | 4,894 / 9,788 | 67,687 | 435 | 79.484% (79.052–79.861) | 78.686% (78.227–79.128) | 89.823% (89.387–89.959) | 89.381% (89.070–89.582) |
+| 40% | 3,915 / 9,788 | 81,525 | 413 | 78.965% (77.747–79.261) | 78.230% (76.855–78.519) | 89.389% (88.553–89.843) | 87.073% (86.277–87.204) |
+| 30% | 2,936 / 9,788 | 95,378 | 360 | 77.935% (76.280–78.779) | 77.056% (75.315–78.031) | 88.763% (88.263–89.409) | 84.248% (83.133–84.873) |
+| 20% | 1,958 / 9,788 | 109,039 | 265 | 75.934% (74.893–77.333) | 75.052% (73.871–76.479) | 88.214% (87.347–88.560) | 80.633% (79.675–81.699) |
+| 10% | 979 / 9,788 | 122,831 | 127 | 72.790% (72.183–75.657) | 71.795% (71.133–74.901) | 85.672% (85.477–87.760) | 75.465% (74.818–77.963) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Magahi</strong> — <code>mag-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 938 / 938 | 0 | 0 | n/a | n/a | n/a | 99.451% |
+| 90% | 844 / 938 | 142 | 1 | 71.831% (69.427–78.125) | 26.562% (22.449–34.286) | 97.849% (94.624–98.925) | 96.703% (96.154–97.527) |
+| 80% | 750 / 938 | 301 | 3 | 69.204% (66.369–83.333) | 24.561% (19.608–33.962) | 97.326% (96.257–98.930) | 93.201% (91.690–96.703) |
+| 70% | 657 / 938 | 450 | 4 | 69.333% (65.574–74.265) | 21.154% (18.248–22.656) | 97.857% (96.786–98.571) | 89.973% (88.049–92.239) |
+| 60% | 563 / 938 | 584 | 4 | 69.692% (67.203–74.259) | 20.565% (18.957–21.687) | 97.594% (96.247–98.391) | 87.294% (85.577–89.904) |
+| 50% | 469 / 938 | 713 | 4 | 70.407% (67.624–72.957) | 20.000% (16.290–22.819) | 96.154% (95.931–97.430) | 84.959% (82.555–86.951) |
+| 40% | 375 / 938 | 863 | 5 | 69.177% (66.447–74.055) | 20.601% (18.991–22.837) | 95.907% (94.643–98.046) | 81.319% (78.709–85.302) |
+| 30% | 281 / 938 | 1,013 | 4 | 69.200% (66.350–71.310) | 19.218% (16.573–20.168) | 95.725% (94.656–97.717) | 78.297% (75.412–80.563) |
+| 20% | 188 / 938 | 1,142 | 3 | 68.860% (67.075–71.074) | 16.552% (14.504–20.370) | 96.800% (94.660–96.933) | 75.343% (73.283–77.335) |
+| 10% | 94 / 938 | 1,297 | 1 | 67.896% (66.492–69.011) | 17.031% (12.373–19.604) | 95.498% (94.187–98.104) | 71.223% (69.231–72.253) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Malagasy</strong> — <code>mg-mg-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 159 / 159 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 90% | 143 / 159 | 64 | 0 | 25.000% | 0.000% | 100.000% | 92.453% |
+| 80% | 127 / 159 | 128 | 0 | 25.000% | 0.000% | 100.000% | 84.906% |
+| 70% | 111 / 159 | 192 | 0 | 25.000% | 0.000% | 100.000% | 77.358% |
+| 60% | 95 / 159 | 256 | 0 | 25.000% | 0.000% | 100.000% | 69.811% |
+| 50% | 80 / 159 | 316 | 0 | 25.000% | 0.000% | 100.000% | 62.736% |
+| 40% | 64 / 159 | 380 | 0 | 25.000% | 0.000% | 100.000% | 55.189% |
+| 30% | 48 / 159 | 444 | 0 | 25.000% | 0.000% | 100.000% | 47.642% |
+| 20% | 32 / 159 | 508 | 0 | 25.000% | 0.000% | 100.000% | 40.094% |
+| 10% | 16 / 159 | 572 | 0 | 25.000% | 0.000% | 100.000% | 32.547% |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Maltese</strong> — <code>mt-mt-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 112 / 112 | 0 | 0 | n/a | n/a | n/a | 99.400% |
+| 90% | 101 / 112 | 144 | 0 | 16.667% (8.333–21.333) | 10.072% (1.493–15.108) | 100.000% (90.909–100.000) | 91.333% (90.600–92.133) |
+| 80% | 90 / 112 | 295 | 0 | 13.851% (11.702–15.333) | 7.299% (4.981–8.633) | 95.455% (95.238–100.000) | 82.533% (82.400–82.800) |
+| 70% | 78 / 112 | 454 | 1 | 11.184% (10.538–14.097) | 4.662% (3.632–7.143) | 96.970% (91.176–100.000) | 72.800% (71.933–74.000) |
+| 60% | 67 / 112 | 592 | 8 | 13.007% (10.697–14.480) | 6.093% (4.021–7.749) | 95.556% (95.455–100.000) | 65.067% (62.667–66.000) |
+| 50% | 56 / 112 | 737 | 8 | 12.278% (11.564–13.067) | 5.579% (4.965–6.110) | 96.364% (90.909–100.000) | 56.533% (54.533–57.333) |
+| 40% | 45 / 112 | 887 | 8 | 11.939% (11.387–12.059) | 5.153% (4.390–5.370) | 96.970% (92.424–97.015) | 47.000% (45.800–47.933) |
+| 30% | 34 / 112 | 1,038 | 8 | 12.694% (11.542–13.572) | 5.765% (4.932–7.121) | 94.872% (92.308–97.436) | 38.600% (38.267–39.600) |
+| 20% | 22 / 112 | 1,199 | 8 | 12.010% (10.461–13.756) | 5.140% (3.648–7.918) | 95.556% (86.667–100.000) | 29.400% (27.533–29.667) |
+| 10% | 11 / 112 | 1,351 | 0 | 12.435% (8.754–12.702) | 5.578% (1.363–6.027) | 98.020% (88.119–100.000) | 20.733% (18.000–21.533) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Manx</strong> — <code>gv-im-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 1 / 1 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 90% | 1 / 1 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 80% | 1 / 1 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 70% | 1 / 1 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 60% | 1 / 1 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 50% | 1 / 1 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 40% | 1 / 1 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 30% | 1 / 1 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 20% | 1 / 1 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 10% | 1 / 1 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Maori</strong> — <code>mi-nz-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 104 / 104 | 0 | 0 | n/a | n/a | n/a | 99.519% |
+| 90% | 94 / 104 | 20 | 0 | 70.000% (65.000–75.000) | 40.000% (30.000–50.000) | 100.000% (90.000–100.000) | 96.635% (96.154–97.115) |
+| 80% | 83 / 104 | 41 | 1 | 68.293% (58.537–76.190) | 40.000% (20.000–52.381) | 95.238% (95.238–100.000) | 93.269% (91.346–95.192) |
+| 70% | 73 / 104 | 61 | 1 | 67.213% (62.903–75.410) | 33.333% (25.806–50.000) | 100.000% (96.774–100.000) | 89.904% (88.942–92.308) |
+| 60% | 62 / 104 | 83 | 1 | 68.675% (63.095–71.084) | 36.585% (28.571–41.463) | 97.619% (97.619–100.000) | 87.019% (85.096–87.981) |
+| 50% | 52 / 104 | 103 | 1 | 67.308% (66.346–72.816) | 36.538% (33.333–47.059) | 98.077% (96.154–100.000) | 83.654% (83.173–86.058) |
+| 40% | 42 / 104 | 124 | 0 | 70.161% (66.935–79.675) | 41.935% (33.871–60.656) | 98.387% (96.774–100.000) | 82.212% (80.288–87.500) |
+| 30% | 31 / 104 | 146 | 0 | 67.123% (64.384–77.931) | 34.247% (30.137–58.333) | 98.630% (94.521–100.000) | 76.923% (75.000–84.135) |
+| 20% | 21 / 104 | 166 | 0 | 67.470% (62.048–75.904) | 39.759% (25.301–54.217) | 97.590% (95.181–98.795) | 74.038% (69.712–80.769) |
+| 10% | 10 / 104 | 188 | 0 | 70.213% (57.447–75.000) | 41.489% (17.021–52.128) | 97.872% (97.872–98.936) | 73.077% (61.538–77.404) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Mapudungun</strong> — <code>arn-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 26 / 26 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 90% | 23 / 26 | 63 | 0 | 46.032% (33.333–88.889) | 43.333% (30.000–88.333) | 100.000% | 93.796% (92.336–98.723) |
+| 80% | 21 / 26 | 105 | 0 | 48.571% (33.962–73.333) | 46.000% (30.693–72.000) | 100.000% | 90.146% (87.226–94.891) |
+| 70% | 18 / 26 | 169 | 0 | 59.172% (43.787–69.643) | 57.143% (40.994–68.125) | 100.000% | 87.409% (82.664–90.693) |
+| 60% | 16 / 26 | 211 | 0 | 56.398% (51.659–66.351) | 54.229% (49.254–64.677) | 100.000% | 83.212% (81.387–87.044) |
+| 50% | 13 / 26 | 275 | 0 | 58.909% (48.175–67.636) | 56.870% (45.594–66.031) | 100.000% | 79.380% (74.088–83.759) |
+| 40% | 10 / 26 | 338 | 0 | 55.621% (49.704–67.160) | 53.583% (47.205–65.528) | 100.000% (93.750–100.000) | 72.628% (68.978–79.745) |
+| 30% | 8 / 26 | 380 | 0 | 53.421% (32.368–62.105) | 51.105% (29.006–60.221) | 100.000% (94.444–100.000) | 67.701% (53.102–73.723) |
+| 20% | 5 / 26 | 443 | 0 | 40.858% (35.892–58.916) | 37.915% (32.701–56.872) | 100.000% | 52.190% (48.175–66.788) |
+| 10% | 3 / 26 | 485 | 0 | 37.320% (25.361–54.227) | 34.199% (21.645–51.948) | 100.000% | 44.526% (33.942–59.489) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Middle French</strong> — <code>frm-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 603 / 603 | 0 | 0 | n/a | n/a | n/a | 99.224% |
+| 90% | 543 / 603 | 2,680 | 24 | 80.852% (77.393–87.793) | 80.409% (76.870–87.519) | 100.000% | 97.364% (97.075–98.005) |
+| 80% | 482 / 603 | 5,421 | 54 | 79.722% (78.325–83.290) | 79.283% (77.824–82.909) | 100.000% (99.174–100.000) | 95.160% (95.017–95.954) |
+| 70% | 422 / 603 | 8,121 | 79 | 78.997% (76.000–83.088) | 78.533% (75.466–82.715) | 99.448% (99.448–100.000) | 92.952% (92.242–94.234) |
+| 60% | 362 / 603 | 10,807 | 114 | 77.429% (75.058–81.635) | 76.913% (74.495–81.226) | 99.585% (99.585–100.000) | 90.368% (89.569–92.000) |
+| 50% | 302 / 603 | 13,506 | 126 | 78.703% (76.302–79.765) | 78.235% (75.765–79.311) | 99.668% (99.336–99.668) | 88.709% (87.662–89.445) |
+| 40% | 241 / 603 | 16,254 | 122 | 77.925% (76.130–80.706) | 77.429% (75.615–80.280) | 99.448% (97.514–99.724) | 86.318% (85.147–87.981) |
+| 30% | 181 / 603 | 18,971 | 96 | 77.220% (72.026–80.620) | 76.750% (71.416–80.192) | 99.526% (97.867–99.763) | 83.748% (80.113–86.223) |
+| 20% | 121 / 603 | 21,738 | 72 | 76.858% (71.690–78.493) | 76.334% (71.072–78.018) | 99.378% (98.133–100.000) | 81.277% (77.100–82.482) |
+| 10% | 60 / 603 | 24,561 | 13 | 75.715% (67.286–79.763) | 75.188% (66.585–79.341) | 98.343% (98.343–100.000) | 78.117% (70.546–81.511) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Middle High German</strong> — <code>gmh-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 29 / 29 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 90% | 26 / 29 | 40 | 0 | 52.500% (6.000–84.000) | 48.649% (0.000–81.818) | 100.000% | 95.052% (87.760–98.958) |
+| 80% | 23 / 29 | 84 | 0 | 60.870% (29.167–75.000) | 58.140% (22.727–73.077) | 100.000% (83.333–100.000) | 90.625% (86.719–94.531) |
+| 70% | 20 / 29 | 119 | 0 | 51.376% (40.146–74.545) | 47.000% (35.938–72.277) | 100.000% (88.889–100.000) | 86.198% (78.646–92.708) |
+| 60% | 17 / 29 | 157 | 0 | 37.736% (16.340–60.000) | 32.653% (12.057–56.391) | 100.000% (66.667–100.000) | 74.219% (64.323–84.896) |
+| 50% | 15 / 29 | 190 | 0 | 33.333% (16.959–43.455) | 27.027% (12.102–38.983) | 100.000% (71.429–100.000) | 69.271% (58.594–71.875) |
+| 40% | 12 / 29 | 230 | 0 | 34.634% (20.796–51.082) | 28.723% (16.746–47.196) | 100.000% (70.588–100.000) | 63.542% (53.385–70.573) |
+| 30% | 9 / 29 | 270 | 0 | 27.737% (20.641–38.148) | 24.016% (15.041–33.200) | 95.000% (75.000–100.000) | 48.438% (41.927–57.031) |
+| 20% | 6 / 29 | 307 | 0 | 19.544% (10.280–24.912) | 14.437% (4.362–18.321) | 95.652% (82.609–100.000) | 35.677% (25.000–44.271) |
+| 10% | 3 / 29 | 347 | 0 | 16.954% (6.761–23.631) | 10.248% (0.608–17.445) | 100.000% (84.615–100.000) | 27.344% (13.802–33.594) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Middle Low German</strong> — <code>gml-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 54 / 54 | 0 | 0 | n/a | n/a | n/a | 98.268% |
+| 90% | 49 / 54 | 68 | 0 | 54.717% (23.529–77.778) | 50.000% (17.460–76.119) | 100.000% | 93.543% (90.079–95.748) |
+| 80% | 43 / 54 | 135 | 5 | 52.778% (38.519–79.710) | 48.872% (33.065–78.125) | 100.000% (90.909–100.000) | 87.559% (85.197–93.858) |
+| 70% | 38 / 54 | 187 | 5 | 51.429% (28.877–66.495) | 47.423% (23.392–63.483) | 100.000% (87.500–100.000) | 82.205% (78.110–88.189) |
+| 60% | 32 / 54 | 249 | 5 | 44.726% (38.824–60.241) | 39.535% (33.906–58.150) | 90.909% (77.273–95.455) | 77.638% (73.701–82.835) |
+| 50% | 27 / 54 | 310 | 5 | 44.194% (35.938–50.489) | 40.283% (30.375–47.143) | 96.296% (85.185–96.296) | 71.181% (67.559–75.591) |
+| 40% | 22 / 54 | 372 | 5 | 42.473% (26.250–48.943) | 37.941% (20.652–44.242) | 90.625% (87.097–96.875) | 64.724% (53.386–71.811) |
+| 30% | 16 / 54 | 443 | 1 | 38.600% (19.136–45.455) | 33.744% (13.839–41.732) | 84.211% (81.579–92.105) | 55.591% (37.953–62.520) |
+| 20% | 11 / 54 | 510 | 1 | 32.941% (19.378–35.903) | 27.778% (13.095–32.589) | 90.476% (69.767–95.349) | 45.354% (30.551–49.606) |
+| 10% | 5 / 54 | 580 | 0 | 27.692% (9.866–30.690) | 22.761% (3.097–25.612) | 81.633% (75.510–85.714) | 33.386% (15.118–36.693) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Mongolian</strong> — <code>mn-mn-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 2,140 / 2,140 | 0 | 0 | n/a | n/a | n/a | 97.942% |
+| 90% | 1,926 / 2,140 | 1,714 | 56 | 76.453% (72.769–80.955) | 73.677% (69.631–78.721) | 95.755% (93.839–96.682) | 95.691% (95.311–96.146) |
+| 80% | 1,712 / 2,140 | 3,409 | 118 | 75.666% (74.572–77.183) | 72.905% (71.589–74.456) | 95.735% (94.258–96.690) | 93.287% (93.009–93.583) |
+| 70% | 1,498 / 2,140 | 5,132 | 173 | 74.603% (73.765–76.228) | 71.662% (70.738–73.587) | 95.404% (93.968–95.741) | 90.678% (90.388–91.156) |
+| 60% | 1,284 / 2,140 | 6,853 | 203 | 73.499% (73.147–74.946) | 70.647% (70.032–72.323) | 94.104% (93.832–95.171) | 87.853% (87.654–88.439) |
+| 50% | 1,070 / 2,140 | 8,623 | 185 | 72.898% (71.936–73.941) | 70.038% (68.929–71.161) | 93.585% (93.302–94.821) | 85.159% (84.698–85.528) |
+| 40% | 856 / 2,140 | 10,368 | 176 | 72.162% (71.020–73.814) | 69.196% (67.738–71.116) | 93.087% (92.543–94.362) | 82.118% (81.430–83.175) |
+| 30% | 642 / 2,140 | 12,165 | 150 | 70.675% (69.396–72.765) | 67.631% (65.971–69.943) | 92.632% (92.550–93.830) | 78.793% (77.633–80.299) |
+| 20% | 428 / 2,140 | 13,953 | 142 | 69.689% (66.518–70.499) | 66.477% (62.702–67.386) | 92.907% (92.443–93.783) | 75.405% (72.540–75.678) |
+| 10% | 214 / 2,140 | 15,767 | 54 | 65.937% (64.789–68.812) | 62.041% (60.864–65.609) | 93.080% (91.150–94.072) | 69.130% (67.805–71.631) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Murrinh-Patha</strong> — <code>mwf-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 29 / 29 | 0 | 0 | n/a | n/a | n/a | 82.222% |
+| 90% | 26 / 29 | 55 | 20 | 3.636% (1.724–5.128) | 0.000% (0.000–2.222) | 100.000% (66.667–100.000) | 74.861% (73.889–77.083) |
+| 80% | 23 / 29 | 116 | 37 | 3.922% (0.833–4.902) | 0.000% (0.000–1.020) | 100.000% (33.333–100.000) | 67.639% (65.694–69.028) |
+| 70% | 20 / 29 | 165 | 55 | 3.553% (2.484–4.575) | 0.000% (0.000–0.680) | 85.714% (66.667–100.000) | 60.417% (56.667–62.500) |
+| 60% | 17 / 29 | 228 | 69 | 3.774% (2.881–4.386) | 0.000% (0.000–0.490) | 87.500% (75.000–90.909) | 54.583% (52.222–55.833) |
+| 50% | 15 / 29 | 271 | 79 | 3.610% (2.951–4.833) | 0.375% (0.000–0.781) | 90.000% (84.615–90.909) | 50.139% (46.250–52.500) |
+| 40% | 12 / 29 | 343 | 77 | 3.499% (3.306–4.545) | 0.000% (0.000–0.633) | 92.308% (92.308–100.000) | 40.139% (39.722–42.917) |
+| 30% | 9 / 29 | 419 | 68 | 3.890% (3.341–4.423) | 0.000% (0.000–0.513) | 94.118% (93.333–100.000) | 32.639% (31.667–35.000) |
+| 20% | 6 / 29 | 519 | 61 | 3.861% (3.238–3.940) | 0.000% (0.000–0.195) | 100.000% (94.444–100.000) | 21.806% (20.000–23.194) |
+| 10% | 3 / 29 | 629 | 31 | 3.975% (3.698–4.114) | 0.000% | 100.000% (95.833–100.000) | 11.806% (11.389–14.861) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Navajo</strong> — <code>nav-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 627 / 627 | 0 | 0 | n/a | n/a | n/a | 97.082% |
+| 90% | 564 / 627 | 1,152 | 48 | 5.690% (5.628–6.813) | 0.310% (0.236–0.456) | 96.825% (96.774–100.000) | 87.537% (87.467–89.638) |
+| 80% | 502 / 627 | 2,137 | 101 | 6.036% (5.573–6.495) | 0.440% (0.186–0.510) | 97.521% (95.200–99.194) | 79.469% (78.300–81.332) |
+| 70% | 439 / 627 | 3,251 | 140 | 5.875% (5.652–6.360) | 0.384% (0.341–0.464) | 97.849% (94.681–98.370) | 70.469% (68.958–72.587) |
+| 60% | 376 / 627 | 4,382 | 155 | 5.728% (5.527–6.041) | 0.367% (0.218–0.416) | 97.561% (95.600–99.190) | 61.426% (60.125–61.839) |
+| 50% | 314 / 627 | 5,560 | 157 | 5.713% (5.586–5.735) | 0.265% (0.228–0.381) | 97.106% (95.820–98.387) | 51.617% (51.389–52.496) |
+| 40% | 251 / 627 | 6,745 | 176 | 5.663% (5.579–5.713) | 0.235% (0.173–0.378) | 96.533% (96.247–99.198) | 42.283% (41.528–42.784) |
+| 30% | 188 / 627 | 7,954 | 145 | 5.577% (5.529–5.746) | 0.213% (0.174–0.267) | 98.165% (97.025–99.315) | 32.967% (32.000–33.143) |
+| 20% | 125 / 627 | 9,161 | 108 | 5.502% (5.392–5.672) | 0.127% (0.069–0.186) | 98.802% (97.809–99.800) | 22.737% (22.517–23.774) |
+| 10% | 63 / 627 | 10,276 | 54 | 5.465% (5.378–5.563) | 0.041% (0.010–0.062) | 99.468% (98.404–99.823) | 13.939% (13.377–15.038) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Neapolitan</strong> — <code>nap-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 40 / 40 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 90% | 36 / 40 | 150 | 0 | 69.677% (46.452–87.838) | 68.874% (45.033–87.500) | 100.000% | 96.860% (94.456–98.798) |
+| 80% | 32 / 40 | 302 | 0 | 72.277% (65.563–90.541) | 71.525% (64.626–90.278) | 100.000% | 94.389% (93.053–98.130) |
+| 70% | 28 / 40 | 450 | 0 | 72.406% (69.111–85.809) | 71.655% (68.265–85.421) | 100.000% | 91.650% (90.715–95.725) |
+| 60% | 24 / 40 | 601 | 0 | 73.579% (65.455–81.940) | 72.852% (64.516–81.443) | 100.000% | 89.446% (86.039–92.786) |
+| 50% | 20 / 40 | 753 | 0 | 72.112% (61.355–76.944) | 71.351% (60.300–76.309) | 100.000% (95.000–100.000) | 85.972% (80.561–88.510) |
+| 40% | 16 / 40 | 901 | 0 | 71.207% (65.338–75.973) | 70.421% (64.391–75.429) | 100.000% (95.833–100.000) | 82.632% (79.092–85.571) |
+| 30% | 12 / 40 | 1,051 | 0 | 69.933% (66.063–76.543) | 69.110% (65.132–76.000) | 100.000% (96.429–100.000) | 78.891% (76.219–83.500) |
+| 20% | 8 / 40 | 1,201 | 0 | 73.439% (56.797–75.187) | 72.797% (55.613–74.679) | 96.875% (93.750–100.000) | 78.691% (65.397–80.094) |
+| 10% | 4 / 40 | 1,349 | 0 | 72.128% (55.902–74.351) | 71.516% (54.691–73.724) | 97.222% (94.444–100.000) | 74.883% (60.321–76.887) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>North Frisian</strong> — <code>frr-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 51 / 51 | 0 | 0 | n/a | n/a | n/a | 95.538% |
+| 90% | 46 / 51 | 37 | 3 | 29.730% (21.739–53.571) | 25.000% (14.286–47.826) | 80.000% (66.667–100.000) | 88.714% (86.089–92.126) |
+| 80% | 41 / 51 | 71 | 7 | 31.818% (22.667–56.923) | 22.222% (13.636–50.909) | 88.889% (62.500–100.000) | 83.727% (80.315–88.976) |
+| 70% | 36 / 51 | 111 | 10 | 30.357% (29.245–48.571) | 23.232% (17.582–41.111) | 92.857% (84.615–100.000) | 76.378% (74.803–82.415) |
+| 60% | 31 / 51 | 142 | 10 | 32.394% (30.263–39.437) | 23.387% (21.053–31.707) | 94.737% (77.778–100.000) | 70.866% (68.766–72.966) |
+| 50% | 26 / 51 | 186 | 12 | 32.447% (26.064–40.909) | 24.324% (15.854–32.895) | 91.667% (86.364–100.000) | 65.092% (60.105–69.029) |
+| 40% | 20 / 51 | 229 | 9 | 33.188% (26.316–38.565) | 24.020% (15.736–31.088) | 93.548% (86.667–100.000) | 56.693% (52.756–61.155) |
+| 30% | 15 / 51 | 264 | 8 | 33.333% (27.138–37.879) | 24.336% (19.313–28.384) | 91.429% (77.778–100.000) | 51.706% (46.719–54.331) |
+| 20% | 10 / 51 | 300 | 6 | 29.934% (28.620–35.333) | 20.833% (18.359–25.769) | 92.500% (87.805–97.500) | 42.520% (41.732–47.769) |
+| 10% | 5 / 51 | 340 | 3 | 23.599% (20.235–27.059) | 15.593% (8.475–18.056) | 91.304% (77.273–95.652) | 30.971% (27.822–34.383) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Northern Sami</strong> — <code>sme-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 2,103 / 2,103 | 0 | 0 | n/a | n/a | n/a | 98.924% |
+| 90% | 1,893 / 2,103 | 5,366 | 96 | 62.095% (56.824–63.831) | 60.992% (55.424–62.783) | 88.177% (86.275–91.584) | 95.248% (94.832–95.562) |
+| 80% | 1,682 / 2,103 | 10,878 | 206 | 60.772% (59.449–62.776) | 59.723% (58.278–61.760) | 87.864% (86.861–88.725) | 91.212% (91.135–91.822) |
+| 70% | 1,472 / 2,103 | 16,554 | 240 | 59.450% (59.165–61.858) | 58.241% (58.093–60.868) | 87.379% (86.795–89.756) | 87.190% (86.708–87.508) |
+| 60% | 1,262 / 2,103 | 22,029 | 296 | 58.137% (56.674–60.620) | 56.985% (55.400–59.630) | 88.043% (86.061–88.429) | 82.496% (82.102–83.281) |
+| 50% | 1,052 / 2,103 | 27,761 | 290 | 57.337% (56.084–58.632) | 56.116% (54.859–57.504) | 87.838% (86.847–88.406) | 77.949% (77.318–78.309) |
+| 40% | 841 / 2,103 | 33,181 | 254 | 55.050% (53.662–55.707) | 53.884% (52.362–54.500) | 86.597% (84.603–86.960) | 72.506% (71.347–72.835) |
+| 30% | 631 / 2,103 | 38,585 | 255 | 52.142% (50.575–52.662) | 50.800% (49.298–51.367) | 85.332% (82.841–86.342) | 65.539% (65.048–66.515) |
+| 20% | 421 / 2,103 | 43,995 | 213 | 48.326% (46.637–49.281) | 46.944% (45.145–47.929) | 83.962% (80.419–84.661) | 58.299% (56.852–58.462) |
+| 10% | 210 / 2,103 | 49,599 | 117 | 38.261% (37.210–41.489) | 36.509% (35.569–40.044) | 78.674% (73.181–82.160) | 44.500% (43.291–47.088) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Norwegian Bokmål</strong> — <code>nb-no-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 17,929 / 17,929 | 0 | 0 | n/a | n/a | n/a | 96.852% |
+| 90% | 16,136 / 17,929 | 8,682 | 459 | 79.616% (79.365–80.300) | 70.546% (70.229–71.560) | 95.510% (95.252–96.117) | 95.003% (94.958–95.075) |
+| 80% | 14,343 / 17,929 | 17,357 | 841 | 79.282% (78.391–79.483) | 69.935% (68.919–70.124) | 95.430% (94.759–95.606) | 93.090% (92.952–93.176) |
+| 70% | 12,550 / 17,929 | 26,235 | 1,056 | 78.616% (78.247–79.117) | 69.080% (68.975–69.623) | 95.083% (94.075–95.571) | 91.051% (91.022–91.248) |
+| 60% | 10,757 / 17,929 | 35,053 | 1,222 | 77.888% (77.692–78.230) | 68.373% (67.934–68.693) | 94.544% (94.086–94.654) | 88.955% (88.860–89.131) |
+| 50% | 8,965 / 17,929 | 44,080 | 1,264 | 77.213% (76.853–77.697) | 67.372% (67.269–68.071) | 94.168% (93.324–94.348) | 86.679% (86.578–86.965) |
+| 40% | 7,172 / 17,929 | 53,261 | 1,266 | 76.064% (75.979–76.622) | 66.133% (65.907–66.950) | 93.222% (92.859–93.532) | 84.095% (83.980–84.407) |
+| 30% | 5,379 / 17,929 | 62,496 | 1,103 | 74.899% (74.255–75.323) | 64.655% (63.730–65.508) | 92.387% (92.185–92.553) | 81.227% (80.765–81.536) |
+| 20% | 3,586 / 17,929 | 71,829 | 816 | 72.851% (72.564–73.423) | 62.099% (61.353–62.757) | 91.669% (91.337–91.848) | 77.536% (77.244–77.940) |
+| 10% | 1,793 / 17,929 | 81,276 | 458 | 69.855% (69.368–70.692) | 58.598% (57.048–59.496) | 89.971% (89.224–90.585) | 72.437% (72.066–73.258) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Norwegian Nynorsk</strong> — <code>nn-no-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 4,688 / 4,688 | 0 | 0 | n/a | n/a | n/a | 93.089% |
+| 90% | 4,219 / 4,688 | 1,714 | 230 | 78.608% (77.071–80.395) | 70.338% (68.194–72.339) | 95.153% (94.473–97.054) | 91.278% (91.145–91.385) |
+| 80% | 3,750 / 4,688 | 3,536 | 415 | 77.555% (74.632–78.831) | 68.830% (64.861–70.289) | 95.308% (94.801–95.734) | 89.242% (88.866–89.563) |
+| 70% | 3,282 / 4,688 | 5,324 | 569 | 76.278% (73.779–76.665) | 67.482% (64.436–68.148) | 94.306% (93.322–94.907) | 87.054% (86.449–87.247) |
+| 60% | 2,813 / 4,688 | 7,179 | 679 | 74.537% (72.582–75.521) | 65.383% (63.050–67.121) | 93.672% (92.678–95.267) | 84.601% (83.970–84.998) |
+| 50% | 2,344 / 4,688 | 9,118 | 713 | 72.900% (71.989–74.007) | 62.808% (62.003–65.179) | 93.120% (92.681–94.000) | 81.986% (81.456–82.423) |
+| 40% | 1,875 / 4,688 | 11,136 | 714 | 70.773% (70.483–71.887) | 60.799% (60.313–62.437) | 92.181% (91.901–92.710) | 78.826% (78.688–79.390) |
+| 30% | 1,406 / 4,688 | 13,143 | 632 | 68.697% (68.490–69.926) | 58.151% (57.820–59.696) | 91.536% (91.227–91.890) | 75.436% (75.045–75.894) |
+| 20% | 938 / 4,688 | 15,283 | 459 | 66.030% (65.878–67.801) | 55.037% (54.754–57.088) | 90.572% (89.668–91.189) | 71.121% (70.882–72.220) |
+| 10% | 469 / 4,688 | 17,401 | 298 | 63.195% (60.378–64.426) | 51.191% (46.485–52.220) | 90.470% (89.673–91.396) | 66.017% (63.259–66.984) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Old English</strong> — <code>ang-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 7,146 / 7,146 | 0 | 0 | n/a | n/a | n/a | 96.417% |
+| 90% | 6,431 / 7,146 | 6,338 | 397 | 71.726% (71.108–72.675) | 69.192% (68.500–70.111) | 93.165% (92.411–94.239) | 93.823% (93.705–93.982) |
+| 80% | 5,717 / 7,146 | 12,643 | 736 | 70.859% (70.075–71.520) | 68.234% (67.428–68.905) | 93.121% (92.697–93.426) | 91.026% (90.922–91.244) |
+| 70% | 5,002 / 7,146 | 19,195 | 968 | 69.947% (69.610–70.711) | 67.258% (66.829–68.201) | 92.401% (91.925–92.794) | 88.200% (88.192–88.356) |
+| 60% | 4,288 / 7,146 | 25,758 | 1,187 | 69.249% (68.853–69.679) | 66.496% (66.006–67.028) | 92.057% (91.584–92.640) | 85.211% (85.170–85.412) |
+| 50% | 3,573 / 7,146 | 32,512 | 1,214 | 68.137% (67.024–68.767) | 65.406% (64.127–66.110) | 91.149% (90.475–91.327) | 81.902% (81.562–82.403) |
+| 40% | 2,858 / 7,146 | 39,286 | 1,174 | 66.297% (65.430–67.131) | 63.492% (62.471–64.417) | 90.213% (89.488–90.647) | 77.996% (77.671–78.721) |
+| 30% | 2,144 / 7,146 | 46,184 | 1,026 | 63.972% (63.769–65.295) | 61.002% (60.870–62.615) | 88.711% (87.932–89.145) | 73.499% (73.431–74.477) |
+| 20% | 1,429 / 7,146 | 53,259 | 789 | 61.886% (60.677–62.167) | 58.922% (57.636–59.197) | 86.910% (86.072–87.259) | 68.685% (67.556–68.833) |
+| 10% | 715 / 7,146 | 60,396 | 490 | 56.959% (55.121–57.888) | 53.692% (51.571–54.728) | 84.802% (83.819–85.219) | 60.821% (59.058–61.727) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Old French</strong> — <code>fro-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 1,699 / 1,699 | 0 | 0 | n/a | n/a | n/a | 89.552% |
+| 90% | 1,529 / 1,699 | 8,939 | 1,567 | 62.647% (61.241–69.134) | 61.934% (60.538–68.588) | 98.802% (97.024–99.412) | 86.459% (86.272–86.858) |
+| 80% | 1,359 / 1,699 | 17,509 | 3,630 | 60.379% (59.208–64.509) | 59.639% (58.441–63.859) | 98.516% (97.619–99.410) | 83.326% (82.968–83.734) |
+| 70% | 1,189 / 1,699 | 26,962 | 4,699 | 59.828% (56.031–63.092) | 59.118% (55.187–62.421) | 98.232% (96.844–98.622) | 79.891% (79.344–80.759) |
+| 60% | 1,019 / 1,699 | 36,496 | 5,497 | 58.144% (56.789–60.305) | 57.415% (56.015–59.601) | 98.375% (97.194–98.817) | 76.563% (75.720–76.942) |
+| 50% | 850 / 1,699 | 46,933 | 5,640 | 56.933% (55.066–58.912) | 56.175% (54.285–58.190) | 98.109% (97.041–99.052) | 73.089% (71.720–73.472) |
+| 40% | 680 / 1,699 | 57,657 | 5,275 | 55.738% (54.333–57.653) | 54.971% (53.561–56.913) | 98.223% (97.635–98.912) | 69.036% (67.836–69.346) |
+| 30% | 510 / 1,699 | 68,813 | 4,752 | 54.061% (51.720–56.253) | 53.283% (50.899–55.532) | 97.980% (96.967–98.730) | 63.876% (62.378–65.923) |
+| 20% | 340 / 1,699 | 80,446 | 3,828 | 50.496% (48.800–53.073) | 49.689% (47.945–52.332) | 98.007% (96.174–98.817) | 57.647% (56.616–60.584) |
+| 10% | 170 / 1,699 | 92,661 | 2,100 | 43.123% (41.897–48.241) | 42.195% (40.956–47.457) | 98.103% (95.095–98.168) | 47.642% (46.457–52.655) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Old High German</strong> — <code>goh-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 482 / 482 | 0 | 0 | n/a | n/a | n/a | 99.377% |
+| 90% | 434 / 482 | 548 | 8 | 43.163% (40.855–53.467) | 38.998% (36.499–49.400) | 91.667% (89.362–95.833) | 93.868% (92.849–94.566) |
+| 80% | 386 / 482 | 1,161 | 20 | 48.079% (43.997–52.810) | 44.353% (39.875–49.467) | 91.667% (86.170–92.708) | 88.283% (86.528–90.189) |
+| 70% | 337 / 482 | 1,658 | 13 | 47.587% (44.229–51.365) | 43.536% (40.100–47.311) | 90.278% (85.915–93.007) | 83.094% (81.264–84.642) |
+| 60% | 289 / 482 | 2,220 | 15 | 45.991% (43.664–47.752) | 42.033% (39.426–44.179) | 89.062% (84.127–90.576) | 77.321% (75.774–78.774) |
+| 50% | 241 / 482 | 2,758 | 10 | 46.240% (44.337–47.186) | 42.241% (40.243–43.062) | 88.235% (86.555–89.167) | 72.019% (70.264–73.057) |
+| 40% | 193 / 482 | 3,264 | 11 | 46.067% (42.737–47.243) | 41.818% (38.396–43.265) | 88.153% (87.500–88.502) | 67.302% (63.906–68.094) |
+| 30% | 145 / 482 | 3,773 | 9 | 44.719% (41.114–45.826) | 40.588% (36.563–41.739) | 87.240% (86.228–89.286) | 60.208% (58.094–62.906) |
+| 20% | 96 / 482 | 4,265 | 8 | 43.312% (38.656–44.003) | 39.203% (33.865–39.589) | 87.013% (84.715–87.728) | 54.189% (50.434–55.679) |
+| 10% | 48 / 482 | 4,764 | 3 | 39.216% (32.200–41.498) | 34.440% (26.628–37.096) | 85.714% (82.949–87.788) | 45.170% (38.962–47.094) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Old Irish</strong> — <code>sga-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 49 / 49 | 0 | 0 | n/a | n/a | n/a | 99.786% |
+| 90% | 44 / 49 | 89 | 0 | 6.173% (3.356–11.236) | 1.316% (0.000–7.143) | 80.000% | 91.346% (84.402–93.697) |
+| 80% | 39 / 49 | 192 | 1 | 6.618% (5.189–8.854) | 1.554% (0.794–4.396) | 80.000% (80.000–90.000) | 81.090% (78.312–86.218) |
+| 70% | 34 / 49 | 260 | 1 | 8.032% (6.050–9.554) | 2.991% (0.562–5.686) | 86.667% (80.000–93.333) | 74.359% (69.444–80.556) |
+| 60% | 29 / 49 | 330 | 1 | 6.536% (6.140–7.309) | 1.423% (1.049–1.835) | 95.000% (85.000–100.000) | 66.987% (54.060–69.979) |
+| 50% | 25 / 49 | 434 | 1 | 6.250% (5.069–6.849) | 1.336% (0.732–1.760) | 83.333% (79.167–95.833) | 56.090% (44.551–63.462) |
+| 40% | 20 / 49 | 520 | 1 | 6.318% (4.944–8.269) | 1.590% (0.836–3.666) | 86.207% (79.310–89.655) | 48.825% (36.218–53.953) |
+| 30% | 15 / 49 | 587 | 1 | 6.228% (5.095–8.589) | 1.670% (0.306–4.531) | 82.353% (70.588–97.059) | 41.987% (30.235–43.697) |
+| 20% | 10 / 49 | 736 | 0 | 7.473% (5.263–8.431) | 3.443% (0.395–4.098) | 87.179% (69.231–100.000) | 27.244% (19.231–35.043) |
+| 10% | 5 / 49 | 819 | 0 | 6.716% (4.612–8.036) | 1.935% (0.355–3.919) | 86.364% (77.273–90.909) | 18.269% (9.402–22.863) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Old Norse</strong> — <code>non-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 2,666 / 2,666 | 0 | 0 | n/a | n/a | n/a | 92.899% |
+| 90% | 2,399 / 2,666 | 4,348 | 637 | 46.596% (44.961–49.988) | 44.439% (42.738–47.839) | 89.450% (88.679–93.665) | 88.273% (88.019–88.698) |
+| 80% | 2,133 / 2,666 | 8,876 | 1,179 | 45.910% (45.779–46.884) | 43.900% (43.617–44.722) | 87.991% (85.550–90.092) | 83.431% (83.248–83.893) |
+| 70% | 1,866 / 2,666 | 13,327 | 1,498 | 44.786% (44.147–46.037) | 42.590% (41.780–43.909) | 87.068% (85.843–88.872) | 78.570% (78.273–79.195) |
+| 60% | 1,600 / 2,666 | 18,126 | 1,734 | 43.529% (42.618–45.647) | 41.354% (40.298–43.612) | 85.998% (83.721–87.459) | 73.250% (73.057–74.345) |
+| 50% | 1,333 / 2,666 | 23,125 | 1,846 | 42.483% (41.700–43.802) | 40.146% (39.393–41.689) | 84.820% (82.549–86.364) | 68.187% (67.085–68.802) |
+| 40% | 1,066 / 2,666 | 28,118 | 1,815 | 41.538% (40.030–42.327) | 39.194% (37.653–40.129) | 84.397% (82.867–85.875) | 61.628% (61.346–62.850) |
+| 30% | 800 / 2,666 | 33,516 | 1,598 | 40.345% (38.141–40.542) | 38.013% (35.756–38.168) | 82.751% (81.680–84.687) | 55.839% (55.158–56.627) |
+| 20% | 533 / 2,666 | 38,752 | 1,220 | 38.020% (35.648–38.940) | 35.608% (33.167–36.554) | 82.289% (80.188–83.094) | 48.623% (47.589–50.274) |
+| 10% | 267 / 2,666 | 44,071 | 703 | 34.624% (33.189–35.549) | 32.147% (30.466–33.161) | 80.128% (77.687–81.657) | 40.389% (39.702–41.976) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Old Saxon</strong> — <code>osx-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 863 / 863 | 0 | 0 | n/a | n/a | n/a | 98.470% |
+| 90% | 777 / 863 | 1,168 | 33 | 53.448% (46.839–53.853) | 50.220% (43.487–50.693) | 94.118% (91.139–97.590) | 93.930% (93.715–94.201) |
+| 80% | 690 / 863 | 2,407 | 56 | 54.343% (47.071–56.514) | 51.569% (43.605–53.827) | 92.442% (91.765–95.152) | 89.405% (88.329–90.265) |
+| 70% | 604 / 863 | 3,616 | 80 | 52.431% (49.034–55.515) | 49.493% (45.697–52.512) | 93.333% (90.980–94.902) | 84.601% (83.852–85.820) |
+| 60% | 518 / 863 | 4,808 | 92 | 48.605% (45.812–53.078) | 45.007% (42.048–50.067) | 93.491% (92.647–96.481) | 78.985% (77.448–80.738) |
+| 50% | 432 / 863 | 6,086 | 97 | 48.011% (45.939–52.136) | 44.683% (42.393–49.134) | 92.453% (91.962–94.588) | 73.927% (72.023–75.663) |
+| 40% | 345 / 863 | 7,379 | 89 | 44.655% (42.126–46.890) | 41.008% (38.397–43.485) | 93.177% (90.607–94.347) | 66.566% (64.415–67.737) |
+| 30% | 259 / 863 | 8,715 | 84 | 40.138% (38.413–41.793) | 36.122% (34.265–38.163) | 94.649% (90.970–95.826) | 57.628% (56.003–58.854) |
+| 20% | 173 / 863 | 10,026 | 64 | 35.174% (31.506–36.436) | 30.773% (26.789–32.407) | 93.851% (90.497–96.070) | 47.758% (44.436–49.502) |
+| 10% | 86 / 863 | 11,355 | 33 | 26.441% (23.697–30.817) | 21.522% (18.468–26.220) | 93.264% (91.495–95.484) | 33.124% (30.383–37.617) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Oodham</strong> — <code>ood-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 368 / 368 | 0 | 0 | n/a | n/a | n/a | 98.387% |
+| 90% | 331 / 368 | 109 | 1 | 34.862% (31.746–41.748) | 9.722% (8.989–12.500) | 91.667% (83.784–94.595) | 92.360% (91.087–93.294) |
+| 80% | 294 / 368 | 231 | 4 | 35.169% (32.035–38.288) | 8.025% (6.329–12.865) | 87.838% (86.486–94.595) | 85.569% (84.975–86.757) |
+| 70% | 258 / 368 | 338 | 6 | 36.152% (34.320–38.663) | 9.009% (6.987–11.966) | 91.743% (88.991–95.455) | 80.475% (79.711–80.900) |
+| 60% | 221 / 368 | 454 | 9 | 35.759% (33.333–36.585) | 8.418% (7.792–11.148) | 89.726% (89.116–96.575) | 74.024% (71.986–74.448) |
+| 50% | 184 / 368 | 576 | 8 | 36.458% (35.195–38.143) | 8.906% (7.084–13.810) | 93.478% (90.164–95.628) | 67.572% (66.129–69.015) |
+| 40% | 147 / 368 | 705 | 8 | 36.018% (34.468–37.912) | 7.930% (6.834–13.976) | 93.665% (91.364–95.909) | 60.781% (59.508–63.328) |
+| 30% | 110 / 368 | 811 | 6 | 36.059% (33.539–38.509) | 10.182% (7.052–12.409) | 91.440% (89.494–94.163) | 53.480% (52.971–56.876) |
+| 20% | 74 / 368 | 936 | 4 | 35.543% (34.342–37.074) | 10.280% (8.466–11.493) | 92.177% (89.796–95.222) | 48.132% (46.095–51.019) |
+| 10% | 37 / 368 | 1,047 | 3 | 36.381% (32.625–36.682) | 8.616% (5.816–11.549) | 92.727% (89.728–95.770) | 42.530% (39.898–43.803) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Pashto</strong> — <code>ps-af-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 381 / 381 | 0 | 0 | n/a | n/a | n/a | 96.116% |
+| 90% | 343 / 381 | 266 | 11 | 23.030% (21.053–34.300) | 12.915% (10.044–21.176) | 94.595% (89.189–97.368) | 89.262% (87.826–91.678) |
+| 80% | 305 / 381 | 531 | 32 | 25.697% (22.242–27.637) | 13.828% (12.134–15.824) | 91.892% (86.667–95.946) | 83.453% (81.886–84.922) |
+| 70% | 267 / 381 | 770 | 44 | 24.120% (22.934–28.344) | 12.960% (12.500–17.290) | 91.892% (91.071–93.805) | 77.121% (74.543–78.655) |
+| 60% | 229 / 381 | 1,018 | 50 | 25.540% (23.182–27.473) | 13.924% (12.159–16.334) | 91.391% (90.667–96.667) | 71.377% (68.603–72.454) |
+| 50% | 191 / 381 | 1,389 | 60 | 23.055% (21.855–25.714) | 11.750% (11.084–14.019) | 91.579% (89.362–95.213) | 61.097% (59.824–65.796) |
+| 40% | 152 / 381 | 1,737 | 76 | 21.416% (20.849–25.967) | 11.074% (10.473–14.341) | 92.105% (89.381–93.833) | 51.860% (50.065–59.530) |
+| 30% | 114 / 381 | 1,984 | 76 | 21.573% (19.838–25.892) | 10.529% (9.205–14.276) | 93.208% (91.353–93.585) | 45.855% (42.004–53.819) |
+| 20% | 76 / 381 | 2,324 | 80 | 20.654% (19.205–21.550) | 10.045% (8.245–10.884) | 92.409% (90.789–94.408) | 36.521% (34.040–41.090) |
+| 10% | 38 / 381 | 2,645 | 88 | 21.190% (18.601–31.194) | 10.807% (7.425–22.136) | 92.690% (88.889–94.737) | 31.070% (26.664–37.565) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Persian</strong> — <code>fa-ir-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 69 / 69 | 0 | 0 | n/a | n/a | n/a | 95.836% |
+| 90% | 62 / 69 | 360 | 29 | 18.699% (15.833–19.469) | 15.493% (13.006–16.000) | 100.000% (85.714–100.000) | 87.878% (87.745–88.594) |
+| 80% | 55 / 69 | 743 | 29 | 15.979% (13.625–19.473) | 12.725% (10.400–16.031) | 100.000% (92.857–100.000) | 79.204% (79.072–81.247) |
+| 70% | 48 / 69 | 1,054 | 87 | 17.998% (15.808–19.631) | 14.544% (12.546–16.211) | 100.000% (95.238–100.000) | 73.077% (71.645–73.899) |
+| 60% | 41 / 69 | 1,474 | 87 | 17.843% (16.847–18.371) | 14.739% (13.572–15.175) | 100.000% (96.429–100.000) | 65.305% (63.740–65.517) |
+| 50% | 35 / 69 | 1,792 | 88 | 17.796% (17.243–18.567) | 14.596% (13.979–15.398) | 100.000% (97.059–100.000) | 58.090% (57.560–59.019) |
+| 40% | 28 / 69 | 2,185 | 87 | 18.078% (16.720–19.128) | 14.979% (13.479–15.918) | 100.000% (97.561–100.000) | 50.186% (49.310–51.114) |
+| 30% | 21 / 69 | 2,604 | 58 | 17.696% (16.285–18.267) | 14.648% (13.104–15.153) | 100.000% (97.917–100.000) | 41.857% (40.981–42.706) |
+| 20% | 14 / 69 | 2,978 | 30 | 18.105% (16.353–18.304) | 15.009% (13.145–15.164) | 100.000% (98.182–100.000) | 34.058% (33.130–34.536) |
+| 10% | 7 / 69 | 3,367 | 29 | 17.987% (16.428–17.998) | 14.842% (13.213–14.887) | 100.000% | 25.995% (24.987–26.180) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Polish</strong> — <code>pl-pl-unimorph</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 9,990 / 9,990 | 0 | 0 | n/a | n/a | n/a | 98.837% |
+| 90% | 8,991 / 9,990 | 12,963 | 296 | 73.070% (71.833–74.408) | 69.972% (68.686–71.530) | 90.985% (89.948–91.264) | 96.179% (96.039–96.269) |
+| 80% | 7,992 / 9,990 | 25,947 | 560 | 73.298% (71.542–73.560) | 70.344% (68.301–70.589) | 90.225% (89.430–90.594) | 93.619% (93.242–93.666) |
+| 70% | 6,993 / 9,990 | 39,130 | 726 | 72.217% (71.567–72.865) | 69.231% (68.375–70.016) | 89.428% (88.990–89.729) | 90.660% (90.510–90.869) |
+| 60% | 5,994 / 9,990 | 52,228 | 831 | 71.858% (71.219–72.488) | 68.868% (68.033–69.513) | 89.261% (88.574–89.356) | 87.814% (87.665–88.121) |
+| 50% | 4,995 / 9,990 | 65,434 | 856 | 71.160% (70.975–71.928) | 68.066% (67.805–69.051) | 88.517% (88.330–89.078) | 84.847% (84.707–85.211) |
+| 40% | 3,996 / 9,990 | 78,368 | 857 | 70.344% (69.790–71.025) | 67.165% (66.449–67.971) | 88.216% (88.039–88.588) | 81.587% (81.365–81.994) |
+| 30% | 2,997 / 9,990 | 91,501 | 747 | 69.088% (68.299–70.063) | 65.677% (64.723–66.909) | 88.256% (87.960–88.753) | 78.020% (77.407–78.451) |
+| 20% | 1,998 / 9,990 | 105,119 | 565 | 68.340% (67.352–68.619) | 64.876% (63.830–65.097) | 87.827% (87.200–88.611) | 74.371% (73.606–74.469) |
+| 10% | 999 / 9,990 | 118,711 | 291 | 65.763% (65.395–67.001) | 61.996% (61.861–63.283) | 86.633% (85.310–87.965) | 69.126% (68.742–70.110) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Portuguese</strong> — <code>pt-pt-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 4,001 / 4,001 | 0 | 0 | n/a | n/a | n/a | 99.815% |
+| 90% | 3,601 / 4,001 | 21,519 | 66 | 94.620% (93.771–95.072) | 94.412% (93.530–94.882) | 100.000% (99.749–100.000) | 99.279% (99.194–99.326) |
+| 80% | 3,201 / 4,001 | 43,021 | 113 | 94.273% (93.721–94.673) | 94.056% (93.478–94.472) | 100.000% (99.875–100.000) | 98.675% (98.576–98.762) |
+| 70% | 2,801 / 4,001 | 64,495 | 146 | 93.982% (93.731–94.081) | 93.752% (93.489–93.852) | 100.000% (99.917–100.000) | 98.035% (97.952–98.049) |
+| 60% | 2,401 / 4,001 | 85,970 | 216 | 93.987% (93.386–94.179) | 93.758% (93.130–93.957) | 99.938% (99.937–100.000) | 97.430% (97.204–97.509) |
+| 50% | 2,001 / 4,001 | 107,492 | 243 | 93.601% (93.084–94.280) | 93.358% (92.822–94.061) | 99.900% (99.800–100.000) | 96.646% (96.404–97.001) |
+| 40% | 1,600 / 4,001 | 129,092 | 242 | 93.154% (92.758–94.021) | 92.889% (92.480–93.795) | 99.958% (99.875–100.000) | 95.791% (95.526–96.281) |
+| 30% | 1,200 / 4,001 | 150,645 | 201 | 92.584% (91.628–93.106) | 92.297% (91.306–92.854) | 99.893% (99.643–100.000) | 94.717% (94.035–95.078) |
+| 20% | 800 / 4,001 | 172,258 | 154 | 91.775% (90.674–92.113) | 91.460% (90.315–91.809) | 99.938% (99.750–100.000) | 93.379% (92.481–93.616) |
+| 10% | 400 / 4,001 | 193,881 | 56 | 90.423% (88.898–91.005) | 90.053% (88.476–90.666) | 99.833% (99.778–100.000) | 91.313% (89.988–91.878) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Quechua</strong> — <code>que-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 1,003 / 1,003 | 0 | 0 | n/a | n/a | n/a | 98.439% |
+| 90% | 903 / 1,003 | 14,159 | 418 | 87.614% (87.075–89.562) | 87.601% (87.049–89.541) | 90.244% (87.952–92.771) | 97.283% (97.249–97.538) |
+| 80% | 802 / 1,003 | 29,049 | 731 | 88.048% (87.025–88.824) | 88.043% (87.005–88.829) | 90.270% (88.068–90.909) | 96.179% (95.964–96.469) |
+| 70% | 702 / 1,003 | 44,103 | 969 | 87.543% (86.009–87.942) | 87.527% (85.992–87.940) | 88.760% (87.591–90.262) | 94.939% (94.595–95.065) |
+| 60% | 602 / 1,003 | 58,894 | 1,202 | 87.450% (85.650–87.905) | 87.437% (85.628–87.895) | 89.532% (87.805–89.694) | 93.609% (93.174–93.850) |
+| 50% | 502 / 1,003 | 73,322 | 1,194 | 86.869% (84.692–87.841) | 86.860% (84.666–87.839) | 88.428% (88.110–89.760) | 92.251% (91.467–92.953) |
+| 40% | 401 / 1,003 | 89,682 | 1,101 | 85.822% (82.961–86.710) | 85.807% (82.930–86.707) | 88.142% (87.211–88.947) | 90.452% (88.877–91.030) |
+| 30% | 301 / 1,003 | 104,007 | 992 | 86.333% (78.919–88.228) | 86.346% (78.859–88.236) | 87.988% (84.234–88.774) | 89.665% (84.398–91.179) |
+| 20% | 201 / 1,003 | 119,570 | 751 | 83.805% (76.610–88.329) | 83.786% (76.516–88.349) | 85.219% (82.368–90.909) | 86.626% (80.757–90.082) |
+| 10% | 100 / 1,003 | 135,287 | 379 | 78.591% (75.365–82.879) | 78.545% (75.321–82.877) | 85.459% (81.910–87.838) | 80.407% (77.818–84.307) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Romanian</strong> — <code>ro-ro-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 4,387 / 4,387 | 0 | 0 | n/a | n/a | n/a | 99.175% |
+| 90% | 3,948 / 4,387 | 4,742 | 70 | 74.659% (73.030–76.354) | 73.593% (71.508–75.100) | 88.235% (85.176–89.202) | 96.784% (96.488–96.925) |
+| 80% | 3,510 / 4,387 | 9,657 | 127 | 74.645% (73.922–76.412) | 73.446% (72.512–75.392) | 87.044% (85.936–88.318) | 94.223% (94.123–94.517) |
+| 70% | 3,071 / 4,387 | 14,679 | 170 | 74.310% (73.479–75.682) | 73.032% (72.291–74.557) | 87.412% (85.927–87.549) | 91.538% (91.303–91.923) |
+| 60% | 2,632 / 4,387 | 19,466 | 195 | 73.800% (72.871–74.215) | 72.529% (71.488–72.936) | 87.002% (86.480–87.507) | 88.877% (88.685–89.015) |
+| 50% | 2,194 / 4,387 | 24,311 | 207 | 72.983% (72.075–73.477) | 71.728% (70.710–72.145) | 86.034% (84.669–87.216) | 85.887% (85.552–86.142) |
+| 40% | 1,755 / 4,387 | 29,095 | 194 | 71.893% (70.956–72.457) | 70.609% (69.717–71.169) | 85.218% (83.617–85.808) | 82.526% (82.260–82.938) |
+| 30% | 1,316 / 4,387 | 34,201 | 176 | 70.530% (69.863–71.305) | 69.211% (68.443–69.923) | 84.309% (83.894–85.278) | 79.021% (78.448–79.720) |
+| 20% | 877 / 4,387 | 39,061 | 139 | 68.598% (67.720–70.052) | 66.948% (66.061–68.629) | 84.587% (83.133–85.443) | 74.678% (73.710–75.830) |
+| 10% | 439 / 4,387 | 44,011 | 84 | 65.973% (65.003–67.866) | 64.222% (63.190–66.537) | 83.189% (81.399–83.825) | 69.223% (68.270–71.010) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Russian</strong> — <code>ru-ru-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 37,410 / 37,410 | 0 | 0 | n/a | n/a | n/a | 98.807% |
+| 90% | 33,669 / 37,410 | 79,477 | 1,784 | 82.007% (81.842–82.574) | 80.744% (80.515–81.372) | 94.506% (94.455–94.967) | 97.047% (97.012–97.088) |
+| 80% | 29,928 / 37,410 | 158,745 | 3,165 | 81.665% (81.300–82.243) | 80.343% (79.956–81.034) | 94.359% (93.988–94.456) | 95.273% (95.172–95.333) |
+| 70% | 26,187 / 37,410 | 238,018 | 4,273 | 81.575% (81.414–81.952) | 80.296% (80.115–80.758) | 94.006% (93.750–94.173) | 93.468% (93.397–93.575) |
+| 60% | 22,446 / 37,410 | 318,625 | 4,709 | 81.182% (80.809–81.611) | 79.915% (79.506–80.371) | 93.632% (93.525–93.734) | 91.517% (91.394–91.717) |
+| 50% | 18,705 / 37,410 | 398,725 | 5,007 | 80.854% (80.304–81.219) | 79.578% (78.976–79.972) | 93.316% (93.219–93.447) | 89.614% (89.371–89.752) |
+| 40% | 14,964 / 37,410 | 479,920 | 4,763 | 80.312% (79.666–80.706) | 79.019% (78.331–79.445) | 93.010% (92.649–93.088) | 87.513% (87.184–87.695) |
+| 30% | 11,223 / 37,410 | 560,108 | 4,217 | 79.739% (79.241–79.963) | 78.423% (77.874–78.684) | 92.571% (92.488–92.673) | 85.305% (84.992–85.415) |
+| 20% | 7,482 / 37,410 | 640,968 | 3,381 | 78.648% (78.391–79.002) | 77.254% (76.977–77.673) | 91.951% (91.925–92.274) | 82.544% (82.404–82.868) |
+| 10% | 3,741 / 37,410 | 723,165 | 1,916 | 76.473% (76.268–77.001) | 74.948% (74.735–75.545) | 91.268% (90.883–91.995) | 78.669% (78.459–79.115) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Seneca</strong> — <code>see-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 139 / 139 | 0 | 0 | n/a | n/a | n/a | 99.756% |
+| 90% | 125 / 139 | 537 | 1 | 2.602% (2.403–2.612) | 0.000% | 100.000% | 89.946% (89.852–89.964) |
+| 80% | 111 / 139 | 1,072 | 1 | 2.519% (2.505–2.634) | 0.000% | 100.000% (96.429–100.000) | 80.154% (80.060–80.341) |
+| 70% | 97 / 139 | 1,604 | 11 | 2.556% (2.413–2.628) | 0.000% | 100.000% (92.857–100.000) | 70.437% (70.174–70.568) |
+| 60% | 83 / 139 | 2,148 | 11 | 2.568% (2.466–2.607) | 0.000% | 100.000% (94.643–100.000) | 60.514% (60.364–60.758) |
+| 50% | 70 / 139 | 2,650 | 11 | 2.585% (2.487–2.604) | 0.000% | 100.000% (95.652–100.000) | 51.341% (51.229–51.754) |
+| 40% | 56 / 139 | 3,180 | 11 | 2.583% (2.513–2.618) | 0.000% | 100.000% (96.386–100.000) | 41.643% (41.268–41.887) |
+| 30% | 42 / 139 | 3,718 | 11 | 2.588% (2.526–2.615) | 0.000% | 100.000% (96.907–100.000) | 31.814% (31.589–32.002) |
+| 20% | 28 / 139 | 4,245 | 11 | 2.594% (2.526–2.600) | 0.000% | 100.000% (97.297–100.000) | 22.191% (21.797–22.435) |
+| 10% | 14 / 139 | 4,789 | 1 | 2.608% (2.593–2.611) | 0.000% | 100.000% (99.200–100.000) | 12.418% (12.324–12.549) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Serbo-Croatian</strong> — <code>hbs-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 24,385 / 24,385 | 0 | 0 | n/a | n/a | n/a | 98.061% |
+| 90% | 21,947 / 24,385 | 26,814 | 932 | 64.884% (64.413–65.313) | 61.807% (61.429–62.316) | 95.286% (94.607–95.719) | 94.672% (94.586–94.780) |
+| 80% | 19,508 / 24,385 | 53,548 | 1,684 | 64.092% (63.870–64.678) | 61.086% (60.771–61.689) | 94.741% (94.595–95.279) | 91.179% (91.162–91.366) |
+| 70% | 17,070 / 24,385 | 80,657 | 2,232 | 63.722% (62.794–64.345) | 60.710% (59.603–61.356) | 94.584% (94.288–95.180) | 87.676% (87.453–87.940) |
+| 60% | 14,631 / 24,385 | 108,349 | 2,570 | 63.351% (62.478–63.666) | 60.291% (59.297–60.632) | 94.594% (94.200–95.051) | 84.068% (83.723–84.309) |
+| 50% | 12,193 / 24,385 | 135,615 | 2,778 | 62.642% (62.355–63.020) | 59.558% (59.269–59.959) | 94.199% (93.989–94.334) | 80.220% (80.098–80.582) |
+| 40% | 9,754 / 24,385 | 163,679 | 2,657 | 61.776% (61.212–62.010) | 58.682% (58.066–58.880) | 94.064% (93.545–94.107) | 76.106% (75.884–76.389) |
+| 30% | 7,316 / 24,385 | 192,170 | 2,289 | 60.705% (60.324–61.019) | 57.558% (57.074–57.877) | 93.449% (93.345–93.927) | 71.716% (71.578–72.056) |
+| 20% | 4,877 / 24,385 | 220,464 | 1,800 | 59.329% (59.153–59.632) | 56.068% (55.892–56.408) | 93.044% (92.817–93.173) | 67.046% (66.947–67.197) |
+| 10% | 2,439 / 24,385 | 248,808 | 1,084 | 57.218% (56.568–58.465) | 53.844% (53.138–55.190) | 92.409% (92.058–92.723) | 61.263% (60.766–62.413) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Shipibo-Conibo</strong> — <code>shp-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 1,999 / 1,999 | 0 | 0 | n/a | n/a | n/a | 96.952% |
+| 90% | 1,799 / 1,999 | 757 | 41 | 84.655% (83.884–85.411) | 81.206% (79.965–82.526) | 95.722% (93.717–97.895) | 95.547% (95.459–95.597) |
+| 80% | 1,599 / 1,999 | 1,526 | 70 | 84.112% (83.083–84.712) | 80.617% (80.225–81.195) | 94.947% (91.799–95.789) | 94.029% (93.778–94.117) |
+| 70% | 1,399 / 1,999 | 2,297 | 94 | 83.777% (83.123–84.491) | 80.264% (79.191–81.312) | 94.170% (91.930–95.288) | 92.423% (92.198–92.737) |
+| 60% | 1,199 / 1,999 | 3,084 | 106 | 82.821% (82.754–83.722) | 79.699% (78.784–80.242) | 93.995% (92.298–94.961) | 90.655% (90.580–91.182) |
+| 50% | 1,000 / 1,999 | 3,858 | 125 | 83.044% (82.271–83.589) | 79.736% (78.549–80.201) | 93.458% (92.340–93.821) | 89.200% (89.024–89.551) |
+| 40% | 800 / 1,999 | 4,649 | 138 | 82.709% (82.111–82.970) | 79.503% (78.469–79.769) | 92.255% (90.987–93.035) | 87.506% (87.293–87.908) |
+| 30% | 600 / 1,999 | 5,451 | 129 | 81.968% (81.563–83.087) | 78.568% (78.042–80.336) | 91.860% (91.256–92.347) | 85.775% (85.462–86.327) |
+| 20% | 400 / 1,999 | 6,276 | 112 | 81.886% (80.003–83.270) | 78.450% (76.312–81.085) | 91.001% (89.784–92.094) | 84.295% (82.526–85.261) |
+| 10% | 200 / 1,999 | 7,089 | 89 | 81.553% (79.770–82.254) | 78.067% (76.244–79.762) | 90.711% (87.787–91.895) | 82.514% (80.682–82.928) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Shona</strong> — <code>sn-zw-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 86 / 86 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 90% | 77 / 86 | 279 | 0 | 3.226% (3.226–3.557) | 0.000% | 100.000% | 89.773% (89.773–90.758) |
+| 80% | 69 / 86 | 527 | 0 | 3.226% (3.226–3.393) | 0.000% | 100.000% | 80.682% (80.682–81.667) |
+| 70% | 60 / 86 | 806 | 0 | 3.226% (3.226–3.333) | 0.000% | 100.000% | 70.455% (70.455–71.439) |
+| 60% | 52 / 86 | 1,054 | 0 | 3.226% (3.226–3.307) | 0.000% | 100.000% | 61.364% (61.364–62.348) |
+| 50% | 43 / 86 | 1,307 | 0 | 3.290% (3.226–3.290) | 0.000% | 100.000% | 52.121% (51.136–52.121) |
+| 40% | 34 / 86 | 1,586 | 0 | 3.279% (3.226–3.279) | 0.000% | 100.000% | 41.894% (40.909–41.894) |
+| 30% | 26 / 86 | 1,834 | 0 | 3.272% (3.226–3.272) | 0.000% | 100.000% | 32.803% (31.818–32.803) |
+| 20% | 17 / 86 | 2,113 | 0 | 3.265% (3.226–3.265) | 0.000% | 100.000% | 22.576% (21.591–22.576) |
+| 10% | 9 / 86 | 2,361 | 0 | 3.261% (3.226–3.261) | 0.000% | 100.000% | 13.485% (12.500–13.485) |
 
 </div>
 </details>
 
 <details markdown="block">
-<summary><strong>Danish</strong> — <code>da-dk-default</code> 1.0.0</summary>
+<summary><strong>Sotho, Southern</strong> — <code>st-za-default</code> 1.0.0</summary>
 <div markdown="1">
 
 | Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
 | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
-| 100% | 4,179 / 4,179 | 0 | 0 | n/a | n/a | n/a | 99.371% |
-| 90% | 3,761 / 4,179 | 3,232 | 38 | 74.192% (73.356–75.186) | 67.504% (66.790–69.750) | 92.718% (90.865–93.961) | 96.813% (96.698–96.897) |
-| 80% | 3,343 / 4,179 | 6,419 | 59 | 74.015% (73.293–74.244) | 67.477% (67.434–67.862) | 92.377% (90.132–93.494) | 94.212% (94.063–94.358) |
-| 70% | 2,925 / 4,179 | 9,622 | 72 | 72.901% (72.673–74.059) | 66.648% (65.699–67.713) | 91.066% (90.273–92.613) | 91.447% (91.276–91.726) |
-| 60% | 2,507 / 4,179 | 12,858 | 89 | 72.406% (72.124–72.935) | 65.740% (65.579–67.061) | 90.565% (89.988–91.933) | 88.498% (88.442–88.641) |
-| 50% | 2,090 / 4,179 | 16,078 | 86 | 71.433% (70.792–72.016) | 64.706% (63.871–65.934) | 90.561% (89.489–90.758) | 85.333% (85.042–85.612) |
-| 40% | 1,672 / 4,179 | 19,316 | 87 | 70.449% (69.207–70.919) | 63.495% (61.895–64.640) | 90.106% (88.956–90.560) | 81.882% (81.244–82.196) |
-| 30% | 1,254 / 4,179 | 22,543 | 70 | 68.316% (67.096–69.463) | 60.962% (59.342–62.755) | 89.335% (88.687–90.062) | 77.558% (76.727–78.444) |
-| 20% | 836 / 4,179 | 25,755 | 61 | 65.725% (64.691–66.441) | 57.555% (56.445–58.672) | 89.232% (88.279–90.082) | 72.436% (71.615–73.034) |
-| 10% | 418 / 4,179 | 29,022 | 28 | 61.104% (59.899–62.277) | 51.055% (49.995–53.218) | 88.238% (87.507–89.957) | 64.847% (63.836–65.997) |
+| 100% | 26 / 26 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 90% | 23 / 26 | 48 | 0 | 12.500% | 6.667% | 100.000% | 89.904% |
+| 80% | 21 / 26 | 80 | 0 | 12.500% | 6.667% | 100.000% | 83.173% |
+| 70% | 18 / 26 | 128 | 0 | 12.500% | 6.667% | 100.000% | 73.077% |
+| 60% | 16 / 26 | 160 | 0 | 12.500% | 6.667% | 100.000% | 66.346% |
+| 50% | 13 / 26 | 208 | 0 | 12.500% | 6.667% | 100.000% | 56.250% |
+| 40% | 10 / 26 | 256 | 0 | 12.500% | 6.667% | 100.000% | 46.154% |
+| 30% | 8 / 26 | 288 | 0 | 12.500% | 6.667% | 100.000% | 39.423% |
+| 20% | 5 / 26 | 336 | 0 | 12.500% | 6.667% | 100.000% | 29.327% |
+| 10% | 3 / 26 | 368 | 0 | 12.500% | 6.667% | 100.000% | 22.596% |
 
 </div>
 </details>
 
 <details markdown="block">
-<summary><strong>German</strong> — <code>de-de-default</code> 1.0.0</summary>
+<summary><strong>Southern Kurdish</strong> — <code>sdh-default</code> 1.0.0</summary>
 <div markdown="1">
 
 | Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
 | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
-| 100% | 54,092 / 54,092 | 0 | 0 | n/a | n/a | n/a | 92.725% |
-| 90% | 48,683 / 54,092 | 27,967 | 5,327 | 67.322% (66.336–67.716) | 59.108% (57.552–59.642) | 93.117% (92.600–93.345) | 89.908% (89.841–89.987) |
-| 80% | 43,274 / 54,092 | 57,109 | 9,524 | 66.531% (65.981–66.706) | 58.016% (57.177–58.316) | 92.441% (91.771–92.581) | 87.066% (86.980–87.121) |
-| 70% | 37,864 / 54,092 | 87,177 | 12,653 | 65.784% (65.442–66.138) | 56.948% (56.648–57.396) | 91.897% (91.743–91.984) | 84.105% (83.974–84.251) |
-| 60% | 32,455 / 54,092 | 118,916 | 14,631 | 65.062% (64.946–65.321) | 56.115% (55.912–56.303) | 91.372% (91.222–91.619) | 81.048% (80.993–81.213) |
-| 50% | 27,046 / 54,092 | 151,254 | 15,476 | 64.493% (64.083–64.603) | 55.371% (54.795–55.510) | 90.896% (90.661–91.008) | 78.014% (77.799–78.062) |
-| 40% | 21,637 / 54,092 | 184,273 | 15,105 | 63.564% (63.219–63.981) | 54.206% (53.646–54.739) | 90.246% (89.885–90.446) | 74.680% (74.544–74.924) |
-| 30% | 16,228 / 54,092 | 219,296 | 13,318 | 62.696% (62.282–63.101) | 53.083% (52.613–53.663) | 89.436% (88.996–89.500) | 71.309% (71.004–71.509) |
-| 20% | 10,818 / 54,092 | 255,708 | 10,520 | 61.569% (60.724–61.785) | 51.959% (50.533–52.161) | 88.244% (88.017–88.723) | 67.504% (66.844–67.696) |
-| 10% | 5,409 / 54,092 | 293,618 | 6,095 | 59.280% (58.219–59.702) | 49.432% (48.026–49.731) | 86.150% (85.836–86.733) | 62.456% (61.579–62.888) |
+| 100% | 1 / 1 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 90% | 1 / 1 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 80% | 1 / 1 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 70% | 1 / 1 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 60% | 1 / 1 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 50% | 1 / 1 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 40% | 1 / 1 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 30% | 1 / 1 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 20% | 1 / 1 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 10% | 1 / 1 | 0 | 0 | n/a | n/a | n/a | 100.000% |
 
 </div>
 </details>
@@ -134,336 +2574,354 @@ evidence. At 100% there is no held-out set, so unseen metrics are `n/a`.
 </details>
 
 <details markdown="block">
-<summary><strong>Persian</strong> — <code>fa-ir-default</code> 1.0.0</summary>
+<summary><strong>Swedish</strong> — <code>sv-se-default</code> 1.0.0</summary>
 <div markdown="1">
 
 | Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
 | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
-| 100% | 69 / 69 | 0 | 0 | n/a | n/a | n/a | 95.836% |
-| 90% | 62 / 69 | 360 | 29 | 18.699% (15.833–19.469) | 15.493% (13.006–16.000) | 100.000% (85.714–100.000) | 87.878% (87.745–88.594) |
-| 80% | 55 / 69 | 743 | 29 | 15.979% (13.625–19.473) | 12.725% (10.400–16.031) | 100.000% (92.857–100.000) | 79.204% (79.072–81.247) |
-| 70% | 48 / 69 | 1,054 | 87 | 17.998% (15.808–19.631) | 14.544% (12.546–16.211) | 100.000% (95.238–100.000) | 73.077% (71.645–73.899) |
-| 60% | 41 / 69 | 1,474 | 87 | 17.843% (16.847–18.371) | 14.739% (13.572–15.175) | 100.000% (96.429–100.000) | 65.305% (63.740–65.517) |
-| 50% | 35 / 69 | 1,792 | 88 | 17.796% (17.243–18.567) | 14.596% (13.979–15.398) | 100.000% (97.059–100.000) | 58.090% (57.560–59.019) |
-| 40% | 28 / 69 | 2,185 | 87 | 18.078% (16.720–19.128) | 14.979% (13.479–15.918) | 100.000% (97.561–100.000) | 50.186% (49.310–51.114) |
-| 30% | 21 / 69 | 2,604 | 58 | 17.696% (16.285–18.267) | 14.648% (13.104–15.153) | 100.000% (97.917–100.000) | 41.857% (40.981–42.706) |
-| 20% | 14 / 69 | 2,978 | 30 | 18.105% (16.353–18.304) | 15.009% (13.145–15.164) | 100.000% (98.182–100.000) | 34.058% (33.130–34.536) |
-| 10% | 7 / 69 | 3,367 | 29 | 17.987% (16.428–17.998) | 14.842% (13.213–14.887) | 100.000% | 25.995% (24.987–26.180) |
+| 100% | 12,371 / 12,371 | 0 | 0 | n/a | n/a | n/a | 96.713% |
+| 90% | 11,134 / 12,371 | 10,386 | 675 | 82.188% (81.351–82.660) | 78.458% (77.818–79.022) | 94.926% (93.807–95.482) | 95.085% (95.025–95.156) |
+| 80% | 9,897 / 12,371 | 20,899 | 1,144 | 81.612% (81.194–82.294) | 78.137% (77.386–78.776) | 94.577% (93.139–94.972) | 93.421% (93.317–93.570) |
+| 70% | 8,660 / 12,371 | 31,613 | 1,500 | 81.378% (80.737–81.694) | 77.755% (77.158–78.370) | 93.450% (93.333–94.124) | 91.702% (91.610–91.808) |
+| 60% | 7,423 / 12,371 | 42,503 | 1,726 | 80.584% (80.441–80.710) | 77.113% (76.939–77.166) | 93.114% (92.742–93.192) | 89.845% (89.812–89.898) |
+| 50% | 6,186 / 12,371 | 53,431 | 1,736 | 79.553% (79.081–79.924) | 75.813% (75.376–76.278) | 92.204% (92.067–92.705) | 87.777% (87.484–87.889) |
+| 40% | 4,948 / 12,371 | 64,540 | 1,677 | 78.689% (78.113–78.970) | 74.894% (74.347–75.197) | 91.778% (91.278–92.147) | 85.502% (85.283–85.636) |
+| 30% | 3,711 / 12,371 | 75,916 | 1,458 | 77.249% (76.663–77.381) | 73.248% (72.517–73.345) | 91.335% (90.486–91.670) | 82.749% (82.497–82.863) |
+| 20% | 2,474 / 12,371 | 87,316 | 1,095 | 75.205% (74.488–75.890) | 70.861% (70.211–72.027) | 89.668% (89.346–90.344) | 79.323% (78.735–79.852) |
+| 10% | 1,237 / 12,371 | 98,793 | 636 | 72.010% (70.983–72.505) | 67.214% (66.144–68.191) | 87.789% (86.974–88.659) | 74.396% (73.426–74.785) |
 
 </div>
 </details>
 
 <details markdown="block">
-<summary><strong>Finnish</strong> — <code>fi-fi-default</code> 1.0.0</summary>
+<summary><strong>Tagalog</strong> — <code>tl-ph-default</code> 1.0.0</summary>
 <div markdown="1">
 
 | Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
 | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
-| 100% | 57,027 / 57,027 | 0 | 0 | n/a | n/a | n/a | 98.661% |
-| 90% | 51,324 / 57,027 | 181,248 | 5,149 | 86.644% (86.280–86.823) | 86.177% (85.790–86.342) | 94.835% (94.730–95.762) | 97.363% (97.337–97.369) |
-| 80% | 45,622 / 57,027 | 363,208 | 9,479 | 86.152% (86.055–86.432) | 85.686% (85.568–85.979) | 94.265% (94.183–94.726) | 95.981% (95.972–96.026) |
-| 70% | 39,919 / 57,027 | 546,876 | 12,346 | 85.689% (85.586–85.804) | 85.219% (85.114–85.335) | 93.710% (93.575–93.808) | 94.531% (94.518–94.560) |
-| 60% | 34,216 / 57,027 | 730,977 | 14,257 | 85.172% (85.110–85.293) | 84.708% (84.639–84.829) | 93.033% (92.895–93.427) | 92.999% (92.978–93.035) |
-| 50% | 28,514 / 57,027 | 917,680 | 14,946 | 84.681% (84.520–84.723) | 84.232% (84.039–84.280) | 92.203% (92.123–92.566) | 91.382% (91.295–91.404) |
-| 40% | 22,811 / 57,027 | 1,103,628 | 14,607 | 83.991% (83.783–84.165) | 83.556% (83.313–83.715) | 91.299% (91.199–91.628) | 89.582% (89.463–89.675) |
-| 30% | 17,108 / 57,027 | 1,292,480 | 12,981 | 83.290% (82.866–83.366) | 82.848% (82.399–82.936) | 90.499% (90.312–90.537) | 87.658% (87.378–87.720) |
-| 20% | 11,405 / 57,027 | 1,481,274 | 10,279 | 81.958% (81.525–81.970) | 81.481% (81.032–81.508) | 89.459% (89.355–89.678) | 85.129% (84.780–85.155) |
-| 10% | 5,703 / 57,027 | 1,672,373 | 6,067 | 79.210% (78.547–79.594) | 78.644% (77.966–79.067) | 88.247% (87.860–88.496) | 81.064% (80.468–81.422) |
+| 100% | 342 / 342 | 0 | 0 | n/a | n/a | n/a | 99.682% |
+| 90% | 308 / 342 | 247 | 1 | 16.194% (14.340–17.121) | 2.817% (1.531–4.484) | 100.000% | 91.461% (90.786–92.017) |
+| 80% | 274 / 342 | 501 | 3 | 15.496% (14.532–16.331) | 1.782% (1.683–3.037) | 100.000% | 82.963% (82.049–83.439) |
+| 70% | 239 / 342 | 774 | 3 | 14.987% (14.861–15.668) | 1.937% (1.902–2.171) | 100.000% | 73.670% (72.955–75.099) |
+| 60% | 205 / 342 | 1,021 | 2 | 15.452% (14.908–15.529) | 1.984% (1.810–2.466) | 100.000% | 65.330% (64.893–66.322) |
+| 50% | 171 / 342 | 1,265 | 3 | 15.361% (14.571–15.494) | 2.204% (1.648–2.285) | 100.000% | 57.029% (55.560–57.506) |
+| 40% | 137 / 342 | 1,519 | 2 | 15.076% (14.516–15.251) | 1.783% (1.487–1.828) | 100.000% | 48.610% (47.180–49.404) |
+| 30% | 103 / 342 | 1,777 | 3 | 14.836% (14.270–15.006) | 1.607% (0.973–1.703) | 100.000% | 39.793% (39.118–40.905) |
+| 20% | 68 / 342 | 2,013 | 2 | 14.655% (14.328–15.037) | 1.208% (0.914–1.817) | 100.000% | 31.652% (31.096–31.970) |
+| 10% | 34 / 342 | 2,275 | 1 | 14.485% (13.846–15.040) | 1.210% (0.356–1.780) | 100.000% (99.675–100.000) | 22.796% (22.121–23.193) |
 
 </div>
 </details>
 
 <details markdown="block">
-<summary><strong>French</strong> — <code>fr-fr-default</code> 1.0.0</summary>
+<summary><strong>Temoaya Otomi</strong> — <code>ote-default</code> 1.0.0</summary>
 <div markdown="1">
 
 | Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
 | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
-| 100% | 59,240 / 59,240 | 0 | 0 | n/a | n/a | n/a | 94.831% |
-| 90% | 53,316 / 59,240 | 41,750 | 5,161 | 83.469% (82.278–83.711) | 80.031% (78.595–80.412) | 96.664% (96.351–96.813) | 93.302% (93.201–93.351) |
-| 80% | 47,392 / 59,240 | 85,315 | 9,281 | 82.721% (81.861–83.091) | 79.345% (78.141–79.693) | 96.074% (95.896–96.266) | 91.731% (91.568–91.794) |
-| 70% | 41,468 / 59,240 | 130,120 | 12,345 | 82.101% (81.643–82.329) | 78.644% (78.015–78.839) | 95.431% (95.257–95.560) | 90.086% (89.942–90.135) |
-| 60% | 35,544 / 59,240 | 175,326 | 14,017 | 81.224% (80.630–81.431) | 77.560% (76.779–77.867) | 94.813% (94.641–94.831) | 88.270% (88.228–88.450) |
-| 50% | 29,620 / 59,240 | 222,533 | 14,808 | 80.344% (79.709–80.528) | 76.569% (75.778–76.888) | 94.006% (93.785–94.099) | 86.507% (86.336–86.609) |
-| 40% | 23,696 / 59,240 | 270,536 | 14,411 | 79.255% (79.039–79.552) | 75.314% (75.088–75.702) | 93.386% (93.086–93.418) | 84.453% (84.319–84.752) |
-| 30% | 17,772 / 59,240 | 318,984 | 12,728 | 78.161% (78.141–78.274) | 74.079% (73.976–74.276) | 92.537% (92.395–92.906) | 82.316% (82.244–82.378) |
-| 20% | 11,848 / 59,240 | 369,337 | 9,783 | 76.769% (76.583–76.946) | 72.504% (72.339–72.772) | 91.463% (90.996–91.884) | 79.741% (79.646–79.873) |
-| 10% | 5,924 / 59,240 | 420,578 | 5,742 | 74.276% (73.276–74.926) | 69.627% (68.400–70.528) | 90.013% (89.987–90.359) | 75.978% (75.101–76.554) |
+| 100% | 2,026 / 2,026 | 0 | 0 | n/a | n/a | n/a | 95.875% |
+| 90% | 1,823 / 2,026 | 329 | 26 | 61.515% (60.182–63.665) | 7.463% (5.344–14.179) | 96.907% (95.000–98.469) | 92.396% (92.200–92.649) |
+| 80% | 1,621 / 2,026 | 668 | 49 | 60.546% (58.555–62.267) | 8.519% (7.609–10.887) | 96.134% (94.444–97.201) | 88.805% (88.187–89.198) |
+| 70% | 1,418 / 2,026 | 1,014 | 63 | 59.724% (59.002–60.926) | 8.451% (7.294–11.083) | 96.441% (94.128–97.293) | 84.792% (84.400–85.241) |
+| 60% | 1,216 / 2,026 | 1,361 | 73 | 59.442% (58.948–59.752) | 7.958% (7.570–9.725) | 96.207% (93.962–97.478) | 81.117% (80.752–81.397) |
+| 50% | 1,013 / 2,026 | 1,709 | 80 | 58.919% (58.534–59.450) | 8.146% (6.863–9.843) | 95.762% (94.260–96.690) | 77.217% (76.880–77.722) |
+| 40% | 810 / 2,026 | 2,076 | 70 | 58.402% (58.382–58.903) | 8.121% (7.859–10.162) | 95.409% (93.395–95.424) | 73.232% (72.952–73.485) |
+| 30% | 608 / 2,026 | 2,441 | 57 | 57.918% (57.681–58.251) | 8.317% (7.157–10.634) | 94.872% (93.138–95.306) | 69.136% (68.771–69.557) |
+| 20% | 405 / 2,026 | 2,805 | 43 | 57.553% (57.468–57.596) | 8.870% (7.305–10.067) | 93.540% (92.742–94.776) | 65.039% (64.955–65.208) |
+| 10% | 203 / 2,026 | 3,179 | 26 | 57.049% (56.368–58.131) | 7.778% (6.250–9.824) | 93.985% (93.341–94.615) | 60.915% (60.269–61.925) |
 
 </div>
 </details>
 
 <details markdown="block">
-<summary><strong>Hebrew</strong> — <code>he-il-default</code> 1.0.0</summary>
+<summary><strong>Turkish</strong> — <code>tr-tr-default</code> 1.0.0</summary>
 <div markdown="1">
 
 | Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
 | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
-| 100% | 2,358 / 2,358 | 0 | 0 | n/a | n/a | n/a | 98.250% |
-| 90% | 2,122 / 2,358 | 5,914 | 196 | 21.813% (21.558–23.195) | 16.362% (15.846–17.917) | 88.785% (87.416–93.636) | 90.688% (90.578–90.888) |
-| 80% | 1,886 / 2,358 | 11,868 | 390 | 21.715% (19.936–22.602) | 16.441% (14.182–17.242) | 89.628% (87.585–91.818) | 83.007% (82.678–83.339) |
-| 70% | 1,651 / 2,358 | 17,846 | 499 | 21.488% (20.548–21.734) | 16.059% (15.072–16.256) | 88.713% (88.166–89.569) | 75.314% (75.106–75.538) |
-| 60% | 1,415 / 2,358 | 23,856 | 551 | 20.530% (20.267–21.311) | 15.043% (14.721–15.824) | 88.702% (87.417–89.456) | 67.302% (67.264–67.705) |
-| 50% | 1,179 / 2,358 | 29,985 | 576 | 20.205% (19.833–20.807) | 14.533% (14.185–15.123) | 88.771% (87.642–90.375) | 59.531% (59.185–59.763) |
-| 40% | 943 / 2,358 | 36,073 | 547 | 19.758% (19.628–20.304) | 14.068% (13.841–14.608) | 89.236% (88.497–90.062) | 51.381% (51.289–51.799) |
-| 30% | 707 / 2,358 | 42,248 | 465 | 19.125% (18.459–19.270) | 13.345% (12.730–13.462) | 88.961% (87.481–90.603) | 43.063% (42.627–43.264) |
-| 20% | 472 / 2,358 | 48,464 | 365 | 17.734% (17.573–18.304) | 11.846% (11.656–12.416) | 89.000% (88.647–89.233) | 34.046% (33.884–34.388) |
-| 10% | 236 / 2,358 | 54,723 | 206 | 16.532% (15.398–18.499) | 10.423% (9.169–12.678) | 89.756% (88.191–91.158) | 24.765% (23.666–26.661) |
+| 100% | 3,017 / 3,017 | 0 | 0 | n/a | n/a | n/a | 99.263% |
+| 90% | 2,715 / 3,017 | 22,024 | 288 | 82.355% (79.663–85.071) | 82.279% (79.549–85.021) | 88.776% (88.014–92.177) | 97.526% (97.031–97.798) |
+| 80% | 2,414 / 3,017 | 43,638 | 442 | 81.792% (81.129–82.694) | 81.709% (81.009–82.607) | 88.985% (88.055–89.932) | 95.804% (95.608–95.959) |
+| 70% | 2,112 / 3,017 | 66,129 | 600 | 81.756% (81.048–83.798) | 81.660% (80.939–83.734) | 88.737% (87.585–91.023) | 93.949% (93.831–94.503) |
+| 60% | 1,810 / 3,017 | 88,540 | 733 | 81.961% (81.480–83.611) | 81.884% (81.381–83.550) | 88.067% (87.026–90.678) | 92.279% (92.054–92.910) |
+| 50% | 1,509 / 3,017 | 111,440 | 911 | 81.608% (79.858–83.287) | 81.531% (79.726–83.240) | 87.754% (86.752–89.744) | 90.244% (89.262–91.086) |
+| 40% | 1,207 / 3,017 | 133,132 | 900 | 81.522% (79.947–82.252) | 81.444% (79.832–82.181) | 87.290% (87.073–88.373) | 88.411% (87.502–89.019) |
+| 30% | 905 / 3,017 | 155,798 | 763 | 81.433% (79.698–81.551) | 81.355% (79.593–81.479) | 87.189% (84.405–87.542) | 86.724% (85.562–86.851) |
+| 20% | 603 / 3,017 | 178,334 | 575 | 79.471% (77.942–80.778) | 79.417% (77.814–80.700) | 86.502% (83.375–87.385) | 83.443% (82.135–84.477) |
+| 10% | 302 / 3,017 | 201,851 | 285 | 77.397% (74.873–79.731) | 77.296% (74.748–79.658) | 84.869% (83.617–85.031) | 79.456% (77.232–81.684) |
 
 </div>
 </details>
 
 <details markdown="block">
-<summary><strong>Hungarian</strong> — <code>hu-hu-default</code> 1.0.0</summary>
+<summary><strong>Ukrainian</strong> — <code>uk-ua-default</code> 1.0.0</summary>
 <div markdown="1">
 
 | Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
 | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
-| 100% | 19,406 / 19,406 | 0 | 0 | n/a | n/a | n/a | 99.222% |
-| 90% | 17,465 / 19,406 | 92,253 | 1,360 | 92.212% (91.551–92.510) | 92.203% (91.476–92.460) | 93.446% (92.426–93.786) | 98.479% (98.407–98.493) |
-| 80% | 15,525 / 19,406 | 184,401 | 2,417 | 91.771% (91.652–92.138) | 91.756% (91.609–92.123) | 92.536% (92.147–92.726) | 97.646% (97.626–97.712) |
-| 70% | 13,584 / 19,406 | 277,343 | 3,300 | 91.756% (91.390–92.053) | 91.744% (91.350–92.058) | 91.940% (90.827–92.376) | 96.861% (96.757–96.947) |
-| 60% | 11,644 / 19,406 | 370,552 | 3,836 | 91.156% (91.058–91.366) | 91.187% (91.074–91.418) | 90.673% (90.092–91.129) | 95.888% (95.829–95.941) |
-| 50% | 9,703 / 19,406 | 463,693 | 3,952 | 90.729% (90.537–90.756) | 90.777% (90.567–90.810) | 89.530% (89.200–89.806) | 94.834% (94.749–94.863) |
-| 40% | 7,762 / 19,406 | 557,829 | 3,717 | 90.002% (89.679–90.181) | 90.053% (89.707–90.254) | 88.776% (88.318–89.016) | 93.526% (93.390–93.683) |
-| 30% | 5,822 / 19,406 | 651,802 | 3,207 | 89.047% (88.909–89.354) | 89.126% (89.006–89.437) | 87.178% (86.603–87.420) | 92.006% (91.882–92.204) |
-| 20% | 3,881 / 19,406 | 746,292 | 2,401 | 88.280% (87.847–88.842) | 88.393% (87.931–88.968) | 85.865% (85.589–85.920) | 90.395% (90.054–90.844) |
-| 10% | 1,941 / 19,406 | 840,551 | 1,399 | 86.761% (86.100–87.668) | 86.876% (86.177–87.841) | 84.086% (83.633–85.050) | 87.963% (87.354–88.786) |
+| 100% | 1,493 / 1,493 | 0 | 0 | n/a | n/a | n/a | 99.307% |
+| 90% | 1,344 / 1,493 | 1,576 | 12 | 63.233% (61.940–65.736) | 57.218% (55.192–59.594) | 90.476% (88.514–93.243) | 95.673% (95.418–95.876) |
+| 80% | 1,194 / 1,493 | 3,166 | 34 | 62.236% (61.727–64.985) | 56.335% (54.703–59.359) | 88.475% (87.879–93.255) | 91.987% (91.580–92.514) |
+| 70% | 1,045 / 1,493 | 4,675 | 36 | 62.139% (60.712–63.062) | 55.456% (54.248–56.768) | 89.367% (85.811–90.562) | 87.990% (87.526–88.575) |
+| 60% | 896 / 1,493 | 6,256 | 53 | 61.416% (58.976–62.308) | 54.708% (52.041–55.903) | 90.211% (87.669–91.525) | 84.076% (83.332–84.393) |
+| 50% | 747 / 1,493 | 7,838 | 50 | 60.214% (59.232–62.314) | 53.268% (51.925–55.648) | 90.135% (88.140–91.180) | 79.793% (79.138–80.587) |
+| 40% | 597 / 1,493 | 9,392 | 66 | 58.933% (57.994–60.117) | 51.880% (50.380–53.150) | 89.013% (87.486–90.578) | 75.097% (74.309–75.599) |
+| 30% | 448 / 1,493 | 10,951 | 60 | 58.222% (57.492–59.390) | 51.048% (50.564–52.388) | 87.380% (86.917–89.403) | 70.566% (69.918–71.240) |
+| 20% | 299 / 1,493 | 12,528 | 27 | 55.281% (54.207–58.249) | 47.743% (46.925–51.631) | 87.400% (85.313–88.161) | 64.002% (63.220–66.404) |
+| 10% | 149 / 1,493 | 14,144 | 20 | 50.982% (50.365–54.332) | 42.530% (42.187–47.222) | 85.464% (84.667–88.227) | 55.754% (55.227–58.849) |
 
 </div>
 </details>
 
 <details markdown="block">
-<summary><strong>Italian</strong> — <code>it-it-default</code> 1.0.0</summary>
+<summary><strong>Uyghur</strong> — <code>ug-cn-default</code> 1.0.0</summary>
 <div markdown="1">
 
 | Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
 | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
-| 100% | 10,009 / 10,009 | 0 | 0 | n/a | n/a | n/a | 99.056% |
-| 90% | 9,008 / 10,009 | 33,156 | 558 | 93.965% (93.205–94.474) | 93.600% (92.783–94.134) | 99.900% (99.600–100.000) | 98.473% (98.385–98.520) |
-| 80% | 8,007 / 10,009 | 66,376 | 1,008 | 93.594% (93.308–93.801) | 93.192% (92.886–93.418) | 99.900% (99.750–99.950) | 97.816% (97.782–97.848) |
-| 70% | 7,006 / 10,009 | 99,459 | 1,285 | 93.065% (92.894–93.176) | 92.631% (92.450–92.747) | 99.867% (99.800–99.900) | 97.072% (97.043–97.115) |
-| 60% | 6,005 / 10,009 | 132,730 | 1,556 | 92.510% (92.196–92.768) | 92.037% (91.712–92.312) | 99.863% (99.750–99.900) | 96.267% (96.099–96.348) |
-| 50% | 5,005 / 10,009 | 166,665 | 1,637 | 91.882% (91.717–92.283) | 91.375% (91.195–91.801) | 99.840% (99.780–99.920) | 95.296% (95.190–95.476) |
-| 40% | 4,004 / 10,009 | 199,950 | 1,479 | 91.606% (91.107–91.863) | 91.077% (90.547–91.356) | 99.842% (99.783–99.883) | 94.458% (94.135–94.521) |
-| 30% | 3,003 / 10,009 | 234,119 | 1,370 | 90.846% (90.583–91.131) | 90.272% (89.989–90.573) | 99.879% (99.800–99.929) | 93.137% (92.971–93.395) |
-| 20% | 2,002 / 10,009 | 268,626 | 1,035 | 89.611% (89.163–90.122) | 88.964% (88.479–89.500) | 99.888% (99.725–99.963) | 91.388% (91.005–91.839) |
-| 10% | 1,001 / 10,009 | 303,036 | 513 | 87.447% (87.014–87.906) | 86.665% (86.198–87.161) | 99.794% (99.556–99.917) | 88.555% (88.194–88.979) |
+| 100% | 85 / 85 | 0 | 0 | n/a | n/a | n/a | 99.516% |
+| 90% | 77 / 85 | 600 | 0 | 58.586% (47.333–59.500) | 58.020% (46.622–58.953) | 100.000% | 95.593% (94.577–95.718) |
+| 80% | 68 / 85 | 1,287 | 0 | 56.410% (48.796–63.907) | 55.827% (48.110–63.407) | 100.000% | 90.795% (89.217–92.499) |
+| 70% | 60 / 85 | 1,887 | 6 | 52.358% (46.962–60.201) | 51.719% (46.253–59.667) | 100.000% | 85.857% (83.826–87.779) |
+| 60% | 51 / 85 | 2,562 | 6 | 53.747% (47.541–57.426) | 53.125% (46.835–56.845) | 100.000% | 80.997% (78.512–82.716) |
+| 50% | 43 / 85 | 3,143 | 31 | 53.293% (47.211–63.662) | 52.660% (46.494–63.173) | 100.000% | 76.574% (73.637–81.950) |
+| 40% | 34 / 85 | 3,818 | 31 | 54.861% (38.240–60.907) | 54.248% (37.404–60.380) | 100.000% | 72.668% (62.666–76.465) |
+| 30% | 26 / 85 | 4,418 | 25 | 53.128% (36.151–58.328) | 52.493% (35.286–57.766) | 100.000% | 67.292% (55.587–71.011) |
+| 20% | 17 / 85 | 5,099 | 25 | 55.295% (27.123–56.631) | 54.693% (26.138–56.047) | 100.000% | 64.151% (41.538–65.260) |
+| 10% | 9 / 85 | 5,718 | 6 | 45.628% (44.741–46.523) | 44.895% (43.998–45.804) | 100.000% | 51.321% (50.570–52.164) |
 
 </div>
 </details>
 
 <details markdown="block">
-<summary><strong>Norwegian Bokmål</strong> — <code>nb-no-default</code> 1.0.0</summary>
+<summary><strong>Uzbek</strong> — <code>uz-uz-default</code> 1.0.0</summary>
 <div markdown="1">
 
 | Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
 | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
-| 100% | 17,929 / 17,929 | 0 | 0 | n/a | n/a | n/a | 96.852% |
-| 90% | 16,136 / 17,929 | 8,682 | 459 | 79.616% (79.365–80.300) | 70.546% (70.229–71.560) | 95.510% (95.252–96.117) | 95.003% (94.958–95.075) |
-| 80% | 14,343 / 17,929 | 17,357 | 841 | 79.282% (78.391–79.483) | 69.935% (68.919–70.124) | 95.430% (94.759–95.606) | 93.090% (92.952–93.176) |
-| 70% | 12,550 / 17,929 | 26,235 | 1,056 | 78.616% (78.247–79.117) | 69.080% (68.975–69.623) | 95.083% (94.075–95.571) | 91.051% (91.022–91.248) |
-| 60% | 10,757 / 17,929 | 35,053 | 1,222 | 77.888% (77.692–78.230) | 68.373% (67.934–68.693) | 94.544% (94.086–94.654) | 88.955% (88.860–89.131) |
-| 50% | 8,965 / 17,929 | 44,080 | 1,264 | 77.213% (76.853–77.697) | 67.372% (67.269–68.071) | 94.168% (93.324–94.348) | 86.679% (86.578–86.965) |
-| 40% | 7,172 / 17,929 | 53,261 | 1,266 | 76.064% (75.979–76.622) | 66.133% (65.907–66.950) | 93.222% (92.859–93.532) | 84.095% (83.980–84.407) |
-| 30% | 5,379 / 17,929 | 62,496 | 1,103 | 74.899% (74.255–75.323) | 64.655% (63.730–65.508) | 92.387% (92.185–92.553) | 81.227% (80.765–81.536) |
-| 20% | 3,586 / 17,929 | 71,829 | 816 | 72.851% (72.564–73.423) | 62.099% (61.353–62.757) | 91.669% (91.337–91.848) | 77.536% (77.244–77.940) |
-| 10% | 1,793 / 17,929 | 81,276 | 458 | 69.855% (69.368–70.692) | 58.598% (57.048–59.496) | 89.971% (89.224–90.585) | 72.437% (72.066–73.258) |
+| 100% | 295 / 295 | 0 | 0 | n/a | n/a | n/a | 99.970% |
+| 90% | 266 / 295 | 979 | 0 | 94.592% (90.693–95.403) | 94.427% (90.723–95.263) | 100.000% (89.655–100.000) | 99.431% (98.973–99.512) |
+| 80% | 236 / 295 | 1,949 | 0 | 92.825% (90.808–94.144) | 92.760% (90.510–93.966) | 100.000% (94.915–100.000) | 98.526% (98.181–98.790) |
+| 70% | 207 / 295 | 2,928 | 0 | 89.857% (84.879–93.959) | 89.577% (84.539–93.877) | 98.864% (96.591–100.000) | 96.971% (95.162–98.191) |
+| 60% | 177 / 295 | 3,916 | 1 | 91.037% (86.785–94.061) | 90.758% (86.470–93.878) | 99.153% (97.458–100.000) | 96.422% (94.450–97.591) |
+| 50% | 148 / 295 | 4,940 | 1 | 89.758% (88.350–92.571) | 89.548% (88.063–92.364) | 98.639% (96.599–99.320) | 94.816% (93.983–96.239) |
+| 40% | 118 / 295 | 5,951 | 2 | 90.388% (88.432–93.508) | 90.145% (88.165–93.325) | 98.305% (97.175–99.435) | 94.186% (92.936–96.087) |
+| 30% | 89 / 295 | 6,899 | 0 | 89.955% (86.577–93.185) | 89.691% (86.226–92.989) | 98.544% (98.058–99.515) | 92.925% (90.527–95.243) |
+| 20% | 59 / 295 | 7,897 | 1 | 88.920% (81.786–91.118) | 88.631% (81.346–90.883) | 98.729% (96.186–99.576) | 91.096% (85.251–92.905) |
+| 10% | 30 / 295 | 8,871 | 0 | 88.164% (82.260–90.524) | 87.869% (81.784–90.242) | 98.868% (97.736–100.000) | 89.327% (83.970–91.513) |
 
 </div>
 </details>
 
 <details markdown="block">
-<summary><strong>Dutch</strong> — <code>nl-nl-default</code> 1.0.0</summary>
+<summary><strong>Voro</strong> — <code>vro-default</code> 1.0.0</summary>
 <div markdown="1">
 
 | Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
 | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
-| 100% | 4,992 / 4,992 | 0 | 0 | n/a | n/a | n/a | 99.120% |
-| 90% | 4,493 / 4,992 | 3,085 | 57 | 74.468% (73.250–76.272) | 65.891% (63.190–67.730) | 94.512% (92.371–95.697) | 96.574% (96.437–96.777) |
-| 80% | 3,994 / 4,992 | 6,160 | 110 | 74.369% (74.091–75.340) | 64.865% (64.550–66.383) | 94.408% (92.901–95.806) | 94.083% (94.025–94.245) |
-| 70% | 3,494 / 4,992 | 9,298 | 148 | 74.137% (73.135–74.845) | 64.553% (63.215–66.338) | 94.412% (93.155–95.287) | 91.461% (91.190–91.734) |
-| 60% | 2,995 / 4,992 | 12,412 | 171 | 73.828% (73.295–73.872) | 64.368% (63.661–64.668) | 94.242% (93.471–95.070) | 88.817% (88.524–88.845) |
-| 50% | 2,496 / 4,992 | 15,581 | 174 | 73.408% (73.027–73.660) | 63.952% (63.070–64.202) | 94.186% (93.539–94.678) | 86.068% (85.791–86.153) |
-| 40% | 1,997 / 4,992 | 18,714 | 171 | 72.501% (71.803–73.110) | 62.655% (61.524–63.525) | 93.853% (93.540–94.029) | 82.969% (82.495–83.280) |
-| 30% | 1,498 / 4,992 | 21,856 | 141 | 71.097% (70.602–72.004) | 61.026% (59.937–61.731) | 93.565% (92.739–94.166) | 79.432% (79.003–79.969) |
-| 20% | 998 / 4,992 | 25,065 | 109 | 69.543% (68.997–70.471) | 58.908% (57.908–60.208) | 92.781% (92.430–93.088) | 75.447% (74.855–76.168) |
-| 10% | 499 / 4,992 | 28,267 | 62 | 67.039% (65.411–67.577) | 55.524% (53.812–55.891) | 92.616% (90.386–92.866) | 70.231% (68.760–70.692) |
+| 100% | 63 / 63 | 0 | 0 | n/a | n/a | n/a | 99.758% |
+| 90% | 57 / 63 | 41 | 0 | 16.667% (11.111–23.256) | 2.564% (0.000–10.811) | 100.000% (66.667–100.000) | 91.768% (90.073–92.494) |
+| 80% | 50 / 63 | 81 | 0 | 15.957% (14.444–18.519) | 3.077% (0.000–4.938) | 92.308% (76.923–100.000) | 83.777% (80.630–84.262) |
+| 70% | 44 / 63 | 123 | 0 | 16.667% (15.200–17.778) | 2.105% (1.031–6.897) | 89.474% (84.211–94.737) | 75.061% (72.881–76.755) |
+| 60% | 38 / 63 | 162 | 1 | 16.352% (15.476–18.519) | 2.985% (1.504–5.839) | 88.000% (88.000–96.000) | 67.554% (65.375–67.797) |
+| 50% | 32 / 63 | 203 | 1 | 16.256% (15.897–16.432) | 2.326% (1.220–4.396) | 93.548% (87.097–96.774) | 58.596% (56.659–60.048) |
+| 40% | 25 / 63 | 249 | 1 | 15.936% (13.765–16.929) | 2.778% (1.493–3.318) | 86.842% (76.316–97.368) | 48.910% (48.184–51.332) |
+| 30% | 19 / 63 | 288 | 1 | 16.495% (13.732–16.897) | 2.128% (0.410–3.644) | 88.636% (81.818–97.727) | 40.920% (39.952–43.584) |
+| 20% | 13 / 63 | 331 | 1 | 16.566% (14.154–16.928) | 1.859% (0.712–3.546) | 92.000% (84.000–98.000) | 32.688% (31.477–35.593) |
+| 10% | 6 / 63 | 373 | 0 | 15.385% (13.793–17.158) | 1.250% (0.000–2.532) | 94.737% (87.719–98.246) | 23.971% (21.308–24.939) |
 
 </div>
 </details>
 
 <details markdown="block">
-<summary><strong>Norwegian Nynorsk</strong> — <code>nn-no-default</code> 1.0.0</summary>
+<summary><strong>Western Highland Chatino</strong> — <code>ctp-default</code> 1.0.0</summary>
 <div markdown="1">
 
 | Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
 | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
-| 100% | 4,688 / 4,688 | 0 | 0 | n/a | n/a | n/a | 93.089% |
-| 90% | 4,219 / 4,688 | 1,714 | 230 | 78.608% (77.071–80.395) | 70.338% (68.194–72.339) | 95.153% (94.473–97.054) | 91.278% (91.145–91.385) |
-| 80% | 3,750 / 4,688 | 3,536 | 415 | 77.555% (74.632–78.831) | 68.830% (64.861–70.289) | 95.308% (94.801–95.734) | 89.242% (88.866–89.563) |
-| 70% | 3,282 / 4,688 | 5,324 | 569 | 76.278% (73.779–76.665) | 67.482% (64.436–68.148) | 94.306% (93.322–94.907) | 87.054% (86.449–87.247) |
-| 60% | 2,813 / 4,688 | 7,179 | 679 | 74.537% (72.582–75.521) | 65.383% (63.050–67.121) | 93.672% (92.678–95.267) | 84.601% (83.970–84.998) |
-| 50% | 2,344 / 4,688 | 9,118 | 713 | 72.900% (71.989–74.007) | 62.808% (62.003–65.179) | 93.120% (92.681–94.000) | 81.986% (81.456–82.423) |
-| 40% | 1,875 / 4,688 | 11,136 | 714 | 70.773% (70.483–71.887) | 60.799% (60.313–62.437) | 92.181% (91.901–92.710) | 78.826% (78.688–79.390) |
-| 30% | 1,406 / 4,688 | 13,143 | 632 | 68.697% (68.490–69.926) | 58.151% (57.820–59.696) | 91.536% (91.227–91.890) | 75.436% (75.045–75.894) |
-| 20% | 938 / 4,688 | 15,283 | 459 | 66.030% (65.878–67.801) | 55.037% (54.754–57.088) | 90.572% (89.668–91.189) | 71.121% (70.882–72.220) |
-| 10% | 469 / 4,688 | 17,401 | 298 | 63.195% (60.378–64.426) | 51.191% (46.485–52.220) | 90.470% (89.673–91.396) | 66.017% (63.259–66.984) |
+| 100% | 220 / 220 | 0 | 0 | n/a | n/a | n/a | 95.830% |
+| 90% | 198 / 220 | 252 | 18 | 9.865% (6.567–11.792) | 0.000% (0.000–1.667) | 100.000% | 87.343% (84.280–89.004) |
+| 80% | 176 / 220 | 494 | 37 | 9.278% (7.513–10.230) | 0.227% (0.000–1.660) | 100.000% | 79.410% (75.498–80.037) |
+| 70% | 154 / 220 | 791 | 44 | 8.492% (7.933–9.355) | 0.138% (0.000–1.103) | 100.000% | 69.557% (67.749–71.107) |
+| 60% | 132 / 220 | 1,044 | 46 | 8.867% (8.246–9.387) | 0.443% (0.216–1.046) | 100.000% | 61.402% (57.897–63.321) |
+| 50% | 110 / 220 | 1,297 | 52 | 8.999% (8.538–9.484) | 0.509% (0.366–0.927) | 100.000% | 53.579% (50.148–56.642) |
+| 40% | 88 / 220 | 1,582 | 57 | 8.910% (8.661–9.471) | 0.602% (0.490–0.929) | 100.000% | 44.022% (42.177–46.494) |
+| 30% | 66 / 220 | 1,830 | 49 | 9.244% (8.502–9.344) | 0.731% (0.347–1.014) | 100.000% | 36.421% (34.022–37.934) |
+| 20% | 44 / 220 | 2,109 | 40 | 9.226% (8.772–9.247) | 0.840% (0.466–1.164) | 100.000% | 27.122% (26.347–29.336) |
+| 10% | 22 / 220 | 2,412 | 19 | 9.000% (8.708–9.731) | 0.859% (0.687–1.669) | 100.000% | 18.413% (16.384–19.299) |
 
 </div>
 </details>
 
 <details markdown="block">
-<summary><strong>Polish</strong> — <code>pl-pl-unimorph</code> 1.0.0</summary>
+<summary><strong>Xibe</strong> — <code>sjo-default</code> 1.0.0</summary>
 <div markdown="1">
 
 | Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
 | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
-| 100% | 9,990 / 9,990 | 0 | 0 | n/a | n/a | n/a | 98.837% |
-| 90% | 8,991 / 9,990 | 12,963 | 296 | 73.070% (71.833–74.408) | 69.972% (68.686–71.530) | 90.985% (89.948–91.264) | 96.179% (96.039–96.269) |
-| 80% | 7,992 / 9,990 | 25,947 | 560 | 73.298% (71.542–73.560) | 70.344% (68.301–70.589) | 90.225% (89.430–90.594) | 93.619% (93.242–93.666) |
-| 70% | 6,993 / 9,990 | 39,130 | 726 | 72.217% (71.567–72.865) | 69.231% (68.375–70.016) | 89.428% (88.990–89.729) | 90.660% (90.510–90.869) |
-| 60% | 5,994 / 9,990 | 52,228 | 831 | 71.858% (71.219–72.488) | 68.868% (68.033–69.513) | 89.261% (88.574–89.356) | 87.814% (87.665–88.121) |
-| 50% | 4,995 / 9,990 | 65,434 | 856 | 71.160% (70.975–71.928) | 68.066% (67.805–69.051) | 88.517% (88.330–89.078) | 84.847% (84.707–85.211) |
-| 40% | 3,996 / 9,990 | 78,368 | 857 | 70.344% (69.790–71.025) | 67.165% (66.449–67.971) | 88.216% (88.039–88.588) | 81.587% (81.365–81.994) |
-| 30% | 2,997 / 9,990 | 91,501 | 747 | 69.088% (68.299–70.063) | 65.677% (64.723–66.909) | 88.256% (87.960–88.753) | 78.020% (77.407–78.451) |
-| 20% | 1,998 / 9,990 | 105,119 | 565 | 68.340% (67.352–68.619) | 64.876% (63.830–65.097) | 87.827% (87.200–88.611) | 74.371% (73.606–74.469) |
-| 10% | 999 / 9,990 | 118,711 | 291 | 65.763% (65.395–67.001) | 61.996% (61.861–63.283) | 86.633% (85.310–87.965) | 69.126% (68.742–70.110) |
+| 100% | 1,892 / 1,892 | 0 | 0 | n/a | n/a | n/a | 98.561% |
+| 90% | 1,703 / 1,892 | 318 | 9 | 79.866% (73.030–81.290) | 52.679% (43.537–61.333) | 96.721% (96.237–98.396) | 96.528% (95.777–96.747) |
+| 80% | 1,514 / 1,892 | 635 | 16 | 76.850% (73.293–79.276) | 49.412% (43.007–50.704) | 96.748% (95.957–97.838) | 93.963% (93.150–94.683) |
+| 70% | 1,324 / 1,892 | 961 | 22 | 75.739% (74.298–78.906) | 47.393% (42.750–49.112) | 96.791% (95.528–97.138) | 91.398% (90.960–92.774) |
+| 60% | 1,135 / 1,892 | 1,248 | 24 | 75.594% (75.492–78.565) | 48.055% (44.356–50.538) | 96.390% (93.307–96.770) | 89.240% (88.771–90.835) |
+| 50% | 946 / 1,892 | 1,563 | 25 | 75.446% (74.344–76.933) | 46.114% (43.153–48.837) | 95.708% (94.984–96.245) | 86.612% (86.519–87.989) |
+| 40% | 757 / 1,892 | 1,900 | 26 | 75.258% (73.895–75.592) | 45.600% (43.445–47.842) | 95.009% (94.831–96.515) | 84.360% (83.516–84.673) |
+| 30% | 568 / 1,892 | 2,233 | 21 | 74.563% (73.753–76.312) | 46.630% (44.066–48.229) | 94.985% (94.136–95.814) | 81.514% (81.045–83.015) |
+| 20% | 378 / 1,892 | 2,554 | 15 | 73.761% (72.122–75.020) | 43.824% (41.563–46.476) | 93.563% (93.422–96.021) | 78.449% (77.166–80.044) |
+| 10% | 189 / 1,892 | 2,870 | 8 | 72.485% (71.533–75.079) | 42.002% (36.325–46.127) | 94.879% (91.995–96.351) | 74.820% (74.195–77.385) |
 
 </div>
 </details>
 
 <details markdown="block">
-<summary><strong>Portuguese</strong> — <code>pt-pt-default</code> 1.0.0</summary>
+<summary><strong>Xipinawa</strong> — <code>xty-default</code> 1.0.0</summary>
 <div markdown="1">
 
 | Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
 | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
-| 100% | 4,001 / 4,001 | 0 | 0 | n/a | n/a | n/a | 99.815% |
-| 90% | 3,601 / 4,001 | 21,519 | 66 | 94.620% (93.771–95.072) | 94.412% (93.530–94.882) | 100.000% (99.749–100.000) | 99.279% (99.194–99.326) |
-| 80% | 3,201 / 4,001 | 43,021 | 113 | 94.273% (93.721–94.673) | 94.056% (93.478–94.472) | 100.000% (99.875–100.000) | 98.675% (98.576–98.762) |
-| 70% | 2,801 / 4,001 | 64,495 | 146 | 93.982% (93.731–94.081) | 93.752% (93.489–93.852) | 100.000% (99.917–100.000) | 98.035% (97.952–98.049) |
-| 60% | 2,401 / 4,001 | 85,970 | 216 | 93.987% (93.386–94.179) | 93.758% (93.130–93.957) | 99.938% (99.937–100.000) | 97.430% (97.204–97.509) |
-| 50% | 2,001 / 4,001 | 107,492 | 243 | 93.601% (93.084–94.280) | 93.358% (92.822–94.061) | 99.900% (99.800–100.000) | 96.646% (96.404–97.001) |
-| 40% | 1,600 / 4,001 | 129,092 | 242 | 93.154% (92.758–94.021) | 92.889% (92.480–93.795) | 99.958% (99.875–100.000) | 95.791% (95.526–96.281) |
-| 30% | 1,200 / 4,001 | 150,645 | 201 | 92.584% (91.628–93.106) | 92.297% (91.306–92.854) | 99.893% (99.643–100.000) | 94.717% (94.035–95.078) |
-| 20% | 800 / 4,001 | 172,258 | 154 | 91.775% (90.674–92.113) | 91.460% (90.315–91.809) | 99.938% (99.750–100.000) | 93.379% (92.481–93.616) |
-| 10% | 400 / 4,001 | 193,881 | 56 | 90.423% (88.898–91.005) | 90.053% (88.476–90.666) | 99.833% (99.778–100.000) | 91.313% (89.988–91.878) |
+| 100% | 585 / 585 | 0 | 0 | n/a | n/a | n/a | 99.029% |
+| 90% | 527 / 585 | 269 | 5 | 28.839% (25.455–32.584) | 12.440% (10.599–16.746) | 89.474% (81.034–91.228) | 92.197% (91.658–92.557) |
+| 80% | 468 / 585 | 545 | 9 | 28.962% (28.336–30.882) | 12.500% (11.601–14.988) | 89.744% (88.889–94.690) | 85.257% (84.933–85.509) |
+| 70% | 410 / 585 | 816 | 11 | 28.360% (27.406–28.571) | 11.076% (10.526–12.112) | 89.714% (88.506–94.152) | 78.029% (77.778–78.317) |
+| 60% | 351 / 585 | 1,098 | 14 | 26.993% (26.648–27.778) | 10.115% (9.186–10.814) | 90.598% (88.793–93.913) | 70.262% (70.191–70.802) |
+| 50% | 293 / 585 | 1,378 | 17 | 26.084% (25.752–26.763) | 9.065% (8.489–9.446) | 91.034% (88.660–92.734) | 62.855% (62.531–62.999) |
+| 40% | 234 / 585 | 1,656 | 15 | 25.260% (24.880–25.604) | 7.573% (6.870–8.187) | 90.831% (90.000–92.286) | 54.908% (54.585–55.304) |
+| 30% | 176 / 585 | 1,930 | 14 | 24.271% (23.575–25.538) | 6.415% (5.581–7.842) | 91.667% (90.441–92.647) | 46.926% (46.422–47.717) |
+| 20% | 117 / 585 | 2,208 | 12 | 23.535% (22.703–24.784) | 5.422% (4.107–6.452) | 91.863% (90.558–92.934) | 38.583% (38.044–39.842) |
+| 10% | 59 / 585 | 2,490 | 8 | 22.427% (21.526–23.387) | 3.406% (3.093–3.478) | 94.667% (89.333–97.524) | 30.133% (29.414–31.140) |
 
 </div>
 </details>
 
 <details markdown="block">
-<summary><strong>Russian</strong> — <code>ru-ru-default</code> 1.0.0</summary>
+<summary><strong>Yamana</strong> — <code>ame-default</code> 1.0.0</summary>
 <div markdown="1">
 
 | Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
 | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
-| 100% | 37,410 / 37,410 | 0 | 0 | n/a | n/a | n/a | 98.807% |
-| 90% | 33,669 / 37,410 | 79,477 | 1,784 | 82.007% (81.842–82.574) | 80.744% (80.515–81.372) | 94.506% (94.455–94.967) | 97.047% (97.012–97.088) |
-| 80% | 29,928 / 37,410 | 158,745 | 3,165 | 81.665% (81.300–82.243) | 80.343% (79.956–81.034) | 94.359% (93.988–94.456) | 95.273% (95.172–95.333) |
-| 70% | 26,187 / 37,410 | 238,018 | 4,273 | 81.575% (81.414–81.952) | 80.296% (80.115–80.758) | 94.006% (93.750–94.173) | 93.468% (93.397–93.575) |
-| 60% | 22,446 / 37,410 | 318,625 | 4,709 | 81.182% (80.809–81.611) | 79.915% (79.506–80.371) | 93.632% (93.525–93.734) | 91.517% (91.394–91.717) |
-| 50% | 18,705 / 37,410 | 398,725 | 5,007 | 80.854% (80.304–81.219) | 79.578% (78.976–79.972) | 93.316% (93.219–93.447) | 89.614% (89.371–89.752) |
-| 40% | 14,964 / 37,410 | 479,920 | 4,763 | 80.312% (79.666–80.706) | 79.019% (78.331–79.445) | 93.010% (92.649–93.088) | 87.513% (87.184–87.695) |
-| 30% | 11,223 / 37,410 | 560,108 | 4,217 | 79.739% (79.241–79.963) | 78.423% (77.874–78.684) | 92.571% (92.488–92.673) | 85.305% (84.992–85.415) |
-| 20% | 7,482 / 37,410 | 640,968 | 3,381 | 78.648% (78.391–79.002) | 77.254% (76.977–77.673) | 91.951% (91.925–92.274) | 82.544% (82.404–82.868) |
-| 10% | 3,741 / 37,410 | 723,165 | 1,916 | 76.473% (76.268–77.001) | 74.948% (74.735–75.545) | 91.268% (90.883–91.995) | 78.669% (78.459–79.115) |
+| 100% | 326 / 326 | 0 | 0 | n/a | n/a | n/a | 98.763% |
+| 90% | 293 / 326 | 259 | 5 | 16.732% (16.484–17.899) | 5.000% (4.762–6.250) | 96.970% (96.970–100.000) | 90.705% (90.217–90.855) |
+| 80% | 261 / 326 | 516 | 7 | 16.667% (16.176–17.485) | 4.922% (4.595–5.856) | 98.462% (96.923–100.000) | 82.759% (81.672–83.021) |
+| 70% | 228 / 326 | 781 | 14 | 16.710% (15.507–17.562) | 5.362% (4.161–5.417) | 96.939% (95.918–100.000) | 74.475% (72.939–75.187) |
+| 60% | 196 / 326 | 1,039 | 18 | 16.957% (15.508–16.977) | 4.949% (3.961–5.476) | 97.692% (97.692–99.231) | 66.754% (65.180–67.091) |
+| 50% | 163 / 326 | 1,307 | 17 | 16.333% (15.264–16.718) | 4.720% (3.729–5.340) | 97.546% (96.933–99.387) | 58.283% (56.259–58.508) |
+| 40% | 130 / 326 | 1,577 | 18 | 16.337% (15.255–16.489) | 4.566% (3.617–5.147) | 98.469% (96.939–98.980) | 49.213% (47.939–50.637) |
+| 30% | 98 / 326 | 1,846 | 18 | 15.855% (14.855–16.502) | 4.483% (3.006–4.825) | 98.246% (95.614–99.561) | 40.855% (39.843–42.466) |
+| 20% | 65 / 326 | 2,118 | 12 | 14.932% (13.651–17.117) | 3.507% (2.716–5.628) | 97.318% (92.337–98.467) | 31.897% (30.210–34.183) |
+| 10% | 33 / 326 | 2,398 | 6 | 14.807% (14.012–16.548) | 3.872% (2.670–5.048) | 98.976% (93.857–99.317) | 23.126% (22.151–25.037) |
 
 </div>
 </details>
 
 <details markdown="block">
-<summary><strong>Swedish</strong> — <code>sv-se-default</code> 1.0.0</summary>
+<summary><strong>Yiddish</strong> — <code>yi-default</code> 1.0.0</summary>
 <div markdown="1">
 
 | Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
 | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
-| 100% | 12,371 / 12,371 | 0 | 0 | n/a | n/a | n/a | 96.713% |
-| 90% | 11,134 / 12,371 | 10,386 | 675 | 82.188% (81.351–82.660) | 78.458% (77.818–79.022) | 94.926% (93.807–95.482) | 95.085% (95.025–95.156) |
-| 80% | 9,897 / 12,371 | 20,899 | 1,144 | 81.612% (81.194–82.294) | 78.137% (77.386–78.776) | 94.577% (93.139–94.972) | 93.421% (93.317–93.570) |
-| 70% | 8,660 / 12,371 | 31,613 | 1,500 | 81.378% (80.737–81.694) | 77.755% (77.158–78.370) | 93.450% (93.333–94.124) | 91.702% (91.610–91.808) |
-| 60% | 7,423 / 12,371 | 42,503 | 1,726 | 80.584% (80.441–80.710) | 77.113% (76.939–77.166) | 93.114% (92.742–93.192) | 89.845% (89.812–89.898) |
-| 50% | 6,186 / 12,371 | 53,431 | 1,736 | 79.553% (79.081–79.924) | 75.813% (75.376–76.278) | 92.204% (92.067–92.705) | 87.777% (87.484–87.889) |
-| 40% | 4,948 / 12,371 | 64,540 | 1,677 | 78.689% (78.113–78.970) | 74.894% (74.347–75.197) | 91.778% (91.278–92.147) | 85.502% (85.283–85.636) |
-| 30% | 3,711 / 12,371 | 75,916 | 1,458 | 77.249% (76.663–77.381) | 73.248% (72.517–73.345) | 91.335% (90.486–91.670) | 82.749% (82.497–82.863) |
-| 20% | 2,474 / 12,371 | 87,316 | 1,095 | 75.205% (74.488–75.890) | 70.861% (70.211–72.027) | 89.668% (89.346–90.344) | 79.323% (78.735–79.852) |
-| 10% | 1,237 / 12,371 | 98,793 | 636 | 72.010% (70.983–72.505) | 67.214% (66.144–68.191) | 87.789% (86.974–88.659) | 74.396% (73.426–74.785) |
+| 100% | 802 / 802 | 0 | 0 | n/a | n/a | n/a | 98.698% |
+| 90% | 722 / 802 | 413 | 12 | 60.291% (58.962–66.986) | 45.318% (42.007–51.673) | 93.007% (87.500–94.631) | 94.907% (94.651–95.558) |
+| 80% | 642 / 802 | 823 | 24 | 60.402% (59.246–64.234) | 45.180% (42.938–49.067) | 90.508% (84.028–92.388) | 90.930% (90.907–91.767) |
+| 70% | 561 / 802 | 1,258 | 26 | 61.924% (60.048–63.121) | 46.748% (43.499–48.335) | 90.562% (86.577–92.511) | 87.512% (87.070–87.860) |
+| 60% | 481 / 802 | 1,687 | 34 | 59.833% (59.576–62.234) | 45.164% (42.091–46.740) | 89.933% (86.464–91.792) | 83.070% (82.767–83.930) |
+| 50% | 401 / 802 | 2,115 | 38 | 60.189% (57.729–64.180) | 44.428% (39.314–49.554) | 90.483% (88.108–91.345) | 79.256% (77.953–81.326) |
+| 40% | 321 / 802 | 2,542 | 34 | 60.386% (59.274–63.676) | 43.712% (43.647–48.674) | 88.840% (87.723–91.081) | 75.419% (74.884–77.488) |
+| 30% | 241 / 802 | 2,983 | 26 | 57.138% (55.881–62.209) | 39.886% (37.598–46.865) | 88.752% (87.027–90.258) | 69.535% (68.535–72.930) |
+| 20% | 160 / 802 | 3,423 | 19 | 56.968% (54.863–59.445) | 38.615% (36.885–43.362) | 88.588% (86.935–90.646) | 65.186% (63.488–67.140) |
+| 10% | 80 / 802 | 3,853 | 14 | 56.133% (53.401–59.770) | 37.767% (32.786–44.588) | 89.466% (84.791–91.028) | 60.233% (57.884–63.419) |
 
 </div>
 </details>
 
 <details markdown="block">
-<summary><strong>Ukrainian</strong> — <code>uk-ua-default</code> 1.0.0</summary>
+<summary><strong>Zacatepec Chatino</strong> — <code>zpv-default</code> 1.0.0</summary>
 <div markdown="1">
 
 | Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
 | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
-| 100% | 1,493 / 1,493 | 0 | 0 | n/a | n/a | n/a | 99.307% |
-| 90% | 1,344 / 1,493 | 1,576 | 12 | 63.233% (61.940–65.736) | 57.218% (55.192–59.594) | 90.476% (88.514–93.243) | 95.673% (95.418–95.876) |
-| 80% | 1,194 / 1,493 | 3,166 | 34 | 62.236% (61.727–64.985) | 56.335% (54.703–59.359) | 88.475% (87.879–93.255) | 91.987% (91.580–92.514) |
-| 70% | 1,045 / 1,493 | 4,675 | 36 | 62.139% (60.712–63.062) | 55.456% (54.248–56.768) | 89.367% (85.811–90.562) | 87.990% (87.526–88.575) |
-| 60% | 896 / 1,493 | 6,256 | 53 | 61.416% (58.976–62.308) | 54.708% (52.041–55.903) | 90.211% (87.669–91.525) | 84.076% (83.332–84.393) |
-| 50% | 747 / 1,493 | 7,838 | 50 | 60.214% (59.232–62.314) | 53.268% (51.925–55.648) | 90.135% (88.140–91.180) | 79.793% (79.138–80.587) |
-| 40% | 597 / 1,493 | 9,392 | 66 | 58.933% (57.994–60.117) | 51.880% (50.380–53.150) | 89.013% (87.486–90.578) | 75.097% (74.309–75.599) |
-| 30% | 448 / 1,493 | 10,951 | 60 | 58.222% (57.492–59.390) | 51.048% (50.564–52.388) | 87.380% (86.917–89.403) | 70.566% (69.918–71.240) |
-| 20% | 299 / 1,493 | 12,528 | 27 | 55.281% (54.207–58.249) | 47.743% (46.925–51.631) | 87.400% (85.313–88.161) | 64.002% (63.220–66.404) |
-| 10% | 149 / 1,493 | 14,144 | 20 | 50.982% (50.365–54.332) | 42.530% (42.187–47.222) | 85.464% (84.667–88.227) | 55.754% (55.227–58.849) |
+| 100% | 204 / 204 | 0 | 0 | n/a | n/a | n/a | 99.272% |
+| 90% | 184 / 204 | 79 | 2 | 25.000% (23.077–25.641) | 0.000% | 100.000% | 91.990% (91.990–92.233) |
+| 80% | 163 / 204 | 164 | 2 | 24.845% (23.171–25.625) | 0.000% | 100.000% (97.436–100.000) | 84.345% (83.981–84.830) |
+| 70% | 143 / 204 | 243 | 3 | 24.797% (23.108–25.417) | 0.000% | 100.000% (96.667–100.000) | 77.184% (75.971–77.549) |
+| 60% | 122 / 204 | 328 | 4 | 24.847% (23.582–25.231) | 0.000% | 98.780% (97.531–100.000) | 69.175% (68.325–69.782) |
+| 50% | 102 / 204 | 409 | 3 | 24.450% (23.798–24.938) | 0.000% | 98.039% (98.020–99.020) | 61.893% (60.922–62.500) |
+| 40% | 82 / 204 | 489 | 3 | 24.089% (23.313–24.846) | 0.000% | 98.347% (94.215–99.180) | 54.005% (53.762–54.976) |
+| 30% | 61 / 204 | 576 | 3 | 24.394% (23.438–24.956) | 0.000% | 98.601% (95.070–100.000) | 46.602% (45.995–47.451) |
+| 20% | 41 / 204 | 657 | 2 | 24.468% (23.860–24.924) | 0.000% | 98.773% (96.319–100.000) | 39.442% (38.956–40.170) |
+| 10% | 20 / 204 | 741 | 0 | 24.462% (24.324–24.865) | 0.000% | 98.913% (97.826–100.000) | 31.796% (31.553–32.524) |
 
 </div>
 </details>
 
 <details markdown="block">
-<summary><strong>English</strong> — <code>us-uk-default</code> 1.0.1</summary>
+<summary><strong>Zarma</strong> — <code>dje-default</code> 1.0.0</summary>
 <div markdown="1">
 
 | Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
 | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
-| 100% | 396,939 / 396,939 | 0 | 0 | n/a | n/a | n/a | 97.668% |
-| 90% | 357,245 / 396,939 | 96,702 | 3,746 | 94.680% (94.609–94.847) | 78.854% (78.538–79.603) | 98.744% (98.642–98.790) | 97.241% (97.237–97.256) |
-| 80% | 317,551 / 396,939 | 193,781 | 6,659 | 94.608% (94.514–94.637) | 78.795% (78.580–79.033) | 98.592% (98.577–98.658) | 96.818% (96.816–96.831) |
-| 70% | 277,857 / 396,939 | 291,831 | 8,726 | 94.425% (94.290–94.438) | 78.505% (77.971–78.614) | 98.454% (98.436–98.500) | 96.386% (96.367–96.394) |
-| 60% | 238,163 / 396,939 | 390,750 | 10,090 | 94.178% (94.121–94.242) | 77.976% (77.824–78.311) | 98.314% (98.268–98.319) | 95.919% (95.916–95.953) |
-| 50% | 198,470 / 396,939 | 490,475 | 10,590 | 93.934% (93.888–94.011) | 77.643% (77.410–77.726) | 98.133% (98.089–98.192) | 95.434% (95.421–95.484) |
-| 40% | 158,776 / 396,939 | 591,318 | 10,198 | 93.699% (93.682–93.712) | 77.308% (77.149–77.479) | 97.940% (97.912–97.948) | 94.946% (94.916–94.960) |
-| 30% | 119,082 / 396,939 | 692,940 | 8,930 | 93.440% (93.409–93.459) | 76.944% (76.715–76.961) | 97.741% (97.718–97.755) | 94.413% (94.372–94.433) |
-| 20% | 79,388 / 396,939 | 795,287 | 6,816 | 93.155% (93.122–93.220) | 76.608% (76.306–76.669) | 97.509% (97.464–97.541) | 93.824% (93.807–93.898) |
-| 10% | 39,694 / 396,939 | 898,572 | 3,741 | 92.805% (92.757–92.843) | 76.010% (75.505–76.378) | 97.184% (97.156–97.293) | 93.157% (93.123–93.211) |
+| 100% | 27 / 27 | 0 | 0 | n/a | n/a | n/a | 100.000% |
+| 90% | 24 / 27 | 9 | 0 | 33.333% | 0.000% | 100.000% | 92.593% |
+| 80% | 22 / 27 | 15 | 0 | 33.333% | 0.000% | 100.000% | 87.654% |
+| 70% | 19 / 27 | 24 | 0 | 33.333% | 0.000% | 100.000% | 80.247% |
+| 60% | 16 / 27 | 33 | 0 | 33.333% | 0.000% | 100.000% | 72.840% |
+| 50% | 14 / 27 | 39 | 0 | 33.333% | 0.000% | 100.000% | 67.901% |
+| 40% | 11 / 27 | 48 | 0 | 33.333% | 0.000% | 100.000% | 60.494% |
+| 30% | 8 / 27 | 57 | 0 | 33.333% | 0.000% | 100.000% | 53.086% |
+| 20% | 5 / 27 | 66 | 0 | 33.333% | 0.000% | 100.000% | 45.679% |
+| 10% | 3 / 27 | 72 | 0 | 33.333% | 0.000% | 100.000% | 40.741% |
 
 </div>
 </details>
 
 <details markdown="block">
-<summary><strong>Yiddish</strong> — <code>yi-default</code> 1.0.0</summary>
+<summary><strong>Zenzontepec Chatino</strong> — <code>czn-default</code> 1.0.0</summary>
 <div markdown="1">
 
 | Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
 | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
-| 100% | 802 / 802 | 0 | 0 | n/a | n/a | n/a | 98.698% |
-| 90% | 722 / 802 | 413 | 12 | 60.291% (58.962–66.986) | 45.318% (42.007–51.673) | 93.007% (87.500–94.631) | 94.907% (94.651–95.558) |
-| 80% | 642 / 802 | 823 | 24 | 60.402% (59.246–64.234) | 45.180% (42.938–49.067) | 90.508% (84.028–92.388) | 90.930% (90.907–91.767) |
-| 70% | 561 / 802 | 1,258 | 26 | 61.924% (60.048–63.121) | 46.748% (43.499–48.335) | 90.562% (86.577–92.511) | 87.512% (87.070–87.860) |
-| 60% | 481 / 802 | 1,687 | 34 | 59.833% (59.576–62.234) | 45.164% (42.091–46.740) | 89.933% (86.464–91.792) | 83.070% (82.767–83.930) |
-| 50% | 401 / 802 | 2,115 | 38 | 60.189% (57.729–64.180) | 44.428% (39.314–49.554) | 90.483% (88.108–91.345) | 79.256% (77.953–81.326) |
-| 40% | 321 / 802 | 2,542 | 34 | 60.386% (59.274–63.676) | 43.712% (43.647–48.674) | 88.840% (87.723–91.081) | 75.419% (74.884–77.488) |
-| 30% | 241 / 802 | 2,983 | 26 | 57.138% (55.881–62.209) | 39.886% (37.598–46.865) | 88.752% (87.027–90.258) | 69.535% (68.535–72.930) |
-| 20% | 160 / 802 | 3,423 | 19 | 56.968% (54.863–59.445) | 38.615% (36.885–43.362) | 88.588% (86.935–90.646) | 65.186% (63.488–67.140) |
-| 10% | 80 / 802 | 3,853 | 14 | 56.133% (53.401–59.770) | 37.767% (32.786–44.588) | 89.466% (84.791–91.028) | 60.233% (57.884–63.419) |
+| 100% | 386 / 386 | 0 | 0 | n/a | n/a | n/a | 97.337% |
+| 90% | 347 / 386 | 185 | 9 | 20.000% (19.149–20.000) | 0.000% | 94.737% (94.737–97.368) | 89.661% (89.399–89.817) |
+| 80% | 309 / 386 | 365 | 17 | 19.780% (19.452–20.596) | 0.000% | 96.000% (94.667–98.701) | 82.089% (81.723–82.298) |
+| 70% | 270 / 386 | 553 | 21 | 20.253% (19.821–20.615) | 0.000% | 97.391% (95.690–99.130) | 74.465% (74.047–74.674) |
+| 60% | 232 / 386 | 739 | 25 | 20.298% (20.134–20.353) | 0.000% | 98.039% (97.368–98.693) | 66.945% (66.632–67.206) |
+| 50% | 193 / 386 | 931 | 26 | 20.364% (20.151–20.408) | 0.000% | 98.446% (97.906–98.958) | 59.164% (58.956–59.582) |
+| 40% | 154 / 386 | 1,126 | 26 | 20.213% (20.125–20.321) | 0.000% | 98.701% (98.261–99.134) | 51.227% (51.175–51.593) |
+| 30% | 116 / 386 | 1,318 | 23 | 20.243% (20.091–20.258) | 0.000% | 99.254% (98.513–99.627) | 43.655% (43.603–43.864) |
+| 20% | 77 / 386 | 1,515 | 19 | 20.198% (19.947–20.305) | 0.000% | 99.673% (98.377–100.000) | 35.822% (35.718–35.979) |
+| 10% | 39 / 386 | 1,710 | 9 | 20.175% (19.977–20.258) | 0.000% | 99.711% (98.844–100.000) | 28.094% (27.937–28.251) |
+
+</div>
+</details>
+
+<details markdown="block">
+<summary><strong>Zulu</strong> — <code>zu-za-default</code> 1.0.0</summary>
+<div markdown="1">
+
+| Training | Selected / total rows | Median unseen occurrences | Overlap excluded | Unseen all exact | Unseen changed exact | Unseen root preserved | Whole all exact |
+| ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
+| 100% | 563 / 563 | 0 | 0 | n/a | n/a | n/a | 98.931% |
+| 90% | 507 / 563 | 3,174 | 75 | 2.148% (2.079–2.192) | 0.417% (0.186–0.657) | 100.000% (98.182–100.000) | 89.436% (88.300–90.728) |
+| 80% | 450 / 563 | 6,423 | 133 | 2.143% (2.114–2.191) | 0.441% (0.327–0.534) | 100.000% (99.083–100.000) | 79.810% (79.654–81.304) |
+| 70% | 394 / 563 | 9,576 | 172 | 2.141% (2.062–2.200) | 0.424% (0.390–0.514) | 100.000% (98.780–100.000) | 70.398% (69.478–71.745) |
+| 60% | 338 / 563 | 12,719 | 181 | 2.123% (2.112–2.171) | 0.408% (0.399–0.499) | 100.000% (99.095–100.000) | 61.089% (60.698–61.881) |
+| 50% | 282 / 563 | 15,762 | 174 | 2.115% (2.049–2.155) | 0.384% (0.304–0.495) | 99.639% (99.632–100.000) | 52.099% (50.626–52.276) |
+| 40% | 225 / 563 | 19,546 | 187 | 2.089% (2.047–2.159) | 0.379% (0.315–0.463) | 99.698% (99.398–100.000) | 40.939% (40.078–42.903) |
+| 30% | 169 / 563 | 22,899 | 156 | 2.114% (2.055–2.162) | 0.387% (0.359–0.474) | 99.239% (98.966–100.000) | 30.861% (29.877–33.369) |
+| 20% | 113 / 563 | 26,177 | 87 | 2.116% (2.021–2.137) | 0.406% (0.345–0.438) | 99.552% (98.876–99.555) | 21.271% (20.272–24.018) |
+| 10% | 56 / 563 | 29,332 | 43 | 2.041% (1.957–2.090) | 0.320% (0.262–0.378) | 99.604% (99.208–99.803) | 11.951% (11.147–13.167) |
 
 </div>
 </details>
 
 ## Provenance
 
-- Radixor/Java: `4.2.0-6-g84e57fb`
-- Core source revision: `84e57fb27ae40913569e826858a8ebb07cf2ea01`
-- Source state: `dirty`
-- Generalization generator SHA-256: `126964f216a48d2164d136a490698301111aae7f050209ced21caf73f8c01fb8`
-- Measured-source manifest: [`dictionary-generalization-sources.sha256`](data/dictionary-generalization-sources.sha256)
+- Measurement provenance is recorded per row. The active snapshot deliberately combines the frozen 20-language archive with the separately measured 123-language continuation.
+- Historical 20-language source manifest: [`dictionary-generalization-sources.sha256`](data/dictionary-generalization-sources.sha256)
+- Continuation source manifest: [`dictionary-generalization-2026-09-11-sources.txt`](data/dictionary-generalization-2026-09-11-sources.txt)
 - Split protocol: `radixor-generalization-v1`
 - Splits per coverage level: 5
-- Authoritative raw counters: [`dictionary-generalization.csv`](data/dictionary-generalization.csv)
-- CSV SHA-256: `e6479840b9307ae03bd0873e55f397811e975125d621a8b8716d4c1a166b3ff2`
+- Authoritative raw counters: [`dictionary-generalization-2026-09-11.csv`](data/dictionary-generalization-2026-09-11.csv)
+- CSV SHA-256: `187a9fe7b7bb292ac8d36b43784dc63a6e6e749942e86f20e7cb26364b7ea8d5`
 - Model artifact IDs, independent versions, and SHA-256 values are recorded on every raw row.
 - Runtime speed is intentionally excluded: speed does not establish generalization. The
   [English coverage deep dive](reference/english-coverage.md) retains its separately measured
diff --git a/docs/benchmarks/index.md b/docs/benchmarks/index.md
index 9a4eb65..93c2fbb 100644
--- a/docs/benchmarks/index.md
+++ b/docs/benchmarks/index.md
@@ -54,7 +54,7 @@ Each language page contains:
 - throughput metrics,
 - pairwise linguistic-quality metrics showing whether same-group forms share a stem and forms with
   no shared dictionary-group membership remain separated,
-- a language-specific edit-cost and dictionary-knowledge evidence table, factor analysis, and conclusion,
+- for the historical 20-language cohort only, the separately retained edit-cost and dictionary-knowledge analysis,
 - interpretation notes for the compared stemmers.
 
 Open [Language Benchmark Pages](languages/index.md) for the complete language list.
@@ -63,8 +63,8 @@ Open [Language Benchmark Pages](languages/index.md) for the complete language li
 
 The English dictionary coverage benchmark shows the current contracted-trie operating curve. With
 the full English dictionary, Radixor reaches `97.668%` all-token exactness and `98.110%`
-changed-token exactness at `97.9 ns/token`. Even with a deterministic 10% dictionary slice, it
-keeps `93.057%` all-token exactness and `77.327%` changed-token exactness at `68.9 ns/token`.
+changed-token exactness at `120.4 ns/token`. Even with a deterministic 10% dictionary slice, it
+keeps `93.057%` all-token exactness and `77.327%` changed-token exactness at `71.6 ns/token`.
 
 Those figures should not be reduced to a single speed badge. The professional interpretation is a
 quality/speed envelope: the amount and quality of dictionary knowledge affect stemming precision,
@@ -72,7 +72,7 @@ while contracted tries reduce lookup cost in uniform regions of the compiled gra
 
 ## Quality versus performance
 
-Each language page keeps exact-root accuracy, JMH latency, and pairwise linguistic-quality results in separate tables. No undocumented scalar combines them. The 2026-08-25 language tables are generated from the current corpus/command report, exact-root and speed JMH reports, and pairwise-quality snapshot produced for this refresh. The Snowball 3.1.0 matrix includes direct Czech, Persian, and Polish stemmers; all published Java stemmers were measured in the same run. Readers should inspect the quality and speed dimensions side by side; no cross-language Pareto ranking is inferred from workloads with different dictionaries and token counts.
+Each language page keeps exact-root accuracy, JMH latency, and pairwise linguistic-quality results in separate tables. No undocumented scalar combines them. The 2026-09-10 Java tables are generated from the published corpus/command, exact-root, and speed reports; the richer pairwise-quality sections on the original 20 pages remain bound to their separately frozen snapshot. The Snowball 3.1.0 rows appear only where the implementation exactly matches the model language. Readers should inspect quality and speed side by side; no cross-language Pareto ranking is inferred from workloads with different dictionaries and token counts.
 
 ### New Snowball 3.1.0 rows
 
@@ -88,55 +88,301 @@ These rows describe exact agreement with each Radixor model dictionary and the m
 
 ## Pairwise Quality Findings
 
-The validated snapshot is a broad multilingual comparison covering the complete 20-language Radixor default-model universe, with one benchmark page per language. The direct ranking below uses only deterministic `PRIMARY_OUTPUT` rows over identical per-language inputs. Candidate-aware rows are intentionally excluded from this claim.
+The validated snapshot is a broad multilingual comparison covering the complete 143-language Radixor default-model universe, with one benchmark page per language. The direct ranking below uses only deterministic `PRIMARY_OUTPUT` rows over identical per-language inputs. Candidate-aware rows are intentionally excluded from this claim.
 
 !!! success "Evidence-based primary-output result"
-    Radixor achieved the highest balanced accuracy among the evaluated deterministic stemmers for every documented language in both `ALL_WORDS` and `LOWERCASE_GROUPS_ONLY`: **first place in all 40 evaluated language-mode matrices, with no exact first-place ties**. 38 matrices include at least one direct comparator; the two Hebrew modes report Radixor independently because no same-language adapter is configured. This statement is limited to the evaluated implementations, versions, dictionaries, adapters, and balanced-accuracy metric; it is not a universal claim about every stemming use case.
+    Radixor achieved the highest balanced accuracy among the evaluated deterministic stemmers for every documented language in both `ALL_WORDS` and `LOWERCASE_GROUPS_ONLY`: **first place in all 282 evaluated language-mode matrices, with no exact first-place ties**. 60 matrices include at least one direct comparator; the two Hebrew modes report Radixor independently because no same-language adapter is configured. This statement is limited to the evaluated implementations, versions, dictionaries, adapters, and balanced-accuracy metric; it is not a universal claim about every stemming use case.
 
 ### Per-language winner matrix
 
 | Language | Dictionary mode | Winner | Balanced accuracy | Runner-up | Difference | Exact tie | Deterministic stemmers |
 |---|---|---|---:|---|---:|---|---:|
+|Adyghe (`ADY`)|ALL_WORDS|Radixor|0.999051|n/a|n/a|no|1|
+|Adyghe (`ADY`)|LOWERCASE_GROUPS_ONLY|Radixor|0.999051|n/a|n/a|no|1|
+|Afrikaans (`AF_ZA`)|ALL_WORDS|Radixor|0.982452|n/a|n/a|no|1|
+|Afrikaans (`AF_ZA`)|LOWERCASE_GROUPS_ONLY|Radixor|0.982452|n/a|n/a|no|1|
+|Aimele (`AIL`)|ALL_WORDS|Radixor|0.965734|n/a|n/a|no|1|
+|Aimele (`AIL`)|LOWERCASE_GROUPS_ONLY|Radixor|0.965734|n/a|n/a|no|1|
+|Akan (`AK`)|ALL_WORDS|Radixor|0.997522|n/a|n/a|no|1|
+|Akan (`AK`)|LOWERCASE_GROUPS_ONLY|Radixor|0.997522|n/a|n/a|no|1|
+|Albanian (`SQ_AL`)|ALL_WORDS|Radixor|0.996960|n/a|n/a|no|1|
+|Albanian (`SQ_AL`)|LOWERCASE_GROUPS_ONLY|Radixor|0.996960|n/a|n/a|no|1|
+|Alsatian (`GSW`)|ALL_WORDS|Radixor|0.995913|n/a|n/a|no|1|
+|Alsatian (`GSW`)|LOWERCASE_GROUPS_ONLY|Radixor|0.995913|n/a|n/a|no|1|
+|Amharic (`AM_ET`)|ALL_WORDS|Radixor|0.980052|n/a|n/a|no|1|
+|Amharic (`AM_ET`)|LOWERCASE_GROUPS_ONLY|Radixor|0.980052|n/a|n/a|no|1|
+|Amol Azerbaijani (`AZG`)|ALL_WORDS|Radixor|0.994697|n/a|n/a|no|1|
+|Amol Azerbaijani (`AZG`)|LOWERCASE_GROUPS_ONLY|Radixor|0.994697|n/a|n/a|no|1|
+|Ancient Greek (`GRC`)|ALL_WORDS|Radixor|0.995540|n/a|n/a|no|1|
+|Ancient Greek (`GRC`)|LOWERCASE_GROUPS_ONLY|Radixor|0.995540|n/a|n/a|no|1|
+|Anglo-Norman (`XNO`)|ALL_WORDS|Radixor|1.000000|n/a|n/a|no|1|
+|Anglo-Norman (`XNO`)|LOWERCASE_GROUPS_ONLY|Radixor|1.000000|n/a|n/a|no|1|
+|Arabic (`AR`)|ALL_WORDS|Radixor|0.987101|SNOWBALL ARABIC DIRECT|0.315268705|no|3|
+|Arabic (`AR`)|LOWERCASE_GROUPS_ONLY|Radixor|0.987101|SNOWBALL ARABIC DIRECT|0.315268705|no|3|
+|Armenian (`HY_AM`)|ALL_WORDS|Radixor|0.993984|SNOWBALL ARMENIAN DIRECT|0.427825813|no|3|
+|Armenian (`HY_AM`)|LOWERCASE_GROUPS_ONLY|Radixor|0.993984|SNOWBALL ARMENIAN DIRECT|0.427825813|no|3|
+|Ashaninka (`CNI`)|ALL_WORDS|Radixor|0.998172|n/a|n/a|no|1|
+|Ashaninka (`CNI`)|LOWERCASE_GROUPS_ONLY|Radixor|0.998172|n/a|n/a|no|1|
+|Assamese (`AS_IN`)|ALL_WORDS|Radixor|0.991853|n/a|n/a|no|1|
+|Assamese (`AS_IN`)|LOWERCASE_GROUPS_ONLY|Radixor|0.991853|n/a|n/a|no|1|
+|Asturian (`AST`)|ALL_WORDS|Radixor|0.994745|n/a|n/a|no|1|
+|Asturian (`AST`)|LOWERCASE_GROUPS_ONLY|Radixor|0.994745|n/a|n/a|no|1|
+|Aymara (`AYM`)|ALL_WORDS|Radixor|0.996564|n/a|n/a|no|1|
+|Aymara (`AYM`)|LOWERCASE_GROUPS_ONLY|Radixor|0.996564|n/a|n/a|no|1|
+|Azerbaijani (`AZ_AZ`)|ALL_WORDS|Radixor|0.999948|n/a|n/a|no|1|
+|Azerbaijani (`AZ_AZ`)|LOWERCASE_GROUPS_ONLY|Radixor|0.999948|n/a|n/a|no|1|
+|Bashkir (`BAK`)|ALL_WORDS|Radixor|0.998782|n/a|n/a|no|1|
+|Bashkir (`BAK`)|LOWERCASE_GROUPS_ONLY|Radixor|0.998782|n/a|n/a|no|1|
+|Belarusian (`BE_BY`)|ALL_WORDS|Radixor|0.996350|n/a|n/a|no|1|
+|Belarusian (`BE_BY`)|LOWERCASE_GROUPS_ONLY|Radixor|0.996350|n/a|n/a|no|1|
+|Bengali (`BN_BD`)|ALL_WORDS|Radixor|0.997025|n/a|n/a|no|1|
+|Bengali (`BN_BD`)|LOWERCASE_GROUPS_ONLY|Radixor|0.997025|n/a|n/a|no|1|
+|Bininj Kun-wok (`GUP`)|ALL_WORDS|Radixor|0.997625|n/a|n/a|no|1|
+|Bininj Kun-wok (`GUP`)|LOWERCASE_GROUPS_ONLY|Radixor|0.997625|n/a|n/a|no|1|
+|Braj (`BRA`)|ALL_WORDS|Radixor|0.989171|n/a|n/a|no|1|
+|Braj (`BRA`)|LOWERCASE_GROUPS_ONLY|Radixor|0.989171|n/a|n/a|no|1|
+|Breton (`BRE`)|ALL_WORDS|Radixor|1.000000|n/a|n/a|no|1|
+|Breton (`BRE`)|LOWERCASE_GROUPS_ONLY|Radixor|1.000000|n/a|n/a|no|1|
+|Bulgarian (`BG_BG`)|ALL_WORDS|Radixor|0.998003|n/a|n/a|no|1|
+|Bulgarian (`BG_BG`)|LOWERCASE_GROUPS_ONLY|Radixor|0.998003|n/a|n/a|no|1|
+|Catalan (`CA_ES`)|ALL_WORDS|Radixor|0.987704|SNOWBALL CATALAN DIRECT|0.076784766|no|3|
+|Catalan (`CA_ES`)|LOWERCASE_GROUPS_ONLY|Radixor|0.987704|SNOWBALL CATALAN DIRECT|0.076784766|no|3|
+|Cebuano (`CEB`)|ALL_WORDS|Radixor|1.000000|n/a|n/a|no|1|
+|Cebuano (`CEB`)|LOWERCASE_GROUPS_ONLY|Radixor|1.000000|n/a|n/a|no|1|
+|Chichewa (`NY_MW`)|ALL_WORDS|Radixor|1.000000|n/a|n/a|no|1|
+|Chichewa (`NY_MW`)|LOWERCASE_GROUPS_ONLY|Radixor|1.000000|n/a|n/a|no|1|
+|Chukchi (`CKT`)|ALL_WORDS|Radixor|0.990506|n/a|n/a|no|1|
+|Chukchi (`CKT`)|LOWERCASE_GROUPS_ONLY|Radixor|0.990506|n/a|n/a|no|1|
+|Church Slavonic (`CHU`)|ALL_WORDS|Radixor|0.918731|n/a|n/a|no|1|
+|Church Slavonic (`CHU`)|LOWERCASE_GROUPS_ONLY|Radixor|0.918731|n/a|n/a|no|1|
+|Classical Armenian (`XCL`)|ALL_WORDS|Radixor|0.949840|n/a|n/a|no|1|
+|Classical Armenian (`XCL`)|LOWERCASE_GROUPS_ONLY|Radixor|0.949840|n/a|n/a|no|1|
+|Classical Syriac (`SYC`)|ALL_WORDS|Radixor|0.993956|n/a|n/a|no|1|
+|Classical Syriac (`SYC`)|LOWERCASE_GROUPS_ONLY|Radixor|0.993956|n/a|n/a|no|1|
+|Congo Swahili (`SWC`)|ALL_WORDS|Radixor|1.000000|n/a|n/a|no|1|
+|Congo Swahili (`SWC`)|LOWERCASE_GROUPS_ONLY|Radixor|1.000000|n/a|n/a|no|1|
+|Copala Triqui (`CPA`)|ALL_WORDS|Radixor|0.895664|n/a|n/a|no|1|
+|Copala Triqui (`CPA`)|LOWERCASE_GROUPS_ONLY|Radixor|0.895664|n/a|n/a|no|1|
+|Cornish (`COR`)|ALL_WORDS|Radixor|0.997378|n/a|n/a|no|1|
+|Cornish (`COR`)|LOWERCASE_GROUPS_ONLY|Radixor|0.997378|n/a|n/a|no|1|
+|Cree (`CRE`)|ALL_WORDS|Radixor|1.000000|n/a|n/a|no|1|
+|Cree (`CRE`)|LOWERCASE_GROUPS_ONLY|Radixor|1.000000|n/a|n/a|no|1|
+|Crimean Tatar (`CRH`)|ALL_WORDS|Radixor|0.999524|n/a|n/a|no|1|
+|Crimean Tatar (`CRH`)|LOWERCASE_GROUPS_ONLY|Radixor|0.999524|n/a|n/a|no|1|
 |Czech (`CS_CZ`)|ALL_WORDS|Radixor|0.996617|HUNSPELL CZECH LUCENE FILTER|0.142485045|no|4|
 |Czech (`CS_CZ`)|LOWERCASE_GROUPS_ONLY|Radixor|0.997195|HUNSPELL CZECH LUCENE FILTER|0.144045088|no|4|
+|Dakota (`DAK`)|ALL_WORDS|Radixor|0.994385|n/a|n/a|no|1|
+|Dakota (`DAK`)|LOWERCASE_GROUPS_ONLY|Radixor|0.994385|n/a|n/a|no|1|
 |Danish (`DA_DK`)|ALL_WORDS|Radixor|0.996243|SNOWBALL DANISH DIRECT|0.053760569|no|3|
 |Danish (`DA_DK`)|LOWERCASE_GROUPS_ONLY|Radixor|0.996482|SNOWBALL DANISH DIRECT|0.054099342|no|3|
 |Dutch (`NL_NL`)|ALL_WORDS|Radixor|0.988733|SNOWBALL DUTCH DIRECT|0.261639748|no|4|
 |Dutch (`NL_NL`)|LOWERCASE_GROUPS_ONLY|Radixor|0.989114|SNOWBALL DUTCH DIRECT|0.258605347|no|4|
+|Eastern Chatino (`CLY`)|ALL_WORDS|Radixor|0.924863|n/a|n/a|no|1|
+|Eastern Chatino (`CLY`)|LOWERCASE_GROUPS_ONLY|Radixor|0.924863|n/a|n/a|no|1|
+|Egyptian Arabic (`ARZ`)|ALL_WORDS|Radixor|0.978579|n/a|n/a|no|1|
+|Egyptian Arabic (`ARZ`)|LOWERCASE_GROUPS_ONLY|Radixor|0.978579|n/a|n/a|no|1|
 |English (`US_UK`)|ALL_WORDS|Radixor|0.976120|ENGLISH LUCENE PORTER COPIED|0.010985401|no|11|
 |English (`US_UK`)|LOWERCASE_GROUPS_ONLY|Radixor|0.976863|ENGLISH LUCENE PORTER COPIED|0.011393230|no|11|
+|Estonian (`ET_EE`)|ALL_WORDS|Radixor|0.997030|SNOWBALL ESTONIAN DIRECT|0.255774903|no|3|
+|Estonian (`ET_EE`)|LOWERCASE_GROUPS_ONLY|Radixor|0.997030|SNOWBALL ESTONIAN DIRECT|0.255774903|no|3|
+|Evenki (`EVN`)|ALL_WORDS|Radixor|0.996169|n/a|n/a|no|1|
+|Evenki (`EVN`)|LOWERCASE_GROUPS_ONLY|Radixor|0.996169|n/a|n/a|no|1|
+|Faroese (`FO_FO`)|ALL_WORDS|Radixor|0.979257|n/a|n/a|no|1|
+|Faroese (`FO_FO`)|LOWERCASE_GROUPS_ONLY|Radixor|0.979257|n/a|n/a|no|1|
 |Finnish (`FI_FI`)|ALL_WORDS|Radixor|0.984838|SNOWBALL FINNISH LUCENE FILTER|0.244558928|no|4|
 |Finnish (`FI_FI`)|LOWERCASE_GROUPS_ONLY|Radixor|0.988242|SNOWBALL FINNISH DIRECT|0.249699076|no|4|
 |French (`FR_FR`)|ALL_WORDS|Radixor|0.958627|SNOWBALL FRENCH DIRECT|0.109964908|no|6|
 |French (`FR_FR`)|LOWERCASE_GROUPS_ONLY|Radixor|0.958856|SNOWBALL FRENCH DIRECT|0.110030565|no|6|
+|Friulian (`FUR`)|ALL_WORDS|Radixor|0.997938|n/a|n/a|no|1|
+|Friulian (`FUR`)|LOWERCASE_GROUPS_ONLY|Radixor|0.997938|n/a|n/a|no|1|
+|Ga (`GAA`)|ALL_WORDS|Radixor|0.997872|n/a|n/a|no|1|
+|Ga (`GAA`)|LOWERCASE_GROUPS_ONLY|Radixor|0.997872|n/a|n/a|no|1|
+|Galolen (`GAL`)|ALL_WORDS|Radixor|0.983508|n/a|n/a|no|1|
+|Galolen (`GAL`)|LOWERCASE_GROUPS_ONLY|Radixor|0.983508|n/a|n/a|no|1|
 |German (`DE_DE`)|ALL_WORDS|Radixor|0.910445|GERMAN CISTEM|0.031918024|no|8|
 |German (`DE_DE`)|LOWERCASE_GROUPS_ONLY|Radixor|0.966959|GERMAN CISTEM|0.052231588|no|8|
+|Gothic (`GOT`)|ALL_WORDS|Radixor|0.981471|n/a|n/a|no|1|
+|Gothic (`GOT`)|LOWERCASE_GROUPS_ONLY|Radixor|0.981471|n/a|n/a|no|1|
+|Greek (`EL_GR`)|ALL_WORDS|Radixor|0.978296|SNOWBALL GREEK DIRECT|0.106884319|no|3|
+|Greek (`EL_GR`)|LOWERCASE_GROUPS_ONLY|Radixor|0.978296|SNOWBALL GREEK DIRECT|0.106884319|no|3|
+|Gulf Arabic (`AFB`)|ALL_WORDS|Radixor|0.986901|n/a|n/a|no|1|
+|Gulf Arabic (`AFB`)|LOWERCASE_GROUPS_ONLY|Radixor|0.986901|n/a|n/a|no|1|
+|Haida (`HAI`)|ALL_WORDS|Radixor|1.000000|n/a|n/a|no|1|
+|Haida (`HAI`)|LOWERCASE_GROUPS_ONLY|Radixor|1.000000|n/a|n/a|no|1|
 |Hebrew (`HE_IL`)|ALL_WORDS|Radixor|0.986253|n/a|n/a|no|1|
 |Hebrew (`HE_IL`)|LOWERCASE_GROUPS_ONLY|Radixor|0.986253|n/a|n/a|no|1|
+|Hiligaynon (`HIL`)|ALL_WORDS|Radixor|0.985981|n/a|n/a|no|1|
+|Hiligaynon (`HIL`)|LOWERCASE_GROUPS_ONLY|Radixor|0.985981|n/a|n/a|no|1|
+|Hsilimo (`HSI`)|ALL_WORDS|Radixor|0.996914|n/a|n/a|no|1|
+|Hsilimo (`HSI`)|LOWERCASE_GROUPS_ONLY|Radixor|0.996914|n/a|n/a|no|1|
 |Hungarian (`HU_HU`)|ALL_WORDS|Radixor|0.995555|SNOWBALL HUNGARIAN LUCENE FILTER|0.172591951|no|4|
 |Hungarian (`HU_HU`)|LOWERCASE_GROUPS_ONLY|Radixor|0.996227|SNOWBALL HUNGARIAN DIRECT|0.174150583|no|4|
+|Icelandic (`IS_IS`)|ALL_WORDS|Radixor|0.981831|n/a|n/a|no|1|
+|Icelandic (`IS_IS`)|LOWERCASE_GROUPS_ONLY|Radixor|0.981831|n/a|n/a|no|1|
+|Indonesian (`ID_ID`)|ALL_WORDS|Radixor|0.999914|SNOWBALL INDONESIAN DIRECT|0.128496605|no|3|
+|Indonesian (`ID_ID`)|LOWERCASE_GROUPS_ONLY|Radixor|0.999914|SNOWBALL INDONESIAN DIRECT|0.128496605|no|3|
+|Ingrian (`IZH`)|ALL_WORDS|Radixor|1.000000|n/a|n/a|no|1|
+|Ingrian (`IZH`)|LOWERCASE_GROUPS_ONLY|Radixor|1.000000|n/a|n/a|no|1|
+|Irish (`GA_IE`)|ALL_WORDS|Radixor|0.975677|SNOWBALL IRISH DIRECT|0.462889616|no|3|
+|Irish (`GA_IE`)|LOWERCASE_GROUPS_ONLY|Radixor|0.975677|SNOWBALL IRISH DIRECT|0.462889616|no|3|
 |Italian (`IT_IT`)|ALL_WORDS|Radixor|0.996651|SNOWBALL ITALIAN DIRECT|0.130360651|no|4|
 |Italian (`IT_IT`)|LOWERCASE_GROUPS_ONLY|Radixor|0.996656|SNOWBALL ITALIAN DIRECT|0.130349693|no|4|
-|Norwegian Bokmal (`NB_NO`)|ALL_WORDS|Radixor|0.976021|SNOWBALL NORWEGIAN BOKMAL DIRECT|0.101762107|no|5|
-|Norwegian Bokmal (`NB_NO`)|LOWERCASE_GROUPS_ONLY|Radixor|0.976240|SNOWBALL NORWEGIAN BOKMAL DIRECT|0.101954266|no|5|
+|Itelmen (`ITL`)|ALL_WORDS|Radixor|0.999389|n/a|n/a|no|1|
+|Itelmen (`ITL`)|LOWERCASE_GROUPS_ONLY|Radixor|0.999389|n/a|n/a|no|1|
+|Japanese (`JA_JP`)|ALL_WORDS|Radixor|0.919503|n/a|n/a|no|1|
+|Japanese (`JA_JP`)|LOWERCASE_GROUPS_ONLY|Radixor|0.919503|n/a|n/a|no|1|
+|Kabardian (`KBD`)|ALL_WORDS|Radixor|1.000000|n/a|n/a|no|1|
+|Kabardian (`KBD`)|LOWERCASE_GROUPS_ONLY|Radixor|1.000000|n/a|n/a|no|1|
+|Kalaallisut (`KL_GL`)|ALL_WORDS|Radixor|1.000000|n/a|n/a|no|1|
+|Kalaallisut (`KL_GL`)|LOWERCASE_GROUPS_ONLY|Radixor|1.000000|n/a|n/a|no|1|
+|Kannada (`KN_IN`)|ALL_WORDS|Radixor|0.898695|n/a|n/a|no|1|
+|Kannada (`KN_IN`)|LOWERCASE_GROUPS_ONLY|Radixor|0.898695|n/a|n/a|no|1|
+|Karelian (`KRL`)|ALL_WORDS|Radixor|1.000000|n/a|n/a|no|1|
+|Karelian (`KRL`)|LOWERCASE_GROUPS_ONLY|Radixor|1.000000|n/a|n/a|no|1|
+|Kashubian (`CSB`)|ALL_WORDS|Radixor|0.995801|n/a|n/a|no|1|
+|Kashubian (`CSB`)|LOWERCASE_GROUPS_ONLY|Radixor|0.995801|n/a|n/a|no|1|
+|Kazakh (`KK_KZ`)|ALL_WORDS|Radixor|0.998095|n/a|n/a|no|1|
+|Kazakh (`KK_KZ`)|LOWERCASE_GROUPS_ONLY|Radixor|0.998095|n/a|n/a|no|1|
+|Khakas (`KJH`)|ALL_WORDS|Radixor|1.000000|n/a|n/a|no|1|
+|Khakas (`KJH`)|LOWERCASE_GROUPS_ONLY|Radixor|1.000000|n/a|n/a|no|1|
+|Khinalugh (`KLR`)|ALL_WORDS|Radixor|0.944247|n/a|n/a|no|1|
+|Khinalugh (`KLR`)|LOWERCASE_GROUPS_ONLY|Radixor|0.944247|n/a|n/a|no|1|
+|Kodi (`KOD`)|ALL_WORDS|Radixor|1.000000|n/a|n/a|no|1|
+|Kodi (`KOD`)|LOWERCASE_GROUPS_ONLY|Radixor|1.000000|n/a|n/a|no|1|
+|Kongo (`KON`)|ALL_WORDS|Radixor|0.998208|n/a|n/a|no|1|
+|Kongo (`KON`)|LOWERCASE_GROUPS_ONLY|Radixor|0.998208|n/a|n/a|no|1|
+|Kyrgyz (`KY_KG`)|ALL_WORDS|Radixor|1.000000|n/a|n/a|no|1|
+|Kyrgyz (`KY_KG`)|LOWERCASE_GROUPS_ONLY|Radixor|1.000000|n/a|n/a|no|1|
+|Ladin (`LLD`)|ALL_WORDS|Radixor|0.997844|n/a|n/a|no|1|
+|Ladin (`LLD`)|LOWERCASE_GROUPS_ONLY|Radixor|0.997844|n/a|n/a|no|1|
+|Latin (`LA`)|ALL_WORDS|Radixor|0.963898|n/a|n/a|no|1|
+|Latin (`LA`)|LOWERCASE_GROUPS_ONLY|Radixor|0.963898|n/a|n/a|no|1|
+|Latvian (`LV_LV`)|ALL_WORDS|Radixor|0.941147|n/a|n/a|no|1|
+|Latvian (`LV_LV`)|LOWERCASE_GROUPS_ONLY|Radixor|0.941147|n/a|n/a|no|1|
+|Lingala (`LIN`)|ALL_WORDS|Radixor|1.000000|n/a|n/a|no|1|
+|Lingala (`LIN`)|LOWERCASE_GROUPS_ONLY|Radixor|1.000000|n/a|n/a|no|1|
+|Lithuanian (`LT_LT`)|ALL_WORDS|Radixor|0.998532|SNOWBALL LITHUANIAN DIRECT|0.212439399|no|3|
+|Lithuanian (`LT_LT`)|LOWERCASE_GROUPS_ONLY|Radixor|0.998532|SNOWBALL LITHUANIAN DIRECT|0.212439399|no|3|
+|Livonian (`LIV`)|ALL_WORDS|Radixor|0.997482|n/a|n/a|no|1|
+|Livonian (`LIV`)|LOWERCASE_GROUPS_ONLY|Radixor|0.997482|n/a|n/a|no|1|
+|Low German (`NDS`)|ALL_WORDS|Radixor|0.980207|n/a|n/a|no|1|
+|Low German (`NDS`)|LOWERCASE_GROUPS_ONLY|Radixor|0.980207|n/a|n/a|no|1|
+|Lower Sorbian (`DSB`)|ALL_WORDS|Radixor|0.992702|n/a|n/a|no|1|
+|Lower Sorbian (`DSB`)|LOWERCASE_GROUPS_ONLY|Radixor|0.992702|n/a|n/a|no|1|
+|Luganda (`LG_UG`)|ALL_WORDS|Radixor|0.994127|n/a|n/a|no|1|
+|Luganda (`LG_UG`)|LOWERCASE_GROUPS_ONLY|Radixor|0.994127|n/a|n/a|no|1|
+|Macedonian (`MK_MK`)|ALL_WORDS|Radixor|0.996059|n/a|n/a|no|1|
+|Macedonian (`MK_MK`)|LOWERCASE_GROUPS_ONLY|Radixor|0.996059|n/a|n/a|no|1|
+|Magahi (`MAG`)|ALL_WORDS|Radixor|0.986749|n/a|n/a|no|1|
+|Magahi (`MAG`)|LOWERCASE_GROUPS_ONLY|Radixor|0.986749|n/a|n/a|no|1|
+|Malagasy (`MG_MG`)|ALL_WORDS|Radixor|1.000000|n/a|n/a|no|1|
+|Malagasy (`MG_MG`)|LOWERCASE_GROUPS_ONLY|Radixor|1.000000|n/a|n/a|no|1|
+|Maltese (`MT_MT`)|ALL_WORDS|Radixor|0.996750|n/a|n/a|no|1|
+|Maltese (`MT_MT`)|LOWERCASE_GROUPS_ONLY|Radixor|0.996750|n/a|n/a|no|1|
+|Manx (`GV_IM`)|ALL_WORDS|n/a|n/a|n/a|n/a|n/a|0|
+|Manx (`GV_IM`)|LOWERCASE_GROUPS_ONLY|n/a|n/a|n/a|n/a|n/a|0|
+|Maori (`MI_NZ`)|ALL_WORDS|Radixor|0.995192|n/a|n/a|no|1|
+|Maori (`MI_NZ`)|LOWERCASE_GROUPS_ONLY|Radixor|0.995192|n/a|n/a|no|1|
+|Mapudungun (`ARN`)|ALL_WORDS|Radixor|1.000000|n/a|n/a|no|1|
+|Mapudungun (`ARN`)|LOWERCASE_GROUPS_ONLY|Radixor|1.000000|n/a|n/a|no|1|
+|Middle French (`FRM`)|ALL_WORDS|Radixor|0.996418|n/a|n/a|no|1|
+|Middle French (`FRM`)|LOWERCASE_GROUPS_ONLY|Radixor|0.996418|n/a|n/a|no|1|
+|Middle High German (`GMH`)|ALL_WORDS|Radixor|1.000000|n/a|n/a|no|1|
+|Middle High German (`GMH`)|LOWERCASE_GROUPS_ONLY|Radixor|1.000000|n/a|n/a|no|1|
+|Middle Low German (`GML`)|ALL_WORDS|Radixor|0.978543|n/a|n/a|no|1|
+|Middle Low German (`GML`)|LOWERCASE_GROUPS_ONLY|Radixor|0.978543|n/a|n/a|no|1|
+|Mongolian (`MN_MN`)|ALL_WORDS|Radixor|0.979541|n/a|n/a|no|1|
+|Mongolian (`MN_MN`)|LOWERCASE_GROUPS_ONLY|Radixor|0.979541|n/a|n/a|no|1|
+|Murrinh-Patha (`MWF`)|ALL_WORDS|Radixor|0.861762|n/a|n/a|no|1|
+|Murrinh-Patha (`MWF`)|LOWERCASE_GROUPS_ONLY|Radixor|0.861762|n/a|n/a|no|1|
+|Navajo (`NAV`)|ALL_WORDS|Radixor|0.976188|n/a|n/a|no|1|
+|Navajo (`NAV`)|LOWERCASE_GROUPS_ONLY|Radixor|0.976188|n/a|n/a|no|1|
+|Neapolitan (`NAP`)|ALL_WORDS|Radixor|1.000000|n/a|n/a|no|1|
+|Neapolitan (`NAP`)|LOWERCASE_GROUPS_ONLY|Radixor|1.000000|n/a|n/a|no|1|
+|North Frisian (`FRR`)|ALL_WORDS|Radixor|0.954614|n/a|n/a|no|1|
+|North Frisian (`FRR`)|LOWERCASE_GROUPS_ONLY|Radixor|0.954614|n/a|n/a|no|1|
+|Northern Sami (`SME`)|ALL_WORDS|Radixor|0.991171|n/a|n/a|no|1|
+|Northern Sami (`SME`)|LOWERCASE_GROUPS_ONLY|Radixor|0.991171|n/a|n/a|no|1|
+|Norwegian Bokmål (`NB_NO`)|ALL_WORDS|Radixor|0.976021|SNOWBALL NORWEGIAN BOKMAL DIRECT|0.101762107|no|5|
+|Norwegian Bokmål (`NB_NO`)|LOWERCASE_GROUPS_ONLY|Radixor|0.976240|SNOWBALL NORWEGIAN BOKMAL DIRECT|0.101954266|no|5|
 |Norwegian Nynorsk (`NN_NO`)|ALL_WORDS|Radixor|0.950991|SNOWBALL NORWEGIAN NYNORSK DIRECT|0.082896791|no|3|
 |Norwegian Nynorsk (`NN_NO`)|LOWERCASE_GROUPS_ONLY|Radixor|0.951104|SNOWBALL NORWEGIAN NYNORSK DIRECT|0.082851757|no|3|
+|Old English (`ANG`)|ALL_WORDS|Radixor|0.970779|n/a|n/a|no|1|
+|Old English (`ANG`)|LOWERCASE_GROUPS_ONLY|Radixor|0.970779|n/a|n/a|no|1|
+|Old French (`FRO`)|ALL_WORDS|Radixor|0.965650|n/a|n/a|no|1|
+|Old French (`FRO`)|LOWERCASE_GROUPS_ONLY|Radixor|0.965650|n/a|n/a|no|1|
+|Old High German (`GOH`)|ALL_WORDS|Radixor|0.992668|n/a|n/a|no|1|
+|Old High German (`GOH`)|LOWERCASE_GROUPS_ONLY|Radixor|0.992668|n/a|n/a|no|1|
+|Old Irish (`SGA`)|ALL_WORDS|Radixor|0.997869|n/a|n/a|no|1|
+|Old Irish (`SGA`)|LOWERCASE_GROUPS_ONLY|Radixor|0.997869|n/a|n/a|no|1|
+|Old Norse (`NON`)|ALL_WORDS|Radixor|0.940817|n/a|n/a|no|1|
+|Old Norse (`NON`)|LOWERCASE_GROUPS_ONLY|Radixor|0.940817|n/a|n/a|no|1|
+|Old Saxon (`OSX`)|ALL_WORDS|Radixor|0.987927|n/a|n/a|no|1|
+|Old Saxon (`OSX`)|LOWERCASE_GROUPS_ONLY|Radixor|0.987927|n/a|n/a|no|1|
+|Oodham (`OOD`)|ALL_WORDS|Radixor|0.988769|n/a|n/a|no|1|
+|Oodham (`OOD`)|LOWERCASE_GROUPS_ONLY|Radixor|0.988769|n/a|n/a|no|1|
+|Pashto (`PS_AF`)|ALL_WORDS|Radixor|0.955597|n/a|n/a|no|1|
+|Pashto (`PS_AF`)|LOWERCASE_GROUPS_ONLY|Radixor|0.955597|n/a|n/a|no|1|
 |Persian (`FA_IR`)|ALL_WORDS|Radixor|0.975610|SNOWBALL PERSIAN DIRECT|0.440486794|no|3|
 |Persian (`FA_IR`)|LOWERCASE_GROUPS_ONLY|Radixor|0.975610|SNOWBALL PERSIAN DIRECT|0.440486794|no|3|
 |Polish (`PL_PL`)|ALL_WORDS|Radixor|0.991105|POLISH LUCENE MORFOLOGIK FILTER|0.042712804|no|6|
 |Polish (`PL_PL`)|LOWERCASE_GROUPS_ONLY|Radixor|0.991301|POLISH LUCENE MORFOLOGIK FILTER|0.042883749|no|6|
 |Portuguese (`PT_PT`)|ALL_WORDS|Radixor|0.998542|SNOWBALL PORTUGUESE DIRECT|0.059619854|no|6|
 |Portuguese (`PT_PT`)|LOWERCASE_GROUPS_ONLY|Radixor|0.998542|SNOWBALL PORTUGUESE DIRECT|0.059619854|no|6|
+|Quechua (`QUE`)|ALL_WORDS|Radixor|0.986835|n/a|n/a|no|1|
+|Quechua (`QUE`)|LOWERCASE_GROUPS_ONLY|Radixor|0.986835|n/a|n/a|no|1|
+|Romanian (`RO_RO`)|ALL_WORDS|Radixor|0.994415|SNOWBALL ROMANIAN DIRECT|0.145283621|no|3|
+|Romanian (`RO_RO`)|LOWERCASE_GROUPS_ONLY|Radixor|0.994415|SNOWBALL ROMANIAN DIRECT|0.145283621|no|3|
 |Russian (`RU_RU`)|ALL_WORDS|Radixor|0.990188|SNOWBALL RUSSIAN LUCENE FILTER|0.155623602|no|4|
 |Russian (`RU_RU`)|LOWERCASE_GROUPS_ONLY|Radixor|0.990213|SNOWBALL RUSSIAN DIRECT|0.155670422|no|4|
+|Seneca (`SEE`)|ALL_WORDS|Radixor|0.998027|n/a|n/a|no|1|
+|Seneca (`SEE`)|LOWERCASE_GROUPS_ONLY|Radixor|0.998027|n/a|n/a|no|1|
+|Serbo-Croatian (`HBS`)|ALL_WORDS|Radixor|0.987756|n/a|n/a|no|1|
+|Serbo-Croatian (`HBS`)|LOWERCASE_GROUPS_ONLY|Radixor|0.987756|n/a|n/a|no|1|
+|Shipibo-Conibo (`SHP`)|ALL_WORDS|Radixor|0.978699|n/a|n/a|no|1|
+|Shipibo-Conibo (`SHP`)|LOWERCASE_GROUPS_ONLY|Radixor|0.978699|n/a|n/a|no|1|
+|Shona (`SN_ZW`)|ALL_WORDS|Radixor|1.000000|n/a|n/a|no|1|
+|Shona (`SN_ZW`)|LOWERCASE_GROUPS_ONLY|Radixor|1.000000|n/a|n/a|no|1|
+|Sotho, Southern (`ST_ZA`)|ALL_WORDS|Radixor|1.000000|SNOWBALL SESOTHO DIRECT|0.496582532|no|2|
+|Sotho, Southern (`ST_ZA`)|LOWERCASE_GROUPS_ONLY|Radixor|1.000000|SNOWBALL SESOTHO DIRECT|0.496582532|no|2|
+|Southern Kurdish (`SDH`)|ALL_WORDS|n/a|n/a|n/a|n/a|n/a|0|
+|Southern Kurdish (`SDH`)|LOWERCASE_GROUPS_ONLY|n/a|n/a|n/a|n/a|n/a|0|
 |Spanish (`ES_ES`)|ALL_WORDS|Radixor|0.989448|SNOWBALL SPANISH LUCENE FILTER|0.337009985|no|7|
 |Spanish (`ES_ES`)|LOWERCASE_GROUPS_ONLY|Radixor|0.989580|SNOWBALL SPANISH DIRECT|0.337037572|no|7|
 |Swedish (`SV_SE`)|ALL_WORDS|Radixor|0.977619|SNOWBALL SWEDISH DIRECT|0.169075635|no|5|
 |Swedish (`SV_SE`)|LOWERCASE_GROUPS_ONLY|Radixor|0.977573|SNOWBALL SWEDISH DIRECT|0.168961385|no|5|
+|Tagalog (`TL_PH`)|ALL_WORDS|Radixor|0.997116|n/a|n/a|no|1|
+|Tagalog (`TL_PH`)|LOWERCASE_GROUPS_ONLY|Radixor|0.997116|n/a|n/a|no|1|
+|Temoaya Otomi (`OTE`)|ALL_WORDS|Radixor|0.953277|n/a|n/a|no|1|
+|Temoaya Otomi (`OTE`)|LOWERCASE_GROUPS_ONLY|Radixor|0.953277|n/a|n/a|no|1|
+|Turkish (`TR_TR`)|ALL_WORDS|Radixor|0.991555|SNOWBALL TURKISH DIRECT|0.378971891|no|3|
+|Turkish (`TR_TR`)|LOWERCASE_GROUPS_ONLY|Radixor|0.991555|SNOWBALL TURKISH DIRECT|0.378971891|no|3|
 |Ukrainian (`UK_UA`)|ALL_WORDS|Radixor|0.995816|UKRAINIAN LUCENE MORFOLOGIK FILTER|0.066909929|no|4|
 |Ukrainian (`UK_UA`)|LOWERCASE_GROUPS_ONLY|Radixor|0.995815|UKRAINIAN LUCENE MORFOLOGIK FILTER|0.066926372|no|4|
+|Uyghur (`UG_CN`)|ALL_WORDS|Radixor|0.996210|n/a|n/a|no|1|
+|Uyghur (`UG_CN`)|LOWERCASE_GROUPS_ONLY|Radixor|0.996210|n/a|n/a|no|1|
+|Uzbek (`UZ_UZ`)|ALL_WORDS|Radixor|0.999762|n/a|n/a|no|1|
+|Uzbek (`UZ_UZ`)|LOWERCASE_GROUPS_ONLY|Radixor|0.999762|n/a|n/a|no|1|
+|Voro (`VRO`)|ALL_WORDS|Radixor|0.998340|n/a|n/a|no|1|
+|Voro (`VRO`)|LOWERCASE_GROUPS_ONLY|Radixor|0.998340|n/a|n/a|no|1|
+|Western Highland Chatino (`CTP`)|ALL_WORDS|Radixor|0.971433|n/a|n/a|no|1|
+|Western Highland Chatino (`CTP`)|LOWERCASE_GROUPS_ONLY|Radixor|0.971433|n/a|n/a|no|1|
+|Xibe (`SJO`)|ALL_WORDS|Radixor|0.970067|n/a|n/a|no|1|
+|Xibe (`SJO`)|LOWERCASE_GROUPS_ONLY|Radixor|0.970067|n/a|n/a|no|1|
+|Xipinawa (`XTY`)|ALL_WORDS|Radixor|0.990824|n/a|n/a|no|1|
+|Xipinawa (`XTY`)|LOWERCASE_GROUPS_ONLY|Radixor|0.990824|n/a|n/a|no|1|
+|Yamana (`AME`)|ALL_WORDS|Radixor|0.991071|n/a|n/a|no|1|
+|Yamana (`AME`)|LOWERCASE_GROUPS_ONLY|Radixor|0.991071|n/a|n/a|no|1|
 |Yiddish (`YI`)|ALL_WORDS|Radixor|0.989079|SNOWBALL YIDDISH DIRECT|0.097960961|no|3|
 |Yiddish (`YI`)|LOWERCASE_GROUPS_ONLY|Radixor|0.989079|SNOWBALL YIDDISH DIRECT|0.097960961|no|3|
+|Zacatepec Chatino (`ZPV`)|ALL_WORDS|Radixor|0.992863|n/a|n/a|no|1|
+|Zacatepec Chatino (`ZPV`)|LOWERCASE_GROUPS_ONLY|Radixor|0.992863|n/a|n/a|no|1|
+|Zarma (`DJE`)|ALL_WORDS|Radixor|1.000000|n/a|n/a|no|1|
+|Zarma (`DJE`)|LOWERCASE_GROUPS_ONLY|Radixor|1.000000|n/a|n/a|no|1|
+|Zenzontepec Chatino (`CZN`)|ALL_WORDS|Radixor|0.973968|n/a|n/a|no|1|
+|Zenzontepec Chatino (`CZN`)|LOWERCASE_GROUPS_ONLY|Radixor|0.973968|n/a|n/a|no|1|
+|Zulu (`ZU_ZA`)|ALL_WORDS|Radixor|0.993145|n/a|n/a|no|1|
+|Zulu (`ZU_ZA`)|LOWERCASE_GROUPS_ONLY|Radixor|0.993145|n/a|n/a|no|1|
 
 ### Secondary-metric trade-offs
 
@@ -157,7 +403,7 @@ Counts use `PRIMARY_OUTPUT` only and retain each adapter configuration as a sepa
 
 | Stemmer | Evaluated languages | Wins | Exact first-place ties | Top-three placements | Average rank | Median rank |
 |---|---:|---:|---:|---:|---:|---:|
-|Radixor|20|20|0|20|1.000|1.000|
+|Radixor|141|141|0|141|1.000|1.000|
 |CZECH LUCENE CZECH STEM FILTER|1|0|0|1|3.000|3.000|
 |ENGLISH LUCENE KSTEM FILTER|1|0|0|0|8.000|8.000|
 |ENGLISH LUCENE MINIMAL FILTER|1|0|0|0|9.000|9.000|
@@ -195,21 +441,37 @@ Counts use `PRIMARY_OUTPUT` only and retain each adapter configuration as a sepa
 |PORTUGUESE LUCENE PORTUGUESE MINIMAL STEM FILTER|1|0|0|0|6.000|6.000|
 |PORTUGUESE LUCENE PORTUGUESE STEM FILTER|1|0|0|0|4.000|4.000|
 |RUSSIAN LUCENE RUSSIAN LIGHT STEM FILTER|1|0|0|0|4.000|4.000|
+|SNOWBALL ARABIC DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL ARABIC LUCENE FILTER|1|0|0|1|3.000|3.000|
+|SNOWBALL ARMENIAN DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL ARMENIAN LUCENE FILTER|1|0|0|1|3.000|3.000|
+|SNOWBALL CATALAN DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL CATALAN LUCENE FILTER|1|0|0|1|3.000|3.000|
 |SNOWBALL CZECH DIRECT|1|0|0|0|4.000|4.000|
 |SNOWBALL DANISH DIRECT|1|0|0|1|2.000|2.000|
 |SNOWBALL DANISH LUCENE FILTER|1|0|0|1|3.000|3.000|
 |SNOWBALL DUTCH DIRECT|1|0|0|1|2.000|2.000|
 |SNOWBALL DUTCH LUCENE FILTER|1|0|0|0|4.000|4.000|
+|SNOWBALL ESTONIAN DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL ESTONIAN LUCENE FILTER|1|0|0|1|3.000|3.000|
 |SNOWBALL FINNISH DIRECT|1|0|0|1|3.000|3.000|
 |SNOWBALL FINNISH LUCENE FILTER|1|0|0|1|2.000|2.000|
 |SNOWBALL FRENCH DIRECT|1|0|0|1|2.000|2.000|
 |SNOWBALL FRENCH LUCENE FILTER|1|0|0|1|3.000|3.000|
 |SNOWBALL GERMAN DIRECT|1|0|0|1|3.000|3.000|
 |SNOWBALL GERMAN LUCENE FILTER|1|0|0|0|4.000|4.000|
+|SNOWBALL GREEK DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL GREEK LUCENE FILTER|1|0|0|1|3.000|3.000|
 |SNOWBALL HUNGARIAN DIRECT|1|0|0|1|3.000|3.000|
 |SNOWBALL HUNGARIAN LUCENE FILTER|1|0|0|1|2.000|2.000|
+|SNOWBALL INDONESIAN DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL INDONESIAN LUCENE FILTER|1|0|0|1|3.000|3.000|
+|SNOWBALL IRISH DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL IRISH LUCENE FILTER|1|0|0|1|3.000|3.000|
 |SNOWBALL ITALIAN DIRECT|1|0|0|1|2.000|2.000|
 |SNOWBALL ITALIAN LUCENE FILTER|1|0|0|1|3.000|3.000|
+|SNOWBALL LITHUANIAN DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL LITHUANIAN LUCENE FILTER|1|0|0|1|3.000|3.000|
 |SNOWBALL NORWEGIAN BOKMAL DIRECT|1|0|0|1|2.000|2.000|
 |SNOWBALL NORWEGIAN BOKMAL LUCENE FILTER|1|0|0|1|3.000|3.000|
 |SNOWBALL NORWEGIAN NYNORSK DIRECT|1|0|0|1|2.000|2.000|
@@ -218,12 +480,17 @@ Counts use `PRIMARY_OUTPUT` only and retain each adapter configuration as a sepa
 |SNOWBALL POLISH DIRECT|1|0|0|0|6.000|6.000|
 |SNOWBALL PORTUGUESE DIRECT|1|0|0|1|2.000|2.000|
 |SNOWBALL PORTUGUESE LUCENE FILTER|1|0|0|1|3.000|3.000|
+|SNOWBALL ROMANIAN DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL ROMANIAN LUCENE FILTER|1|0|0|1|3.000|3.000|
 |SNOWBALL RUSSIAN DIRECT|1|0|0|1|3.000|3.000|
 |SNOWBALL RUSSIAN LUCENE FILTER|1|0|0|1|2.000|2.000|
+|SNOWBALL SESOTHO DIRECT|1|0|0|1|2.000|2.000|
 |SNOWBALL SPANISH DIRECT|1|0|0|1|3.000|3.000|
 |SNOWBALL SPANISH LUCENE FILTER|1|0|0|1|2.000|2.000|
 |SNOWBALL SWEDISH DIRECT|1|0|0|1|2.000|2.000|
 |SNOWBALL SWEDISH LUCENE FILTER|1|0|0|1|3.000|3.000|
+|SNOWBALL TURKISH DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL TURKISH LUCENE FILTER|1|0|0|1|3.000|3.000|
 |SNOWBALL YIDDISH DIRECT|1|0|0|1|2.000|2.000|
 |SNOWBALL YIDDISH LUCENE FILTER|1|0|0|1|3.000|3.000|
 |SPANISH LUCENE SPANISH LIGHT STEM FILTER|1|0|0|0|5.000|5.000|
@@ -240,7 +507,7 @@ Counts use `PRIMARY_OUTPUT` only and retain each adapter configuration as a sepa
 
 | Stemmer | Evaluated languages | Wins | Exact first-place ties | Top-three placements | Average rank | Median rank |
 |---|---:|---:|---:|---:|---:|---:|
-|Radixor|20|20|0|20|1.000|1.000|
+|Radixor|141|141|0|141|1.000|1.000|
 |CZECH LUCENE CZECH STEM FILTER|1|0|0|1|3.000|3.000|
 |ENGLISH LUCENE KSTEM FILTER|1|0|0|0|8.000|8.000|
 |ENGLISH LUCENE MINIMAL FILTER|1|0|0|0|9.000|9.000|
@@ -278,21 +545,37 @@ Counts use `PRIMARY_OUTPUT` only and retain each adapter configuration as a sepa
 |PORTUGUESE LUCENE PORTUGUESE MINIMAL STEM FILTER|1|0|0|0|6.000|6.000|
 |PORTUGUESE LUCENE PORTUGUESE STEM FILTER|1|0|0|0|4.000|4.000|
 |RUSSIAN LUCENE RUSSIAN LIGHT STEM FILTER|1|0|0|0|4.000|4.000|
+|SNOWBALL ARABIC DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL ARABIC LUCENE FILTER|1|0|0|1|3.000|3.000|
+|SNOWBALL ARMENIAN DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL ARMENIAN LUCENE FILTER|1|0|0|1|3.000|3.000|
+|SNOWBALL CATALAN DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL CATALAN LUCENE FILTER|1|0|0|1|3.000|3.000|
 |SNOWBALL CZECH DIRECT|1|0|0|0|4.000|4.000|
 |SNOWBALL DANISH DIRECT|1|0|0|1|2.000|2.000|
 |SNOWBALL DANISH LUCENE FILTER|1|0|0|1|3.000|3.000|
 |SNOWBALL DUTCH DIRECT|1|0|0|1|2.000|2.000|
 |SNOWBALL DUTCH LUCENE FILTER|1|0|0|0|4.000|4.000|
+|SNOWBALL ESTONIAN DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL ESTONIAN LUCENE FILTER|1|0|0|1|3.000|3.000|
 |SNOWBALL FINNISH DIRECT|1|0|0|1|2.000|2.000|
 |SNOWBALL FINNISH LUCENE FILTER|1|0|0|1|3.000|3.000|
 |SNOWBALL FRENCH DIRECT|1|0|0|1|2.000|2.000|
 |SNOWBALL FRENCH LUCENE FILTER|1|0|0|1|3.000|3.000|
 |SNOWBALL GERMAN DIRECT|1|0|0|1|3.000|3.000|
 |SNOWBALL GERMAN LUCENE FILTER|1|0|0|0|4.000|4.000|
+|SNOWBALL GREEK DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL GREEK LUCENE FILTER|1|0|0|1|3.000|3.000|
 |SNOWBALL HUNGARIAN DIRECT|1|0|0|1|2.000|2.000|
 |SNOWBALL HUNGARIAN LUCENE FILTER|1|0|0|1|3.000|3.000|
+|SNOWBALL INDONESIAN DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL INDONESIAN LUCENE FILTER|1|0|0|1|3.000|3.000|
+|SNOWBALL IRISH DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL IRISH LUCENE FILTER|1|0|0|1|3.000|3.000|
 |SNOWBALL ITALIAN DIRECT|1|0|0|1|2.000|2.000|
 |SNOWBALL ITALIAN LUCENE FILTER|1|0|0|1|3.000|3.000|
+|SNOWBALL LITHUANIAN DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL LITHUANIAN LUCENE FILTER|1|0|0|1|3.000|3.000|
 |SNOWBALL NORWEGIAN BOKMAL DIRECT|1|0|0|1|2.000|2.000|
 |SNOWBALL NORWEGIAN BOKMAL LUCENE FILTER|1|0|0|1|3.000|3.000|
 |SNOWBALL NORWEGIAN NYNORSK DIRECT|1|0|0|1|2.000|2.000|
@@ -301,12 +584,17 @@ Counts use `PRIMARY_OUTPUT` only and retain each adapter configuration as a sepa
 |SNOWBALL POLISH DIRECT|1|0|0|0|6.000|6.000|
 |SNOWBALL PORTUGUESE DIRECT|1|0|0|1|2.000|2.000|
 |SNOWBALL PORTUGUESE LUCENE FILTER|1|0|0|1|3.000|3.000|
+|SNOWBALL ROMANIAN DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL ROMANIAN LUCENE FILTER|1|0|0|1|3.000|3.000|
 |SNOWBALL RUSSIAN DIRECT|1|0|0|1|2.000|2.000|
 |SNOWBALL RUSSIAN LUCENE FILTER|1|0|0|1|3.000|3.000|
+|SNOWBALL SESOTHO DIRECT|1|0|0|1|2.000|2.000|
 |SNOWBALL SPANISH DIRECT|1|0|0|1|2.000|2.000|
 |SNOWBALL SPANISH LUCENE FILTER|1|0|0|1|3.000|3.000|
 |SNOWBALL SWEDISH DIRECT|1|0|0|1|2.000|2.000|
 |SNOWBALL SWEDISH LUCENE FILTER|1|0|0|1|3.000|3.000|
+|SNOWBALL TURKISH DIRECT|1|0|0|1|2.000|2.000|
+|SNOWBALL TURKISH LUCENE FILTER|1|0|0|1|3.000|3.000|
 |SNOWBALL YIDDISH DIRECT|1|0|0|1|2.000|2.000|
 |SNOWBALL YIDDISH LUCENE FILTER|1|0|0|1|3.000|3.000|
 |SPANISH LUCENE SPANISH LIGHT STEM FILTER|1|0|0|0|5.000|5.000|
@@ -322,17 +610,17 @@ Counts use `PRIMARY_OUTPUT` only and retain each adapter configuration as a sepa
 
 ### Radixor full-coverage aggregates
 
-These aggregates cover all 20 documented languages. Macro balanced accuracy gives each language equal weight. Micro metrics first sum raw pair counts across languages. Unsupported third-party languages are never inserted as zero results, so this full-coverage table is not presented as a cross-stemmer common-language ranking.
+These aggregates cover all 143 documented languages. Macro balanced accuracy gives each language equal weight. Micro metrics first sum raw pair counts across languages. Unsupported third-party languages are never inserted as zero results, so this full-coverage table is not presented as a cross-stemmer common-language ranking.
 
 | Dictionary mode | Languages | Macro balanced accuracy | Micro balanced accuracy | Micro precision | Micro recall | Micro F1 |
 |---|---:|---:|---:|---:|---:|---:|
-|ALL_WORDS|20|0.981225|0.988003|0.999988|0.976005|0.987851|
-|LOWERCASE_GROUPS_ONLY|20|0.984395|0.989641|0.999992|0.979282|0.989529|
+|ALL_WORDS|141 / 143|0.984680|0.987120|0.999993|0.974239|0.986948|
+|LOWERCASE_GROUPS_ONLY|141 / 143|0.985130|0.988026|0.999996|0.976053|0.987879|
 
 ### Reproducible data
 
-- [Machine-readable quality snapshot](data/stemming-quality.csv)
-- SHA-256: `85763189eab4d0fbb047c2d5d3554c66abf9732182bd0d8fd758d7aef680e66f`
+- [Machine-readable quality snapshot](data/stemming-quality-2026-09-11.csv)
+- SHA-256: `24bddfeed06a60bb3eeed58e1bfc93aed46c1d32e7bebd293aec2dacfddfef5b`
 - [Linguistic quality methodology](reference/linguistic-quality.md)
 - [Tested stemmer inventory](reference/tested-stemmers.md)
 - [Reproducibility and raw data](reference/reproducibility.md)
diff --git a/docs/benchmarks/languages/czech.md b/docs/benchmarks/languages/czech.md
index ef7e261..894f474 100644
--- a/docs/benchmarks/languages/czech.md
+++ b/docs/benchmarks/languages/czech.md
@@ -1,8 +1,12 @@
-# Czech Stemmer Benchmarks
+# Czech Stemmer Benchmarks <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 51,401 distinct usable word forms" title="Relative dictionary size 4 of 5; 51,401 distinct usable word forms">★★★★☆</span>
 
 This page reports same-language stemming benchmarks for Czech. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
 
-All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-08-25 Radixor/Java `4.2.0-6-g84e57fb` snapshot. Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+<!-- DICTIONARY-SIZE-RATING:START -->
+Dictionary size: <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 51,401 distinct usable word forms" title="Relative dictionary size 4 of 5; 51,401 distinct usable word forms">★★★★☆</span>. The exact count is **51,401 distinct usable word forms** after parser-compatible filtering and exact, case-preserved deduplication. Stars rank dictionary size relative to all 167 user-facing models in five nearly equal groups; they do **not** measure linguistic quality or benchmark accuracy.
+<!-- DICTIONARY-SIZE-RATING:END -->
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-09-11 Radixor/Java `4.3.0.dirty` snapshot. Speed benchmark operations process changed tokens. Accuracy uses the complete Radixor dictionary for the language.
 
 <!-- BENCHMARK-EVIDENCE-MAP:START -->
 !!! info "How to read this page"
@@ -13,9 +17,9 @@ Runtime and exact-root agreement measure different properties. Light, minimal, p
 
 ## Dictionary Corpus
 
-| Model ID | Model version | Language | Dictionary rows | Complete quality tokens | Already-root tokens | Changed tokens | JMH timing tokens |
-| --- | --- | --- | ---: | ---: | ---: | ---: | ---: |
-| `cs-cz-default` | `1.0.0` | `CS_CZ` | 5,113 | 56,612 | 10,049 | 46,563 | 46,563 |
+| Model ID | Model version | Language | Dictionary rows | Distinct usable forms | Complete quality tokens | Already-root tokens | Changed tokens | Timing workload | JMH timing tokens |
+| --- | --- | --- | ---: | ---: | ---: | ---: | ---: | --- | ---: |
+| `cs-cz-default` | `1.0.0` | `CS_CZ` | 5,113 | 51,401 | 56,612 | 10,049 | 46,563 | changed tokens | 46,563 |
 
 ## Radixor Patch Command Distribution
 
@@ -42,14 +46,14 @@ Accuracy is computed from JMH auxiliary counters in the current report. The coun
 
 ## Speed
 
-Speed uses JMH average time, 5 warmup iterations, 7 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
 
 | Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
 | --- | --- | ---: | ---: | ---: | ---: | --- |
-| Radixor | `czechRadixor` | 3.292 | 0.065 | 70.7 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
-| Lucene HunspellStemFilter | `luceneHunspellStemFilter` | 378.764 | 38.312 | 8134.4 | 115.055 | Benchmark-only Czech Hunspell dictionary compared via Lucene HunspellStemFilter. |
-| Lucene CzechStemFilter | `czechLuceneCzechStemFilter` | 2.958 | 0.050 | 63.5 | 0.898 | Czech suffix stemmer implemented as a Lucene TokenFilter. |
-| Official Snowball direct | `snowballDirect[CZECH]` | 4.057 | 0.516 | 87.1 | 1.232 | Official Snowball 3.1.0 generated Java stemmer; direct API. |
+| Radixor | `czechRadixor` | 3.361 | 0.088 | 72.2 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
+| Lucene HunspellStemFilter | `luceneHunspellStemFilter` | 352.480 | 40.873 | 7570.0 | 104.877 | Benchmark-only Czech Hunspell dictionary compared via Lucene HunspellStemFilter. |
+| Lucene CzechStemFilter | `czechLuceneCzechStemFilter` | 3.134 | 0.042 | 67.3 | 0.933 | Czech suffix stemmer implemented as a Lucene TokenFilter. |
+| Official Snowball direct | `snowballDirect[CZECH]` | 4.308 | 0.654 | 92.5 | 1.282 | Official Snowball 3.1.0 generated Java stemmer; direct API. |
 
 ## Interpretation Notes
 
@@ -87,18 +91,15 @@ also appeared in training. Parentheses show the observed split minimum–maximum
 
 ### Generalization conclusion
 
-- Median exactness on genuinely unseen changed forms moves from **65.439%**
-  at 10% training knowledge to **69.900%** at 90%, a measured
-  **+4.461 percentage-point** change for this dictionary.
-- Over the same endpoints, unseen all-form exactness changes by **+3.349 pp** and
-  preservation of unseen already-root forms changes by **+1.660 pp**. These separate
-  outcomes show whether the changed-form result coexists with preservation behavior.
+- Median exactness on genuinely unseen changed forms moves from **65.439%** at 10% training knowledge to **69.900%** at 90%, a measured **+4.461 percentage-point** change.
+- Unseen all-form exactness moves from **69.921%** at 10% training knowledge to **73.270%** at 90%, a measured **+3.349 percentage-point** change.
+- Preservation of unseen already-root forms moves from **89.636%** at 10% training knowledge to **91.296%** at 90%, a measured **+1.660 percentage-point** change.
 - The evidence establishes within-resource transfer across withheld dictionary families. It
   does not estimate unrelated domains, misspellings, arbitrary compounds, or external corpora.
 
 The complete ten-level table and split ranges remain in the
 [independent generalization report](../generalization.md); raw counters and provenance are in
-[`dictionary-generalization.csv`](../data/dictionary-generalization.csv). The
+[active machine-readable snapshot](../data/dictionary-generalization-2026-09-11.csv). The
 [frozen methodology](../reference/generalization-methodology.md) defines family-level
 splitting, unseen-surface leakage control, aggregation, and the limits of the claim.
 
@@ -191,7 +192,7 @@ The complete evidence is available in the [raw logical matrix](../data/edit-cost
 
 Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `CS_CZ` using the complete validated stemming-quality result matrix. Every distinct surface form is one evaluated item and can belong to several dictionary groups. Two forms are a positive pair when their group-membership sets intersect and a negative pair when those sets are disjoint. A pair shared through several groups is counted once. Exact equality with a predetermined lemma is not required.
 
-`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality-2026-09-11.csv).
 
 ### Evaluation Scope and Key Findings
 
@@ -460,8 +461,8 @@ Standard ARI, homogeneity, completeness, V-measure, and NMI are not calculated:
 
 ### Provenance
 
-- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
-- Source SHA-256: `85763189eab4d0fbb047c2d5d3554c66abf9732182bd0d8fd758d7aef680e66f`
+- Authoritative source: `docs/benchmarks/data/stemming-quality-2026-09-11.csv`
+- Source SHA-256: `24bddfeed06a60bb3eeed58e1bfc93aed46c1d32e7bebd293aec2dacfddfef5b`
 - Evaluation command: `./gradlew stemmingQuality --no-daemon`
 - Dictionary language: `CS_CZ`
 - Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
diff --git a/docs/benchmarks/languages/danish.md b/docs/benchmarks/languages/danish.md
index 3897a92..a16b4a2 100644
--- a/docs/benchmarks/languages/danish.md
+++ b/docs/benchmarks/languages/danish.md
@@ -1,8 +1,12 @@
-# Danish Stemmer Benchmarks
+# Danish Stemmer Benchmarks <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 27,921 distinct usable word forms" title="Relative dictionary size 4 of 5; 27,921 distinct usable word forms">★★★★☆</span>
 
 This page reports same-language stemming benchmarks for Danish. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
 
-All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-08-25 Radixor/Java `4.2.0-6-g84e57fb` snapshot. Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+<!-- DICTIONARY-SIZE-RATING:START -->
+Dictionary size: <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 27,921 distinct usable word forms" title="Relative dictionary size 4 of 5; 27,921 distinct usable word forms">★★★★☆</span>. The exact count is **27,921 distinct usable word forms** after parser-compatible filtering and exact, case-preserved deduplication. Stars rank dictionary size relative to all 167 user-facing models in five nearly equal groups; they do **not** measure linguistic quality or benchmark accuracy.
+<!-- DICTIONARY-SIZE-RATING:END -->
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-09-11 Radixor/Java `4.3.0.dirty` snapshot. Speed benchmark operations process changed tokens. Accuracy uses the complete Radixor dictionary for the language.
 
 <!-- BENCHMARK-EVIDENCE-MAP:START -->
 !!! info "How to read this page"
@@ -13,9 +17,9 @@ Runtime and exact-root agreement measure different properties. Light, minimal, p
 
 ## Dictionary Corpus
 
-| Model ID | Model version | Language | Dictionary rows | Complete quality tokens | Already-root tokens | Changed tokens | JMH timing tokens |
-| --- | --- | --- | ---: | ---: | ---: | ---: | ---: |
-| `da-dk-default` | `1.0.0` | `DA_DK` | 4,179 | 32,256 | 8,356 | 23,900 | 23,900 |
+| Model ID | Model version | Language | Dictionary rows | Distinct usable forms | Complete quality tokens | Already-root tokens | Changed tokens | Timing workload | JMH timing tokens |
+| --- | --- | --- | ---: | ---: | ---: | ---: | ---: | --- | ---: |
+| `da-dk-default` | `1.0.0` | `DA_DK` | 4,179 | 27,921 | 32,256 | 8,356 | 23,900 | changed tokens | 23,900 |
 
 ## Radixor Patch Command Distribution
 
@@ -41,13 +45,13 @@ Accuracy is computed from JMH auxiliary counters in the current report. The coun
 
 ## Speed
 
-Speed uses JMH average time, 5 warmup iterations, 7 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
 
 | Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
 | --- | --- | ---: | ---: | ---: | ---: | --- |
-| Radixor | `radixor[DANISH]` | 1.146 | 0.171 | 48.0 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
-| Official Snowball direct | `snowballDirect[DANISH]` | 2.617 | 0.255 | 109.5 | 2.283 | Official Snowball generated Java stemmer; direct API. |
-| Lucene SnowballFilter | `luceneSnowballFilter[DANISH]` | 3.013 | 0.332 | 126.1 | 2.629 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
+| Radixor | `radixor[DANISH]` | 1.361 | 0.216 | 57.0 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
+| Official Snowball direct | `snowballDirect[DANISH]` | 2.712 | 0.277 | 113.5 | 1.992 | Official Snowball generated Java stemmer; direct API. |
+| Lucene SnowballFilter | `luceneSnowballFilter[DANISH]` | 3.164 | 0.421 | 132.4 | 2.325 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
 
 ## Interpretation Notes
 
@@ -85,18 +89,15 @@ also appeared in training. Parentheses show the observed split minimum–maximum
 
 ### Generalization conclusion
 
-- Median exactness on genuinely unseen changed forms moves from **51.055%**
-  at 10% training knowledge to **67.504%** at 90%, a measured
-  **+16.449 percentage-point** change for this dictionary.
-- Over the same endpoints, unseen all-form exactness changes by **+13.087 pp** and
-  preservation of unseen already-root forms changes by **+4.480 pp**. These separate
-  outcomes show whether the changed-form result coexists with preservation behavior.
+- Median exactness on genuinely unseen changed forms moves from **51.055%** at 10% training knowledge to **67.504%** at 90%, a measured **+16.449 percentage-point** change.
+- Unseen all-form exactness moves from **61.104%** at 10% training knowledge to **74.192%** at 90%, a measured **+13.087 percentage-point** change.
+- Preservation of unseen already-root forms moves from **88.238%** at 10% training knowledge to **92.718%** at 90%, a measured **+4.480 percentage-point** change.
 - The evidence establishes within-resource transfer across withheld dictionary families. It
   does not estimate unrelated domains, misspellings, arbitrary compounds, or external corpora.
 
 The complete ten-level table and split ranges remain in the
 [independent generalization report](../generalization.md); raw counters and provenance are in
-[`dictionary-generalization.csv`](../data/dictionary-generalization.csv). The
+[active machine-readable snapshot](../data/dictionary-generalization-2026-09-11.csv). The
 [frozen methodology](../reference/generalization-methodology.md) defines family-level
 splitting, unseen-surface leakage control, aggregation, and the limits of the claim.
 
@@ -187,7 +188,7 @@ The complete evidence is available in the [raw logical matrix](../data/edit-cost
 
 Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `DA_DK` using the complete validated stemming-quality result matrix. Every distinct surface form is one evaluated item and can belong to several dictionary groups. Two forms are a positive pair when their group-membership sets intersect and a negative pair when those sets are disjoint. A pair shared through several groups is counted once. Exact equality with a predetermined lemma is not required.
 
-`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality-2026-09-11.csv).
 
 ### Evaluation Scope and Key Findings
 
@@ -434,8 +435,8 @@ Standard ARI, homogeneity, completeness, V-measure, and NMI are not calculated:
 
 ### Provenance
 
-- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
-- Source SHA-256: `85763189eab4d0fbb047c2d5d3554c66abf9732182bd0d8fd758d7aef680e66f`
+- Authoritative source: `docs/benchmarks/data/stemming-quality-2026-09-11.csv`
+- Source SHA-256: `24bddfeed06a60bb3eeed58e1bfc93aed46c1d32e7bebd293aec2dacfddfef5b`
 - Evaluation command: `./gradlew stemmingQuality --no-daemon`
 - Dictionary language: `DA_DK`
 - Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
diff --git a/docs/benchmarks/languages/dutch.md b/docs/benchmarks/languages/dutch.md
index d20ded9..1db4bb1 100644
--- a/docs/benchmarks/languages/dutch.md
+++ b/docs/benchmarks/languages/dutch.md
@@ -1,8 +1,12 @@
-# Dutch Stemmer Benchmarks
+# Dutch Stemmer Benchmarks <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 26,201 distinct usable word forms" title="Relative dictionary size 4 of 5; 26,201 distinct usable word forms">★★★★☆</span>
 
 This page reports same-language stemming benchmarks for Dutch. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
 
-All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-08-25 Radixor/Java `4.2.0-6-g84e57fb` snapshot. Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+<!-- DICTIONARY-SIZE-RATING:START -->
+Dictionary size: <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 26,201 distinct usable word forms" title="Relative dictionary size 4 of 5; 26,201 distinct usable word forms">★★★★☆</span>. The exact count is **26,201 distinct usable word forms** after parser-compatible filtering and exact, case-preserved deduplication. Stars rank dictionary size relative to all 167 user-facing models in five nearly equal groups; they do **not** measure linguistic quality or benchmark accuracy.
+<!-- DICTIONARY-SIZE-RATING:END -->
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-09-11 Radixor/Java `4.3.0.dirty` snapshot. Speed benchmark operations process changed tokens. Accuracy uses the complete Radixor dictionary for the language.
 
 <!-- BENCHMARK-EVIDENCE-MAP:START -->
 !!! info "How to read this page"
@@ -13,9 +17,9 @@ Runtime and exact-root agreement measure different properties. Light, minimal, p
 
 ## Dictionary Corpus
 
-| Model ID | Model version | Language | Dictionary rows | Complete quality tokens | Already-root tokens | Changed tokens | JMH timing tokens |
-| --- | --- | --- | ---: | ---: | ---: | ---: | ---: |
-| `nl-nl-default` | `1.0.0` | `NL_NL` | 4,992 | 31,466 | 9,981 | 21,485 | 21,485 |
+| Model ID | Model version | Language | Dictionary rows | Distinct usable forms | Complete quality tokens | Already-root tokens | Changed tokens | Timing workload | JMH timing tokens |
+| --- | --- | --- | ---: | ---: | ---: | ---: | ---: | --- | ---: |
+| `nl-nl-default` | `1.0.0` | `NL_NL` | 4,992 | 26,201 | 31,466 | 9,981 | 21,485 | changed tokens | 21,485 |
 
 ## Radixor Patch Command Distribution
 
@@ -42,14 +46,14 @@ Accuracy is computed from JMH auxiliary counters in the current report. The coun
 
 ## Speed
 
-Speed uses JMH average time, 5 warmup iterations, 7 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
 
 | Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
 | --- | --- | ---: | ---: | ---: | ---: | --- |
-| Radixor | `radixor[DUTCH]` | 1.395 | 0.185 | 64.9 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
-| Lucene HunspellStemFilter | `luceneHunspellStemFilter` | 26.133 | 4.062 | 1216.3 | 18.735 | Benchmark-only Dutch Hunspell dictionary compared via Lucene HunspellStemFilter. |
-| Official Snowball direct | `snowballDirect[DUTCH]` | 4.299 | 0.274 | 200.1 | 3.082 | Official Snowball generated Java stemmer; direct API. |
-| Lucene SnowballFilter | `luceneSnowballFilter[DUTCH]` | 7.270 | 0.290 | 338.4 | 5.212 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
+| Radixor | `radixor[DUTCH]` | 1.540 | 0.278 | 71.7 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
+| Lucene HunspellStemFilter | `luceneHunspellStemFilter` | 28.377 | 4.971 | 1320.8 | 18.426 | Benchmark-only Dutch Hunspell dictionary compared via Lucene HunspellStemFilter. |
+| Official Snowball direct | `snowballDirect[DUTCH]` | 4.102 | 0.293 | 190.9 | 2.663 | Official Snowball generated Java stemmer; direct API. |
+| Lucene SnowballFilter | `luceneSnowballFilter[DUTCH]` | 7.211 | 0.376 | 335.6 | 4.682 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
 
 ## Interpretation Notes
 
@@ -87,18 +91,15 @@ also appeared in training. Parentheses show the observed split minimum–maximum
 
 ### Generalization conclusion
 
-- Median exactness on genuinely unseen changed forms moves from **55.524%**
-  at 10% training knowledge to **65.891%** at 90%, a measured
-  **+10.366 percentage-point** change for this dictionary.
-- Over the same endpoints, unseen all-form exactness changes by **+7.429 pp** and
-  preservation of unseen already-root forms changes by **+1.896 pp**. These separate
-  outcomes show whether the changed-form result coexists with preservation behavior.
+- Median exactness on genuinely unseen changed forms moves from **55.524%** at 10% training knowledge to **65.891%** at 90%, a measured **+10.366 percentage-point** change.
+- Unseen all-form exactness moves from **67.039%** at 10% training knowledge to **74.468%** at 90%, a measured **+7.429 percentage-point** change.
+- Preservation of unseen already-root forms moves from **92.616%** at 10% training knowledge to **94.512%** at 90%, a measured **+1.896 percentage-point** change.
 - The evidence establishes within-resource transfer across withheld dictionary families. It
   does not estimate unrelated domains, misspellings, arbitrary compounds, or external corpora.
 
 The complete ten-level table and split ranges remain in the
 [independent generalization report](../generalization.md); raw counters and provenance are in
-[`dictionary-generalization.csv`](../data/dictionary-generalization.csv). The
+[active machine-readable snapshot](../data/dictionary-generalization-2026-09-11.csv). The
 [frozen methodology](../reference/generalization-methodology.md) defines family-level
 splitting, unseen-surface leakage control, aggregation, and the limits of the claim.
 
@@ -191,7 +192,7 @@ The complete evidence is available in the [raw logical matrix](../data/edit-cost
 
 Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `NL_NL` using the complete validated stemming-quality result matrix. Every distinct surface form is one evaluated item and can belong to several dictionary groups. Two forms are a positive pair when their group-membership sets intersect and a negative pair when those sets are disjoint. A pair shared through several groups is counted once. Exact equality with a predetermined lemma is not required.
 
-`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality-2026-09-11.csv).
 
 ### Evaluation Scope and Key Findings
 
@@ -460,8 +461,8 @@ Standard ARI, homogeneity, completeness, V-measure, and NMI are not calculated:
 
 ### Provenance
 
-- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
-- Source SHA-256: `85763189eab4d0fbb047c2d5d3554c66abf9732182bd0d8fd758d7aef680e66f`
+- Authoritative source: `docs/benchmarks/data/stemming-quality-2026-09-11.csv`
+- Source SHA-256: `24bddfeed06a60bb3eeed58e1bfc93aed46c1d32e7bebd293aec2dacfddfef5b`
 - Evaluation command: `./gradlew stemmingQuality --no-daemon`
 - Dictionary language: `NL_NL`
 - Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
diff --git a/docs/benchmarks/languages/english.md b/docs/benchmarks/languages/english.md
index faf0ccb..94fdbaa 100644
--- a/docs/benchmarks/languages/english.md
+++ b/docs/benchmarks/languages/english.md
@@ -1,8 +1,12 @@
-# English Stemmer Benchmarks
+# English Stemmer Benchmarks <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 591,946 distinct usable word forms" title="Relative dictionary size 5 of 5; 591,946 distinct usable word forms">★★★★★</span>
 
 This page reports same-language stemming benchmarks for English. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
 
-All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-08-25 Radixor/Java `4.2.0-6-g84e57fb` snapshot. Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+<!-- DICTIONARY-SIZE-RATING:START -->
+Dictionary size: <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 591,946 distinct usable word forms" title="Relative dictionary size 5 of 5; 591,946 distinct usable word forms">★★★★★</span>. The exact count is **591,946 distinct usable word forms** after parser-compatible filtering and exact, case-preserved deduplication. Stars rank dictionary size relative to all 167 user-facing models in five nearly equal groups; they do **not** measure linguistic quality or benchmark accuracy.
+<!-- DICTIONARY-SIZE-RATING:END -->
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-09-11 Radixor/Java `4.3.0.dirty` snapshot. Speed benchmark operations process changed tokens. Accuracy uses the complete Radixor dictionary for the language.
 
 <!-- BENCHMARK-EVIDENCE-MAP:START -->
 !!! info "How to read this page"
@@ -13,9 +17,9 @@ Runtime and exact-root agreement measure different properties. Light, minimal, p
 
 ## Dictionary Corpus
 
-| Model ID | Model version | Language | Dictionary rows | Complete quality tokens | Already-root tokens | Changed tokens | JMH timing tokens |
-| --- | --- | --- | ---: | ---: | ---: | ---: | ---: |
-| `us-uk-default` | `1.0.1` | `US_UK` | 396,939 | 1,002,414 | 793,874 | 208,540 | 208,540 |
+| Model ID | Model version | Language | Dictionary rows | Distinct usable forms | Complete quality tokens | Already-root tokens | Changed tokens | Timing workload | JMH timing tokens |
+| --- | --- | --- | ---: | ---: | ---: | ---: | ---: | --- | ---: |
+| `us-uk-default` | `1.0.1` | `US_UK` | 396,939 | 591,946 | 1,002,414 | 793,874 | 208,540 | changed tokens | 208,540 |
 
 ## Radixor Patch Command Distribution
 
@@ -49,26 +53,26 @@ Accuracy is computed from JMH auxiliary counters in the current report. The coun
 
 ## Speed
 
-Speed uses JMH average time, 5 warmup iterations, 7 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
 
 <!-- ENGLISH-SPEED-SUITES:START -->
 !!! note "Separate English speed suites"
-    The `80.8 ns/token` Radixor value below is from the multilingual same-language comparison suite. The [coverage experiment](../reference/english-coverage.md) reports its own full-knowledge point from a separate benchmark method and run. Treat both as suite-specific estimates with their published uncertainty, not as interchangeable values.
+    The `84.7 ns/token` Radixor value below is from the multilingual same-language comparison suite. The [coverage experiment](../reference/english-coverage.md) reports its own full-knowledge point from a separate benchmark method and run. Treat both as suite-specific estimates with their published uncertainty, not as interchangeable values.
 <!-- ENGLISH-SPEED-SUITES:END -->
 
 | Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
 | --- | --- | ---: | ---: | ---: | ---: | --- |
-| Radixor | `radixorUsUkProfiPreferredStem` | 16.840 | 1.732 | 80.8 | 1.000 | Full dictionary patch-command stemmer using compiled patch commands. |
-| Lucene EnglishPossessiveFilter | `luceneEnglishPossessiveFilter` | 16.439 | 0.771 | 78.8 | 0.976 | Possessive-ending remover only; not a full stemmer. |
-| Lucene EnglishMinimalStemFilter | `luceneEnglishMinimalStemFilter` | 18.665 | 0.717 | 89.5 | 1.108 | Narrow plural reduction filter; not a full stemmer. |
-| Lucene PorterStemmer direct copy | `lucenePorterStemmerCopied` | 16.770 | 0.260 | 80.4 | 0.996 | Benchmark-only generated copy of Lucene package-private Porter implementation. |
-| OpenNLP PorterStemmer | `opennlpPorterStemmer` | 16.888 | 0.247 | 81.0 | 1.003 | Apache OpenNLP Porter implementation. |
-| Snowball original Porter | `snowballOriginalPorter` | 32.842 | 2.612 | 157.5 | 1.950 | Classic Porter suffix-rule stemmer; historical English baseline, not a dictionary-equivalent stemmer. |
-| Lucene PorterStemFilter | `lucenePorterStemFilter` | 31.412 | 0.750 | 150.6 | 1.865 | Lucene TokenFilter integration path for Porter; includes TokenStream overhead. |
-| Lucene KStemFilter | `luceneKStemFilter` | 43.402 | 1.146 | 208.1 | 2.577 | Krovetz-style English TokenFilter; broader than minimal suffix filters. |
-| Lucene HunspellStemFilter | `luceneHunspellStemFilter` | 72.329 | 1.052 | 346.8 | 4.295 | Benchmark-only English Hunspell comparison using the benchmark Hunspell corpus. |
-| Snowball English / Porter2 | `snowballEnglishPorter2` | 46.742 | 2.724 | 224.1 | 2.776 | Porter2 suffix-rule stemmer, distinct from original Porter. |
-| Paice/Husk Lancaster | `paiceHuskLancaster` | 138.158 | 3.983 | 662.5 | 8.204 | Aggressive rule-based English stemmer. |
+| Radixor | `radixorUsUkProfiPreferredStem` | 17.654 | 2.053 | 84.7 | 1.000 | Full dictionary patch-command stemmer using compiled patch commands. |
+| Lucene EnglishPossessiveFilter | `luceneEnglishPossessiveFilter` | 15.732 | 0.282 | 75.4 | 0.891 | Possessive-ending remover only; not a full stemmer. |
+| Lucene EnglishMinimalStemFilter | `luceneEnglishMinimalStemFilter` | 17.249 | 0.344 | 82.7 | 0.977 | Narrow plural reduction filter; not a full stemmer. |
+| Lucene PorterStemmer direct copy | `lucenePorterStemmerCopied` | 19.164 | 3.414 | 91.9 | 1.086 | Benchmark-only generated copy of Lucene package-private Porter implementation. |
+| OpenNLP PorterStemmer | `opennlpPorterStemmer` | 18.592 | 2.908 | 89.2 | 1.053 | Apache OpenNLP Porter implementation. |
+| Snowball original Porter | `snowballOriginalPorter` | 34.518 | 2.150 | 165.5 | 1.955 | Classic Porter suffix-rule stemmer; historical English baseline, not a dictionary-equivalent stemmer. |
+| Lucene PorterStemFilter | `lucenePorterStemFilter` | 29.831 | 0.457 | 143.0 | 1.690 | Lucene TokenFilter integration path for Porter; includes TokenStream overhead. |
+| Lucene KStemFilter | `luceneKStemFilter` | 42.618 | 1.848 | 204.4 | 2.414 | Krovetz-style English TokenFilter; broader than minimal suffix filters. |
+| Lucene HunspellStemFilter | `luceneHunspellStemFilter` | 74.214 | 6.598 | 355.9 | 4.204 | Benchmark-only English Hunspell comparison using the benchmark Hunspell corpus. |
+| Snowball English / Porter2 | `snowballEnglishPorter2` | 46.211 | 1.606 | 221.6 | 2.618 | Porter2 suffix-rule stemmer, distinct from original Porter. |
+| Paice/Husk Lancaster | `paiceHuskLancaster` | 139.249 | 3.975 | 667.7 | 7.888 | Aggressive rule-based English stemmer. |
 
 ## Interpretation Notes
 
@@ -106,18 +110,15 @@ also appeared in training. Parentheses show the observed split minimum–maximum
 
 ### Generalization conclusion
 
-- Median exactness on genuinely unseen changed forms moves from **76.010%**
-  at 10% training knowledge to **78.854%** at 90%, a measured
-  **+2.843 percentage-point** change for this dictionary.
-- Over the same endpoints, unseen all-form exactness changes by **+1.875 pp** and
-  preservation of unseen already-root forms changes by **+1.560 pp**. These separate
-  outcomes show whether the changed-form result coexists with preservation behavior.
+- Median exactness on genuinely unseen changed forms moves from **76.010%** at 10% training knowledge to **78.854%** at 90%, a measured **+2.843 percentage-point** change.
+- Unseen all-form exactness moves from **92.805%** at 10% training knowledge to **94.680%** at 90%, a measured **+1.875 percentage-point** change.
+- Preservation of unseen already-root forms moves from **97.184%** at 10% training knowledge to **98.744%** at 90%, a measured **+1.560 percentage-point** change.
 - The evidence establishes within-resource transfer across withheld dictionary families. It
   does not estimate unrelated domains, misspellings, arbitrary compounds, or external corpora.
 
 The complete ten-level table and split ranges remain in the
 [independent generalization report](../generalization.md); raw counters and provenance are in
-[`dictionary-generalization.csv`](../data/dictionary-generalization.csv). The
+[active machine-readable snapshot](../data/dictionary-generalization-2026-09-11.csv). The
 [frozen methodology](../reference/generalization-methodology.md) defines family-level
 splitting, unseen-surface leakage control, aggregation, and the limits of the claim.
 
@@ -210,7 +211,7 @@ The complete evidence is available in the [raw logical matrix](../data/edit-cost
 
 Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `US_UK` using the complete validated stemming-quality result matrix. Every distinct surface form is one evaluated item and can belong to several dictionary groups. Two forms are a positive pair when their group-membership sets intersect and a negative pair when those sets are disjoint. A pair shared through several groups is counted once. Exact equality with a predetermined lemma is not required.
 
-`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality-2026-09-11.csv).
 
 ### Evaluation Scope and Key Findings
 
@@ -535,8 +536,8 @@ Standard ARI, homogeneity, completeness, V-measure, and NMI are not calculated:
 
 ### Provenance
 
-- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
-- Source SHA-256: `85763189eab4d0fbb047c2d5d3554c66abf9732182bd0d8fd758d7aef680e66f`
+- Authoritative source: `docs/benchmarks/data/stemming-quality-2026-09-11.csv`
+- Source SHA-256: `24bddfeed06a60bb3eeed58e1bfc93aed46c1d32e7bebd293aec2dacfddfef5b`
 - Evaluation command: `./gradlew stemmingQuality --no-daemon`
 - Dictionary language: `US_UK`
 - Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
diff --git a/docs/benchmarks/languages/finnish.md b/docs/benchmarks/languages/finnish.md
index 2881f9e..0921a2d 100644
--- a/docs/benchmarks/languages/finnish.md
+++ b/docs/benchmarks/languages/finnish.md
@@ -1,8 +1,12 @@
-# Finnish Stemmer Benchmarks
+# Finnish Stemmer Benchmarks <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 1,788,784 distinct usable word forms" title="Relative dictionary size 5 of 5; 1,788,784 distinct usable word forms">★★★★★</span>
 
 This page reports same-language stemming benchmarks for Finnish. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
 
-All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-08-25 Radixor/Java `4.2.0-6-g84e57fb` snapshot. Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+<!-- DICTIONARY-SIZE-RATING:START -->
+Dictionary size: <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 1,788,784 distinct usable word forms" title="Relative dictionary size 5 of 5; 1,788,784 distinct usable word forms">★★★★★</span>. The exact count is **1,788,784 distinct usable word forms** after parser-compatible filtering and exact, case-preserved deduplication. Stars rank dictionary size relative to all 167 user-facing models in five nearly equal groups; they do **not** measure linguistic quality or benchmark accuracy.
+<!-- DICTIONARY-SIZE-RATING:END -->
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-09-11 Radixor/Java `4.3.0.dirty` snapshot. Speed benchmark operations process changed tokens. Accuracy uses the complete Radixor dictionary for the language.
 
 <!-- BENCHMARK-EVIDENCE-MAP:START -->
 !!! info "How to read this page"
@@ -13,9 +17,9 @@ Runtime and exact-root agreement measure different properties. Light, minimal, p
 
 ## Dictionary Corpus
 
-| Model ID | Model version | Language | Dictionary rows | Complete quality tokens | Already-root tokens | Changed tokens | JMH timing tokens |
-| --- | --- | --- | ---: | ---: | ---: | ---: | ---: |
-| `fi-fi-default` | `1.0.0` | `FI_FI` | 57,027 | 1,865,215 | 110,525 | 1,754,690 | 1,754,690 |
+| Model ID | Model version | Language | Dictionary rows | Distinct usable forms | Complete quality tokens | Already-root tokens | Changed tokens | Timing workload | JMH timing tokens |
+| --- | --- | --- | ---: | ---: | ---: | ---: | ---: | --- | ---: |
+| `fi-fi-default` | `1.0.0` | `FI_FI` | 57,027 | 1,788,784 | 1,865,215 | 110,525 | 1,754,690 | changed tokens | 1,754,690 |
 
 ## Radixor Patch Command Distribution
 
@@ -42,14 +46,14 @@ Accuracy is computed from JMH auxiliary counters in the current report. The coun
 
 ## Speed
 
-Speed uses JMH average time, 5 warmup iterations, 7 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
 
 | Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
 | --- | --- | ---: | ---: | ---: | ---: | --- |
-| Radixor | `finnishRadixor` | 261.169 | 4.729 | 148.8 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
-| Lucene FinnishLightStemFilter | `finnishLuceneFinnishLightStemFilter` | 169.322 | 7.190 | 96.5 | 0.648 | Light Finnish suffix stemmer. |
-| Official Snowball direct | `snowballDirect[FINNISH]` | 263.094 | 24.236 | 149.9 | 1.007 | Official Snowball generated Java stemmer; direct API. |
-| Lucene SnowballFilter | `luceneSnowballFilter[FINNISH]` | 322.750 | 12.740 | 183.9 | 1.236 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
+| Radixor | `finnishRadixor` | 296.399 | 10.829 | 168.9 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
+| Lucene FinnishLightStemFilter | `finnishLuceneFinnishLightStemFilter` | 177.154 | 8.445 | 101.0 | 0.598 | Light Finnish suffix stemmer. |
+| Official Snowball direct | `snowballDirect[FINNISH]` | 286.169 | 15.290 | 163.1 | 0.965 | Official Snowball generated Java stemmer; direct API. |
+| Lucene SnowballFilter | `luceneSnowballFilter[FINNISH]` | 311.412 | 18.031 | 177.5 | 1.051 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
 
 ## Interpretation Notes
 
@@ -87,18 +91,15 @@ also appeared in training. Parentheses show the observed split minimum–maximum
 
 ### Generalization conclusion
 
-- Median exactness on genuinely unseen changed forms moves from **78.644%**
-  at 10% training knowledge to **86.177%** at 90%, a measured
-  **+7.532 percentage-point** change for this dictionary.
-- Over the same endpoints, unseen all-form exactness changes by **+7.434 pp** and
-  preservation of unseen already-root forms changes by **+6.588 pp**. These separate
-  outcomes show whether the changed-form result coexists with preservation behavior.
+- Median exactness on genuinely unseen changed forms moves from **78.644%** at 10% training knowledge to **86.177%** at 90%, a measured **+7.532 percentage-point** change.
+- Unseen all-form exactness moves from **79.210%** at 10% training knowledge to **86.644%** at 90%, a measured **+7.434 percentage-point** change.
+- Preservation of unseen already-root forms moves from **88.247%** at 10% training knowledge to **94.835%** at 90%, a measured **+6.588 percentage-point** change.
 - The evidence establishes within-resource transfer across withheld dictionary families. It
   does not estimate unrelated domains, misspellings, arbitrary compounds, or external corpora.
 
 The complete ten-level table and split ranges remain in the
 [independent generalization report](../generalization.md); raw counters and provenance are in
-[`dictionary-generalization.csv`](../data/dictionary-generalization.csv). The
+[active machine-readable snapshot](../data/dictionary-generalization-2026-09-11.csv). The
 [frozen methodology](../reference/generalization-methodology.md) defines family-level
 splitting, unseen-surface leakage control, aggregation, and the limits of the claim.
 
@@ -191,7 +192,7 @@ The complete evidence is available in the [raw logical matrix](../data/edit-cost
 
 Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `FI_FI` using the complete validated stemming-quality result matrix. Every distinct surface form is one evaluated item and can belong to several dictionary groups. Two forms are a positive pair when their group-membership sets intersect and a negative pair when those sets are disjoint. A pair shared through several groups is counted once. Exact equality with a predetermined lemma is not required.
 
-`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality-2026-09-11.csv).
 
 ### Evaluation Scope and Key Findings
 
@@ -446,8 +447,8 @@ Standard ARI, homogeneity, completeness, V-measure, and NMI are not calculated:
 
 ### Provenance
 
-- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
-- Source SHA-256: `85763189eab4d0fbb047c2d5d3554c66abf9732182bd0d8fd758d7aef680e66f`
+- Authoritative source: `docs/benchmarks/data/stemming-quality-2026-09-11.csv`
+- Source SHA-256: `24bddfeed06a60bb3eeed58e1bfc93aed46c1d32e7bebd293aec2dacfddfef5b`
 - Evaluation command: `./gradlew stemmingQuality --no-daemon`
 - Dictionary language: `FI_FI`
 - Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
diff --git a/docs/benchmarks/languages/french.md b/docs/benchmarks/languages/french.md
index f3426aa..f534134 100644
--- a/docs/benchmarks/languages/french.md
+++ b/docs/benchmarks/languages/french.md
@@ -1,8 +1,12 @@
-# French Stemmer Benchmarks
+# French Stemmer Benchmarks <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 404,011 distinct usable word forms" title="Relative dictionary size 5 of 5; 404,011 distinct usable word forms">★★★★★</span>
 
 This page reports same-language stemming benchmarks for French. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
 
-All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-08-25 Radixor/Java `4.2.0-6-g84e57fb` snapshot. Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+<!-- DICTIONARY-SIZE-RATING:START -->
+Dictionary size: <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 404,011 distinct usable word forms" title="Relative dictionary size 5 of 5; 404,011 distinct usable word forms">★★★★★</span>. The exact count is **404,011 distinct usable word forms** after parser-compatible filtering and exact, case-preserved deduplication. Stars rank dictionary size relative to all 167 user-facing models in five nearly equal groups; they do **not** measure linguistic quality or benchmark accuracy.
+<!-- DICTIONARY-SIZE-RATING:END -->
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-09-11 Radixor/Java `4.3.0.dirty` snapshot. Speed benchmark operations process changed tokens. Accuracy uses the complete Radixor dictionary for the language.
 
 <!-- BENCHMARK-EVIDENCE-MAP:START -->
 !!! info "How to read this page"
@@ -13,9 +17,9 @@ Runtime and exact-root agreement measure different properties. Light, minimal, p
 
 ## Dictionary Corpus
 
-| Model ID | Model version | Language | Dictionary rows | Complete quality tokens | Already-root tokens | Changed tokens | JMH timing tokens |
-| --- | --- | --- | ---: | ---: | ---: | ---: | ---: |
-| `fr-fr-default` | `1.0.0` | `FR_FR` | 59,240 | 474,110 | 108,141 | 365,969 | 365,969 |
+| Model ID | Model version | Language | Dictionary rows | Distinct usable forms | Complete quality tokens | Already-root tokens | Changed tokens | Timing workload | JMH timing tokens |
+| --- | --- | --- | ---: | ---: | ---: | ---: | ---: | --- | ---: |
+| `fr-fr-default` | `1.0.0` | `FR_FR` | 59,240 | 404,011 | 474,110 | 108,141 | 365,969 | changed tokens | 365,969 |
 
 ## Radixor Patch Command Distribution
 
@@ -44,16 +48,16 @@ Accuracy is computed from JMH auxiliary counters in the current report. The coun
 
 ## Speed
 
-Speed uses JMH average time, 5 warmup iterations, 7 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
 
 | Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
 | --- | --- | ---: | ---: | ---: | ---: | --- |
-| Radixor | `frenchRadixor` | 42.603 | 1.019 | 116.4 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
-| Lucene HunspellStemFilter | `luceneHunspellStemFilter` | 1699.026 | 35.808 | 4642.5 | 39.881 | Benchmark-only French Hunspell dictionary compared via Lucene HunspellStemFilter. |
-| Lucene FrenchMinimalStemFilter | `frenchLuceneFrenchMinimalStemFilter` | 17.239 | 0.422 | 47.1 | 0.405 | Minimal French suffix reducer; narrow baseline. |
-| Lucene FrenchLightStemFilter | `frenchLuceneFrenchLightStemFilter` | 28.180 | 1.108 | 77.0 | 0.661 | Light French suffix stemmer. |
-| Official Snowball direct | `snowballDirect[FRENCH]` | 110.345 | 5.041 | 301.5 | 2.590 | Official Snowball generated Java stemmer; direct API. |
-| Lucene SnowballFilter | `luceneSnowballFilter[FRENCH]` | 124.104 | 5.294 | 339.1 | 2.913 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
+| Radixor | `frenchRadixor` | 46.930 | 1.367 | 128.2 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
+| Lucene HunspellStemFilter | `luceneHunspellStemFilter` | 1762.626 | 129.299 | 4816.3 | 37.559 | Benchmark-only French Hunspell dictionary compared via Lucene HunspellStemFilter. |
+| Lucene FrenchMinimalStemFilter | `frenchLuceneFrenchMinimalStemFilter` | 18.828 | 1.281 | 51.4 | 0.401 | Minimal French suffix reducer; narrow baseline. |
+| Lucene FrenchLightStemFilter | `frenchLuceneFrenchLightStemFilter` | 29.206 | 1.693 | 79.8 | 0.622 | Light French suffix stemmer. |
+| Official Snowball direct | `snowballDirect[FRENCH]` | 107.915 | 5.830 | 294.9 | 2.299 | Official Snowball generated Java stemmer; direct API. |
+| Lucene SnowballFilter | `luceneSnowballFilter[FRENCH]` | 121.681 | 5.945 | 332.5 | 2.593 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
 
 ## Interpretation Notes
 
@@ -91,18 +95,15 @@ also appeared in training. Parentheses show the observed split minimum–maximum
 
 ### Generalization conclusion
 
-- Median exactness on genuinely unseen changed forms moves from **69.627%**
-  at 10% training knowledge to **80.031%** at 90%, a measured
-  **+10.404 percentage-point** change for this dictionary.
-- Over the same endpoints, unseen all-form exactness changes by **+9.193 pp** and
-  preservation of unseen already-root forms changes by **+6.651 pp**. These separate
-  outcomes show whether the changed-form result coexists with preservation behavior.
+- Median exactness on genuinely unseen changed forms moves from **69.627%** at 10% training knowledge to **80.031%** at 90%, a measured **+10.404 percentage-point** change.
+- Unseen all-form exactness moves from **74.276%** at 10% training knowledge to **83.469%** at 90%, a measured **+9.193 percentage-point** change.
+- Preservation of unseen already-root forms moves from **90.013%** at 10% training knowledge to **96.664%** at 90%, a measured **+6.651 percentage-point** change.
 - The evidence establishes within-resource transfer across withheld dictionary families. It
   does not estimate unrelated domains, misspellings, arbitrary compounds, or external corpora.
 
 The complete ten-level table and split ranges remain in the
 [independent generalization report](../generalization.md); raw counters and provenance are in
-[`dictionary-generalization.csv`](../data/dictionary-generalization.csv). The
+[active machine-readable snapshot](../data/dictionary-generalization-2026-09-11.csv). The
 [frozen methodology](../reference/generalization-methodology.md) defines family-level
 splitting, unseen-surface leakage control, aggregation, and the limits of the claim.
 
@@ -195,7 +196,7 @@ The complete evidence is available in the [raw logical matrix](../data/edit-cost
 
 Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `FR_FR` using the complete validated stemming-quality result matrix. Every distinct surface form is one evaluated item and can belong to several dictionary groups. Two forms are a positive pair when their group-membership sets intersect and a negative pair when those sets are disjoint. A pair shared through several groups is counted once. Exact equality with a predetermined lemma is not required.
 
-`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality-2026-09-11.csv).
 
 ### Evaluation Scope and Key Findings
 
@@ -480,8 +481,8 @@ Standard ARI, homogeneity, completeness, V-measure, and NMI are not calculated:
 
 ### Provenance
 
-- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
-- Source SHA-256: `85763189eab4d0fbb047c2d5d3554c66abf9732182bd0d8fd758d7aef680e66f`
+- Authoritative source: `docs/benchmarks/data/stemming-quality-2026-09-11.csv`
+- Source SHA-256: `24bddfeed06a60bb3eeed58e1bfc93aed46c1d32e7bebd293aec2dacfddfef5b`
 - Evaluation command: `./gradlew stemmingQuality --no-daemon`
 - Dictionary language: `FR_FR`
 - Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
diff --git a/docs/benchmarks/languages/german.md b/docs/benchmarks/languages/german.md
index a4ba23e..671aa18 100644
--- a/docs/benchmarks/languages/german.md
+++ b/docs/benchmarks/languages/german.md
@@ -1,8 +1,12 @@
-# German Stemmer Benchmarks
+# German Stemmer Benchmarks <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 277,266 distinct usable word forms" title="Relative dictionary size 5 of 5; 277,266 distinct usable word forms">★★★★★</span>
 
 This page reports same-language stemming benchmarks for German. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
 
-All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-08-25 Radixor/Java `4.2.0-6-g84e57fb` snapshot. Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+<!-- DICTIONARY-SIZE-RATING:START -->
+Dictionary size: <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 277,266 distinct usable word forms" title="Relative dictionary size 5 of 5; 277,266 distinct usable word forms">★★★★★</span>. The exact count is **277,266 distinct usable word forms** after parser-compatible filtering and exact, case-preserved deduplication. Stars rank dictionary size relative to all 167 user-facing models in five nearly equal groups; they do **not** measure linguistic quality or benchmark accuracy.
+<!-- DICTIONARY-SIZE-RATING:END -->
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-09-11 Radixor/Java `4.3.0.dirty` snapshot. Speed benchmark operations process changed tokens. Accuracy uses the complete Radixor dictionary for the language.
 
 <!-- BENCHMARK-EVIDENCE-MAP:START -->
 !!! info "How to read this page"
@@ -13,9 +17,9 @@ Runtime and exact-root agreement measure different properties. Light, minimal, p
 
 ## Dictionary Corpus
 
-| Model ID | Model version | Language | Dictionary rows | Complete quality tokens | Already-root tokens | Changed tokens | JMH timing tokens |
-| --- | --- | --- | ---: | ---: | ---: | ---: | ---: |
-| `de-de-default` | `1.0.0` | `DE_DE` | 54,092 | 333,036 | 90,535 | 242,501 | 242,501 |
+| Model ID | Model version | Language | Dictionary rows | Distinct usable forms | Complete quality tokens | Already-root tokens | Changed tokens | Timing workload | JMH timing tokens |
+| --- | --- | --- | ---: | ---: | ---: | ---: | ---: | --- | ---: |
+| `de-de-default` | `1.0.0` | `DE_DE` | 54,092 | 277,266 | 333,036 | 90,535 | 242,501 | changed tokens | 242,501 |
 
 ## Radixor Patch Command Distribution
 
@@ -46,18 +50,18 @@ Accuracy is computed from JMH auxiliary counters in the current report. The coun
 
 ## Speed
 
-Speed uses JMH average time, 5 warmup iterations, 7 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
 
 | Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
 | --- | --- | ---: | ---: | ---: | ---: | --- |
-| Radixor | `germanRadixor` | 36.523 | 1.174 | 150.6 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
-| CISTEM | `germanCistem` | 289.327 | 8.094 | 1193.1 | 7.922 | Benchmark-only CISTEM implementation. |
-| Lucene HunspellStemFilter | `luceneHunspellStemFilter` | 276.775 | 22.204 | 1141.3 | 7.578 | Benchmark-only German Hunspell dictionary compared via Lucene HunspellStemFilter. |
-| Lucene GermanMinimalStemFilter | `germanLuceneGermanMinimalStemFilter` | 22.697 | 0.295 | 93.6 | 0.621 | Minimal German suffix reduction; narrow baseline. |
-| Lucene GermanLightStemFilter | `germanLuceneGermanLightStemFilter` | 23.477 | 0.264 | 96.8 | 0.643 | Light German suffix stemmer; narrower than Radixor's lexicon-trained transformation model. |
-| Lucene GermanStemFilter | `germanLuceneGermanStemFilter` | 67.524 | 0.851 | 278.4 | 1.849 | Older German stemming TokenFilter with normalization requirements. |
-| Lucene SnowballFilter | `luceneSnowballFilter[GERMAN]` | 104.691 | 3.041 | 431.7 | 2.866 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
-| Official Snowball direct | `snowballDirect[GERMAN]` | 95.094 | 3.149 | 392.1 | 2.604 | Official Snowball generated Java stemmer; direct API. |
+| Radixor | `germanRadixor` | 38.055 | 1.638 | 156.9 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
+| CISTEM | `germanCistem` | 289.668 | 11.748 | 1194.5 | 7.612 | Benchmark-only CISTEM implementation. |
+| Lucene HunspellStemFilter | `luceneHunspellStemFilter` | 308.274 | 46.165 | 1271.2 | 8.101 | Benchmark-only German Hunspell dictionary compared via Lucene HunspellStemFilter. |
+| Lucene GermanMinimalStemFilter | `germanLuceneGermanMinimalStemFilter` | 23.477 | 2.303 | 96.8 | 0.617 | Minimal German suffix reduction; narrow baseline. |
+| Lucene GermanLightStemFilter | `germanLuceneGermanLightStemFilter` | 23.050 | 0.269 | 95.1 | 0.606 | Light German suffix stemmer; narrower than Radixor's lexicon-trained transformation model. |
+| Lucene GermanStemFilter | `germanLuceneGermanStemFilter` | 66.984 | 2.175 | 276.2 | 1.760 | Older German stemming TokenFilter with normalization requirements. |
+| Lucene SnowballFilter | `luceneSnowballFilter[GERMAN]` | 103.452 | 6.867 | 426.6 | 2.718 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
+| Official Snowball direct | `snowballDirect[GERMAN]` | 93.896 | 4.248 | 387.2 | 2.467 | Official Snowball generated Java stemmer; direct API. |
 
 ## Interpretation Notes
 
@@ -95,18 +99,15 @@ also appeared in training. Parentheses show the observed split minimum–maximum
 
 ### Generalization conclusion
 
-- Median exactness on genuinely unseen changed forms moves from **49.432%**
-  at 10% training knowledge to **59.108%** at 90%, a measured
-  **+9.676 percentage-point** change for this dictionary.
-- Over the same endpoints, unseen all-form exactness changes by **+8.042 pp** and
-  preservation of unseen already-root forms changes by **+6.967 pp**. These separate
-  outcomes show whether the changed-form result coexists with preservation behavior.
+- Median exactness on genuinely unseen changed forms moves from **49.432%** at 10% training knowledge to **59.108%** at 90%, a measured **+9.676 percentage-point** change.
+- Unseen all-form exactness moves from **59.280%** at 10% training knowledge to **67.322%** at 90%, a measured **+8.042 percentage-point** change.
+- Preservation of unseen already-root forms moves from **86.150%** at 10% training knowledge to **93.117%** at 90%, a measured **+6.967 percentage-point** change.
 - The evidence establishes within-resource transfer across withheld dictionary families. It
   does not estimate unrelated domains, misspellings, arbitrary compounds, or external corpora.
 
 The complete ten-level table and split ranges remain in the
 [independent generalization report](../generalization.md); raw counters and provenance are in
-[`dictionary-generalization.csv`](../data/dictionary-generalization.csv). The
+[active machine-readable snapshot](../data/dictionary-generalization-2026-09-11.csv). The
 [frozen methodology](../reference/generalization-methodology.md) defines family-level
 splitting, unseen-surface leakage control, aggregation, and the limits of the claim.
 
@@ -199,7 +200,7 @@ The complete evidence is available in the [raw logical matrix](../data/edit-cost
 
 Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `DE_DE` using the complete validated stemming-quality result matrix. Every distinct surface form is one evaluated item and can belong to several dictionary groups. Two forms are a positive pair when their group-membership sets intersect and a negative pair when those sets are disjoint. A pair shared through several groups is counted once. Exact equality with a predetermined lemma is not required.
 
-`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality-2026-09-11.csv).
 
 ### Evaluation Scope and Key Findings
 
@@ -500,8 +501,8 @@ Standard ARI, homogeneity, completeness, V-measure, and NMI are not calculated:
 
 ### Provenance
 
-- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
-- Source SHA-256: `85763189eab4d0fbb047c2d5d3554c66abf9732182bd0d8fd758d7aef680e66f`
+- Authoritative source: `docs/benchmarks/data/stemming-quality-2026-09-11.csv`
+- Source SHA-256: `24bddfeed06a60bb3eeed58e1bfc93aed46c1d32e7bebd293aec2dacfddfef5b`
 - Evaluation command: `./gradlew stemmingQuality --no-daemon`
 - Dictionary language: `DE_DE`
 - Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
diff --git a/docs/benchmarks/languages/hebrew.md b/docs/benchmarks/languages/hebrew.md
index 8cdc6f3..9cffb96 100644
--- a/docs/benchmarks/languages/hebrew.md
+++ b/docs/benchmarks/languages/hebrew.md
@@ -1,8 +1,12 @@
-# Hebrew Stemmer Benchmarks
+# Hebrew Stemmer Benchmarks <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 57,658 distinct usable word forms" title="Relative dictionary size 4 of 5; 57,658 distinct usable word forms">★★★★☆</span>
 
 This page reports same-language stemming benchmarks for Hebrew. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
 
-All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-08-25 Radixor/Java `4.2.0-6-g84e57fb` snapshot. Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+<!-- DICTIONARY-SIZE-RATING:START -->
+Dictionary size: <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 57,658 distinct usable word forms" title="Relative dictionary size 4 of 5; 57,658 distinct usable word forms">★★★★☆</span>. The exact count is **57,658 distinct usable word forms** after parser-compatible filtering and exact, case-preserved deduplication. Stars rank dictionary size relative to all 167 user-facing models in five nearly equal groups; they do **not** measure linguistic quality or benchmark accuracy.
+<!-- DICTIONARY-SIZE-RATING:END -->
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-09-11 Radixor/Java `4.3.0.dirty` snapshot. Speed benchmark operations process changed tokens. Accuracy uses the complete Radixor dictionary for the language.
 
 <!-- BENCHMARK-EVIDENCE-MAP:START -->
 !!! info "How to read this page"
@@ -13,9 +17,9 @@ The default Hebrew model currently has no same-language third-party adapter in t
 
 ## Dictionary Corpus
 
-| Model ID | Model version | Language | Dictionary rows | Complete quality tokens | Already-root tokens | Changed tokens | JMH timing tokens |
-| --- | --- | --- | ---: | ---: | ---: | ---: | ---: |
-| `he-il-default` | `1.0.0` | `HE_IL` | 2,358 | 61,071 | 4,715 | 56,356 | 56,356 |
+| Model ID | Model version | Language | Dictionary rows | Distinct usable forms | Complete quality tokens | Already-root tokens | Changed tokens | Timing workload | JMH timing tokens |
+| --- | --- | --- | ---: | ---: | ---: | ---: | ---: | --- | ---: |
+| `he-il-default` | `1.0.0` | `HE_IL` | 2,358 | 57,658 | 61,071 | 4,715 | 56,356 | changed tokens | 56,356 |
 
 ## Radixor Patch Command Distribution
 
@@ -39,11 +43,11 @@ Accuracy is computed from JMH auxiliary counters in the current report. The coun
 
 ## Speed
 
-Speed uses JMH average time, 5 warmup iterations, 7 measurement iterations, 3 independent forks, and 1 thread.
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 3 independent forks, and 1 thread.
 
 | Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
 | --- | --- | ---: | ---: | ---: | ---: | --- |
-| Radixor | `hebrewRadixor` | 4.966 | 0.167 | 88.1 | 1.000 | Full default-model Radixor dictionary patch-command stemmer. |
+| Radixor | `hebrewRadixor` | 5.073 | 0.131 | 90.0 | 1.000 | Full default-model Radixor dictionary patch-command stemmer. |
 
 ## Interpretation Notes
 
@@ -80,18 +84,15 @@ also appeared in training. Parentheses show the observed split minimum–maximum
 
 ### Generalization conclusion
 
-- Median exactness on genuinely unseen changed forms moves from **10.423%**
-  at 10% training knowledge to **16.362%** at 90%, a measured
-  **+5.939 percentage-point** change for this dictionary.
-- Over the same endpoints, unseen all-form exactness changes by **+5.281 pp** and
-  preservation of unseen already-root forms changes by **-0.971 pp**. These separate
-  outcomes show whether the changed-form result coexists with preservation behavior.
+- Median exactness on genuinely unseen changed forms moves from **10.423%** at 10% training knowledge to **16.362%** at 90%, a measured **+5.939 percentage-point** change.
+- Unseen all-form exactness moves from **16.532%** at 10% training knowledge to **21.813%** at 90%, a measured **+5.281 percentage-point** change.
+- Preservation of unseen already-root forms moves from **89.756%** at 10% training knowledge to **88.785%** at 90%, a measured **-0.971 percentage-point** change.
 - The evidence establishes within-resource transfer across withheld dictionary families. It
   does not estimate unrelated domains, misspellings, arbitrary compounds, or external corpora.
 
 The complete ten-level table and split ranges remain in the
 [independent generalization report](../generalization.md); raw counters and provenance are in
-[`dictionary-generalization.csv`](../data/dictionary-generalization.csv). The
+[active machine-readable snapshot](../data/dictionary-generalization-2026-09-11.csv). The
 [frozen methodology](../reference/generalization-methodology.md) defines family-level
 splitting, unseen-surface leakage control, aggregation, and the limits of the claim.
 
@@ -184,7 +185,7 @@ The complete evidence is available in the [raw logical matrix](../data/edit-cost
 
 Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `HE_IL` using the complete validated stemming-quality result matrix. Every distinct surface form is one evaluated item and can belong to several dictionary groups. Two forms are a positive pair when their group-membership sets intersect and a negative pair when those sets are disjoint. A pair shared through several groups is counted once. Exact equality with a predetermined lemma is not required.
 
-`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality-2026-09-11.csv).
 
 ### Evaluation Scope and Key Findings
 
@@ -415,8 +416,8 @@ Standard ARI, homogeneity, completeness, V-measure, and NMI are not calculated:
 
 ### Provenance
 
-- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
-- Source SHA-256: `85763189eab4d0fbb047c2d5d3554c66abf9732182bd0d8fd758d7aef680e66f`
+- Authoritative source: `docs/benchmarks/data/stemming-quality-2026-09-11.csv`
+- Source SHA-256: `24bddfeed06a60bb3eeed58e1bfc93aed46c1d32e7bebd293aec2dacfddfef5b`
 - Evaluation command: `./gradlew stemmingQuality --no-daemon`
 - Dictionary language: `HE_IL`
 - Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
diff --git a/docs/benchmarks/languages/hungarian.md b/docs/benchmarks/languages/hungarian.md
index c1b86ce..06ee7b5 100644
--- a/docs/benchmarks/languages/hungarian.md
+++ b/docs/benchmarks/languages/hungarian.md
@@ -1,8 +1,12 @@
-# Hungarian Stemmer Benchmarks
+# Hungarian Stemmer Benchmarks <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 910,688 distinct usable word forms" title="Relative dictionary size 5 of 5; 910,688 distinct usable word forms">★★★★★</span>
 
 This page reports same-language stemming benchmarks for Hungarian. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
 
-All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-08-25 Radixor/Java `4.2.0-6-g84e57fb` snapshot. Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+<!-- DICTIONARY-SIZE-RATING:START -->
+Dictionary size: <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 910,688 distinct usable word forms" title="Relative dictionary size 5 of 5; 910,688 distinct usable word forms">★★★★★</span>. The exact count is **910,688 distinct usable word forms** after parser-compatible filtering and exact, case-preserved deduplication. Stars rank dictionary size relative to all 167 user-facing models in five nearly equal groups; they do **not** measure linguistic quality or benchmark accuracy.
+<!-- DICTIONARY-SIZE-RATING:END -->
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-09-11 Radixor/Java `4.3.0.dirty` snapshot. Speed benchmark operations process changed tokens. Accuracy uses the complete Radixor dictionary for the language.
 
 <!-- BENCHMARK-EVIDENCE-MAP:START -->
 !!! info "How to read this page"
@@ -13,9 +17,9 @@ Runtime and exact-root agreement measure different properties. Light, minimal, p
 
 ## Dictionary Corpus
 
-| Model ID | Model version | Language | Dictionary rows | Complete quality tokens | Already-root tokens | Changed tokens | JMH timing tokens |
-| --- | --- | --- | ---: | ---: | ---: | ---: | ---: |
-| `hu-hu-default` | `1.0.0` | `HU_HU` | 19,406 | 935,713 | 38,775 | 896,938 | 896,938 |
+| Model ID | Model version | Language | Dictionary rows | Distinct usable forms | Complete quality tokens | Already-root tokens | Changed tokens | Timing workload | JMH timing tokens |
+| --- | --- | --- | ---: | ---: | ---: | ---: | ---: | --- | ---: |
+| `hu-hu-default` | `1.0.0` | `HU_HU` | 19,406 | 910,688 | 935,713 | 38,775 | 896,938 | changed tokens | 896,938 |
 
 ## Radixor Patch Command Distribution
 
@@ -42,14 +46,14 @@ Accuracy is computed from JMH auxiliary counters in the current report. The coun
 
 ## Speed
 
-Speed uses JMH average time, 5 warmup iterations, 7 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
 
 | Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
 | --- | --- | ---: | ---: | ---: | ---: | --- |
-| Radixor | `hungarianRadixor` | 55.040 | 0.877 | 61.4 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
-| Lucene HungarianLightStemFilter | `hungarianLuceneHungarianLightStemFilter` | 88.848 | 3.854 | 99.1 | 1.614 | Light Hungarian suffix stemmer. |
-| Official Snowball direct | `snowballDirect[HUNGARIAN]` | 167.949 | 9.811 | 187.2 | 3.051 | Official Snowball generated Java stemmer; direct API. |
-| Lucene SnowballFilter | `luceneSnowballFilter[HUNGARIAN]` | 186.896 | 11.344 | 208.4 | 3.396 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
+| Radixor | `hungarianRadixor` | 57.672 | 1.428 | 64.3 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
+| Lucene HungarianLightStemFilter | `hungarianLuceneHungarianLightStemFilter` | 92.463 | 5.822 | 103.1 | 1.603 | Light Hungarian suffix stemmer. |
+| Official Snowball direct | `snowballDirect[HUNGARIAN]` | 160.725 | 14.076 | 179.2 | 2.787 | Official Snowball generated Java stemmer; direct API. |
+| Lucene SnowballFilter | `luceneSnowballFilter[HUNGARIAN]` | 195.427 | 22.425 | 217.9 | 3.389 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
 
 ## Interpretation Notes
 
@@ -87,18 +91,15 @@ also appeared in training. Parentheses show the observed split minimum–maximum
 
 ### Generalization conclusion
 
-- Median exactness on genuinely unseen changed forms moves from **86.876%**
-  at 10% training knowledge to **92.203%** at 90%, a measured
-  **+5.327 percentage-point** change for this dictionary.
-- Over the same endpoints, unseen all-form exactness changes by **+5.450 pp** and
-  preservation of unseen already-root forms changes by **+9.360 pp**. These separate
-  outcomes show whether the changed-form result coexists with preservation behavior.
+- Median exactness on genuinely unseen changed forms moves from **86.876%** at 10% training knowledge to **92.203%** at 90%, a measured **+5.327 percentage-point** change.
+- Unseen all-form exactness moves from **86.761%** at 10% training knowledge to **92.212%** at 90%, a measured **+5.450 percentage-point** change.
+- Preservation of unseen already-root forms moves from **84.086%** at 10% training knowledge to **93.446%** at 90%, a measured **+9.360 percentage-point** change.
 - The evidence establishes within-resource transfer across withheld dictionary families. It
   does not estimate unrelated domains, misspellings, arbitrary compounds, or external corpora.
 
 The complete ten-level table and split ranges remain in the
 [independent generalization report](../generalization.md); raw counters and provenance are in
-[`dictionary-generalization.csv`](../data/dictionary-generalization.csv). The
+[active machine-readable snapshot](../data/dictionary-generalization-2026-09-11.csv). The
 [frozen methodology](../reference/generalization-methodology.md) defines family-level
 splitting, unseen-surface leakage control, aggregation, and the limits of the claim.
 
@@ -191,7 +192,7 @@ The complete evidence is available in the [raw logical matrix](../data/edit-cost
 
 Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `HU_HU` using the complete validated stemming-quality result matrix. Every distinct surface form is one evaluated item and can belong to several dictionary groups. Two forms are a positive pair when their group-membership sets intersect and a negative pair when those sets are disjoint. A pair shared through several groups is counted once. Exact equality with a predetermined lemma is not required.
 
-`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality-2026-09-11.csv).
 
 ### Evaluation Scope and Key Findings
 
@@ -446,8 +447,8 @@ Standard ARI, homogeneity, completeness, V-measure, and NMI are not calculated:
 
 ### Provenance
 
-- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
-- Source SHA-256: `85763189eab4d0fbb047c2d5d3554c66abf9732182bd0d8fd758d7aef680e66f`
+- Authoritative source: `docs/benchmarks/data/stemming-quality-2026-09-11.csv`
+- Source SHA-256: `24bddfeed06a60bb3eeed58e1bfc93aed46c1d32e7bebd293aec2dacfddfef5b`
 - Evaluation command: `./gradlew stemmingQuality --no-daemon`
 - Dictionary language: `HU_HU`
 - Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
diff --git a/docs/benchmarks/languages/index.md b/docs/benchmarks/languages/index.md
index eae21fc..8aecf02 100644
--- a/docs/benchmarks/languages/index.md
+++ b/docs/benchmarks/languages/index.md
@@ -1,8 +1,8 @@
 # Language Benchmark Pages
 
-This section splits Radixor stemmer benchmark results by language. Each of the 20 registered default models has one language page containing the refreshed corpus, patch-command distribution, exact-root accuracy, runtime performance, pairwise stemming-quality tables, and a language-specific interpretation of the edit-cost and dictionary-knowledge experiment.
+This section covers all 143 registered language defaults. The original 20 pages retain their complete measured and experimental sections; newly added pages publish corpus, accuracy, and speed evidence from the current checked-in canonical reports.
 
-The experimental sections do not merely repeat the cross-language macro values. They publish each language's five-split 10%–90% knowledge curve, exact command-equivalence census, selected edit costs, command-count effect, within-language factor associations, evidence coverage, and bounded conclusion. Non-baseline settings are exploratory candidates until independently validated.
+The original 20 measured pages include experimental sections that do not merely repeat the cross-language macro values. They publish each language's five-split 10%–90% knowledge curve, exact command-equivalence census, selected edit costs, command-count effect, within-language factor associations, evidence coverage, and bounded conclusion. Non-baseline settings are exploratory candidates until independently validated. Equivalent sections for newly registered models remain reserved for later deterministic publication.
 
 !!! info "Two knowledge curves, two questions"
     The dictionary-family curve tests transfer to withheld lexical families under the fixed
@@ -26,28 +26,151 @@ The experimental sections do not merely repeat the cross-language macro values.
 
 ## Languages
 
-| Language | Resource | Benchmark page |
-| --- | --- | --- |
-| Czech | `CS_CZ` | [Czech](czech.md) |
-| Danish | `DA_DK` | [Danish](danish.md) |
-| Dutch | `NL_NL` | [Dutch](dutch.md) |
-| English | `US_UK` | [English](english.md) |
-| Finnish | `FI_FI` | [Finnish](finnish.md) |
-| French | `FR_FR` | [French](french.md) |
-| German | `DE_DE` | [German](german.md) |
-| Hebrew | `HE_IL` | [Hebrew](hebrew.md) |
-| Hungarian | `HU_HU` | [Hungarian](hungarian.md) |
-| Italian | `IT_IT` | [Italian](italian.md) |
-| Norwegian Bokmal | `NB_NO` | [Norwegian Bokmal](norwegian-bokmal.md) |
-| Norwegian Nynorsk | `NN_NO` | [Norwegian Nynorsk](norwegian-nynorsk.md) |
-| Persian | `FA_IR` | [Persian](persian.md) |
-| Polish | `PL_PL` | [Polish](polish.md) |
-| Portuguese | `PT_PT` | [Portuguese](portuguese.md) |
-| Russian | `RU_RU` | [Russian](russian.md) |
-| Spanish | `ES_ES` | [Spanish](spanish.md) |
-| Swedish | `SV_SE` | [Swedish](swedish.md) |
-| Ukrainian | `UK_UA` | [Ukrainian](ukrainian.md) |
-| Yiddish | `YI` | [Yiddish](yiddish.md) |
+| Language | Java enum | Model ID | Relative dictionary size | Availability | Benchmark page |
+| --- | --- | --- | --- | --- | --- |
+| Adyghe | `ADY` | `ady-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 20,334 distinct usable word forms" title="Relative dictionary size 4 of 5; 20,334 distinct usable word forms">★★★★☆</span> (20,334) | Individual artifact; outside standard aggregate | [Adyghe](ady.md) |
+| Afrikaans | `AF_ZA` | `af-za-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 289,303 distinct usable word forms" title="Relative dictionary size 5 of 5; 289,303 distinct usable word forms">★★★★★</span> (289,303) | Individual artifact; outside standard aggregate | [Afrikaans](af-za.md) |
+| Aimele | `AIL` | `ail-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 2,978 distinct usable word forms" title="Relative dictionary size 3 of 5; 2,978 distinct usable word forms">★★★☆☆</span> (2,978) | Individual artifact; outside standard aggregate | [Aimele](ail.md) |
+| Akan | `AK` | `aka-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 1,890 distinct usable word forms" title="Relative dictionary size 2 of 5; 1,890 distinct usable word forms">★★☆☆☆</span> (1,890) | Individual artifact; outside standard aggregate | [Akan](aka.md) |
+| Albanian | `SQ_AL` | `sq-al-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 10,218 distinct usable word forms" title="Relative dictionary size 3 of 5; 10,218 distinct usable word forms">★★★☆☆</span> (10,218) | Individual artifact; outside standard aggregate | [Albanian](sq-al.md) |
+| Alsatian | `GSW` | `gsw-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 1,171 distinct usable word forms" title="Relative dictionary size 2 of 5; 1,171 distinct usable word forms">★★☆☆☆</span> (1,171) | Individual artifact; outside standard aggregate | [Alsatian](gsw.md) |
+| Amharic | `AM_ET` | `am-et-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 41,308 distinct usable word forms" title="Relative dictionary size 4 of 5; 41,308 distinct usable word forms">★★★★☆</span> (41,308) | Individual artifact; outside standard aggregate | [Amharic](am-et.md) |
+| Amol Azerbaijani | `AZG` | `azg-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 9,357 distinct usable word forms" title="Relative dictionary size 3 of 5; 9,357 distinct usable word forms">★★★☆☆</span> (9,357) | Individual artifact; outside standard aggregate | [Amol Azerbaijani](azg.md) |
+| Ancient Greek | `GRC` | `grc-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 11,109 distinct usable word forms" title="Relative dictionary size 4 of 5; 11,109 distinct usable word forms">★★★★☆</span> (11,109) | Individual artifact; outside standard aggregate | [Ancient Greek](grc.md) |
+| Anglo-Norman | `XNO` | `xno-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 186 distinct usable word forms" title="Relative dictionary size 1 of 5; 186 distinct usable word forms">★☆☆☆☆</span> (186) | Individual artifact; outside standard aggregate | [Anglo-Norman](xno.md) |
+| Arabic | `AR` | `ar-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 452,974 distinct usable word forms" title="Relative dictionary size 5 of 5; 452,974 distinct usable word forms">★★★★★</span> (452,974) | Individual artifact; outside standard aggregate | [Arabic](ar.md) |
+| Armenian | `HY_AM` | `hy-am-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 242,091 distinct usable word forms" title="Relative dictionary size 5 of 5; 242,091 distinct usable word forms">★★★★★</span> (242,091) | Individual artifact; outside standard aggregate | [Armenian](hy-am.md) |
+| Ashaninka | `CNI` | `cni-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 10,095 distinct usable word forms" title="Relative dictionary size 3 of 5; 10,095 distinct usable word forms">★★★☆☆</span> (10,095) | Individual artifact; outside standard aggregate | [Ashaninka](cni.md) |
+| Assamese | `AS_IN` | `as-in-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 328 distinct usable word forms" title="Relative dictionary size 1 of 5; 328 distinct usable word forms">★☆☆☆☆</span> (328) | Individual artifact; outside standard aggregate | [Assamese](as-in.md) |
+| Asturian | `AST` | `ast-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 21,584 distinct usable word forms" title="Relative dictionary size 4 of 5; 21,584 distinct usable word forms">★★★★☆</span> (21,584) | Individual artifact; outside standard aggregate | [Asturian](ast.md) |
+| Aymara | `AYM` | `aym-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 231,247 distinct usable word forms" title="Relative dictionary size 5 of 5; 231,247 distinct usable word forms">★★★★★</span> (231,247) | Individual artifact; outside standard aggregate | [Aymara](aym.md) |
+| Azerbaijani | `AZ_AZ` | `az-az-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 6,669 distinct usable word forms" title="Relative dictionary size 3 of 5; 6,669 distinct usable word forms">★★★☆☆</span> (6,669) | Individual artifact; outside standard aggregate | [Azerbaijani](az-az.md) |
+| Bashkir | `BAK` | `bak-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 10,616 distinct usable word forms" title="Relative dictionary size 4 of 5; 10,616 distinct usable word forms">★★★★☆</span> (10,616) | Individual artifact; outside standard aggregate | [Bashkir](bak.md) |
+| Belarusian | `BE_BY` | `be-by-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 678 distinct usable word forms" title="Relative dictionary size 2 of 5; 678 distinct usable word forms">★★☆☆☆</span> (678) | Individual artifact; outside standard aggregate | [Belarusian](be-by.md) |
+| Bengali | `BN_BD` | `bn-bd-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 3 distinct usable word forms" title="Relative dictionary size 1 of 5; 3 distinct usable word forms">★☆☆☆☆</span> (3) | Individual artifact; outside standard aggregate | [Bengali](bn-bd.md) |
+| Bininj Kun-wok | `GUP` | `gup-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 370 distinct usable word forms" title="Relative dictionary size 1 of 5; 370 distinct usable word forms">★☆☆☆☆</span> (370) | Individual artifact; outside standard aggregate | [Bininj Kun-wok](gup.md) |
+| Braj | `BRA` | `bra-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 115 distinct usable word forms" title="Relative dictionary size 1 of 5; 115 distinct usable word forms">★☆☆☆☆</span> (115) | Individual artifact; outside standard aggregate | [Braj](bra.md) |
+| Breton | `BRE` | `bre-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 1,682 distinct usable word forms" title="Relative dictionary size 2 of 5; 1,682 distinct usable word forms">★★☆☆☆</span> (1,682) | Individual artifact; outside standard aggregate | [Breton](bre.md) |
+| Bulgarian | `BG_BG` | `bg-bg-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 37,335 distinct usable word forms" title="Relative dictionary size 4 of 5; 37,335 distinct usable word forms">★★★★☆</span> (37,335) | Individual artifact; outside standard aggregate | [Bulgarian](bg-bg.md) |
+| Catalan | `CA_ES` | `ca-es-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 61,495 distinct usable word forms" title="Relative dictionary size 5 of 5; 61,495 distinct usable word forms">★★★★★</span> (61,495) | Individual artifact; outside standard aggregate | [Catalan](ca-es.md) |
+| Cebuano | `CEB` | `ceb-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 392 distinct usable word forms" title="Relative dictionary size 1 of 5; 392 distinct usable word forms">★☆☆☆☆</span> (392) | Individual artifact; outside standard aggregate | [Cebuano](ceb.md) |
+| Chichewa | `NY_MW` | `ny-mw-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 3,150 distinct usable word forms" title="Relative dictionary size 3 of 5; 3,150 distinct usable word forms">★★★☆☆</span> (3,150) | Individual artifact; outside standard aggregate | [Chichewa](ny-mw.md) |
+| Chukchi | `CKT` | `ckt-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 302 distinct usable word forms" title="Relative dictionary size 1 of 5; 302 distinct usable word forms">★☆☆☆☆</span> (302) | Individual artifact; outside standard aggregate | [Chukchi](ckt.md) |
+| Church Slavonic | `CHU` | `chu-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 1,632 distinct usable word forms" title="Relative dictionary size 2 of 5; 1,632 distinct usable word forms">★★☆☆☆</span> (1,632) | Individual artifact; outside standard aggregate | [Church Slavonic](chu.md) |
+| Classical Armenian | `XCL` | `xcl-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 55,946 distinct usable word forms" title="Relative dictionary size 4 of 5; 55,946 distinct usable word forms">★★★★☆</span> (55,946) | Individual artifact; outside standard aggregate | [Classical Armenian](xcl.md) |
+| Classical Syriac | `SYC` | `syc-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 2,139 distinct usable word forms" title="Relative dictionary size 3 of 5; 2,139 distinct usable word forms">★★★☆☆</span> (2,139) | Individual artifact; outside standard aggregate | [Classical Syriac](syc.md) |
+| Congo Swahili | `SWC` | `swc-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 6,700 distinct usable word forms" title="Relative dictionary size 3 of 5; 6,700 distinct usable word forms">★★★☆☆</span> (6,700) | Individual artifact; outside standard aggregate | [Congo Swahili](swc.md) |
+| Copala Triqui | `CPA` | `cpa-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 1,751 distinct usable word forms" title="Relative dictionary size 2 of 5; 1,751 distinct usable word forms">★★☆☆☆</span> (1,751) | Individual artifact; outside standard aggregate | [Copala Triqui](cpa.md) |
+| Cornish | `COR` | `cor-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 159 distinct usable word forms" title="Relative dictionary size 1 of 5; 159 distinct usable word forms">★☆☆☆☆</span> (159) | Individual artifact; outside standard aggregate | [Cornish](cor.md) |
+| Cree | `CRE` | `cre-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 514 distinct usable word forms" title="Relative dictionary size 2 of 5; 514 distinct usable word forms">★★☆☆☆</span> (514) | Individual artifact; outside standard aggregate | [Cree](cre.md) |
+| Crimean Tatar | `CRH` | `crh-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 7,183 distinct usable word forms" title="Relative dictionary size 3 of 5; 7,183 distinct usable word forms">★★★☆☆</span> (7,183) | Individual artifact; outside standard aggregate | [Crimean Tatar](crh.md) |
+| Czech | `CS_CZ` | `cs-cz-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 51,401 distinct usable word forms" title="Relative dictionary size 4 of 5; 51,401 distinct usable word forms">★★★★☆</span> (51,401) | Standard aggregate | [Czech](czech.md) |
+| Dakota | `DAK` | `dak-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 2,942 distinct usable word forms" title="Relative dictionary size 3 of 5; 2,942 distinct usable word forms">★★★☆☆</span> (2,942) | Individual artifact; outside standard aggregate | [Dakota](dak.md) |
+| Danish | `DA_DK` | `da-dk-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 27,921 distinct usable word forms" title="Relative dictionary size 4 of 5; 27,921 distinct usable word forms">★★★★☆</span> (27,921) | Standard aggregate | [Danish](danish.md) |
+| Dutch | `NL_NL` | `nl-nl-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 26,201 distinct usable word forms" title="Relative dictionary size 4 of 5; 26,201 distinct usable word forms">★★★★☆</span> (26,201) | Standard aggregate | [Dutch](dutch.md) |
+| Eastern Chatino | `CLY` | `cly-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 2,030 distinct usable word forms" title="Relative dictionary size 2 of 5; 2,030 distinct usable word forms">★★☆☆☆</span> (2,030) | Individual artifact; outside standard aggregate | [Eastern Chatino](cly.md) |
+| Egyptian Arabic | `ARZ` | `arz-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 135 distinct usable word forms" title="Relative dictionary size 1 of 5; 135 distinct usable word forms">★☆☆☆☆</span> (135) | Individual artifact; outside standard aggregate | [Egyptian Arabic](arz.md) |
+| English | `US_UK` | `us-uk-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 591,946 distinct usable word forms" title="Relative dictionary size 5 of 5; 591,946 distinct usable word forms">★★★★★</span> (591,946) | Standard aggregate | [English](english.md) |
+| Estonian | `ET_EE` | `et-ee-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 24,811 distinct usable word forms" title="Relative dictionary size 4 of 5; 24,811 distinct usable word forms">★★★★☆</span> (24,811) | Individual artifact; outside standard aggregate | [Estonian](et-ee.md) |
+| Evenki | `EVN` | `evn-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 11,965 distinct usable word forms" title="Relative dictionary size 4 of 5; 11,965 distinct usable word forms">★★★★☆</span> (11,965) | Individual artifact; outside standard aggregate | [Evenki](evn.md) |
+| Faroese | `FO_FO` | `fo-fo-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 31,366 distinct usable word forms" title="Relative dictionary size 4 of 5; 31,366 distinct usable word forms">★★★★☆</span> (31,366) | Individual artifact; outside standard aggregate | [Faroese](fo-fo.md) |
+| Finnish | `FI_FI` | `fi-fi-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 1,788,784 distinct usable word forms" title="Relative dictionary size 5 of 5; 1,788,784 distinct usable word forms">★★★★★</span> (1,788,784) | Standard aggregate | [Finnish](finnish.md) |
+| French | `FR_FR` | `fr-fr-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 404,011 distinct usable word forms" title="Relative dictionary size 5 of 5; 404,011 distinct usable word forms">★★★★★</span> (404,011) | Standard aggregate | [French](french.md) |
+| Friulian | `FUR` | `fur-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 5,007 distinct usable word forms" title="Relative dictionary size 3 of 5; 5,007 distinct usable word forms">★★★☆☆</span> (5,007) | Individual artifact; outside standard aggregate | [Friulian](fur.md) |
+| Ga | `GAA` | `gaa-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 469 distinct usable word forms" title="Relative dictionary size 1 of 5; 469 distinct usable word forms">★☆☆☆☆</span> (469) | Individual artifact; outside standard aggregate | [Ga](gaa.md) |
+| Galolen | `GAL` | `gal-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 25,436 distinct usable word forms" title="Relative dictionary size 4 of 5; 25,436 distinct usable word forms">★★★★☆</span> (25,436) | Individual artifact; outside standard aggregate | [Galolen](gal.md) |
+| German | `DE_DE` | `de-de-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 277,266 distinct usable word forms" title="Relative dictionary size 5 of 5; 277,266 distinct usable word forms">★★★★★</span> (277,266) | Standard aggregate | [German](german.md) |
+| Gothic | `GOT` | `got-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 134,162 distinct usable word forms" title="Relative dictionary size 5 of 5; 134,162 distinct usable word forms">★★★★★</span> (134,162) | Individual artifact; outside standard aggregate | [Gothic](got.md) |
+| Greek | `EL_GR` | `el-gr-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 76,851 distinct usable word forms" title="Relative dictionary size 5 of 5; 76,851 distinct usable word forms">★★★★★</span> (76,851) | Individual artifact; outside standard aggregate | [Greek](el-gr.md) |
+| Gulf Arabic | `AFB` | `afb-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 355 distinct usable word forms" title="Relative dictionary size 1 of 5; 355 distinct usable word forms">★☆☆☆☆</span> (355) | Individual artifact; outside standard aggregate | [Gulf Arabic](afb.md) |
+| Haida | `HAI` | `hai-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 520 distinct usable word forms" title="Relative dictionary size 2 of 5; 520 distinct usable word forms">★★☆☆☆</span> (520) | Individual artifact; outside standard aggregate | [Haida](hai.md) |
+| Hebrew | `HE_IL` | `he-il-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 57,658 distinct usable word forms" title="Relative dictionary size 4 of 5; 57,658 distinct usable word forms">★★★★☆</span> (57,658) | Standard aggregate | [Hebrew](hebrew.md) |
+| Hiligaynon | `HIL` | `hil-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 308 distinct usable word forms" title="Relative dictionary size 1 of 5; 308 distinct usable word forms">★☆☆☆☆</span> (308) | Individual artifact; outside standard aggregate | [Hiligaynon](hil.md) |
+| Hsilimo | `HSI` | `hsi-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 158 distinct usable word forms" title="Relative dictionary size 1 of 5; 158 distinct usable word forms">★☆☆☆☆</span> (158) | Individual artifact; outside standard aggregate | [Hsilimo](hsi.md) |
+| Hungarian | `HU_HU` | `hu-hu-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 910,688 distinct usable word forms" title="Relative dictionary size 5 of 5; 910,688 distinct usable word forms">★★★★★</span> (910,688) | Standard aggregate | [Hungarian](hungarian.md) |
+| Icelandic | `IS_IS` | `is-is-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 52,132 distinct usable word forms" title="Relative dictionary size 4 of 5; 52,132 distinct usable word forms">★★★★☆</span> (52,132) | Individual artifact; outside standard aggregate | [Icelandic](is-is.md) |
+| Indonesian | `ID_ID` | `id-id-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 21,296 distinct usable word forms" title="Relative dictionary size 4 of 5; 21,296 distinct usable word forms">★★★★☆</span> (21,296) | Individual artifact; outside standard aggregate | [Indonesian](id-id.md) |
+| Ingrian | `IZH` | `izh-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 1,024 distinct usable word forms" title="Relative dictionary size 2 of 5; 1,024 distinct usable word forms">★★☆☆☆</span> (1,024) | Individual artifact; outside standard aggregate | [Ingrian](izh.md) |
+| Irish | `GA_IE` | `ga-ie-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 21,829 distinct usable word forms" title="Relative dictionary size 4 of 5; 21,829 distinct usable word forms">★★★★☆</span> (21,829) | Individual artifact; outside standard aggregate | [Irish](ga-ie.md) |
+| Italian | `IT_IT` | `it-it-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 324,366 distinct usable word forms" title="Relative dictionary size 5 of 5; 324,366 distinct usable word forms">★★★★★</span> (324,366) | Standard aggregate | [Italian](italian.md) |
+| Itelmen | `ITL` | `itl-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 3,546 distinct usable word forms" title="Relative dictionary size 3 of 5; 3,546 distinct usable word forms">★★★☆☆</span> (3,546) | Individual artifact; outside standard aggregate | [Itelmen](itl.md) |
+| Japanese | `JA_JP` | `ja-jp-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 10,848 distinct usable word forms" title="Relative dictionary size 4 of 5; 10,848 distinct usable word forms">★★★★☆</span> (10,848) | Individual artifact; outside standard aggregate | [Japanese](ja-jp.md) |
+| Kabardian | `KBD` | `kbd-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 3,054 distinct usable word forms" title="Relative dictionary size 3 of 5; 3,054 distinct usable word forms">★★★☆☆</span> (3,054) | Individual artifact; outside standard aggregate | [Kabardian](kbd.md) |
+| Kalaallisut | `KL_GL` | `kl-gl-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 321 distinct usable word forms" title="Relative dictionary size 1 of 5; 321 distinct usable word forms">★☆☆☆☆</span> (321) | Individual artifact; outside standard aggregate | [Kalaallisut](kl-gl.md) |
+| Kannada | `KN_IN` | `kn-in-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 11 distinct usable word forms" title="Relative dictionary size 1 of 5; 11 distinct usable word forms">★☆☆☆☆</span> (11) | Individual artifact; outside standard aggregate | [Kannada](kn-in.md) |
+| Karelian | `KRL` | `krl-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 566 distinct usable word forms" title="Relative dictionary size 2 of 5; 566 distinct usable word forms">★★☆☆☆</span> (566) | Individual artifact; outside standard aggregate | [Karelian](krl.md) |
+| Kashubian | `CSB` | `csb-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 350 distinct usable word forms" title="Relative dictionary size 1 of 5; 350 distinct usable word forms">★☆☆☆☆</span> (350) | Individual artifact; outside standard aggregate | [Kashubian](csb.md) |
+| Kazakh | `KK_KZ` | `kk-kz-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 357 distinct usable word forms" title="Relative dictionary size 1 of 5; 357 distinct usable word forms">★☆☆☆☆</span> (357) | Individual artifact; outside standard aggregate | [Kazakh](kk-kz.md) |
+| Khakas | `KJH` | `kjh-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 1,172 distinct usable word forms" title="Relative dictionary size 2 of 5; 1,172 distinct usable word forms">★★☆☆☆</span> (1,172) | Individual artifact; outside standard aggregate | [Khakas](kjh.md) |
+| Khinalugh | `KLR` | `klr-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 41,185 distinct usable word forms" title="Relative dictionary size 4 of 5; 41,185 distinct usable word forms">★★★★☆</span> (41,185) | Individual artifact; outside standard aggregate | [Khinalugh](klr.md) |
+| Kodi | `KOD` | `kod-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 524 distinct usable word forms" title="Relative dictionary size 2 of 5; 524 distinct usable word forms">★★☆☆☆</span> (524) | Individual artifact; outside standard aggregate | [Kodi](kod.md) |
+| Kongo | `KON` | `kon-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 585 distinct usable word forms" title="Relative dictionary size 2 of 5; 585 distinct usable word forms">★★☆☆☆</span> (585) | Individual artifact; outside standard aggregate | [Kongo](kon.md) |
+| Kyrgyz | `KY_KG` | `ky-kg-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 2,997 distinct usable word forms" title="Relative dictionary size 3 of 5; 2,997 distinct usable word forms">★★★☆☆</span> (2,997) | Individual artifact; outside standard aggregate | [Kyrgyz](ky-kg.md) |
+| Ladin | `LLD` | `lld-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 4,819 distinct usable word forms" title="Relative dictionary size 3 of 5; 4,819 distinct usable word forms">★★★☆☆</span> (4,819) | Individual artifact; outside standard aggregate | [Ladin](lld.md) |
+| Latin | `LA` | `la-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 485,259 distinct usable word forms" title="Relative dictionary size 5 of 5; 485,259 distinct usable word forms">★★★★★</span> (485,259) | Individual artifact; outside standard aggregate | [Latin](la.md) |
+| Latvian | `LV_LV` | `lv-lv-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 75,492 distinct usable word forms" title="Relative dictionary size 5 of 5; 75,492 distinct usable word forms">★★★★★</span> (75,492) | Individual artifact; outside standard aggregate | [Latvian](lv-lv.md) |
+| Lingala | `LIN` | `lin-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 230 distinct usable word forms" title="Relative dictionary size 1 of 5; 230 distinct usable word forms">★☆☆☆☆</span> (230) | Individual artifact; outside standard aggregate | [Lingala](lin.md) |
+| Lithuanian | `LT_LT` | `lt-lt-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 27,335 distinct usable word forms" title="Relative dictionary size 4 of 5; 27,335 distinct usable word forms">★★★★☆</span> (27,335) | Individual artifact; outside standard aggregate | [Lithuanian](lt-lt.md) |
+| Livonian | `LIV` | `liv-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 2,841 distinct usable word forms" title="Relative dictionary size 3 of 5; 2,841 distinct usable word forms">★★★☆☆</span> (2,841) | Individual artifact; outside standard aggregate | [Livonian](liv.md) |
+| Low German | `NDS` | `nds-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 2,493 distinct usable word forms" title="Relative dictionary size 3 of 5; 2,493 distinct usable word forms">★★★☆☆</span> (2,493) | Individual artifact; outside standard aggregate | [Low German](nds.md) |
+| Lower Sorbian | `DSB` | `dsb-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 12,038 distinct usable word forms" title="Relative dictionary size 4 of 5; 12,038 distinct usable word forms">★★★★☆</span> (12,038) | Individual artifact; outside standard aggregate | [Lower Sorbian](dsb.md) |
+| Luganda | `LG_UG` | `lg-ug-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 4,671 distinct usable word forms" title="Relative dictionary size 3 of 5; 4,671 distinct usable word forms">★★★☆☆</span> (4,671) | Individual artifact; outside standard aggregate | [Luganda](lg-ug.md) |
+| Macedonian | `MK_MK` | `mk-mk-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 135,590 distinct usable word forms" title="Relative dictionary size 5 of 5; 135,590 distinct usable word forms">★★★★★</span> (135,590) | Individual artifact; outside standard aggregate | [Macedonian](mk-mk.md) |
+| Magahi | `MAG` | `mag-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 206 distinct usable word forms" title="Relative dictionary size 1 of 5; 206 distinct usable word forms">★☆☆☆☆</span> (206) | Individual artifact; outside standard aggregate | [Magahi](mag.md) |
+| Malagasy | `MG_MG` | `mg-mg-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 636 distinct usable word forms" title="Relative dictionary size 2 of 5; 636 distinct usable word forms">★★☆☆☆</span> (636) | Individual artifact; outside standard aggregate | [Malagasy](mg-mg.md) |
+| Maltese | `MT_MT` | `mt-mt-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 1,421 distinct usable word forms" title="Relative dictionary size 2 of 5; 1,421 distinct usable word forms">★★☆☆☆</span> (1,421) | Individual artifact; outside standard aggregate | [Maltese](mt-mt.md) |
+| Manx | `GV_IM` | `gv-im-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 15 distinct usable word forms" title="Relative dictionary size 1 of 5; 15 distinct usable word forms">★☆☆☆☆</span> (15) | Individual artifact; outside standard aggregate | [Manx](gv-im.md) |
+| Maori | `MI_NZ` | `mi-nz-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 207 distinct usable word forms" title="Relative dictionary size 1 of 5; 207 distinct usable word forms">★☆☆☆☆</span> (207) | Individual artifact; outside standard aggregate | [Maori](mi-nz.md) |
+| Mapudungun | `ARN` | `arn-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 548 distinct usable word forms" title="Relative dictionary size 2 of 5; 548 distinct usable word forms">★★☆☆☆</span> (548) | Individual artifact; outside standard aggregate | [Mapudungun](arn.md) |
+| Middle French | `FRM` | `frm-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 27,002 distinct usable word forms" title="Relative dictionary size 4 of 5; 27,002 distinct usable word forms">★★★★☆</span> (27,002) | Individual artifact; outside standard aggregate | [Middle French](frm.md) |
+| Middle High German | `GMH` | `gmh-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 384 distinct usable word forms" title="Relative dictionary size 1 of 5; 384 distinct usable word forms">★☆☆☆☆</span> (384) | Individual artifact; outside standard aggregate | [Middle High German](gmh.md) |
+| Middle Low German | `GML` | `gml-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 565 distinct usable word forms" title="Relative dictionary size 2 of 5; 565 distinct usable word forms">★★☆☆☆</span> (565) | Individual artifact; outside standard aggregate | [Middle Low German](gml.md) |
+| Mongolian | `MN_MN` | `mn-mn-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 17,231 distinct usable word forms" title="Relative dictionary size 4 of 5; 17,231 distinct usable word forms">★★★★☆</span> (17,231) | Individual artifact; outside standard aggregate | [Mongolian](mn-mn.md) |
+| Murrinh-Patha | `MWF` | `mwf-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 592 distinct usable word forms" title="Relative dictionary size 2 of 5; 592 distinct usable word forms">★★☆☆☆</span> (592) | Individual artifact; outside standard aggregate | [Murrinh-Patha](mwf.md) |
+| Navajo | `NAV` | `nav-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 9,826 distinct usable word forms" title="Relative dictionary size 3 of 5; 9,826 distinct usable word forms">★★★☆☆</span> (9,826) | Individual artifact; outside standard aggregate | [Navajo](nav.md) |
+| Neapolitan | `NAP` | `nap-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 1,497 distinct usable word forms" title="Relative dictionary size 2 of 5; 1,497 distinct usable word forms">★★☆☆☆</span> (1,497) | Individual artifact; outside standard aggregate | [Neapolitan](nap.md) |
+| North Frisian | `FRR` | `frr-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 340 distinct usable word forms" title="Relative dictionary size 1 of 5; 340 distinct usable word forms">★☆☆☆☆</span> (340) | Individual artifact; outside standard aggregate | [North Frisian](frr.md) |
+| Northern Sami | `SME` | `sme-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 53,034 distinct usable word forms" title="Relative dictionary size 4 of 5; 53,034 distinct usable word forms">★★★★☆</span> (53,034) | Individual artifact; outside standard aggregate | [Northern Sami](sme.md) |
+| Norwegian Bokmål | `NB_NO` | `nb-no-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 73,170 distinct usable word forms" title="Relative dictionary size 5 of 5; 73,170 distinct usable word forms">★★★★★</span> (73,170) | Standard aggregate | [Norwegian Bokmål](norwegian-bokmal.md) |
+| Norwegian Nynorsk | `NN_NO` | `nn-no-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 16,937 distinct usable word forms" title="Relative dictionary size 4 of 5; 16,937 distinct usable word forms">★★★★☆</span> (16,937) | Standard aggregate | [Norwegian Nynorsk](norwegian-nynorsk.md) |
+| Old English | `ANG` | `ang-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 65,100 distinct usable word forms" title="Relative dictionary size 5 of 5; 65,100 distinct usable word forms">★★★★★</span> (65,100) | Individual artifact; outside standard aggregate | [Old English](ang.md) |
+| Old French | `FRO` | `fro-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 93,998 distinct usable word forms" title="Relative dictionary size 5 of 5; 93,998 distinct usable word forms">★★★★★</span> (93,998) | Individual artifact; outside standard aggregate | [Old French](fro.md) |
+| Old High German | `GOH` | `goh-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 5,267 distinct usable word forms" title="Relative dictionary size 3 of 5; 5,267 distinct usable word forms">★★★☆☆</span> (5,267) | Individual artifact; outside standard aggregate | [Old High German](goh.md) |
+| Old Irish | `SGA` | `sga-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 929 distinct usable word forms" title="Relative dictionary size 2 of 5; 929 distinct usable word forms">★★☆☆☆</span> (929) | Individual artifact; outside standard aggregate | [Old Irish](sga.md) |
+| Old Norse | `NON` | `non-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 46,067 distinct usable word forms" title="Relative dictionary size 4 of 5; 46,067 distinct usable word forms">★★★★☆</span> (46,067) | Individual artifact; outside standard aggregate | [Old Norse](non.md) |
+| Old Saxon | `OSX` | `osx-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 12,361 distinct usable word forms" title="Relative dictionary size 4 of 5; 12,361 distinct usable word forms">★★★★☆</span> (12,361) | Individual artifact; outside standard aggregate | [Old Saxon](osx.md) |
+| Oodham | `OOD` | `ood-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 455 distinct usable word forms" title="Relative dictionary size 1 of 5; 455 distinct usable word forms">★☆☆☆☆</span> (455) | Individual artifact; outside standard aggregate | [Oodham](ood.md) |
+| Pashto | `PS_AF` | `ps-af-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 2,945 distinct usable word forms" title="Relative dictionary size 3 of 5; 2,945 distinct usable word forms">★★★☆☆</span> (2,945) | Individual artifact; outside standard aggregate | [Pashto](ps-af.md) |
+| Persian | `FA_IR` | `fa-ir-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 3,544 distinct usable word forms" title="Relative dictionary size 3 of 5; 3,544 distinct usable word forms">★★★☆☆</span> (3,544) | Standard aggregate | [Persian](persian.md) |
+| Polish | `PL_PL` | `pl-pl-unimorph`; optional `pl-pl-polimorf` (<span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 4,668,685 distinct usable word forms" title="Relative dictionary size 5 of 5; 4,668,685 distinct usable word forms">★★★★★</span>) | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 120,867 distinct usable word forms" title="Relative dictionary size 5 of 5; 120,867 distinct usable word forms">★★★★★</span> (120,867) | Standard aggregate; optional model is individual only | [Polish](polish.md) |
+| Portuguese | `PT_PT` | `pt-pt-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 211,091 distinct usable word forms" title="Relative dictionary size 5 of 5; 211,091 distinct usable word forms">★★★★★</span> (211,091) | Standard aggregate | [Portuguese](portuguese.md) |
+| Quechua | `QUE` | `que-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 122,820 distinct usable word forms" title="Relative dictionary size 5 of 5; 122,820 distinct usable word forms">★★★★★</span> (122,820) | Individual artifact; outside standard aggregate | [Quechua](que.md) |
+| Romanian | `RO_RO` | `ro-ro-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 48,497 distinct usable word forms" title="Relative dictionary size 4 of 5; 48,497 distinct usable word forms">★★★★☆</span> (48,497) | Individual artifact; outside standard aggregate | [Romanian](ro-ro.md) |
+| Russian | `RU_RU` | `ru-ru-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 759,333 distinct usable word forms" title="Relative dictionary size 5 of 5; 759,333 distinct usable word forms">★★★★★</span> (759,333) | Standard aggregate | [Russian](russian.md) |
+| Seneca | `SEE` | `see-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 483 distinct usable word forms" title="Relative dictionary size 1 of 5; 483 distinct usable word forms">★☆☆☆☆</span> (483) | Individual artifact; outside standard aggregate | [Seneca](see.md) |
+| Serbo-Croatian | `HBS` | `hbs-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 272,515 distinct usable word forms" title="Relative dictionary size 5 of 5; 272,515 distinct usable word forms">★★★★★</span> (272,515) | Individual artifact; outside standard aggregate | [Serbo-Croatian](hbs.md) |
+| Shipibo-Conibo | `SHP` | `shp-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 7,705 distinct usable word forms" title="Relative dictionary size 3 of 5; 7,705 distinct usable word forms">★★★☆☆</span> (7,705) | Individual artifact; outside standard aggregate | [Shipibo-Conibo](shp.md) |
+| Shona | `SN_ZW` | `sn-zw-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 2,640 distinct usable word forms" title="Relative dictionary size 3 of 5; 2,640 distinct usable word forms">★★★☆☆</span> (2,640) | Individual artifact; outside standard aggregate | [Shona](sn-zw.md) |
+| Sotho, Southern | `ST_ZA` | `st-za-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 416 distinct usable word forms" title="Relative dictionary size 1 of 5; 416 distinct usable word forms">★☆☆☆☆</span> (416) | Individual artifact; outside standard aggregate | [Sotho, Southern](st-za.md) |
+| Southern Kurdish | `SDH` | `sdh-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 165 distinct usable word forms" title="Relative dictionary size 1 of 5; 165 distinct usable word forms">★☆☆☆☆</span> (165) | Individual artifact; outside standard aggregate | [Southern Kurdish](sdh.md) |
+| Spanish | `ES_ES` | `es-es-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 849,661 distinct usable word forms" title="Relative dictionary size 5 of 5; 849,661 distinct usable word forms">★★★★★</span> (849,661) | Standard aggregate | [Spanish](spanish.md) |
+| Swedish | `SV_SE` | `sv-se-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 95,181 distinct usable word forms" title="Relative dictionary size 5 of 5; 95,181 distinct usable word forms">★★★★★</span> (95,181) | Standard aggregate | [Swedish](swedish.md) |
+| Tagalog | `TL_PH` | `tl-ph-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 2,042 distinct usable word forms" title="Relative dictionary size 3 of 5; 2,042 distinct usable word forms">★★★☆☆</span> (2,042) | Individual artifact; outside standard aggregate | [Tagalog](tl-ph.md) |
+| Temoaya Otomi | `OTE` | `ote-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 1,785 distinct usable word forms" title="Relative dictionary size 2 of 5; 1,785 distinct usable word forms">★★☆☆☆</span> (1,785) | Individual artifact; outside standard aggregate | [Temoaya Otomi](ote.md) |
+| Turkish | `TR_TR` | `tr-tr-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 222,207 distinct usable word forms" title="Relative dictionary size 5 of 5; 222,207 distinct usable word forms">★★★★★</span> (222,207) | Individual artifact; outside standard aggregate | [Turkish](tr-tr.md) |
+| Ukrainian | `UK_UA` | `uk-ua-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 14,150 distinct usable word forms" title="Relative dictionary size 4 of 5; 14,150 distinct usable word forms">★★★★☆</span> (14,150) | Standard aggregate | [Ukrainian](ukrainian.md) |
+| Uyghur | `UG_CN` | `ug-cn-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 6,200 distinct usable word forms" title="Relative dictionary size 3 of 5; 6,200 distinct usable word forms">★★★☆☆</span> (6,200) | Individual artifact; outside standard aggregate | [Uyghur](ug-cn.md) |
+| Uzbek | `UZ_UZ` | `uz-uz-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 1,080 distinct usable word forms" title="Relative dictionary size 2 of 5; 1,080 distinct usable word forms">★★☆☆☆</span> (1,080) | Individual artifact; outside standard aggregate | [Uzbek](uz-uz.md) |
+| Voro | `VRO` | `vro-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 277 distinct usable word forms" title="Relative dictionary size 1 of 5; 277 distinct usable word forms">★☆☆☆☆</span> (277) | Individual artifact; outside standard aggregate | [Voro](vro.md) |
+| Western Highland Chatino | `CTP` | `ctp-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 2,456 distinct usable word forms" title="Relative dictionary size 3 of 5; 2,456 distinct usable word forms">★★★☆☆</span> (2,456) | Individual artifact; outside standard aggregate | [Western Highland Chatino](ctp.md) |
+| Xibe | `SJO` | `sjo-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 3,151 distinct usable word forms" title="Relative dictionary size 3 of 5; 3,151 distinct usable word forms">★★★☆☆</span> (3,151) | Individual artifact; outside standard aggregate | [Xibe](sjo.md) |
+| Xipinawa | `XTY` | `xty-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 1,536 distinct usable word forms" title="Relative dictionary size 2 of 5; 1,536 distinct usable word forms">★★☆☆☆</span> (1,536) | Individual artifact; outside standard aggregate | [Xipinawa](xty.md) |
+| Yamana | `AME` | `ame-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 1,489 distinct usable word forms" title="Relative dictionary size 2 of 5; 1,489 distinct usable word forms">★★☆☆☆</span> (1,489) | Individual artifact; outside standard aggregate | [Yamana](ame.md) |
+| Yiddish | `YI` | `yi-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 3,532 distinct usable word forms" title="Relative dictionary size 3 of 5; 3,532 distinct usable word forms">★★★☆☆</span> (3,532) | Standard aggregate | [Yiddish](yiddish.md) |
+| Zacatepec Chatino | `ZPV` | `zpv-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 812 distinct usable word forms" title="Relative dictionary size 2 of 5; 812 distinct usable word forms">★★☆☆☆</span> (812) | Individual artifact; outside standard aggregate | [Zacatepec Chatino](zpv.md) |
+| Zarma | `DJE` | `dje-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 75 distinct usable word forms" title="Relative dictionary size 1 of 5; 75 distinct usable word forms">★☆☆☆☆</span> (75) | Individual artifact; outside standard aggregate | [Zarma](dje.md) |
+| Zenzontepec Chatino | `CZN` | `czn-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 68 distinct usable word forms" title="Relative dictionary size 1 of 5; 68 distinct usable word forms">★☆☆☆☆</span> (68) | Individual artifact; outside standard aggregate | [Zenzontepec Chatino](czn.md) |
+| Zulu | `ZU_ZA` | `zu-za-default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 32,384 distinct usable word forms" title="Relative dictionary size 4 of 5; 32,384 distinct usable word forms">★★★★☆</span> (32,384) | Individual artifact; outside standard aggregate | [Zulu](zu-za.md) |
 
 ## Methodology Notes
 
diff --git a/docs/benchmarks/languages/italian.md b/docs/benchmarks/languages/italian.md
index 0a7b806..2e62bc0 100644
--- a/docs/benchmarks/languages/italian.md
+++ b/docs/benchmarks/languages/italian.md
@@ -1,8 +1,12 @@
-# Italian Stemmer Benchmarks
+# Italian Stemmer Benchmarks <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 324,366 distinct usable word forms" title="Relative dictionary size 5 of 5; 324,366 distinct usable word forms">★★★★★</span>
 
 This page reports same-language stemming benchmarks for Italian. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
 
-All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-08-25 Radixor/Java `4.2.0-6-g84e57fb` snapshot. Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+<!-- DICTIONARY-SIZE-RATING:START -->
+Dictionary size: <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 324,366 distinct usable word forms" title="Relative dictionary size 5 of 5; 324,366 distinct usable word forms">★★★★★</span>. The exact count is **324,366 distinct usable word forms** after parser-compatible filtering and exact, case-preserved deduplication. Stars rank dictionary size relative to all 167 user-facing models in five nearly equal groups; they do **not** measure linguistic quality or benchmark accuracy.
+<!-- DICTIONARY-SIZE-RATING:END -->
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-09-11 Radixor/Java `4.3.0.dirty` snapshot. Speed benchmark operations process changed tokens. Accuracy uses the complete Radixor dictionary for the language.
 
 <!-- BENCHMARK-EVIDENCE-MAP:START -->
 !!! info "How to read this page"
@@ -13,9 +17,9 @@ Runtime and exact-root agreement measure different properties. Light, minimal, p
 
 ## Dictionary Corpus
 
-| Model ID | Model version | Language | Dictionary rows | Complete quality tokens | Already-root tokens | Changed tokens | JMH timing tokens |
-| --- | --- | --- | ---: | ---: | ---: | ---: | ---: |
-| `it-it-default` | `1.0.0` | `IT_IT` | 10,009 | 337,546 | 20,004 | 317,542 | 317,542 |
+| Model ID | Model version | Language | Dictionary rows | Distinct usable forms | Complete quality tokens | Already-root tokens | Changed tokens | Timing workload | JMH timing tokens |
+| --- | --- | --- | ---: | ---: | ---: | ---: | ---: | --- | ---: |
+| `it-it-default` | `1.0.0` | `IT_IT` | 10,009 | 324,366 | 337,546 | 20,004 | 317,542 | changed tokens | 317,542 |
 
 ## Radixor Patch Command Distribution
 
@@ -41,14 +45,14 @@ Accuracy is computed from JMH auxiliary counters in the current report. The coun
 
 ## Speed
 
-Speed uses JMH average time, 5 warmup iterations, 7 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
 
 | Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
 | --- | --- | ---: | ---: | ---: | ---: | --- |
-| Radixor | `italianRadixor` | 23.234 | 0.511 | 73.2 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
-| Lucene ItalianLightStemFilter | `italianLuceneItalianLightStemFilter` | 14.961 | 0.201 | 47.1 | 0.644 | Light Italian suffix stemmer. |
-| Official Snowball direct | `snowballDirect[ITALIAN]` | 110.443 | 3.741 | 347.8 | 4.754 | Official Snowball generated Java stemmer; direct API. |
-| Lucene SnowballFilter | `luceneSnowballFilter[ITALIAN]` | 115.937 | 4.929 | 365.1 | 4.990 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
+| Radixor | `italianRadixor` | 24.797 | 0.581 | 78.1 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
+| Lucene ItalianLightStemFilter | `italianLuceneItalianLightStemFilter` | 16.243 | 0.373 | 51.2 | 0.655 | Light Italian suffix stemmer. |
+| Official Snowball direct | `snowballDirect[ITALIAN]` | 110.766 | 5.309 | 348.8 | 4.467 | Official Snowball generated Java stemmer; direct API. |
+| Lucene SnowballFilter | `luceneSnowballFilter[ITALIAN]` | 112.716 | 3.913 | 355.0 | 4.545 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
 
 ## Interpretation Notes
 
@@ -86,18 +90,15 @@ also appeared in training. Parentheses show the observed split minimum–maximum
 
 ### Generalization conclusion
 
-- Median exactness on genuinely unseen changed forms moves from **86.665%**
-  at 10% training knowledge to **93.600%** at 90%, a measured
-  **+6.935 percentage-point** change for this dictionary.
-- Over the same endpoints, unseen all-form exactness changes by **+6.518 pp** and
-  preservation of unseen already-root forms changes by **+0.105 pp**. These separate
-  outcomes show whether the changed-form result coexists with preservation behavior.
+- Median exactness on genuinely unseen changed forms moves from **86.665%** at 10% training knowledge to **93.600%** at 90%, a measured **+6.935 percentage-point** change.
+- Unseen all-form exactness moves from **87.447%** at 10% training knowledge to **93.965%** at 90%, a measured **+6.518 percentage-point** change.
+- Preservation of unseen already-root forms moves from **99.794%** at 10% training knowledge to **99.900%** at 90%, a measured **+0.105 percentage-point** change.
 - The evidence establishes within-resource transfer across withheld dictionary families. It
   does not estimate unrelated domains, misspellings, arbitrary compounds, or external corpora.
 
 The complete ten-level table and split ranges remain in the
 [independent generalization report](../generalization.md); raw counters and provenance are in
-[`dictionary-generalization.csv`](../data/dictionary-generalization.csv). The
+[active machine-readable snapshot](../data/dictionary-generalization-2026-09-11.csv). The
 [frozen methodology](../reference/generalization-methodology.md) defines family-level
 splitting, unseen-surface leakage control, aggregation, and the limits of the claim.
 
@@ -190,7 +191,7 @@ The complete evidence is available in the [raw logical matrix](../data/edit-cost
 
 Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `IT_IT` using the complete validated stemming-quality result matrix. Every distinct surface form is one evaluated item and can belong to several dictionary groups. Two forms are a positive pair when their group-membership sets intersect and a negative pair when those sets are disjoint. A pair shared through several groups is counted once. Exact equality with a predetermined lemma is not required.
 
-`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality-2026-09-11.csv).
 
 ### Evaluation Scope and Key Findings
 
@@ -445,8 +446,8 @@ Standard ARI, homogeneity, completeness, V-measure, and NMI are not calculated:
 
 ### Provenance
 
-- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
-- Source SHA-256: `85763189eab4d0fbb047c2d5d3554c66abf9732182bd0d8fd758d7aef680e66f`
+- Authoritative source: `docs/benchmarks/data/stemming-quality-2026-09-11.csv`
+- Source SHA-256: `24bddfeed06a60bb3eeed58e1bfc93aed46c1d32e7bebd293aec2dacfddfef5b`
 - Evaluation command: `./gradlew stemmingQuality --no-daemon`
 - Dictionary language: `IT_IT`
 - Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
diff --git a/docs/benchmarks/languages/norwegian-bokmal.md b/docs/benchmarks/languages/norwegian-bokmal.md
index 5a5baa1..f157622 100644
--- a/docs/benchmarks/languages/norwegian-bokmal.md
+++ b/docs/benchmarks/languages/norwegian-bokmal.md
@@ -1,8 +1,12 @@
-# Norwegian Bokmal Stemmer Benchmarks
+# Norwegian Bokmal Stemmer Benchmarks <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 73,170 distinct usable word forms" title="Relative dictionary size 5 of 5; 73,170 distinct usable word forms">★★★★★</span>
 
 This page reports same-language stemming benchmarks for Norwegian Bokmal. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
 
-All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-08-25 Radixor/Java `4.2.0-6-g84e57fb` snapshot. Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+<!-- DICTIONARY-SIZE-RATING:START -->
+Dictionary size: <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 73,170 distinct usable word forms" title="Relative dictionary size 5 of 5; 73,170 distinct usable word forms">★★★★★</span>. The exact count is **73,170 distinct usable word forms** after parser-compatible filtering and exact, case-preserved deduplication. Stars rank dictionary size relative to all 167 user-facing models in five nearly equal groups; they do **not** measure linguistic quality or benchmark accuracy.
+<!-- DICTIONARY-SIZE-RATING:END -->
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-09-11 Radixor/Java `4.3.0.dirty` snapshot. Speed benchmark operations process changed tokens. Accuracy uses the complete Radixor dictionary for the language.
 
 <!-- BENCHMARK-EVIDENCE-MAP:START -->
 !!! info "How to read this page"
@@ -13,9 +17,9 @@ Runtime and exact-root agreement measure different properties. Light, minimal, p
 
 ## Dictionary Corpus
 
-| Model ID | Model version | Language | Dictionary rows | Complete quality tokens | Already-root tokens | Changed tokens | JMH timing tokens |
-| --- | --- | --- | ---: | ---: | ---: | ---: | ---: |
-| `nb-no-default` | `1.0.0` | `NB_NO` | 17,929 | 90,757 | 33,376 | 57,381 | 57,381 |
+| Model ID | Model version | Language | Dictionary rows | Distinct usable forms | Complete quality tokens | Already-root tokens | Changed tokens | Timing workload | JMH timing tokens |
+| --- | --- | --- | ---: | ---: | ---: | ---: | ---: | --- | ---: |
+| `nb-no-default` | `1.0.0` | `NB_NO` | 17,929 | 73,170 | 90,757 | 33,376 | 57,381 | changed tokens | 57,381 |
 
 ## Radixor Patch Command Distribution
 
@@ -43,15 +47,15 @@ Accuracy is computed from JMH auxiliary counters in the current report. The coun
 
 ## Speed
 
-Speed uses JMH average time, 5 warmup iterations, 7 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
 
 | Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
 | --- | --- | ---: | ---: | ---: | ---: | --- |
-| Radixor | `norwegianBokmalRadixor` | 3.315 | 0.067 | 57.8 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
-| Lucene NorwegianMinimalStemFilter | `norwegianBokmalLuceneNorwegianMinimalStemFilter` | 2.828 | 0.055 | 49.3 | 0.853 | Minimal Norwegian suffix reducer. |
-| Lucene NorwegianLightStemFilter | `norwegianBokmalLuceneNorwegianLightStemFilter` | 3.260 | 0.093 | 56.8 | 0.983 | Light Norwegian suffix stemmer. |
-| Official Snowball direct | `snowballDirect[NORWEGIAN_BOKMAL]` | 4.903 | 0.592 | 85.4 | 1.479 | Official Snowball generated Java stemmer; direct API. |
-| Lucene SnowballFilter | `luceneSnowballFilter[NORWEGIAN_BOKMAL]` | 5.856 | 0.384 | 102.1 | 1.766 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
+| Radixor | `norwegianBokmalRadixor` | 3.217 | 0.114 | 56.1 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
+| Lucene NorwegianMinimalStemFilter | `norwegianBokmalLuceneNorwegianMinimalStemFilter` | 2.991 | 0.079 | 52.1 | 0.930 | Minimal Norwegian suffix reducer. |
+| Lucene NorwegianLightStemFilter | `norwegianBokmalLuceneNorwegianLightStemFilter` | 3.389 | 0.059 | 59.1 | 1.054 | Light Norwegian suffix stemmer. |
+| Official Snowball direct | `snowballDirect[NORWEGIAN_BOKMAL]` | 5.233 | 0.760 | 91.2 | 1.627 | Official Snowball generated Java stemmer; direct API. |
+| Lucene SnowballFilter | `luceneSnowballFilter[NORWEGIAN_BOKMAL]` | 6.614 | 1.014 | 115.3 | 2.056 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
 
 ## Interpretation Notes
 
@@ -89,18 +93,15 @@ also appeared in training. Parentheses show the observed split minimum–maximum
 
 ### Generalization conclusion
 
-- Median exactness on genuinely unseen changed forms moves from **58.598%**
-  at 10% training knowledge to **70.546%** at 90%, a measured
-  **+11.949 percentage-point** change for this dictionary.
-- Over the same endpoints, unseen all-form exactness changes by **+9.761 pp** and
-  preservation of unseen already-root forms changes by **+5.538 pp**. These separate
-  outcomes show whether the changed-form result coexists with preservation behavior.
+- Median exactness on genuinely unseen changed forms moves from **58.598%** at 10% training knowledge to **70.546%** at 90%, a measured **+11.949 percentage-point** change.
+- Unseen all-form exactness moves from **69.855%** at 10% training knowledge to **79.616%** at 90%, a measured **+9.761 percentage-point** change.
+- Preservation of unseen already-root forms moves from **89.971%** at 10% training knowledge to **95.510%** at 90%, a measured **+5.538 percentage-point** change.
 - The evidence establishes within-resource transfer across withheld dictionary families. It
   does not estimate unrelated domains, misspellings, arbitrary compounds, or external corpora.
 
 The complete ten-level table and split ranges remain in the
 [independent generalization report](../generalization.md); raw counters and provenance are in
-[`dictionary-generalization.csv`](../data/dictionary-generalization.csv). The
+[active machine-readable snapshot](../data/dictionary-generalization-2026-09-11.csv). The
 [frozen methodology](../reference/generalization-methodology.md) defines family-level
 splitting, unseen-surface leakage control, aggregation, and the limits of the claim.
 
@@ -193,7 +194,7 @@ The complete evidence is available in the [raw logical matrix](../data/edit-cost
 
 Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `NB_NO` using the complete validated stemming-quality result matrix. Every distinct surface form is one evaluated item and can belong to several dictionary groups. Two forms are a positive pair when their group-membership sets intersect and a negative pair when those sets are disjoint. A pair shared through several groups is counted once. Exact equality with a predetermined lemma is not required.
 
-`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality-2026-09-11.csv).
 
 ### Evaluation Scope and Key Findings
 
@@ -456,8 +457,8 @@ Standard ARI, homogeneity, completeness, V-measure, and NMI are not calculated:
 
 ### Provenance
 
-- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
-- Source SHA-256: `85763189eab4d0fbb047c2d5d3554c66abf9732182bd0d8fd758d7aef680e66f`
+- Authoritative source: `docs/benchmarks/data/stemming-quality-2026-09-11.csv`
+- Source SHA-256: `24bddfeed06a60bb3eeed58e1bfc93aed46c1d32e7bebd293aec2dacfddfef5b`
 - Evaluation command: `./gradlew stemmingQuality --no-daemon`
 - Dictionary language: `NB_NO`
 - Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
diff --git a/docs/benchmarks/languages/norwegian-nynorsk.md b/docs/benchmarks/languages/norwegian-nynorsk.md
index ba29b2e..0fe4438 100644
--- a/docs/benchmarks/languages/norwegian-nynorsk.md
+++ b/docs/benchmarks/languages/norwegian-nynorsk.md
@@ -1,8 +1,12 @@
-# Norwegian Nynorsk Stemmer Benchmarks
+# Norwegian Nynorsk Stemmer Benchmarks <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 16,937 distinct usable word forms" title="Relative dictionary size 4 of 5; 16,937 distinct usable word forms">★★★★☆</span>
 
 This page reports same-language stemming benchmarks for Norwegian Nynorsk. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
 
-All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-08-25 Radixor/Java `4.2.0-6-g84e57fb` snapshot. Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+<!-- DICTIONARY-SIZE-RATING:START -->
+Dictionary size: <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 16,937 distinct usable word forms" title="Relative dictionary size 4 of 5; 16,937 distinct usable word forms">★★★★☆</span>. The exact count is **16,937 distinct usable word forms** after parser-compatible filtering and exact, case-preserved deduplication. Stars rank dictionary size relative to all 167 user-facing models in five nearly equal groups; they do **not** measure linguistic quality or benchmark accuracy.
+<!-- DICTIONARY-SIZE-RATING:END -->
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-09-11 Radixor/Java `4.3.0.dirty` snapshot. Speed benchmark operations process changed tokens. Accuracy uses the complete Radixor dictionary for the language.
 
 <!-- BENCHMARK-EVIDENCE-MAP:START -->
 !!! info "How to read this page"
@@ -13,9 +17,9 @@ Runtime and exact-root agreement measure different properties. Light, minimal, p
 
 ## Dictionary Corpus
 
-| Model ID | Model version | Language | Dictionary rows | Complete quality tokens | Already-root tokens | Changed tokens | JMH timing tokens |
-| --- | --- | --- | ---: | ---: | ---: | ---: | ---: |
-| `nn-no-default` | `1.0.0` | `NN_NO` | 4,688 | 19,651 | 6,089 | 13,562 | 13,562 |
+| Model ID | Model version | Language | Dictionary rows | Distinct usable forms | Complete quality tokens | Already-root tokens | Changed tokens | Timing workload | JMH timing tokens |
+| --- | --- | --- | ---: | ---: | ---: | ---: | ---: | --- | ---: |
+| `nn-no-default` | `1.0.0` | `NN_NO` | 4,688 | 16,937 | 19,651 | 6,089 | 13,562 | changed tokens | 13,562 |
 
 ## Radixor Patch Command Distribution
 
@@ -41,13 +45,13 @@ Accuracy is computed from JMH auxiliary counters in the current report. The coun
 
 ## Speed
 
-Speed uses JMH average time, 5 warmup iterations, 7 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
 
 | Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
 | --- | --- | ---: | ---: | ---: | ---: | --- |
-| Radixor | `radixor[NORWEGIAN_NYNORSK]` | 0.596 | 0.083 | 44.0 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
-| Official Snowball direct | `snowballDirect[NORWEGIAN_NYNORSK]` | 1.130 | 0.142 | 83.3 | 1.895 | Official Snowball generated Java stemmer; direct API. |
-| Lucene SnowballFilter | `luceneSnowballFilter[NORWEGIAN_NYNORSK]` | 1.306 | 0.146 | 96.3 | 2.191 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
+| Radixor | `radixor[NORWEGIAN_NYNORSK]` | 0.681 | 0.133 | 50.2 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
+| Official Snowball direct | `snowballDirect[NORWEGIAN_NYNORSK]` | 1.197 | 0.171 | 88.2 | 1.757 | Official Snowball generated Java stemmer; direct API. |
+| Lucene SnowballFilter | `luceneSnowballFilter[NORWEGIAN_NYNORSK]` | 1.461 | 0.255 | 107.7 | 2.145 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
 
 ## Interpretation Notes
 
@@ -85,18 +89,15 @@ also appeared in training. Parentheses show the observed split minimum–maximum
 
 ### Generalization conclusion
 
-- Median exactness on genuinely unseen changed forms moves from **51.191%**
-  at 10% training knowledge to **70.338%** at 90%, a measured
-  **+19.147 percentage-point** change for this dictionary.
-- Over the same endpoints, unseen all-form exactness changes by **+15.413 pp** and
-  preservation of unseen already-root forms changes by **+4.683 pp**. These separate
-  outcomes show whether the changed-form result coexists with preservation behavior.
+- Median exactness on genuinely unseen changed forms moves from **51.191%** at 10% training knowledge to **70.338%** at 90%, a measured **+19.147 percentage-point** change.
+- Unseen all-form exactness moves from **63.195%** at 10% training knowledge to **78.608%** at 90%, a measured **+15.413 percentage-point** change.
+- Preservation of unseen already-root forms moves from **90.470%** at 10% training knowledge to **95.153%** at 90%, a measured **+4.683 percentage-point** change.
 - The evidence establishes within-resource transfer across withheld dictionary families. It
   does not estimate unrelated domains, misspellings, arbitrary compounds, or external corpora.
 
 The complete ten-level table and split ranges remain in the
 [independent generalization report](../generalization.md); raw counters and provenance are in
-[`dictionary-generalization.csv`](../data/dictionary-generalization.csv). The
+[active machine-readable snapshot](../data/dictionary-generalization-2026-09-11.csv). The
 [frozen methodology](../reference/generalization-methodology.md) defines family-level
 splitting, unseen-surface leakage control, aggregation, and the limits of the claim.
 
@@ -189,7 +190,7 @@ The complete evidence is available in the [raw logical matrix](../data/edit-cost
 
 Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `NN_NO` using the complete validated stemming-quality result matrix. Every distinct surface form is one evaluated item and can belong to several dictionary groups. Two forms are a positive pair when their group-membership sets intersect and a negative pair when those sets are disjoint. A pair shared through several groups is counted once. Exact equality with a predetermined lemma is not required.
 
-`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality-2026-09-11.csv).
 
 ### Evaluation Scope and Key Findings
 
@@ -436,8 +437,8 @@ Standard ARI, homogeneity, completeness, V-measure, and NMI are not calculated:
 
 ### Provenance
 
-- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
-- Source SHA-256: `85763189eab4d0fbb047c2d5d3554c66abf9732182bd0d8fd758d7aef680e66f`
+- Authoritative source: `docs/benchmarks/data/stemming-quality-2026-09-11.csv`
+- Source SHA-256: `24bddfeed06a60bb3eeed58e1bfc93aed46c1d32e7bebd293aec2dacfddfef5b`
 - Evaluation command: `./gradlew stemmingQuality --no-daemon`
 - Dictionary language: `NN_NO`
 - Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
diff --git a/docs/benchmarks/languages/persian.md b/docs/benchmarks/languages/persian.md
index 7eac6ea..7e575d0 100644
--- a/docs/benchmarks/languages/persian.md
+++ b/docs/benchmarks/languages/persian.md
@@ -1,8 +1,12 @@
-# Persian Stemmer Benchmarks
+# Persian Stemmer Benchmarks <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 3,544 distinct usable word forms" title="Relative dictionary size 3 of 5; 3,544 distinct usable word forms">★★★☆☆</span>
 
 This page reports same-language stemming benchmarks for Persian. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
 
-All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-08-25 Radixor/Java `4.2.0-6-g84e57fb` snapshot. Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+<!-- DICTIONARY-SIZE-RATING:START -->
+Dictionary size: <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 3,544 distinct usable word forms" title="Relative dictionary size 3 of 5; 3,544 distinct usable word forms">★★★☆☆</span>. The exact count is **3,544 distinct usable word forms** after parser-compatible filtering and exact, case-preserved deduplication. Stars rank dictionary size relative to all 167 user-facing models in five nearly equal groups; they do **not** measure linguistic quality or benchmark accuracy.
+<!-- DICTIONARY-SIZE-RATING:END -->
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-09-11 Radixor/Java `4.3.0.dirty` snapshot. Speed benchmark operations process changed tokens. Accuracy uses the complete Radixor dictionary for the language.
 
 <!-- BENCHMARK-EVIDENCE-MAP:START -->
 !!! info "How to read this page"
@@ -13,9 +17,9 @@ Runtime and exact-root agreement measure different properties. Light, minimal, p
 
 ## Dictionary Corpus
 
-| Model ID | Model version | Language | Dictionary rows | Complete quality tokens | Already-root tokens | Changed tokens | JMH timing tokens |
-| --- | --- | --- | ---: | ---: | ---: | ---: | ---: |
-| `fa-ir-default` | `1.0.0` | `FA_IR` | 69 | 3,770 | 138 | 3,632 | 5,000 |
+| Model ID | Model version | Language | Dictionary rows | Distinct usable forms | Complete quality tokens | Already-root tokens | Changed tokens | Timing workload | JMH timing tokens |
+| --- | --- | --- | ---: | ---: | ---: | ---: | ---: | --- | ---: |
+| `fa-ir-default` | `1.0.0` | `FA_IR` | 69 | 3,544 | 3,770 | 138 | 3,632 | changed tokens | 5,000 |
 
 ## Radixor Patch Command Distribution
 
@@ -41,13 +45,13 @@ Accuracy is computed from JMH auxiliary counters in the current report. The coun
 
 ## Speed
 
-Speed uses JMH average time, 5 warmup iterations, 7 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
 
 | Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
 | --- | --- | ---: | ---: | ---: | ---: | --- |
-| Radixor | `persianRadixor` | 0.265 | 0.006 | 53.0 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
-| Lucene PersianStemFilter | `persianLucenePersianStemFilter` | 0.458 | 0.008 | 91.6 | 1.728 | Persian suffix stemmer with Lucene normalization in the measured path. |
-| Official Snowball direct | `snowballDirect[PERSIAN]` | 1.531 | 0.073 | 306.2 | 5.777 | Official Snowball 3.1.0 generated Java stemmer; direct API. |
+| Radixor | `persianRadixor` | 0.280 | 0.007 | 56.1 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
+| Lucene PersianStemFilter | `persianLucenePersianStemFilter` | 0.477 | 0.009 | 95.4 | 1.702 | Persian suffix stemmer with Lucene normalization in the measured path. |
+| Official Snowball direct | `snowballDirect[PERSIAN]` | 1.511 | 0.089 | 302.3 | 5.392 | Official Snowball 3.1.0 generated Java stemmer; direct API. |
 
 ## Interpretation Notes
 
@@ -85,18 +89,15 @@ also appeared in training. Parentheses show the observed split minimum–maximum
 
 ### Generalization conclusion
 
-- Median exactness on genuinely unseen changed forms moves from **14.842%**
-  at 10% training knowledge to **15.493%** at 90%, a measured
-  **+0.651 percentage-point** change for this dictionary.
-- Over the same endpoints, unseen all-form exactness changes by **+0.712 pp** and
-  preservation of unseen already-root forms changes by **+0.000 pp**. These separate
-  outcomes show whether the changed-form result coexists with preservation behavior.
+- Median exactness on genuinely unseen changed forms moves from **14.842%** at 10% training knowledge to **15.493%** at 90%, a measured **+0.651 percentage-point** change.
+- Unseen all-form exactness moves from **17.987%** at 10% training knowledge to **18.699%** at 90%, a measured **+0.712 percentage-point** change.
+- Preservation of unseen already-root forms moves from **100.000%** at 10% training knowledge to **100.000%** at 90%, a measured **+0.000 percentage-point** change.
 - The evidence establishes within-resource transfer across withheld dictionary families. It
   does not estimate unrelated domains, misspellings, arbitrary compounds, or external corpora.
 
 The complete ten-level table and split ranges remain in the
 [independent generalization report](../generalization.md); raw counters and provenance are in
-[`dictionary-generalization.csv`](../data/dictionary-generalization.csv). The
+[active machine-readable snapshot](../data/dictionary-generalization-2026-09-11.csv). The
 [frozen methodology](../reference/generalization-methodology.md) defines family-level
 splitting, unseen-surface leakage control, aggregation, and the limits of the claim.
 
@@ -189,7 +190,7 @@ The complete evidence is available in the [raw logical matrix](../data/edit-cost
 
 Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `FA_IR` using the complete validated stemming-quality result matrix. Every distinct surface form is one evaluated item and can belong to several dictionary groups. Two forms are a positive pair when their group-membership sets intersect and a negative pair when those sets are disjoint. A pair shared through several groups is counted once. Exact equality with a predetermined lemma is not required.
 
-`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality-2026-09-11.csv).
 
 ### Evaluation Scope and Key Findings
 
@@ -436,8 +437,8 @@ Standard ARI, homogeneity, completeness, V-measure, and NMI are not calculated:
 
 ### Provenance
 
-- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
-- Source SHA-256: `85763189eab4d0fbb047c2d5d3554c66abf9732182bd0d8fd758d7aef680e66f`
+- Authoritative source: `docs/benchmarks/data/stemming-quality-2026-09-11.csv`
+- Source SHA-256: `24bddfeed06a60bb3eeed58e1bfc93aed46c1d32e7bebd293aec2dacfddfef5b`
 - Evaluation command: `./gradlew stemmingQuality --no-daemon`
 - Dictionary language: `FA_IR`
 - Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
diff --git a/docs/benchmarks/languages/polish.md b/docs/benchmarks/languages/polish.md
index 740695f..fd3861c 100644
--- a/docs/benchmarks/languages/polish.md
+++ b/docs/benchmarks/languages/polish.md
@@ -1,8 +1,12 @@
-# Polish Stemmer Benchmarks
+# Polish Stemmer Benchmarks <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 120,867 distinct usable word forms" title="Relative dictionary size 5 of 5; 120,867 distinct usable word forms">★★★★★</span>
 
 This page reports same-language stemming benchmarks for Polish. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
 
-All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-08-25 Radixor/Java `4.2.0-6-g84e57fb` snapshot. Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+<!-- DICTIONARY-SIZE-RATING:START -->
+Dictionary size: <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 120,867 distinct usable word forms" title="Relative dictionary size 5 of 5; 120,867 distinct usable word forms">★★★★★</span>. The exact count is **120,867 distinct usable word forms** after parser-compatible filtering and exact, case-preserved deduplication. Stars rank dictionary size relative to all 167 user-facing models in five nearly equal groups; they do **not** measure linguistic quality or benchmark accuracy.
+<!-- DICTIONARY-SIZE-RATING:END -->
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-09-11 Radixor/Java `4.3.0.dirty` snapshot. Speed benchmark operations process changed tokens. Accuracy uses the complete Radixor dictionary for the language.
 
 <!-- BENCHMARK-EVIDENCE-MAP:START -->
 !!! info "How to read this page"
@@ -13,9 +17,9 @@ Runtime and exact-root agreement measure different properties. Light, minimal, p
 
 ## Dictionary Corpus
 
-| Model ID | Model version | Language | Dictionary rows | Complete quality tokens | Already-root tokens | Changed tokens | JMH timing tokens |
-| --- | --- | --- | ---: | ---: | ---: | ---: | ---: |
-| `pl-pl-unimorph` | `1.0.0` | `PL_PL` | 9,990 | 132,308 | 19,957 | 112,351 | 112,351 |
+| Model ID | Model version | Language | Dictionary rows | Distinct usable forms | Complete quality tokens | Already-root tokens | Changed tokens | Timing workload | JMH timing tokens |
+| --- | --- | --- | ---: | ---: | ---: | ---: | ---: | --- | ---: |
+| `pl-pl-unimorph` | `1.0.0` | `PL_PL` | 9,990 | 120,867 | 132,308 | 19,957 | 112,351 | changed tokens | 112,351 |
 
 ## Radixor Patch Command Distribution
 
@@ -44,16 +48,16 @@ Accuracy is computed from JMH auxiliary counters in the current report. The coun
 
 ## Speed
 
-Speed uses JMH average time, 5 warmup iterations, 7 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
 
 | Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
 | --- | --- | ---: | ---: | ---: | ---: | --- |
-| Radixor | `polishRadixor` | 8.470 | 0.176 | 75.4 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
-| Lucene HunspellStemFilter | `luceneHunspellStemFilter` | 537.060 | 32.708 | 4780.2 | 63.408 | Benchmark-only Polish Hunspell dictionary compared via Lucene HunspellStemFilter. |
-| Lucene StempelStemmer direct | `polishLuceneStempelStemmerDirect` | 33.209 | 0.662 | 295.6 | 3.921 | Direct table-driven Polish Stempel stemmer API. |
-| Lucene StempelFilter | `polishLuceneStempelFilter` | 41.041 | 0.656 | 365.3 | 4.846 | Lucene TokenFilter integration path for table-driven Polish Stempel. |
-| Lucene MorfologikFilter | `polishLuceneMorfologikFilter` | 140.092 | 1.931 | 1246.9 | 16.540 | Dictionary-based Morfologik TokenFilter; may emit multiple terms. |
-| Official Snowball direct | `snowballDirect[POLISH]` | 10.168 | 1.442 | 90.5 | 1.201 | Official Snowball 3.1.0 generated Java stemmer; direct API. |
+| Radixor | `polishRadixor` | 8.121 | 0.140 | 72.3 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
+| Lucene HunspellStemFilter | `luceneHunspellStemFilter` | 544.375 | 51.396 | 4845.3 | 67.036 | Benchmark-only Polish Hunspell dictionary compared via Lucene HunspellStemFilter. |
+| Lucene StempelStemmer direct | `polishLuceneStempelStemmerDirect` | 33.200 | 1.703 | 295.5 | 4.088 | Direct table-driven Polish Stempel stemmer API. |
+| Lucene StempelFilter | `polishLuceneStempelFilter` | 40.165 | 0.872 | 357.5 | 4.946 | Lucene TokenFilter integration path for table-driven Polish Stempel. |
+| Lucene MorfologikFilter | `polishLuceneMorfologikFilter` | 145.100 | 7.257 | 1291.5 | 17.868 | Dictionary-based Morfologik TokenFilter; may emit multiple terms. |
+| Official Snowball direct | `snowballDirect[POLISH]` | 10.920 | 1.787 | 97.2 | 1.345 | Official Snowball 3.1.0 generated Java stemmer; direct API. |
 
 ## Interpretation Notes
 
@@ -63,6 +67,37 @@ Speed uses JMH average time, 5 warmup iterations, 7 measurement iterations, 3 in
 - Morfologik rows are dictionary-based and can emit multiple terms for one input token. Quality rows use the first returned term when no ranking weight is available.
 - Snowball rows are rule-based generated suffix stemmers; they are useful algorithmic baselines, not dictionary-root equivalence guarantees.
 
+<!-- POLIMORF-BENCHMARK:START -->
+
+## Optional PoliMorf Model
+
+PoliMorf is an independently available optional Polish model and is not the `PL_PL` default or a member of either standard aggregate. Its evidence is kept separate from the UniMorph default above.
+
+Optional dictionary size: <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 4,668,685 distinct usable word forms" title="Relative dictionary size 5 of 5; 4,668,685 distinct usable word forms">★★★★★</span>. The exact count is **4,668,685 distinct usable word forms**; the rating is relative size, not linguistic quality.
+
+### PoliMorf Corpus
+
+| Model ID | Model version | Language | Dictionary rows | Distinct usable forms | Complete quality tokens | Already-root tokens | Changed tokens | Timing workload | JMH timing tokens |
+| --- | --- | --- | ---: | ---: | ---: | ---: | ---: | --- | ---: |
+| `pl-pl-polimorf` | `1.0.0` | `PL_PL` | 315,639 | 4,668,685 | 4,812,023 | 315,639 | 4,496,384 | changed tokens | 4,496,384 |
+
+### PoliMorf Exact-root Accuracy
+
+| Stemmer | All exact | Changed exact | Root preserved | Note |
+| --- | ---: | ---: | ---: | --- |
+| Radixor PoliMorf | 98.837% | 98.744% | 99.359% | Exact optional model. |
+| PoliMorf Lucene MorfologikFilter | 87.729% | 86.606% | 94.047% | Matching Polish dictionary adapter. |
+| PoliMorf Official Snowball direct | 22.315% | 20.225% | 34.078% | Matching Polish Snowball algorithm. |
+
+### PoliMorf Speed
+
+| Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor PoliMorf | Note |
+| --- | --- | ---: | ---: | ---: | ---: | --- |
+| Radixor PoliMorf | `radixor[pl-pl-polimorf]` | 802.691 | 12.562 | 178.5 | 1.000 | Exact optional-model baseline. |
+| Lucene MorfologikFilter | `polishPolimorfLuceneMorfologikFilter` | 6199.653 | 99.810 | 1378.8 | 7.724 | PoliMorf-derived corpus. |
+
+<!-- POLIMORF-BENCHMARK:END -->
+
 <!-- DICTIONARY-GENERALIZATION:START -->
 
 ## Dictionary-Family Generalization Conclusion
@@ -91,18 +126,15 @@ also appeared in training. Parentheses show the observed split minimum–maximum
 
 ### Generalization conclusion
 
-- Median exactness on genuinely unseen changed forms moves from **61.996%**
-  at 10% training knowledge to **69.972%** at 90%, a measured
-  **+7.976 percentage-point** change for this dictionary.
-- Over the same endpoints, unseen all-form exactness changes by **+7.307 pp** and
-  preservation of unseen already-root forms changes by **+4.352 pp**. These separate
-  outcomes show whether the changed-form result coexists with preservation behavior.
+- Median exactness on genuinely unseen changed forms moves from **61.996%** at 10% training knowledge to **69.972%** at 90%, a measured **+7.976 percentage-point** change.
+- Unseen all-form exactness moves from **65.763%** at 10% training knowledge to **73.070%** at 90%, a measured **+7.307 percentage-point** change.
+- Preservation of unseen already-root forms moves from **86.633%** at 10% training knowledge to **90.985%** at 90%, a measured **+4.352 percentage-point** change.
 - The evidence establishes within-resource transfer across withheld dictionary families. It
   does not estimate unrelated domains, misspellings, arbitrary compounds, or external corpora.
 
 The complete ten-level table and split ranges remain in the
 [independent generalization report](../generalization.md); raw counters and provenance are in
-[`dictionary-generalization.csv`](../data/dictionary-generalization.csv). The
+[active machine-readable snapshot](../data/dictionary-generalization-2026-09-11.csv). The
 [frozen methodology](../reference/generalization-methodology.md) defines family-level
 splitting, unseen-surface leakage control, aggregation, and the limits of the claim.
 
@@ -195,7 +227,7 @@ The complete evidence is available in the [raw logical matrix](../data/edit-cost
 
 Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `PL_PL` using the complete validated stemming-quality result matrix. Every distinct surface form is one evaluated item and can belong to several dictionary groups. Two forms are a positive pair when their group-membership sets intersect and a negative pair when those sets are disjoint. A pair shared through several groups is counted once. Exact equality with a predetermined lemma is not required.
 
-`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality-2026-09-11.csv).
 
 ### Evaluation Scope and Key Findings
 
@@ -494,8 +526,8 @@ Standard ARI, homogeneity, completeness, V-measure, and NMI are not calculated:
 
 ### Provenance
 
-- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
-- Source SHA-256: `85763189eab4d0fbb047c2d5d3554c66abf9732182bd0d8fd758d7aef680e66f`
+- Authoritative source: `docs/benchmarks/data/stemming-quality-2026-09-11.csv`
+- Source SHA-256: `24bddfeed06a60bb3eeed58e1bfc93aed46c1d32e7bebd293aec2dacfddfef5b`
 - Evaluation command: `./gradlew stemmingQuality --no-daemon`
 - Dictionary language: `PL_PL`
 - Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
diff --git a/docs/benchmarks/languages/portuguese.md b/docs/benchmarks/languages/portuguese.md
index db9d7a2..7ec13e0 100644
--- a/docs/benchmarks/languages/portuguese.md
+++ b/docs/benchmarks/languages/portuguese.md
@@ -1,8 +1,12 @@
-# Portuguese Stemmer Benchmarks
+# Portuguese Stemmer Benchmarks <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 211,091 distinct usable word forms" title="Relative dictionary size 5 of 5; 211,091 distinct usable word forms">★★★★★</span>
 
 This page reports same-language stemming benchmarks for Portuguese. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
 
-All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-08-25 Radixor/Java `4.2.0-6-g84e57fb` snapshot. Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+<!-- DICTIONARY-SIZE-RATING:START -->
+Dictionary size: <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 211,091 distinct usable word forms" title="Relative dictionary size 5 of 5; 211,091 distinct usable word forms">★★★★★</span>. The exact count is **211,091 distinct usable word forms** after parser-compatible filtering and exact, case-preserved deduplication. Stars rank dictionary size relative to all 167 user-facing models in five nearly equal groups; they do **not** measure linguistic quality or benchmark accuracy.
+<!-- DICTIONARY-SIZE-RATING:END -->
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-09-11 Radixor/Java `4.3.0.dirty` snapshot. Speed benchmark operations process changed tokens. Accuracy uses the complete Radixor dictionary for the language.
 
 <!-- BENCHMARK-EVIDENCE-MAP:START -->
 !!! info "How to read this page"
@@ -13,9 +17,9 @@ Runtime and exact-root agreement measure different properties. Light, minimal, p
 
 ## Dictionary Corpus
 
-| Model ID | Model version | Language | Dictionary rows | Complete quality tokens | Already-root tokens | Changed tokens | JMH timing tokens |
-| --- | --- | --- | ---: | ---: | ---: | ---: | ---: |
-| `pt-pt-default` | `1.0.0` | `PT_PT` | 4,001 | 215,490 | 8,002 | 207,488 | 207,488 |
+| Model ID | Model version | Language | Dictionary rows | Distinct usable forms | Complete quality tokens | Already-root tokens | Changed tokens | Timing workload | JMH timing tokens |
+| --- | --- | --- | ---: | ---: | ---: | ---: | ---: | --- | ---: |
+| `pt-pt-default` | `1.0.0` | `PT_PT` | 4,001 | 211,091 | 215,490 | 8,002 | 207,488 | changed tokens | 207,488 |
 
 ## Radixor Patch Command Distribution
 
@@ -44,16 +48,16 @@ Accuracy is computed from JMH auxiliary counters in the current report. The coun
 
 ## Speed
 
-Speed uses JMH average time, 5 warmup iterations, 7 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
 
 | Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
 | --- | --- | ---: | ---: | ---: | ---: | --- |
-| Radixor | `portugueseRadixor` | 11.444 | 0.204 | 55.2 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
-| Lucene PortugueseLightStemFilter | `portugueseLucenePortugueseLightStemFilter` | 10.557 | 0.211 | 50.9 | 0.923 | Light Portuguese suffix stemmer. |
-| Lucene PortugueseMinimalStemFilter | `portugueseLucenePortugueseMinimalStemFilter` | 14.744 | 0.203 | 71.1 | 1.288 | Minimal Portuguese suffix reducer. |
-| Official Snowball direct | `snowballDirect[PORTUGUESE]` | 51.852 | 2.250 | 249.9 | 4.531 | Official Snowball generated Java stemmer; direct API. |
-| Lucene SnowballFilter | `luceneSnowballFilter[PORTUGUESE]` | 62.553 | 1.666 | 301.5 | 5.466 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
-| Lucene PortugueseStemFilter | `portugueseLucenePortugueseStemFilter` | 144.379 | 3.988 | 695.8 | 12.616 | Portuguese RSLP-style Lucene TokenFilter. |
+| Radixor | `portugueseRadixor` | 10.856 | 0.302 | 52.3 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
+| Lucene PortugueseLightStemFilter | `portugueseLucenePortugueseLightStemFilter` | 10.333 | 0.232 | 49.8 | 0.952 | Light Portuguese suffix stemmer. |
+| Lucene PortugueseMinimalStemFilter | `portugueseLucenePortugueseMinimalStemFilter` | 14.489 | 0.250 | 69.8 | 1.335 | Minimal Portuguese suffix reducer. |
+| Official Snowball direct | `snowballDirect[PORTUGUESE]` | 51.535 | 2.448 | 248.4 | 4.747 | Official Snowball generated Java stemmer; direct API. |
+| Lucene SnowballFilter | `luceneSnowballFilter[PORTUGUESE]` | 59.921 | 4.127 | 288.8 | 5.520 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
+| Lucene PortugueseStemFilter | `portugueseLucenePortugueseStemFilter` | 145.578 | 9.659 | 701.6 | 13.410 | Portuguese RSLP-style Lucene TokenFilter. |
 
 ## Interpretation Notes
 
@@ -91,18 +95,15 @@ also appeared in training. Parentheses show the observed split minimum–maximum
 
 ### Generalization conclusion
 
-- Median exactness on genuinely unseen changed forms moves from **90.053%**
-  at 10% training knowledge to **94.412%** at 90%, a measured
-  **+4.359 percentage-point** change for this dictionary.
-- Over the same endpoints, unseen all-form exactness changes by **+4.197 pp** and
-  preservation of unseen already-root forms changes by **+0.167 pp**. These separate
-  outcomes show whether the changed-form result coexists with preservation behavior.
+- Median exactness on genuinely unseen changed forms moves from **90.053%** at 10% training knowledge to **94.412%** at 90%, a measured **+4.359 percentage-point** change.
+- Unseen all-form exactness moves from **90.423%** at 10% training knowledge to **94.620%** at 90%, a measured **+4.197 percentage-point** change.
+- Preservation of unseen already-root forms moves from **99.833%** at 10% training knowledge to **100.000%** at 90%, a measured **+0.167 percentage-point** change.
 - The evidence establishes within-resource transfer across withheld dictionary families. It
   does not estimate unrelated domains, misspellings, arbitrary compounds, or external corpora.
 
 The complete ten-level table and split ranges remain in the
 [independent generalization report](../generalization.md); raw counters and provenance are in
-[`dictionary-generalization.csv`](../data/dictionary-generalization.csv). The
+[active machine-readable snapshot](../data/dictionary-generalization-2026-09-11.csv). The
 [frozen methodology](../reference/generalization-methodology.md) defines family-level
 splitting, unseen-surface leakage control, aggregation, and the limits of the claim.
 
@@ -195,7 +196,7 @@ The complete evidence is available in the [raw logical matrix](../data/edit-cost
 
 Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `PT_PT` using the complete validated stemming-quality result matrix. Every distinct surface form is one evaluated item and can belong to several dictionary groups. Two forms are a positive pair when their group-membership sets intersect and a negative pair when those sets are disjoint. A pair shared through several groups is counted once. Exact equality with a predetermined lemma is not required.
 
-`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality-2026-09-11.csv).
 
 ### Evaluation Scope and Key Findings
 
@@ -466,8 +467,8 @@ Standard ARI, homogeneity, completeness, V-measure, and NMI are not calculated:
 
 ### Provenance
 
-- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
-- Source SHA-256: `85763189eab4d0fbb047c2d5d3554c66abf9732182bd0d8fd758d7aef680e66f`
+- Authoritative source: `docs/benchmarks/data/stemming-quality-2026-09-11.csv`
+- Source SHA-256: `24bddfeed06a60bb3eeed58e1bfc93aed46c1d32e7bebd293aec2dacfddfef5b`
 - Evaluation command: `./gradlew stemmingQuality --no-daemon`
 - Dictionary language: `PT_PT`
 - Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
diff --git a/docs/benchmarks/languages/russian.md b/docs/benchmarks/languages/russian.md
index 83dbe9d..6adca5a 100644
--- a/docs/benchmarks/languages/russian.md
+++ b/docs/benchmarks/languages/russian.md
@@ -1,8 +1,12 @@
-# Russian Stemmer Benchmarks
+# Russian Stemmer Benchmarks <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 759,333 distinct usable word forms" title="Relative dictionary size 5 of 5; 759,333 distinct usable word forms">★★★★★</span>
 
 This page reports same-language stemming benchmarks for Russian. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
 
-All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-08-25 Radixor/Java `4.2.0-6-g84e57fb` snapshot. Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+<!-- DICTIONARY-SIZE-RATING:START -->
+Dictionary size: <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 759,333 distinct usable word forms" title="Relative dictionary size 5 of 5; 759,333 distinct usable word forms">★★★★★</span>. The exact count is **759,333 distinct usable word forms** after parser-compatible filtering and exact, case-preserved deduplication. Stars rank dictionary size relative to all 167 user-facing models in five nearly equal groups; they do **not** measure linguistic quality or benchmark accuracy.
+<!-- DICTIONARY-SIZE-RATING:END -->
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-09-11 Radixor/Java `4.3.0.dirty` snapshot. Speed benchmark operations process changed tokens. Accuracy uses the complete Radixor dictionary for the language.
 
 <!-- BENCHMARK-EVIDENCE-MAP:START -->
 !!! info "How to read this page"
@@ -13,9 +17,9 @@ Runtime and exact-root agreement measure different properties. Light, minimal, p
 
 ## Dictionary Corpus
 
-| Model ID | Model version | Language | Dictionary rows | Complete quality tokens | Already-root tokens | Changed tokens | JMH timing tokens |
-| --- | --- | --- | ---: | ---: | ---: | ---: | ---: |
-| `ru-ru-default` | `1.0.0` | `RU_RU` | 37,410 | 806,279 | 74,808 | 731,471 | 731,471 |
+| Model ID | Model version | Language | Dictionary rows | Distinct usable forms | Complete quality tokens | Already-root tokens | Changed tokens | Timing workload | JMH timing tokens |
+| --- | --- | --- | ---: | ---: | ---: | ---: | ---: | --- | ---: |
+| `ru-ru-default` | `1.0.0` | `RU_RU` | 37,410 | 759,333 | 806,279 | 74,808 | 731,471 | changed tokens | 731,471 |
 
 ## Radixor Patch Command Distribution
 
@@ -42,14 +46,14 @@ Accuracy is computed from JMH auxiliary counters in the current report. The coun
 
 ## Speed
 
-Speed uses JMH average time, 5 warmup iterations, 7 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
 
 | Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
 | --- | --- | ---: | ---: | ---: | ---: | --- |
-| Radixor | `russianRadixor` | 78.162 | 2.240 | 106.9 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
-| Lucene RussianLightStemFilter | `russianLuceneRussianLightStemFilter` | 57.502 | 2.358 | 78.6 | 0.736 | Light Russian suffix stemmer. |
-| Official Snowball direct | `snowballDirect[RUSSIAN]` | 107.685 | 11.649 | 147.2 | 1.378 | Official Snowball generated Java stemmer; direct API. |
-| Lucene SnowballFilter | `luceneSnowballFilter[RUSSIAN]` | 133.100 | 6.117 | 182.0 | 1.703 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
+| Radixor | `russianRadixor` | 75.341 | 2.119 | 103.0 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
+| Lucene RussianLightStemFilter | `russianLuceneRussianLightStemFilter` | 56.914 | 3.211 | 77.8 | 0.755 | Light Russian suffix stemmer. |
+| Official Snowball direct | `snowballDirect[RUSSIAN]` | 119.356 | 12.591 | 163.2 | 1.584 | Official Snowball generated Java stemmer; direct API. |
+| Lucene SnowballFilter | `luceneSnowballFilter[RUSSIAN]` | 141.677 | 16.761 | 193.7 | 1.880 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
 
 ## Interpretation Notes
 
@@ -87,18 +91,15 @@ also appeared in training. Parentheses show the observed split minimum–maximum
 
 ### Generalization conclusion
 
-- Median exactness on genuinely unseen changed forms moves from **74.948%**
-  at 10% training knowledge to **80.744%** at 90%, a measured
-  **+5.796 percentage-point** change for this dictionary.
-- Over the same endpoints, unseen all-form exactness changes by **+5.534 pp** and
-  preservation of unseen already-root forms changes by **+3.238 pp**. These separate
-  outcomes show whether the changed-form result coexists with preservation behavior.
+- Median exactness on genuinely unseen changed forms moves from **74.948%** at 10% training knowledge to **80.744%** at 90%, a measured **+5.796 percentage-point** change.
+- Unseen all-form exactness moves from **76.473%** at 10% training knowledge to **82.007%** at 90%, a measured **+5.534 percentage-point** change.
+- Preservation of unseen already-root forms moves from **91.268%** at 10% training knowledge to **94.506%** at 90%, a measured **+3.238 percentage-point** change.
 - The evidence establishes within-resource transfer across withheld dictionary families. It
   does not estimate unrelated domains, misspellings, arbitrary compounds, or external corpora.
 
 The complete ten-level table and split ranges remain in the
 [independent generalization report](../generalization.md); raw counters and provenance are in
-[`dictionary-generalization.csv`](../data/dictionary-generalization.csv). The
+[active machine-readable snapshot](../data/dictionary-generalization-2026-09-11.csv). The
 [frozen methodology](../reference/generalization-methodology.md) defines family-level
 splitting, unseen-surface leakage control, aggregation, and the limits of the claim.
 
@@ -191,7 +192,7 @@ The complete evidence is available in the [raw logical matrix](../data/edit-cost
 
 Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `RU_RU` using the complete validated stemming-quality result matrix. Every distinct surface form is one evaluated item and can belong to several dictionary groups. Two forms are a positive pair when their group-membership sets intersect and a negative pair when those sets are disjoint. A pair shared through several groups is counted once. Exact equality with a predetermined lemma is not required.
 
-`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality-2026-09-11.csv).
 
 ### Evaluation Scope and Key Findings
 
@@ -446,8 +447,8 @@ Standard ARI, homogeneity, completeness, V-measure, and NMI are not calculated:
 
 ### Provenance
 
-- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
-- Source SHA-256: `85763189eab4d0fbb047c2d5d3554c66abf9732182bd0d8fd758d7aef680e66f`
+- Authoritative source: `docs/benchmarks/data/stemming-quality-2026-09-11.csv`
+- Source SHA-256: `24bddfeed06a60bb3eeed58e1bfc93aed46c1d32e7bebd293aec2dacfddfef5b`
 - Evaluation command: `./gradlew stemmingQuality --no-daemon`
 - Dictionary language: `RU_RU`
 - Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
diff --git a/docs/benchmarks/languages/spanish.md b/docs/benchmarks/languages/spanish.md
index 4b80d68..c384913 100644
--- a/docs/benchmarks/languages/spanish.md
+++ b/docs/benchmarks/languages/spanish.md
@@ -1,8 +1,12 @@
-# Spanish Stemmer Benchmarks
+# Spanish Stemmer Benchmarks <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 849,661 distinct usable word forms" title="Relative dictionary size 5 of 5; 849,661 distinct usable word forms">★★★★★</span>
 
 This page reports same-language stemming benchmarks for Spanish. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
 
-All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-08-25 Radixor/Java `4.2.0-6-g84e57fb` snapshot. Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+<!-- DICTIONARY-SIZE-RATING:START -->
+Dictionary size: <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 849,661 distinct usable word forms" title="Relative dictionary size 5 of 5; 849,661 distinct usable word forms">★★★★★</span>. The exact count is **849,661 distinct usable word forms** after parser-compatible filtering and exact, case-preserved deduplication. Stars rank dictionary size relative to all 167 user-facing models in five nearly equal groups; they do **not** measure linguistic quality or benchmark accuracy.
+<!-- DICTIONARY-SIZE-RATING:END -->
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-09-11 Radixor/Java `4.3.0.dirty` snapshot. Speed benchmark operations process changed tokens. Accuracy uses the complete Radixor dictionary for the language.
 
 <!-- BENCHMARK-EVIDENCE-MAP:START -->
 !!! info "How to read this page"
@@ -13,9 +17,9 @@ Runtime and exact-root agreement measure different properties. Light, minimal, p
 
 ## Dictionary Corpus
 
-| Model ID | Model version | Language | Dictionary rows | Complete quality tokens | Already-root tokens | Changed tokens | JMH timing tokens |
-| --- | --- | --- | ---: | ---: | ---: | ---: | ---: |
-| `es-es-default` | `1.0.0` | `ES_ES` | 65,059 | 926,393 | 120,121 | 806,272 | 806,272 |
+| Model ID | Model version | Language | Dictionary rows | Distinct usable forms | Complete quality tokens | Already-root tokens | Changed tokens | Timing workload | JMH timing tokens |
+| --- | --- | --- | ---: | ---: | ---: | ---: | ---: | --- | ---: |
+| `es-es-default` | `1.0.0` | `ES_ES` | 65,059 | 849,661 | 926,393 | 120,121 | 806,272 | changed tokens | 806,272 |
 
 ## Radixor Patch Command Distribution
 
@@ -45,17 +49,17 @@ Accuracy is computed from JMH auxiliary counters in the current report. The coun
 
 ## Speed
 
-Speed uses JMH average time, 5 warmup iterations, 7 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
 
 | Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
 | --- | --- | ---: | ---: | ---: | ---: | --- |
-| Radixor | `spanishRadixor` | 70.807 | 1.533 | 87.8 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
-| Lucene HunspellStemFilter | `luceneHunspellStemFilter` | 1983.491 | 40.315 | 2460.1 | 28.013 | Benchmark-only Spanish Hunspell dictionary compared via Lucene HunspellStemFilter. |
-| Lucene SpanishMinimalStemFilter | `spanishLuceneSpanishMinimalStemFilter` | 39.667 | 1.296 | 49.2 | 0.560 | Minimal Spanish suffix reducer; narrow baseline. |
-| Lucene SpanishLightStemFilter | `spanishLuceneSpanishLightStemFilter` | 42.394 | 1.768 | 52.6 | 0.599 | Light Spanish suffix stemmer. |
-| Lucene SpanishPluralStemFilter | `spanishLuceneSpanishPluralStemFilter` | 93.092 | 4.240 | 115.5 | 1.315 | Plural-oriented Spanish suffix reducer. |
-| Official Snowball direct | `snowballDirect[SPANISH]` | 180.797 | 8.287 | 224.2 | 2.553 | Official Snowball generated Java stemmer; direct API. |
-| Lucene SnowballFilter | `luceneSnowballFilter[SPANISH]` | 201.614 | 17.917 | 250.1 | 2.847 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
+| Radixor | `spanishRadixor` | 64.311 | 0.840 | 79.8 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
+| Lucene HunspellStemFilter | `luceneHunspellStemFilter` | 2038.148 | 88.781 | 2527.9 | 31.692 | Benchmark-only Spanish Hunspell dictionary compared via Lucene HunspellStemFilter. |
+| Lucene SpanishMinimalStemFilter | `spanishLuceneSpanishMinimalStemFilter` | 40.128 | 2.577 | 49.8 | 0.624 | Minimal Spanish suffix reducer; narrow baseline. |
+| Lucene SpanishLightStemFilter | `spanishLuceneSpanishLightStemFilter` | 41.989 | 2.354 | 52.1 | 0.653 | Light Spanish suffix stemmer. |
+| Lucene SpanishPluralStemFilter | `spanishLuceneSpanishPluralStemFilter` | 93.070 | 6.243 | 115.4 | 1.447 | Plural-oriented Spanish suffix reducer. |
+| Official Snowball direct | `snowballDirect[SPANISH]` | 186.104 | 10.262 | 230.8 | 2.894 | Official Snowball generated Java stemmer; direct API. |
+| Lucene SnowballFilter | `luceneSnowballFilter[SPANISH]` | 206.679 | 23.811 | 256.3 | 3.214 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
 
 ## Interpretation Notes
 
@@ -93,18 +97,15 @@ also appeared in training. Parentheses show the observed split minimum–maximum
 
 ### Generalization conclusion
 
-- Median exactness on genuinely unseen changed forms moves from **80.995%**
-  at 10% training knowledge to **89.176%** at 90%, a measured
-  **+8.181 percentage-point** change for this dictionary.
-- Over the same endpoints, unseen all-form exactness changes by **+7.735 pp** and
-  preservation of unseen already-root forms changes by **+5.348 pp**. These separate
-  outcomes show whether the changed-form result coexists with preservation behavior.
+- Median exactness on genuinely unseen changed forms moves from **80.995%** at 10% training knowledge to **89.176%** at 90%, a measured **+8.181 percentage-point** change.
+- Unseen all-form exactness moves from **82.066%** at 10% training knowledge to **89.801%** at 90%, a measured **+7.735 percentage-point** change.
+- Preservation of unseen already-root forms moves from **89.364%** at 10% training knowledge to **94.711%** at 90%, a measured **+5.348 percentage-point** change.
 - The evidence establishes within-resource transfer across withheld dictionary families. It
   does not estimate unrelated domains, misspellings, arbitrary compounds, or external corpora.
 
 The complete ten-level table and split ranges remain in the
 [independent generalization report](../generalization.md); raw counters and provenance are in
-[`dictionary-generalization.csv`](../data/dictionary-generalization.csv). The
+[active machine-readable snapshot](../data/dictionary-generalization-2026-09-11.csv). The
 [frozen methodology](../reference/generalization-methodology.md) defines family-level
 splitting, unseen-surface leakage control, aggregation, and the limits of the claim.
 
@@ -197,7 +198,7 @@ The complete evidence is available in the [raw logical matrix](../data/edit-cost
 
 Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `ES_ES` using the complete validated stemming-quality result matrix. Every distinct surface form is one evaluated item and can belong to several dictionary groups. Two forms are a positive pair when their group-membership sets intersect and a negative pair when those sets are disjoint. A pair shared through several groups is counted once. Exact equality with a predetermined lemma is not required.
 
-`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality-2026-09-11.csv).
 
 ### Evaluation Scope and Key Findings
 
@@ -490,8 +491,8 @@ Standard ARI, homogeneity, completeness, V-measure, and NMI are not calculated:
 
 ### Provenance
 
-- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
-- Source SHA-256: `85763189eab4d0fbb047c2d5d3554c66abf9732182bd0d8fd758d7aef680e66f`
+- Authoritative source: `docs/benchmarks/data/stemming-quality-2026-09-11.csv`
+- Source SHA-256: `24bddfeed06a60bb3eeed58e1bfc93aed46c1d32e7bebd293aec2dacfddfef5b`
 - Evaluation command: `./gradlew stemmingQuality --no-daemon`
 - Dictionary language: `ES_ES`
 - Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
diff --git a/docs/benchmarks/languages/swedish.md b/docs/benchmarks/languages/swedish.md
index 5421b6d..75fbcf0 100644
--- a/docs/benchmarks/languages/swedish.md
+++ b/docs/benchmarks/languages/swedish.md
@@ -1,8 +1,12 @@
-# Swedish Stemmer Benchmarks
+# Swedish Stemmer Benchmarks <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 95,181 distinct usable word forms" title="Relative dictionary size 5 of 5; 95,181 distinct usable word forms">★★★★★</span>
 
 This page reports same-language stemming benchmarks for Swedish. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
 
-All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-08-25 Radixor/Java `4.2.0-6-g84e57fb` snapshot. Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+<!-- DICTIONARY-SIZE-RATING:START -->
+Dictionary size: <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 95,181 distinct usable word forms" title="Relative dictionary size 5 of 5; 95,181 distinct usable word forms">★★★★★</span>. The exact count is **95,181 distinct usable word forms** after parser-compatible filtering and exact, case-preserved deduplication. Stars rank dictionary size relative to all 167 user-facing models in five nearly equal groups; they do **not** measure linguistic quality or benchmark accuracy.
+<!-- DICTIONARY-SIZE-RATING:END -->
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-09-11 Radixor/Java `4.3.0.dirty` snapshot. Speed benchmark operations process changed tokens. Accuracy uses the complete Radixor dictionary for the language.
 
 <!-- BENCHMARK-EVIDENCE-MAP:START -->
 !!! info "How to read this page"
@@ -13,9 +17,9 @@ Runtime and exact-root agreement measure different properties. Light, minimal, p
 
 ## Dictionary Corpus
 
-| Model ID | Model version | Language | Dictionary rows | Complete quality tokens | Already-root tokens | Changed tokens | JMH timing tokens |
-| --- | --- | --- | ---: | ---: | ---: | ---: | ---: |
-| `sv-se-default` | `1.0.0` | `SV_SE` | 12,371 | 110,468 | 24,731 | 85,737 | 85,737 |
+| Model ID | Model version | Language | Dictionary rows | Distinct usable forms | Complete quality tokens | Already-root tokens | Changed tokens | Timing workload | JMH timing tokens |
+| --- | --- | --- | ---: | ---: | ---: | ---: | ---: | --- | ---: |
+| `sv-se-default` | `1.0.0` | `SV_SE` | 12,371 | 95,181 | 110,468 | 24,731 | 85,737 | changed tokens | 85,737 |
 
 ## Radixor Patch Command Distribution
 
@@ -43,15 +47,15 @@ Accuracy is computed from JMH auxiliary counters in the current report. The coun
 
 ## Speed
 
-Speed uses JMH average time, 5 warmup iterations, 7 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
 
 | Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
 | --- | --- | ---: | ---: | ---: | ---: | --- |
-| Radixor | `swedishRadixor` | 5.232 | 0.112 | 61.0 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
-| Lucene SwedishMinimalStemFilter | `swedishLuceneSwedishMinimalStemFilter` | 4.504 | 0.065 | 52.5 | 0.861 | Minimal Swedish suffix reducer. |
-| Lucene SwedishLightStemFilter | `swedishLuceneSwedishLightStemFilter` | 4.682 | 0.092 | 54.6 | 0.895 | Light Swedish suffix stemmer. |
-| Official Snowball direct | `snowballDirect[SWEDISH]` | 8.027 | 0.979 | 93.6 | 1.534 | Official Snowball generated Java stemmer; direct API. |
-| Lucene SnowballFilter | `luceneSnowballFilter[SWEDISH]` | 10.296 | 0.977 | 120.1 | 1.968 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
+| Radixor | `swedishRadixor` | 4.890 | 0.124 | 57.0 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
+| Lucene SwedishMinimalStemFilter | `swedishLuceneSwedishMinimalStemFilter` | 4.436 | 0.130 | 51.7 | 0.907 | Minimal Swedish suffix reducer. |
+| Lucene SwedishLightStemFilter | `swedishLuceneSwedishLightStemFilter` | 4.668 | 0.198 | 54.4 | 0.955 | Light Swedish suffix stemmer. |
+| Official Snowball direct | `snowballDirect[SWEDISH]` | 8.255 | 1.129 | 96.3 | 1.688 | Official Snowball generated Java stemmer; direct API. |
+| Lucene SnowballFilter | `luceneSnowballFilter[SWEDISH]` | 11.391 | 1.699 | 132.9 | 2.329 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
 
 ## Interpretation Notes
 
@@ -89,18 +93,15 @@ also appeared in training. Parentheses show the observed split minimum–maximum
 
 ### Generalization conclusion
 
-- Median exactness on genuinely unseen changed forms moves from **67.214%**
-  at 10% training knowledge to **78.458%** at 90%, a measured
-  **+11.244 percentage-point** change for this dictionary.
-- Over the same endpoints, unseen all-form exactness changes by **+10.177 pp** and
-  preservation of unseen already-root forms changes by **+7.137 pp**. These separate
-  outcomes show whether the changed-form result coexists with preservation behavior.
+- Median exactness on genuinely unseen changed forms moves from **67.214%** at 10% training knowledge to **78.458%** at 90%, a measured **+11.244 percentage-point** change.
+- Unseen all-form exactness moves from **72.010%** at 10% training knowledge to **82.188%** at 90%, a measured **+10.177 percentage-point** change.
+- Preservation of unseen already-root forms moves from **87.789%** at 10% training knowledge to **94.926%** at 90%, a measured **+7.137 percentage-point** change.
 - The evidence establishes within-resource transfer across withheld dictionary families. It
   does not estimate unrelated domains, misspellings, arbitrary compounds, or external corpora.
 
 The complete ten-level table and split ranges remain in the
 [independent generalization report](../generalization.md); raw counters and provenance are in
-[`dictionary-generalization.csv`](../data/dictionary-generalization.csv). The
+[active machine-readable snapshot](../data/dictionary-generalization-2026-09-11.csv). The
 [frozen methodology](../reference/generalization-methodology.md) defines family-level
 splitting, unseen-surface leakage control, aggregation, and the limits of the claim.
 
@@ -194,7 +195,7 @@ The complete evidence is available in the [raw logical matrix](../data/edit-cost
 
 Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `SV_SE` using the complete validated stemming-quality result matrix. Every distinct surface form is one evaluated item and can belong to several dictionary groups. Two forms are a positive pair when their group-membership sets intersect and a negative pair when those sets are disjoint. A pair shared through several groups is counted once. Exact equality with a predetermined lemma is not required.
 
-`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality-2026-09-11.csv).
 
 ### Evaluation Scope and Key Findings
 
@@ -457,8 +458,8 @@ Standard ARI, homogeneity, completeness, V-measure, and NMI are not calculated:
 
 ### Provenance
 
-- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
-- Source SHA-256: `85763189eab4d0fbb047c2d5d3554c66abf9732182bd0d8fd758d7aef680e66f`
+- Authoritative source: `docs/benchmarks/data/stemming-quality-2026-09-11.csv`
+- Source SHA-256: `24bddfeed06a60bb3eeed58e1bfc93aed46c1d32e7bebd293aec2dacfddfef5b`
 - Evaluation command: `./gradlew stemmingQuality --no-daemon`
 - Dictionary language: `SV_SE`
 - Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
diff --git a/docs/benchmarks/languages/ukrainian.md b/docs/benchmarks/languages/ukrainian.md
index b86fd23..7fc60e1 100644
--- a/docs/benchmarks/languages/ukrainian.md
+++ b/docs/benchmarks/languages/ukrainian.md
@@ -1,8 +1,12 @@
-# Ukrainian Stemmer Benchmarks
+# Ukrainian Stemmer Benchmarks <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 14,150 distinct usable word forms" title="Relative dictionary size 4 of 5; 14,150 distinct usable word forms">★★★★☆</span>
 
 This page reports same-language stemming benchmarks for Ukrainian. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
 
-All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-08-25 Radixor/Java `4.2.0-6-g84e57fb` snapshot. Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+<!-- DICTIONARY-SIZE-RATING:START -->
+Dictionary size: <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 14,150 distinct usable word forms" title="Relative dictionary size 4 of 5; 14,150 distinct usable word forms">★★★★☆</span>. The exact count is **14,150 distinct usable word forms** after parser-compatible filtering and exact, case-preserved deduplication. Stars rank dictionary size relative to all 167 user-facing models in five nearly equal groups; they do **not** measure linguistic quality or benchmark accuracy.
+<!-- DICTIONARY-SIZE-RATING:END -->
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-09-11 Radixor/Java `4.3.0.dirty` snapshot. Speed benchmark operations process changed tokens. Accuracy uses the complete Radixor dictionary for the language.
 
 <!-- BENCHMARK-EVIDENCE-MAP:START -->
 !!! info "How to read this page"
@@ -13,9 +17,9 @@ Runtime and exact-root agreement measure different properties. Light, minimal, p
 
 ## Dictionary Corpus
 
-| Model ID | Model version | Language | Dictionary rows | Complete quality tokens | Already-root tokens | Changed tokens | JMH timing tokens |
-| --- | --- | --- | ---: | ---: | ---: | ---: | ---: |
-| `uk-ua-default` | `1.0.0` | `UK_UA` | 1,493 | 15,737 | 2,985 | 12,752 | 12,752 |
+| Model ID | Model version | Language | Dictionary rows | Distinct usable forms | Complete quality tokens | Already-root tokens | Changed tokens | Timing workload | JMH timing tokens |
+| --- | --- | --- | ---: | ---: | ---: | ---: | ---: | --- | ---: |
+| `uk-ua-default` | `1.0.0` | `UK_UA` | 1,493 | 14,150 | 15,737 | 2,985 | 12,752 | changed tokens | 12,752 |
 
 ## Radixor Patch Command Distribution
 
@@ -42,14 +46,14 @@ Accuracy is computed from JMH auxiliary counters in the current report. The coun
 
 ## Speed
 
-Speed uses JMH average time, 5 warmup iterations, 7 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
 
 | Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
 | --- | --- | ---: | ---: | ---: | ---: | --- |
-| Radixor | `ukrainianRadixor` | 0.625 | 0.017 | 49.0 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
-| Lucene HunspellStemFilter | `luceneHunspellStemFilter` | 49.515 | 6.147 | 3883.0 | 79.197 | Benchmark-only Ukrainian Hunspell dictionary compared via Lucene HunspellStemFilter. |
-| Morfologik direct | `ukrainianMorfologikDirect` | 8.299 | 0.124 | 650.8 | 13.275 | Direct Morfologik dictionary lookup; first returned stem is used for quality. |
-| Lucene MorfologikFilter | `ukrainianLuceneMorfologikFilter` | 14.580 | 0.206 | 1143.4 | 23.320 | Dictionary-based Morfologik TokenFilter; may emit multiple terms. |
+| Radixor | `ukrainianRadixor` | 0.620 | 0.011 | 48.6 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
+| Lucene HunspellStemFilter | `luceneHunspellStemFilter` | 55.033 | 8.824 | 4315.7 | 88.803 | Benchmark-only Ukrainian Hunspell dictionary compared via Lucene HunspellStemFilter. |
+| Morfologik direct | `ukrainianMorfologikDirect` | 8.224 | 0.157 | 644.9 | 13.270 | Direct Morfologik dictionary lookup; first returned stem is used for quality. |
+| Lucene MorfologikFilter | `ukrainianLuceneMorfologikFilter` | 14.808 | 0.346 | 1161.2 | 23.895 | Dictionary-based Morfologik TokenFilter; may emit multiple terms. |
 
 ## Interpretation Notes
 
@@ -87,18 +91,15 @@ also appeared in training. Parentheses show the observed split minimum–maximum
 
 ### Generalization conclusion
 
-- Median exactness on genuinely unseen changed forms moves from **42.530%**
-  at 10% training knowledge to **57.218%** at 90%, a measured
-  **+14.689 percentage-point** change for this dictionary.
-- Over the same endpoints, unseen all-form exactness changes by **+12.251 pp** and
-  preservation of unseen already-root forms changes by **+5.012 pp**. These separate
-  outcomes show whether the changed-form result coexists with preservation behavior.
+- Median exactness on genuinely unseen changed forms moves from **42.530%** at 10% training knowledge to **57.218%** at 90%, a measured **+14.689 percentage-point** change.
+- Unseen all-form exactness moves from **50.982%** at 10% training knowledge to **63.233%** at 90%, a measured **+12.251 percentage-point** change.
+- Preservation of unseen already-root forms moves from **85.464%** at 10% training knowledge to **90.476%** at 90%, a measured **+5.012 percentage-point** change.
 - The evidence establishes within-resource transfer across withheld dictionary families. It
   does not estimate unrelated domains, misspellings, arbitrary compounds, or external corpora.
 
 The complete ten-level table and split ranges remain in the
 [independent generalization report](../generalization.md); raw counters and provenance are in
-[`dictionary-generalization.csv`](../data/dictionary-generalization.csv). The
+[active machine-readable snapshot](../data/dictionary-generalization-2026-09-11.csv). The
 [frozen methodology](../reference/generalization-methodology.md) defines family-level
 splitting, unseen-surface leakage control, aggregation, and the limits of the claim.
 
@@ -190,7 +191,7 @@ The complete evidence is available in the [raw logical matrix](../data/edit-cost
 
 Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `UK_UA` using the complete validated stemming-quality result matrix. Every distinct surface form is one evaluated item and can belong to several dictionary groups. Two forms are a positive pair when their group-membership sets intersect and a negative pair when those sets are disjoint. A pair shared through several groups is counted once. Exact equality with a predetermined lemma is not required.
 
-`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality-2026-09-11.csv).
 
 ### Evaluation Scope and Key Findings
 
@@ -487,8 +488,8 @@ Standard ARI, homogeneity, completeness, V-measure, and NMI are not calculated:
 
 ### Provenance
 
-- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
-- Source SHA-256: `85763189eab4d0fbb047c2d5d3554c66abf9732182bd0d8fd758d7aef680e66f`
+- Authoritative source: `docs/benchmarks/data/stemming-quality-2026-09-11.csv`
+- Source SHA-256: `24bddfeed06a60bb3eeed58e1bfc93aed46c1d32e7bebd293aec2dacfddfef5b`
 - Evaluation command: `./gradlew stemmingQuality --no-daemon`
 - Dictionary language: `UK_UA`
 - Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
diff --git a/docs/benchmarks/languages/yiddish.md b/docs/benchmarks/languages/yiddish.md
index 948c0d7..b4a61cc 100644
--- a/docs/benchmarks/languages/yiddish.md
+++ b/docs/benchmarks/languages/yiddish.md
@@ -1,8 +1,12 @@
-# Yiddish Stemmer Benchmarks
+# Yiddish Stemmer Benchmarks <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 3,532 distinct usable word forms" title="Relative dictionary size 3 of 5; 3,532 distinct usable word forms">★★★☆☆</span>
 
 This page reports same-language stemming benchmarks for Yiddish. Accuracy is listed first because speed without root agreement is not enough to interpret stemmer quality.
 
-All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-08-25 Radixor/Java `4.2.0-6-g84e57fb` snapshot. Speed benchmark operations process changed dictionary tokens only. Accuracy uses the complete Radixor dictionary for the language.
+<!-- DICTIONARY-SIZE-RATING:START -->
+Dictionary size: <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 3,532 distinct usable word forms" title="Relative dictionary size 3 of 5; 3,532 distinct usable word forms">★★★☆☆</span>. The exact count is **3,532 distinct usable word forms** after parser-compatible filtering and exact, case-preserved deduplication. Stars rank dictionary size relative to all 167 user-facing models in five nearly equal groups; they do **not** measure linguistic quality or benchmark accuracy.
+<!-- DICTIONARY-SIZE-RATING:END -->
+
+All speed values are environment-specific and were measured on the hardware and JVM listed in the [benchmark overview](../index.md). The command distribution, exact-root accuracy, and speed tables belong to the published 2026-09-11 Radixor/Java `4.3.0.dirty` snapshot. Speed benchmark operations process changed tokens. Accuracy uses the complete Radixor dictionary for the language.
 
 <!-- BENCHMARK-EVIDENCE-MAP:START -->
 !!! info "How to read this page"
@@ -13,9 +17,9 @@ Runtime and exact-root agreement measure different properties. Light, minimal, p
 
 ## Dictionary Corpus
 
-| Model ID | Model version | Language | Dictionary rows | Complete quality tokens | Already-root tokens | Changed tokens | JMH timing tokens |
-| --- | --- | --- | ---: | ---: | ---: | ---: | ---: |
-| `yi-default` | `1.0.0` | `YI` | 802 | 4,300 | 1,524 | 2,776 | 5,000 |
+| Model ID | Model version | Language | Dictionary rows | Distinct usable forms | Complete quality tokens | Already-root tokens | Changed tokens | Timing workload | JMH timing tokens |
+| --- | --- | --- | ---: | ---: | ---: | ---: | ---: | --- | ---: |
+| `yi-default` | `1.0.0` | `YI` | 802 | 3,532 | 4,300 | 1,524 | 2,776 | changed tokens | 5,000 |
 
 ## Radixor Patch Command Distribution
 
@@ -41,13 +45,13 @@ Accuracy is computed from JMH auxiliary counters in the current report. The coun
 
 ## Speed
 
-Speed uses JMH average time, 5 warmup iterations, 7 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 3 independent forks, and 1 thread. Relative factor is computed against the single Radixor row on this language page. Values below 1.000 are faster than that Radixor baseline; values above 1.000 are slower.
 
 | Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
 | --- | --- | ---: | ---: | ---: | ---: | --- |
-| Radixor | `radixor[YIDDISH]` | 0.214 | 0.002 | 42.8 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
-| Official Snowball direct | `snowballDirect[YIDDISH]` | 1.506 | 0.074 | 301.3 | 7.044 | Official Snowball generated Java stemmer; direct API. |
-| Lucene SnowballFilter | `luceneSnowballFilter[YIDDISH]` | 1.744 | 0.094 | 348.8 | 8.155 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
+| Radixor | `radixor[YIDDISH]` | 0.257 | 0.080 | 51.3 | 1.000 | Radixor dictionary-trained patch-command stemmer. |
+| Official Snowball direct | `snowballDirect[YIDDISH]` | 1.523 | 0.096 | 304.6 | 5.936 | Official Snowball generated Java stemmer; direct API. |
+| Lucene SnowballFilter | `luceneSnowballFilter[YIDDISH]` | 1.709 | 0.113 | 341.7 | 6.659 | Lucene TokenFilter path around Snowball; includes TokenStream overhead. |
 
 ## Interpretation Notes
 
@@ -85,18 +89,15 @@ also appeared in training. Parentheses show the observed split minimum–maximum
 
 ### Generalization conclusion
 
-- Median exactness on genuinely unseen changed forms moves from **37.767%**
-  at 10% training knowledge to **45.318%** at 90%, a measured
-  **+7.551 percentage-point** change for this dictionary.
-- Over the same endpoints, unseen all-form exactness changes by **+4.158 pp** and
-  preservation of unseen already-root forms changes by **+3.541 pp**. These separate
-  outcomes show whether the changed-form result coexists with preservation behavior.
+- Median exactness on genuinely unseen changed forms moves from **37.767%** at 10% training knowledge to **45.318%** at 90%, a measured **+7.551 percentage-point** change.
+- Unseen all-form exactness moves from **56.133%** at 10% training knowledge to **60.291%** at 90%, a measured **+4.158 percentage-point** change.
+- Preservation of unseen already-root forms moves from **89.466%** at 10% training knowledge to **93.007%** at 90%, a measured **+3.541 percentage-point** change.
 - The evidence establishes within-resource transfer across withheld dictionary families. It
   does not estimate unrelated domains, misspellings, arbitrary compounds, or external corpora.
 
 The complete ten-level table and split ranges remain in the
 [independent generalization report](../generalization.md); raw counters and provenance are in
-[`dictionary-generalization.csv`](../data/dictionary-generalization.csv). The
+[active machine-readable snapshot](../data/dictionary-generalization-2026-09-11.csv). The
 [frozen methodology](../reference/generalization-methodology.md) defines family-level
 splitting, unseen-surface leakage control, aggregation, and the limits of the claim.
 
@@ -189,7 +190,7 @@ The complete evidence is available in the [raw logical matrix](../data/edit-cost
 
 Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `YI` using the complete validated stemming-quality result matrix. Every distinct surface form is one evaluated item and can belong to several dictionary groups. Two forms are a positive pair when their group-membership sets intersect and a negative pair when those sets are disjoint. A pair shared through several groups is counted once. Exact equality with a predetermined lemma is not required.
 
-`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).
+`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality-2026-09-11.csv).
 
 ### Evaluation Scope and Key Findings
 
@@ -436,8 +437,8 @@ Standard ARI, homogeneity, completeness, V-measure, and NMI are not calculated:
 
 ### Provenance
 
-- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`
-- Source SHA-256: `85763189eab4d0fbb047c2d5d3554c66abf9732182bd0d8fd758d7aef680e66f`
+- Authoritative source: `docs/benchmarks/data/stemming-quality-2026-09-11.csv`
+- Source SHA-256: `24bddfeed06a60bb3eeed58e1bfc93aed46c1d32e7bebd293aec2dacfddfef5b`
 - Evaluation command: `./gradlew stemmingQuality --no-daemon`
 - Dictionary language: `YI`
 - Processing modes: `ALL_WORDS`, `LOWERCASE_GROUPS_ONLY`
diff --git a/docs/benchmarks/reference/corpora.md b/docs/benchmarks/reference/corpora.md
index 669bc15..98295a3 100644
--- a/docs/benchmarks/reference/corpora.md
+++ b/docs/benchmarks/reference/corpora.md
@@ -1,28 +1,175 @@
 # Benchmark Corpora
 
-The table below describes the Radixor resources used to build speed and quality corpora. `Total tokens` is the complete dictionary token count used by quality benchmarks. `Already-root tokens` counts fields where the token is already equal to the line root. `Changed tokens` is the speed workload before the minimum-size repeat rule.
+The table below describes the Radixor resources used to build speed and quality corpora. `Total benchmark tokens` is the complete dictionary token count used by quality benchmarks. `Changed benchmark tokens` counts fields whose token differs from the line root. Timing normally uses that changed population; a root-only model instead uses all root-preservation tokens. `Timing workload` makes this choice explicit.
 
-| Default model ID | Version | SHA-256 | Language | Dictionary rows | Total tokens | Already-root tokens | Changed tokens | Speed timing tokens |
-| --- | --- | --- | --- | ---: | ---: | ---: | ---: | ---: |
-| `cs-cz-default` | `1.0.0` | `62afdaa6dc7a721b54a0dc278a0c648a63ad52a34a412d27b5b52fbcde9c1ce4` | `CS_CZ` | 5,113 | 56,612 | 10,049 | 46,563 | 46,563 |
-| `da-dk-default` | `1.0.0` | `3f7b670a0e7b872bda0381f5154ce058a4656297b39b7157b4ccf6560257cb90` | `DA_DK` | 4,179 | 32,256 | 8,356 | 23,900 | 23,900 |
-| `nl-nl-default` | `1.0.0` | `c098034adc42da2ca3e419160e6dd2c2b3868f8af334303b3a191e09caadaf5e` | `NL_NL` | 4,992 | 31,466 | 9,981 | 21,485 | 21,485 |
-| `us-uk-default` | `1.0.1` | `0a3ec1fddd5ddb859f5ffa8003efe9200a5ef6203df1fddaf53a58003f33582c` | `US_UK` | 396,939 | 1,002,414 | 793,874 | 208,540 | 208,540 |
-| `fi-fi-default` | `1.0.0` | `ca2628b3db31fee92f1b612ebbbd5e956a6dbbfb10e721325e55ef528f26072f` | `FI_FI` | 57,027 | 1,865,215 | 110,525 | 1,754,690 | 1,754,690 |
-| `fr-fr-default` | `1.0.0` | `a988658758952fd599dc7360e0234178a6d65ac46e5cedc7dcd325a7cb7e71d9` | `FR_FR` | 59,240 | 474,110 | 108,141 | 365,969 | 365,969 |
-| `de-de-default` | `1.0.0` | `cbfa038122823f02e4bdb54b0035492c356b6ecd80f11eb11290d7a7248a59f5` | `DE_DE` | 54,092 | 333,036 | 90,535 | 242,501 | 242,501 |
-| `he-il-default` | `1.0.0` | `9a47dc69bb7dab21aba0266b73cd74cdaeb17db94363796a0a56111ac8518256` | `HE_IL` | 2,358 | 61,071 | 4,715 | 56,356 | 56,356 |
-| `hu-hu-default` | `1.0.0` | `359d46a01d751ec823705ad7f3dd1cc8f6663feb1a9d13cb04d0c6fb51ab646e` | `HU_HU` | 19,406 | 935,713 | 38,775 | 896,938 | 896,938 |
-| `it-it-default` | `1.0.0` | `5e03be31c9761e30dbf24a47a5ced3d6ec949dabd31e92632fdd9f7c67fc2e12` | `IT_IT` | 10,009 | 337,546 | 20,004 | 317,542 | 317,542 |
-| `nb-no-default` | `1.0.0` | `f495bffb44e79d27993e6e2e65d4b1204b29365dc93f481b2d8b96766fc90fd9` | `NB_NO` | 17,929 | 90,757 | 33,376 | 57,381 | 57,381 |
-| `nn-no-default` | `1.0.0` | `900cf2005605aea2a3d8d731ec0b0c1f47fb4469b4ba6b9134145d4d026a0398` | `NN_NO` | 4,688 | 19,651 | 6,089 | 13,562 | 13,562 |
-| `fa-ir-default` | `1.0.0` | `b29a0d168a6a97f980666aa40b74a0edd8b6be4ab3320a7abfbb76b3529f4ea1` | `FA_IR` | 69 | 3,770 | 138 | 3,632 | 5,000 |
-| `pl-pl-unimorph` | `1.0.0` | `8191ed727097839cc808cbc5c56a1bd78b3c851e7733ad226ad9a51519a54721` | `PL_PL` | 9,990 | 132,308 | 19,957 | 112,351 | 112,351 |
-| `pt-pt-default` | `1.0.0` | `7a035ff330a6f0548f446cd0d6617bc1cf4751292125a3564d3a255c5d6f516d` | `PT_PT` | 4,001 | 215,490 | 8,002 | 207,488 | 207,488 |
-| `ru-ru-default` | `1.0.0` | `df7ea25e63a875eeec7a4185be685bd5372a3c568db85c34c44fdf5d8d980a40` | `RU_RU` | 37,410 | 806,279 | 74,808 | 731,471 | 731,471 |
-| `es-es-default` | `1.0.0` | `7a1ec94cfdb1e9a95431289d62dc5579cb2a532d99532eeda90290072e569721` | `ES_ES` | 65,059 | 926,393 | 120,121 | 806,272 | 806,272 |
-| `sv-se-default` | `1.0.0` | `d9be72e3d67c776622c4281e04e4063b9381e8f84a823d98ebf08888b82dff0c` | `SV_SE` | 12,371 | 110,468 | 24,731 | 85,737 | 85,737 |
-| `uk-ua-default` | `1.0.0` | `cf3f612cfff16cb7763f99c55851069489b883c3bdd1a6576cd8c57a97e07eae` | `UK_UA` | 1,493 | 15,737 | 2,985 | 12,752 | 12,752 |
-| `yi-default` | `1.0.0` | `f47de665c27dcd72833a82904e49c68a945bb5aca769a7ec5a0164e2c981a6d3` | `YI` | 802 | 4,300 | 1,524 | 2,776 | 5,000 |
+| Model ID | Role | Version | SHA-256 | Language | Dictionary rows | Distinct usable forms | Total benchmark tokens | Changed benchmark tokens | Timing workload | Speed timing tokens |
+| --- | --- | --- | --- | --- | ---: | ---: | ---: | ---: | --- | ---: |
+| `ady-default` | `standalone` | `1.0.0` | `952fbbe1fc0a4f58e110d5ebed3848939c31b6f414199c0658398f369a84ba3d` | `ADY` | 1,634 | 20,334 | 21,988 | 18,720 | changed tokens | 18,720 |
+| `af-za-default` | `standalone` | `1.0.0` | `2360e30adf722e9a8a6fd428427ce650ac6bd883c8f186e27c4f2c72d09400d9` | `AF_ZA` | 166,498 | 289,303 | 460,165 | 127,169 | changed tokens | 127,169 |
+| `afb-default` | `standalone` | `1.0.0` | `8ad86d7a99e8c0a42c1f799f9b20caa5367e6b6504ef319a580a87b4533a78a9` | `AFB` | 198 | 355 | 553 | 157 | changed tokens | 5,000 |
+| `ail-default` | `standalone` | `1.0.0` | `364ae64f3c6df69cc6ed22d892762ffe7c875651b08cfef95abfbb388e0ebacd` | `AIL` | 622 | 2,978 | 3,672 | 2,428 | changed tokens | 5,000 |
+| `aka-default` | `standalone` | `1.0.0` | `364fbdf2e356f1c5474d8ff2c83666bd6b8eb5193983f53437d2e3c556fe3c02` | `AK` | 79 | 1,890 | 1,975 | 1,817 | changed tokens | 5,000 |
+| `am-et-default` | `standalone` | `1.0.0` | `ff154b83f4e30c28811b0d05933d455c40d45e2264b985ca90fcb0ad655f2b4c` | `AM_ET` | 2,461 | 41,308 | 44,923 | 40,001 | changed tokens | 40,001 |
+| `ame-default` | `standalone` | `1.0.0` | `4876a2f73d99478d1c7c7572729e817738246d2f3ace4f51f0284bda82430dc0` | `AME` | 212 | 1,489 | 1,711 | 1,287 | changed tokens | 5,000 |
+| `ang-default` | `standalone` | `1.0.0` | `ed45aecbcb8e394df8739e6f48e9333383c9d542f900b2ad3e3c8c3d7dc4d996` | `ANG` | 7,135 | 65,100 | 74,638 | 60,368 | changed tokens | 60,368 |
+| `ar-default` | `standalone` | `1.0.1` | `36fe6a7a80e922f423e80761e27f21c61d86cc7f1b2837a1989761af4d51c904` | `AR` | 12,815 | 452,974 | 460,343 | 447,528 | changed tokens | 447,528 |
+| `arn-default` | `standalone` | `1.0.0` | `aa1472b03d2cf04094814484e2d57a908c95b761c5c2ea6955623536bbe84077` | `ARN` | 26 | 548 | 574 | 522 | changed tokens | 5,000 |
+| `arz-default` | `standalone` | `1.0.0` | `a8024efa9fb52a91209a97a17589ae3b52e2b5547de98c6cc5e2cb82a37b58eb` | `ARZ` | 105 | 135 | 241 | 31 | changed tokens | 5,000 |
+| `as-in-default` | `standalone` | `1.0.0` | `5211ee56f3d91029bf9d7e9626aaf56cec45d163ab51086250883ac3cde75c11` | `AS_IN` | 81 | 328 | 409 | 247 | changed tokens | 5,000 |
+| `ast-default` | `standalone` | `1.0.0` | `263bca1a264ad6a2b00616fcfaa064f9e43e6a433224220af884340b5ad0b86d` | `AST` | 436 | 21,584 | 22,256 | 21,384 | changed tokens | 21,384 |
+| `aym-default` | `standalone` | `1.0.0` | `2fd438798d8ceac034af4aa2eb6a52e27ded6d84b84117d2476906fc8cf79400` | `AYM` | 2,482 | 231,247 | 234,796 | 229,832 | changed tokens | 229,832 |
+| `az-az-default` | `standalone` | `1.0.0` | `5327f8dde212ac34c3c4573cdd2fb6b21f908cd7c0cdf919e400b6f59c36abb9` | `AZ_AZ` | 340 | 6,669 | 7,011 | 6,331 | changed tokens | 6,331 |
+| `azg-default` | `standalone` | `1.0.0` | `43594f9d451d5e7ee630895efe7b452dc54fec3d98c610392faed1a8ed1b16fe` | `AZG` | 258 | 9,357 | 9,685 | 9,169 | changed tokens | 9,169 |
+| `bak-default` | `standalone` | `1.0.0` | `46b76859f6bbbcaa9bc22ff228bbadb9d0149048c17a47f415c7efb37d6ad1f8` | `BAK` | 1,048 | 10,616 | 11,683 | 9,587 | changed tokens | 9,587 |
+| `be-by-default` | `standalone` | `1.0.0` | `52ccbdd9a5105910ac06fc279d02d184a21e5a27a7ddbb84e5a3d25df646bf86` | `BE_BY` | 196 | 678 | 876 | 484 | changed tokens | 5,000 |
+| `bg-bg-default` | `standalone` | `1.0.0` | `fcc4bb4e765a2a2fdf2c8631e9f065bd0b7f269e60beac68f101dea1bc91975e` | `BG_BG` | 2,412 | 37,335 | 39,868 | 35,044 | changed tokens | 35,044 |
+| `bn-bd-default` | `standalone` | `1.0.0` | `dcaff09bdd0a1c10bf6da6c51b7608edc3021380ebbfe3707cf5284002d8f880` | `BN_BD` | 3 | 3 | 6 | 0 | root-only tokens | 5,000 |
+| `bra-default` | `standalone` | `1.0.0` | `46d9f30d4509fddb9a9c91b4de97f4eca2348d7191ee99261af3183808baef3c` | `BRA` | 88 | 115 | 204 | 28 | changed tokens | 5,000 |
+| `bre-default` | `standalone` | `1.0.0` | `184e58faab57319b5877bb2d6057cfb87aa0f2b6c17b3c1fdb2ce2c4e76e37d6` | `BRE` | 42 | 1,682 | 1,724 | 1,640 | changed tokens | 5,000 |
+| `ca-es-default` | `standalone` | `1.0.0` | `e8736229250fedd586a1651a730ea99036f7adb8a7f3e85ed6234af742dcff6d` | `CA_ES` | 1,480 | 61,495 | 63,314 | 60,354 | changed tokens | 60,354 |
+| `ceb-default` | `standalone` | `1.0.0` | `12e337c1d22c7efec4012e6473516973c972f9c4a0d5fd8d1a2c1312ade45ea9` | `CEB` | 90 | 392 | 482 | 302 | changed tokens | 5,000 |
+| `chu-default` | `standalone` | `1.0.0` | `064d51bafb5ebf7fdbb1fbdf3a5f471dde7457164a4f680ae6448c1b2207baa0` | `CHU` | 150 | 1,632 | 2,142 | 1,842 | changed tokens | 5,000 |
+| `ckt-default` | `standalone` | `1.0.0` | `1315617dbf5ba569eea7a05e6413052754fb929b18bed329c0a577cc7c837ca2` | `CKT` | 191 | 302 | 495 | 113 | changed tokens | 5,000 |
+| `cly-default` | `standalone` | `1.0.0` | `5fb7fa5ea1061e47ca22e37119f3316bac9608ad0b3c4390e41c5c1c83fd3f66` | `CLY` | 185 | 2,030 | 2,431 | 2,061 | changed tokens | 5,000 |
+| `cni-default` | `standalone` | `1.0.0` | `e793997cce34ccfa1786088d89421128857b3c9d8b50a5643d8def8533183c91` | `CNI` | 407 | 10,095 | 10,536 | 9,722 | changed tokens | 9,722 |
+| `cor-default` | `standalone` | `1.0.0` | `c6c88a61c66ce8c846ff92180e2a469d5a3dbee36897ef85ecbe5ad1480ec575` | `COR` | 8 | 159 | 177 | 161 | changed tokens | 5,000 |
+| `cpa-default` | `standalone` | `1.0.0` | `c52b1d26c06a11eb82192ad374df94666317a67b465ee5e0e429df61e0af1d24` | `CPA` | 555 | 1,751 | 2,535 | 1,425 | changed tokens | 5,000 |
+| `cre-default` | `standalone` | `1.0.0` | `2a858e3097c89e458a36661e39b5cca5d73db3f421e50df8cdfdc7609344b030` | `CRE` | 27 | 514 | 541 | 487 | changed tokens | 5,000 |
+| `crh-default` | `standalone` | `1.0.0` | `828cdc3a52e98f0a93739e57a1400e880bdb22e564bda7cfc51aac5dcf418fd8` | `CRH` | 1,208 | 7,183 | 8,395 | 5,979 | changed tokens | 5,979 |
+| `cs-cz-default` | `default` | `1.0.0` | `62afdaa6dc7a721b54a0dc278a0c648a63ad52a34a412d27b5b52fbcde9c1ce4` | `CS_CZ` | 5,113 | 51,401 | 56,612 | 46,563 | changed tokens | 46,563 |
+| `csb-default` | `standalone` | `1.0.0` | `ef7b728f0963360c1e6fd7fdbc8c44f87c31d0ad95e424f127db10936a381bd5` | `CSB` | 37 | 350 | 392 | 318 | changed tokens | 5,000 |
+| `ctp-default` | `standalone` | `1.0.0` | `a8cc8316dc040985e757cd8eb47894dc72128dd3fdd30a2b629b54a26f3239fe` | `CTP` | 217 | 2,456 | 2,786 | 2,352 | changed tokens | 5,000 |
+| `cy-gb-default` | `standalone` | `1.0.0` | `394475eedc3c0fe26d395d06821efa2e8ff0c2ea4b5c63b72bbe3c8a4efd3d5d` | `CY_GB` | 183 | 6,240 | 6,425 | 6,059 | changed tokens | 6,059 |
+| `czn-default` | `standalone` | `1.0.0` | `30e6473c472452d4105c99e162142949a2c1ff975ca35035a5ee439c977f07ba` | `CZN` | 39 | 68 | 108 | 30 | changed tokens | 5,000 |
+| `da-dk-default` | `default` | `1.0.0` | `3f7b670a0e7b872bda0381f5154ce058a4656297b39b7157b4ccf6560257cb90` | `DA_DK` | 4,179 | 27,921 | 32,256 | 23,900 | changed tokens | 23,900 |
+| `dak-default` | `standalone` | `1.0.0` | `5cc5ea06b28df6b18aca534fe06701fac953023f34217aae5df6452c335af53f` | `DAK` | 419 | 2,942 | 3,386 | 2,548 | changed tokens | 5,000 |
+| `de-de-default` | `default` | `1.0.0` | `cbfa038122823f02e4bdb54b0035492c356b6ecd80f11eb11290d7a7248a59f5` | `DE_DE` | 54,092 | 277,266 | 333,036 | 242,501 | changed tokens | 242,501 |
+| `dje-default` | `standalone` | `1.0.0` | `1efb975e9f6ab2c58583eaf573a4a3f150a2d6a1ffb9bbfeb2a224e0d51f205d` | `DJE` | 25 | 75 | 100 | 50 | changed tokens | 5,000 |
+| `dsb-default` | `standalone` | `1.0.0` | `c251bbd00618946b5d40c49605a2b6d3fe305019c691d2859d0dad19910237e8` | `DSB` | 990 | 12,038 | 13,185 | 11,205 | changed tokens | 11,205 |
+| `el-gr-default` | `standalone` | `1.0.0` | `995d9f80ae3e4810447696407d01e9b6b10b7d5088d123df8b8175cf616974aa` | `EL_GR` | 11,792 | 76,851 | 91,268 | 67,684 | changed tokens | 67,684 |
+| `es-es-default` | `default` | `1.0.0` | `7a1ec94cfdb1e9a95431289d62dc5579cb2a532d99532eeda90290072e569721` | `ES_ES` | 65,059 | 849,661 | 926,393 | 806,272 | changed tokens | 806,272 |
+| `et-ee-default` | `standalone` | `1.0.0` | `3179b5135c9efb4fd83d1f4840de337c5b8d8e330243f4b69c0289cdec435fdf` | `ET_EE` | 886 | 24,811 | 25,776 | 24,004 | changed tokens | 24,004 |
+| `eus-default` | `standalone` | `1.0.0` | `29ea20d68e085f9403c32330259f02599f6c510f47db0475f1e3ea73a276a98b` | `EUS` | 26 | 10,401 | 11,535 | 11,483 | changed tokens | 11,483 |
+| `evn-default` | `standalone` | `1.0.0` | `bf3dd82443f2dc033e553fe9685cc7d9df8806fcbf86b724953f3b7c6d74b258` | `EVN` | 4,072 | 11,965 | 16,135 | 7,991 | changed tokens | 7,991 |
+| `fa-ir-default` | `default` | `1.0.0` | `b29a0d168a6a97f980666aa40b74a0edd8b6be4ab3320a7abfbb76b3529f4ea1` | `FA_IR` | 69 | 3,544 | 3,770 | 3,632 | changed tokens | 5,000 |
+| `fi-fi-default` | `default` | `1.0.0` | `ca2628b3db31fee92f1b612ebbbd5e956a6dbbfb10e721325e55ef528f26072f` | `FI_FI` | 57,027 | 1,788,784 | 1,865,215 | 1,754,690 | changed tokens | 1,754,690 |
+| `fo-fo-default` | `standalone` | `1.0.0` | `7d52c5ef7a69d8a4b3f5fd4725f8c2f75e1e978f021351df1b67773cc58ebb13` | `FO_FO` | 3,077 | 31,366 | 36,410 | 30,256 | changed tokens | 30,256 |
+| `fr-fr-default` | `default` | `1.0.0` | `a988658758952fd599dc7360e0234178a6d65ac46e5cedc7dcd325a7cb7e71d9` | `FR_FR` | 59,240 | 404,011 | 474,110 | 365,969 | changed tokens | 365,969 |
+| `frm-default` | `standalone` | `1.0.0` | `ebe1ce025d74d7cf5eee3ce328840911d01f227ff6834fd039461763c9f3ac18` | `FRM` | 602 | 27,002 | 27,816 | 26,612 | changed tokens | 26,612 |
+| `fro-default` | `standalone` | `1.0.0` | `58e05dd60f4c616db2c28e34394171c6015c07de2d28bfeae7cfd2c60d7393d8` | `FRO` | 1,699 | 93,998 | 106,654 | 103,256 | changed tokens | 103,256 |
+| `frr-default` | `standalone` | `1.0.0` | `8ad47d2806f6ef2926e2270389609b3fc77b696ee1f09df6c1c395a06d9b64bb` | `FRR` | 51 | 340 | 408 | 306 | changed tokens | 5,000 |
+| `fur-default` | `standalone` | `1.0.0` | `ee6eb34acd2db625e91282629040da78f2a2f596ef34d12686acf76b4d3d9f68` | `FUR` | 168 | 5,007 | 5,185 | 4,849 | changed tokens | 5,000 |
+| `fy-nl-default` | `standalone` | `1.0.0` | `24dbc346fc258d7315048b02d24e8cc3675739e6563613695dac3c548fcfab68` | `FY_NL` | 85 | 696 | 784 | 614 | changed tokens | 5,000 |
+| `ga-ie-default` | `standalone` | `1.0.0` | `cc4ffc97b026107936fee34938b02273330ec23e61a1c9835cb3fd38b5319702` | `GA_IE` | 7,023 | 21,829 | 29,342 | 15,296 | changed tokens | 15,296 |
+| `gaa-default` | `standalone` | `1.0.0` | `052af2955659a954f1c3036e564b405ec75d93fa858e96c5cb8f2ce8bac8a534` | `GAA` | 94 | 469 | 564 | 376 | changed tokens | 5,000 |
+| `gal-default` | `standalone` | `1.0.0` | `91fcd0fe4215d310749795879d4d441fa23b12f4bcbbc3734ae83565c9c8b3c4` | `GAL` | 486 | 25,436 | 26,681 | 25,709 | changed tokens | 25,709 |
+| `gd-gb-default` | `standalone` | `1.0.0` | `089c9cb590ef564fee033fcaa59c50b7317aaed5e55ebd5493679a60b6458c33` | `GD_GB` | 70 | 265 | 335 | 195 | changed tokens | 5,000 |
+| `gmh-default` | `standalone` | `1.0.0` | `91ccdb971ed6905cf710362b208537c12e7aed0b39fdf2e881cef7e3114cd109` | `GMH` | 29 | 384 | 413 | 355 | changed tokens | 5,000 |
+| `gml-default` | `standalone` | `1.0.0` | `7a6d0ce9b18d3a0aa6cf8c6d4179d8473ba86c9d8b36133eb2370dfd4d257aec` | `GML` | 50 | 565 | 624 | 524 | changed tokens | 5,000 |
+| `goh-default` | `standalone` | `1.0.0` | `7dc85288accc2a0b66298254e461a7b5328e6443e8334e6e5ad0746d62d00e6b` | `GOH` | 482 | 5,267 | 5,782 | 4,818 | changed tokens | 5,000 |
+| `got-default` | `standalone` | `1.0.0` | `1e68dbfe6c47af09142952282110d5f953c416d34e54185ddc4459c16850eb09` | `GOT` | 6,225 | 134,162 | 143,969 | 131,519 | changed tokens | 131,519 |
+| `grc-default` | `standalone` | `1.0.0` | `ea83c5b51c7795f6c8e628ec6cac95578948175726c51d71710e751940f11aca` | `GRC` | 1,909 | 11,109 | 13,092 | 9,274 | changed tokens | 9,274 |
+| `gsw-default` | `standalone` | `1.0.0` | `770d3af9a6c62efb3c90bf8654066b262491d8d957e7bdb230a1b04c7a21f17f` | `GSW` | 143 | 1,171 | 1,318 | 1,032 | changed tokens | 5,000 |
+| `gup-default` | `standalone` | `1.0.0` | `c9dfd43c26952aad6abd283b8ec2982d2d71763d92e85b9db14542126b3b7876` | `GUP` | 73 | 370 | 444 | 298 | changed tokens | 5,000 |
+| `gv-im-default` | `standalone` | `1.0.0` | `3f51e5929a22c339dc21d309f0b7b8f73b2e3f95207af8514f591c50d37e9e15` | `GV_IM` | 1 | 15 | 16 | 14 | changed tokens | 5,000 |
+| `hai-default` | `standalone` | `1.0.0` | `27de8438d05cc3c6f86087145212bbfbf044041f155e973c0ef1aad49baf2cfe` | `HAI` | 28 | 520 | 548 | 492 | changed tokens | 5,000 |
+| `hbs-default` | `standalone` | `1.0.0` | `edeb7fff29f64fbe76b2087818d4838d36f4606857acbadc577a28774419f213` | `HBS` | 24,382 | 272,515 | 302,286 | 253,522 | changed tokens | 253,522 |
+| `he-il-default` | `default` | `1.0.0` | `9a47dc69bb7dab21aba0266b73cd74cdaeb17db94363796a0a56111ac8518256` | `HE_IL` | 2,358 | 57,658 | 61,071 | 56,356 | changed tokens | 56,356 |
+| `hil-default` | `standalone` | `1.0.0` | `43ccfab3a54e725d44a1c0c03154dcb474ffa33a716e6fd92ab5f59ba5d57d5c` | `HIL` | 80 | 308 | 388 | 228 | changed tokens | 5,000 |
+| `hsb-default` | `standalone` | `1.0.0` | `dd085232541899920df2acfcb11ef8fd914fea6894351f08a1e99f8fb84c3950` | `HSB` | 343 | 662 | 1,006 | 320 | changed tokens | 5,000 |
+| `hsi-default` | `standalone` | `1.0.0` | `40099c2852ee4407f1da7e2ebc9a396765d51cfeee6868a83c04abe169d1b057` | `HSI` | 49 | 158 | 208 | 110 | changed tokens | 5,000 |
+| `hu-hu-default` | `default` | `1.0.0` | `359d46a01d751ec823705ad7f3dd1cc8f6663feb1a9d13cb04d0c6fb51ab646e` | `HU_HU` | 19,406 | 910,688 | 935,713 | 896,938 | changed tokens | 896,938 |
+| `hy-am-default` | `standalone` | `1.0.0` | `dd2cef8bd372c2fc32a3127f0fa5ba7a4829c59e54f508b14ffd054e04fc1c00` | `HY_AM` | 6,946 | 242,091 | 250,040 | 236,148 | changed tokens | 236,148 |
+| `id-id-default` | `standalone` | `1.0.0` | `dc285b60aaa4e149f3d77fc93fede8bb6d20bae599b01304a8f552cab925ddde` | `ID_ID` | 3,877 | 21,296 | 25,175 | 17,421 | changed tokens | 17,421 |
+| `is-is-default` | `standalone` | `1.0.0` | `920ca13c98fe07f141f6b646c5858c5e0143ec94cfde08fd885ae66923076ebf` | `IS_IS` | 4,766 | 52,132 | 58,089 | 48,557 | changed tokens | 48,557 |
+| `it-it-default` | `default` | `1.0.0` | `5e03be31c9761e30dbf24a47a5ced3d6ec949dabd31e92632fdd9f7c67fc2e12` | `IT_IT` | 10,009 | 324,366 | 337,546 | 317,542 | changed tokens | 317,542 |
+| `itl-default` | `standalone` | `1.0.0` | `e794b033969008c47de41f60506075e2e8f3e4a50120a30c708e74ba012abceb` | `ITL` | 1,574 | 3,546 | 5,124 | 1,976 | changed tokens | 5,000 |
+| `izh-default` | `standalone` | `1.0.0` | `31b8cdd20bb1be97b44a83c8d3b529119dae4b9204bd93866894d6c30d0ce2b2` | `IZH` | 50 | 1,024 | 1,074 | 974 | changed tokens | 5,000 |
+| `ja-jp-default` | `standalone` | `1.0.0` | `4eb7421a20914b847177e9fd02c798fe58e088d2bb212136efcce576c9dd180b` | `JA_JP` | 107 | 10,848 | 11,699 | 11,485 | changed tokens | 11,485 |
+| `ka-ge-default` | `standalone` | `1.0.0` | `f8205f3c31e3dc6a0f908b3839956c72688ae04b672ed75ba9083e4280b99c3b` | `KA_GE` | 3,847 | 75,914 | 79,967 | 72,273 | changed tokens | 72,273 |
+| `kbd-default` | `standalone` | `1.0.0` | `2cdd9c1b8d88369667243c7b27c9eab06c02497893d6190deb7850d245b2fb89` | `KBD` | 249 | 3,054 | 3,303 | 2,805 | changed tokens | 5,000 |
+| `kca-default` | `standalone` | `1.0.0` | `22be165f4e2d9805a8db4b84ca943fe66eb6e1c3dbf390bd8f135a03678107e9` | `KCA` | 661 | 2,175 | 2,859 | 1,537 | changed tokens | 5,000 |
+| `ket-default` | `standalone` | `1.0.0` | `4916e12863d306cbda2240606bd667e77d1c0d96bb8a3981c1d72a7d565887eb` | `KET` | 519 | 1,100 | 1,622 | 584 | changed tokens | 5,000 |
+| `kjh-default` | `standalone` | `1.0.0` | `241daa9381a3a3d0cb28defa7872ec3f913195782bc343d185d60e28de09b622` | `KJH` | 74 | 1,172 | 1,246 | 1,098 | changed tokens | 5,000 |
+| `kk-kz-default` | `standalone` | `1.0.0` | `46bada2596fbdc80e42617ae1c4fbf638b436271a37c2d4253072b399e5c3d66` | `KK_KZ` | 26 | 357 | 383 | 331 | changed tokens | 5,000 |
+| `kl-gl-default` | `standalone` | `1.0.0` | `e82c2bf70b7e7378002469f79dab280af2d14f8db558cec395b776ae423f2deb` | `KL_GL` | 23 | 321 | 344 | 298 | changed tokens | 5,000 |
+| `klr-default` | `standalone` | `1.0.0` | `f1395204bd22af53c50b07e632e4c330877c4018b67bc4441cee5aa199bbaddc` | `KLR` | 591 | 41,185 | 45,299 | 44,117 | changed tokens | 44,117 |
+| `kmr-default` | `standalone` | `1.0.0` | `147729f9f3eb36dd1eaf35095cc57086e0aebf42ad4780aaf80034a5c0492948` | `KMR` | 13,981 | 191,104 | 206,245 | 178,283 | changed tokens | 178,283 |
+| `kn-in-default` | `standalone` | `1.0.0` | `c6922ab4612042b71af377b03692c6787e07cec25ddd1b1cea757f2e0de095c8` | `KN_IN` | 4 | 11 | 15 | 7 | changed tokens | 5,000 |
+| `ko-kr-default` | `standalone` | `1.0.0` | `bbcb468402ee54327dfd2101df5ef90869db06fe0543d85c1f62ac4d3beba155` | `KO_KR` | 2,672 | 192,673 | 196,302 | 190,958 | changed tokens | 190,958 |
+| `kod-default` | `standalone` | `1.0.0` | `b0a2de585785e54665a193e707bfd4b90e9f0c675f44817d5465278553880867` | `KOD` | 65 | 524 | 589 | 459 | changed tokens | 5,000 |
+| `kon-default` | `standalone` | `1.0.0` | `0adce159fa3b67b38dbc181eb3d2d2e0d6016cb62b3266fe6e47e901bd2a80e0` | `KON` | 200 | 585 | 786 | 386 | changed tokens | 5,000 |
+| `krl-default` | `standalone` | `1.0.0` | `a98c4446dd7eaec464369f6c2a6691748c7844629b74efbebf7437a8454c4c5a` | `KRL` | 20 | 566 | 586 | 546 | changed tokens | 5,000 |
+| `ky-kg-default` | `standalone` | `1.0.0` | `ef11ea53a79367b3c156ffb284d3e9101f30948e705c77660fe01810b0b00b86` | `KY_KG` | 72 | 2,997 | 3,069 | 2,925 | changed tokens | 5,000 |
+| `la-default` | `standalone` | `1.0.0` | `fd906e27def14dcb23ae0771a968555c1abf9f8d1de6bb156099be2425d3b4af` | `LA` | 46,067 | 485,259 | 545,673 | 453,539 | changed tokens | 453,539 |
+| `lg-ug-default` | `standalone` | `1.0.0` | `6a9974b58b7b5b5781c61efa728ef599c87d921f4f016ffdecf820c1b65e2cbb` | `LG_UG` | 87 | 4,671 | 4,831 | 4,657 | changed tokens | 5,000 |
+| `lin-default` | `standalone` | `1.0.0` | `f6eda2e6cce620baf0bd1baaa6ee8142497e702a094a5b92b8295c4aa9732439` | `LIN` | 57 | 230 | 287 | 173 | changed tokens | 5,000 |
+| `liv-default` | `standalone` | `1.0.0` | `9545c84208002654360f49184d96f8553a9e9b8abd0cddb85db41b1109397e33` | `LIV` | 201 | 2,841 | 3,059 | 2,657 | changed tokens | 5,000 |
+| `lld-default` | `standalone` | `1.0.0` | `2866be97f9b5ff688865d927aa1c47975dd1a69a151f9e643f2e6221fbb89e51` | `LLD` | 180 | 4,819 | 5,034 | 4,674 | changed tokens | 5,000 |
+| `lt-lt-default` | `standalone` | `1.0.0` | `5702f06a6340b06ebb98443686b7053968d1b1b51ec178866488a6f7a6791257` | `LT_LT` | 1,348 | 27,335 | 28,771 | 26,075 | changed tokens | 26,075 |
+| `lv-lv-default` | `standalone` | `1.0.0` | `157444ef19affe7ed5d1b0d1e356683f025f611b3cb687a8c8e8be26bb24287f` | `LV_LV` | 7,238 | 75,492 | 87,465 | 72,989 | changed tokens | 72,989 |
+| `mag-default` | `standalone` | `1.0.0` | `9cbefd751c52eb3f6b322f2ed17b34edb9c46ac3dd2358c89434c3dca2ce91f6` | `MAG` | 110 | 206 | 318 | 98 | changed tokens | 5,000 |
+| `mg-mg-default` | `standalone` | `1.0.0` | `d446dd8d28922e7a3d54e3f059845e2165bfbe758d8b3869021ad47e8e7c46f2` | `MG_MG` | 159 | 636 | 795 | 477 | changed tokens | 5,000 |
+| `mi-nz-default` | `standalone` | `1.0.0` | `4f55a59ba208cd87743e18fbca89af50c731baed6148ff7940ddc51c95acc9de` | `MI_NZ` | 104 | 207 | 312 | 104 | changed tokens | 5,000 |
+| `mk-mk-default` | `standalone` | `1.0.0` | `56aed4f1e6df14829a2bbc92961247b002404cd8a73068ec05f27a52cda972bd` | `MK_MK` | 9,776 | 135,590 | 146,166 | 126,614 | changed tokens | 126,614 |
+| `mn-mn-default` | `standalone` | `1.0.0` | `ac94039864a6b0e4fb2008dcba2a695875409e6382091f4c820d917cb1707ecb` | `MN_MN` | 2,140 | 17,231 | 19,733 | 15,453 | changed tokens | 15,453 |
+| `mt-mt-default` | `standalone` | `1.0.0` | `c3ff6a3014b3765a698c48b6c85d84c729fb4c552c50bb40cc641002dc959beb` | `MT_MT` | 107 | 1,421 | 1,537 | 1,323 | changed tokens | 5,000 |
+| `mwf-default` | `standalone` | `1.0.0` | `6247cb8e87c3d534c07e6843f2f7671fc958a785712bc47d607fb86233fcf897` | `MWF` | 29 | 592 | 749 | 691 | changed tokens | 5,000 |
+| `nap-default` | `standalone` | `1.0.0` | `3cb6cfa9bd14b91af99e845cbb75f0ffa62bd4c9916e8c3733be9c313b95e097` | `NAP` | 40 | 1,497 | 1,537 | 1,457 | changed tokens | 5,000 |
+| `nav-default` | `standalone` | `1.0.0` | `ad0db7cb940f7d76d2c1ffbc4da52b0bffa725592f0357a09cf1434c88ef8b84` | `NAV` | 585 | 9,826 | 10,706 | 9,536 | changed tokens | 9,536 |
+| `nb-no-default` | `default` | `1.0.0` | `f495bffb44e79d27993e6e2e65d4b1204b29365dc93f481b2d8b96766fc90fd9` | `NB_NO` | 17,929 | 73,170 | 90,757 | 57,381 | changed tokens | 57,381 |
+| `nds-default` | `standalone` | `1.0.0` | `358c11faeb80d31662f3a3cca65a87c7f764382381a00e715add8f78b83ae8eb` | `NDS` | 325 | 2,493 | 2,868 | 2,218 | changed tokens | 5,000 |
+| `nl-nl-default` | `default` | `1.0.0` | `c098034adc42da2ca3e419160e6dd2c2b3868f8af334303b3a191e09caadaf5e` | `NL_NL` | 4,992 | 26,201 | 31,466 | 21,485 | changed tokens | 21,485 |
+| `nn-no-default` | `default` | `1.0.0` | `900cf2005605aea2a3d8d731ec0b0c1f47fb4469b4ba6b9134145d4d026a0398` | `NN_NO` | 4,688 | 16,937 | 19,651 | 13,562 | changed tokens | 13,562 |
+| `non-default` | `standalone` | `1.0.0` | `0e4cc499f1414b304a0a91d74e1315c951a93344a74ca70c27c9d339c90daa52` | `NON` | 2,664 | 46,067 | 52,239 | 46,911 | changed tokens | 46,911 |
+| `ny-mw-default` | `standalone` | `1.0.0` | `2f6bf10a59475dbf8c925fc27cdd23712c3d914d6b0fa5a9d6542f51873afadd` | `NY_MW` | 225 | 3,150 | 3,375 | 2,925 | changed tokens | 5,000 |
+| `oci-default` | `standalone` | `1.0.0` | `7c5dcd14b47020d7676a1fc634a1d9f38a06728d69407386d8864cd909046c3c` | `OCI` | 174 | 7,465 | 7,784 | 7,436 | changed tokens | 7,436 |
+| `om-et-default` | `standalone` | `1.0.0` | `df01f908af60ba4a63933a91845d1241a37d366467026bd826ebec08689406bb` | `OM_ET` | 84 | 656 | 740 | 572 | changed tokens | 5,000 |
+| `ood-default` | `standalone` | `1.0.0` | `41a559d1a6ae83c0f2579793a2a4b883b2e974cfd705f90ad856b395346c368b` | `OOD` | 156 | 455 | 614 | 302 | changed tokens | 5,000 |
+| `osx-default` | `standalone` | `1.0.0` | `7ad2453964fa001ec17b73eb1156c25513961a4683d5d7abaaaec477d3e8755e` | `OSX` | 863 | 12,361 | 13,416 | 11,690 | changed tokens | 11,690 |
+| `ote-default` | `standalone` | `1.0.0` | `8cd8e5d6bfe7c5848e39ff334b49580a769ba78d1576d2861436641e8cec5a58` | `OTE` | 1,110 | 1,785 | 2,991 | 771 | changed tokens | 5,000 |
+| `pl-pl-polimorf` | `optional` | `1.0.0` | `4fe4bf5e6c22c1beea5b3d57f1ce4c9ea5aac1ed8ab24c616fb06df745e40d15` | `PL_PL` | 315,639 | 4,668,685 | 4,812,023 | 4,496,384 | changed tokens | 4,496,384 |
+| `pl-pl-unimorph` | `default` | `1.0.0` | `8191ed727097839cc808cbc5c56a1bd78b3c851e7733ad226ad9a51519a54721` | `PL_PL` | 9,990 | 120,867 | 132,308 | 112,351 | changed tokens | 112,351 |
+| `ps-af-default` | `standalone` | `1.0.0` | `34c08ecbe51c274996a2e8be119cbf323317fe2eb390454730e60a4b776e9d33` | `PS_AF` | 381 | 2,945 | 3,445 | 2,683 | changed tokens | 5,000 |
+| `pt-pt-default` | `default` | `1.0.0` | `7a035ff330a6f0548f446cd0d6617bc1cf4751292125a3564d3a255c5d6f516d` | `PT_PT` | 4,001 | 211,091 | 215,490 | 207,488 | changed tokens | 207,488 |
+| `que-default` | `standalone` | `1.0.0` | `4aca805b20dce58d1cbd80a824efca214ada72a8955df738a271b08a5bfbeca8` | `QUE` | 814 | 122,820 | 125,491 | 123,863 | changed tokens | 123,863 |
+| `ro-ro-default` | `standalone` | `1.0.0` | `fe5215778c84f871d0f6408dcb663a7cdef6571aaa0b96dba88ed7bfd09a3461` | `RO_RO` | 4,377 | 48,497 | 53,278 | 44,524 | changed tokens | 44,524 |
+| `ru-ru-default` | `default` | `1.0.0` | `df7ea25e63a875eeec7a4185be685bd5372a3c568db85c34c44fdf5d8d980a40` | `RU_RU` | 37,410 | 759,333 | 806,279 | 731,471 | changed tokens | 731,471 |
+| `sah-default` | `standalone` | `1.0.0` | `c5b361c9e1df64d60813c96cbea4d228d97fbc76e71bbaccb87609a13e6b67e8` | `SAH` | 5,622 | 526,431 | 534,266 | 523,022 | changed tokens | 523,022 |
+| `san-default` | `standalone` | `1.0.0` | `bf94e3c36639ce16b0c11a6e9ab62baebc8acea18d14926a597585cbb179e708` | `SAN` | 54 | 55 | 109 | 1 | changed tokens | 5,000 |
+| `sdh-default` | `standalone` | `1.0.0` | `c19f455c815ae94776d54dba1c504d41abea342c15ba9e8714dd370cad98be2a` | `SDH` | 1 | 165 | 166 | 164 | changed tokens | 5,000 |
+| `see-default` | `standalone` | `1.0.0` | `329f57fb8c306ab2883f5640641c9a361f04a0a50fc3e9f62c0d8c6e40ddf380` | `SEE` | 26 | 483 | 509 | 457 | changed tokens | 5,000 |
+| `sga-default` | `standalone` | `1.0.0` | `01ea73ea683f0864976e43375d3ff828c009349e17c5b4c34fb832950f29e7ca` | `SGA` | 49 | 929 | 980 | 882 | changed tokens | 5,000 |
+| `shp-default` | `standalone` | `1.0.0` | `6139156cc7e79c30520d70b5a40efbcf784ea878c8e2668b278a01dee9ebf073` | `SHP` | 1,997 | 7,705 | 9,940 | 5,946 | changed tokens | 5,946 |
+| `sjo-default` | `standalone` | `1.0.0` | `22101d045412a65a834bce25963181ec1cc3b7bf7f2ae8a1cf1d31950b11840c` | `SJO` | 1,892 | 3,151 | 5,089 | 1,305 | changed tokens | 5,000 |
+| `sl-si-default` | `standalone` | `1.0.0` | `b59e7199dce258b404729e4d708172c535733cfae430a26f1c4faa3d25dc323e` | `SL_SI` | 2,531 | 25,100 | 27,734 | 22,672 | changed tokens | 22,672 |
+| `slp-default` | `standalone` | `1.0.0` | `fb4cc41d286579f2c0992e282bbfa7e0ccc251526119cf51ad752f363ac352be` | `SLP` | 21 | 41 | 62 | 20 | changed tokens | 5,000 |
+| `sme-default` | `standalone` | `1.0.0` | `77abb417f0c67f00cf44a1d129b8b578b9381a2e6b223558124cb39954830685` | `SME` | 2,082 | 53,034 | 55,703 | 51,539 | changed tokens | 51,539 |
+| `sn-zw-default` | `standalone` | `1.0.0` | `a9aae37550b9230283ca7c62056ba6c7d0b5cf3f3bf20f30905f723c50addd98` | `SN_ZW` | 86 | 2,640 | 2,726 | 2,554 | changed tokens | 5,000 |
+| `sq-al-default` | `standalone` | `1.0.0` | `dda602d80fe10911f4aec2f70196f429eed1a5c232ece4620065fe96274512c4` | `SQ_AL` | 587 | 10,218 | 10,851 | 9,677 | changed tokens | 9,677 |
+| `st-za-default` | `standalone` | `1.0.0` | `86230d154bada51128d1c03a631fd182170e5d2b8070d57b65ebee2db91c6ce9` | `ST_ZA` | 26 | 416 | 442 | 390 | changed tokens | 5,000 |
+| `sv-se-default` | `default` | `1.0.0` | `d9be72e3d67c776622c4281e04e4063b9381e8f84a823d98ebf08888b82dff0c` | `SV_SE` | 12,371 | 95,181 | 110,468 | 85,737 | changed tokens | 85,737 |
+| `swc-default` | `standalone` | `1.0.0` | `28f8e90aea44ced44da8ca797ea40c3e822d580bd6aaa71f3dce6d847a02b9a9` | `SWC` | 100 | 6,700 | 6,800 | 6,600 | changed tokens | 6,600 |
+| `syc-default` | `standalone` | `1.0.0` | `8eda4d03d168a0908033eacff07d2fbe3332692157c9d1f862cd8ca07fcafdd1` | `SYC` | 215 | 2,139 | 2,365 | 1,935 | changed tokens | 5,000 |
+| `tat-default` | `standalone` | `1.0.0` | `1617b08152da422aa295ef7f990e9e30ec8834eeeb9555cba805d96142462779` | `TAT` | 1,258 | 7,483 | 8,745 | 6,229 | changed tokens | 6,229 |
+| `tg-tj-default` | `standalone` | `1.0.0` | `4809cdc83d745d0db145d74c6654e4732b83bf70e32df3a208aac22a0db4d6b1` | `TG_TJ` | 75 | 78 | 153 | 3 | changed tokens | 5,000 |
+| `tk-tm-default` | `standalone` | `1.0.0` | `3821730499cd42fa7974c907d08377763e1cc1a6f85784b638d22f87124317e2` | `TK_TM` | 68 | 804 | 872 | 736 | changed tokens | 5,000 |
+| `tl-ph-default` | `standalone` | `1.0.0` | `e9c291d64e9359d58257cdd5342b7a4fdb94b123e5b23be3343a97073661aa78` | `TL_PH` | 289 | 2,042 | 2,336 | 1,758 | changed tokens | 5,000 |
+| `tr-tr-default` | `standalone` | `1.0.0` | `29db7f67de1b5560187ab496998aaf886277b96f8c10cec7f49790c2afd23be1` | `TR_TR` | 3,014 | 222,207 | 226,873 | 220,845 | changed tokens | 220,845 |
+| `tyv-default` | `standalone` | `1.0.0` | `7b4189ea74cbc151a0dfa22c01dd30b105a53f9f1be561033184df0d5fecd7b6` | `TYV` | 5,032 | 491,421 | 499,024 | 488,960 | changed tokens | 488,960 |
+| `ug-cn-default` | `standalone` | `1.0.0` | `a76f8114c54be66c7ea564c79bc81f88d62a9ab9dd2c7028e542aa1d04e9b22f` | `UG_CN` | 85 | 6,200 | 6,316 | 6,146 | changed tokens | 6,146 |
+| `uk-ua-default` | `default` | `1.0.0` | `cf3f612cfff16cb7763f99c55851069489b883c3bdd1a6576cd8c57a97e07eae` | `UK_UA` | 1,493 | 14,150 | 15,737 | 12,752 | changed tokens | 12,752 |
+| `ur-pk-default` | `standalone` | `1.0.0` | `588e4e2644f121073adc8a62f8f5c14e9fcaa5e57ed29ada787596c0248eb426` | `UR_PK` | 162 | 881 | 1,047 | 723 | changed tokens | 5,000 |
+| `us-uk-default` | `default` | `1.0.1` | `0a3ec1fddd5ddb859f5ffa8003efe9200a5ef6203df1fddaf53a58003f33582c` | `US_UK` | 396,939 | 591,946 | 1,002,414 | 208,540 | changed tokens | 208,540 |
+| `uz-uz-default` | `standalone` | `1.0.0` | `f8e520d4f2741f4c8a52013923850696838cdffa9c433827e942cc779d86f725` | `UZ_UZ` | 15 | 1,080 | 1,095 | 1,065 | changed tokens | 5,000 |
+| `vec-default` | `standalone` | `1.0.0` | `29d9dee4846c6fee65c69116c9ac2cfa370719af04cf206050ce08b9c5acfb0d` | `VEC` | 368 | 8,595 | 9,056 | 8,320 | changed tokens | 8,320 |
+| `vot-default` | `standalone` | `1.0.0` | `f05693d86412bf46ca79606c033018d488553c30a180f3d00bee394867ad667b` | `VOT` | 55 | 1,392 | 1,448 | 1,338 | changed tokens | 5,000 |
+| `vro-default` | `standalone` | `1.0.0` | `01a1408466bf5c549721a9bffa976dda00fb66d89a1a70d91022296162d9ba94` | `VRO` | 50 | 277 | 328 | 228 | changed tokens | 5,000 |
+| `xcl-default` | `standalone` | `1.0.0` | `aca4a0452f876a3c7fc4372c8253f6d73c77e493191e5e5f62aeb69cecbc92e4` | `XCL` | 4,299 | 55,946 | 63,418 | 54,820 | changed tokens | 54,820 |
+| `xno-default` | `standalone` | `1.0.0` | `eb6fa9f98fbc2b3d74714032555f5ab4bad394fdec1ddb2d386dabeed1812fbf` | `XNO` | 5 | 186 | 191 | 181 | changed tokens | 5,000 |
+| `xty-default` | `standalone` | `1.0.0` | `f79b31e96b351170504e1a28ca6665be22725f1c50e3f2b30324f37de84f5431` | `XTY` | 417 | 1,536 | 1,965 | 1,131 | changed tokens | 5,000 |
+| `yi-default` | `default` | `1.0.0` | `f47de665c27dcd72833a82904e49c68a945bb5aca769a7ec5a0164e2c981a6d3` | `YI` | 802 | 3,532 | 4,300 | 2,776 | changed tokens | 5,000 |
+| `zpv-default` | `standalone` | `1.0.0` | `0dba154558eede2bec318eb5c33edc856483b1ebd999f433f0d16508d5cc62e9` | `ZPV` | 203 | 812 | 1,021 | 615 | changed tokens | 5,000 |
+| `zu-za-default` | `standalone` | `1.0.0` | `da7922f2232206c74f0fe22bdc539c5b6f28a2fc781b40945a13be9bb1da4c75` | `ZU_ZA` | 563 | 32,384 | 33,297 | 32,171 | changed tokens | 32,171 |
 
-Speed benchmarks process the complete changed-token dictionary sequence for the language. Only resources with fewer than 5,000 changed tokens are repeated to reach the minimum timing size; larger resources are not sampled or truncated.
+Speed benchmarks process the complete selected timing sequence for the model. Changed tokens are selected whenever any exist. For root-only dictionaries, the complete root-preservation sequence is selected so the model still receives a speed measurement. Only selected populations with fewer than 5,000 tokens are repeated to reach the minimum timing size; larger populations are not sampled or truncated.
diff --git a/docs/benchmarks/reference/english-coverage.md b/docs/benchmarks/reference/english-coverage.md
index e1ac0de..9975880 100644
--- a/docs/benchmarks/reference/english-coverage.md
+++ b/docs/benchmarks/reference/english-coverage.md
@@ -15,21 +15,21 @@ The speed cells are JMH point estimates with their 99.9% score errors. The 100%
 
 <!-- ENGLISH-SPEED-SUITES:START -->
 !!! note "Separate English speed suites"
-    The full-knowledge `97.9 ns/token` point on this page belongs to the coverage suite. The [English language comparison](../languages/english.md#speed) reports `80.8 ns/token` from a separate JMH method and run. Their uncertainty intervals overlap; neither point estimate should replace the other.
+    The full-knowledge `120.4 ns/token` point on this page belongs to the coverage suite. The [English language comparison](../languages/english.md#speed) reports `84.7 ns/token` from a separate JMH method and run. Their uncertainty intervals overlap; neither point estimate should replace the other.
 <!-- ENGLISH-SPEED-SUITES:END -->
 
 | Used rows | Actual row ratio | All exact | Changed exact | Root preserved | Speed ms/op | Error ms | ns/token |
 | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
-| 100% | 100.000% | 97.668% | 98.110% | 97.552% | 20.425 | 3.636 | 97.9 |
-| 90% | 90.000% | 97.239% | 95.821% | 97.612% | 17.779 | 1.827 | 85.3 |
-| 80% | 80.000% | 96.827% | 93.673% | 97.656% | 15.343 | 1.321 | 73.6 |
-| 70% | 70.000% | 96.392% | 91.430% | 97.695% | 16.444 | 2.027 | 78.9 |
-| 60% | 60.000% | 95.935% | 89.244% | 97.693% | 14.330 | 1.350 | 68.7 |
-| 50% | 50.000% | 95.453% | 86.979% | 97.678% | 14.953 | 2.625 | 71.7 |
-| 40% | 40.000% | 94.939% | 84.667% | 97.638% | 12.919 | 1.155 | 62.0 |
-| 30% | 30.000% | 94.398% | 82.443% | 97.538% | 12.166 | 1.305 | 58.3 |
-| 20% | 20.000% | 93.821% | 80.174% | 97.406% | 11.549 | 1.535 | 55.4 |
-| 10% | 10.000% | 93.057% | 77.327% | 97.190% | 14.360 | 3.524 | 68.9 |
+| 100% | 100.000% | 97.668% | 98.110% | 97.552% | 25.107 | 4.358 | 120.4 |
+| 90% | 90.000% | 97.239% | 95.821% | 97.612% | 24.021 | 1.865 | 115.2 |
+| 80% | 80.000% | 96.827% | 93.673% | 97.656% | 23.593 | 2.656 | 113.1 |
+| 70% | 70.000% | 96.392% | 91.430% | 97.695% | 18.965 | 1.559 | 90.9 |
+| 60% | 60.000% | 95.935% | 89.244% | 97.693% | 18.588 | 1.040 | 89.1 |
+| 50% | 50.000% | 95.453% | 86.979% | 97.678% | 18.479 | 2.304 | 88.6 |
+| 40% | 40.000% | 94.939% | 84.667% | 97.638% | 17.345 | 1.504 | 83.2 |
+| 30% | 30.000% | 94.398% | 82.443% | 97.538% | 17.164 | 1.252 | 82.3 |
+| 20% | 20.000% | 93.821% | 80.174% | 97.406% | 14.912 | 1.512 | 71.5 |
+| 10% | 10.000% | 93.057% | 77.327% | 97.190% | 14.937 | 2.320 | 71.6 |
 
 ## Column Meanings
 
diff --git a/docs/benchmarks/reference/environment.md b/docs/benchmarks/reference/environment.md
index b1644d9..8883221 100644
--- a/docs/benchmarks/reference/environment.md
+++ b/docs/benchmarks/reference/environment.md
@@ -6,16 +6,20 @@ The values below are environment-specific and must not be read as universal perf
 
 The generalization report is deterministic and publishes no runtime score, so
 CPU frequency, background load, and elapsed time do not affect its accuracy
-counters. It was nevertheless generated on the same identified workstation:
+counters. The active snapshot combines the immutable 20-language campaign with
+the separately measured 146-language continuation:
 
 | Item | Value |
 | --- | --- |
-| Evaluation date | 2026-08-25 (Europe/Prague) |
-| Command | `./gradlew --no-daemon -PdictionaryGeneralizationReleaseVersion=4.2.0-6-g84e57fb dictionaryGeneralization` |
-| Radixor/Java source identity | `4.2.0-6-g84e57fb` |
-| Report | `build/reports/generalization/dictionary-generalization.csv` |
-| Published snapshot | `docs/benchmarks/data/dictionary-generalization.csv` |
-| Scenarios | 1,000: 20 default models × 10 coverage levels × 5 frozen splits |
+| Evaluation dates | Historical 20 languages: 2026-08-25; 146-language continuation: 2026-09-11 (Europe/Prague) |
+| Continuation command | `./gradlew --no-daemon -PdictionaryGeneralizationReleaseVersion=4.3.0.dirty dictionaryGeneralizationStandalone` |
+| Merge/publication command | `./gradlew --no-daemon mergeDictionaryGeneralizationSnapshots publishDictionaryGeneralizationDocumentation` |
+| Radixor/Java source identities | Historical rows: `4.2.0-6-g84e57fb`; continuation rows: `4.3.0.dirty` at base revision `14f61beb106bca21d78f828e2f8e35fb457d3d90` |
+| Continuation report | `build/reports/generalization/dictionary-generalization-standalone-2026-09-11.csv` |
+| Active published snapshot | `docs/benchmarks/data/dictionary-generalization-2026-09-11.csv` |
+| Historical archive | `docs/benchmarks/data/dictionary-generalization.csv` |
+| Scenarios | 8,300: 166 default models × 10 coverage levels × 5 frozen splits |
+| Continuation runtime | 6 min 34 s; informational only |
 | Evaluation type | Raw deterministic exact-root counters; no elapsed-time value is interpreted |
 | Java runtime | OpenJDK 25.0.4, Red Hat build 25.0.4+7 |
 | Operating system | Fedora Linux 44 (MATE-Compiz) |
@@ -23,12 +27,12 @@ counters. It was nevertheless generated on the same identified workstation:
 | CPU | AMD Ryzen 5 8600G, 6 physical / 12 logical CPUs |
 | Architecture | x86_64 |
 | Split protocol | `radixor-generalization-v1`; exact-size nested row prefixes |
-| Provenance | Every CSV row records model ID, independent model version, and compressed-resource SHA-256 |
+| Provenance | Every CSV row records model ID, independent model version, compressed-resource SHA-256, source revision/state, and generator SHA-256; each campaign has its own source manifest |
 
 The CPU was configured with the `performance` governor, but that fact is not
 used to support any generalization claim. The new report deliberately excludes
 speed; the Java timing environment below remains the provenance for the
-published 2026-08-25 JMH tables.
+published 2026-09-10 JMH tables.
 
 ## Edit-Cost Sensitivity Run
 
@@ -52,49 +56,50 @@ Elapsed time and memory describe the cost of reproducing the experiment and are
 
 | Item | Value |
 | --- | --- |
-| Benchmark date | 2026-08-25 (Europe/Prague) |
+| Benchmark date | 2026-09-10 (the run completed on 2026-09-11, Europe/Prague) |
 | Corpus command | `./gradlew benchmarkCorpusReport --no-daemon` |
-| Exact-root accuracy command | `tools/run-published-accuracy-benchmarks.sh 2026-08-25`; all four `*BenchmarkQuality` classes are selected and timing scores are discarded |
-| Stemming-quality command | `./gradlew stemmingQuality --no-daemon` |
-| Published speed command | `tools/run-published-speed-benchmarks.sh 2026-08-25 4.2.0-6-g84e57fb` |
-| Published speed run interval | 2026-08-25 14:33:15 to 16:27:55 Europe/Prague (1 h 54 min 40 s, including idle intervals and both JMH suites) |
-| Stabilization intervals | 30 s before the main speed matrix; 15 s between the main matrix and coverage-speed suite |
+| Exact-root accuracy command | `tools/run-published-accuracy-benchmarks.sh 2026-09-10`; the exact-root benchmark suite is selected and timing scores are discarded |
+| Pairwise stemming quality | Complete 166-language default-model snapshot published as `stemming-quality-2026-09-11.csv`; the prior 20-language CSV remains an immutable archive |
+| Published speed command | `tools/run-published-speed-benchmarks.sh 2026-09-10 4.3.0.dirty` |
+| Main speed run | Started 2026-09-10 22:21:14; the completed main CSV was written on 2026-09-11 at approximately 01:21:00 Europe/Prague |
+| Clean coverage-speed run | 2026-09-11 02:26:14 to 02:31:25 Europe/Prague |
+| Stabilization intervals | 30 s before the main speed matrix; no additional delay before the clean recovery run because the preceding failed post-run validation had already left the machine idle |
 | Corpus and command report | `build/reports/jmh/benchmark-corpora.csv` |
-| Exact-root reports | `build/reports/jmh/stemmer-accuracy-2026-08-25.csv` and `.txt` |
-| Speed reports | `build/reports/jmh/stemmer-speed-2026-08-25.csv` and `.txt` |
-| English coverage accuracy reports | `build/reports/jmh/english-coverage-accuracy-2026-08-25.csv` and `.txt` |
-| English coverage speed reports | `build/reports/jmh/english-coverage-speed-2026-08-25.csv` and `.txt` |
+| Exact-root reports | `build/reports/jmh/stemmer-accuracy-2026-09-10.csv` and `.txt` |
+| Speed reports | `build/reports/jmh/stemmer-speed-2026-09-10.csv` and `.txt` |
+| English coverage accuracy reports | `build/reports/jmh/english-coverage-accuracy-2026-09-10.csv` and `.txt` |
+| English coverage speed reports | `build/reports/jmh/english-coverage-speed-2026-09-10.csv` and `.txt` |
 | Stemming-quality reports | `build/reports/stemming-quality/stemming-quality.csv` and `.md` |
-| Environment report | `build/reports/jmh/performance-environment-2026-08-25.txt` |
-| Selected speed methods | `build/reports/jmh/published-speed-benchmarks-2026-08-25.txt` |
-| Comparison scope | Same-language methods used by the 20 language pages; `PolishPolimorfStemmerComparisonBenchmark`, all quality methods, the separate CISTEM gold-standard experiment, and internal trie microbenchmarks are excluded |
-| Model scope | Exactly the 20 IDs declared by `Language.defaultModelId()`; Polish uses `pl-pl-unimorph`, and `pl-pl-polimorf` is not measured |
-| Core base commit | `84e57fb27ae40913569e826858a8ebb07cf2ea01` |
-| Release identity | Radixor/Java `4.2.0-6-g84e57fb`; exact measured tracked changes and untracked-source checksums are retained as `measured-source-2026-08-25.patch` and `measured-untracked-2026-08-25.sha256` |
+| Environment report | `build/reports/jmh/performance-environment-2026-09-10.txt` |
+| Selected speed methods | `build/reports/jmh/published-speed-benchmarks-2026-09-10.txt` |
+| Comparison scope | Parameterized Radixor measurements for all 167 user-facing model IDs plus approved authoritative same-language comparator paths, including the separately identified optional PoliMorf Morfologik row; quality methods, the CISTEM gold-standard experiment, ambiguous duplicate baselines, and internal trie microbenchmarks are excluded |
+| Model scope | All 167 user-facing IDs: 166 language defaults plus the optional `pl-pl-polimorf`; default and optional Polish results retain distinct model identities |
+| Core base commit | `14f61beb106bca21d78f828e2f8e35fb457d3d90` |
+| Release identity | Radixor/Java `4.3.0.dirty`; exact measured tracked changes and untracked-source checksums are retained as `measured-source-2026-09-10.patch` and `measured-untracked-2026-09-10.sha256` |
 | JMH version | 1.37 |
 | Speed benchmark mode | Average time, `time/op` |
 | Score unit | `ns/op`; language pages additionally derive `ms/op` and `ns/token` |
-| Speed warmup | 5 iterations, 1 s each, independently in every fork |
-| Speed measurement | 7 iterations, 1 s each, independently in every fork |
+| Speed warmup | 3 iterations, 1 s each, independently in every fork |
+| Speed measurement | 5 iterations, 1 s each, independently in every fork |
 | Speed forks | 3 independent JVM forks |
 | Speed threads | 1 |
 | Speed fork heap | Fixed `-Xms6g -Xmx6g` |
-| Reported uncertainty | JMH `Score Error (99.9%)` over 21 measured samples |
-| Observed relative uncertainty | Main speed matrix: maximum 15.542%, with 18 of 105 rows above 10% and none above 20%; coverage-speed curve: maximum 24.542%, with 7 of 10 rows above 10% and 1 above 20% |
+| Reported uncertainty | JMH `Score Error (99.9%)` over 15 measured samples |
+| Observed relative uncertainty | Main speed matrix: maximum 34.021%, with 207 of 307 rows above 10% and 106 above 20%; clean coverage-speed curve: maximum 17.358%, with 5 of 10 rows above 10% and none above 20% |
 | Deterministic measurements | Corpus, patch-command distribution, exact-root counters, coverage accuracy, and pairwise stemming quality are evaluated without interpreting runtime scores; no warmup is required |
-| JVM reported by JMH | JDK 25.0.4, OpenJDK 64-Bit Server VM, 25.0.4+7 |
-| Java runtime | OpenJDK Runtime Environment, Red Hat build 25.0.4+7 |
+| JVM reported by JMH | JDK 25.0.4.1, OpenJDK 64-Bit Server VM, 25.0.4.1+1 |
+| Java runtime | OpenJDK Runtime Environment, Red Hat build 25.0.4.1+1 |
 | JVM invoker | `/usr/lib/jvm/java-25-openjdk/bin/java` |
 | Operating system | Fedora Linux 44 (MATE-Compiz) |
-| Kernel | Linux 7.1.8-200.fc44.x86_64 |
+| Kernel | Linux 7.1.13-200.fc44.x86_64 |
 | Architecture | x86_64 |
 | CPU | AMD Ryzen 5 8600G w/ Radeon 760M Graphics |
 | Physical / logical CPUs | 6 / 12 |
 | CPU frequency policy | `amd-pstate-epp`; governor `performance` on every logical CPU; EPP `performance`; boost enabled |
 | CPU affinity | Scheduler default; no explicit pinning |
 | Installed memory | 61 GiB reported by the operating system |
-| Pre-run idle state | Load average 0.32 / 0.56 / 1.14 after the 30 s idle interval; CPU Tctl 42.8 degrees Celsius; swap unused |
-| End-of-run state | Load average 1.17 / 1.21 / 1.27; CPU Tctl 68.9 degrees Celsius |
+| Pre-run idle state | Load average 0.43 / 0.47 / 0.59 after the 30 s idle interval; CPU Tctl 46.6 degrees Celsius; swap unused |
+| End of clean coverage run | Load average 1.28 / 0.97 / 0.56; CPU Tctl 61.4 degrees Celsius; swap unused |
 | Power and idle policy | Developer workstation on stable power; screensaver, suspend, and hibernation disabled |
 | Concurrent project work | None during the published speed and coverage-speed run |
 
@@ -106,33 +111,33 @@ All Radixor rows use contracted compiled patch tries. During compilation, a subt
 
 ## Model And Source Identity
 
-`benchmark-corpora.csv` records the model ID, independent artifact version, and descriptor SHA-256 for every language. Every stemming-quality CSV row repeats the same three fields. The performance environment report additionally records checksums of the executable JMH JAR, runtime classpath manifest, corpus report, quality report, measured source patch, and untracked-source manifest.
+`benchmark-corpora.csv` records the model ID, independent artifact version, and descriptor SHA-256 for all 167 user-facing models. Every active stemming-quality CSV row repeats the same three fields for its complete 166-default-language scope. The performance environment report additionally records checksums of the executable JMH JAR, runtime classpath manifest, corpus report, quality report, measured source patch, and untracked-source manifest.
 
-The JMH runtime classpath contains the optional model artifact because it is a separately testable project dependency. It is not selected by any published benchmark. The selected-method manifest rejects `PolishPolimorf`, and the corpus/quality publication validators reject any non-default Polish model.
+The parameterized Radixor speed benchmark selects every one of the 167 model IDs, including optional PoliMorf under its separate identity. Pairwise stemming quality covers all 166 default languages; the prior 20-language file is retained only as an immutable archive. A complete 189-entry runtime-classpath content manifest and the immutable main speed CSV were pinned retrospectively after the completed main run and before the clean coverage recovery. Their SHA-256 values are `0ae88fef5c30d58b90679ed149c6570e2d7bf2eb79f347d2daaf61c8d00f6a12` and `165b653523cdf04ebf8430044876efbade80aaddb5893edaad943b24858547c0`, respectively. This retrospective pin protects the recovery boundary but is explicitly not contemporaneous main-run provenance.
 
 ## Report Files
 
 Generated local report files for this benchmark update:
 
-- `build/reports/generalization/dictionary-generalization.csv`
+- `build/reports/generalization/dictionary-generalization-standalone-2026-09-11.csv`
 - `build/reports/generalization/edit-cost-sensitivity.csv`
 - `build/reports/jmh/benchmark-corpora.csv`
-- `build/reports/jmh/stemmer-accuracy-2026-08-25.csv`
-- `build/reports/jmh/stemmer-accuracy-2026-08-25.txt`
-- `build/reports/jmh/stemmer-speed-2026-08-25.csv`
-- `build/reports/jmh/stemmer-speed-2026-08-25.txt`
-- `build/reports/jmh/english-coverage-accuracy-2026-08-25.csv`
-- `build/reports/jmh/english-coverage-accuracy-2026-08-25.txt`
-- `build/reports/jmh/english-coverage-speed-2026-08-25.csv`
-- `build/reports/jmh/english-coverage-speed-2026-08-25.txt`
-- `build/reports/jmh/performance-environment-2026-08-25.txt`
+- `build/reports/jmh/stemmer-accuracy-2026-09-10.csv`
+- `build/reports/jmh/stemmer-accuracy-2026-09-10.txt`
+- `build/reports/jmh/stemmer-speed-2026-09-10.csv`
+- `build/reports/jmh/stemmer-speed-2026-09-10.txt`
+- `build/reports/jmh/english-coverage-accuracy-2026-09-10.csv`
+- `build/reports/jmh/english-coverage-accuracy-2026-09-10.txt`
+- `build/reports/jmh/english-coverage-speed-2026-09-10.csv`
+- `build/reports/jmh/english-coverage-speed-2026-09-10.txt`
+- `build/reports/jmh/performance-environment-2026-09-10.txt`
 - `build/reports/stemming-quality/stemming-quality.csv`
 - `build/reports/stemming-quality/stemming-quality.md`
 - `build/reports/stemming-quality/metric-correlations-pearson.csv`
 - `build/reports/stemming-quality/metric-correlations-spearman.csv`
 
 The versioned documentation snapshots under `docs/benchmarks/data/` preserve the complete
-stemming-quality matrix, all 1,000 generalization scenarios, the compressed edit-cost raw matrix
+stemming-quality matrix, all 8,300 active generalization scenarios, the compressed historical edit-cost raw matrix
 plus derived analyses, and the dated CSV inputs for the published Java and Python performance
 tables. The Python provenance JSON records its environment and run parameters. Detailed JMH TXT
 logs and machine-state reports remain local build artifacts. The complete versioned input list and
diff --git a/docs/benchmarks/reference/generalization-methodology.md b/docs/benchmarks/reference/generalization-methodology.md
index 80c936c..45c1bb2 100644
--- a/docs/benchmarks/reference/generalization-methodology.md
+++ b/docs/benchmarks/reference/generalization-methodology.md
@@ -78,13 +78,17 @@ English and Finnish dictionaries from dominating small resources.
 Run:
 
 ```bash
-./gradlew dictionaryGeneralization
+./gradlew dictionaryGeneralizationStandalone
+./gradlew mergeDictionaryGeneralizationSnapshots
 ./gradlew publishDictionaryGeneralizationDocumentation
 ./gradlew verifyDictionaryGeneralizationDocumentation
 ```
 
-The first task writes `build/reports/generalization/dictionary-generalization.csv`.
-Publication validates the complete 20 × 10 × 5 matrix, exact selected-row
+The first task writes the topology-derived 146-language continuation without
+rerunning the frozen original 20 defaults. The merge task validates disjoint
+20/146 model sets, exact 1,000 + 7,300 scenario counts, headers, corpus identities,
+and row-level provenance before creating the dated 8,300-row active snapshot.
+Publication validates the complete 166 × 10 × 5 matrix, exact selected-row
 cardinality, count partitions, overlap arithmetic, model provenance, and the
 100% boundary before checking in the CSV and its SHA-256 checksum. Verification
 re-renders the page from the checked-in counters and fails if either the page or
diff --git a/docs/benchmarks/reference/methodology.md b/docs/benchmarks/reference/methodology.md
index 48a25fc..12c9537 100644
--- a/docs/benchmarks/reference/methodology.md
+++ b/docs/benchmarks/reference/methodology.md
@@ -2,26 +2,28 @@
 
 The stemmer comparison suite measures Radixor and Java stemmers on the same language and deterministic Radixor model dictionary-derived data. Published Radixor rows in this refresh use contracted compiled patch tries, where uniform preferred-command subtrees are collapsed into accepting leaves before the trie is frozen for lookup. For each language, the registered default model resource stores the expected root as the first tab-separated field on a line and its surface forms on the same line. Every single-token field on that line can therefore be paired with the same expected root.
 
-Published speed results come only from the exact method selection retained in `published-speed-benchmarks-2026-08-25.txt`. Internal `FrequencyTrie*` microbenchmarks, quality methods, the CISTEM gold-standard experiment, and the optional `PolishPolimorfStemmerComparisonBenchmark` are not part of those results. The Snowball 3.1.0 matrix includes direct Czech, Persian, and Polish workloads; all published Java comparators were measured in the same refresh.
+Published speed results come only from the exact method selection retained in `published-speed-benchmarks-2026-09-10.txt`. The parameterized Radixor method covers all 167 user-facing model IDs, while comparator rows are limited to authoritative same-language implementations; this includes the separately identified optional PoliMorf Morfologik row. Internal `FrequencyTrie*` microbenchmarks, quality methods, the CISTEM gold-standard experiment, and the ambiguous PoliMorf Radixor comparison method are excluded.
 
 ## Benchmark Passes
 
 There are two distinct benchmark passes:
 
-- Speed benchmarks process only changed dictionary pairs where `token != expectedRoot`. This removes already-root tokens from timing so a stemmer is measured on words that actually require a transformation. If a language has fewer than 5,000 changed pairs, the complete changed-pair sequence is repeated in stable order until the timing corpus has at least 5,000 tokens. Larger changed-pair corpora are not sampled or truncated.
+- Speed benchmarks preferentially process changed dictionary pairs where `token != expectedRoot`. This removes already-root tokens when the dictionary contains a transformation. A root-only model has no changed population, so its complete root-preservation corpus is used instead; the corpus report and language page identify that basis explicitly. If the selected population has fewer than 5,000 pairs, its complete sequence is repeated in stable order until the timing corpus has at least 5,000 tokens. Larger populations are not sampled or truncated.
 - Quality benchmarks process the complete dictionary for the language. They report exact agreement over all tokens, exact agreement over changed tokens only, and preservation of tokens that are already roots.
 
 Timing corpora are generated once per JMH JVM and kept in memory as shared `{token, expectedRoot}` arrays. Corpus construction, dictionary loading, trie loading, table loading, and analyzer construction are setup work and are not included in measured benchmark methods.
 
-The deterministic and timed workloads are executed separately. Corpus statistics, patch-command counts, exact-root counters, coverage accuracy, and pairwise quality do not use or interpret warmup or runtime scores. Published speed and coverage-speed methods use three independent forks, five one-second warmup iterations and seven one-second measurement iterations per fork, one benchmark thread, and a fixed 6 GiB heap.
+The deterministic and timed workloads are executed separately. Corpus statistics, patch-command counts, exact-root counters, coverage accuracy, and pairwise quality do not use or interpret warmup or runtime scores. Published speed and coverage-speed methods use three independent forks, three one-second warmup iterations and five one-second measurement iterations per fork, one benchmark thread, and a fixed 6 GiB heap.
 
 Performance is interpreted as average time per input token:
 
 ```text
-timePerChangedTokenNs = JMH score ns/op / changedTimingTokenCount
+timePerTimingTokenNs = JMH score ns/op / timingTokenCount
 ```
 
-This is necessary because Radixor dictionaries have different token counts by language.
+For transformed dictionaries, `timingTokenCount` is the changed-form timing population. For a
+root-only dictionary it is the explicitly reported complete root-preservation timing population.
+This normalization is necessary because Radixor dictionaries have different token counts by language.
 
 ## Exact-root quality and interpretation
 
@@ -41,11 +43,11 @@ Trie metadata records the language writing direction for inspection and intercha
 
 The quality pass reports exact-root agreement against the expected root from the default-model dictionary line. External-stemmer counters are written locally to:
 
-- `build/reports/jmh/stemmer-accuracy-2026-08-25.csv`
-- `build/reports/jmh/stemmer-accuracy-2026-08-25.txt`
+- `build/reports/jmh/stemmer-accuracy-2026-09-10.csv`
+- `build/reports/jmh/stemmer-accuracy-2026-09-10.txt`
 
 The tables in this documentation are verified against the checked-in
-[dated accuracy CSV](../data/java-stemmer-accuracy-2026-08-25.csv), not against the mutable local
+[dated accuracy CSV](../data/java-stemmer-accuracy-2026-09-10.csv), not against the mutable local
 report directory.
 
 Accuracy is computed from standard JMH secondary rows:
@@ -60,7 +62,9 @@ rootPreservedPercent = rootPreservedMatches / rootEvaluatedTokens * 100
 
 Morfologik can emit multiple terms for one input token. The quality benchmark uses the first emitted term for exact-root accounting when no ranking weight is exposed. Throughput benchmarks for Morfologik TokenFilter paths consume all emitted terms.
 
-External-stemmer quality reports use JMH auxiliary counter rows from one deterministic evaluation. Radixor exact-root counts are computed directly while the default-model corpus and preferred patch commands are audited, so all 20 default models have the same coverage even where no older JMH quality adapter existed. Documentation uses counter ratios and does not interpret quality benchmark timing scores.
+When `changedEvaluatedTokens` is zero, changed exactness is reported as `n/a`; an empty changed population is not a measured 0% or 100%. The overall and root-preservation ratios remain numeric when their populations are non-empty.
+
+External-stemmer quality reports use JMH auxiliary counter rows from one deterministic evaluation. Radixor exact-root counts are computed directly while the model corpus and preferred patch commands are audited, so all registered models have auditable exact-root coverage even where no older JMH quality adapter existed. Documentation uses counter ratios and does not interpret quality benchmark timing scores.
 
 Pairwise over-stemming, under-stemming, candidate-aware policies, and relation metrics are a separate analytical evaluation. See [Linguistic Quality Methodology](linguistic-quality.md); exact-root accuracy must not be interpreted as the complement of pairwise under-stemming.
 Default rows use `Language.defaultModelId()`. Optional variants require a separate model field; `pl-pl-unimorph` and `pl-pl-polimorf` must never share an ambiguous Polish label. The benchmark runtime receives each resource exactly once from its individual model JAR through direct JMH runtime dependencies. See [Model Selection and Loading](../../model-selection-and-loading.md).
diff --git a/docs/benchmarks/reference/reproducibility.md b/docs/benchmarks/reference/reproducibility.md
index 659811b..01484dc 100644
--- a/docs/benchmarks/reference/reproducibility.md
+++ b/docs/benchmarks/reference/reproducibility.md
@@ -2,24 +2,32 @@
 
 ## Published quality snapshot
 
-- Machine-readable CSV: [stemming-quality.csv](../data/stemming-quality.csv)
-- SHA-256 record: [stemming-quality.sha256](../data/stemming-quality.sha256)
-- SHA-256: `85763189eab4d0fbb047c2d5d3554c66abf9732182bd0d8fd758d7aef680e66f`
-- Complete scenarios: 314
-- Authoritative language universe: 20 languages
-- Language-page scenarios: 314 across 20 benchmark pages
+- Active machine-readable CSV: [stemming-quality-2026-09-11.csv](../data/stemming-quality-2026-09-11.csv)
+- SHA-256 record: [stemming-quality-2026-09-11.sha256](../data/stemming-quality-2026-09-11.sha256)
+- SHA-256: `1e893f8e16b0e96616d5b16c4b58d0a6e6c55b560e69bd06d433a3d213bbc772`
+- Complete scenarios: 1,236
+- Authoritative universe: 166 language defaults; optional PoliMorf is explicitly outside this snapshot
+- Historical archive: [stemming-quality.csv](../data/stemming-quality.csv), preserved byte-for-byte with SHA-256 `85763189eab4d0fbb047c2d5d3554c66abf9732182bd0d8fd758d7aef680e66f`
 
-The CSV contains the model ID, independent model version, descriptor SHA-256, raw pair counts, raw over/under numerators and denominators, candidate statistics, and relation metrics. Reserved partition-metric columns remain empty because the gold standard is an overlapping cover. Documentation is regenerated from this file rather than manually transcribed. Publication fails when any row uses a model other than the language's registered default.
+The CSV contains the model ID, independent model version, descriptor SHA-256, raw pair counts, raw over/under numerators and denominators, candidate statistics, and relation metrics. Reserved partition-metric columns remain empty because the gold standard is an overlapping cover. Documentation is regenerated from this file rather than manually transcribed. Publication fails when any row uses a model other than the language's registered default. Empty metric fields with zero denominators render as `n/a`; the raw counts remain published. `ANY_CANDIDATE` is an oracle-bound policy and has no coherent confusion matrix.
+
+The [active snapshot manifest](../data/active-snapshots.properties) selects the complete current
+performance campaign (corpus, exact-root accuracy, speed, and both English coverage inputs),
+quality, generalization, its measured-source manifest, and the comparator catalog. A refresh adds
+dated files and changes the pointers; it never repurposes an archived file name.
 
 ## Published generalization snapshot
 
-- Machine-readable CSV: [dictionary-generalization.csv](../data/dictionary-generalization.csv)
-- SHA-256 record: [dictionary-generalization.sha256](../data/dictionary-generalization.sha256)
-- SHA-256: `e6479840b9307ae03bd0873e55f397811e975125d621a8b8716d4c1a166b3ff2`
-- Measured-source manifest: [dictionary-generalization-sources.sha256](../data/dictionary-generalization-sources.sha256)
-- Complete scenarios: 1,000
-- Matrix: 20 default models × 10 coverage levels × 5 frozen splits
-- Measured source identity: Radixor/Java `4.2.0-6-g84e57fb`
+- Active machine-readable CSV: [dictionary-generalization-2026-09-11.csv](../data/dictionary-generalization-2026-09-11.csv)
+- SHA-256 record: [dictionary-generalization-2026-09-11.sha256](../data/dictionary-generalization-2026-09-11.sha256)
+- SHA-256: `0c64dda032a48a04c3e890b0669d00dd8e323000bf66274105ec7a34e87790a5`
+- Continuation measured-source manifest: [dictionary-generalization-2026-09-11-sources.txt](../data/dictionary-generalization-2026-09-11-sources.txt)
+- Continuation source-manifest checksum: [dictionary-generalization-2026-09-11-sources.sha256](../data/dictionary-generalization-2026-09-11-sources.sha256), SHA-256 `5b1612856d22ee6436864332968a726eedc0f625cede1d4f2b3a88f13392a222`
+- Complete scenarios: 8,300
+- Matrix: 166 default models × 10 coverage levels × 5 frozen splits
+- Measured source identities: the archived 20-language Radixor/Java `4.2.0-6-g84e57fb` campaign plus the separately recorded 146-language `4.3.0.dirty` continuation
+- Historical archive: [dictionary-generalization.csv](../data/dictionary-generalization.csv), preserved byte-for-byte with SHA-256 `e6479840b9307ae03bd0873e55f397811e975125d621a8b8716d4c1a166b3ff2`
+- Historical measured-source manifest: [dictionary-generalization-sources.sha256](../data/dictionary-generalization-sources.sha256)
 
 This CSV retains integer numerators and denominators for complete-dictionary,
 withheld-row, and unseen-surface scopes. It also records selected and total rows,
@@ -27,10 +35,10 @@ overlap exclusions, split seed, protocol version, and exact model provenance.
 The [generalization methodology](generalization-methodology.md) defines the
 frozen split and the limits of the claim.
 
-The accompanying manifest records the byte identity of the generator and the
-Java implementation files that determine trie construction, traversal, patch
-encoding, and lookup. The base revision and measured source state are retained
-instead of inferring a clean release tag for this experiment.
+The two campaign-specific manifests record the byte identity of the generator and the Java
+implementation files that determine trie construction, traversal, patch encoding, and lookup.
+Every row retains its campaign's base revision and measured source state; the active snapshot does
+not infer a clean release tag for the dirty continuation.
 
 ## Published edit-cost snapshot
 
@@ -54,11 +62,16 @@ The Java comparison tables, English coverage curve, Python runtime page, landing
 and technology summary are regenerated from the dated inputs below rather than from disposable
 files under `build/reports/`:
 
-- [Java corpus and command report](../data/java-benchmark-corpora-2026-08-25.csv)
-- [Java exact-root report](../data/java-stemmer-accuracy-2026-08-25.csv)
-- [Java speed report](../data/java-stemmer-speed-2026-08-25.csv)
-- [English coverage accuracy report](../data/java-english-coverage-accuracy-2026-08-25.csv)
-- [English coverage speed report](../data/java-english-coverage-speed-2026-08-25.csv)
+- [Current Java corpus and command report](../data/java-benchmark-corpora-2026-09-10.csv)
+- [Current Java exact-root report](../data/java-stemmer-accuracy-2026-09-10.csv)
+- [Current Java speed report](../data/java-stemmer-speed-2026-09-10.csv)
+- [Current English coverage accuracy report](../data/java-english-coverage-accuracy-2026-09-10.csv)
+- [Current English coverage speed report](../data/java-english-coverage-speed-2026-09-10.csv)
+- [Previous Java corpus and command report](../data/java-benchmark-corpora-2026-08-25.csv)
+- [Previous Java exact-root report](../data/java-stemmer-accuracy-2026-08-25.csv)
+- [Previous Java speed report](../data/java-stemmer-speed-2026-08-25.csv)
+- [Previous English coverage accuracy report](../data/java-english-coverage-accuracy-2026-08-25.csv)
+- [Previous English coverage speed report](../data/java-english-coverage-speed-2026-08-25.csv)
 - [Python all-language batch CSV](../data/python-all-languages-batch-2026-08-25.csv)
 - [Python all-language provenance JSON](../data/python-all-languages-batch-2026-08-25.json)
 - [SHA-256 manifest](../data/performance-snapshots.sha256)
@@ -76,12 +89,14 @@ these checked-in inputs and verifies every checksum in the manifest.
 ./gradlew --no-daemon publishStemmingQualityDocumentation \
     verifyStemmingQualityDocumentation
 ./gradlew --no-daemon \
-    -PdictionaryGeneralizationReleaseVersion=4.2.0-6-g84e57fb \
-    dictionaryGeneralization
+    -PdictionaryGeneralizationReleaseVersion=4.3.0.dirty \
+    dictionaryGeneralizationStandalone
+./gradlew --no-daemon mergeDictionaryGeneralizationSnapshots \
+    publishDictionaryGeneralizationDocumentation
 python3 tools/update-generalization-documentation.py \
-    build/reports/generalization/dictionary-generalization.csv docs update
-python3 tools/update-generalization-documentation.py \
-    docs/benchmarks/data/dictionary-generalization.csv docs verify
+    docs/benchmarks/data/dictionary-generalization-2026-09-11.csv docs update \
+    --corpus docs/benchmarks/data/java-benchmark-corpora-2026-09-10.csv \
+    --snapshot-name dictionary-generalization-2026-09-11.csv
 ./gradlew --no-daemon \
     -PdictionaryGeneralizationReleaseVersion=4.2.0-6-g84e57fb \
     editCostSensitivity
@@ -90,19 +105,19 @@ python3 tools/update-edit-cost-documentation.py \
 python3 tools/update-edit-cost-documentation.py \
     docs/benchmarks/data/edit-cost-sensitivity.csv.gz docs verify
 ./gradlew --no-daemon benchmarkCorpusReport writeJmhRuntimeClasspath
-tools/run-published-accuracy-benchmarks.sh 2026-08-25
-tools/run-published-speed-benchmarks.sh 2026-08-25 4.2.0-6-g84e57fb
+tools/run-published-accuracy-benchmarks.sh 2026-09-10
+tools/run-published-speed-benchmarks.sh 2026-09-10 4.3.0.dirty
 ./gradlew --no-daemon pythonBenchmarkAllLanguagesBatch
 cp build/reports/jmh/benchmark-corpora.csv \
-    docs/benchmarks/data/java-benchmark-corpora-2026-08-25.csv
-cp build/reports/jmh/stemmer-accuracy-2026-08-25.csv \
-    docs/benchmarks/data/java-stemmer-accuracy-2026-08-25.csv
-cp build/reports/jmh/stemmer-speed-2026-08-25.csv \
-    docs/benchmarks/data/java-stemmer-speed-2026-08-25.csv
-cp build/reports/jmh/english-coverage-accuracy-2026-08-25.csv \
-    docs/benchmarks/data/java-english-coverage-accuracy-2026-08-25.csv
-cp build/reports/jmh/english-coverage-speed-2026-08-25.csv \
-    docs/benchmarks/data/java-english-coverage-speed-2026-08-25.csv
+    docs/benchmarks/data/java-benchmark-corpora-2026-09-10.csv
+cp build/reports/jmh/stemmer-accuracy-2026-09-10.csv \
+    docs/benchmarks/data/java-stemmer-accuracy-2026-09-10.csv
+cp build/reports/jmh/stemmer-speed-2026-09-10.csv \
+    docs/benchmarks/data/java-stemmer-speed-2026-09-10.csv
+cp build/reports/jmh/english-coverage-accuracy-2026-09-10.csv \
+    docs/benchmarks/data/java-english-coverage-accuracy-2026-09-10.csv
+cp build/reports/jmh/english-coverage-speed-2026-09-10.csv \
+    docs/benchmarks/data/java-english-coverage-speed-2026-09-10.csv
 cp build/reports/python-benchmarks/all-languages-batch.csv \
     docs/benchmarks/data/python-all-languages-batch-2026-08-25.csv
 cp build/reports/python-benchmarks/all-languages-batch.json \
@@ -115,17 +130,22 @@ sed -i "s#${PWD}#<repository>#g" \
     java-stemmer-speed-2026-08-25.csv \
     java-english-coverage-accuracy-2026-08-25.csv \
     java-english-coverage-speed-2026-08-25.csv \
+    java-benchmark-corpora-2026-09-10.csv \
+    java-stemmer-accuracy-2026-09-10.csv \
+    java-stemmer-speed-2026-09-10.csv \
+    java-english-coverage-accuracy-2026-09-10.csv \
+    java-english-coverage-speed-2026-09-10.csv \
     python-all-languages-batch-2026-08-25.csv \
     python-all-languages-batch-2026-08-25.json \
     > performance-snapshots.sha256)
 python3 tools/update-benchmark-documentation.py \
-    --corpus docs/benchmarks/data/java-benchmark-corpora-2026-08-25.csv \
-    --accuracy docs/benchmarks/data/java-stemmer-accuracy-2026-08-25.csv \
-    --speed docs/benchmarks/data/java-stemmer-speed-2026-08-25.csv \
-    --coverage-accuracy docs/benchmarks/data/java-english-coverage-accuracy-2026-08-25.csv \
-    --coverage-speed docs/benchmarks/data/java-english-coverage-speed-2026-08-25.csv \
-    --date 2026-08-25 \
-    --release-version 4.2.0-6-g84e57fb \
+    --corpus docs/benchmarks/data/java-benchmark-corpora-2026-09-10.csv \
+    --accuracy docs/benchmarks/data/java-stemmer-accuracy-2026-09-10.csv \
+    --speed docs/benchmarks/data/java-stemmer-speed-2026-09-10.csv \
+    --coverage-accuracy docs/benchmarks/data/java-english-coverage-accuracy-2026-09-10.csv \
+    --coverage-speed docs/benchmarks/data/java-english-coverage-speed-2026-09-10.csv \
+    --date 2026-09-10 \
+    --release-version 4.3.0.dirty \
     --mode update
 python3 tools/update-python-benchmark-documentation.py \
     --csv docs/benchmarks/data/python-all-languages-batch-2026-08-25.csv \
@@ -157,7 +177,7 @@ the checked-in catalog and must never rewrite it ad hoc.
 
 Audit mode is enabled with `-PstemmingQualityAudit=true`. Language, stemmer, dictionary-mode, output-policy, and ranking filters are documented on the central [stemming-quality page](../../stemming-quality.md). Filtered reports use separate filenames and cannot be accepted as publication sources.
 
-`publishStemmingQualityDocumentation` validates the complete build CSV, copies a versioned documentation snapshot, and replaces only marked generated sections. `verifyStemmingQualityDocumentation` re-renders from the checked-in snapshot and fails on changed values, ordering, missing pages, duplicate keys, arithmetic inconsistencies, policy violations, or stale sections.
+`publishStemmingQualityDocumentation` validates the complete build CSV, copies it to a new dated documentation snapshot, updates the active pointer, and replaces only marked generated sections. It never overwrites the historical `stemming-quality.csv`. `verifyStemmingQualityDocumentation` re-renders from the active checked-in snapshot and fails on changed values, ordering, missing pages, duplicate keys, arithmetic inconsistencies, policy violations, or stale sections.
 
 The staged site under `build/` is disposable output. The canonical model catalog is the reviewed,
 checked-in `docs/stemmer-model-catalog.md`; the staged copy must remain identical to it.
@@ -181,10 +201,11 @@ not recorded for the legacy UniMorph imports; that limitation is disclosed with
 upstream revision and source-archive checksum. This reproducibility limitation does not replace or
 weaken the packaged license and attribution requirements.
 
-Each UniMorph-derived model artifact carries its own notice with the canonical CC BY-SA 3.0 URI,
-upstream attribution, transformations, ShareAlike statement, and Leo Galambos contribution notice.
-The full CC legal text is not duplicated or presented as a root-project license. PoliMorf retains
-its separately packaged BSD-2-Clause license.
+Each UniMorph-derived model artifact carries its own notice with the exact audited license URI,
+upstream attribution, transformations, applicable distribution statement, and Leo Galambos
+contribution notice. Most are CC BY-SA 3.0; CC BY-SA 4.0, CC BY 4.0, and LGPLLR exceptions are
+identified exactly. Khaling additionally packages the canonical LGPLLR text and legible dictionary
+form. PoliMorf retains its separately packaged BSD-2-Clause license.
 
 For a future full PoliMorf measurement, also record the startup heap separately from benchmark parameters. Complete runtime construction is currently verified with a dedicated 6 GiB maximum heap; this limit is neither a retained-trie measurement nor a setting applied to ordinary JMH runs.
 
@@ -196,15 +217,15 @@ The current accuracy, speed, and coverage commands are:
 
 ```bash
 ./gradlew --no-daemon benchmarkCorpusReport writeJmhRuntimeClasspath
-tools/run-published-accuracy-benchmarks.sh 2026-08-25
-tools/run-published-speed-benchmarks.sh 2026-08-25 4.2.0-6-g84e57fb
+tools/run-published-accuracy-benchmarks.sh 2026-09-10
+tools/run-published-speed-benchmarks.sh 2026-09-10 4.3.0.dirty
 ```
 
-The speed runner refuses to start unless every CPU uses the `performance` governor, materializes the exact selected benchmark list, rejects quality/Polimorf/gold-standard methods, and requires the Hebrew speed path. It records hardware, JVM, source-state, JAR, classpath, corpus, quality, load, temperature, and governor provenance before running. The accuracy runner evaluates all four exact-root benchmark classes and verifies that every new Snowball 3.1.0 candidate exposes all six accuracy counters. The exact JMH configuration is listed in [Environment and reports](environment.md). Quality and performance reports are separate datasets and are not combined into an undocumented scalar.
+The speed runner refuses to start unless every CPU uses the `performance` governor, materializes the exact selected benchmark list, and rejects quality, gold-standard, and internal microbenchmark methods. It validates that the parameterized Radixor method covers all 167 user-facing model IDs and selects only the approved same-language comparator paths, including the separately identified optional PoliMorf comparator. It records hardware, JVM, source state, executable JAR, the complete runtime-classpath content manifest, corpus, quality, load, temperature, and governor provenance. After the main CSV passes model-coverage validation, the runner also records its SHA-256 before the coverage run. The accuracy runner evaluates the exact-root benchmark suite and verifies the complete counter rows for every selected Snowball 3.1.0 candidate. The exact JMH configuration is listed in [Environment and reports](environment.md). Quality and performance reports are separate datasets and are not combined into an undocumented scalar.
 
 ## Recorded and unavailable provenance
 
-The performance documentation records its 2026-08-25 environment, JDK, operating system, hardware, base revision, exact measured-source patch, untracked-source checksums, executable JMH JAR checksum, and model descriptor checksums. The quality CSV embeds model identity and checksum in every row; run date, core source state, JVM, OS, and hardware are shared provenance on the environment page.
+The performance documentation records its 2026-09-10 environment, JDK, operating system, hardware, base revision, exact measured-source patch, untracked-source checksums, executable JMH JAR checksum, complete runtime-classpath content manifest, main-report checksum, and model descriptor checksums. The quality CSV embeds model identity and checksum in every row; run date, core source state, JVM, OS, and hardware are shared provenance on the environment page.
 
 Exact immutable upstream revisions were not recorded for every legacy UniMorph import. That limitation remains explicit in model descriptors and cannot be repaired from filesystem timestamps. Dependency versions reproducible from repository configuration include Apache Lucene 10.5.0, Morfologik 2.1.9, the Ukrainian dictionary artifact 4.9.1, and JMH 1.37.
 
diff --git a/docs/benchmarks/reference/tested-stemmers.md b/docs/benchmarks/reference/tested-stemmers.md
index 90100e6..2c62d6e 100644
--- a/docs/benchmarks/reference/tested-stemmers.md
+++ b/docs/benchmarks/reference/tested-stemmers.md
@@ -4,10 +4,10 @@ The JMH adapter registry is authoritative for evaluated implementations and lang
 
 | Family or implementation | Upstream / attribution | Tested version or revision | Evaluated scope | Output capability and adapter behaviour | Interpretation notes |
 | --- | --- | --- | --- | --- | --- |
-| Radixor | Egothor / Radixor project | Base commit and measured working-tree state recorded on the environment page | All 20 reconciled default model languages; all 20 have benchmark pages | Deterministic preferred patch via `get`; ranked distinct alternatives via `getAll`; primary is always included | Dictionary-trained compiled patch trie. Default rows use each language's stable default model ID. |
+| Radixor | Egothor / Radixor project | Base commit and measured working-tree state recorded on the environment page | All 166 default languages have quality and benchmark pages; speed covers 167 user-facing models including optional PoliMorf | Deterministic preferred patch via `get`; ranked distinct alternatives via `getAll`; primary is always included | Dictionary-trained compiled patch trie. Default rows use each language's stable default model ID. |
 | Apache Lucene language stem filters | Apache Lucene project | 10.5.0 | Adapter-declared language-specific subsets | TokenFilter lifecycle and language normalization match JMH; normally single-output | Light, minimal, possessive, and language stem filters deliberately implement different scopes. Narrow scope is not a defect. |
 | Apache Lucene SnowballFilter | Apache Lucene project using Snowball algorithms | Lucene 10.5.0 | Snowball-supported subset of Radixor languages | Single primary token emitted through the Lucene TokenFilter path | Includes TokenStream overhead and required normalization. |
-| Official Snowball Java | Snowball project | 3.1.0 source distribution; SHA-256 `5dab34d491f55f47b6e971569ffe6aadf5991512c648ddfe5d331b494cf6d655` | 17 same-language direct adapters, including the Czech, Persian, and Polish stemmers added in 3.1.0 | Direct generated Java API; single output | Rule-based suffix algorithms provide broad baselines rather than dictionary-root guarantees. Lucene 10.5.0 does not yet expose the three new algorithms through `SnowballFilter`, so those rows are direct-only. |
+| Official Snowball Java | Snowball project | 3.1.0 source distribution; SHA-256 `5dab34d491f55f47b6e971569ffe6aadf5991512c648ddfe5d331b494cf6d655` | 29 exact-language direct adapters; 28 participate in the homepage aggregate and root-only Arabic is charted separately | Direct generated Java API; single output | Rule-based suffix algorithms provide broad baselines rather than dictionary-root guarantees. Lucene 10.5.0 exposes 25 of these algorithms through `SnowballFilter`; the remaining four comparisons use the direct API only. |
 | Lucene Stempel | Apache Lucene / Polish stemming tables | Lucene 10.5.0 | Polish | Direct and TokenFilter paths where registered; single primary output | Table-driven Polish implementation. |
 | Morfologik | Morfologik project; Lucene integration by Apache Lucene | Morfologik 2.1.9, Lucene integration 10.5.0; Ukrainian dictionary artifact 4.9.1 | Registered Polish and Ukrainian paths | Deterministic first lemma for primary comparison; all distinct lemma strings for candidate policies | Several analyses may share a lemma and are deduplicated by exact string equality. |
 | Hunspell via Lucene | Hunspell dictionaries from the `wooorm/dictionaries` repository; adapter by Apache Lucene | Lucene 10.5.0; dictionary repository revision was not recorded | Configured German, English, Spanish, French, Dutch, Polish, and Ukrainian dictionaries | First emitted stem is primary; all distinct stems at the token position are candidates | Dictionary content and affix rules differ by language. |
diff --git a/docs/built-in-languages.md b/docs/built-in-languages.md
index 8eb027f..68eeb21 100644
--- a/docs/built-in-languages.md
+++ b/docs/built-in-languages.md
@@ -1,10 +1,10 @@
 # Built-in Languages and Default Models
 
-“Supported language” means that the repository maintains a default dictionary
-and runtime mapping for that language. Packaging differs by runtime: Java keeps
-the core dictionary-free and resolves external model artifacts, while the
-Python installs a separate `radixor-models-standard` data package containing
-the 20 default dictionaries in precompiled version 7 form.
+“Registered language” means that the repository maintains a dictionary and a
+stable Java runtime mapping for that language. Packaging differs by runtime:
+Java keeps the core dictionary-free and resolves individual model artifacts,
+while the Java and Python `radixor-models-standard` aggregates intentionally
+remain limited to the original 20 standard dictionaries.
 
 The Java language enum carries language identity, writing direction, a legacy
 resource-directory name, and the stable default model ID. A Java model
@@ -13,33 +13,158 @@ Python accepts the short alias or the same full model ID. See
 [Model Selection and Loading](model-selection-and-loading.md) for Java and
 [Python Usage and API](python/usage.md) for Python.
 
-## Defaults and variants
-
-| Language | Java enum | Python alias | Default model ID | Java default artifact | Optional variants |
-|---|---|---|---|---|---|
-| Czech | `CS_CZ` | `cs` | `cs-cz-default` | `org.egothor:radixor-model-cs-cz-default` | — |
-| Danish | `DA_DK` | `da` | `da-dk-default` | `org.egothor:radixor-model-da-dk-default` | — |
-| German | `DE_DE` | `de` | `de-de-default` | `org.egothor:radixor-model-de-de-default` | — |
-| Spanish | `ES_ES` | `es` | `es-es-default` | `org.egothor:radixor-model-es-es-default` | — |
-| Persian | `FA_IR` | `fa` | `fa-ir-default` | `org.egothor:radixor-model-fa-ir-default` | — |
-| Finnish | `FI_FI` | `fi` | `fi-fi-default` | `org.egothor:radixor-model-fi-fi-default` | — |
-| French | `FR_FR` | `fr` | `fr-fr-default` | `org.egothor:radixor-model-fr-fr-default` | — |
-| Hebrew | `HE_IL` | `he` | `he-il-default` | `org.egothor:radixor-model-he-il-default` | — |
-| Hungarian | `HU_HU` | `hu` | `hu-hu-default` | `org.egothor:radixor-model-hu-hu-default` | — |
-| Italian | `IT_IT` | `it` | `it-it-default` | `org.egothor:radixor-model-it-it-default` | — |
-| Norwegian Bokmål | `NB_NO` | `nb` | `nb-no-default` | `org.egothor:radixor-model-nb-no-default` | — |
-| Dutch | `NL_NL` | `nl` | `nl-nl-default` | `org.egothor:radixor-model-nl-nl-default` | — |
-| Norwegian Nynorsk | `NN_NO` | `nn` | `nn-no-default` | `org.egothor:radixor-model-nn-no-default` | — |
-| Polish | `PL_PL` | `pl` | `pl-pl-unimorph` | `org.egothor:radixor-model-pl-pl-unimorph` | `pl-pl-polimorf` / `org.egothor:radixor-model-pl-pl-polimorf` |
-| Portuguese | `PT_PT` | `pt` | `pt-pt-default` | `org.egothor:radixor-model-pt-pt-default` | — |
-| Russian | `RU_RU` | `ru` | `ru-ru-default` | `org.egothor:radixor-model-ru-ru-default` | — |
-| Swedish | `SV_SE` | `sv` | `sv-se-default` | `org.egothor:radixor-model-sv-se-default` | — |
-| Ukrainian | `UK_UA` | `uk` | `uk-ua-default` | `org.egothor:radixor-model-uk-ua-default` | — |
-| English | `US_UK` | `en` | `us-uk-default` | `org.egothor:radixor-model-us-uk-default` | — |
-| Yiddish | `YI` | `yi` | `yi-default` | `org.egothor:radixor-model-yi-default` | — |
-
-The maintained table deliberately avoids duplicating mutable provenance and checksum fields. Those values come from module metadata and are generated into the model catalog.
-
+## Registered languages and models
+
+The standard Java and Python aggregates remain fixed at the original 20 `default` entries. Every `standalone` entry is a registered Java language default and an individually publishable artifact, but is not installed by either standard aggregate. The optional PoliMorf model is also individual only.
+
+| Language | Java enum | Model ID | Topology role | Relative dictionary size |
+| --- | --- | --- | --- | --- |
+| Adyghe | `ADY` | `ady-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 20,334 distinct usable word forms" title="Relative dictionary size 4 of 5; 20,334 distinct usable word forms">★★★★☆</span> (20,334 distinct forms) |
+| Afrikaans | `AF_ZA` | `af-za-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 289,303 distinct usable word forms" title="Relative dictionary size 5 of 5; 289,303 distinct usable word forms">★★★★★</span> (289,303 distinct forms) |
+| Aimele | `AIL` | `ail-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 2,978 distinct usable word forms" title="Relative dictionary size 3 of 5; 2,978 distinct usable word forms">★★★☆☆</span> (2,978 distinct forms) |
+| Akan | `AK` | `aka-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 1,890 distinct usable word forms" title="Relative dictionary size 2 of 5; 1,890 distinct usable word forms">★★☆☆☆</span> (1,890 distinct forms) |
+| Albanian | `SQ_AL` | `sq-al-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 10,218 distinct usable word forms" title="Relative dictionary size 3 of 5; 10,218 distinct usable word forms">★★★☆☆</span> (10,218 distinct forms) |
+| Alsatian | `GSW` | `gsw-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 1,171 distinct usable word forms" title="Relative dictionary size 2 of 5; 1,171 distinct usable word forms">★★☆☆☆</span> (1,171 distinct forms) |
+| Amharic | `AM_ET` | `am-et-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 41,308 distinct usable word forms" title="Relative dictionary size 4 of 5; 41,308 distinct usable word forms">★★★★☆</span> (41,308 distinct forms) |
+| Amol Azerbaijani | `AZG` | `azg-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 9,357 distinct usable word forms" title="Relative dictionary size 3 of 5; 9,357 distinct usable word forms">★★★☆☆</span> (9,357 distinct forms) |
+| Ancient Greek | `GRC` | `grc-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 11,109 distinct usable word forms" title="Relative dictionary size 4 of 5; 11,109 distinct usable word forms">★★★★☆</span> (11,109 distinct forms) |
+| Anglo-Norman | `XNO` | `xno-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 186 distinct usable word forms" title="Relative dictionary size 1 of 5; 186 distinct usable word forms">★☆☆☆☆</span> (186 distinct forms) |
+| Arabic | `AR` | `ar-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 452,974 distinct usable word forms" title="Relative dictionary size 5 of 5; 452,974 distinct usable word forms">★★★★★</span> (452,974 distinct forms) |
+| Armenian | `HY_AM` | `hy-am-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 242,091 distinct usable word forms" title="Relative dictionary size 5 of 5; 242,091 distinct usable word forms">★★★★★</span> (242,091 distinct forms) |
+| Ashaninka | `CNI` | `cni-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 10,095 distinct usable word forms" title="Relative dictionary size 3 of 5; 10,095 distinct usable word forms">★★★☆☆</span> (10,095 distinct forms) |
+| Assamese | `AS_IN` | `as-in-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 328 distinct usable word forms" title="Relative dictionary size 1 of 5; 328 distinct usable word forms">★☆☆☆☆</span> (328 distinct forms) |
+| Asturian | `AST` | `ast-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 21,584 distinct usable word forms" title="Relative dictionary size 4 of 5; 21,584 distinct usable word forms">★★★★☆</span> (21,584 distinct forms) |
+| Aymara | `AYM` | `aym-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 231,247 distinct usable word forms" title="Relative dictionary size 5 of 5; 231,247 distinct usable word forms">★★★★★</span> (231,247 distinct forms) |
+| Azerbaijani | `AZ_AZ` | `az-az-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 6,669 distinct usable word forms" title="Relative dictionary size 3 of 5; 6,669 distinct usable word forms">★★★☆☆</span> (6,669 distinct forms) |
+| Bashkir | `BAK` | `bak-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 10,616 distinct usable word forms" title="Relative dictionary size 4 of 5; 10,616 distinct usable word forms">★★★★☆</span> (10,616 distinct forms) |
+| Belarusian | `BE_BY` | `be-by-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 678 distinct usable word forms" title="Relative dictionary size 2 of 5; 678 distinct usable word forms">★★☆☆☆</span> (678 distinct forms) |
+| Bengali | `BN_BD` | `bn-bd-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 3 distinct usable word forms" title="Relative dictionary size 1 of 5; 3 distinct usable word forms">★☆☆☆☆</span> (3 distinct forms) |
+| Bininj Kun-wok | `GUP` | `gup-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 370 distinct usable word forms" title="Relative dictionary size 1 of 5; 370 distinct usable word forms">★☆☆☆☆</span> (370 distinct forms) |
+| Braj | `BRA` | `bra-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 115 distinct usable word forms" title="Relative dictionary size 1 of 5; 115 distinct usable word forms">★☆☆☆☆</span> (115 distinct forms) |
+| Breton | `BRE` | `bre-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 1,682 distinct usable word forms" title="Relative dictionary size 2 of 5; 1,682 distinct usable word forms">★★☆☆☆</span> (1,682 distinct forms) |
+| Bulgarian | `BG_BG` | `bg-bg-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 37,335 distinct usable word forms" title="Relative dictionary size 4 of 5; 37,335 distinct usable word forms">★★★★☆</span> (37,335 distinct forms) |
+| Catalan | `CA_ES` | `ca-es-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 61,495 distinct usable word forms" title="Relative dictionary size 5 of 5; 61,495 distinct usable word forms">★★★★★</span> (61,495 distinct forms) |
+| Cebuano | `CEB` | `ceb-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 392 distinct usable word forms" title="Relative dictionary size 1 of 5; 392 distinct usable word forms">★☆☆☆☆</span> (392 distinct forms) |
+| Chichewa | `NY_MW` | `ny-mw-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 3,150 distinct usable word forms" title="Relative dictionary size 3 of 5; 3,150 distinct usable word forms">★★★☆☆</span> (3,150 distinct forms) |
+| Chukchi | `CKT` | `ckt-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 302 distinct usable word forms" title="Relative dictionary size 1 of 5; 302 distinct usable word forms">★☆☆☆☆</span> (302 distinct forms) |
+| Church Slavonic | `CHU` | `chu-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 1,632 distinct usable word forms" title="Relative dictionary size 2 of 5; 1,632 distinct usable word forms">★★☆☆☆</span> (1,632 distinct forms) |
+| Classical Armenian | `XCL` | `xcl-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 55,946 distinct usable word forms" title="Relative dictionary size 4 of 5; 55,946 distinct usable word forms">★★★★☆</span> (55,946 distinct forms) |
+| Classical Syriac | `SYC` | `syc-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 2,139 distinct usable word forms" title="Relative dictionary size 3 of 5; 2,139 distinct usable word forms">★★★☆☆</span> (2,139 distinct forms) |
+| Congo Swahili | `SWC` | `swc-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 6,700 distinct usable word forms" title="Relative dictionary size 3 of 5; 6,700 distinct usable word forms">★★★☆☆</span> (6,700 distinct forms) |
+| Copala Triqui | `CPA` | `cpa-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 1,751 distinct usable word forms" title="Relative dictionary size 2 of 5; 1,751 distinct usable word forms">★★☆☆☆</span> (1,751 distinct forms) |
+| Cornish | `COR` | `cor-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 159 distinct usable word forms" title="Relative dictionary size 1 of 5; 159 distinct usable word forms">★☆☆☆☆</span> (159 distinct forms) |
+| Cree | `CRE` | `cre-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 514 distinct usable word forms" title="Relative dictionary size 2 of 5; 514 distinct usable word forms">★★☆☆☆</span> (514 distinct forms) |
+| Crimean Tatar | `CRH` | `crh-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 7,183 distinct usable word forms" title="Relative dictionary size 3 of 5; 7,183 distinct usable word forms">★★★☆☆</span> (7,183 distinct forms) |
+| Czech | `CS_CZ` | `cs-cz-default` | `default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 51,401 distinct usable word forms" title="Relative dictionary size 4 of 5; 51,401 distinct usable word forms">★★★★☆</span> (51,401 distinct forms) |
+| Dakota | `DAK` | `dak-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 2,942 distinct usable word forms" title="Relative dictionary size 3 of 5; 2,942 distinct usable word forms">★★★☆☆</span> (2,942 distinct forms) |
+| Danish | `DA_DK` | `da-dk-default` | `default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 27,921 distinct usable word forms" title="Relative dictionary size 4 of 5; 27,921 distinct usable word forms">★★★★☆</span> (27,921 distinct forms) |
+| Dutch | `NL_NL` | `nl-nl-default` | `default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 26,201 distinct usable word forms" title="Relative dictionary size 4 of 5; 26,201 distinct usable word forms">★★★★☆</span> (26,201 distinct forms) |
+| Eastern Chatino | `CLY` | `cly-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 2,030 distinct usable word forms" title="Relative dictionary size 2 of 5; 2,030 distinct usable word forms">★★☆☆☆</span> (2,030 distinct forms) |
+| Egyptian Arabic | `ARZ` | `arz-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 135 distinct usable word forms" title="Relative dictionary size 1 of 5; 135 distinct usable word forms">★☆☆☆☆</span> (135 distinct forms) |
+| English | `US_UK` | `us-uk-default` | `default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 591,946 distinct usable word forms" title="Relative dictionary size 5 of 5; 591,946 distinct usable word forms">★★★★★</span> (591,946 distinct forms) |
+| Estonian | `ET_EE` | `et-ee-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 24,811 distinct usable word forms" title="Relative dictionary size 4 of 5; 24,811 distinct usable word forms">★★★★☆</span> (24,811 distinct forms) |
+| Evenki | `EVN` | `evn-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 11,965 distinct usable word forms" title="Relative dictionary size 4 of 5; 11,965 distinct usable word forms">★★★★☆</span> (11,965 distinct forms) |
+| Faroese | `FO_FO` | `fo-fo-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 31,366 distinct usable word forms" title="Relative dictionary size 4 of 5; 31,366 distinct usable word forms">★★★★☆</span> (31,366 distinct forms) |
+| Finnish | `FI_FI` | `fi-fi-default` | `default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 1,788,784 distinct usable word forms" title="Relative dictionary size 5 of 5; 1,788,784 distinct usable word forms">★★★★★</span> (1,788,784 distinct forms) |
+| French | `FR_FR` | `fr-fr-default` | `default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 404,011 distinct usable word forms" title="Relative dictionary size 5 of 5; 404,011 distinct usable word forms">★★★★★</span> (404,011 distinct forms) |
+| Friulian | `FUR` | `fur-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 5,007 distinct usable word forms" title="Relative dictionary size 3 of 5; 5,007 distinct usable word forms">★★★☆☆</span> (5,007 distinct forms) |
+| Ga | `GAA` | `gaa-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 469 distinct usable word forms" title="Relative dictionary size 1 of 5; 469 distinct usable word forms">★☆☆☆☆</span> (469 distinct forms) |
+| Galolen | `GAL` | `gal-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 25,436 distinct usable word forms" title="Relative dictionary size 4 of 5; 25,436 distinct usable word forms">★★★★☆</span> (25,436 distinct forms) |
+| German | `DE_DE` | `de-de-default` | `default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 277,266 distinct usable word forms" title="Relative dictionary size 5 of 5; 277,266 distinct usable word forms">★★★★★</span> (277,266 distinct forms) |
+| Gothic | `GOT` | `got-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 134,162 distinct usable word forms" title="Relative dictionary size 5 of 5; 134,162 distinct usable word forms">★★★★★</span> (134,162 distinct forms) |
+| Greek | `EL_GR` | `el-gr-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 76,851 distinct usable word forms" title="Relative dictionary size 5 of 5; 76,851 distinct usable word forms">★★★★★</span> (76,851 distinct forms) |
+| Gulf Arabic | `AFB` | `afb-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 355 distinct usable word forms" title="Relative dictionary size 1 of 5; 355 distinct usable word forms">★☆☆☆☆</span> (355 distinct forms) |
+| Haida | `HAI` | `hai-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 520 distinct usable word forms" title="Relative dictionary size 2 of 5; 520 distinct usable word forms">★★☆☆☆</span> (520 distinct forms) |
+| Hebrew | `HE_IL` | `he-il-default` | `default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 57,658 distinct usable word forms" title="Relative dictionary size 4 of 5; 57,658 distinct usable word forms">★★★★☆</span> (57,658 distinct forms) |
+| Hiligaynon | `HIL` | `hil-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 308 distinct usable word forms" title="Relative dictionary size 1 of 5; 308 distinct usable word forms">★☆☆☆☆</span> (308 distinct forms) |
+| Hsilimo | `HSI` | `hsi-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 158 distinct usable word forms" title="Relative dictionary size 1 of 5; 158 distinct usable word forms">★☆☆☆☆</span> (158 distinct forms) |
+| Hungarian | `HU_HU` | `hu-hu-default` | `default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 910,688 distinct usable word forms" title="Relative dictionary size 5 of 5; 910,688 distinct usable word forms">★★★★★</span> (910,688 distinct forms) |
+| Icelandic | `IS_IS` | `is-is-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 52,132 distinct usable word forms" title="Relative dictionary size 4 of 5; 52,132 distinct usable word forms">★★★★☆</span> (52,132 distinct forms) |
+| Indonesian | `ID_ID` | `id-id-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 21,296 distinct usable word forms" title="Relative dictionary size 4 of 5; 21,296 distinct usable word forms">★★★★☆</span> (21,296 distinct forms) |
+| Ingrian | `IZH` | `izh-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 1,024 distinct usable word forms" title="Relative dictionary size 2 of 5; 1,024 distinct usable word forms">★★☆☆☆</span> (1,024 distinct forms) |
+| Irish | `GA_IE` | `ga-ie-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 21,829 distinct usable word forms" title="Relative dictionary size 4 of 5; 21,829 distinct usable word forms">★★★★☆</span> (21,829 distinct forms) |
+| Italian | `IT_IT` | `it-it-default` | `default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 324,366 distinct usable word forms" title="Relative dictionary size 5 of 5; 324,366 distinct usable word forms">★★★★★</span> (324,366 distinct forms) |
+| Itelmen | `ITL` | `itl-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 3,546 distinct usable word forms" title="Relative dictionary size 3 of 5; 3,546 distinct usable word forms">★★★☆☆</span> (3,546 distinct forms) |
+| Japanese | `JA_JP` | `ja-jp-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 10,848 distinct usable word forms" title="Relative dictionary size 4 of 5; 10,848 distinct usable word forms">★★★★☆</span> (10,848 distinct forms) |
+| Kabardian | `KBD` | `kbd-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 3,054 distinct usable word forms" title="Relative dictionary size 3 of 5; 3,054 distinct usable word forms">★★★☆☆</span> (3,054 distinct forms) |
+| Kalaallisut | `KL_GL` | `kl-gl-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 321 distinct usable word forms" title="Relative dictionary size 1 of 5; 321 distinct usable word forms">★☆☆☆☆</span> (321 distinct forms) |
+| Kannada | `KN_IN` | `kn-in-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 11 distinct usable word forms" title="Relative dictionary size 1 of 5; 11 distinct usable word forms">★☆☆☆☆</span> (11 distinct forms) |
+| Karelian | `KRL` | `krl-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 566 distinct usable word forms" title="Relative dictionary size 2 of 5; 566 distinct usable word forms">★★☆☆☆</span> (566 distinct forms) |
+| Kashubian | `CSB` | `csb-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 350 distinct usable word forms" title="Relative dictionary size 1 of 5; 350 distinct usable word forms">★☆☆☆☆</span> (350 distinct forms) |
+| Kazakh | `KK_KZ` | `kk-kz-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 357 distinct usable word forms" title="Relative dictionary size 1 of 5; 357 distinct usable word forms">★☆☆☆☆</span> (357 distinct forms) |
+| Khakas | `KJH` | `kjh-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 1,172 distinct usable word forms" title="Relative dictionary size 2 of 5; 1,172 distinct usable word forms">★★☆☆☆</span> (1,172 distinct forms) |
+| Khinalugh | `KLR` | `klr-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 41,185 distinct usable word forms" title="Relative dictionary size 4 of 5; 41,185 distinct usable word forms">★★★★☆</span> (41,185 distinct forms) |
+| Kodi | `KOD` | `kod-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 524 distinct usable word forms" title="Relative dictionary size 2 of 5; 524 distinct usable word forms">★★☆☆☆</span> (524 distinct forms) |
+| Kongo | `KON` | `kon-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 585 distinct usable word forms" title="Relative dictionary size 2 of 5; 585 distinct usable word forms">★★☆☆☆</span> (585 distinct forms) |
+| Kyrgyz | `KY_KG` | `ky-kg-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 2,997 distinct usable word forms" title="Relative dictionary size 3 of 5; 2,997 distinct usable word forms">★★★☆☆</span> (2,997 distinct forms) |
+| Ladin | `LLD` | `lld-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 4,819 distinct usable word forms" title="Relative dictionary size 3 of 5; 4,819 distinct usable word forms">★★★☆☆</span> (4,819 distinct forms) |
+| Latin | `LA` | `la-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 485,259 distinct usable word forms" title="Relative dictionary size 5 of 5; 485,259 distinct usable word forms">★★★★★</span> (485,259 distinct forms) |
+| Latvian | `LV_LV` | `lv-lv-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 75,492 distinct usable word forms" title="Relative dictionary size 5 of 5; 75,492 distinct usable word forms">★★★★★</span> (75,492 distinct forms) |
+| Lingala | `LIN` | `lin-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 230 distinct usable word forms" title="Relative dictionary size 1 of 5; 230 distinct usable word forms">★☆☆☆☆</span> (230 distinct forms) |
+| Lithuanian | `LT_LT` | `lt-lt-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 27,335 distinct usable word forms" title="Relative dictionary size 4 of 5; 27,335 distinct usable word forms">★★★★☆</span> (27,335 distinct forms) |
+| Livonian | `LIV` | `liv-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 2,841 distinct usable word forms" title="Relative dictionary size 3 of 5; 2,841 distinct usable word forms">★★★☆☆</span> (2,841 distinct forms) |
+| Low German | `NDS` | `nds-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 2,493 distinct usable word forms" title="Relative dictionary size 3 of 5; 2,493 distinct usable word forms">★★★☆☆</span> (2,493 distinct forms) |
+| Lower Sorbian | `DSB` | `dsb-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 12,038 distinct usable word forms" title="Relative dictionary size 4 of 5; 12,038 distinct usable word forms">★★★★☆</span> (12,038 distinct forms) |
+| Luganda | `LG_UG` | `lg-ug-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 4,671 distinct usable word forms" title="Relative dictionary size 3 of 5; 4,671 distinct usable word forms">★★★☆☆</span> (4,671 distinct forms) |
+| Macedonian | `MK_MK` | `mk-mk-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 135,590 distinct usable word forms" title="Relative dictionary size 5 of 5; 135,590 distinct usable word forms">★★★★★</span> (135,590 distinct forms) |
+| Magahi | `MAG` | `mag-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 206 distinct usable word forms" title="Relative dictionary size 1 of 5; 206 distinct usable word forms">★☆☆☆☆</span> (206 distinct forms) |
+| Malagasy | `MG_MG` | `mg-mg-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 636 distinct usable word forms" title="Relative dictionary size 2 of 5; 636 distinct usable word forms">★★☆☆☆</span> (636 distinct forms) |
+| Maltese | `MT_MT` | `mt-mt-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 1,421 distinct usable word forms" title="Relative dictionary size 2 of 5; 1,421 distinct usable word forms">★★☆☆☆</span> (1,421 distinct forms) |
+| Manx | `GV_IM` | `gv-im-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 15 distinct usable word forms" title="Relative dictionary size 1 of 5; 15 distinct usable word forms">★☆☆☆☆</span> (15 distinct forms) |
+| Maori | `MI_NZ` | `mi-nz-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 207 distinct usable word forms" title="Relative dictionary size 1 of 5; 207 distinct usable word forms">★☆☆☆☆</span> (207 distinct forms) |
+| Mapudungun | `ARN` | `arn-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 548 distinct usable word forms" title="Relative dictionary size 2 of 5; 548 distinct usable word forms">★★☆☆☆</span> (548 distinct forms) |
+| Middle French | `FRM` | `frm-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 27,002 distinct usable word forms" title="Relative dictionary size 4 of 5; 27,002 distinct usable word forms">★★★★☆</span> (27,002 distinct forms) |
+| Middle High German | `GMH` | `gmh-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 384 distinct usable word forms" title="Relative dictionary size 1 of 5; 384 distinct usable word forms">★☆☆☆☆</span> (384 distinct forms) |
+| Middle Low German | `GML` | `gml-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 565 distinct usable word forms" title="Relative dictionary size 2 of 5; 565 distinct usable word forms">★★☆☆☆</span> (565 distinct forms) |
+| Mongolian | `MN_MN` | `mn-mn-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 17,231 distinct usable word forms" title="Relative dictionary size 4 of 5; 17,231 distinct usable word forms">★★★★☆</span> (17,231 distinct forms) |
+| Murrinh-Patha | `MWF` | `mwf-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 592 distinct usable word forms" title="Relative dictionary size 2 of 5; 592 distinct usable word forms">★★☆☆☆</span> (592 distinct forms) |
+| Navajo | `NAV` | `nav-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 9,826 distinct usable word forms" title="Relative dictionary size 3 of 5; 9,826 distinct usable word forms">★★★☆☆</span> (9,826 distinct forms) |
+| Neapolitan | `NAP` | `nap-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 1,497 distinct usable word forms" title="Relative dictionary size 2 of 5; 1,497 distinct usable word forms">★★☆☆☆</span> (1,497 distinct forms) |
+| North Frisian | `FRR` | `frr-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 340 distinct usable word forms" title="Relative dictionary size 1 of 5; 340 distinct usable word forms">★☆☆☆☆</span> (340 distinct forms) |
+| Northern Sami | `SME` | `sme-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 53,034 distinct usable word forms" title="Relative dictionary size 4 of 5; 53,034 distinct usable word forms">★★★★☆</span> (53,034 distinct forms) |
+| Norwegian Bokmål | `NB_NO` | `nb-no-default` | `default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 73,170 distinct usable word forms" title="Relative dictionary size 5 of 5; 73,170 distinct usable word forms">★★★★★</span> (73,170 distinct forms) |
+| Norwegian Nynorsk | `NN_NO` | `nn-no-default` | `default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 16,937 distinct usable word forms" title="Relative dictionary size 4 of 5; 16,937 distinct usable word forms">★★★★☆</span> (16,937 distinct forms) |
+| Old English | `ANG` | `ang-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 65,100 distinct usable word forms" title="Relative dictionary size 5 of 5; 65,100 distinct usable word forms">★★★★★</span> (65,100 distinct forms) |
+| Old French | `FRO` | `fro-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 93,998 distinct usable word forms" title="Relative dictionary size 5 of 5; 93,998 distinct usable word forms">★★★★★</span> (93,998 distinct forms) |
+| Old High German | `GOH` | `goh-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 5,267 distinct usable word forms" title="Relative dictionary size 3 of 5; 5,267 distinct usable word forms">★★★☆☆</span> (5,267 distinct forms) |
+| Old Irish | `SGA` | `sga-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 929 distinct usable word forms" title="Relative dictionary size 2 of 5; 929 distinct usable word forms">★★☆☆☆</span> (929 distinct forms) |
+| Old Norse | `NON` | `non-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 46,067 distinct usable word forms" title="Relative dictionary size 4 of 5; 46,067 distinct usable word forms">★★★★☆</span> (46,067 distinct forms) |
+| Old Saxon | `OSX` | `osx-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 12,361 distinct usable word forms" title="Relative dictionary size 4 of 5; 12,361 distinct usable word forms">★★★★☆</span> (12,361 distinct forms) |
+| Oodham | `OOD` | `ood-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 455 distinct usable word forms" title="Relative dictionary size 1 of 5; 455 distinct usable word forms">★☆☆☆☆</span> (455 distinct forms) |
+| Pashto | `PS_AF` | `ps-af-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 2,945 distinct usable word forms" title="Relative dictionary size 3 of 5; 2,945 distinct usable word forms">★★★☆☆</span> (2,945 distinct forms) |
+| Persian | `FA_IR` | `fa-ir-default` | `default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 3,544 distinct usable word forms" title="Relative dictionary size 3 of 5; 3,544 distinct usable word forms">★★★☆☆</span> (3,544 distinct forms) |
+| Polish | `PL_PL` | `pl-pl-polimorf` | `optional` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 4,668,685 distinct usable word forms" title="Relative dictionary size 5 of 5; 4,668,685 distinct usable word forms">★★★★★</span> (4,668,685 distinct forms) |
+| Polish | `PL_PL` | `pl-pl-unimorph` | `default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 120,867 distinct usable word forms" title="Relative dictionary size 5 of 5; 120,867 distinct usable word forms">★★★★★</span> (120,867 distinct forms) |
+| Portuguese | `PT_PT` | `pt-pt-default` | `default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 211,091 distinct usable word forms" title="Relative dictionary size 5 of 5; 211,091 distinct usable word forms">★★★★★</span> (211,091 distinct forms) |
+| Quechua | `QUE` | `que-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 122,820 distinct usable word forms" title="Relative dictionary size 5 of 5; 122,820 distinct usable word forms">★★★★★</span> (122,820 distinct forms) |
+| Romanian | `RO_RO` | `ro-ro-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 48,497 distinct usable word forms" title="Relative dictionary size 4 of 5; 48,497 distinct usable word forms">★★★★☆</span> (48,497 distinct forms) |
+| Russian | `RU_RU` | `ru-ru-default` | `default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 759,333 distinct usable word forms" title="Relative dictionary size 5 of 5; 759,333 distinct usable word forms">★★★★★</span> (759,333 distinct forms) |
+| Seneca | `SEE` | `see-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 483 distinct usable word forms" title="Relative dictionary size 1 of 5; 483 distinct usable word forms">★☆☆☆☆</span> (483 distinct forms) |
+| Serbo-Croatian | `HBS` | `hbs-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 272,515 distinct usable word forms" title="Relative dictionary size 5 of 5; 272,515 distinct usable word forms">★★★★★</span> (272,515 distinct forms) |
+| Shipibo-Conibo | `SHP` | `shp-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 7,705 distinct usable word forms" title="Relative dictionary size 3 of 5; 7,705 distinct usable word forms">★★★☆☆</span> (7,705 distinct forms) |
+| Shona | `SN_ZW` | `sn-zw-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 2,640 distinct usable word forms" title="Relative dictionary size 3 of 5; 2,640 distinct usable word forms">★★★☆☆</span> (2,640 distinct forms) |
+| Sotho, Southern | `ST_ZA` | `st-za-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 416 distinct usable word forms" title="Relative dictionary size 1 of 5; 416 distinct usable word forms">★☆☆☆☆</span> (416 distinct forms) |
+| Southern Kurdish | `SDH` | `sdh-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 165 distinct usable word forms" title="Relative dictionary size 1 of 5; 165 distinct usable word forms">★☆☆☆☆</span> (165 distinct forms) |
+| Spanish | `ES_ES` | `es-es-default` | `default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 849,661 distinct usable word forms" title="Relative dictionary size 5 of 5; 849,661 distinct usable word forms">★★★★★</span> (849,661 distinct forms) |
+| Swedish | `SV_SE` | `sv-se-default` | `default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 95,181 distinct usable word forms" title="Relative dictionary size 5 of 5; 95,181 distinct usable word forms">★★★★★</span> (95,181 distinct forms) |
+| Tagalog | `TL_PH` | `tl-ph-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 2,042 distinct usable word forms" title="Relative dictionary size 3 of 5; 2,042 distinct usable word forms">★★★☆☆</span> (2,042 distinct forms) |
+| Temoaya Otomi | `OTE` | `ote-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 1,785 distinct usable word forms" title="Relative dictionary size 2 of 5; 1,785 distinct usable word forms">★★☆☆☆</span> (1,785 distinct forms) |
+| Turkish | `TR_TR` | `tr-tr-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 5 of 5; 222,207 distinct usable word forms" title="Relative dictionary size 5 of 5; 222,207 distinct usable word forms">★★★★★</span> (222,207 distinct forms) |
+| Ukrainian | `UK_UA` | `uk-ua-default` | `default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 14,150 distinct usable word forms" title="Relative dictionary size 4 of 5; 14,150 distinct usable word forms">★★★★☆</span> (14,150 distinct forms) |
+| Uyghur | `UG_CN` | `ug-cn-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 6,200 distinct usable word forms" title="Relative dictionary size 3 of 5; 6,200 distinct usable word forms">★★★☆☆</span> (6,200 distinct forms) |
+| Uzbek | `UZ_UZ` | `uz-uz-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 1,080 distinct usable word forms" title="Relative dictionary size 2 of 5; 1,080 distinct usable word forms">★★☆☆☆</span> (1,080 distinct forms) |
+| Voro | `VRO` | `vro-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 277 distinct usable word forms" title="Relative dictionary size 1 of 5; 277 distinct usable word forms">★☆☆☆☆</span> (277 distinct forms) |
+| Western Highland Chatino | `CTP` | `ctp-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 2,456 distinct usable word forms" title="Relative dictionary size 3 of 5; 2,456 distinct usable word forms">★★★☆☆</span> (2,456 distinct forms) |
+| Xibe | `SJO` | `sjo-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 3,151 distinct usable word forms" title="Relative dictionary size 3 of 5; 3,151 distinct usable word forms">★★★☆☆</span> (3,151 distinct forms) |
+| Xipinawa | `XTY` | `xty-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 1,536 distinct usable word forms" title="Relative dictionary size 2 of 5; 1,536 distinct usable word forms">★★☆☆☆</span> (1,536 distinct forms) |
+| Yamana | `AME` | `ame-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 1,489 distinct usable word forms" title="Relative dictionary size 2 of 5; 1,489 distinct usable word forms">★★☆☆☆</span> (1,489 distinct forms) |
+| Yiddish | `YI` | `yi-default` | `default` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 3 of 5; 3,532 distinct usable word forms" title="Relative dictionary size 3 of 5; 3,532 distinct usable word forms">★★★☆☆</span> (3,532 distinct forms) |
+| Zacatepec Chatino | `ZPV` | `zpv-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 2 of 5; 812 distinct usable word forms" title="Relative dictionary size 2 of 5; 812 distinct usable word forms">★★☆☆☆</span> (812 distinct forms) |
+| Zarma | `DJE` | `dje-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 75 distinct usable word forms" title="Relative dictionary size 1 of 5; 75 distinct usable word forms">★☆☆☆☆</span> (75 distinct forms) |
+| Zenzontepec Chatino | `CZN` | `czn-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 1 of 5; 68 distinct usable word forms" title="Relative dictionary size 1 of 5; 68 distinct usable word forms">★☆☆☆☆</span> (68 distinct forms) |
+| Zulu | `ZU_ZA` | `zu-za-default` | `standalone` | <span class="dictionary-rating" role="img" aria-label="Relative dictionary size 4 of 5; 32,384 distinct usable word forms" title="Relative dictionary size 4 of 5; 32,384 distinct usable word forms">★★★★☆</span> (32,384 distinct forms) |
+
+The maintained table deliberately avoids duplicating mutable provenance fields. Those values come from model module metadata and the generated model catalog.
 ## The Polish dual-model case
 
 `PL_PL` represents Polish. It is not an alias for either source dictionary.
@@ -73,7 +198,7 @@ dependencies {
 }
 ```
 
-All documented defaults:
+All 20 standard-aggregate defaults:
 
 ```groovy
 dependencies {
@@ -85,11 +210,13 @@ dependencies {
 The standard pack is metadata-only and excludes optional PoliMorf.
 
 Every individual model artifact carries its own provenance and licensing material. UniMorph
-models carry different model-specific CC BY-SA 3.0 notices because their official language
-repositories identify different lexical sources and contributors. Each notice preserves upstream
+models carry model-specific audited licenses and notices because their official language
+repositories identify different lexical sources, contributors, and applicable terms. Each notice preserves upstream
 attribution and records the Radixor transformations and Leo Galambos contribution statement.
 Legacy imports disclose when an exact historical revision was not recorded; this is a
-reproducibility limitation, not a claim that the source or license is unknown.
+reproducibility limitation, not a claim that the source or license is unknown. The active new-model
+set includes CC BY-SA 3.0, CC BY-SA 4.0, CC BY 4.0, and LGPLLR material; unsupported evidence is
+quarantined rather than assigned a generic license.
 
 ## Loading a language default
 
@@ -105,7 +232,7 @@ The call discovers the default descriptor from the runtime classpath, verifies i
 
 ## Writing direction
 
-Persian, Hebrew, and Yiddish declare right-to-left writing metadata for presentation. Writing direction does not reorder characters in a Java `String`: all built-in natural-language models therefore use backward traversal from the stored sequence end, where suffixes remain located. Explicit forward traversal is reserved for deliberately prefix-oriented custom data. The selected traversal must remain aligned across dictionary parsing, trie lookup, patch generation, persistence, and application; model identity and writing direction are separate concerns.
+Arabic (including Gulf and Egyptian Arabic), Persian, Hebrew, Pashto, Southern Kurdish, Classical Syriac, Uyghur, Urdu, and Yiddish declare right-to-left writing metadata for presentation. Writing direction does not reorder characters in a Java `String`: all built-in natural-language models therefore use backward traversal from the stored sequence end, where suffixes remain located. Explicit forward traversal is reserved for deliberately prefix-oriented custom data. The selected traversal must remain aligned across dictionary parsing, trie lookup, patch generation, persistence, and application; model identity and writing direction are separate concerns.
 
 ## Custom and persisted alternatives
 
diff --git a/docs/dictionary-format.md b/docs/dictionary-format.md
index ce9fffd..4da0e31 100644
--- a/docs/dictionary-format.md
+++ b/docs/dictionary-format.md
@@ -23,8 +23,8 @@ Consequently, compressed size is not a construction-memory estimate. The PoliMor
 
 Comment headers in maintained model inputs summarize provenance but do not replace packaged legal
 material. Each UniMorph-derived artifact includes a language-specific notice describing its
-official repository, lexical source, upstream attribution, CC BY-SA 3.0 canonical URI, ShareAlike
-status, Radixor transformations, and Leo Galambos's protectable model-data contributions. The
+official repository, lexical source, upstream attribution, exact audited license and canonical
+URI, Radixor transformations, and Leo Galambos's protectable model-data contributions. The
 notice does not claim ownership over the underlying data. GZip packaging and descriptor/checksum
 generation are disclosed transformations; the in-memory trie is a Radixor runtime structure.
 
diff --git a/docs/index.md b/docs/index.md
index daa37e9..0e527ea 100644
--- a/docs/index.md
+++ b/docs/index.md
@@ -1,12 +1,12 @@
 ---
 template: landing.html
 title: Radixor
-description: Java flagship and two native Python runtimes using interoperable compiled patch-command tries across 20 languages.
+description: Java flagship and two native Python runtimes using interoperable compiled patch-command tries across 143 language defaults.
 ---
 
 # Radixor
 
 Learned transformation stemming with a flagship Java implementation and two
 native Python choices: batch-oriented `radixor` and scalar-oriented
-`radixor-c`. All use compact reduced tries of patch commands across 20
-languages.
+`radixor-c`. Java publishes 144 individually available models across 143
+language pages; the standard aggregate remains the original 20-model pack.
diff --git a/docs/integration-deep-dive.md b/docs/integration-deep-dive.md
index 67a52cf..c367138 100644
--- a/docs/integration-deep-dive.md
+++ b/docs/integration-deep-dive.md
@@ -78,7 +78,13 @@ module example.search {
 
 The core ships no language dictionary. Add one or more `radixor-model-<model-id>` artifacts, or the optional metadata-only standard pack. Each model JAR contains an indexed descriptor and a namespaced GZip dictionary. `StemmerPatchTrieLoader.Language` represents language properties and a stable default model ID; it does not own embedded data.
 
-The standard option is specifically a POM-only runtime dependency aggregate, not an all-model binary JAR. It resolves one default model JAR per language and excludes optional PoliMorf. The separate POM-only `radixor-models-bom` manages recommended versions without adding runtime artifacts. Repository tests and JMH attach individual model projects directly to non-production configurations, so neither path changes the root publication's dependency graph.
+The standard option is specifically a POM-only runtime dependency aggregate,
+not an all-model binary JAR. It resolves the original 20 standard model JARs
+and excludes the 123 standalone language models and optional PoliMorf. The
+separate POM-only `radixor-models-bom` manages recommended versions for all 144
+user-facing models without adding runtime artifacts. Repository tests and JMH
+attach individual model projects directly to non-production configurations, so
+neither path changes the root publication's dependency graph.
 
 For minimal deployments choose only required model artifacts. For multiple Polish variants add both `pl-pl-unimorph` and `pl-pl-polimorf`, retain UniMorph as the language default, and request PoliMorf explicitly. See [Model Selection and Loading](model-selection-and-loading.md) for complete dependencies and [Built-in Languages](built-in-languages.md) for mappings.
 
diff --git a/docs/migration-and-backward-compatibility.md b/docs/migration-and-backward-compatibility.md
index a6eddb4..4e32b4d 100644
--- a/docs/migration-and-backward-compatibility.md
+++ b/docs/migration-and-backward-compatibility.md
@@ -92,7 +92,7 @@ UniMorph and PoliMorf are not interchangeable quality datasets. They can differ
 
 Model migration does not erase source obligations. Each migrated UniMorph artifact packages its
 language-specific notice with upstream attribution, Radixor modifications and contribution
-statement, ShareAlike terms, and the canonical CC BY-SA 3.0 URI. The original imports did not
+statement, applicable license terms, and the canonical license URI. The original imports did not
 record exact UniMorph commits, so descriptors use
 `source.revision=not-recorded-in-legacy-import` and disclose that fact. Future model imports must
 record an exact upstream revision and source-archive checksum.
@@ -135,7 +135,12 @@ git tag -a "models-catalog@<catalog-version>" -m "Release Radixor model catalog
 
 A core tag publishes only the root `org.egothor:radixor` software artifacts, never model JARs. A model tag validates and publishes exactly its matching module, never core, standard, BOM, JMH, or the multilingual quality suite. A catalog tag publishes only BOM and standard aggregate metadata. Local model dry-run:
 
-The catalog artifacts are POM-only: `radixor-models-standard` carries runtime dependencies on the 20 defaults, while `radixor-models-bom` carries dependency-management constraints for all 21 individual models. Neither publishes an empty binary, sources, or Javadoc JAR. This Maven BOM is distinct from the root CycloneDX SBOM report under `build/reports/sbom/`.
+The catalog artifacts are POM-only: `radixor-models-standard` carries runtime
+dependencies on the original 20 standard models, while `radixor-models-bom`
+carries dependency-management constraints for all 144 user-facing individual
+models. Neither publishes an empty binary, sources, or Javadoc JAR. This Maven
+BOM is distinct from the root CycloneDX SBOM report under
+`build/reports/sbom/`.
 
 ```bash
 ./tools/parse-model-release-tag.sh "model/pl-pl-polimorf@<model-version>" .
diff --git a/docs/model-selection-and-loading.md b/docs/model-selection-and-loading.md
index c11b8dc..9fc61f4 100644
--- a/docs/model-selection-and-loading.md
+++ b/docs/model-selection-and-loading.md
@@ -27,8 +27,8 @@ These identifiers answer different questions:
 One language can have several models. `Language.PL_PL` is neither UniMorph nor PoliMorf. `loadCompiled(Language.PL_PL, ...)` resolves the stable default ID declared by `Language.defaultModelId()`. An explicit lookup requests exactly one ID. Registry ordering never changes either decision.
 
 Licensing follows the selected artifact. Radixor Java software is BSD-3-Clause; UniMorph-derived
-model data carries a model-specific CC BY-SA 3.0 notice, while PoliMorf carries its separate
-BSD-2-Clause license. The UniMorph notice preserves upstream attribution and identifies the
+model data carries its model-specific audited license and notice, while PoliMorf carries its
+separate BSD-2-Clause license. The UniMorph notice preserves upstream attribution and identifies the
 Radixor transformations and limited protectable contributions without claiming the underlying data.
 
 ## Choose runtime dependencies
@@ -78,7 +78,10 @@ dependencies {
 }
 ```
 
-The standard aggregate is POM-only. Its POM supplies exactly one default model per supported language as transitive runtime dependencies and excludes optional PoliMorf. It publishes no empty binary JAR.
+The standard aggregate is POM-only. Its POM supplies exactly the original 20
+standard models as transitive runtime dependencies. It excludes the 123
+standalone language models and optional PoliMorf; all remain available through
+individual artifacts. It publishes no empty binary JAR.
 
 ### BOM-managed versions
 
diff --git a/docs/overrides/landing.html b/docs/overrides/landing.html
index 9a987f7..65c5f3b 100644
--- a/docs/overrides/landing.html
+++ b/docs/overrides/landing.html
@@ -41,7 +41,7 @@
       <div class="rx2-hero-copy">
         <div class="rx2-kicker"><span></span>Dictionary-trained. Not dictionary-bound.</div>
         <h1>High-quality multilingual stemming.<br><span>Built for <b>Java</b> and <em>Python</em>.</span></h1>
-        <p>Radixor learns word-to-stem transformations from lexical evidence, compiles them into compact tries of patch commands, and applies those transformations algorithmically at runtime. It combines deterministic results, small deployable models, and high throughput across 20 languages. The training dictionary supplies evidence&mdash;it does not define a closed runtime vocabulary.</p>
+        <p>Radixor learns word-to-stem transformations from lexical evidence, compiles them into compact tries of patch commands, and applies those transformations algorithmically at runtime. It combines deterministic results, small deployable models, and high throughput across 166 language defaults. The training dictionary supplies evidence&mdash;it does not define a closed runtime vocabulary.</p>
       </div>
 
       <div class="rx2-hero-actions" aria-label="Get Radixor and project assurances">
@@ -69,13 +69,53 @@
       <aside class="rx2-highlight-card">
         <h2>Measured quality and runtime leadership</h2>
         <div class="rx2-highlight-stats">
-          <div class="rx2-highlight-stat teal"><i><svg viewBox="0 0 32 32" aria-hidden="true"><path d="M10 4h12v6c0 7-3 10-6 10s-6-3-6-10V4Z" fill="none" stroke="currentColor" stroke-width="2.2"/><path d="M10 7H5c0 6 2 9 7 9M22 7h5c0 6-2 9-7 9M16 20v5M11 27h10" fill="none" stroke="currentColor" stroke-width="2.2" stroke-linecap="round"/></svg></i><strong>40 / 40</strong><span>First-ranked balanced<br>accuracy results</span></div>
-          <div class="rx2-highlight-stat purple"><i><svg viewBox="0 0 32 32" aria-hidden="true"><path d="M10 4h12v6c0 7-3 10-6 10s-6-3-6-10V4Z" fill="none" stroke="currentColor" stroke-width="2.2"/><path d="M10 7H5c0 6 2 9 7 9M22 7h5c0 6-2 9-7 9M16 20v5M11 27h10" fill="none" stroke="currentColor" stroke-width="2.2" stroke-linecap="round"/></svg></i><strong>1,000</strong><span>Java training-coverage<br>scenarios</span></div>
+          <div class="rx2-highlight-stat teal"><i><svg viewBox="0 0 32 32" aria-hidden="true"><path d="M10 4h12v6c0 7-3 10-6 10s-6-3-6-10V4Z" fill="none" stroke="currentColor" stroke-width="2.2"/><path d="M10 7H5c0 6 2 9 7 9M22 7h5c0 6-2 9-7 9M16 20v5M11 27h10" fill="none" stroke="currentColor" stroke-width="2.2" stroke-linecap="round"/></svg></i><strong>326 / 326</strong><span>First-ranked balanced<br>accuracy results</span></div>
+          <div class="rx2-highlight-stat purple"><i><svg viewBox="0 0 32 32" aria-hidden="true"><path d="M10 4h12v6c0 7-3 10-6 10s-6-3-6-10V4Z" fill="none" stroke="currentColor" stroke-width="2.2"/><path d="M10 7H5c0 6 2 9 7 9M22 7h5c0 6-2 9-7 9M16 20v5M11 27h10" fill="none" stroke="currentColor" stroke-width="2.2" stroke-linecap="round"/></svg></i><strong>8,300</strong><span>Java training-coverage<br>scenarios</span></div>
           <div class="rx2-highlight-stat blue"><i><svg viewBox="0 0 32 32" aria-hidden="true"><path d="M5 23a12 12 0 1 1 22 0" fill="none" stroke="currentColor" stroke-width="2.2" stroke-linecap="round"/><path d="m16 20 6-8" stroke="currentColor" stroke-width="2.2" stroke-linecap="round"/><circle cx="16" cy="20" r="2" fill="currentColor"/></svg></i><strong>2.19×</strong><span>PyO3 geometric-<br>mean speedup</span></div>
           <div class="rx2-highlight-stat blue"><i><svg viewBox="0 0 32 32" aria-hidden="true"><path d="M5 23a12 12 0 1 1 22 0" fill="none" stroke="currentColor" stroke-width="2.2" stroke-linecap="round"/><path d="m16 20 6-8" stroke="currentColor" stroke-width="2.2" stroke-linecap="round"/><circle cx="16" cy="20" r="2" fill="currentColor"/></svg></i><strong>2.28×</strong><span>Python-C geometric-<br>mean speedup</span></div>
         </div>
-        <p class="rx2-evidence-line"><strong>What quality means here:</strong> the principal benchmark asks whether forms in the same annotated dictionary group&mdash;a morphological family, not a semantic or synonym set&mdash;receive the same stem, and whether forms with no shared group remain separated. Exact agreement with a prescribed dictionary root is reported separately. On this pairwise measure, Radixor ranks first in all 40 evaluated deterministic language-mode matrices, with no exact first-place ties; 38 include direct comparators and the two Hebrew modes report Radixor independently. The claim is limited to the published implementations, versions, adapters, dictionaries and metric. A <a href="{{ base_url }}/benchmarks/generalization/">20-language, five-split experiment</a> measures transfer to dictionary families withheld from Java model training. A separate <a href="{{ base_url }}/benchmarks/edit-cost-sensitivity/">234,000-observation edit-cost analysis</a> publishes dictionary-specific cost sensitivity, factor associations, and evidence-backed conclusions on every language page.</p>
-        <div class="rx2-chart-title"><strong>Python (PyO3) performance</strong> <span>(batch size N=100)</span></div>
+        <p class="rx2-evidence-line"><strong>What quality means here:</strong> the principal benchmark asks whether forms in the same annotated dictionary group&mdash;a morphological family, not a semantic or synonym set&mdash;receive the same stem, and whether forms with no shared group remain separated. Exact agreement with a prescribed dictionary root is reported separately. On this pairwise measure, Radixor is the sole leader in 326 of 326 mathematically defined deterministic language-mode matrices; 62 include direct comparators. 6 matrices with an absent pair class publish OI/UI and raw counts but no balanced-accuracy rank. The claim is limited to the published implementations, versions, adapters, dictionaries and metric. A <a href="{{ base_url }}/benchmarks/generalization/">166-language, five-split experiment</a> measures transfer to dictionary families withheld from Java model training. The historical 20-language edit-cost analysis remains separately published and is not extrapolated to the new dictionaries.</p>
+        <!-- JAVA-SNOWBALL-COMPARISON:START -->
+<div class="rx2-chart-title"><strong>Java Radixor and official direct Snowball</strong> <span>(exact-language cases)</span></div>
+<div class="rx2-throughput"><b>2.223×</b> <span>Snowball / Radixor geometric-mean runtime ratio</span></div>
+<div class="rx2-ratio-chart" aria-label="Log-scaled Snowball divided by Radixor nanoseconds per token for 29 exact-language cases">
+<div class="rx2-ratio-axis"><span>Radixor slower</span><b>1×</b><span>Snowball slower</span></div>
+<div class="rx2-ratio-row"><span>Arabic</span><span class="rx2-ratio-track" role="img" aria-label="Arabic: Snowball to Radixor runtime ratio 2.151" title="Arabic: 2.151×"><i style="--left:50.00%;--width:20.42%"></i></span><b>2.151×</b></div>
+<div class="rx2-ratio-row"><span>Armenian</span><span class="rx2-ratio-track" role="img" aria-label="Armenian: Snowball to Radixor runtime ratio 1.791" title="Armenian: 1.791×"><i style="--left:50.00%;--width:15.54%"></i></span><b>1.791×</b></div>
+<div class="rx2-ratio-row"><span>Basque</span><span class="rx2-ratio-track" role="img" aria-label="Basque: Snowball to Radixor runtime ratio 1.627" title="Basque: 1.627×"><i style="--left:50.00%;--width:12.98%"></i></span><b>1.627×</b></div>
+<div class="rx2-ratio-row"><span>Catalan</span><span class="rx2-ratio-track" role="img" aria-label="Catalan: Snowball to Radixor runtime ratio 4.339" title="Catalan: 4.339×"><i style="--left:50.00%;--width:39.12%"></i></span><b>4.339×</b></div>
+<div class="rx2-ratio-row"><span>Czech</span><span class="rx2-ratio-track" role="img" aria-label="Czech: Snowball to Radixor runtime ratio 1.108" title="Czech: 1.108×"><i style="--left:50.00%;--width:2.74%"></i></span><b>1.108×</b></div>
+<div class="rx2-ratio-row"><span>Danish</span><span class="rx2-ratio-track" role="img" aria-label="Danish: Snowball to Radixor runtime ratio 2.092" title="Danish: 2.092×"><i style="--left:50.00%;--width:19.68%"></i></span><b>2.092×</b></div>
+<div class="rx2-ratio-row"><span>Dutch</span><span class="rx2-ratio-track" role="img" aria-label="Dutch: Snowball to Radixor runtime ratio 2.678" title="Dutch: 2.678×"><i style="--left:50.00%;--width:26.26%"></i></span><b>2.678×</b></div>
+<div class="rx2-ratio-row"><span>Estonian</span><span class="rx2-ratio-track" role="img" aria-label="Estonian: Snowball to Radixor runtime ratio 3.233" title="Estonian: 3.233×"><i style="--left:50.00%;--width:31.28%"></i></span><b>3.233×</b></div>
+<div class="rx2-ratio-row"><span>Finnish</span><span class="rx2-ratio-track" role="img" aria-label="Finnish: Snowball to Radixor runtime ratio 1.078" title="Finnish: 1.078×"><i style="--left:50.00%;--width:1.99%"></i></span><b>1.078×</b></div>
+<div class="rx2-ratio-row"><span>French</span><span class="rx2-ratio-track" role="img" aria-label="French: Snowball to Radixor runtime ratio 1.982" title="French: 1.982×"><i style="--left:50.00%;--width:18.23%"></i></span><b>1.982×</b></div>
+<div class="rx2-ratio-row"><span>German</span><span class="rx2-ratio-track" role="img" aria-label="German: Snowball to Radixor runtime ratio 2.420" title="German: 2.420×"><i style="--left:50.00%;--width:23.56%"></i></span><b>2.420×</b></div>
+<div class="rx2-ratio-row"><span>Greek</span><span class="rx2-ratio-track" role="img" aria-label="Greek: Snowball to Radixor runtime ratio 6.524" title="Greek: 6.524×"><i style="--left:50.00%;--width:50.00%"></i></span><b>6.524×</b></div>
+<div class="rx2-ratio-row"><span>Hungarian</span><span class="rx2-ratio-track" role="img" aria-label="Hungarian: Snowball to Radixor runtime ratio 2.968" title="Hungarian: 2.968×"><i style="--left:50.00%;--width:29.00%"></i></span><b>2.968×</b></div>
+<div class="rx2-ratio-row"><span>Indonesian</span><span class="rx2-ratio-track" role="img" aria-label="Indonesian: Snowball to Radixor runtime ratio 1.200" title="Indonesian: 1.200×"><i style="--left:50.00%;--width:4.85%"></i></span><b>1.200×</b></div>
+<div class="rx2-ratio-row"><span>Irish</span><span class="rx2-ratio-track" role="img" aria-label="Irish: Snowball to Radixor runtime ratio 1.376" title="Irish: 1.376×"><i style="--left:50.00%;--width:8.50%"></i></span><b>1.376×</b></div>
+<div class="rx2-ratio-row"><span>Italian</span><span class="rx2-ratio-track" role="img" aria-label="Italian: Snowball to Radixor runtime ratio 4.132" title="Italian: 4.132×"><i style="--left:50.00%;--width:37.82%"></i></span><b>4.132×</b></div>
+<div class="rx2-ratio-row"><span>Lithuanian</span><span class="rx2-ratio-track" role="img" aria-label="Lithuanian: Snowball to Radixor runtime ratio 1.329" title="Lithuanian: 1.329×"><i style="--left:50.00%;--width:7.57%"></i></span><b>1.329×</b></div>
+<div class="rx2-ratio-row"><span>Norwegian Bokmal</span><span class="rx2-ratio-track" role="img" aria-label="Norwegian Bokmal: Snowball to Radixor runtime ratio 1.400" title="Norwegian Bokmal: 1.400×"><i style="--left:50.00%;--width:8.96%"></i></span><b>1.400×</b></div>
+<div class="rx2-ratio-row"><span>Norwegian Nynorsk</span><span class="rx2-ratio-track" role="img" aria-label="Norwegian Nynorsk: Snowball to Radixor runtime ratio 1.753" title="Norwegian Nynorsk: 1.753×"><i style="--left:50.00%;--width:14.96%"></i></span><b>1.753×</b></div>
+<div class="rx2-ratio-row"><span>Persian</span><span class="rx2-ratio-track" role="img" aria-label="Persian: Snowball to Radixor runtime ratio 4.634" title="Persian: 4.634×"><i style="--left:50.00%;--width:40.88%"></i></span><b>4.634×</b></div>
+<div class="rx2-ratio-row"><span>Polish</span><span class="rx2-ratio-track" role="img" aria-label="Polish: Snowball to Radixor runtime ratio 1.084" title="Polish: 1.084×"><i style="--left:50.00%;--width:2.16%"></i></span><b>1.084×</b></div>
+<div class="rx2-ratio-row"><span>Portuguese</span><span class="rx2-ratio-track" role="img" aria-label="Portuguese: Snowball to Radixor runtime ratio 4.208" title="Portuguese: 4.208×"><i style="--left:50.00%;--width:38.31%"></i></span><b>4.208×</b></div>
+<div class="rx2-ratio-row"><span>Russian</span><span class="rx2-ratio-track" role="img" aria-label="Russian: Snowball to Radixor runtime ratio 1.388" title="Russian: 1.388×"><i style="--left:50.00%;--width:8.73%"></i></span><b>1.388×</b></div>
+<div class="rx2-ratio-row"><span>Romanian</span><span class="rx2-ratio-track" role="img" aria-label="Romanian: Snowball to Radixor runtime ratio 3.604" title="Romanian: 3.604×"><i style="--left:50.00%;--width:34.18%"></i></span><b>3.604×</b></div>
+<div class="rx2-ratio-row"><span>Southern Sotho</span><span class="rx2-ratio-track" role="img" aria-label="Southern Sotho: Snowball to Radixor runtime ratio 1.014" title="Southern Sotho: 1.014×"><i style="--left:50.00%;--width:0.36%"></i></span><b>1.014×</b></div>
+<div class="rx2-ratio-row"><span>Spanish</span><span class="rx2-ratio-track" role="img" aria-label="Spanish: Snowball to Radixor runtime ratio 2.553" title="Spanish: 2.553×"><i style="--left:50.00%;--width:24.99%"></i></span><b>2.553×</b></div>
+<div class="rx2-ratio-row"><span>Swedish</span><span class="rx2-ratio-track" role="img" aria-label="Swedish: Snowball to Radixor runtime ratio 1.406" title="Swedish: 1.406×"><i style="--left:50.00%;--width:9.08%"></i></span><b>1.406×</b></div>
+<div class="rx2-ratio-row"><span>Turkish</span><span class="rx2-ratio-track" role="img" aria-label="Turkish: Snowball to Radixor runtime ratio 4.134" title="Turkish: 4.134×"><i style="--left:50.00%;--width:37.84%"></i></span><b>4.134×</b></div>
+<div class="rx2-ratio-row"><span>Yiddish</span><span class="rx2-ratio-track" role="img" aria-label="Yiddish: Snowball to Radixor runtime ratio 5.958" title="Yiddish: 5.958×"><i style="--left:50.00%;--width:47.58%"></i></span><b>5.958×</b></div>
+</div>
+<p class="rx2-note">The 2026-09-11 Java JMH snapshot charts all <b>29</b> official direct Snowball exact-language cases against the exact generic Radixor model row. The geometric mean is <b>2.223× over 29 comparable changed-token cases</b>. A ratio above 1 means Snowball used more time per token. All rows use 3 warmup iterations, 5 measurement iterations, 3 forks, one thread, and the same per-language timing population.</p>
+<details class="quality-details"><summary>Accessible raw Java comparison table</summary>
+<table><thead><tr><th>Language</th><th>Model</th><th>Timing corpus</th><th>Radixor ns/token</th><th>Snowball ns/token</th><th>Snowball JMH error ns/token</th><th>Snowball / Radixor</th><th>Aggregate</th></tr></thead><tbody><tr><td>Arabic</td><td><code>ar-default</code></td><td>changed tokens</td><td>209.3</td><td>450.2</td><td>±18.5</td><td>2.151×</td><td>yes</td></tr><tr><td>Armenian</td><td><code>hy-am-default</code></td><td>changed tokens</td><td>72.8</td><td>130.4</td><td>±20.6</td><td>1.791×</td><td>yes</td></tr><tr><td>Basque</td><td><code>eus-default</code></td><td>changed tokens</td><td>98.3</td><td>160.0</td><td>±15.9</td><td>1.627×</td><td>yes</td></tr><tr><td>Catalan</td><td><code>ca-es-default</code></td><td>changed tokens</td><td>54.9</td><td>238.3</td><td>±13.9</td><td>4.339×</td><td>yes</td></tr><tr><td>Czech</td><td><code>cs-cz-default</code></td><td>changed tokens</td><td>83.5</td><td>92.5</td><td>±14.0</td><td>1.108×</td><td>yes</td></tr><tr><td>Danish</td><td><code>da-dk-default</code></td><td>changed tokens</td><td>54.2</td><td>113.5</td><td>±11.6</td><td>2.092×</td><td>yes</td></tr><tr><td>Dutch</td><td><code>nl-nl-default</code></td><td>changed tokens</td><td>71.3</td><td>190.9</td><td>±13.6</td><td>2.678×</td><td>yes</td></tr><tr><td>Estonian</td><td><code>et-ee-default</code></td><td>changed tokens</td><td>56.7</td><td>183.4</td><td>±12.8</td><td>3.233×</td><td>yes</td></tr><tr><td>Finnish</td><td><code>fi-fi-default</code></td><td>changed tokens</td><td>151.4</td><td>163.1</td><td>±8.7</td><td>1.078×</td><td>yes</td></tr><tr><td>French</td><td><code>fr-fr-default</code></td><td>changed tokens</td><td>148.8</td><td>294.9</td><td>±15.9</td><td>1.982×</td><td>yes</td></tr><tr><td>German</td><td><code>de-de-default</code></td><td>changed tokens</td><td>160.0</td><td>387.2</td><td>±17.5</td><td>2.420×</td><td>yes</td></tr><tr><td>Greek</td><td><code>el-gr-default</code></td><td>changed tokens</td><td>109.4</td><td>713.5</td><td>±31.9</td><td>6.524×</td><td>yes</td></tr><tr><td>Hungarian</td><td><code>hu-hu-default</code></td><td>changed tokens</td><td>60.4</td><td>179.2</td><td>±15.7</td><td>2.968×</td><td>yes</td></tr><tr><td>Indonesian</td><td><code>id-id-default</code></td><td>changed tokens</td><td>84.2</td><td>101.0</td><td>±11.5</td><td>1.200×</td><td>yes</td></tr><tr><td>Irish</td><td><code>ga-ie-default</code></td><td>changed tokens</td><td>94.1</td><td>129.5</td><td>±14.5</td><td>1.376×</td><td>yes</td></tr><tr><td>Italian</td><td><code>it-it-default</code></td><td>changed tokens</td><td>84.4</td><td>348.8</td><td>±16.7</td><td>4.132×</td><td>yes</td></tr><tr><td>Lithuanian</td><td><code>lt-lt-default</code></td><td>changed tokens</td><td>109.6</td><td>145.6</td><td>±12.6</td><td>1.329×</td><td>yes</td></tr><tr><td>Norwegian Bokmal</td><td><code>nb-no-default</code></td><td>changed tokens</td><td>65.2</td><td>91.2</td><td>±13.2</td><td>1.400×</td><td>yes</td></tr><tr><td>Norwegian Nynorsk</td><td><code>nn-no-default</code></td><td>changed tokens</td><td>50.3</td><td>88.2</td><td>±12.6</td><td>1.753×</td><td>yes</td></tr><tr><td>Persian</td><td><code>fa-ir-default</code></td><td>changed tokens</td><td>65.2</td><td>302.3</td><td>±17.7</td><td>4.634×</td><td>yes</td></tr><tr><td>Polish</td><td><code>pl-pl-unimorph</code></td><td>changed tokens</td><td>89.6</td><td>97.2</td><td>±15.9</td><td>1.084×</td><td>yes</td></tr><tr><td>Portuguese</td><td><code>pt-pt-default</code></td><td>changed tokens</td><td>59.0</td><td>248.4</td><td>±11.8</td><td>4.208×</td><td>yes</td></tr><tr><td>Russian</td><td><code>ru-ru-default</code></td><td>changed tokens</td><td>117.6</td><td>163.2</td><td>±17.2</td><td>1.388×</td><td>yes</td></tr><tr><td>Romanian</td><td><code>ro-ro-default</code></td><td>changed tokens</td><td>85.2</td><td>307.1</td><td>±16.1</td><td>3.604×</td><td>yes</td></tr><tr><td>Southern Sotho</td><td><code>st-za-default</code></td><td>changed tokens</td><td>44.7</td><td>45.3</td><td>±10.9</td><td>1.014×</td><td>yes</td></tr><tr><td>Spanish</td><td><code>es-es-default</code></td><td>changed tokens</td><td>90.4</td><td>230.8</td><td>±12.7</td><td>2.553×</td><td>yes</td></tr><tr><td>Swedish</td><td><code>sv-se-default</code></td><td>changed tokens</td><td>68.5</td><td>96.3</td><td>±13.2</td><td>1.406×</td><td>yes</td></tr><tr><td>Turkish</td><td><code>tr-tr-default</code></td><td>changed tokens</td><td>100.1</td><td>413.9</td><td>±20.2</td><td>4.134×</td><td>yes</td></tr><tr><td>Yiddish</td><td><code>yi-default</code></td><td>changed tokens</td><td>51.1</td><td>304.6</td><td>±19.2</td><td>5.958×</td><td>yes</td></tr></tbody></table>
+</details>
+<!-- JAVA-SNOWBALL-COMPARISON:END -->
+<div class="rx2-chart-title"><strong>Python (PyO3) performance</strong> <span>(batch size N=100)</span></div>
         <div class="rx2-throughput"><b>6.85M – 13.87M</b> <span>words per second</span></div>
         <div class="rx2-chart-wrap">
           <div class="rx2-ylabels"><span>15M</span><span>11.25M</span><span>7.5M</span><span>3.75M</span><span>0</span></div>
@@ -109,8 +149,8 @@
   <div class="rx2-shell">
     <section class="rx2-java-banner">
       <img src="{{ base_url }}/assets/images/java-award.svg" alt="Java benchmark award">
-      <div class="rx2-java-copy"><h2>Java: quality leadership<br>at production-grade speed</h2><p>Radixor ranks first in all 40 evaluated balanced-accuracy matrices; 38 include direct comparators and the two Hebrew modes report Radixor independently. It also records the lowest runtime point estimate in 9 of 19 language pages with a direct Java comparator; where narrower light or minimal filters are faster, the adjacent quality table shows the linguistic trade-off.</p></div>
-      <div class="rx2-win-box"><strong>40 / 40</strong><span>Quality first places</span><small>Plus 9 / 19 runtime minima</small></div>
+      <div class="rx2-java-copy"><h2>Java: quality leadership<br>at production-grade speed</h2><p>Radixor is the sole balanced-accuracy leader in 326 of 326 mathematically defined language-mode matrices; 62 include a direct comparator. 6 matrices with an absent pair class publish OI/UI and raw counts but have no defined balanced accuracy. It also records the lowest runtime point estimate in 9 of 19 language pages with a direct Java comparator; where narrower light or minimal filters are faster, the adjacent quality table shows the linguistic trade-off.</p></div>
+      <div class="rx2-win-box"><strong>326 / 326</strong><span>Quality first places</span><small>Plus 9 / 19 runtime minima</small></div>
       <div class="rx2-verify"><b>Reproducible · Transparent · Verifiable</b><p>Every result is public, verifiable, and reproducible.</p><a href="{{ base_url }}/benchmarks/">View all benchmarks →</a></div>
     </section>
 
@@ -123,12 +163,14 @@
     </section>
     <p class="rx2-case-note">Quality comparator: published Snowball Finnish Lucene/Snowball results. The displayed speed comparison is Python (PyO3) versus PyStemmer 3.1.0 at batch size N=100; Python-C is 1.51× faster than the same baseline. The point is the combined quality/performance envelope, not wrapper identity.</p>
 
-    <section class="rx2-languages-strip">
-      <div class="rx2-language-label"><svg viewBox="0 0 32 32" aria-hidden="true"><circle cx="16" cy="16" r="12" fill="none" stroke="currentColor" stroke-width="2"/><path d="M4 16h24M16 4c4 4 5 8 5 12s-1 8-5 12c-4-4-5-8-5-12s1-8 5-12Z" fill="none" stroke="currentColor" stroke-width="2"/></svg><div><strong>20 languages</strong><span>Shared quality · Java runtime</span></div></div>
-      <div class="rx2-language-list"><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/czech/" aria-label="Open Czech benchmark"><img src="{{ base_url }}/assets/images/flags/cs.svg" alt="Czech flag"><span>Czech</span></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/danish/" aria-label="Open Danish benchmark"><img src="{{ base_url }}/assets/images/flags/da.svg" alt="Danish flag"><span>Danish</span></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/dutch/" aria-label="Open Dutch benchmark"><img src="{{ base_url }}/assets/images/flags/nl.svg" alt="Dutch flag"><span>Dutch</span></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/english/" aria-label="Open English benchmark"><img src="{{ base_url }}/assets/images/flags/en.svg" alt="English flag"><span>English</span></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/finnish/" aria-label="Open Finnish benchmark"><img src="{{ base_url }}/assets/images/flags/fi.svg" alt="Finnish flag"><span>Finnish</span></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/french/" aria-label="Open French benchmark"><img src="{{ base_url }}/assets/images/flags/fr.svg" alt="French flag"><span>French</span></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/german/" aria-label="Open German benchmark"><img src="{{ base_url }}/assets/images/flags/de.svg" alt="German flag"><span>German</span></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/hebrew/" aria-label="Open Hebrew benchmark"><img src="{{ base_url }}/assets/images/flags/he.svg" alt="Hebrew flag"><span>Hebrew</span></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/hungarian/" aria-label="Open Hungarian benchmark"><img src="{{ base_url }}/assets/images/flags/hu.svg" alt="Hungarian flag"><span>Hungarian</span></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/italian/" aria-label="Open Italian benchmark"><img src="{{ base_url }}/assets/images/flags/it.svg" alt="Italian flag"><span>Italian</span></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/norwegian-bokmal/" aria-label="Open Norwegian Bokmål benchmark"><img src="{{ base_url }}/assets/images/flags/nb.svg" alt="Norwegian Bokmål flag"><span>Norwegian Bokmål</span></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/norwegian-nynorsk/" aria-label="Open Norwegian Nynorsk benchmark"><img src="{{ base_url }}/assets/images/flags/nn.svg" alt="Norwegian Nynorsk flag"><span>Norwegian Nynorsk</span></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/persian/" aria-label="Open Persian benchmark"><img src="{{ base_url }}/assets/images/flags/fa.svg" alt="Persian flag"><span>Persian</span></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/polish/" aria-label="Open Polish benchmark"><img src="{{ base_url }}/assets/images/flags/pl.svg" alt="Polish flag"><span>Polish</span></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/portuguese/" aria-label="Open Portuguese benchmark"><img src="{{ base_url }}/assets/images/flags/pt.svg" alt="Portuguese flag"><span>Portuguese</span></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/russian/" aria-label="Open Russian benchmark"><img src="{{ base_url }}/assets/images/flags/ru.svg" alt="Russian flag"><span>Russian</span></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/spanish/" aria-label="Open Spanish benchmark"><img src="{{ base_url }}/assets/images/flags/es.svg" alt="Spanish flag"><span>Spanish</span></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/swedish/" aria-label="Open Swedish benchmark"><img src="{{ base_url }}/assets/images/flags/sv.svg" alt="Swedish flag"><span>Swedish</span></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/ukrainian/" aria-label="Open Ukrainian benchmark"><img src="{{ base_url }}/assets/images/flags/uk.svg" alt="Ukrainian flag"><span>Ukrainian</span></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/yiddish/" aria-label="Open Yiddish benchmark"><img src="{{ base_url }}/assets/images/flags/yi.svg" alt="Yiddish flag"><span>Yiddish</span></a></div>
-    </section>
+    <!-- JAVA-LANGUAGE-PAGES:START -->
+<section class="rx2-languages-strip">
+<div class="rx2-language-label"><div><strong>167 models · 166 language pages</strong><span>Complete Java quality evidence; Polish documents both models</span></div></div>
+<div class="rx2-language-list"><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/ady/" aria-label="Open Adyghe benchmark; Russia"><i class="rx2-language-icon" aria-hidden="true" title="Russia">🇷🇺</i><span>Adyghe</span><small>ady-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/af-za/" aria-label="Open Afrikaans benchmark; South Africa"><i class="rx2-language-icon" aria-hidden="true" title="South Africa">🇿🇦</i><span>Afrikaans</span><small>af-za-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/ail/" aria-label="Open Aimele benchmark; Papua New Guinea"><i class="rx2-language-icon" aria-hidden="true" title="Papua New Guinea">🇵🇬</i><span>Aimele</span><small>ail-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/aka/" aria-label="Open Akan benchmark; Ghana"><i class="rx2-language-icon" aria-hidden="true" title="Ghana">🇬🇭</i><span>Akan</span><small>aka-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/sq-al/" aria-label="Open Albanian benchmark; Albania"><i class="rx2-language-icon" aria-hidden="true" title="Albania">🇦🇱</i><span>Albanian</span><small>sq-al-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/gsw/" aria-label="Open Alsatian benchmark; Transnational language"><i class="rx2-language-icon" aria-hidden="true" title="Transnational language">🌐</i><span>Alsatian</span><small>gsw-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/am-et/" aria-label="Open Amharic benchmark; Ethiopia"><i class="rx2-language-icon" aria-hidden="true" title="Ethiopia">🇪🇹</i><span>Amharic</span><small>am-et-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/azg/" aria-label="Open Amol Azerbaijani benchmark; Iran"><i class="rx2-language-icon" aria-hidden="true" title="Iran">🇮🇷</i><span>Amol Azerbaijani</span><small>azg-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/grc/" aria-label="Open Ancient Greek benchmark; Historical language"><i class="rx2-language-icon" aria-hidden="true" title="Historical language">🌐</i><span>Ancient Greek</span><small>grc-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/xno/" aria-label="Open Anglo-Norman benchmark; Historical language"><i class="rx2-language-icon" aria-hidden="true" title="Historical language">🌐</i><span>Anglo-Norman</span><small>xno-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/ar/" aria-label="Open Arabic benchmark; Transnational language"><i class="rx2-language-icon" aria-hidden="true" title="Transnational language">🌐</i><span>Arabic</span><small>ar-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/hy-am/" aria-label="Open Armenian benchmark; Armenia"><i class="rx2-language-icon" aria-hidden="true" title="Armenia">🇦🇲</i><span>Armenian</span><small>hy-am-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/cni/" aria-label="Open Ashaninka benchmark; Peru"><i class="rx2-language-icon" aria-hidden="true" title="Peru">🇵🇪</i><span>Ashaninka</span><small>cni-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/as-in/" aria-label="Open Assamese benchmark; India"><i class="rx2-language-icon" aria-hidden="true" title="India">🇮🇳</i><span>Assamese</span><small>as-in-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/ast/" aria-label="Open Asturian benchmark; Spain"><i class="rx2-language-icon" aria-hidden="true" title="Spain">🇪🇸</i><span>Asturian</span><small>ast-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/aym/" aria-label="Open Aymara benchmark; Transnational language"><i class="rx2-language-icon" aria-hidden="true" title="Transnational language">🌐</i><span>Aymara</span><small>aym-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/az-az/" aria-label="Open Azerbaijani benchmark; Azerbaijan"><i class="rx2-language-icon" aria-hidden="true" title="Azerbaijan">🇦🇿</i><span>Azerbaijani</span><small>az-az-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/bak/" aria-label="Open Bashkir benchmark; Russia"><i class="rx2-language-icon" aria-hidden="true" title="Russia">🇷🇺</i><span>Bashkir</span><small>bak-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/eus/" aria-label="Open Basque benchmark; Transnational language"><i class="rx2-language-icon" aria-hidden="true" title="Transnational language">🌐</i><span>Basque</span><small>eus-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/be-by/" aria-label="Open Belarusian benchmark; Belarus"><i class="rx2-language-icon" aria-hidden="true" title="Belarus">🇧🇾</i><span>Belarusian</span><small>be-by-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/bn-bd/" aria-label="Open Bengali benchmark; Bangladesh"><i class="rx2-language-icon" aria-hidden="true" title="Bangladesh">🇧🇩</i><span>Bengali</span><small>bn-bd-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/gup/" aria-label="Open Bininj Kun-wok benchmark; Australia"><i class="rx2-language-icon" aria-hidden="true" title="Australia">🇦🇺</i><span>Bininj Kun-wok</span><small>gup-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/bra/" aria-label="Open Braj benchmark; India"><i class="rx2-language-icon" aria-hidden="true" title="India">🇮🇳</i><span>Braj</span><small>bra-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/bre/" aria-label="Open Breton benchmark; France"><i class="rx2-language-icon" aria-hidden="true" title="France">🇫🇷</i><span>Breton</span><small>bre-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/bg-bg/" aria-label="Open Bulgarian benchmark; Bulgaria"><i class="rx2-language-icon" aria-hidden="true" title="Bulgaria">🇧🇬</i><span>Bulgarian</span><small>bg-bg-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/ca-es/" aria-label="Open Catalan benchmark; Spain"><i class="rx2-language-icon" aria-hidden="true" title="Spain">🇪🇸</i><span>Catalan</span><small>ca-es-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/ceb/" aria-label="Open Cebuano benchmark; Philippines"><i class="rx2-language-icon" aria-hidden="true" title="Philippines">🇵🇭</i><span>Cebuano</span><small>ceb-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/ny-mw/" aria-label="Open Chichewa benchmark; Malawi"><i class="rx2-language-icon" aria-hidden="true" title="Malawi">🇲🇼</i><span>Chichewa</span><small>ny-mw-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/ckt/" aria-label="Open Chukchi benchmark; Russia"><i class="rx2-language-icon" aria-hidden="true" title="Russia">🇷🇺</i><span>Chukchi</span><small>ckt-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/chu/" aria-label="Open Church Slavonic benchmark; Historical language"><i class="rx2-language-icon" aria-hidden="true" title="Historical language">🌐</i><span>Church Slavonic</span><small>chu-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/xcl/" aria-label="Open Classical Armenian benchmark; Historical language"><i class="rx2-language-icon" aria-hidden="true" title="Historical language">🌐</i><span>Classical Armenian</span><small>xcl-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/syc/" aria-label="Open Classical Syriac benchmark; Historical language"><i class="rx2-language-icon" aria-hidden="true" title="Historical language">🌐</i><span>Classical Syriac</span><small>syc-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/swc/" aria-label="Open Congo Swahili benchmark; Democratic Republic of the Congo"><i class="rx2-language-icon" aria-hidden="true" title="Democratic Republic of the Congo">🇨🇩</i><span>Congo Swahili</span><small>swc-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/cpa/" aria-label="Open Copala Triqui benchmark; Mexico"><i class="rx2-language-icon" aria-hidden="true" title="Mexico">🇲🇽</i><span>Copala Triqui</span><small>cpa-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/cor/" aria-label="Open Cornish benchmark; United Kingdom"><i class="rx2-language-icon" aria-hidden="true" title="United Kingdom">🇬🇧</i><span>Cornish</span><small>cor-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/cre/" aria-label="Open Cree benchmark; Canada"><i class="rx2-language-icon" aria-hidden="true" title="Canada">🇨🇦</i><span>Cree</span><small>cre-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/crh/" aria-label="Open Crimean Tatar benchmark; Transnational or politically ambiguous language"><i class="rx2-language-icon" aria-hidden="true" title="Transnational or politically ambiguous language">🌐</i><span>Crimean Tatar</span><small>crh-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/czech/" aria-label="Open Czech benchmark; Czechia"><i class="rx2-language-icon" aria-hidden="true" title="Czechia">🇨🇿</i><span>Czech</span><small>cs-cz-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/dak/" aria-label="Open Dakota benchmark; United States"><i class="rx2-language-icon" aria-hidden="true" title="United States">🇺🇸</i><span>Dakota</span><small>dak-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/danish/" aria-label="Open Danish benchmark; Denmark"><i class="rx2-language-icon" aria-hidden="true" title="Denmark">🇩🇰</i><span>Danish</span><small>da-dk-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/dutch/" aria-label="Open Dutch benchmark; Netherlands"><i class="rx2-language-icon" aria-hidden="true" title="Netherlands">🇳🇱</i><span>Dutch</span><small>nl-nl-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/cly/" aria-label="Open Eastern Chatino benchmark; Mexico"><i class="rx2-language-icon" aria-hidden="true" title="Mexico">🇲🇽</i><span>Eastern Chatino</span><small>cly-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/arz/" aria-label="Open Egyptian Arabic benchmark; Egypt"><i class="rx2-language-icon" aria-hidden="true" title="Egypt">🇪🇬</i><span>Egyptian Arabic</span><small>arz-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/english/" aria-label="Open English benchmark; Transnational language"><i class="rx2-language-icon" aria-hidden="true" title="Transnational language">🌐</i><span>English</span><small>us-uk-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/et-ee/" aria-label="Open Estonian benchmark; Estonia"><i class="rx2-language-icon" aria-hidden="true" title="Estonia">🇪🇪</i><span>Estonian</span><small>et-ee-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/evn/" aria-label="Open Evenki benchmark; Transnational language"><i class="rx2-language-icon" aria-hidden="true" title="Transnational language">🌐</i><span>Evenki</span><small>evn-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/fo-fo/" aria-label="Open Faroese benchmark; Faroe Islands"><i class="rx2-language-icon" aria-hidden="true" title="Faroe Islands">🇫🇴</i><span>Faroese</span><small>fo-fo-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/finnish/" aria-label="Open Finnish benchmark; Finland"><i class="rx2-language-icon" aria-hidden="true" title="Finland">🇫🇮</i><span>Finnish</span><small>fi-fi-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/french/" aria-label="Open French benchmark; France"><i class="rx2-language-icon" aria-hidden="true" title="France">🇫🇷</i><span>French</span><small>fr-fr-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/fur/" aria-label="Open Friulian benchmark; Italy"><i class="rx2-language-icon" aria-hidden="true" title="Italy">🇮🇹</i><span>Friulian</span><small>fur-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/gaa/" aria-label="Open Ga benchmark; Ghana"><i class="rx2-language-icon" aria-hidden="true" title="Ghana">🇬🇭</i><span>Ga</span><small>gaa-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/gal/" aria-label="Open Galolen benchmark; Timor-Leste"><i class="rx2-language-icon" aria-hidden="true" title="Timor-Leste">🇹🇱</i><span>Galolen</span><small>gal-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/ka-ge/" aria-label="Open Georgian benchmark; Georgia"><i class="rx2-language-icon" aria-hidden="true" title="Georgia">🇬🇪</i><span>Georgian</span><small>ka-ge-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/german/" aria-label="Open German benchmark; Germany"><i class="rx2-language-icon" aria-hidden="true" title="Germany">🇩🇪</i><span>German</span><small>de-de-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/got/" aria-label="Open Gothic benchmark; Historical language"><i class="rx2-language-icon" aria-hidden="true" title="Historical language">🌐</i><span>Gothic</span><small>got-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/el-gr/" aria-label="Open Greek benchmark; Greece"><i class="rx2-language-icon" aria-hidden="true" title="Greece">🇬🇷</i><span>Greek</span><small>el-gr-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/afb/" aria-label="Open Gulf Arabic benchmark; Transnational language"><i class="rx2-language-icon" aria-hidden="true" title="Transnational language">🌐</i><span>Gulf Arabic</span><small>afb-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/hai/" aria-label="Open Haida benchmark; Transnational language"><i class="rx2-language-icon" aria-hidden="true" title="Transnational language">🌐</i><span>Haida</span><small>hai-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/hebrew/" aria-label="Open Hebrew benchmark; Israel"><i class="rx2-language-icon" aria-hidden="true" title="Israel">🇮🇱</i><span>Hebrew</span><small>he-il-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/hil/" aria-label="Open Hiligaynon benchmark; Philippines"><i class="rx2-language-icon" aria-hidden="true" title="Philippines">🇵🇭</i><span>Hiligaynon</span><small>hil-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/hsi/" aria-label="Open Hsilimo benchmark; China"><i class="rx2-language-icon" aria-hidden="true" title="China">🇨🇳</i><span>Hsilimo</span><small>hsi-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/hungarian/" aria-label="Open Hungarian benchmark; Hungary"><i class="rx2-language-icon" aria-hidden="true" title="Hungary">🇭🇺</i><span>Hungarian</span><small>hu-hu-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/is-is/" aria-label="Open Icelandic benchmark; Iceland"><i class="rx2-language-icon" aria-hidden="true" title="Iceland">🇮🇸</i><span>Icelandic</span><small>is-is-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/id-id/" aria-label="Open Indonesian benchmark; Indonesia"><i class="rx2-language-icon" aria-hidden="true" title="Indonesia">🇮🇩</i><span>Indonesian</span><small>id-id-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/izh/" aria-label="Open Ingrian benchmark; Russia"><i class="rx2-language-icon" aria-hidden="true" title="Russia">🇷🇺</i><span>Ingrian</span><small>izh-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/ga-ie/" aria-label="Open Irish benchmark; Ireland"><i class="rx2-language-icon" aria-hidden="true" title="Ireland">🇮🇪</i><span>Irish</span><small>ga-ie-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/italian/" aria-label="Open Italian benchmark; Italy"><i class="rx2-language-icon" aria-hidden="true" title="Italy">🇮🇹</i><span>Italian</span><small>it-it-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/itl/" aria-label="Open Itelmen benchmark; Russia"><i class="rx2-language-icon" aria-hidden="true" title="Russia">🇷🇺</i><span>Itelmen</span><small>itl-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/ja-jp/" aria-label="Open Japanese benchmark; Japan"><i class="rx2-language-icon" aria-hidden="true" title="Japan">🇯🇵</i><span>Japanese</span><small>ja-jp-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/kbd/" aria-label="Open Kabardian benchmark; Russia"><i class="rx2-language-icon" aria-hidden="true" title="Russia">🇷🇺</i><span>Kabardian</span><small>kbd-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/kl-gl/" aria-label="Open Kalaallisut benchmark; Greenland"><i class="rx2-language-icon" aria-hidden="true" title="Greenland">🇬🇱</i><span>Kalaallisut</span><small>kl-gl-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/kn-in/" aria-label="Open Kannada benchmark; India"><i class="rx2-language-icon" aria-hidden="true" title="India">🇮🇳</i><span>Kannada</span><small>kn-in-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/krl/" aria-label="Open Karelian benchmark; Transnational language"><i class="rx2-language-icon" aria-hidden="true" title="Transnational language">🌐</i><span>Karelian</span><small>krl-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/csb/" aria-label="Open Kashubian benchmark; Poland"><i class="rx2-language-icon" aria-hidden="true" title="Poland">🇵🇱</i><span>Kashubian</span><small>csb-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/kk-kz/" aria-label="Open Kazakh benchmark; Kazakhstan"><i class="rx2-language-icon" aria-hidden="true" title="Kazakhstan">🇰🇿</i><span>Kazakh</span><small>kk-kz-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/ket/" aria-label="Open Ket benchmark; Russia"><i class="rx2-language-icon" aria-hidden="true" title="Russia">🇷🇺</i><span>Ket</span><small>ket-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/kjh/" aria-label="Open Khakas benchmark; Russia"><i class="rx2-language-icon" aria-hidden="true" title="Russia">🇷🇺</i><span>Khakas</span><small>kjh-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/kca/" aria-label="Open Khanty benchmark; Russia"><i class="rx2-language-icon" aria-hidden="true" title="Russia">🇷🇺</i><span>Khanty</span><small>kca-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/klr/" aria-label="Open Khinalugh benchmark; Azerbaijan"><i class="rx2-language-icon" aria-hidden="true" title="Azerbaijan">🇦🇿</i><span>Khinalugh</span><small>klr-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/kod/" aria-label="Open Kodi benchmark; Indonesia"><i class="rx2-language-icon" aria-hidden="true" title="Indonesia">🇮🇩</i><span>Kodi</span><small>kod-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/kon/" aria-label="Open Kongo benchmark; Transnational language"><i class="rx2-language-icon" aria-hidden="true" title="Transnational language">🌐</i><span>Kongo</span><small>kon-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/ko-kr/" aria-label="Open Korean benchmark; South Korea"><i class="rx2-language-icon" aria-hidden="true" title="South Korea">🇰🇷</i><span>Korean</span><small>ko-kr-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/kmr/" aria-label="Open Kurmanji Kurdish benchmark; Transnational language"><i class="rx2-language-icon" aria-hidden="true" title="Transnational language">🌐</i><span>Kurmanji Kurdish</span><small>kmr-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/ky-kg/" aria-label="Open Kyrgyz benchmark; Kyrgyzstan"><i class="rx2-language-icon" aria-hidden="true" title="Kyrgyzstan">🇰🇬</i><span>Kyrgyz</span><small>ky-kg-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/slp/" aria-label="Open Lacandon benchmark; Mexico"><i class="rx2-language-icon" aria-hidden="true" title="Mexico">🇲🇽</i><span>Lacandon</span><small>slp-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/lld/" aria-label="Open Ladin benchmark; Italy"><i class="rx2-language-icon" aria-hidden="true" title="Italy">🇮🇹</i><span>Ladin</span><small>lld-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/la/" aria-label="Open Latin benchmark; Historical and transnational language"><i class="rx2-language-icon" aria-hidden="true" title="Historical and transnational language">🌐</i><span>Latin</span><small>la-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/lv-lv/" aria-label="Open Latvian benchmark; Latvia"><i class="rx2-language-icon" aria-hidden="true" title="Latvia">🇱🇻</i><span>Latvian</span><small>lv-lv-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/lin/" aria-label="Open Lingala benchmark; Transnational language"><i class="rx2-language-icon" aria-hidden="true" title="Transnational language">🌐</i><span>Lingala</span><small>lin-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/lt-lt/" aria-label="Open Lithuanian benchmark; Lithuania"><i class="rx2-language-icon" aria-hidden="true" title="Lithuania">🇱🇹</i><span>Lithuanian</span><small>lt-lt-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/liv/" aria-label="Open Livonian benchmark; Latvia"><i class="rx2-language-icon" aria-hidden="true" title="Latvia">🇱🇻</i><span>Livonian</span><small>liv-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/nds/" aria-label="Open Low German benchmark; Transnational language"><i class="rx2-language-icon" aria-hidden="true" title="Transnational language">🌐</i><span>Low German</span><small>nds-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/dsb/" aria-label="Open Lower Sorbian benchmark; Germany"><i class="rx2-language-icon" aria-hidden="true" title="Germany">🇩🇪</i><span>Lower Sorbian</span><small>dsb-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/lg-ug/" aria-label="Open Luganda benchmark; Uganda"><i class="rx2-language-icon" aria-hidden="true" title="Uganda">🇺🇬</i><span>Luganda</span><small>lg-ug-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/mk-mk/" aria-label="Open Macedonian benchmark; North Macedonia"><i class="rx2-language-icon" aria-hidden="true" title="North Macedonia">🇲🇰</i><span>Macedonian</span><small>mk-mk-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/mag/" aria-label="Open Magahi benchmark; India"><i class="rx2-language-icon" aria-hidden="true" title="India">🇮🇳</i><span>Magahi</span><small>mag-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/mg-mg/" aria-label="Open Malagasy benchmark; Madagascar"><i class="rx2-language-icon" aria-hidden="true" title="Madagascar">🇲🇬</i><span>Malagasy</span><small>mg-mg-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/mt-mt/" aria-label="Open Maltese benchmark; Malta"><i class="rx2-language-icon" aria-hidden="true" title="Malta">🇲🇹</i><span>Maltese</span><small>mt-mt-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/gv-im/" aria-label="Open Manx benchmark; Isle of Man"><i class="rx2-language-icon" aria-hidden="true" title="Isle of Man">🇮🇲</i><span>Manx</span><small>gv-im-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/mi-nz/" aria-label="Open Maori benchmark; New Zealand"><i class="rx2-language-icon" aria-hidden="true" title="New Zealand">🇳🇿</i><span>Maori</span><small>mi-nz-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/arn/" aria-label="Open Mapudungun benchmark; Transnational language"><i class="rx2-language-icon" aria-hidden="true" title="Transnational language">🌐</i><span>Mapudungun</span><small>arn-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/frm/" aria-label="Open Middle French benchmark; Historical language"><i class="rx2-language-icon" aria-hidden="true" title="Historical language">🌐</i><span>Middle French</span><small>frm-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/gmh/" aria-label="Open Middle High German benchmark; Historical language"><i class="rx2-language-icon" aria-hidden="true" title="Historical language">🌐</i><span>Middle High German</span><small>gmh-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/gml/" aria-label="Open Middle Low German benchmark; Historical language"><i class="rx2-language-icon" aria-hidden="true" title="Historical language">🌐</i><span>Middle Low German</span><small>gml-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/mn-mn/" aria-label="Open Mongolian benchmark; Mongolia"><i class="rx2-language-icon" aria-hidden="true" title="Mongolia">🇲🇳</i><span>Mongolian</span><small>mn-mn-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/mwf/" aria-label="Open Murrinh-Patha benchmark; Australia"><i class="rx2-language-icon" aria-hidden="true" title="Australia">🇦🇺</i><span>Murrinh-Patha</span><small>mwf-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/nav/" aria-label="Open Navajo benchmark; United States"><i class="rx2-language-icon" aria-hidden="true" title="United States">🇺🇸</i><span>Navajo</span><small>nav-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/nap/" aria-label="Open Neapolitan benchmark; Italy"><i class="rx2-language-icon" aria-hidden="true" title="Italy">🇮🇹</i><span>Neapolitan</span><small>nap-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/frr/" aria-label="Open North Frisian benchmark; Germany"><i class="rx2-language-icon" aria-hidden="true" title="Germany">🇩🇪</i><span>North Frisian</span><small>frr-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/sme/" aria-label="Open Northern Sami benchmark; Transnational language"><i class="rx2-language-icon" aria-hidden="true" title="Transnational language">🌐</i><span>Northern Sami</span><small>sme-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/norwegian-bokmal/" aria-label="Open Norwegian Bokmål benchmark; Norway"><i class="rx2-language-icon" aria-hidden="true" title="Norway">🇳🇴</i><span>Norwegian Bokmål</span><small>nb-no-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/norwegian-nynorsk/" aria-label="Open Norwegian Nynorsk benchmark; Norway"><i class="rx2-language-icon" aria-hidden="true" title="Norway">🇳🇴</i><span>Norwegian Nynorsk</span><small>nn-no-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/oci/" aria-label="Open Occitan benchmark; Transnational language"><i class="rx2-language-icon" aria-hidden="true" title="Transnational language">🌐</i><span>Occitan</span><small>oci-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/ang/" aria-label="Open Old English benchmark; Historical language"><i class="rx2-language-icon" aria-hidden="true" title="Historical language">🌐</i><span>Old English</span><small>ang-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/fro/" aria-label="Open Old French benchmark; Historical language"><i class="rx2-language-icon" aria-hidden="true" title="Historical language">🌐</i><span>Old French</span><small>fro-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/goh/" aria-label="Open Old High German benchmark; Historical language"><i class="rx2-language-icon" aria-hidden="true" title="Historical language">🌐</i><span>Old High German</span><small>goh-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/sga/" aria-label="Open Old Irish benchmark; Historical language"><i class="rx2-language-icon" aria-hidden="true" title="Historical language">🌐</i><span>Old Irish</span><small>sga-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/non/" aria-label="Open Old Norse benchmark; Historical language"><i class="rx2-language-icon" aria-hidden="true" title="Historical language">🌐</i><span>Old Norse</span><small>non-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/osx/" aria-label="Open Old Saxon benchmark; Historical language"><i class="rx2-language-icon" aria-hidden="true" title="Historical language">🌐</i><span>Old Saxon</span><small>osx-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/ood/" aria-label="Open Oodham benchmark; United States"><i class="rx2-language-icon" aria-hidden="true" title="United States">🇺🇸</i><span>Oodham</span><small>ood-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/om-et/" aria-label="Open Oromo benchmark; Transnational language"><i class="rx2-language-icon" aria-hidden="true" title="Transnational language">🌐</i><span>Oromo</span><small>om-et-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/ps-af/" aria-label="Open Pashto benchmark; Transnational language"><i class="rx2-language-icon" aria-hidden="true" title="Transnational language">🌐</i><span>Pashto</span><small>ps-af-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/persian/" aria-label="Open Persian benchmark; Iran"><i class="rx2-language-icon" aria-hidden="true" title="Iran">🇮🇷</i><span>Persian</span><small>fa-ir-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/polish/" aria-label="Open Polish benchmark; Poland"><i class="rx2-language-icon" aria-hidden="true" title="Poland">🇵🇱</i><span>Polish</span><small>pl-pl-unimorph + pl-pl-polimorf</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/portuguese/" aria-label="Open Portuguese benchmark; Portugal"><i class="rx2-language-icon" aria-hidden="true" title="Portugal">🇵🇹</i><span>Portuguese</span><small>pt-pt-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/que/" aria-label="Open Quechua benchmark; Transnational language"><i class="rx2-language-icon" aria-hidden="true" title="Transnational language">🌐</i><span>Quechua</span><small>que-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/ro-ro/" aria-label="Open Romanian benchmark; Romania"><i class="rx2-language-icon" aria-hidden="true" title="Romania">🇷🇴</i><span>Romanian</span><small>ro-ro-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/russian/" aria-label="Open Russian benchmark; Russia"><i class="rx2-language-icon" aria-hidden="true" title="Russia">🇷🇺</i><span>Russian</span><small>ru-ru-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/san/" aria-label="Open Sanskrit benchmark; Historical and transnational language"><i class="rx2-language-icon" aria-hidden="true" title="Historical and transnational language">🌐</i><span>Sanskrit</span><small>san-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/gd-gb/" aria-label="Open Scottish Gaelic benchmark; United Kingdom"><i class="rx2-language-icon" aria-hidden="true" title="United Kingdom">🇬🇧</i><span>Scottish Gaelic</span><small>gd-gb-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/see/" aria-label="Open Seneca benchmark; Transnational language"><i class="rx2-language-icon" aria-hidden="true" title="Transnational language">🌐</i><span>Seneca</span><small>see-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/hbs/" aria-label="Open Serbo-Croatian benchmark; Transnational language"><i class="rx2-language-icon" aria-hidden="true" title="Transnational language">🌐</i><span>Serbo-Croatian</span><small>hbs-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/shp/" aria-label="Open Shipibo-Conibo benchmark; Peru"><i class="rx2-language-icon" aria-hidden="true" title="Peru">🇵🇪</i><span>Shipibo-Conibo</span><small>shp-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/sn-zw/" aria-label="Open Shona benchmark; Zimbabwe"><i class="rx2-language-icon" aria-hidden="true" title="Zimbabwe">🇿🇼</i><span>Shona</span><small>sn-zw-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/sl-si/" aria-label="Open Slovenian benchmark; Slovenia"><i class="rx2-language-icon" aria-hidden="true" title="Slovenia">🇸🇮</i><span>Slovenian</span><small>sl-si-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/st-za/" aria-label="Open Sotho, Southern benchmark; Transnational language"><i class="rx2-language-icon" aria-hidden="true" title="Transnational language">🌐</i><span>Sotho, Southern</span><small>st-za-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/sdh/" aria-label="Open Southern Kurdish benchmark; Transnational language"><i class="rx2-language-icon" aria-hidden="true" title="Transnational language">🌐</i><span>Southern Kurdish</span><small>sdh-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/spanish/" aria-label="Open Spanish benchmark; Spain"><i class="rx2-language-icon" aria-hidden="true" title="Spain">🇪🇸</i><span>Spanish</span><small>es-es-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/swedish/" aria-label="Open Swedish benchmark; Sweden"><i class="rx2-language-icon" aria-hidden="true" title="Sweden">🇸🇪</i><span>Swedish</span><small>sv-se-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/tl-ph/" aria-label="Open Tagalog benchmark; Philippines"><i class="rx2-language-icon" aria-hidden="true" title="Philippines">🇵🇭</i><span>Tagalog</span><small>tl-ph-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/tg-tj/" aria-label="Open Tajik benchmark; Tajikistan"><i class="rx2-language-icon" aria-hidden="true" title="Tajikistan">🇹🇯</i><span>Tajik</span><small>tg-tj-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/tat/" aria-label="Open Tatar benchmark; Russia"><i class="rx2-language-icon" aria-hidden="true" title="Russia">🇷🇺</i><span>Tatar</span><small>tat-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/ote/" aria-label="Open Temoaya Otomi benchmark; Mexico"><i class="rx2-language-icon" aria-hidden="true" title="Mexico">🇲🇽</i><span>Temoaya Otomi</span><small>ote-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/tr-tr/" aria-label="Open Turkish benchmark; Turkey"><i class="rx2-language-icon" aria-hidden="true" title="Turkey">🇹🇷</i><span>Turkish</span><small>tr-tr-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/tk-tm/" aria-label="Open Turkmen benchmark; Turkmenistan"><i class="rx2-language-icon" aria-hidden="true" title="Turkmenistan">🇹🇲</i><span>Turkmen</span><small>tk-tm-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/tyv/" aria-label="Open Tuvan benchmark; Russia"><i class="rx2-language-icon" aria-hidden="true" title="Russia">🇷🇺</i><span>Tuvan</span><small>tyv-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/ukrainian/" aria-label="Open Ukrainian benchmark; Ukraine"><i class="rx2-language-icon" aria-hidden="true" title="Ukraine">🇺🇦</i><span>Ukrainian</span><small>uk-ua-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/hsb/" aria-label="Open Upper Sorbian benchmark; Germany"><i class="rx2-language-icon" aria-hidden="true" title="Germany">🇩🇪</i><span>Upper Sorbian</span><small>hsb-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/ur-pk/" aria-label="Open Urdu benchmark; Pakistan"><i class="rx2-language-icon" aria-hidden="true" title="Pakistan">🇵🇰</i><span>Urdu</span><small>ur-pk-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/ug-cn/" aria-label="Open Uyghur benchmark; China"><i class="rx2-language-icon" aria-hidden="true" title="China">🇨🇳</i><span>Uyghur</span><small>ug-cn-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/uz-uz/" aria-label="Open Uzbek benchmark; Uzbekistan"><i class="rx2-language-icon" aria-hidden="true" title="Uzbekistan">🇺🇿</i><span>Uzbek</span><small>uz-uz-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/vec/" aria-label="Open Venetian benchmark; Italy"><i class="rx2-language-icon" aria-hidden="true" title="Italy">🇮🇹</i><span>Venetian</span><small>vec-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/vro/" aria-label="Open Voro benchmark; Estonia"><i class="rx2-language-icon" aria-hidden="true" title="Estonia">🇪🇪</i><span>Voro</span><small>vro-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/vot/" aria-label="Open Votic benchmark; Russia"><i class="rx2-language-icon" aria-hidden="true" title="Russia">🇷🇺</i><span>Votic</span><small>vot-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/cy-gb/" aria-label="Open Welsh benchmark; United Kingdom"><i class="rx2-language-icon" aria-hidden="true" title="United Kingdom">🇬🇧</i><span>Welsh</span><small>cy-gb-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/fy-nl/" aria-label="Open West Frisian benchmark; Netherlands"><i class="rx2-language-icon" aria-hidden="true" title="Netherlands">🇳🇱</i><span>West Frisian</span><small>fy-nl-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/ctp/" aria-label="Open Western Highland Chatino benchmark; Mexico"><i class="rx2-language-icon" aria-hidden="true" title="Mexico">🇲🇽</i><span>Western Highland Chatino</span><small>ctp-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/sjo/" aria-label="Open Xibe benchmark; Transnational language"><i class="rx2-language-icon" aria-hidden="true" title="Transnational language">🌐</i><span>Xibe</span><small>sjo-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/xty/" aria-label="Open Xipinawa benchmark; Mexico"><i class="rx2-language-icon" aria-hidden="true" title="Mexico">🇲🇽</i><span>Xipinawa</span><small>xty-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/sah/" aria-label="Open Yakut benchmark; Russia"><i class="rx2-language-icon" aria-hidden="true" title="Russia">🇷🇺</i><span>Yakut</span><small>sah-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/ame/" aria-label="Open Yamana benchmark; Peru"><i class="rx2-language-icon" aria-hidden="true" title="Peru">🇵🇪</i><span>Yamana</span><small>ame-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/yiddish/" aria-label="Open Yiddish benchmark; Transnational language"><i class="rx2-language-icon" aria-hidden="true" title="Transnational language">🌐</i><span>Yiddish</span><small>yi-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/zpv/" aria-label="Open Zacatepec Chatino benchmark; Mexico"><i class="rx2-language-icon" aria-hidden="true" title="Mexico">🇲🇽</i><span>Zacatepec Chatino</span><small>zpv-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/dje/" aria-label="Open Zarma benchmark; Niger"><i class="rx2-language-icon" aria-hidden="true" title="Niger">🇳🇪</i><span>Zarma</span><small>dje-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/czn/" aria-label="Open Zenzontepec Chatino benchmark; Mexico"><i class="rx2-language-icon" aria-hidden="true" title="Mexico">🇲🇽</i><span>Zenzontepec Chatino</span><small>czn-default</small></a><a class="rx2-language" href="{{ base_url }}/benchmarks/languages/zu-za/" aria-label="Open Zulu benchmark; South Africa"><i class="rx2-language-icon" aria-hidden="true" title="South Africa">🇿🇦</i><span>Zulu</span><small>zu-za-default</small></a></div>
+</section>
+<!-- JAVA-LANGUAGE-PAGES:END -->
 
-    <section class="rx2-code-grid">
+<section class="rx2-code-grid">
       <article class="rx2-code-card"><div class="rx2-code-title"><span class="java-symbol">♨</span><b>Java</b><button type="button" data-copy="java-code">Copy</button></div><pre id="java-code"><code><span class="kw">import</span> org.egothor.stemmer.StemmerPatchTrieLoader;
 <span class="kw">import</span> org.egothor.stemmer.FrequencyTrie;
 
@@ -151,7 +193,7 @@ stems = stemmer.stem_batch([
       <div><svg viewBox="0 0 32 32" aria-hidden="true"><path d="m7 9 9-5 9 5v14l-9 5-9-5Z" fill="none" stroke="currentColor" stroke-width="2"/><path d="m7 9 9 5 9-5M16 14v14" fill="none" stroke="currentColor" stroke-width="2"/></svg><p><b>Learned transformations</b><span>Lexical evidence becomes<br>compact patch commands</span></p></div>
       <div><svg viewBox="0 0 32 32" aria-hidden="true"><path d="M5 24c4-9 9-13 22-16M8 9h7v7" fill="none" stroke="currentColor" stroke-width="2.2" stroke-linecap="round" stroke-linejoin="round"/><path d="M4 27h24" stroke="currentColor" stroke-width="2.2" stroke-linecap="round"/></svg><p><b>Beyond the dictionary</b><span>Runtime is not a closed<br>word-to-lemma lookup</span></p></div>
       <div><svg viewBox="0 0 32 32" aria-hidden="true"><path d="M18 2 8 18h8l-2 12 10-17h-8Z" fill="none" stroke="currentColor" stroke-width="2.2" stroke-linejoin="round"/></svg><p><b>Native speed</b><span>Multi-million words per<br>second in Python and Java</span></p></div>
-      <div><svg viewBox="0 0 32 32" aria-hidden="true"><circle cx="16" cy="16" r="12" fill="none" stroke="currentColor" stroke-width="2"/><path d="M4 16h24M16 4c4 4 5 8 5 12s-1 8-5 12c-4-4-5-8-5-12s1-8 5-12Z" fill="none" stroke="currentColor" stroke-width="2"/></svg><p><b>20 language models</b><span>One architecture, trained<br>from language-specific data</span></p></div>
+      <div><svg viewBox="0 0 32 32" aria-hidden="true"><circle cx="16" cy="16" r="12" fill="none" stroke="currentColor" stroke-width="2"/><path d="M4 16h24M16 4c4 4 5 8 5 12s-1 8-5 12c-4-4-5-8-5-12s1-8 5-12Z" fill="none" stroke="currentColor" stroke-width="2"/></svg><p><b>167 model artifacts</b><span>One architecture, trained<br>from language-specific data</span></p></div>
       <div><svg viewBox="0 0 32 32" aria-hidden="true"><path d="m16 3 10 4v7c0 8-4 12-10 15C10 26 6 22 6 14V7Z" fill="none" stroke="currentColor" stroke-width="2.1"/><path d="m11 15 3 3 7-7" fill="none" stroke="currentColor" stroke-width="2.1" stroke-linecap="round" stroke-linejoin="round"/></svg><p><b>Production ready</b><span>Deterministic behavior<br>and reproducible evidence</span></p></div>
     </section>
 
diff --git a/docs/stemmer-model-catalog.md b/docs/stemmer-model-catalog.md
index 79faee3..e29bbb9 100644
--- a/docs/stemmer-model-catalog.md
+++ b/docs/stemmer-model-catalog.md
@@ -2,24 +2,147 @@
 
 | Model ID | Language | Default | Coordinates | Version | Source | Repository | Source version | Revision | Revision status | License | Attribution | SHA-256 | Bytes |
 |---|---|---:|---|---:|---|---|---|---|---|---|---|---|---:|
+| ady-default | ADY | true | org.egothor:radixor-model-ady-default | 1.0.0 | UniMorph | https://github.com/unimorph/ady | 1ce60b56fc1ae6c4a242d264aabb4ce76f23e509 | 1ce60b56fc1ae6c4a242d264aabb4ce76f23e509 | recorded | CC-BY-SA-3.0 | UniMorph | a6ec25c18612cfdc77d43aba4bf49ec3b6cf07e3e484af404a788dff7e944e16 | 59000 |
+| af-za-default | AF_ZA | true | org.egothor:radixor-model-af-za-default | 1.0.0 | UniMorph / AfriTools | https://github.com/unimorph/afr | cc2df0c0de1a4e44a96739c1670c52d386cbf401 | cc2df0c0de1a4e44a96739c1670c52d386cbf401 | recorded | CC-BY-SA-3.0 | UniMorph; Peter Dirix | ab23228100b7671fa7a21a23c6c8662a71de69584b277033d5160806542b4bbd | 1023904 |
+| afb-default | AFB | true | org.egothor:radixor-model-afb-default | 1.0.0 | UniMorph | https://github.com/unimorph/afb | aca24ea0d194d5fec56e18ee7cb341bb93be729e | aca24ea0d194d5fec56e18ee7cb341bb93be729e | recorded | CC-BY-SA-3.0 | UniMorph; Salam Khalifa and Nizar Habash ([CAMeL Lab](www.camel-lab.com) @ NYU Abu Dhabi) | 7ba0759227455725063648b569652e0442452501fa58b763f3861a3828655ba7 | 98196 |
+| ail-default | AIL | true | org.egothor:radixor-model-ail-default | 1.0.0 | UniMorph | https://github.com/unimorph/ail | e88b9d137859320ec5bcb406d9c3eb54d8e38c33 | e88b9d137859320ec5bcb406d9c3eb54d8e38c33 | recorded | CC-BY-SA-3.0 | UniMorph; Grant Aiton, Edoardo Maria Ponti, Ekaterina Vylomova | a462d3bc94670a2e37c292d027e70892a35c221441500242b5fa9e350fbfba6f | 11536 |
+| aka-default | AK | true | org.egothor:radixor-model-aka-default | 1.0.0 | UniMorph | https://github.com/unimorph/aka | f30f6bceef61f64dae87f8e78d4c02a7cd920a07 | f30f6bceef61f64dae87f8e78d4c02a7cd920a07 | recorded | CC-BY-SA-3.0 | UniMorph; Tiago Pimentel | 7dd8e1835f51ced8bc3f87409c1ce6eb089ef0e5cde20111c44282ed9ca498b0 | 4333 |
+| am-et-default | AM_ET | true | org.egothor:radixor-model-am-et-default | 1.0.0 | UniMorph | https://github.com/unimorph/amh | 02db5a185339f6d7a3440f05a4f1d1ca0131881a | 02db5a185339f6d7a3440f05a4f1d1ca0131881a | recorded | CC-BY-SA-3.0 | UniMorph | 69e9089fae06a465459f288a4e2074ffcb9656bf2b6fd01772a2af3f71e5bb70 | 130439 |
+| ame-default | AME | true | org.egothor:radixor-model-ame-default | 1.0.0 | UniMorph | https://github.com/unimorph/ame | a3119da64f3b740d7f9b9dcbd4189fb5c8ac4a85 | a3119da64f3b740d7f9b9dcbd4189fb5c8ac4a85 | recorded | CC-BY-SA-3.0 | UniMorph; Arturo Oncevay ,Gema Celeste Silva Villegas, Juan López Bautista, Didier López Francis, Roberto Zariquiey | e59d252da23b0152d213b1ccff688971bfb49971e2ffcc50b991340516d8feaa | 7356 |
+| ang-default | ANG | true | org.egothor:radixor-model-ang-default | 1.0.0 | UniMorph | https://github.com/unimorph/ang | 825a9e35208fd8fcbc340bd8129d69012386e048 | 825a9e35208fd8fcbc340bd8129d69012386e048 | recorded | CC-BY-SA-3.0 | UniMorph | 1cb0c94c9df7b61676399d163f0a6b7edc2c9d4c14b7560d8ad3667b447d70cf | 170423 |
+| ar-default | AR | true | org.egothor:radixor-model-ar-default | 1.0.1 | UniMorph | https://github.com/unimorph/ara | e9b7521ab75e3a7da6f996a428031164993e6f19 | e9b7521ab75e3a7da6f996a428031164993e6f19 | recorded | CC-BY-SA-3.0 | UniMorph; Salam Khalifa and Nizar Habash (CAMeL Lab @ NYU Abu Dhabi) | 36fe6a7a80e922f423e80761e27f21c61d86cc7f1b2837a1989761af4d51c904 | 1329163 |
+| arn-default | ARN | true | org.egothor:radixor-model-arn-default | 1.0.0 | UniMorph | https://github.com/unimorph/arn | f788f46e295e87e23980cc2634c2449040f05b1c | f788f46e295e87e23980cc2634c2449040f05b1c | recorded | CC-BY-SA-3.0 | UniMorph | a9d691ff4f8d5b9a8c87baa400e70754295a107f81a157ffcf9b0063034e9a99 | 1335 |
+| arz-default | ARZ | true | org.egothor:radixor-model-arz-default | 1.0.0 | UniMorph | https://github.com/unimorph/arz | e3fd562845647176b8be9b8e21710732bc76d558 | e3fd562845647176b8be9b8e21710732bc76d558 | recorded | CC-BY-SA-3.0 | UniMorph; Salam Khalifa and Nizar Habash ([CAMeL Lab](www.camel-lab.com) @ NYU Abu Dhabi) | d5269ffb7e66b59abe39c80415817a7a6ff1cafaa196f9a64d66e74408e70a4f | 61436 |
+| as-in-default | AS_IN | true | org.egothor:radixor-model-as-in-default | 1.0.0 | UniMorph | https://github.com/unimorph/asm | 2763a62176e3cf4ce29a6b555624cc13a5df389d | 2763a62176e3cf4ce29a6b555624cc13a5df389d | recorded | CC-BY-SA-3.0 | UniMorph; Khuyagbaatar Batsuren <br />; Aryaman Arora | 07a07c7834b55fe980ed5be459aa9d724f54ba4f2449e714aff6b6bce960cba3 | 202408 |
+| ast-default | AST | true | org.egothor:radixor-model-ast-default | 1.0.0 | UniMorph | https://github.com/unimorph/ast | 04c0e3bfb91704dcd1614ae0b81faec803921c6d | 04c0e3bfb91704dcd1614ae0b81faec803921c6d | recorded | CC-BY-SA-3.0 | UniMorph | ace7643781385ebd1c162eb88c4f2b3321048b2e5d4e733627a7eeba42a76bb2 | 50934 |
+| aym-default | AYM | true | org.egothor:radixor-model-aym-default | 1.0.0 | UniMorph | https://github.com/unimorph/aym | ccd338ca3465c1bce58ebb4e379d1db36d883af4 | ccd338ca3465c1bce58ebb4e379d1db36d883af4 | recorded | CC-BY-SA-3.0 | UniMorph; Matt Coler, Eleanor Chodroff | a6f0af3899e1b7b84080d727a87672822719a6e5a151aed411fda07b7bfbd353 | 766138 |
+| az-az-default | AZ_AZ | true | org.egothor:radixor-model-az-az-default | 1.0.0 | UniMorph | https://github.com/unimorph/aze | fd9fa8f004be414c1a837bf0845697cc94f553a1 | fd9fa8f004be414c1a837bf0845697cc94f553a1 | recorded | CC-BY-SA-3.0 | UniMorph | 4d61808b68c94830b228d712b769216171c0a2c4f47be7026f9c6a61e68c595a | 16388 |
+| azg-default | AZG | true | org.egothor:radixor-model-azg-default | 1.0.0 | UniMorph | https://github.com/unimorph/azg | 9bfbc5daa9aa2cee06367e3293c1068f5ae582d5 | 9bfbc5daa9aa2cee06367e3293c1068f5ae582d5 | recorded | CC-BY-SA-3.0 | UniMorph; Antonis Anastasopoulos | 67312c026040f0d7a2eef4448bbb23e462598893ead820db1cfc6c59530f48da | 29618 |
+| bak-default | BAK | true | org.egothor:radixor-model-bak-default | 1.0.0 | UniMorph | https://github.com/unimorph/bak | 39e1618c37e900a30ed1e06e95baffb98f382878 | 39e1618c37e900a30ed1e06e95baffb98f382878 | recorded | CC-BY-SA-3.0 | UniMorph | ecc3eafd8f269822efc6aab62f8621de561febc14a85f78b56d393e2e40b577f | 33038 |
+| be-by-default | BE_BY | true | org.egothor:radixor-model-be-by-default | 1.0.0 | UniMorph | https://github.com/unimorph/bel | cba29019d3ba581c26d4152ead81c27ff16a6550 | cba29019d3ba581c26d4152ead81c27ff16a6550 | recorded | CC-BY-SA-3.0 | UniMorph | 52b674acaf9e28461b0e638f562e98aee84461ee7acf6c0693d8a36e79f38548 | 55855 |
+| bg-bg-default | BG_BG | true | org.egothor:radixor-model-bg-bg-default | 1.0.0 | UniMorph | https://github.com/unimorph/bul | b60b78dbf1d65867419a260e91ca213dc355c2f3 | b60b78dbf1d65867419a260e91ca213dc355c2f3 | recorded | CC-BY-SA-3.0 | UniMorph | dd6b424edc9d8a54626d3ee9408feda774a8a13215840074c912752ee37620d2 | 135076 |
+| bn-bd-default | BN_BD | true | org.egothor:radixor-model-bn-bd-default | 1.0.0 | UniMorph | https://github.com/unimorph/ben | 55a44fa60e9b7a3a5ff7acbed66b07571ceefa79 | 55a44fa60e9b7a3a5ff7acbed66b07571ceefa79 | recorded | CC-BY-SA-3.0 | UniMorph | 0c5a4e2eab4431ef381e54d74b5b8d5d102fe47902882034a8d03f809090b4a5 | 8611 |
+| bra-default | BRA | true | org.egothor:radixor-model-bra-default | 1.0.0 | UniMorph | https://github.com/unimorph/bra | 090298dda0035e28c2fa9427421c5926df0f0112 | 090298dda0035e28c2fa9427421c5926df0f0112 | recorded | CC-BY-SA-3.0 | UniMorph; Shyam Ratan, Ritesh Kumar | 2c9f7a8f9bbca119cc735633b5dacd9d551267c69c36d0ddfaa3ae78bfef6475 | 8182 |
+| bre-default | BRE | true | org.egothor:radixor-model-bre-default | 1.0.0 | UniMorph | https://github.com/unimorph/bre | 2f20b8cc95217ffce35e0f8f6fa31c29955c271a | 2f20b8cc95217ffce35e0f8f6fa31c29955c271a | recorded | CC-BY-SA-3.0 | UniMorph | f27b933c1a5c37c83a0824ab1b033f2816269d1da107e18232786fe3914e0e02 | 4424 |
+| ca-es-default | CA_ES | true | org.egothor:radixor-model-ca-es-default | 1.0.0 | UniMorph | https://github.com/unimorph/cat | 5fe904e0fb37867b4d9e6099fd0798aeeba37d14 | 5fe904e0fb37867b4d9e6099fd0798aeeba37d14 | recorded | CC-BY-SA-3.0 | UniMorph | f6f76fd51fcfa7619968fecbab2a77cc53389d251e203cf3fc2452a0dee6f63e | 343650 |
+| ceb-default | CEB | true | org.egothor:radixor-model-ceb-default | 1.0.0 | UniMorph | https://github.com/unimorph/ceb | edbada3630dc12c4025f0b80d5a745a256ef25f6 | edbada3630dc12c4025f0b80d5a745a256ef25f6 | recorded | CC-BY-SA-3.0 | UniMorph; Ran Zmigrod | f33a8b4a366ace91154808dd5d51ea089114cca5ba343cdd3d9b1e6c51f57b64 | 1251 |
+| chu-default | CHU | true | org.egothor:radixor-model-chu-default | 1.0.0 | UniMorph | https://github.com/unimorph/chu | 35f1e2af69f3f618fe6a95245198c380852d3337 | 35f1e2af69f3f618fe6a95245198c380852d3337 | recorded | CC-BY-SA-3.0 | UniMorph | 1c5d552abc5ed624bcb7af7533e286e7fd1fa9d44f34e38c32360226cb1f0b90 | 5445 |
+| ckt-default | CKT | true | org.egothor:radixor-model-ckt-default | 1.0.0 | UniMorph | https://github.com/unimorph/ckt | facbd482ef2558916b7283b398c9810a8649e7f2 | facbd482ef2558916b7283b398c9810a8649e7f2 | recorded | CC-BY-SA-3.0 | UniMorph; Karina Sheifer, Maria Ryskina | 1ac8387d6dad002370756edfb57286e2add869d0591b9551e7dfbcf24cda4367 | 1944 |
+| cly-default | CLY | true | org.egothor:radixor-model-cly-default | 1.0.0 | UniMorph | https://github.com/unimorph/cly | b8a09dda02f38b4858f8c77ecb3d9b198cf9d5f9 | b8a09dda02f38b4858f8c77ecb3d9b198cf9d5f9 | recorded | CC-BY-SA-3.0 | UniMorph; Hilaria Cruz and Antonis Anastasopoulos | 7d9e70a7f756343f88bad3ac07d9108da9e0e38659cf3a0ba0d838a55180a6b6 | 5869 |
+| cni-default | CNI | true | org.egothor:radixor-model-cni-default | 1.0.0 | UniMorph | https://github.com/unimorph/cni | 0d8a9abca12da88a5ca00b3e04113fcf4d2382d1 | 0d8a9abca12da88a5ca00b3e04113fcf4d2382d1 | recorded | CC-BY-SA-3.0 | UniMorph; Jaime Rafael Montoya Samame, Esaú Zumaeta Rojas, Delio Siticonatzi C., Roberto Zariquiey, Arturo Oncevay | 13edaa2b6fad0c81eac550d3214fd1ecdf258184a2568a91d5d8c3cd04d44766 | 24712 |
+| cor-default | COR | true | org.egothor:radixor-model-cor-default | 1.0.0 | UniMorph | https://github.com/unimorph/cor | 70830c56c7d5183a041bc176067aed2cba40380a | 70830c56c7d5183a041bc176067aed2cba40380a | recorded | CC-BY-SA-3.0 | UniMorph | 71602207574edad2377b94ef48821f29338c13ccf1ee16419f69f10476e48a5a | 544 |
+| cpa-default | CPA | true | org.egothor:radixor-model-cpa-default | 1.0.0 | UniMorph | https://github.com/unimorph/cpa | 5ecd884b1955cf4d738c8c13c2cab05c7d1cbcc8 | 5ecd884b1955cf4d738c8c13c2cab05c7d1cbcc8 | recorded | CC-BY-SA-3.0 | UniMorph; Antonis Anastasopoulos | 6f2c06058918d2419d53f1698071588a79e0cb5475ad8e1dd286e86f22eeeb89 | 7628 |
+| cre-default | CRE | true | org.egothor:radixor-model-cre-default | 1.0.0 | UniMorph | https://github.com/unimorph/cre | 46a26589123dc5327ee30e2c77375bbb93a5afd7 | 46a26589123dc5327ee30e2c77375bbb93a5afd7 | recorded | CC-BY-SA-3.0 | UniMorph; Eleanor Chodroff | e70a313201ff437268acc5955ea26caa646a4c3d5dca5e898b957b02f691544e | 3097 |
+| crh-default | CRH | true | org.egothor:radixor-model-crh-default | 1.0.0 | UniMorph | https://github.com/unimorph/crh | a926b51a71a5cc7a0eab16056243a82e3439eaa3 | a926b51a71a5cc7a0eab16056243a82e3439eaa3 | recorded | CC-BY-SA-3.0 | UniMorph | 4698c8efadf49f53c89e0fc7b311a5bd70f40490b8dce671394677b9078d8a2c | 20519 |
 | cs-cz-default | CS_CZ | true | org.egothor:radixor-model-cs-cz-default | 1.0.0 | UniMorph | https://github.com/unimorph/ces | not-recorded-in-legacy-import | not-recorded-in-legacy-import | not-recorded-in-legacy-import | CC-BY-SA-3.0 | UniMorph; Witold Kieraś is credited for the separate MorfFlex-CZ conversion | 62afdaa6dc7a721b54a0dc278a0c648a63ad52a34a412d27b5b52fbcde9c1ce4 | 142365 |
+| csb-default | CSB | true | org.egothor:radixor-model-csb-default | 1.0.0 | UniMorph | https://github.com/unimorph/csb | 62793ce5aa6e76c3a7742180f0d0a90a6906ceb7 | 62793ce5aa6e76c3a7742180f0d0a90a6906ceb7 | recorded | CC-BY-SA-3.0 | UniMorph | 32cf30a085941969f1acc4149152d515efa99fd1061c4ddd9eb1049e5cc3097b | 1235 |
+| ctp-default | CTP | true | org.egothor:radixor-model-ctp-default | 1.0.0 | UniMorph | https://github.com/unimorph/ctp | f006ec4502b3e0b1512dbcf1136c44f593fa4910 | f006ec4502b3e0b1512dbcf1136c44f593fa4910 | recorded | CC-BY-SA-3.0 | UniMorph; Antonis Anastasopoulos | 36b5b636f158d0d65762e87c2125617847cf1b8c6fd766ddf4c42e4343b39122 | 7004 |
+| czn-default | CZN | true | org.egothor:radixor-model-czn-default | 1.0.0 | UniMorph | https://github.com/unimorph/czn | 42c150c93bd81252592961e206e09a1aa64865b9 | 42c150c93bd81252592961e206e09a1aa64865b9 | recorded | CC-BY-SA-3.0 | UniMorph; Antonis Anastasopoulos | f8d1d16cfd2a023c322a50965f7437944cad928f716c268a18eccb39c4dab66a | 5817 |
 | da-dk-default | DA_DK | true | org.egothor:radixor-model-da-dk-default | 1.0.0 | UniMorph | https://github.com/unimorph/dan | not-recorded-in-legacy-import | not-recorded-in-legacy-import | not-recorded-in-legacy-import | CC-BY-SA-3.0 | UniMorph and Wikipedia contributors | 3f7b670a0e7b872bda0381f5154ce058a4656297b39b7157b4ccf6560257cb90 | 73030 |
+| dak-default | DAK | true | org.egothor:radixor-model-dak-default | 1.0.0 | UniMorph | https://github.com/unimorph/dak | e374c530a660fcf13252bc473adcec409b9b302b | e374c530a660fcf13252bc473adcec409b9b302b | recorded | CC-BY-SA-3.0 | UniMorph; Eleanor Chodroff | a867bb0f4165fc47b0906c179213d8563088a3e5f13d161120a2556fa1d9032a | 9139 |
 | de-de-default | DE_DE | true | org.egothor:radixor-model-de-de-default | 1.0.0 | UniMorph | https://github.com/unimorph/deu | not-recorded-in-legacy-import | not-recorded-in-legacy-import | not-recorded-in-legacy-import | CC-BY-SA-3.0 | UniMorph and English Wiktionary contributors | cbfa038122823f02e4bdb54b0035492c356b6ecd80f11eb11290d7a7248a59f5 | 838450 |
+| dje-default | DJE | true | org.egothor:radixor-model-dje-default | 1.0.0 | UniMorph | https://github.com/unimorph/dje | 64cb65268fcf1c871b19936f1a279d25a5cc57ce | 64cb65268fcf1c871b19936f1a279d25a5cc57ce | recorded | CC-BY-SA-3.0 | UniMorph; Ran Zmigrod | e4a09508fed786d640720e44077b59abb5dac9d154712a2a0d721f2300ae20df | 377 |
+| dsb-default | DSB | true | org.egothor:radixor-model-dsb-default | 1.0.0 | UniMorph | https://github.com/unimorph/dsb | f94b5f674fcef872fbab5efe24fb279a7246ee6f | f94b5f674fcef872fbab5efe24fb279a7246ee6f | recorded | CC-BY-SA-3.0 | UniMorph | 0910f3977f825457d7f64f988c984a0400824e4d696991261211d3e29ad02d92 | 30546 |
+| el-gr-default | EL_GR | true | org.egothor:radixor-model-el-gr-default | 1.0.0 | UniMorph | https://github.com/unimorph/ell | fce1186efc7c8036fe2b6a13c110c2f2cd140e38 | fce1186efc7c8036fe2b6a13c110c2f2cd140e38 | recorded | CC-BY-SA-3.0 | UniMorph | 05ecc1b60b714c64ab5b6d29aef1a9544f0594f98640e5c66dd9ac324672ac79 | 259340 |
 | es-es-default | ES_ES | true | org.egothor:radixor-model-es-es-default | 1.0.0 | UniMorph | https://github.com/unimorph/spa | not-recorded-in-legacy-import | not-recorded-in-legacy-import | not-recorded-in-legacy-import | CC-BY-SA-3.0 | UniMorph and English Wiktionary contributors | 7a1ec94cfdb1e9a95431289d62dc5579cb2a532d99532eeda90290072e569721 | 2269280 |
-| fa-ir-default | FA_IR | true | org.egothor:radixor-model-fa-ir-default | 1.0.0 | UniMorph | https://github.com/unimorph/fas | not-recorded-in-legacy-import | not-recorded-in-legacy-import | not-recorded-in-legacy-import | CC-BY-SA-3.0 | UniMorph and Wikipedia contributors | b29a0d168a6a97f980666aa40b74a0edd8b6be4ab3320a7abfbb76b3529f4ea1 | 8934 |
+| et-ee-default | ET_EE | true | org.egothor:radixor-model-et-ee-default | 1.0.0 | UniMorph | https://github.com/unimorph/est | 97663990872e6bbc06701839d33462833676ebc2 | 97663990872e6bbc06701839d33462833676ebc2 | recorded | CC-BY-SA-3.0 | UniMorph | 0d7da990392011b0312fdee181720c2778ff07581912c718b8e7c1d07267c735 | 60662 |
+| evn-default | EVN | true | org.egothor:radixor-model-evn-default | 1.0.0 | UniMorph | https://github.com/unimorph/evn | cdbe7b44ba754403d60fd38818ac173bec13ba34 | cdbe7b44ba754403d60fd38818ac173bec13ba34 | recorded | CC-BY-SA-3.0 | UniMorph; Elena Klyachko | 4ff0de56329166d6d84c8443f0d6223507874854930b2ba2a339bc8be5045c0f | 46911 |
+| fa-ir-default | FA_IR | true | org.egothor:radixor-model-fa-ir-default | 1.0.0 | UniMorph | https://github.com/unimorph/fas | not-recorded-in-legacy-import | not-recorded-in-legacy-import | not-recorded-in-legacy-import | CC-BY-SA-3.0 | UniMorph | b29a0d168a6a97f980666aa40b74a0edd8b6be4ab3320a7abfbb76b3529f4ea1 | 8934 |
 | fi-fi-default | FI_FI | true | org.egothor:radixor-model-fi-fi-default | 1.0.0 | UniMorph | https://github.com/unimorph/fin | not-recorded-in-legacy-import | not-recorded-in-legacy-import | not-recorded-in-legacy-import | CC-BY-SA-3.0 | UniMorph and Wikipedia contributors | ca2628b3db31fee92f1b612ebbbd5e956a6dbbfb10e721325e55ef528f26072f | 4867450 |
+| fo-fo-default | FO_FO | true | org.egothor:radixor-model-fo-fo-default | 1.0.0 | UniMorph | https://github.com/unimorph/fao | 82bf05c94fbb54e13a48e0dac49a805d766e834e | 82bf05c94fbb54e13a48e0dac49a805d766e834e | recorded | CC-BY-SA-3.0 | UniMorph | c5ab9a35317d6cfbba64e8a4a441c1a8a5c3a922384631941804d0bf4eb81fc0 | 83579 |
 | fr-fr-default | FR_FR | true | org.egothor:radixor-model-fr-fr-default | 1.0.0 | UniMorph | https://github.com/unimorph/fra | not-recorded-in-legacy-import | not-recorded-in-legacy-import | not-recorded-in-legacy-import | CC-BY-SA-3.0 | UniMorph and Wikipedia contributors | a988658758952fd599dc7360e0234178a6d65ac46e5cedc7dcd325a7cb7e71d9 | 1117956 |
-| he-il-default | HE_IL | true | org.egothor:radixor-model-he-il-default | 1.0.0 | UniMorph | https://github.com/unimorph/heb | not-recorded-in-legacy-import | not-recorded-in-legacy-import | not-recorded-in-legacy-import | CC-BY-SA-3.0 | UniMorph; Omer Goldman (annotator); Wiktionary contributors | 9a47dc69bb7dab21aba0266b73cd74cdaeb17db94363796a0a56111ac8518256 | 179776 |
+| frm-default | FRM | true | org.egothor:radixor-model-frm-default | 1.0.0 | UniMorph | https://github.com/unimorph/frm | 7baa5932dbdd25f5f8c5d55c43a76e972b18cf8f | 7baa5932dbdd25f5f8c5d55c43a76e972b18cf8f | recorded | CC-BY-SA-3.0 | UniMorph | e4ed58dd8e6d2305bccf46d2b8273d663d2def09fce1046063a4e27b88a82cb7 | 62660 |
+| fro-default | FRO | true | org.egothor:radixor-model-fro-default | 1.0.0 | UniMorph | https://github.com/unimorph/fro | 7b0937f7ca90103b390e93f9bfab026e8daedff2 | 7b0937f7ca90103b390e93f9bfab026e8daedff2 | recorded | CC-BY-SA-3.0 | UniMorph | dab25d905a872c0623c952c102a958f72b6cf7fe39665c12e8a320b5455a9418 | 234310 |
+| frr-default | FRR | true | org.egothor:radixor-model-frr-default | 1.0.0 | UniMorph | https://github.com/unimorph/frr | 7882e34f921ba75a87166a648759764b702b2a26 | 7882e34f921ba75a87166a648759764b702b2a26 | recorded | CC-BY-SA-3.0 | UniMorph | d92dacea73857c8ddef8113c70a0c72eb5b1d1f6796ef1a4b640993aabc544c4 | 1178 |
+| fur-default | FUR | true | org.egothor:radixor-model-fur-default | 1.0.0 | UniMorph | https://github.com/unimorph/fur | cba86dc9e8d980e415ce4d0b89fa90a3eca33648 | cba86dc9e8d980e415ce4d0b89fa90a3eca33648 | recorded | CC-BY-SA-3.0 | UniMorph | 279631dc8998d6d55a2707d7746ba0e50fe9e5b7f1279e128ac48a70dcf06685 | 12239 |
+| ga-ie-default | GA_IE | true | org.egothor:radixor-model-ga-ie-default | 1.0.0 | UniMorph | https://github.com/unimorph/gle | efdca3ead351c804224b97336c3bcef8c78b6015 | efdca3ead351c804224b97336c3bcef8c78b6015 | recorded | CC-BY-SA-3.0 | UniMorph | b669438f4375857896dcf9adc535f46f25ddc12c2c163b6980dc33564c48991b | 79229 |
+| gaa-default | GAA | true | org.egothor:radixor-model-gaa-default | 1.0.0 | UniMorph | https://github.com/unimorph/gaa | 38a58c09fae385f486ff0df662728ce5380d3f14 | 38a58c09fae385f486ff0df662728ce5380d3f14 | recorded | CC-BY-SA-3.0 | UniMorph; Tiago Pimentel | 981733932ac87c90b142c9230e721513d642bac1d25f131a9c36643015b68108 | 1197 |
+| gal-default | GAL | true | org.egothor:radixor-model-gal-default | 1.0.0 | UniMorph | https://github.com/unimorph/gal | 7927b7b3b2379df0870c23fafb6481bda214c572 | 7927b7b3b2379df0870c23fafb6481bda214c572 | recorded | CC-BY-SA-3.0 | UniMorph | d57ad44d8863d70fb6c28152de8e96b1480ae66b9bacd59a27f7f0e9aa44f566 | 58522 |
+| gmh-default | GMH | true | org.egothor:radixor-model-gmh-default | 1.0.0 | UniMorph | https://github.com/unimorph/gmh | 768cb5bafa6343f2f99b1422e42ddde539517a35 | 768cb5bafa6343f2f99b1422e42ddde539517a35 | recorded | CC-BY-SA-3.0 | UniMorph | 9dfcff6fca22db34b3206e35ab50e1e7d91fd4ce81fd50b6a7461d8f6fe3e5d6 | 1088 |
+| gml-default | GML | true | org.egothor:radixor-model-gml-default | 1.0.0 | UniMorph | https://github.com/unimorph/gml | 8a6c09f6b88e61fd93293298b6d7be3ac2014c9a | 8a6c09f6b88e61fd93293298b6d7be3ac2014c9a | recorded | CC-BY-SA-3.0 | UniMorph | 3461579139d22f110969530d3a88820895c0e6818efc988c025549787e7c54da | 1741 |
+| goh-default | GOH | true | org.egothor:radixor-model-goh-default | 1.0.0 | UniMorph | https://github.com/unimorph/goh | d3c23c10fe0a1418182f1023c5db30fd44159709 | d3c23c10fe0a1418182f1023c5db30fd44159709 | recorded | CC-BY-SA-3.0 | UniMorph | 42f175991354ce30b3cf6563b9f8c7a0c1ecfc7593a8151fd1320b09575b0208 | 12889 |
+| got-default | GOT | true | org.egothor:radixor-model-got-default | 1.0.0 | UniMorph | https://github.com/unimorph/got | d3aa0b4386646cd6c7c24eb36796726d234ab418 | d3aa0b4386646cd6c7c24eb36796726d234ab418 | recorded | CC-BY-SA-3.0 | UniMorph; Khuyagbaatar Batsuren | d7f0bb16c028f9e80151e8f33b4e56e786c099a514a393d7ddb079861e0ba321 | 382477 |
+| grc-default | GRC | true | org.egothor:radixor-model-grc-default | 1.0.0 | UniMorph | https://github.com/unimorph/grc | 7f4a58df733726c75c1355dd3a038e950d5e308f | 7f4a58df733726c75c1355dd3a038e950d5e308f | recorded | CC-BY-SA-3.0 | UniMorph | 8c50d12e6489acff903ed5a9e5cf463da9e543845d48989255d99609f9a4e2e9 | 93002 |
+| gsw-default | GSW | true | org.egothor:radixor-model-gsw-default | 1.0.0 | UniMorph | https://github.com/unimorph/gsw | dc1e9fce09ae14c2ef22b06ee10da3ab943388fe | dc1e9fce09ae14c2ef22b06ee10da3ab943388fe | recorded | CC-BY-SA-3.0 | UniMorph; Ryan Cotterell | 65ffa4b644e8adef0fe6e484a478a5017bb9c079e015f6a02dd478a4f6c9a871 | 3110 |
+| gup-default | GUP | true | org.egothor:radixor-model-gup-default | 1.0.0 | UniMorph | https://github.com/unimorph/gup | 2d911a4381163e0c7afeb302acf4f4edf82f9880 | 2d911a4381163e0c7afeb302acf4f4edf82f9880 | recorded | CC-BY-SA-3.0 | UniMorph; William Lane | 9a9193fc4b921801a12c91e585a3fb71824c243a0f66138675e5be6ec731772b | 1467 |
+| gv-im-default | GV_IM | true | org.egothor:radixor-model-gv-im-default | 1.0.0 | UniMorph | https://github.com/unimorph/glv | 92565416550aebef179078fd0256777c189afce5 | 92565416550aebef179078fd0256777c189afce5 | recorded | CC-BY-SA-3.0 | UniMorph | da48ff4553af69f5f7dce09951a0d0e93a35ac649034de9d88f4067bac48a1ea | 203 |
+| hai-default | HAI | true | org.egothor:radixor-model-hai-default | 1.0.0 | UniMorph | https://github.com/unimorph/hai | 98b7ddb7534122068f375ef8aa8b1ca8ec57d2a0 | 98b7ddb7534122068f375ef8aa8b1ca8ec57d2a0 | recorded | CC-BY-SA-3.0 | UniMorph | 71a80067a05106b90668da2a78c8f9412d64406036ef13e24de732124fff69a5 | 12994 |
+| hbs-default | HBS | true | org.egothor:radixor-model-hbs-default | 1.0.0 | UniMorph | https://github.com/unimorph/hbs | f41348090bf8ec35e41fee61d7c52d12ae1feff1 | f41348090bf8ec35e41fee61d7c52d12ae1feff1 | recorded | CC-BY-SA-3.0 | UniMorph | be225aef70bd01ee8f988ac5e6a58f3d6e1cddc284a6564dd15f93d4f1634dda | 710473 |
+| he-il-default | HE_IL | true | org.egothor:radixor-model-he-il-default | 1.0.0 | UniMorph | https://github.com/unimorph/heb | not-recorded-in-legacy-import | not-recorded-in-legacy-import | not-recorded-in-legacy-import | CC-BY-SA-3.0 | UniMorph; Omer Goldman | 9a47dc69bb7dab21aba0266b73cd74cdaeb17db94363796a0a56111ac8518256 | 179776 |
+| hil-default | HIL | true | org.egothor:radixor-model-hil-default | 1.0.0 | UniMorph | https://github.com/unimorph/hil | 0ae9f3eca62de21bb948b34b2c03af44dff1f6ca | 0ae9f3eca62de21bb948b34b2c03af44dff1f6ca | recorded | CC-BY-SA-3.0 | UniMorph; Ran Zmigrod | d8114c609f5a38623a2fa3c7c3cf2ba74a2b5d9e78258cf0c428b50afcb9e031 | 1201 |
+| hsi-default | HSI | true | org.egothor:radixor-model-hsi-default | 1.0.0 | UniMorph | https://github.com/unimorph/hsi | 31bd9e0fed419b6257a722364c0d8a4e56f5c373 | 31bd9e0fed419b6257a722364c0d8a4e56f5c373 | recorded | CC-BY-SA-4.0 | UniMorph; Aryaman Arora (Georgetown University) | 82f3518b9ef41c53429248865abdd539aa0436cba9a5f0517f333fe18e4c9146 | 694 |
 | hu-hu-default | HU_HU | true | org.egothor:radixor-model-hu-hu-default | 1.0.0 | UniMorph | https://github.com/unimorph/hun | not-recorded-in-legacy-import | not-recorded-in-legacy-import | not-recorded-in-legacy-import | CC-BY-SA-3.0 | UniMorph; Christo Kirov, Ryan Cotterell, and Khuyagbaatar Batsuren (conversion); Judit Ács and Gábor Bella (validation); English Wiktionary contributors | 359d46a01d751ec823705ad7f3dd1cc8f6663feb1a9d13cb04d0c6fb51ab646e | 2346297 |
+| hy-am-default | HY_AM | true | org.egothor:radixor-model-hy-am-default | 1.0.0 | UniMorph | https://github.com/unimorph/hye | 8d404710b0451dae1576d3aaf77c91e638eff66b | 8d404710b0451dae1576d3aaf77c91e638eff66b | recorded | CC-BY-SA-3.0 | UniMorph | 42eea1564db57d86b2f09fec3d6d19104253cd389769b834c823b1684f21a757 | 683956 |
+| id-id-default | ID_ID | true | org.egothor:radixor-model-id-id-default | 1.0.0 | UniMorph | https://github.com/unimorph/ind | 4663cf3b8127fb85cb7e8a0c1a4f07e96103fe34 | 4663cf3b8127fb85cb7e8a0c1a4f07e96103fe34 | recorded | CC-BY-SA-3.0 | UniMorph; Clara Vania; Totok Suhardijanto; Zahroh Nuriah | ed7ab0190fcd098ff4e871605f61056cf43b54a6967e8c40652584720ddb5044 | 58989 |
+| is-is-default | IS_IS | true | org.egothor:radixor-model-is-is-default | 1.0.0 | UniMorph | https://github.com/unimorph/isl | 71a748303ef3fcc30810f90687a543313563fa1a | 71a748303ef3fcc30810f90687a543313563fa1a | recorded | CC-BY-SA-3.0 | UniMorph | 0d16a717f7c71b5fc333ec8687b1e44d6840092a5e3ce88785830ea9083e9b60 | 139861 |
 | it-it-default | IT_IT | true | org.egothor:radixor-model-it-it-default | 1.0.0 | UniMorph | https://github.com/unimorph/ita | not-recorded-in-legacy-import | not-recorded-in-legacy-import | not-recorded-in-legacy-import | CC-BY-SA-3.0 | UniMorph and Wikipedia contributors | 5e03be31c9761e30dbf24a47a5ced3d6ec949dabd31e92632fdd9f7c67fc2e12 | 841105 |
+| itl-default | ITL | true | org.egothor:radixor-model-itl-default | 1.0.0 | UniMorph | https://github.com/unimorph/itl | 251a11eb5fdcd44d01e498209632c903d2ce3d3e | 251a11eb5fdcd44d01e498209632c903d2ce3d3e | recorded | CC-BY-SA-3.0 | UniMorph; Karina Sheifer, Sofya Ganieva, Matvey Plugaryov | 4664dc2087cf79102555fecdf978c6a2b7e830192f4e7c01e4f7642d167e8175 | 13191 |
+| izh-default | IZH | true | org.egothor:radixor-model-izh-default | 1.0.0 | UniMorph | https://github.com/unimorph/izh | be351d32224a63c6611ffda5034d31982fe49814 | be351d32224a63c6611ffda5034d31982fe49814 | recorded | CC-BY-SA-3.0 | UniMorph | 294754b0e212cd5045fcced35de8e0016a77cb46732d285e2babef0d1e3c5d84 | 2728 |
+| ja-jp-default | JA_JP | true | org.egothor:radixor-model-ja-jp-default | 1.0.0 | UniMorph | https://github.com/unimorph/jpn | ef6d7bf780d177d4b0802568cff45a76056396e5 | ef6d7bf780d177d4b0802568cff45a76056396e5 | recorded | CC-BY-4.0 | UniMorph | 852e1431af196108ac5c79a46e8aa11caead2931dd0960fc4b653cde08fbc161 | 33108 |
+| kbd-default | KBD | true | org.egothor:radixor-model-kbd-default | 1.0.0 | UniMorph | https://github.com/unimorph/kbd | ee843e1110be705a121495bd4656bf1635b2c4d2 | ee843e1110be705a121495bd4656bf1635b2c4d2 | recorded | CC-BY-SA-3.0 | UniMorph | 36208c790afe93152677ba6a4cabb3adfc420b521857ea462d81933e1f38cdb4 | 8835 |
+| kjh-default | KJH | true | org.egothor:radixor-model-kjh-default | 1.0.0 | UniMorph | https://github.com/unimorph/kjh | fbef46743e079bb7ff0138b2497ebf242aa3c4cb | fbef46743e079bb7ff0138b2497ebf242aa3c4cb | recorded | CC-BY-SA-3.0 | UniMorph | 210b01be8a529b3bd8611f82d1050dc9702a41784226c6eab31ceebb77e652e2 | 3570 |
+| kk-kz-default | KK_KZ | true | org.egothor:radixor-model-kk-kz-default | 1.0.0 | UniMorph | https://github.com/unimorph/kaz | 91885d96011368d91397033f4a8f8b0f9c2b60a0 | 91885d96011368d91397033f4a8f8b0f9c2b60a0 | recorded | CC-BY-SA-3.0 | UniMorph | 73216ba98bc3e3c240d626fc47f4dc3ff95148ba086c8a9e8d89bd6cfbdc648b | 104556 |
+| kl-gl-default | KL_GL | true | org.egothor:radixor-model-kl-gl-default | 1.0.0 | UniMorph | https://github.com/unimorph/kal | c1532eb12ca3f13ec351fb541f839defb190aa3a | c1532eb12ca3f13ec351fb541f839defb190aa3a | recorded | CC-BY-SA-3.0 | UniMorph | 19212cb272cf695b77a6e90cb5c6394689cf7ea738830cab29b4ea629865d782 | 980 |
+| klr-default | KLR | true | org.egothor:radixor-model-klr-default | 1.0.0 | UniMorph | https://github.com/unimorph/klr | e9c1cd9e6def47a33759d2f4331dd886bacd5964 | e9c1cd9e6def47a33759d2f4331dd886bacd5964 | recorded | LGPLLR | UniMorph | dba18e7da6b469e0e7d3c0a4b49280a88bb1845c8466f1b6db994c9747a2eb2d | 157818 |
+| kn-in-default | KN_IN | true | org.egothor:radixor-model-kn-in-default | 1.0.0 | UniMorph | https://github.com/unimorph/kan | dc17627bc86d650a885217f84faa3c90179e71fb | dc17627bc86d650a885217f84faa3c90179e71fb | recorded | CC-BY-SA-3.0 | UniMorph | aba11fd379a9d89366d045dc87dcf1e2a1b3d5b1e1425ada894caf54517be265 | 12141 |
+| kod-default | KOD | true | org.egothor:radixor-model-kod-default | 1.0.0 | UniMorph | https://github.com/unimorph/kod | da840ac504da546c7bd84bf330314fe7e481c57d | da840ac504da546c7bd84bf330314fe7e481c57d | recorded | CC-BY-SA-3.0 | UniMorph; Yustinus Ghanggo Ate, Garrett Nicolai | 38c0562b78e339601d44d91cb4cb510039495710fe162d4797df2b973dcc0196 | 1672 |
+| kon-default | KON | true | org.egothor:radixor-model-kon-default | 1.0.0 | UniMorph | https://github.com/unimorph/kon | 4a89718138a97673750689e2e0dfca4376d2a385 | 4a89718138a97673750689e2e0dfca4376d2a385 | recorded | CC-BY-SA-3.0 | UniMorph; Jennifer White | 08f082c093e50f6cc95c9e80fe08a0b2f6e249894159eb1a503d557b4eadc645 | 1304 |
+| krl-default | KRL | true | org.egothor:radixor-model-krl-default | 1.0.0 | UniMorph | https://github.com/unimorph/krl | d72ba9566b70640b415a2cecc95595fe239e17d3 | d72ba9566b70640b415a2cecc95595fe239e17d3 | recorded | CC-BY-SA-3.0 | UniMorph | 8f99c9c57ba8bb642be3c033a973d37e6b1064be186f0b9c1f43a900ad1a1706 | 1578 |
+| ky-kg-default | KY_KG | true | org.egothor:radixor-model-ky-kg-default | 1.0.0 | UniMorph | https://github.com/unimorph/kir | 3eba9b72cc8222bf964a06842e4ae31c45ce21bf | 3eba9b72cc8222bf964a06842e4ae31c45ce21bf | recorded | CC-BY-SA-3.0 | UniMorph; Eleanor Chodroff | 3b7d57494e647317afd9075559f7727c38d9d98041f968e6f0237c649b63ebed | 9243 |
+| la-default | LA | true | org.egothor:radixor-model-la-default | 1.0.0 | UniMorph | https://github.com/unimorph/lat | 050924c9f013e6aba4d7c87be243a1d79b6c5b10 | 050924c9f013e6aba4d7c87be243a1d79b6c5b10 | recorded | CC-BY-SA-3.0 | UniMorph | 1ecc8889f0cf91f5005f8c1434dd2d12b03c34ec11f16336896544ef676709d1 | 1291334 |
+| lg-ug-default | LG_UG | true | org.egothor:radixor-model-lg-ug-default | 1.0.0 | UniMorph | https://github.com/unimorph/lug | 5bb5dc567e071e2e3111d5eb29bf438244a9b0e1 | 5bb5dc567e071e2e3111d5eb29bf438244a9b0e1 | recorded | CC-BY-SA-3.0 | UniMorph; Edoardo Ponti | bdf75bc28bc51c41d1b2d4eab15988a165824a0859b737c9eb25e5f03016108f | 11105 |
+| lin-default | LIN | true | org.egothor:radixor-model-lin-default | 1.0.0 | UniMorph | https://github.com/unimorph/lin | d56576af7cfc2ea79343e9a12ae1be82450b4e0a | d56576af7cfc2ea79343e9a12ae1be82450b4e0a | recorded | CC-BY-SA-3.0 | UniMorph | d784b140022b3cdff26dcb83d1ff9132eb4db12a6f6b5bb5e1d0bae53f956909 | 731 |
+| liv-default | LIV | true | org.egothor:radixor-model-liv-default | 1.0.0 | UniMorph | https://github.com/unimorph/liv | 1fc0245ac28f62daf10cd7c7daa5b0fdb8465c56 | 1fc0245ac28f62daf10cd7c7daa5b0fdb8465c56 | recorded | CC-BY-SA-3.0 | UniMorph | 3f8f654cbcd4b2ada1db6be64c7eeed8d8ea4c09a961de34fe907a1d01e22a7f | 7874 |
+| lld-default | LLD | true | org.egothor:radixor-model-lld-default | 1.0.0 | UniMorph | https://github.com/unimorph/lld | 8366d54e483e5334d8d761b68826b79a9a9333c2 | 8366d54e483e5334d8d761b68826b79a9a9333c2 | recorded | CC-BY-SA-3.0 | UniMorph | 014be5ac5327c64f68287595d371735da93a3f36d3ae22e8209b3db4f8331ab1 | 11694 |
+| lt-lt-default | LT_LT | true | org.egothor:radixor-model-lt-lt-default | 1.0.0 | UniMorph | https://github.com/unimorph/lit | 8367e0ab8cac32555e2f040da468bbfd7d476586 | 8367e0ab8cac32555e2f040da468bbfd7d476586 | recorded | CC-BY-SA-3.0 | UniMorph | c7b8536235bc7de9d101393ec41a829a4b750b4c45d3ab7324d8ef974c277d30 | 76555 |
+| lv-lv-default | LV_LV | true | org.egothor:radixor-model-lv-lv-default | 1.0.0 | UniMorph | https://github.com/unimorph/lav | 0103e8ae2f0921378763c7d7875765a3ca3dc8df | 0103e8ae2f0921378763c7d7875765a3ca3dc8df | recorded | CC-BY-SA-3.0 | UniMorph | 6ad88bdd8283f4fbc98093a5dfc08fe69b4f9baf9bb0f8cf156edeaa114216f3 | 193067 |
+| mag-default | MAG | true | org.egothor:radixor-model-mag-default | 1.0.0 | UniMorph | https://github.com/unimorph/mag | c7570a85f4ef711885971231207890c1d94c4bf0 | c7570a85f4ef711885971231207890c1d94c4bf0 | recorded | CC-BY-SA-3.0 | UniMorph; Mohit Raj, Ritesh Kumar | a3cf63802ff531265bd5867123cfe0dc305ec57aea29f2d8654f7c19f9df1c85 | 5653 |
+| mg-mg-default | MG_MG | true | org.egothor:radixor-model-mg-mg-default | 1.0.0 | UniMorph | https://github.com/unimorph/mlg | 5d543a609b86ae149e4be295af68fdf4f61a8f27 | 5d543a609b86ae149e4be295af68fdf4f61a8f27 | recorded | CC-BY-SA-3.0 | UniMorph; Jennifer White | f45cb5edae2859c0f6730edf4e0614be0fb6256d7e3e22dfbd97a12cb559133c | 1787 |
+| mi-nz-default | MI_NZ | true | org.egothor:radixor-model-mi-nz-default | 1.0.0 | UniMorph | https://github.com/unimorph/mao | d12063aa44d4d38a5283729c6c0247cb50cb56d5 | d12063aa44d4d38a5283729c6c0247cb50cb56d5 | recorded | CC-BY-SA-3.0 | UniMorph; Jennifer White | 0de22749a54aa35d4a788fe9b07a3e7e2cb1ce6774258eb69475b91c97666c1f | 748 |
+| mk-mk-default | MK_MK | true | org.egothor:radixor-model-mk-mk-default | 1.0.0 | UniMorph | https://github.com/unimorph/mkd | d7a3968a6c70041ccbbddf47a4651ddfd75e4109 | d7a3968a6c70041ccbbddf47a4651ddfd75e4109 | recorded | CC-BY-SA-3.0 | UniMorph | 16f138c9b020209290f4bd88aa16a1b08f766e793b8f4f7a51078ff18c29f875 | 397508 |
+| mn-mn-default | MN_MN | true | org.egothor:radixor-model-mn-mn-default | 1.0.0 | UniMorph | https://github.com/unimorph/khk | c062afd9bf2dc36f54cc8f16efecfd97fc4a6b8c | c062afd9bf2dc36f54cc8f16efecfd97fc4a6b8c | recorded | CC-BY-SA-3.0 | UniMorph; Khuyagbaatar Batsuren | e178aca6228449ecc967e6a00e3ee209ba9e1bc4bf43935ff29bb63a49fa18e3 | 54457 |
+| mt-mt-default | MT_MT | true | org.egothor:radixor-model-mt-mt-default | 1.0.0 | UniMorph | https://github.com/unimorph/mlt | 9f60b87c1d391b5316caa6d868eb118341d06043 | 9f60b87c1d391b5316caa6d868eb118341d06043 | recorded | CC-BY-SA-3.0 | UniMorph | 3be6ab250e1f9489c0b28d24a5fae8a43da6ddb2528c9d74281dac6a620618b6 | 3886 |
+| mwf-default | MWF | true | org.egothor:radixor-model-mwf-default | 1.0.0 | UniMorph | https://github.com/unimorph/mwf | 7ae9b2386c21275056cae0eb48916cc38154dadd | 7ae9b2386c21275056cae0eb48916cc38154dadd | recorded | CC-BY-SA-3.0 | UniMorph | cc20412e152352ea121257c724b63f9a4dbc575beb90bfa3b10d664c32917f69 | 1609 |
+| nap-default | NAP | true | org.egothor:radixor-model-nap-default | 1.0.0 | UniMorph | https://github.com/unimorph/nap | d0d1edda9214ae3543272b82471a568c6df9dad5 | d0d1edda9214ae3543272b82471a568c6df9dad5 | recorded | CC-BY-SA-3.0 | UniMorph | 7caaaac7ed1ceedf5f8eca36a1ded21371e7003189fdb4b43a215072f600dfdc | 3539 |
+| nav-default | NAV | true | org.egothor:radixor-model-nav-default | 1.0.0 | UniMorph | https://github.com/unimorph/nav | 7ca00e86d03bbd3aaedd5291c53b3813ef521832 | 7ca00e86d03bbd3aaedd5291c53b3813ef521832 | recorded | CC-BY-SA-3.0 | UniMorph | 0015e0e7d39918770a8f94ac21469cf0f3cb0f9d62aa8e56b36763af80e36735 | 33166 |
 | nb-no-default | NB_NO | true | org.egothor:radixor-model-nb-no-default | 1.0.0 | UniMorph | https://github.com/unimorph/nob | not-recorded-in-legacy-import | not-recorded-in-legacy-import | not-recorded-in-legacy-import | CC-BY-SA-3.0 | UniMorph and Wikipedia contributors | f495bffb44e79d27993e6e2e65d4b1204b29365dc93f481b2d8b96766fc90fd9 | 205645 |
+| nds-default | NDS | true | org.egothor:radixor-model-nds-default | 1.0.0 | UniMorph | https://github.com/unimorph/nds | 531ffb061f0462f1c42d09001ba2caf8227d118d | 531ffb061f0462f1c42d09001ba2caf8227d118d | recorded | CC-BY-SA-3.0 | UniMorph | 262c5a5aa1ab0bcb2145e43968b1cffeace44a614aa9cab0496c4fdc5b1f1a14 | 6391 |
 | nl-nl-default | NL_NL | true | org.egothor:radixor-model-nl-nl-default | 1.0.0 | UniMorph | https://github.com/unimorph/nld | not-recorded-in-legacy-import | not-recorded-in-legacy-import | not-recorded-in-legacy-import | CC-BY-SA-3.0 | UniMorph and Wikipedia contributors | c098034adc42da2ca3e419160e6dd2c2b3868f8af334303b3a191e09caadaf5e | 70600 |
 | nn-no-default | NN_NO | true | org.egothor:radixor-model-nn-no-default | 1.0.0 | UniMorph | https://github.com/unimorph/nno | not-recorded-in-legacy-import | not-recorded-in-legacy-import | not-recorded-in-legacy-import | CC-BY-SA-3.0 | UniMorph and Wikipedia contributors | 900cf2005605aea2a3d8d731ec0b0c1f47fb4469b4ba6b9134145d4d026a0398 | 51244 |
+| non-default | NON | true | org.egothor:radixor-model-non-default | 1.0.0 | UniMorph | https://github.com/unimorph/non | d23b7967e3f1aedb846c7a356420141b2f56e1bf | d23b7967e3f1aedb846c7a356420141b2f56e1bf | recorded | CC-BY-SA-3.0 | UniMorph | d231245493a2d59d541247546133e102e021661ddcd9d3ef486b76a0900bf7b8 | 115713 |
+| ny-mw-default | NY_MW | true | org.egothor:radixor-model-ny-mw-default | 1.0.0 | UniMorph | https://github.com/unimorph/nya | 4f7699bfac803b3422e79bac1f317352547b8c3b | 4f7699bfac803b3422e79bac1f317352547b8c3b | recorded | CC-BY-SA-3.0 | UniMorph; Ryan Cotterell | 7d7487a8ed6db1d058f5f6871d35cece850205423e77134553dae767927fba99 | 7857 |
+| ood-default | OOD | true | org.egothor:radixor-model-ood-default | 1.0.0 | UniMorph | https://github.com/unimorph/ood | e6d1af34d2a7ef22e713af6629fd74dfb80b5a80 | e6d1af34d2a7ef22e713af6629fd74dfb80b5a80 | recorded | CC-BY-SA-3.0 | UniMorph; Eleanor Chodroff | e4bc4feb122a2bab40e981b3916f260cc13e4a2750796cf88d9bb9eba155448e | 3608 |
+| osx-default | OSX | true | org.egothor:radixor-model-osx-default | 1.0.0 | UniMorph | https://github.com/unimorph/osx | 5cd3166c5ff5e3cb1c40911003c259f47467f8c1 | 5cd3166c5ff5e3cb1c40911003c259f47467f8c1 | recorded | CC-BY-SA-3.0 | UniMorph | 1f6b3cc29edb4b6c25872133805f212f8fcdc72b139ba96fee8b3715de729a01 | 31188 |
+| ote-default | OTE | true | org.egothor:radixor-model-ote-default | 1.0.0 | UniMorph | https://github.com/unimorph/ote | aade58b0be48c21adda52ef1d01dd8761050b2dd | aade58b0be48c21adda52ef1d01dd8761050b2dd | recorded | CC-BY-SA-3.0 | UniMorph; Antonis Anastasopoulos | c7af96b64a51275622783265da6eb8ed30cfc5a1e60a9aad1ddced08874d7e4d | 11695 |
 | pl-pl-polimorf | PL_PL | false | org.egothor:radixor-model-pl-pl-polimorf | 1.0.0 | PoliMorf 2.1 | https://github.com/morfologik/morfologik-stemming | 2.1 | 6e63b53 | recorded | BSD-2-Clause | Copyright (c) 2016, Marcin Miłkowski | 4fe4bf5e6c22c1beea5b3d57f1ce4c9ea5aac1ed8ab24c616fb06df745e40d15 | 12624997 |
 | pl-pl-unimorph | PL_PL | true | org.egothor:radixor-model-pl-pl-unimorph | 1.0.0 | UniMorph | https://github.com/unimorph/pol | not-recorded-in-legacy-import | not-recorded-in-legacy-import | not-recorded-in-legacy-import | CC-BY-SA-3.0 | UniMorph; SGJP authors Marcin Woliński, Zygmunt Saloni, Robert Wołosz, Włodzimierz Gruszczyński, Danuta Skowrońska, and Zbigniew Bronk; Witold Kieraś (conversion); Wiktionary contributors | 8191ed727097839cc808cbc5c56a1bd78b3c851e7733ad226ad9a51519a54721 | 334951 |
+| ps-af-default | PS_AF | true | org.egothor:radixor-model-ps-af-default | 1.0.0 | UniMorph | https://github.com/unimorph/pus | 94a599739a3f886d9a3639ad45368fd838e8d5b2 | 94a599739a3f886d9a3639ad45368fd838e8d5b2 | recorded | CC-BY-SA-3.0 | UniMorph | 104ddc76f34a17c4f16719280c5fe6e11d7c0a6cea2c79f99226bdb667aebbc5 | 8496 |
 | pt-pt-default | PT_PT | true | org.egothor:radixor-model-pt-pt-default | 1.0.0 | UniMorph | https://github.com/unimorph/por | not-recorded-in-legacy-import | not-recorded-in-legacy-import | not-recorded-in-legacy-import | CC-BY-SA-3.0 | UniMorph and Wikipedia contributors | 7a035ff330a6f0548f446cd0d6617bc1cf4751292125a3564d3a255c5d6f516d | 509965 |
+| que-default | QUE | true | org.egothor:radixor-model-que-default | 1.0.0 | UniMorph | https://github.com/unimorph/que | 78df51d731ec3d70436ff9024a5534a1eeae2553 | 78df51d731ec3d70436ff9024a5534a1eeae2553 | recorded | CC-BY-SA-3.0 | UniMorph | 2a9eb15b84f19ccfbfeb0b323fb93267109f49968914e5723f155f72db1188b8 | 369663 |
+| ro-ro-default | RO_RO | true | org.egothor:radixor-model-ro-ro-default | 1.0.0 | UniMorph | https://github.com/unimorph/ron | 0910e7850baef37f67817060fa8aae7c42cbe269 | 0910e7850baef37f67817060fa8aae7c42cbe269 | recorded | CC-BY-SA-3.0 | UniMorph | 94637c539437fd719436ded52fdef0a5271fc0b91e306efcd7e4fa8deadf6e79 | 128963 |
 | ru-ru-default | RU_RU | true | org.egothor:radixor-model-ru-ru-default | 1.0.0 | UniMorph | https://github.com/unimorph/rus | not-recorded-in-legacy-import | not-recorded-in-legacy-import | not-recorded-in-legacy-import | CC-BY-SA-3.0 | UniMorph and Wikipedia contributors | df7ea25e63a875eeec7a4185be685bd5372a3c568db85c34c44fdf5d8d980a40 | 2414507 |
+| sdh-default | SDH | true | org.egothor:radixor-model-sdh-default | 1.0.0 | UniMorph | https://github.com/unimorph/sdh | 99f7edcace30616a591a79ce01959218e02bb849 | 99f7edcace30616a591a79ce01959218e02bb849 | recorded | CC-BY-SA-3.0 | UniMorph; Ali Salehi | e50b3b8a342c09a07b5df99de263246617724188cb7176420ebaf34cab1ea89b | 530 |
+| see-default | SEE | true | org.egothor:radixor-model-see-default | 1.0.0 | UniMorph | https://github.com/unimorph/see | 8d182ef3039f9e7c497eaa956086f7a0fb7f361e | 8d182ef3039f9e7c497eaa956086f7a0fb7f361e | recorded | CC-BY-SA-3.0 | unavailable | ba81e101e0c2f5c2ef6762a3bcae3e9af95b79eedc7dd9a2bacc9216ea70ed65 | 14806 |
+| sga-default | SGA | true | org.egothor:radixor-model-sga-default | 1.0.0 | UniMorph | https://github.com/unimorph/sga | de9858d9966879b78d44fcded5287858146a6e16 | de9858d9966879b78d44fcded5287858146a6e16 | recorded | CC-BY-SA-3.0 | UniMorph | d5a1f169cfb32c134e287b98a652ffce1a43f71f856a3e875a60475ba0d0989b | 2848 |
+| shp-default | SHP | true | org.egothor:radixor-model-shp-default | 1.0.0 | UniMorph | https://github.com/unimorph/shp | 4471a3e24c17d830ffd55318eceb3e7e76e77beb | 4471a3e24c17d830ffd55318eceb3e7e76e77beb | recorded | CC-BY-SA-3.0 | UniMorph; Arturo Oncevay, Candy Milagros Angulo Pando, Roberto Zariquiey | 213ad85c4a6b892750c90ed6bdbead05a64e56dd2091ad203a47d2b84198bc9b | 21371 |
+| sjo-default | SJO | true | org.egothor:radixor-model-sjo-default | 1.0.0 | UniMorph | https://github.com/unimorph/sjo | cd7ff346bffe85c134167eec0130e282acb6442e | cd7ff346bffe85c134167eec0130e282acb6442e | recorded | CC-BY-SA-3.0 | UniMorph; Elena Klyachko | ca640317c3b84c98e3d296fe6f2dba266181087135475d1320ba2980c66cb3fc | 11777 |
+| sme-default | SME | true | org.egothor:radixor-model-sme-default | 1.0.0 | UniMorph | https://github.com/unimorph/sme | dcb1abed05e25b2ce85f3a806c270e2b192a6dfe | dcb1abed05e25b2ce85f3a806c270e2b192a6dfe | recorded | CC-BY-SA-3.0 | UniMorph | d42d30a9b27e1aae3122ee0c5367f60ebe7afe9cf2bca2472afcae959af972f9 | 147750 |
+| sn-zw-default | SN_ZW | true | org.egothor:radixor-model-sn-zw-default | 1.0.0 | UniMorph | https://github.com/unimorph/sna | 76ec498cca0713125206014eb453224ab761a71f | 76ec498cca0713125206014eb453224ab761a71f | recorded | CC-BY-SA-3.0 | UniMorph; Rowan Hall Maudslay | e0f2e68f72fadf364d9e93fd1210566044981769f3d6f749504402dc7db1b06d | 6422 |
+| sq-al-default | SQ_AL | true | org.egothor:radixor-model-sq-al-default | 1.0.0 | UniMorph | https://github.com/unimorph/sqi | 63fed8fdd510d14a75d2d2dacb1c5ab1475617e7 | 63fed8fdd510d14a75d2d2dacb1c5ab1475617e7 | recorded | CC-BY-SA-3.0 | UniMorph | f36caf9057f6eede49212d2ea0612d8f5db850fcbfc08f7f921573507bf938be | 25808 |
+| st-za-default | ST_ZA | true | org.egothor:radixor-model-st-za-default | 1.0.0 | UniMorph | https://github.com/unimorph/sot | e523f2c896a0de10f0ef19c67e8239f9c7c04cfd | e523f2c896a0de10f0ef19c67e8239f9c7c04cfd | recorded | CC-BY-SA-3.0 | UniMorph | 79cf6bd61d12a2115418e35a185ae6b66a4f99f678fe3690d774ced6be604eb5 | 1091 |
 | sv-se-default | SV_SE | true | org.egothor:radixor-model-sv-se-default | 1.0.0 | UniMorph | https://github.com/unimorph/swe | not-recorded-in-legacy-import | not-recorded-in-legacy-import | not-recorded-in-legacy-import | CC-BY-SA-3.0 | UniMorph and English Wiktionary contributors | d9be72e3d67c776622c4281e04e4063b9381e8f84a823d98ebf08888b82dff0c | 256300 |
+| swc-default | SWC | true | org.egothor:radixor-model-swc-default | 1.0.0 | UniMorph | https://github.com/unimorph/swc | 02a2bdec0e5cb0dc93b6ef11db4d54a82c34b224 | 02a2bdec0e5cb0dc93b6ef11db4d54a82c34b224 | recorded | CC-BY-SA-3.0 | UniMorph; Jennifer White | a00d480708771bd7c70fcc82e013255444624c4b7a20dc95ea346a8a584c561a | 26652 |
+| syc-default | SYC | true | org.egothor:radixor-model-syc-default | 1.0.0 | UniMorph | https://github.com/unimorph/syc | 2e94e69b8c20f57e2593b70ff4c73fe3d5a2066e | 2e94e69b8c20f57e2593b70ff4c73fe3d5a2066e | recorded | CC-BY-SA-3.0 | UniMorph | 62bd6eaf022709cd4188c99bd491301c31dbcce5075877981b3a5ea421366a8f | 95208 |
+| tl-ph-default | TL_PH | true | org.egothor:radixor-model-tl-ph-default | 1.0.0 | UniMorph | https://github.com/unimorph/tgl | 799d79de338d1370ebfbeabb9823a0e8f86a2b6a | 799d79de338d1370ebfbeabb9823a0e8f86a2b6a | recorded | CC-BY-SA-3.0 | UniMorph; Jennifer White; `tgl`: contains some archaic forms | 7ff4fcdf97eb7e2f432480d2bfa4cceaa54acb7e2a1426737c7911c972f429ed | 6754 |
+| tr-tr-default | TR_TR | true | org.egothor:radixor-model-tr-tr-default | 1.0.0 | UniMorph | https://github.com/unimorph/tur | 6c179ace7d2f3d7f3484020e5304c1544d07bb6b | 6c179ace7d2f3d7f3484020e5304c1544d07bb6b | recorded | CC-BY-SA-3.0 | UniMorph | a3605d0abd12c4bd077c72bdc0be52153400601dbd985897cb5c4ec424330200 | 567241 |
+| ug-cn-default | UG_CN | true | org.egothor:radixor-model-ug-cn-default | 1.0.0 | UniMorph | https://github.com/unimorph/uig | 0786d110b042c9e64599f53a6a0ac80889eb7f31 | 0786d110b042c9e64599f53a6a0ac80889eb7f31 | recorded | CC-BY-SA-3.0 | UniMorph; Eleanor Chodroff | 26d181f6d1114ed9ba838608a77d1e7849086e3ad90bc1f2c246e4b4034e63ad | 17583 |
 | uk-ua-default | UK_UA | true | org.egothor:radixor-model-uk-ua-default | 1.0.0 | UniMorph | https://github.com/unimorph/ukr | not-recorded-in-legacy-import | not-recorded-in-legacy-import | not-recorded-in-legacy-import | CC-BY-SA-3.0 | UniMorph; Witold Kieraś and Maria Shvedova are credited for the separate VESUM conversion; Wiktionary contributors | cf3f612cfff16cb7763f99c55851069489b883c3bdd1a6576cd8c57a97e07eae | 47300 |
 | us-uk-default | US_UK | true | org.egothor:radixor-model-us-uk-default | 1.0.1 | UniMorph | https://github.com/unimorph/eng | not-recorded-in-legacy-import | not-recorded-in-legacy-import | not-recorded-in-legacy-import | CC-BY-SA-3.0 | UniMorph and Wikipedia contributors | 0a3ec1fddd5ddb859f5ffa8003efe9200a5ef6203df1fddaf53a58003f33582c | 2708332 |
+| uz-uz-default | UZ_UZ | true | org.egothor:radixor-model-uz-uz-default | 1.0.0 | UniMorph | https://github.com/unimorph/uzb | 0d06920a2aaf8dfb58b4a5492355e8f6425d2e51 | 0d06920a2aaf8dfb58b4a5492355e8f6425d2e51 | recorded | CC-BY-SA-3.0 | UniMorph; Eleanor Chodroff; `uzb_verbs`: also checked by native speakers | caf2d4038429a72be272b400d611f80ac37fd1b0f3c54ea44c7a1159ea42728f | 25418 |
+| vro-default | VRO | true | org.egothor:radixor-model-vro-default | 1.0.0 | UniMorph | https://github.com/unimorph/vro | 51ce9786332835a8b045c9089260e5ba561ab03c | 51ce9786332835a8b045c9089260e5ba561ab03c | recorded | CC-BY-SA-3.0 | UniMorph; Ekaterina Vylomova | 5da7535bd5e629b8ffb99a3b6d906ad0189b1c0d46c6deada72d72f0791f575e | 1338 |
+| xcl-default | XCL | true | org.egothor:radixor-model-xcl-default | 1.0.0 | UniMorph | https://github.com/unimorph/xcl | 847afc046ac9e56294b878cc0ec3ccbdb023c0f7 | 847afc046ac9e56294b878cc0ec3ccbdb023c0f7 | recorded | CC-BY-SA-3.0 | UniMorph | 7a4000404b5f55b67dea02bf5d3e84972979cac3138fb83458b5a6dcf3495864 | 170740 |
+| xno-default | XNO | true | org.egothor:radixor-model-xno-default | 1.0.0 | UniMorph | https://github.com/unimorph/xno | 110c24094bcacd3053f5702fd09f6aa3221eda1e | 110c24094bcacd3053f5702fd09f6aa3221eda1e | recorded | CC-BY-SA-3.0 | UniMorph | 90941356e0bd87195290432052ed9a85011f8e7c1607e24655ba298c519035da | 636 |
+| xty-default | XTY | true | org.egothor:radixor-model-xty-default | 1.0.0 | UniMorph | https://github.com/unimorph/xty | d4b2aad2a976e7f307b7b715cfc5812616691886 | d4b2aad2a976e7f307b7b715cfc5812616691886 | recorded | CC-BY-SA-3.0 | UniMorph; Antonis Anastasopoulos | 9985fe69992665a30309475c30f6734cb24393b21e1c3628543e4810775e711c | 8987 |
 | yi-default | YI | true | org.egothor:radixor-model-yi-default | 1.0.0 | UniMorph | https://github.com/unimorph/yid | not-recorded-in-legacy-import | not-recorded-in-legacy-import | not-recorded-in-legacy-import | CC-BY-SA-3.0 | UniMorph | f47de665c27dcd72833a82904e49c68a945bb5aca769a7ec5a0164e2c981a6d3 | 12222 |
+| zpv-default | ZPV | true | org.egothor:radixor-model-zpv-default | 1.0.0 | UniMorph | https://github.com/unimorph/zpv | d6ff7122f0a457062293890f4befe7f22d227ab7 | d6ff7122f0a457062293890f4befe7f22d227ab7 | recorded | CC-BY-SA-3.0 | UniMorph; Antonis Anastasopoulos | e3cae35dca9b29226dbbbfe49415e4cb80f8379228edbd657f9bfe7c25eb0998 | 2375 |
+| zu-za-default | ZU_ZA | true | org.egothor:radixor-model-zu-za-default | 1.0.0 | UniMorph | https://github.com/unimorph/zul | cc7adc828d0ee63b282a44a105b11689ec5951da | cc7adc828d0ee63b282a44a105b11689ec5951da | recorded | CC-BY-SA-3.0 | UniMorph | 6200464bc24c7519e1a1c42b2c292375fdd9e07bdd66ef994d925d466d636bc9 | 74042 |
diff --git a/docs/stemmer-models.md b/docs/stemmer-models.md
index 21138d4..c621761 100644
--- a/docs/stemmer-models.md
+++ b/docs/stemmer-models.md
@@ -40,10 +40,11 @@ org/egothor/stemmer/models/pl-pl-polimorf/stemmer.gz
 
 Each UniMorph-derived model instead contains one model-specific
 `META-INF/NOTICE/<model-id>-data.txt`. That notice records the upstream attribution, the Radixor
-transformations and contribution statement, the ShareAlike distribution terms, and the canonical
-CC BY-SA 3.0 URI. The repository has no root CC license directory because CC BY-SA applies to
-these model-data artifacts, not to the BSD-3-Clause Radixor Java software. PoliMorf retains only
-its BSD-2-Clause data license.
+transformations and contribution statement, the applicable distribution terms, and the canonical
+license URI. Most use CC BY-SA 3.0; Hsilimo uses CC BY-SA 4.0, Japanese uses CC BY 4.0, and
+Khaling uses LGPLLR with its complete license text and legible dictionary form packaged. The
+repository has no root model-data license because Radixor Java software remains BSD-3-Clause.
+PoliMorf retains only its BSD-2-Clause data license.
 
 `models.index` contains the descriptor path. The descriptor contains the exact resource path. No Java provider class is required, and model modules do not compile against a core API.
 
@@ -133,7 +134,7 @@ The final runtime resource is exactly:
 
 | Project | Published coordinate | Contents and purpose |
 |---|---|---|
-| `models/standard` | `org.egothor:radixor-models-standard:<catalog-version>` | POM-only aggregate with one transitive runtime default per language; excludes PoliMorf |
+| `models/standard` | `org.egothor:radixor-models-standard:<catalog-version>` | POM-only aggregate of the original 20 `default` entries; excludes `standalone` models and PoliMorf |
 | `models/bom` | `org.egothor:radixor-models-bom:<catalog-version>` | POM-only Maven dependency-management constraints for all individual published model versions |
 
 Neither catalog artifact publishes a binary, sources, or Javadoc JAR. The standard aggregate resolves model JARs because its POM contains runtime dependencies. Importing the BOM only manages versions and resolves no model by itself. JMH, tests, and quality evaluation depend directly on individual model projects through non-production Gradle configurations.
@@ -155,7 +156,12 @@ loading procedure, and promotion boundary are documented on the
 3. Apply `org.egothor.radixor.model` in the module build script.
 4. Declare `modelId`, `language`, `displayName`, `defaultModel`, repository, dataset, revision and status, license URI, attribution, verification date, and transformations.
 5. Put immutable `stemmer.gz` and a model-specific `NOTICE-model-data.txt` under `src/modelInput/`. The notice must identify the applicable data license and canonical URI, upstream attribution, transformations, and derived-data contributions without implying that the core software uses that license.
-6. Add the module ID and its `default` or `optional` build-topology role to `models/model-projects.properties`. `settings.gradle`, verification, standard membership, BOM constraints, tests, and JMH all consume that list; descriptor metadata remains authoritative for model identity and language properties.
+6. Add the module ID and its `default`, `standalone`, or `optional`
+   build-topology role to `models/model-projects.properties`. All roles feed
+   settings, verification, BOM constraints, the catalog, individual
+   publication, tests, and JMH. Only `default` feeds the Java and Python
+   standard aggregates. Descriptor metadata remains authoritative for model
+   identity and language properties.
 7. Run:
 
 ```bash
diff --git a/mkdocs.yml b/mkdocs.yml
index 4053212..f961e9a 100644
--- a/mkdocs.yml
+++ b/mkdocs.yml
@@ -70,26 +70,149 @@ nav:
      - Python Runtime Performance: python/performance.md
      - Language Comparisons:
         - Overview: benchmarks/languages/index.md
+        - Adyghe: benchmarks/languages/ady.md
+        - Afrikaans: benchmarks/languages/af-za.md
+        - Aimele: benchmarks/languages/ail.md
+        - Akan: benchmarks/languages/aka.md
+        - Albanian: benchmarks/languages/sq-al.md
+        - Alsatian: benchmarks/languages/gsw.md
+        - Amharic: benchmarks/languages/am-et.md
+        - Amol Azerbaijani: benchmarks/languages/azg.md
+        - Ancient Greek: benchmarks/languages/grc.md
+        - Anglo-Norman: benchmarks/languages/xno.md
+        - Arabic: benchmarks/languages/ar.md
+        - Armenian: benchmarks/languages/hy-am.md
+        - Ashaninka: benchmarks/languages/cni.md
+        - Assamese: benchmarks/languages/as-in.md
+        - Asturian: benchmarks/languages/ast.md
+        - Aymara: benchmarks/languages/aym.md
+        - Azerbaijani: benchmarks/languages/az-az.md
+        - Bashkir: benchmarks/languages/bak.md
+        - Belarusian: benchmarks/languages/be-by.md
+        - Bengali: benchmarks/languages/bn-bd.md
+        - Bininj Kun-wok: benchmarks/languages/gup.md
+        - Braj: benchmarks/languages/bra.md
+        - Breton: benchmarks/languages/bre.md
+        - Bulgarian: benchmarks/languages/bg-bg.md
+        - Catalan: benchmarks/languages/ca-es.md
+        - Cebuano: benchmarks/languages/ceb.md
+        - Chichewa: benchmarks/languages/ny-mw.md
+        - Chukchi: benchmarks/languages/ckt.md
+        - Church Slavonic: benchmarks/languages/chu.md
+        - Classical Armenian: benchmarks/languages/xcl.md
+        - Classical Syriac: benchmarks/languages/syc.md
+        - Congo Swahili: benchmarks/languages/swc.md
+        - Copala Triqui: benchmarks/languages/cpa.md
+        - Cornish: benchmarks/languages/cor.md
+        - Cree: benchmarks/languages/cre.md
+        - Crimean Tatar: benchmarks/languages/crh.md
         - Czech: benchmarks/languages/czech.md
+        - Dakota: benchmarks/languages/dak.md
         - Danish: benchmarks/languages/danish.md
         - Dutch: benchmarks/languages/dutch.md
+        - Eastern Chatino: benchmarks/languages/cly.md
+        - Egyptian Arabic: benchmarks/languages/arz.md
         - English: benchmarks/languages/english.md
+        - Estonian: benchmarks/languages/et-ee.md
+        - Evenki: benchmarks/languages/evn.md
+        - Faroese: benchmarks/languages/fo-fo.md
         - Finnish: benchmarks/languages/finnish.md
         - French: benchmarks/languages/french.md
+        - Friulian: benchmarks/languages/fur.md
+        - Ga: benchmarks/languages/gaa.md
+        - Galolen: benchmarks/languages/gal.md
         - German: benchmarks/languages/german.md
+        - Gothic: benchmarks/languages/got.md
+        - Greek: benchmarks/languages/el-gr.md
+        - Gulf Arabic: benchmarks/languages/afb.md
+        - Haida: benchmarks/languages/hai.md
         - Hebrew: benchmarks/languages/hebrew.md
+        - Hiligaynon: benchmarks/languages/hil.md
+        - Hsilimo: benchmarks/languages/hsi.md
         - Hungarian: benchmarks/languages/hungarian.md
+        - Icelandic: benchmarks/languages/is-is.md
+        - Indonesian: benchmarks/languages/id-id.md
+        - Ingrian: benchmarks/languages/izh.md
+        - Irish: benchmarks/languages/ga-ie.md
         - Italian: benchmarks/languages/italian.md
-        - Norwegian Bokmal: benchmarks/languages/norwegian-bokmal.md
+        - Itelmen: benchmarks/languages/itl.md
+        - Japanese: benchmarks/languages/ja-jp.md
+        - Kabardian: benchmarks/languages/kbd.md
+        - Kalaallisut: benchmarks/languages/kl-gl.md
+        - Kannada: benchmarks/languages/kn-in.md
+        - Karelian: benchmarks/languages/krl.md
+        - Kashubian: benchmarks/languages/csb.md
+        - Kazakh: benchmarks/languages/kk-kz.md
+        - Khakas: benchmarks/languages/kjh.md
+        - Khinalugh: benchmarks/languages/klr.md
+        - Kodi: benchmarks/languages/kod.md
+        - Kongo: benchmarks/languages/kon.md
+        - Kyrgyz: benchmarks/languages/ky-kg.md
+        - Ladin: benchmarks/languages/lld.md
+        - Latin: benchmarks/languages/la.md
+        - Latvian: benchmarks/languages/lv-lv.md
+        - Lingala: benchmarks/languages/lin.md
+        - Lithuanian: benchmarks/languages/lt-lt.md
+        - Livonian: benchmarks/languages/liv.md
+        - Low German: benchmarks/languages/nds.md
+        - Lower Sorbian: benchmarks/languages/dsb.md
+        - Luganda: benchmarks/languages/lg-ug.md
+        - Macedonian: benchmarks/languages/mk-mk.md
+        - Magahi: benchmarks/languages/mag.md
+        - Malagasy: benchmarks/languages/mg-mg.md
+        - Maltese: benchmarks/languages/mt-mt.md
+        - Manx: benchmarks/languages/gv-im.md
+        - Maori: benchmarks/languages/mi-nz.md
+        - Mapudungun: benchmarks/languages/arn.md
+        - Middle French: benchmarks/languages/frm.md
+        - Middle High German: benchmarks/languages/gmh.md
+        - Middle Low German: benchmarks/languages/gml.md
+        - Mongolian: benchmarks/languages/mn-mn.md
+        - Murrinh-Patha: benchmarks/languages/mwf.md
+        - Navajo: benchmarks/languages/nav.md
+        - Neapolitan: benchmarks/languages/nap.md
+        - North Frisian: benchmarks/languages/frr.md
+        - Northern Sami: benchmarks/languages/sme.md
+        - Norwegian Bokmål: benchmarks/languages/norwegian-bokmal.md
         - Norwegian Nynorsk: benchmarks/languages/norwegian-nynorsk.md
+        - Old English: benchmarks/languages/ang.md
+        - Old French: benchmarks/languages/fro.md
+        - Old High German: benchmarks/languages/goh.md
+        - Old Irish: benchmarks/languages/sga.md
+        - Old Norse: benchmarks/languages/non.md
+        - Old Saxon: benchmarks/languages/osx.md
+        - Oodham: benchmarks/languages/ood.md
+        - Pashto: benchmarks/languages/ps-af.md
         - Persian: benchmarks/languages/persian.md
         - Polish: benchmarks/languages/polish.md
         - Portuguese: benchmarks/languages/portuguese.md
+        - Quechua: benchmarks/languages/que.md
+        - Romanian: benchmarks/languages/ro-ro.md
         - Russian: benchmarks/languages/russian.md
+        - Seneca: benchmarks/languages/see.md
+        - Serbo-Croatian: benchmarks/languages/hbs.md
+        - Shipibo-Conibo: benchmarks/languages/shp.md
+        - Shona: benchmarks/languages/sn-zw.md
+        - Sotho, Southern: benchmarks/languages/st-za.md
+        - Southern Kurdish: benchmarks/languages/sdh.md
         - Spanish: benchmarks/languages/spanish.md
         - Swedish: benchmarks/languages/swedish.md
+        - Tagalog: benchmarks/languages/tl-ph.md
+        - Temoaya Otomi: benchmarks/languages/ote.md
+        - Turkish: benchmarks/languages/tr-tr.md
         - Ukrainian: benchmarks/languages/ukrainian.md
+        - Uyghur: benchmarks/languages/ug-cn.md
+        - Uzbek: benchmarks/languages/uz-uz.md
+        - Voro: benchmarks/languages/vro.md
+        - Western Highland Chatino: benchmarks/languages/ctp.md
+        - Xibe: benchmarks/languages/sjo.md
+        - Xipinawa: benchmarks/languages/xty.md
+        - Yamana: benchmarks/languages/ame.md
         - Yiddish: benchmarks/languages/yiddish.md
+        - Zacatepec Chatino: benchmarks/languages/zpv.md
+        - Zarma: benchmarks/languages/dje.md
+        - Zenzontepec Chatino: benchmarks/languages/czn.md
+        - Zulu: benchmarks/languages/zu-za.md
      - Methods and Reproduction:
         - Comparison Methodology: benchmarks/reference/methodology.md
         - Linguistic Quality: benchmarks/reference/linguistic-quality.md
diff --git a/models/fa-ir-default/build.gradle b/models/fa-ir-default/build.gradle
index 8314fc9..3be9e65 100644
--- a/models/fa-ir-default/build.gradle
+++ b/models/fa-ir-default/build.gradle
@@ -38,6 +38,7 @@ radixorModel {
     language = 'FA_IR'
     displayName = 'Persian default model'
     defaultModel = true
+    rightToLeft = true
     sourceName = 'UniMorph'
     sourceVersion = 'not-recorded-in-legacy-import'
     sourceRevision = 'not-recorded-in-legacy-import'
@@ -47,7 +48,7 @@ radixorModel {
     sourceRevisionStatus = 'not-recorded-in-legacy-import'
     sourceLicense = 'CC-BY-SA-3.0'
     sourceLicenseUri = 'https://creativecommons.org/licenses/by-sa/3.0/'
-    sourceAttribution = 'UniMorph and Wikipedia contributors'
+    sourceAttribution = 'UniMorph'
     sourceVerificationDate = '2026-07-22'
     transformationsSummary = 'Cleaning, normalization, grouping inflected forms by lemma, deduplication, filtering invalid rows, reformatting into Radixor dictionary groups, GZip packaging, and generation of runtime descriptor and checksum metadata'
     noticeFileName = 'NOTICE-model-data.txt'
diff --git a/models/fa-ir-default/src/modelInput/NOTICE-model-data.txt b/models/fa-ir-default/src/modelInput/NOTICE-model-data.txt
index 92f065b..9289718 100644
--- a/models/fa-ir-default/src/modelInput/NOTICE-model-data.txt
+++ b/models/fa-ir-default/src/modelInput/NOTICE-model-data.txt
@@ -22,12 +22,14 @@ Source project: UniMorph
 Official repository: https://github.com/unimorph/fas
 Upstream dataset: UniMorph Persian morphological dataset (`fas`)
 Upstream lexical source: Wikipedia
-Attribution: UniMorph and Wikipedia contributors
+Attribution: UniMorph
 License:
 Creative Commons Attribution-ShareAlike 3.0 Unported
+
 Canonical license URI: https://creativecommons.org/licenses/by-sa/3.0/
 Source revision: not-recorded-in-legacy-import
 Revision status: not-recorded-in-legacy-import
+Note: license not stated explicitly in repository source; CC-BY-SA-3.0 is the UniMorph project standard.
 
 The exact UniMorph commit used for the original Radixor import was not recorded. The model remains attributed to the official UniMorph language repository and is distributed under the repository's stated data license.
 
diff --git a/models/he-il-default/build.gradle b/models/he-il-default/build.gradle
index 928be55..02cde2a 100644
--- a/models/he-il-default/build.gradle
+++ b/models/he-il-default/build.gradle
@@ -38,6 +38,7 @@ radixorModel {
     language = 'HE_IL'
     displayName = 'Hebrew default model'
     defaultModel = true
+    rightToLeft = true
     sourceName = 'UniMorph'
     sourceVersion = 'not-recorded-in-legacy-import'
     sourceRevision = 'not-recorded-in-legacy-import'
@@ -47,7 +48,7 @@ radixorModel {
     sourceRevisionStatus = 'not-recorded-in-legacy-import'
     sourceLicense = 'CC-BY-SA-3.0'
     sourceLicenseUri = 'https://creativecommons.org/licenses/by-sa/3.0/'
-    sourceAttribution = 'UniMorph; Omer Goldman (annotator); Wiktionary contributors'
+    sourceAttribution = 'UniMorph; Omer Goldman'
     sourceVerificationDate = '2026-07-22'
     transformationsSummary = 'Cleaning, normalization, grouping inflected forms by lemma, deduplication, filtering invalid rows, reformatting into Radixor dictionary groups, GZip packaging, and generation of runtime descriptor and checksum metadata'
     noticeFileName = 'NOTICE-model-data.txt'
diff --git a/models/he-il-default/src/modelInput/NOTICE-model-data.txt b/models/he-il-default/src/modelInput/NOTICE-model-data.txt
index 483c368..94692fc 100644
--- a/models/he-il-default/src/modelInput/NOTICE-model-data.txt
+++ b/models/he-il-default/src/modelInput/NOTICE-model-data.txt
@@ -22,12 +22,14 @@ Source project: UniMorph
 Official repository: https://github.com/unimorph/heb
 Upstream dataset: UniMorph Hebrew morphological dataset (`heb`)
 Upstream lexical source: Wiktionary
-Attribution: UniMorph; Omer Goldman (annotator); Wiktionary contributors
+Attribution: UniMorph; Omer Goldman
 License:
 Creative Commons Attribution-ShareAlike 3.0 Unported
+
 Canonical license URI: https://creativecommons.org/licenses/by-sa/3.0/
 Source revision: not-recorded-in-legacy-import
 Revision status: not-recorded-in-legacy-import
+Note: license not stated explicitly in repository source; CC-BY-SA-3.0 is the UniMorph project standard.
 
 The exact UniMorph commit used for the original Radixor import was not recorded. The model remains attributed to the official UniMorph language repository and is distributed under the repository's stated data license.
 
diff --git a/models/model-projects.properties b/models/model-projects.properties
index 6511de4..3721a08 100644
--- a/models/model-projects.properties
+++ b/models/model-projects.properties
@@ -1,22 +1,148 @@
-# Build-topology membership only. Per-model metadata remains authoritative.
+# Build-topology membership only. "default" selects the original standard-pack
+# members; "standalone" and "optional" remain individually publishable. All
+# roles participate in settings, verification, JMH, BOM, and catalog generation.
+# Per-model metadata remains authoritative.
+ady-default=standalone
+af-za-default=standalone
+afb-default=standalone
+ail-default=standalone
+aka-default=standalone
+am-et-default=standalone
+ame-default=standalone
+ang-default=standalone
+ar-default=standalone
+arn-default=standalone
+arz-default=standalone
+as-in-default=standalone
+ast-default=standalone
+aym-default=standalone
+az-az-default=standalone
+azg-default=standalone
+bak-default=standalone
+be-by-default=standalone
+bg-bg-default=standalone
+bn-bd-default=standalone
+bra-default=standalone
+bre-default=standalone
+ca-es-default=standalone
+ceb-default=standalone
+chu-default=standalone
+ckt-default=standalone
+cly-default=standalone
+cni-default=standalone
+cor-default=standalone
+cpa-default=standalone
+cre-default=standalone
+crh-default=standalone
 cs-cz-default=default
+csb-default=standalone
+ctp-default=standalone
+czn-default=standalone
 da-dk-default=default
+dak-default=standalone
 de-de-default=default
+dje-default=standalone
+dsb-default=standalone
+el-gr-default=standalone
 es-es-default=default
+et-ee-default=standalone
+evn-default=standalone
 fa-ir-default=default
 fi-fi-default=default
+fo-fo-default=standalone
 fr-fr-default=default
+frm-default=standalone
+fro-default=standalone
+frr-default=standalone
+fur-default=standalone
+ga-ie-default=standalone
+gaa-default=standalone
+gal-default=standalone
+gmh-default=standalone
+gml-default=standalone
+goh-default=standalone
+got-default=standalone
+grc-default=standalone
+gsw-default=standalone
+gup-default=standalone
+gv-im-default=standalone
+hai-default=standalone
+hbs-default=standalone
 he-il-default=default
+hil-default=standalone
+hsi-default=standalone
 hu-hu-default=default
+hy-am-default=standalone
+id-id-default=standalone
+is-is-default=standalone
 it-it-default=default
+itl-default=standalone
+izh-default=standalone
+ja-jp-default=standalone
+kbd-default=standalone
+kjh-default=standalone
+kk-kz-default=standalone
+kl-gl-default=standalone
+klr-default=standalone
+kn-in-default=standalone
+kod-default=standalone
+kon-default=standalone
+krl-default=standalone
+ky-kg-default=standalone
+la-default=standalone
+lg-ug-default=standalone
+lin-default=standalone
+liv-default=standalone
+lld-default=standalone
+lt-lt-default=standalone
+lv-lv-default=standalone
+mag-default=standalone
+mg-mg-default=standalone
+mi-nz-default=standalone
+mk-mk-default=standalone
+mn-mn-default=standalone
+mt-mt-default=standalone
+mwf-default=standalone
+nap-default=standalone
+nav-default=standalone
 nb-no-default=default
+nds-default=standalone
 nl-nl-default=default
 nn-no-default=default
-pl-pl-polimorf=optional
+non-default=standalone
+ny-mw-default=standalone
+ood-default=standalone
+osx-default=standalone
+ote-default=standalone
 pl-pl-unimorph=default
+ps-af-default=standalone
 pt-pt-default=default
+que-default=standalone
+ro-ro-default=standalone
 ru-ru-default=default
+sdh-default=standalone
+see-default=standalone
+sga-default=standalone
+shp-default=standalone
+sjo-default=standalone
+sme-default=standalone
+sn-zw-default=standalone
+sq-al-default=standalone
+st-za-default=standalone
 sv-se-default=default
+swc-default=standalone
+syc-default=standalone
+tl-ph-default=standalone
+tr-tr-default=standalone
+ug-cn-default=standalone
 uk-ua-default=default
 us-uk-default=default
+uz-uz-default=standalone
+vro-default=standalone
+xcl-default=standalone
+xno-default=standalone
+xty-default=standalone
 yi-default=default
+zpv-default=standalone
+zu-za-default=standalone
+pl-pl-polimorf=optional
diff --git a/models/standard/build.gradle b/models/standard/build.gradle
index 5455cee..499c2f1 100644
--- a/models/standard/build.gradle
+++ b/models/standard/build.gradle
@@ -63,7 +63,7 @@ publishing {
             artifactId = 'radixor-models-standard'
             pom {
                 name = 'Radixor Standard Stemmer Models'
-                description = 'POM-only runtime aggregate containing one default model for every supported language.'
+                description = 'POM-only runtime aggregate containing the original 20 standard model dependencies.'
                 packaging = 'pom'
                 url = 'https://github.com/leogalambos/Radixor'
                 licenses {
@@ -135,7 +135,7 @@ tasks.register('verifyPomOnlyAggregate') {
             throw new GradleException('radixor-models-standard must publish Maven packaging pom.')
         }
         if (artifactIds != expectedStandardArtifactIds || dependencies.any { Node dependency -> dependency.scope.text() != 'runtime' }) {
-            throw new GradleException('radixor-models-standard must contain exactly the ordered default model runtime dependencies.')
+            throw new GradleException('radixor-models-standard must contain exactly the ordered standard model runtime dependencies.')
         }
         if (artifactIds.contains('radixor-model-pl-pl-polimorf')) {
             throw new GradleException('radixor-models-standard must exclude the optional PoliMorf model.')
diff --git a/models/yi-default/build.gradle b/models/yi-default/build.gradle
index 74e2642..8539581 100644
--- a/models/yi-default/build.gradle
+++ b/models/yi-default/build.gradle
@@ -38,6 +38,7 @@ radixorModel {
     language = 'YI'
     displayName = 'Yiddish default model'
     defaultModel = true
+    rightToLeft = true
     sourceName = 'UniMorph'
     sourceVersion = 'not-recorded-in-legacy-import'
     sourceRevision = 'not-recorded-in-legacy-import'
diff --git a/models/yi-default/src/modelInput/NOTICE-model-data.txt b/models/yi-default/src/modelInput/NOTICE-model-data.txt
index 3d6c76f..b7f4897 100644
--- a/models/yi-default/src/modelInput/NOTICE-model-data.txt
+++ b/models/yi-default/src/modelInput/NOTICE-model-data.txt
@@ -25,9 +25,11 @@ Upstream lexical source: The official repository and catalog do not name a separ
 Attribution: UniMorph
 License:
 Creative Commons Attribution-ShareAlike 3.0 Unported
+
 Canonical license URI: https://creativecommons.org/licenses/by-sa/3.0/
 Source revision: not-recorded-in-legacy-import
 Revision status: not-recorded-in-legacy-import
+Note: license not stated explicitly in repository source; CC-BY-SA-3.0 is the UniMorph project standard.
 
 The exact UniMorph commit used for the original Radixor import was not recorded. The model remains attributed to the official UniMorph language repository and is distributed under the repository's stated data license.
 
diff --git a/python/README.md b/python/README.md
index 8b8feee..8fc9aca 100644
--- a/python/README.md
+++ b/python/README.md
@@ -301,7 +301,7 @@ sdist contains build-ready gzip v7 `.rxc` files, a checksummed provenance
 manifest, and per-model CC BY-SA 3.0 notices. They are generated below `build/`
 from canonical `models/*/src/modelInput/stemmer.gz` inputs and are never stored
 in Git. `./gradlew regeneratePythonStandardModels` performs this deterministic
-generation; repository topology selects the 20 defaults and excludes optional
+generation; repository topology selects the 20 `default` entries and excludes `standalone` and optional
 `pl-pl-polimorf`.
 
 `radixor` requires `radixor-models-standard>=2.0,<3.0`. The Python distribution
diff --git a/python/scripts/build_standard_models.py b/python/scripts/build_standard_models.py
index ea62af5..54c331f 100644
--- a/python/scripts/build_standard_models.py
+++ b/python/scripts/build_standard_models.py
@@ -32,7 +32,9 @@
 
 """Regenerate the pure-Python standard model package deterministically.
 
-The canonical build topology selects ``default`` models. Source dictionaries
+The canonical build topology selects ``default`` models. ``standalone`` and
+``optional`` models remain individually publishable but are deliberately
+excluded from this aggregate. Source dictionaries
 are compiler inputs only and are never copied into either Python distribution.
 Run this script with a built Radixor extension importable by the selected
 Python interpreter.
@@ -90,7 +92,11 @@ def _default_models() -> list[str]:
         if not line or line.startswith("#"):
             continue
         model_id, separator, membership = line.partition("=")
-        if not separator or not model_id or membership not in {"default", "optional"}:
+        if not separator or not model_id or membership not in {
+            "default",
+            "standalone",
+            "optional",
+        }:
             raise ValueError(f"Invalid model topology line: {raw_line!r}")
         entries[model_id] = membership
     model_ids = sorted(
diff --git a/python/scripts/verify_distributions.py b/python/scripts/verify_distributions.py
index 790fc6d..78ef026 100644
--- a/python/scripts/verify_distributions.py
+++ b/python/scripts/verify_distributions.py
@@ -81,7 +81,11 @@ def _tracked_default_model_versions() -> dict[str, str]:
         if not line or line.startswith("#"):
             continue
         model_id, separator, membership = line.partition("=")
-        if not separator or membership not in {"default", "optional"}:
+        if not separator or membership not in {
+            "default",
+            "standalone",
+            "optional",
+        }:
             raise ValueError(f"Invalid model topology line: {raw_line!r}")
         if membership != "default":
             continue
diff --git a/settings.gradle b/settings.gradle
index 92653fa..94e9f25 100644
--- a/settings.gradle
+++ b/settings.gradle
@@ -42,10 +42,37 @@ Properties modelTopology = new Properties()
 file('models/model-projects.properties').withInputStream { InputStream input ->
     modelTopology.load(input)
 }
+final Set<String> acceptedModelRoles = ['default', 'standalone', 'optional'] as Set<String>
+modelTopology.each { Object modelId, Object role ->
+    if (!acceptedModelRoles.contains(role.toString())) {
+        throw new GradleException("Model ${modelId} has unsupported topology role '${role}'. "
+                + "Expected default, standalone, or optional.")
+    }
+}
 modelTopology.stringPropertyNames().toList().sort().each { String modelId ->
     include(":models:${modelId}")
 }
 
+Properties modelQuarantine = new Properties()
+file('models/model-quarantine.properties').withInputStream { InputStream input ->
+    modelQuarantine.load(input)
+}
+final Set<String> quarantineOverlap = modelTopology.stringPropertyNames().intersect(
+        modelQuarantine.stringPropertyNames())
+if (!quarantineOverlap.isEmpty()) {
+    throw new GradleException("Quarantined model IDs must not appear in active topology: "
+            + quarantineOverlap.toList().sort().join(', '))
+}
+modelQuarantine.stringPropertyNames().each { String modelId ->
+    if (!file("models/${modelId}").isDirectory()) {
+        throw new GradleException("Quarantined model directory is missing: models/${modelId}")
+    }
+    String record = modelQuarantine.getProperty(modelId)
+    if (!(record ==~ /https:\/\/github\.com\/unimorph\/[a-z0-9-]+\|\S.*\S/)) {
+        throw new GradleException("Quarantine record for ${modelId} must contain an evidence URI and reason.")
+    }
+}
+
 Properties alternativeModelTopology = new Properties()
 file('models/alternative-model-projects.properties').withInputStream { InputStream input ->
     alternativeModelTopology.load(input)
diff --git a/src/jmh/java/org/egothor/stemmer/benchmark/BenchmarkCorpusReportApplication.java b/src/jmh/java/org/egothor/stemmer/benchmark/BenchmarkCorpusReportApplication.java
index eb8ed45..3e0bf3d 100644
--- a/src/jmh/java/org/egothor/stemmer/benchmark/BenchmarkCorpusReportApplication.java
+++ b/src/jmh/java/org/egothor/stemmer/benchmark/BenchmarkCorpusReportApplication.java
@@ -37,8 +37,10 @@ import java.io.InputStreamReader;
 import java.nio.charset.StandardCharsets;
 import java.nio.file.Files;
 import java.nio.file.Path;
+import java.util.HashSet;
 import java.util.Map;
 import java.util.Objects;
+import java.util.Set;
 import java.util.TreeMap;
 import java.util.zip.GZIPInputStream;
 
@@ -53,13 +55,13 @@ import org.egothor.stemmer.StemmerPatchTrieLoader;
 
 /**
  * Writes deterministic corpus and preferred patch-command counts for every
- * registered default model.
+ * registered user-facing model.
  *
  * <p>
  * This application performs setup-time analysis only; it does not publish or
- * interpret runtime performance. Optional model variants are excluded by
- * resolving every entry through
- * {@link StemmerModelRegistry#requireDefault(StemmerPatchTrieLoader.Language)}.
+ * interpret runtime performance. Models are selected by exact descriptor ID,
+ * including optional user-facing variants and excluding non-registered
+ * experimental candidates.
  * </p>
  */
 public final class BenchmarkCorpusReportApplication {
@@ -90,30 +92,28 @@ public final class BenchmarkCorpusReportApplication {
         }
 
         final StringBuilder csv = new StringBuilder(16_384);
-        csv.append("Language,Model ID,Model version,Model SHA-256,Dictionary rows,Total tokens,Already-root tokens,Changed tokens,")
-                .append("Speed timing tokens,All exact matches,Changed exact matches,Root preserved matches,")
+        csv.append("Language,Model ID,Model version,Model SHA-256,Dictionary rows,Distinct usable forms,Total tokens,Already-root tokens,Changed tokens,")
+                .append("Speed timing workload,Speed timing tokens,All exact matches,Changed exact matches,Root preserved matches,")
                 .append("Command class,Command count\n");
         final StemmerModelRegistry registry = StemmerModelRegistry.fromContextClassLoader();
-        for (StemmerPatchTrieLoader.Language language : StemmerPatchTrieLoader.Language.values()) {
-            appendLanguage(csv, registry, language);
+        for (StemmerModelDescriptor descriptor : registry.models()) {
+            appendModel(csv, descriptor);
         }
         Files.writeString(output, csv, StandardCharsets.UTF_8);
         System.out.println("Benchmark corpus report: " + output.toAbsolutePath());
     }
 
     /**
-     * Appends all command-class rows for one default model.
+     * Appends all command-class rows for one exact model.
      *
      * @param csv      destination
-     * @param registry discovered model registry
-     * @param language language to analyze
+     * @param descriptor exact registered model descriptor
      * @throws IOException if the model cannot be parsed or loaded
      */
-    private static void appendLanguage(final StringBuilder csv, final StemmerModelRegistry registry,
-            final StemmerPatchTrieLoader.Language language) throws IOException {
-        final StemmerModelDescriptor descriptor = registry.requireDefault(language);
-        final int dictionaryRows = countDictionaryRows(descriptor);
-        final LanguageBenchmarkCorpus.Corpus corpus = LanguageBenchmarkCorpus.createFullCorpus(language);
+    private static void appendModel(final StringBuilder csv, final StemmerModelDescriptor descriptor)
+            throws IOException {
+        final CorpusStatistics statistics = countDictionary(descriptor);
+        final LanguageBenchmarkCorpus.Corpus corpus = LanguageBenchmarkCorpus.createFullCorpus(descriptor.id());
         final String[] tokens = corpus.tokens();
         final String[] expectedRoots = corpus.expectedRoots();
         long alreadyRootTokens = 0;
@@ -123,9 +123,11 @@ public final class BenchmarkCorpusReportApplication {
             }
         }
         final long changedTokens = tokens.length - alreadyRootTokens;
-        final int timingTokens = LanguageBenchmarkCorpus.createChangedCorpus(language).tokens().length;
+        final LanguageBenchmarkCorpus.TimingCorpus timingCorpus =
+                LanguageBenchmarkCorpus.createTimingCorpus(descriptor.id());
+        final int timingTokens = timingCorpus.corpus().tokens().length;
 
-        final FrequencyTrie<CompiledPatchCommand> trie = StemmerPatchTrieLoader.loadCompiled(language, true,
+        final FrequencyTrie<CompiledPatchCommand> trie = StemmerPatchTrieLoader.loadCompiled(descriptor.id(), true,
                 ReductionMode.MERGE_SUBTREES_WITH_EQUIVALENT_RANKED_GET_ALL_RESULTS);
         final RadixorBenchmarkStemmer stemmer = new RadixorBenchmarkStemmer(trie);
         final Map<String, Long> commandCounts = new TreeMap<>();
@@ -149,14 +151,16 @@ public final class BenchmarkCorpusReportApplication {
             }
         }
         for (Map.Entry<String, Long> commandCount : commandCounts.entrySet()) {
-            csv.append(language).append(',')
+            csv.append(descriptor.language()).append(',')
                     .append(descriptor.id()).append(',')
                     .append(descriptor.version()).append(',')
                     .append(descriptor.sha256()).append(',')
-                    .append(dictionaryRows).append(',')
+                    .append(statistics.dictionaryRows()).append(',')
+                    .append(statistics.distinctUsableForms()).append(',')
                     .append(tokens.length).append(',')
                     .append(alreadyRootTokens).append(',')
                     .append(changedTokens).append(',')
+                    .append(timingCorpus.basis().reportValue()).append(',')
                     .append(timingTokens).append(',')
                     .append(allExactMatches).append(',')
                     .append(changedExactMatches).append(',')
@@ -167,13 +171,13 @@ public final class BenchmarkCorpusReportApplication {
     }
 
     /**
-     * Counts valid logical rows in one default dictionary.
+     * Counts valid logical rows and distinct case-preserved usable surface forms.
      *
      * @param descriptor model descriptor
-     * @return parsed dictionary-row count
+     * @return parsed dictionary statistics
      * @throws IOException if the dictionary cannot be opened or parsed
      */
-    private static int countDictionaryRows(final StemmerModelDescriptor descriptor) throws IOException {
+    private static CorpusStatistics countDictionary(final StemmerModelDescriptor descriptor) throws IOException {
         final ClassLoader contextClassLoader = Thread.currentThread().getContextClassLoader();
         final ClassLoader classLoader = contextClassLoader == null
                 ? BenchmarkCorpusReportApplication.class.getClassLoader()
@@ -184,12 +188,28 @@ public final class BenchmarkCorpusReportApplication {
                     + descriptor.resource() + ".");
         }
         final int[] rows = {0};
+        final Set<String> distinctForms = new HashSet<>();
         try (InputStream raw = resource;
                 GZIPInputStream gzip = new GZIPInputStream(raw);
                 BufferedReader reader = new BufferedReader(new InputStreamReader(gzip, StandardCharsets.UTF_8))) {
-            StemmerDictionaryParser.parse(reader, descriptor.resource(), CaseProcessingMode.LOWERCASE_WITH_LOCALE_ROOT,
-                    (stem, variants, lineNumber) -> rows[0] = Math.addExact(rows[0], 1));
+            StemmerDictionaryParser.parse(reader, descriptor.resource(), CaseProcessingMode.AS_IS,
+                    (stem, variants, lineNumber) -> {
+                        rows[0] = Math.addExact(rows[0], 1);
+                        distinctForms.add(stem);
+                        for (String variant : variants) {
+                            distinctForms.add(variant);
+                        }
+                    });
         }
-        return rows[0];
+        return new CorpusStatistics(rows[0], distinctForms.size());
+    }
+
+    /**
+     * Dictionary size statistics computed in one streaming parser pass.
+     *
+     * @param dictionaryRows parsed logical row count
+     * @param distinctUsableForms exact distinct case-preserved form count
+     */
+    private record CorpusStatistics(int dictionaryRows, int distinctUsableForms) {
     }
 }
diff --git a/src/jmh/java/org/egothor/stemmer/benchmark/LanguageBenchmarkCorpus.java b/src/jmh/java/org/egothor/stemmer/benchmark/LanguageBenchmarkCorpus.java
index 3146e17..7558cad 100644
--- a/src/jmh/java/org/egothor/stemmer/benchmark/LanguageBenchmarkCorpus.java
+++ b/src/jmh/java/org/egothor/stemmer/benchmark/LanguageBenchmarkCorpus.java
@@ -37,10 +37,9 @@ import java.io.InputStreamReader;
 import java.nio.charset.StandardCharsets;
 import java.util.ArrayList;
 import java.util.EnumMap;
-import java.util.HashMap;
 import java.util.List;
-import java.util.Map;
 import java.util.Locale;
+import java.util.Map;
 import java.util.Objects;
 import java.util.zip.GZIPInputStream;
 
@@ -55,8 +54,11 @@ import org.egothor.stemmer.StemmerPatchTrieLoader;
  * <p>
  * Corpus construction is setup work only. It is intentionally based on the same
  * resource that backs the Radixor benchmark path so every competitor for a
- * language consumes the same changed-token timing workload, while quality
- * benchmarks can still use the complete dictionary workload.
+ * language consumes the same timing workload, while quality benchmarks can
+ * still use the complete dictionary workload. The timing workload contains
+ * changed token/root pairs whenever at least one exists. A root-only dictionary
+ * instead uses its complete root-preservation corpus so that every registered
+ * model remains measurable.
  * </p>
  */
 final class LanguageBenchmarkCorpus {
@@ -75,14 +77,9 @@ final class LanguageBenchmarkCorpus {
     /**
      * Shared changed-token timing corpora keyed by bundled Radixor language.
      */
-    private static final Map<StemmerPatchTrieLoader.Language, Corpus> CHANGED_TIMING_CORPORA =
+    private static final Map<StemmerPatchTrieLoader.Language, TimingCorpus> TIMING_WORKLOADS =
             new EnumMap<>(StemmerPatchTrieLoader.Language.class);
 
-    /**
-     * Shared changed-token timing corpora keyed by explicit bundled model ID.
-     */
-    private static final Map<String, Corpus> MODEL_CHANGED_TIMING_CORPORA = new HashMap<>();
-
     /**
      * Shared complete corpora keyed by bundled Radixor language.
      */
@@ -97,65 +94,84 @@ final class LanguageBenchmarkCorpus {
     }
 
     /**
-     * Creates a deterministic changed-token timing corpus from a bundled language
-     * dictionary.
+     * Creates the canonical timing token array from a bundled language dictionary.
      *
      * <p>
-     * Only token/root pairs where the token differs from the expected root are
-     * included. Smaller changed-token resources are repeated in stable order until
-     * the timing corpus reaches 5,000 tokens.
+     * Changed token/root pairs are preferred. A root-only dictionary falls back to
+     * its complete root-preservation corpus. Smaller resources are repeated in
+     * stable order until the timing corpus reaches 5,000 tokens.
      * </p>
      *
      * @param language bundled Radixor language
-     * @return token array containing changed-token dictionary entries, repeated
-     *         only when the changed-token resource is smaller than 5,000 tokens
+     * @return canonical timing tokens, repeated only when the selected population
+     *         is smaller than 5,000 tokens
      * @throws IOException if the resource cannot be read
      */
     static String[] createTokens(final StemmerPatchTrieLoader.Language language) throws IOException {
-        return createChangedCorpus(language).tokens();
+        return createTimingCorpus(language).corpus().tokens();
     }
 
     /**
-     * Creates a deterministic changed-token timing corpus from an explicitly
-     * selected bundled model dictionary.
+     * Creates the canonical timing token array for an exact model identifier.
      *
      * <p>
-     * Only token/root pairs where the token differs from the expected root are
-     * included. Smaller changed-token resources are repeated in stable order until
-     * the timing corpus reaches 5,000 tokens.
+     * Changed token/root pairs are preferred. A root-only dictionary falls back to
+     * its complete root-preservation corpus. Smaller resources are repeated in
+     * stable order until the timing corpus reaches 5,000 tokens.
      * </p>
      *
      * @param modelId exact bundled model identifier
-     * @return token array containing changed-token dictionary entries, repeated
-     *         only when the changed-token resource is smaller than 5,000 tokens
+     * @return canonical timing tokens, repeated only when the selected population
+     *         is smaller than 5,000 tokens
      * @throws IOException if the resource cannot be read
      */
     static String[] createTokens(final String modelId) throws IOException {
-        return createChangedCorpus(modelId).tokens();
+        return createTimingCorpus(modelId).corpus().tokens();
     }
 
     /**
-     * Creates a deterministic changed-token timing corpus from a bundled language
-     * dictionary.
+     * Creates the canonical timing corpus from a bundled language dictionary.
      *
      * @param language bundled Radixor language
-     * @return changed-token corpus with expected roots
+     * @return changed-token corpus, or the root-preservation fallback
      * @throws IOException if the resource cannot be read
      */
     static Corpus createChangedCorpus(final StemmerPatchTrieLoader.Language language) throws IOException {
-        return cachedChangedCorpus(language);
+        return createTimingCorpus(language).corpus();
     }
 
     /**
-     * Creates a deterministic changed-token timing corpus from an explicitly
-     * selected bundled model dictionary.
+     * Creates the canonical timing corpus for an exact model identifier.
      *
      * @param modelId exact bundled model identifier
-     * @return changed-token corpus with expected roots
+     * @return changed-token corpus, or the root-preservation fallback
      * @throws IOException if the resource cannot be read
      */
     static Corpus createChangedCorpus(final String modelId) throws IOException {
-        return cachedChangedCorpus(modelId);
+        return createTimingCorpus(modelId).corpus();
+    }
+
+    /**
+     * Creates the canonical timing workload for a bundled language default.
+     *
+     * @param language bundled Radixor language
+     * @return timing corpus and its explicit workload basis
+     * @throws IOException if the resource cannot be read
+     */
+    static TimingCorpus createTimingCorpus(final StemmerPatchTrieLoader.Language language) throws IOException {
+        return cachedTimingCorpus(language);
+    }
+
+    /**
+     * Creates the canonical timing workload for an exact model identifier.
+     *
+     * @param modelId exact bundled model identifier
+     * @return timing corpus and its explicit workload basis
+     * @throws IOException if the resource cannot be read
+     */
+    static TimingCorpus createTimingCorpus(final String modelId) throws IOException {
+        Objects.requireNonNull(modelId, "modelId");
+        return buildTimingWorkload(modelId, MINIMUM_TIMING_TOKEN_COUNT);
     }
 
     /**
@@ -209,6 +225,29 @@ final class LanguageBenchmarkCorpus {
         return cachedCorpus(FULL_CORPORA, language, false);
     }
 
+    /**
+     * Creates a complete deterministic corpus for an exact model identifier.
+     *
+     * @param modelId exact bundled model identifier
+     * @return complete token corpus with expected roots
+     * @throws IOException if the resource cannot be read
+     */
+    static Corpus createFullCorpus(final String modelId) throws IOException {
+        Objects.requireNonNull(modelId, "modelId");
+        final List<Entry> candidates = readCandidates(modelId, Integer.MAX_VALUE);
+        if (candidates.isEmpty()) {
+            throw new IllegalStateException("No benchmark corpus tokens were available for " + modelId + ".");
+        }
+        final String[] tokens = new String[candidates.size()];
+        final String[] expectedRoots = new String[candidates.size()];
+        for (int index = 0; index < candidates.size(); index++) {
+            final Entry entry = candidates.get(index);
+            tokens[index] = entry.token();
+            expectedRoots[index] = entry.root();
+        }
+        return new Corpus(tokens, expectedRoots);
+    }
+
     /**
      * Returns a cached corpus, creating it once per JVM when necessary.
      *
@@ -237,47 +276,24 @@ final class LanguageBenchmarkCorpus {
     }
 
     /**
-     * Returns a cached changed-token timing corpus, creating it once per JVM when
+     * Returns a cached canonical timing workload, creating it once per JVM when
      * necessary.
      *
      * @param language bundled Radixor language
-     * @return changed-token timing corpus
+     * @return canonical timing workload
      * @throws IOException if the resource cannot be read
      */
-    private static Corpus cachedChangedCorpus(final StemmerPatchTrieLoader.Language language) throws IOException {
+    private static TimingCorpus cachedTimingCorpus(final StemmerPatchTrieLoader.Language language) throws IOException {
         Objects.requireNonNull(language, "language");
 
         synchronized (LanguageBenchmarkCorpus.class) {
-            final Corpus existing = CHANGED_TIMING_CORPORA.get(language);
-            if (existing != null) {
-                return existing;
-            }
-
-            final Corpus created = buildChangedTimingCorpus(language, MINIMUM_TIMING_TOKEN_COUNT);
-            CHANGED_TIMING_CORPORA.put(language, created);
-            return created;
-        }
-    }
-
-    /**
-     * Returns a cached changed-token timing corpus, creating it once per JVM when
-     * necessary.
-     *
-     * @param modelId exact bundled model identifier
-     * @return changed-token timing corpus
-     * @throws IOException if the resource cannot be read
-     */
-    private static Corpus cachedChangedCorpus(final String modelId) throws IOException {
-        Objects.requireNonNull(modelId, "modelId");
-
-        synchronized (LanguageBenchmarkCorpus.class) {
-            final Corpus existing = MODEL_CHANGED_TIMING_CORPORA.get(modelId);
+            final TimingCorpus existing = TIMING_WORKLOADS.get(language);
             if (existing != null) {
                 return existing;
             }
 
-            final Corpus created = buildChangedTimingCorpus(modelId, MINIMUM_TIMING_TOKEN_COUNT);
-            MODEL_CHANGED_TIMING_CORPORA.put(modelId, created);
+            final TimingCorpus created = buildTimingWorkload(language, MINIMUM_TIMING_TOKEN_COUNT);
+            TIMING_WORKLOADS.put(language, created);
             return created;
         }
     }
@@ -314,45 +330,44 @@ final class LanguageBenchmarkCorpus {
     }
 
     /**
-     * Builds a deterministic changed-token timing corpus from a bundled language
+     * Builds a deterministic canonical timing workload from a bundled language
      * dictionary.
      *
      * @param language bundled Radixor language
      * @param minimumTokenCount minimum token count for timing
-     * @return changed-token corpus with expected roots
+     * @return canonical timing workload and basis
      * @throws IOException if the resource cannot be read
      */
-    private static Corpus buildChangedTimingCorpus(final StemmerPatchTrieLoader.Language language,
+    private static TimingCorpus buildTimingWorkload(final StemmerPatchTrieLoader.Language language,
             final int minimumTokenCount) throws IOException {
         Objects.requireNonNull(language, "language");
-        return buildChangedTimingCorpus(readCandidates(language, Integer.MAX_VALUE), language.toString(),
+        return buildTimingWorkload(readCandidates(language, Integer.MAX_VALUE), language.toString(),
                 minimumTokenCount);
     }
 
     /**
-     * Builds a deterministic changed-token timing corpus from an explicitly
-     * selected bundled model dictionary.
+     * Builds a deterministic canonical timing workload for an exact model.
      *
      * @param modelId exact bundled model identifier
      * @param minimumTokenCount minimum token count for timing
-     * @return changed-token corpus with expected roots
+     * @return canonical timing workload and basis
      * @throws IOException if the resource cannot be read
      */
-    private static Corpus buildChangedTimingCorpus(final String modelId, final int minimumTokenCount)
+    private static TimingCorpus buildTimingWorkload(final String modelId, final int minimumTokenCount)
             throws IOException {
         Objects.requireNonNull(modelId, "modelId");
-        return buildChangedTimingCorpus(readCandidates(modelId, Integer.MAX_VALUE), modelId, minimumTokenCount);
+        return buildTimingWorkload(readCandidates(modelId, Integer.MAX_VALUE), modelId, minimumTokenCount);
     }
 
     /**
-     * Builds a deterministic changed-token timing corpus from parsed entries.
+     * Builds a canonical timing workload from parsed entries.
      *
      * @param allCandidates all valid dictionary entries
      * @param sourceLabel human-readable source label for diagnostics
      * @param minimumTokenCount minimum token count for timing
-     * @return changed-token corpus with expected roots
+     * @return canonical timing workload and basis
      */
-    private static Corpus buildChangedTimingCorpus(final List<Entry> allCandidates, final String sourceLabel,
+    private static TimingCorpus buildTimingWorkload(final List<Entry> allCandidates, final String sourceLabel,
             final int minimumTokenCount) {
         Objects.requireNonNull(allCandidates, "allCandidates");
         Objects.requireNonNull(sourceLabel, "sourceLabel");
@@ -366,20 +381,23 @@ final class LanguageBenchmarkCorpus {
                 changedCandidates.add(entry);
             }
         }
-        if (changedCandidates.isEmpty()) {
-            throw new IllegalStateException("No changed-token benchmark corpus tokens were available for "
-                    + sourceLabel + ".");
+        if (allCandidates.isEmpty()) {
+            throw new IllegalStateException("No benchmark corpus tokens were available for " + sourceLabel + ".");
         }
 
-        final int timingTokenCount = Math.max(changedCandidates.size(), minimumTokenCount);
+        final List<Entry> timingCandidates = changedCandidates.isEmpty() ? allCandidates : changedCandidates;
+        final TimingWorkloadBasis basis = changedCandidates.isEmpty()
+                ? TimingWorkloadBasis.ROOT_ONLY_TOKENS
+                : TimingWorkloadBasis.CHANGED_TOKENS;
+        final int timingTokenCount = Math.max(timingCandidates.size(), minimumTokenCount);
         final String[] tokens = new String[timingTokenCount];
         final String[] expectedRoots = new String[timingTokenCount];
         for (int index = 0; index < tokens.length; index++) {
-            final Entry entry = changedCandidates.get(index % changedCandidates.size());
+            final Entry entry = timingCandidates.get(index % timingCandidates.size());
             tokens[index] = entry.token();
             expectedRoots[index] = entry.root();
         }
-        return new Corpus(tokens, expectedRoots);
+        return new TimingCorpus(new Corpus(tokens, expectedRoots), basis);
     }
 
     /**
@@ -531,6 +549,43 @@ final class LanguageBenchmarkCorpus {
         return false;
     }
 
+    /** Identifies the token population used by a canonical timing workload. */
+    enum TimingWorkloadBasis {
+        /** Token/root pairs whose token differs from its expected root. */
+        CHANGED_TOKENS("changed tokens"),
+
+        /** Complete root-preservation corpus used when no changed pair exists. */
+        ROOT_ONLY_TOKENS("root-only tokens");
+
+        private final String reportValue;
+
+        TimingWorkloadBasis(final String reportValue) {
+            this.reportValue = reportValue;
+        }
+
+        /**
+         * Returns the stable human-readable CSV value.
+         *
+         * @return report value
+         */
+        String reportValue() {
+            return this.reportValue;
+        }
+    }
+
+    /**
+     * Canonical timing corpus together with its auditable token-population basis.
+     *
+     * @param corpus measured token/root pairs
+     * @param basis token-population basis
+     */
+    record TimingCorpus(Corpus corpus, TimingWorkloadBasis basis) {
+        TimingCorpus {
+            Objects.requireNonNull(corpus, "corpus");
+            Objects.requireNonNull(basis, "basis");
+        }
+    }
+
     /**
      * Immutable token corpus with expected roots.
      *
diff --git a/src/jmh/java/org/egothor/stemmer/benchmark/PolishPolimorfStemmerComparisonBenchmark.java b/src/jmh/java/org/egothor/stemmer/benchmark/PolishPolimorfStemmerComparisonBenchmark.java
index 225bb6b..eba5616 100644
--- a/src/jmh/java/org/egothor/stemmer/benchmark/PolishPolimorfStemmerComparisonBenchmark.java
+++ b/src/jmh/java/org/egothor/stemmer/benchmark/PolishPolimorfStemmerComparisonBenchmark.java
@@ -143,7 +143,8 @@ public class PolishPolimorfStemmerComparisonBenchmark {
      * @throws IOException if Lucene token streaming fails
      */
     @Benchmark
-    public void polishLuceneMorfologikFilter(final SharedState sharedState, final LuceneFilterState filterState,
+    public void polishPolimorfLuceneMorfologikFilter(final SharedState sharedState,
+            final LuceneFilterState filterState,
             final Blackhole blackhole) throws IOException {
         filterState.polishMorfologik.run(sharedState.tokens, blackhole);
     }
diff --git a/src/jmh/java/org/egothor/stemmer/benchmark/QualityStemmerMatrix.java b/src/jmh/java/org/egothor/stemmer/benchmark/QualityStemmerMatrix.java
index 9bca0a0..c45ebf2 100644
--- a/src/jmh/java/org/egothor/stemmer/benchmark/QualityStemmerMatrix.java
+++ b/src/jmh/java/org/egothor/stemmer/benchmark/QualityStemmerMatrix.java
@@ -58,6 +58,7 @@ public final class QualityStemmerMatrix {
     public static List<Candidate> candidates() {
         final List<Candidate> candidates = new ArrayList<>();
         Arrays.stream(StemmerComparisonBenchmarkQuality.QualityCandidate.values())
+                .filter(candidate -> !candidate.name().startsWith("POLISH_POLIMORF_"))
                 .map(candidate -> new Candidate(candidate.name(), candidate.radixorLanguage(),
                         () -> adapt(candidate.createStemmer())))
                 .forEach(candidates::add);
diff --git a/src/jmh/java/org/egothor/stemmer/benchmark/SnowballLanguageCase.java b/src/jmh/java/org/egothor/stemmer/benchmark/SnowballLanguageCase.java
index eb382e2..8e37035 100644
--- a/src/jmh/java/org/egothor/stemmer/benchmark/SnowballLanguageCase.java
+++ b/src/jmh/java/org/egothor/stemmer/benchmark/SnowballLanguageCase.java
@@ -31,21 +31,32 @@
 package org.egothor.stemmer.benchmark;
 
 import org.egothor.stemmer.StemmerPatchTrieLoader;
+import org.egothor.stemmer.benchmark.snowball.ext.arabicStemmer;
+import org.egothor.stemmer.benchmark.snowball.ext.armenianStemmer;
+import org.egothor.stemmer.benchmark.snowball.ext.catalanStemmer;
 import org.egothor.stemmer.benchmark.snowball.ext.czechStemmer;
 import org.egothor.stemmer.benchmark.snowball.ext.danishStemmer;
 import org.egothor.stemmer.benchmark.snowball.ext.dutchStemmer;
+import org.egothor.stemmer.benchmark.snowball.ext.estonianStemmer;
 import org.egothor.stemmer.benchmark.snowball.ext.finnishStemmer;
 import org.egothor.stemmer.benchmark.snowball.ext.frenchStemmer;
 import org.egothor.stemmer.benchmark.snowball.ext.germanStemmer;
+import org.egothor.stemmer.benchmark.snowball.ext.greekStemmer;
 import org.egothor.stemmer.benchmark.snowball.ext.hungarianStemmer;
+import org.egothor.stemmer.benchmark.snowball.ext.indonesianStemmer;
+import org.egothor.stemmer.benchmark.snowball.ext.irishStemmer;
 import org.egothor.stemmer.benchmark.snowball.ext.italianStemmer;
+import org.egothor.stemmer.benchmark.snowball.ext.lithuanianStemmer;
 import org.egothor.stemmer.benchmark.snowball.ext.norwegianStemmer;
 import org.egothor.stemmer.benchmark.snowball.ext.persianStemmer;
 import org.egothor.stemmer.benchmark.snowball.ext.polishStemmer;
 import org.egothor.stemmer.benchmark.snowball.ext.portugueseStemmer;
 import org.egothor.stemmer.benchmark.snowball.ext.russianStemmer;
+import org.egothor.stemmer.benchmark.snowball.ext.romanianStemmer;
+import org.egothor.stemmer.benchmark.snowball.ext.sesothoStemmer;
 import org.egothor.stemmer.benchmark.snowball.ext.spanishStemmer;
 import org.egothor.stemmer.benchmark.snowball.ext.swedishStemmer;
+import org.egothor.stemmer.benchmark.snowball.ext.turkishStemmer;
 import org.egothor.stemmer.benchmark.snowball.ext.yiddishStemmer;
 
 /**
@@ -53,6 +64,15 @@ import org.egothor.stemmer.benchmark.snowball.ext.yiddishStemmer;
  */
 enum SnowballLanguageCase {
 
+    /** Arabic Snowball stemming over the Radixor Arabic dictionary. */
+    ARABIC("Arabic", StemmerPatchTrieLoader.Language.AR, arabicStemmer::new, "Arabic"),
+
+    /** Armenian Snowball stemming over the Radixor Armenian dictionary. */
+    ARMENIAN("Armenian", StemmerPatchTrieLoader.Language.HY_AM, armenianStemmer::new, "Armenian"),
+
+    /** Catalan Snowball stemming over the Radixor Catalan dictionary. */
+    CATALAN("Catalan", StemmerPatchTrieLoader.Language.CA_ES, catalanStemmer::new, "Catalan"),
+
     /**
      * Czech Snowball stemming over the Radixor Czech dictionary.
      */
@@ -68,6 +88,9 @@ enum SnowballLanguageCase {
      */
     DUTCH("Dutch", StemmerPatchTrieLoader.Language.NL_NL, dutchStemmer::new, "Dutch"),
 
+    /** Estonian Snowball stemming over the Radixor Estonian dictionary. */
+    ESTONIAN("Estonian", StemmerPatchTrieLoader.Language.ET_EE, estonianStemmer::new, "Estonian"),
+
     /**
      * Finnish Snowball stemming over the Radixor Finnish dictionary.
      */
@@ -83,16 +106,28 @@ enum SnowballLanguageCase {
      */
     GERMAN("German", StemmerPatchTrieLoader.Language.DE_DE, germanStemmer::new, "German"),
 
+    /** Greek Snowball stemming over the Radixor Greek dictionary. */
+    GREEK("Greek", StemmerPatchTrieLoader.Language.EL_GR, greekStemmer::new, "Greek"),
+
     /**
      * Hungarian Snowball stemming over the Radixor Hungarian dictionary.
      */
     HUNGARIAN("Hungarian", StemmerPatchTrieLoader.Language.HU_HU, hungarianStemmer::new, "Hungarian"),
 
+    /** Indonesian Snowball stemming over the Radixor Indonesian dictionary. */
+    INDONESIAN("Indonesian", StemmerPatchTrieLoader.Language.ID_ID, indonesianStemmer::new, "Indonesian"),
+
+    /** Irish Snowball stemming over the Radixor Irish dictionary. */
+    IRISH("Irish", StemmerPatchTrieLoader.Language.GA_IE, irishStemmer::new, "Irish"),
+
     /**
      * Italian Snowball stemming over the Radixor Italian dictionary.
      */
     ITALIAN("Italian", StemmerPatchTrieLoader.Language.IT_IT, italianStemmer::new, "Italian"),
 
+    /** Lithuanian Snowball stemming over the Radixor Lithuanian dictionary. */
+    LITHUANIAN("Lithuanian", StemmerPatchTrieLoader.Language.LT_LT, lithuanianStemmer::new, "Lithuanian"),
+
     /**
      * Norwegian Snowball stemming over the Radixor Bokmal dictionary.
      */
@@ -125,6 +160,12 @@ enum SnowballLanguageCase {
      */
     RUSSIAN("Russian", StemmerPatchTrieLoader.Language.RU_RU, russianStemmer::new, "Russian"),
 
+    /** Romanian Snowball stemming over the Radixor Romanian dictionary. */
+    ROMANIAN("Romanian", StemmerPatchTrieLoader.Language.RO_RO, romanianStemmer::new, "Romanian"),
+
+    /** Southern Sotho Snowball stemming over the matching Radixor dictionary. */
+    SESOTHO("Southern Sotho", StemmerPatchTrieLoader.Language.ST_ZA, sesothoStemmer::new),
+
     /**
      * Spanish Snowball stemming over the Radixor Spanish dictionary.
      */
@@ -135,6 +176,9 @@ enum SnowballLanguageCase {
      */
     SWEDISH("Swedish", StemmerPatchTrieLoader.Language.SV_SE, swedishStemmer::new, "Swedish"),
 
+    /** Turkish Snowball stemming over the Radixor Turkish dictionary. */
+    TURKISH("Turkish", StemmerPatchTrieLoader.Language.TR_TR, turkishStemmer::new, "Turkish"),
+
     /**
      * Yiddish Snowball stemming over the Radixor Yiddish dictionary.
      */
@@ -160,6 +204,9 @@ enum SnowballLanguageCase {
      */
     private final String luceneSnowballName;
 
+    /** Whether this direct case participates in the cross-language homepage aggregate. */
+    private final boolean homepageAggregate;
+
     /**
      * Creates a direct-only language case not provided by the current Lucene
      * Snowball implementation.
@@ -187,6 +234,7 @@ enum SnowballLanguageCase {
         this.radixorLanguage = radixorLanguage;
         this.directFactory = directFactory;
         this.luceneSnowballName = luceneSnowballName;
+        this.homepageAggregate = radixorLanguage != StemmerPatchTrieLoader.Language.AR;
     }
 
     /**
@@ -227,4 +275,34 @@ enum SnowballLanguageCase {
         }
         return this.luceneSnowballName;
     }
+
+    /**
+     * Reports whether the current Lucene integration exposes this exact
+     * Snowball algorithm.
+     *
+     * @return {@code true} when a Lucene SnowballFilter comparison is available
+     */
+    boolean hasLuceneSnowball() {
+        return this.luceneSnowballName != null;
+    }
+
+    /**
+     * Reports whether this official direct case is shown in the homepage chart.
+     *
+     * @return always {@code true}; every exact-language direct case is charted
+     */
+    boolean homepageChart() {
+        return true;
+    }
+
+    /**
+     * Reports whether this case contributes to the homepage geometric mean.
+     * Arabic is charted but excluded because its Radixor timing corpus is
+     * root-only and is not comparable to the changed-token cohort.
+     *
+     * @return {@code true} for comparable changed-token cases
+     */
+    boolean homepageAggregate() {
+        return this.homepageAggregate;
+    }
 }
diff --git a/src/jmh/java/org/egothor/stemmer/benchmark/SnowballLanguageStemmerComparisonBenchmark.java b/src/jmh/java/org/egothor/stemmer/benchmark/SnowballLanguageStemmerComparisonBenchmark.java
index 1cb5b90..9e4b67c 100644
--- a/src/jmh/java/org/egothor/stemmer/benchmark/SnowballLanguageStemmerComparisonBenchmark.java
+++ b/src/jmh/java/org/egothor/stemmer/benchmark/SnowballLanguageStemmerComparisonBenchmark.java
@@ -58,9 +58,10 @@ import org.openjdk.jmh.infra.Blackhole;
  * that has a matching Snowball Java stemmer.
  *
  * <p>
- * Each benchmark operation processes the same changed-token Radixor
- * dictionary-derived language corpus, repeated only when the changed-token
- * resource contains fewer than 5,000 token fields. The direct Snowball method
+ * Each benchmark operation processes the same Radixor dictionary-derived
+ * timing corpus. Changed tokens are preferred; a root-only model uses its
+ * complete root-preservation corpus. Populations smaller than 5,000 token fields
+ * are repeated deterministically. The direct Snowball method
  * measures the isolated benchmark-only Snowball source. The Lucene
  * SnowballFilter method measures Lucene's TokenStream integration path,
  * including lower-case normalization and token attribute overhead.
@@ -81,9 +82,10 @@ public class SnowballLanguageStemmerComparisonBenchmark {
         /**
          * Language/algorithm case under comparison.
          */
-        @Param({ "DANISH", "DUTCH", "FINNISH", "FRENCH", "GERMAN", "HUNGARIAN", "ITALIAN",
-                "NORWEGIAN_BOKMAL", "NORWEGIAN_NYNORSK", "PORTUGUESE", "RUSSIAN", "SPANISH", "SWEDISH",
-                "YIDDISH" })
+        @Param({ "ARABIC", "ARMENIAN", "CATALAN", "DANISH", "DUTCH", "ESTONIAN", "FINNISH",
+                "FRENCH", "GERMAN", "GREEK", "HUNGARIAN", "INDONESIAN", "IRISH", "ITALIAN", "LITHUANIAN",
+                "NORWEGIAN_BOKMAL", "NORWEGIAN_NYNORSK", "PORTUGUESE", "ROMANIAN", "RUSSIAN", "SPANISH",
+                "SWEDISH", "TURKISH", "YIDDISH" })
         public String languageCaseName;
 
         /**
@@ -92,7 +94,7 @@ public class SnowballLanguageStemmerComparisonBenchmark {
         private SnowballLanguageCase languageCase;
 
         /**
-         * Shared deterministic changed-token dictionary corpus.
+         * Shared deterministic canonical timing corpus.
          */
         private String[] tokens;
 
@@ -120,7 +122,7 @@ public class SnowballLanguageStemmerComparisonBenchmark {
      * Shared corpus state for every official direct Snowball implementation.
      *
      * <p>
-     * Czech, Persian, and Polish are available in the official Snowball 3.1.0
+     * Czech, Persian, Polish, and Sesotho are available in the official Snowball 3.1.0
      * distribution but not through the Lucene SnowballFilter version used by
      * this project. Keeping the direct parameter domain separate prevents JMH
      * from constructing unsupported Lucene workloads.
@@ -132,9 +134,10 @@ public class SnowballLanguageStemmerComparisonBenchmark {
         /**
          * Language/algorithm case under comparison.
          */
-        @Param({ "CZECH", "DANISH", "DUTCH", "FINNISH", "FRENCH", "GERMAN", "HUNGARIAN", "ITALIAN",
-                "NORWEGIAN_BOKMAL", "NORWEGIAN_NYNORSK", "PERSIAN", "POLISH", "PORTUGUESE", "RUSSIAN",
-                "SPANISH", "SWEDISH", "YIDDISH" })
+        @Param({ "ARABIC", "ARMENIAN", "CATALAN", "CZECH", "DANISH", "DUTCH", "ESTONIAN",
+                "FINNISH", "FRENCH", "GERMAN", "GREEK", "HUNGARIAN", "INDONESIAN", "IRISH", "ITALIAN",
+                "LITHUANIAN", "NORWEGIAN_BOKMAL", "NORWEGIAN_NYNORSK", "PERSIAN", "POLISH", "PORTUGUESE",
+                "ROMANIAN", "RUSSIAN", "SESOTHO", "SPANISH", "SWEDISH", "TURKISH", "YIDDISH" })
         public String languageCaseName;
 
         /**
@@ -143,7 +146,7 @@ public class SnowballLanguageStemmerComparisonBenchmark {
         private SnowballLanguageCase languageCase;
 
         /**
-         * Shared deterministic changed-token dictionary corpus.
+         * Shared deterministic canonical timing corpus.
          */
         private String[] tokens;
 
diff --git a/src/jmh/java/org/egothor/stemmer/benchmark/StemmerComparisonBenchmarkQuality.java b/src/jmh/java/org/egothor/stemmer/benchmark/StemmerComparisonBenchmarkQuality.java
index 89a296e..464608c 100644
--- a/src/jmh/java/org/egothor/stemmer/benchmark/StemmerComparisonBenchmarkQuality.java
+++ b/src/jmh/java/org/egothor/stemmer/benchmark/StemmerComparisonBenchmarkQuality.java
@@ -167,6 +167,9 @@ public class StemmerComparisonBenchmarkQuality {
                 "POLISH_LUCENE_STEMPEL_DIRECT",
                 "POLISH_LUCENE_STEMPEL_FILTER",
                 "POLISH_LUCENE_MORFOLOGIK_FILTER",
+                "POLISH_POLIMORF_RADIXOR",
+                "POLISH_POLIMORF_LUCENE_MORFOLOGIK_FILTER",
+                "POLISH_POLIMORF_SNOWBALL_DIRECT",
                 "PORTUGUESE_RADIXOR",
                 "PORTUGUESE_LUCENE_PORTUGUESE_STEM_FILTER",
                 "PORTUGUESE_LUCENE_PORTUGUESE_LIGHT_STEM_FILTER",
@@ -179,21 +182,37 @@ public class StemmerComparisonBenchmarkQuality {
                 "UKRAINIAN_RADIXOR",
                 "UKRAINIAN_MORFOLOGIK_DIRECT",
                 "UKRAINIAN_LUCENE_MORFOLOGIK_FILTER",
+                "SNOWBALL_ARABIC_DIRECT",
+                "SNOWBALL_ARABIC_LUCENE_FILTER",
+                "SNOWBALL_ARMENIAN_DIRECT",
+                "SNOWBALL_ARMENIAN_LUCENE_FILTER",
+                "SNOWBALL_CATALAN_DIRECT",
+                "SNOWBALL_CATALAN_LUCENE_FILTER",
                 "SNOWBALL_CZECH_DIRECT",
                 "SNOWBALL_DANISH_DIRECT",
                 "SNOWBALL_DANISH_LUCENE_FILTER",
                 "SNOWBALL_DUTCH_DIRECT",
                 "SNOWBALL_DUTCH_LUCENE_FILTER",
+                "SNOWBALL_ESTONIAN_DIRECT",
+                "SNOWBALL_ESTONIAN_LUCENE_FILTER",
                 "SNOWBALL_FINNISH_DIRECT",
                 "SNOWBALL_FINNISH_LUCENE_FILTER",
                 "SNOWBALL_FRENCH_DIRECT",
                 "SNOWBALL_FRENCH_LUCENE_FILTER",
                 "SNOWBALL_GERMAN_DIRECT",
                 "SNOWBALL_GERMAN_LUCENE_FILTER",
+                "SNOWBALL_GREEK_DIRECT",
+                "SNOWBALL_GREEK_LUCENE_FILTER",
                 "SNOWBALL_HUNGARIAN_DIRECT",
                 "SNOWBALL_HUNGARIAN_LUCENE_FILTER",
+                "SNOWBALL_INDONESIAN_DIRECT",
+                "SNOWBALL_INDONESIAN_LUCENE_FILTER",
+                "SNOWBALL_IRISH_DIRECT",
+                "SNOWBALL_IRISH_LUCENE_FILTER",
                 "SNOWBALL_ITALIAN_DIRECT",
                 "SNOWBALL_ITALIAN_LUCENE_FILTER",
+                "SNOWBALL_LITHUANIAN_DIRECT",
+                "SNOWBALL_LITHUANIAN_LUCENE_FILTER",
                 "SNOWBALL_NORWEGIAN_BOKMAL_DIRECT",
                 "SNOWBALL_NORWEGIAN_BOKMAL_LUCENE_FILTER",
                 "SNOWBALL_NORWEGIAN_NYNORSK_DIRECT",
@@ -204,10 +223,15 @@ public class StemmerComparisonBenchmarkQuality {
                 "SNOWBALL_PORTUGUESE_LUCENE_FILTER",
                 "SNOWBALL_RUSSIAN_DIRECT",
                 "SNOWBALL_RUSSIAN_LUCENE_FILTER",
+                "SNOWBALL_ROMANIAN_DIRECT",
+                "SNOWBALL_ROMANIAN_LUCENE_FILTER",
+                "SNOWBALL_SESOTHO_DIRECT",
                 "SNOWBALL_SPANISH_DIRECT",
                 "SNOWBALL_SPANISH_LUCENE_FILTER",
                 "SNOWBALL_SWEDISH_DIRECT",
                 "SNOWBALL_SWEDISH_LUCENE_FILTER",
+                "SNOWBALL_TURKISH_DIRECT",
+                "SNOWBALL_TURKISH_LUCENE_FILTER",
                 "SNOWBALL_YIDDISH_DIRECT",
                 "SNOWBALL_YIDDISH_LUCENE_FILTER"
         })
@@ -231,7 +255,9 @@ public class StemmerComparisonBenchmarkQuality {
         @Setup(Level.Trial)
         public void setUp() throws IOException {
             final QualityCandidate candidate = QualityCandidate.valueOf(this.candidateName);
-            this.corpus = LanguageBenchmarkCorpus.createFullCorpus(candidate.radixorLanguage());
+            this.corpus = candidate.corpusModelId == null
+                    ? LanguageBenchmarkCorpus.createFullCorpus(candidate.radixorLanguage())
+                    : LanguageBenchmarkCorpus.createFullCorpus(candidate.corpusModelId);
             this.evaluator = candidate.createEvaluator();
         }
     }
@@ -356,6 +382,10 @@ public class StemmerComparisonBenchmarkQuality {
         POLISH_LUCENE_STEMPEL_DIRECT(StemmerPatchTrieLoader.Language.PL_PL),
         POLISH_LUCENE_STEMPEL_FILTER(StemmerPatchTrieLoader.Language.PL_PL),
         POLISH_LUCENE_MORFOLOGIK_FILTER(StemmerPatchTrieLoader.Language.PL_PL),
+        POLISH_POLIMORF_RADIXOR(StemmerPatchTrieLoader.Language.PL_PL, "pl-pl-polimorf"),
+        POLISH_POLIMORF_LUCENE_MORFOLOGIK_FILTER(StemmerPatchTrieLoader.Language.PL_PL, "pl-pl-polimorf"),
+        POLISH_POLIMORF_SNOWBALL_DIRECT(StemmerPatchTrieLoader.Language.PL_PL,
+                SnowballLanguageCase.POLISH, "pl-pl-polimorf"),
         PORTUGUESE_RADIXOR(StemmerPatchTrieLoader.Language.PT_PT),
         PORTUGUESE_LUCENE_PORTUGUESE_STEM_FILTER(StemmerPatchTrieLoader.Language.PT_PT),
         PORTUGUESE_LUCENE_PORTUGUESE_LIGHT_STEM_FILTER(StemmerPatchTrieLoader.Language.PT_PT),
@@ -368,21 +398,37 @@ public class StemmerComparisonBenchmarkQuality {
         UKRAINIAN_RADIXOR(StemmerPatchTrieLoader.Language.UK_UA),
         UKRAINIAN_MORFOLOGIK_DIRECT(StemmerPatchTrieLoader.Language.UK_UA),
         UKRAINIAN_LUCENE_MORFOLOGIK_FILTER(StemmerPatchTrieLoader.Language.UK_UA),
+        SNOWBALL_ARABIC_DIRECT(StemmerPatchTrieLoader.Language.AR, SnowballLanguageCase.ARABIC),
+        SNOWBALL_ARABIC_LUCENE_FILTER(StemmerPatchTrieLoader.Language.AR, SnowballLanguageCase.ARABIC),
+        SNOWBALL_ARMENIAN_DIRECT(StemmerPatchTrieLoader.Language.HY_AM, SnowballLanguageCase.ARMENIAN),
+        SNOWBALL_ARMENIAN_LUCENE_FILTER(StemmerPatchTrieLoader.Language.HY_AM, SnowballLanguageCase.ARMENIAN),
+        SNOWBALL_CATALAN_DIRECT(StemmerPatchTrieLoader.Language.CA_ES, SnowballLanguageCase.CATALAN),
+        SNOWBALL_CATALAN_LUCENE_FILTER(StemmerPatchTrieLoader.Language.CA_ES, SnowballLanguageCase.CATALAN),
         SNOWBALL_CZECH_DIRECT(StemmerPatchTrieLoader.Language.CS_CZ, SnowballLanguageCase.CZECH),
         SNOWBALL_DANISH_DIRECT(StemmerPatchTrieLoader.Language.DA_DK, SnowballLanguageCase.DANISH),
         SNOWBALL_DANISH_LUCENE_FILTER(StemmerPatchTrieLoader.Language.DA_DK, SnowballLanguageCase.DANISH),
         SNOWBALL_DUTCH_DIRECT(StemmerPatchTrieLoader.Language.NL_NL, SnowballLanguageCase.DUTCH),
         SNOWBALL_DUTCH_LUCENE_FILTER(StemmerPatchTrieLoader.Language.NL_NL, SnowballLanguageCase.DUTCH),
+        SNOWBALL_ESTONIAN_DIRECT(StemmerPatchTrieLoader.Language.ET_EE, SnowballLanguageCase.ESTONIAN),
+        SNOWBALL_ESTONIAN_LUCENE_FILTER(StemmerPatchTrieLoader.Language.ET_EE, SnowballLanguageCase.ESTONIAN),
         SNOWBALL_FINNISH_DIRECT(StemmerPatchTrieLoader.Language.FI_FI, SnowballLanguageCase.FINNISH),
         SNOWBALL_FINNISH_LUCENE_FILTER(StemmerPatchTrieLoader.Language.FI_FI, SnowballLanguageCase.FINNISH),
         SNOWBALL_FRENCH_DIRECT(StemmerPatchTrieLoader.Language.FR_FR, SnowballLanguageCase.FRENCH),
         SNOWBALL_FRENCH_LUCENE_FILTER(StemmerPatchTrieLoader.Language.FR_FR, SnowballLanguageCase.FRENCH),
         SNOWBALL_GERMAN_DIRECT(StemmerPatchTrieLoader.Language.DE_DE, SnowballLanguageCase.GERMAN),
         SNOWBALL_GERMAN_LUCENE_FILTER(StemmerPatchTrieLoader.Language.DE_DE, SnowballLanguageCase.GERMAN),
+        SNOWBALL_GREEK_DIRECT(StemmerPatchTrieLoader.Language.EL_GR, SnowballLanguageCase.GREEK),
+        SNOWBALL_GREEK_LUCENE_FILTER(StemmerPatchTrieLoader.Language.EL_GR, SnowballLanguageCase.GREEK),
         SNOWBALL_HUNGARIAN_DIRECT(StemmerPatchTrieLoader.Language.HU_HU, SnowballLanguageCase.HUNGARIAN),
         SNOWBALL_HUNGARIAN_LUCENE_FILTER(StemmerPatchTrieLoader.Language.HU_HU, SnowballLanguageCase.HUNGARIAN),
+        SNOWBALL_INDONESIAN_DIRECT(StemmerPatchTrieLoader.Language.ID_ID, SnowballLanguageCase.INDONESIAN),
+        SNOWBALL_INDONESIAN_LUCENE_FILTER(StemmerPatchTrieLoader.Language.ID_ID, SnowballLanguageCase.INDONESIAN),
+        SNOWBALL_IRISH_DIRECT(StemmerPatchTrieLoader.Language.GA_IE, SnowballLanguageCase.IRISH),
+        SNOWBALL_IRISH_LUCENE_FILTER(StemmerPatchTrieLoader.Language.GA_IE, SnowballLanguageCase.IRISH),
         SNOWBALL_ITALIAN_DIRECT(StemmerPatchTrieLoader.Language.IT_IT, SnowballLanguageCase.ITALIAN),
         SNOWBALL_ITALIAN_LUCENE_FILTER(StemmerPatchTrieLoader.Language.IT_IT, SnowballLanguageCase.ITALIAN),
+        SNOWBALL_LITHUANIAN_DIRECT(StemmerPatchTrieLoader.Language.LT_LT, SnowballLanguageCase.LITHUANIAN),
+        SNOWBALL_LITHUANIAN_LUCENE_FILTER(StemmerPatchTrieLoader.Language.LT_LT, SnowballLanguageCase.LITHUANIAN),
         SNOWBALL_NORWEGIAN_BOKMAL_DIRECT(StemmerPatchTrieLoader.Language.NB_NO,
                 SnowballLanguageCase.NORWEGIAN_BOKMAL),
         SNOWBALL_NORWEGIAN_BOKMAL_LUCENE_FILTER(StemmerPatchTrieLoader.Language.NB_NO,
@@ -397,10 +443,15 @@ public class StemmerComparisonBenchmarkQuality {
         SNOWBALL_PORTUGUESE_LUCENE_FILTER(StemmerPatchTrieLoader.Language.PT_PT, SnowballLanguageCase.PORTUGUESE),
         SNOWBALL_RUSSIAN_DIRECT(StemmerPatchTrieLoader.Language.RU_RU, SnowballLanguageCase.RUSSIAN),
         SNOWBALL_RUSSIAN_LUCENE_FILTER(StemmerPatchTrieLoader.Language.RU_RU, SnowballLanguageCase.RUSSIAN),
+        SNOWBALL_ROMANIAN_DIRECT(StemmerPatchTrieLoader.Language.RO_RO, SnowballLanguageCase.ROMANIAN),
+        SNOWBALL_ROMANIAN_LUCENE_FILTER(StemmerPatchTrieLoader.Language.RO_RO, SnowballLanguageCase.ROMANIAN),
+        SNOWBALL_SESOTHO_DIRECT(StemmerPatchTrieLoader.Language.ST_ZA, SnowballLanguageCase.SESOTHO),
         SNOWBALL_SPANISH_DIRECT(StemmerPatchTrieLoader.Language.ES_ES, SnowballLanguageCase.SPANISH),
         SNOWBALL_SPANISH_LUCENE_FILTER(StemmerPatchTrieLoader.Language.ES_ES, SnowballLanguageCase.SPANISH),
         SNOWBALL_SWEDISH_DIRECT(StemmerPatchTrieLoader.Language.SV_SE, SnowballLanguageCase.SWEDISH),
         SNOWBALL_SWEDISH_LUCENE_FILTER(StemmerPatchTrieLoader.Language.SV_SE, SnowballLanguageCase.SWEDISH),
+        SNOWBALL_TURKISH_DIRECT(StemmerPatchTrieLoader.Language.TR_TR, SnowballLanguageCase.TURKISH),
+        SNOWBALL_TURKISH_LUCENE_FILTER(StemmerPatchTrieLoader.Language.TR_TR, SnowballLanguageCase.TURKISH),
         SNOWBALL_YIDDISH_DIRECT(StemmerPatchTrieLoader.Language.YI, SnowballLanguageCase.YIDDISH),
         SNOWBALL_YIDDISH_LUCENE_FILTER(StemmerPatchTrieLoader.Language.YI, SnowballLanguageCase.YIDDISH);
 
@@ -414,13 +465,26 @@ public class StemmerComparisonBenchmarkQuality {
          */
         private final SnowballLanguageCase snowballLanguageCase;
 
+        /** Exact non-default corpus model, or {@code null} for the language default. */
+        private final String corpusModelId;
+
         /**
          * Creates a candidate.
          *
          * @param radixorLanguage Radixor dictionary language
          */
         QualityCandidate(final StemmerPatchTrieLoader.Language radixorLanguage) {
-            this(radixorLanguage, null);
+            this(radixorLanguage, null, null);
+        }
+
+        /**
+         * Creates a candidate evaluated against an exact non-default model.
+         *
+         * @param radixorLanguage Radixor dictionary language
+         * @param corpusModelId exact corpus model identifier
+         */
+        QualityCandidate(final StemmerPatchTrieLoader.Language radixorLanguage, final String corpusModelId) {
+            this(radixorLanguage, null, corpusModelId);
         }
 
         /**
@@ -431,8 +495,21 @@ public class StemmerComparisonBenchmarkQuality {
          */
         QualityCandidate(final StemmerPatchTrieLoader.Language radixorLanguage,
                 final SnowballLanguageCase snowballLanguageCase) {
+            this(radixorLanguage, snowballLanguageCase, null);
+        }
+
+        /**
+         * Creates a Snowball candidate evaluated against an exact model.
+         *
+         * @param radixorLanguage Radixor dictionary language
+         * @param snowballLanguageCase matching Snowball case
+         * @param corpusModelId exact corpus model identifier
+         */
+        QualityCandidate(final StemmerPatchTrieLoader.Language radixorLanguage,
+                final SnowballLanguageCase snowballLanguageCase, final String corpusModelId) {
             this.radixorLanguage = radixorLanguage;
             this.snowballLanguageCase = snowballLanguageCase;
+            this.corpusModelId = corpusModelId;
         }
 
         /**
@@ -453,7 +530,9 @@ public class StemmerComparisonBenchmarkQuality {
         @SuppressWarnings("deprecation") // Lucene retains SpanishMinimalStemFilter only for compatibility benchmarking.
         CandidateStemmer createStemmer() throws IOException {
             if (name().endsWith("_RADIXOR")) {
-                return radixor(createRadixorStemmer(this.radixorLanguage));
+                return this.corpusModelId == null
+                        ? radixor(createRadixorStemmer(this.radixorLanguage))
+                        : createRadixorQualityStemmer(this.corpusModelId);
             }
             if (name().endsWith("_DIRECT") && this.snowballLanguageCase != null) {
                 return direct(this.snowballLanguageCase.createDirectStemmer()::stem);
@@ -515,7 +594,8 @@ public class StemmerComparisonBenchmarkQuality {
                 }
                 case POLISH_LUCENE_STEMPEL_FILTER ->
                     tokenFilter(input -> new StempelFilter(input, new StempelStemmer(PolishAnalyzer.getDefaultTable())));
-                case POLISH_LUCENE_MORFOLOGIK_FILTER -> tokenFilter(MorfologikFilter::new, true);
+                case POLISH_LUCENE_MORFOLOGIK_FILTER, POLISH_POLIMORF_LUCENE_MORFOLOGIK_FILTER ->
+                    tokenFilter(MorfologikFilter::new, true);
                 case PORTUGUESE_LUCENE_PORTUGUESE_STEM_FILTER ->
                     tokenFilter(input -> new PortugueseStemFilter(lowercase(input)));
                 case PORTUGUESE_LUCENE_PORTUGUESE_LIGHT_STEM_FILTER ->
diff --git a/src/main/java/org/egothor/stemmer/StemmerModelDescriptor.java b/src/main/java/org/egothor/stemmer/StemmerModelDescriptor.java
index 7f51a43..0c50d18 100644
--- a/src/main/java/org/egothor/stemmer/StemmerModelDescriptor.java
+++ b/src/main/java/org/egothor/stemmer/StemmerModelDescriptor.java
@@ -48,6 +48,7 @@ public final class StemmerModelDescriptor implements Comparable<StemmerModelDesc
     private final String displayName;
     private final String resource;
     private final boolean defaultModel;
+    private final boolean rightToLeft;
     private final String format;
     private final int formatVersion;
     private final String sha256;
@@ -57,13 +58,15 @@ public final class StemmerModelDescriptor implements Comparable<StemmerModelDesc
     /** Creates a validated immutable descriptor. */
     StemmerModelDescriptor(final String id, final String version, final StemmerPatchTrieLoader.Language language,
             final String displayName, final String resource, final boolean defaultModel, final String format,
-            final int formatVersion, final String sha256, final URL source, final ClassLoader classLoader) {
+            final int formatVersion, final String sha256, final boolean rightToLeft, final URL source,
+            final ClassLoader classLoader) {
         this.id = Objects.requireNonNull(id, "id");
         this.version = Objects.requireNonNull(version, "version");
         this.language = Objects.requireNonNull(language, "language");
         this.displayName = Objects.requireNonNull(displayName, "displayName");
         this.resource = Objects.requireNonNull(resource, "resource");
         this.defaultModel = defaultModel;
+        this.rightToLeft = rightToLeft;
         this.format = Objects.requireNonNull(format, "format");
         this.formatVersion = formatVersion;
         this.sha256 = Objects.requireNonNull(sha256, "sha256");
@@ -88,6 +91,15 @@ public final class StemmerModelDescriptor implements Comparable<StemmerModelDesc
      * mapping.
      */
     public boolean isDefaultModel() { return this.defaultModel; }
+    /**
+     * Returns whether the model language is conventionally written right-to-left.
+     *
+     * <p>This presentation metadata does not affect dictionary parsing or trie
+     * traversal.</p>
+     *
+     * @return {@code true} for a right-to-left language; {@code false} otherwise
+     */
+    public boolean isRightToLeft() { return this.rightToLeft; }
     /** Returns the dictionary format identifier. */
     public String format() { return this.format; }
     /** Returns the dictionary format version. */
diff --git a/src/main/java/org/egothor/stemmer/StemmerModelRegistry.java b/src/main/java/org/egothor/stemmer/StemmerModelRegistry.java
index 7aa7774..0aee9bf 100644
--- a/src/main/java/org/egothor/stemmer/StemmerModelRegistry.java
+++ b/src/main/java/org/egothor/stemmer/StemmerModelRegistry.java
@@ -32,7 +32,6 @@ package org.egothor.stemmer;
 
 import java.io.BufferedReader;
 import java.io.IOException;
-import java.io.InputStream;
 import java.io.InputStreamReader;
 import java.net.URL;
 import java.nio.charset.StandardCharsets;
@@ -198,9 +197,11 @@ public final class StemmerModelRegistry {
     /** Parses and validates one immutable descriptor. */
     private static StemmerModelDescriptor readDescriptor(final URL source, final ClassLoader classLoader) throws IOException {
         final Properties properties = new Properties();
-        try (InputStream input = source.openStream()) { properties.load(input); }
+        try (InputStreamReader reader = new InputStreamReader(source.openStream(), StandardCharsets.UTF_8)) {
+            properties.load(reader);
+        }
         final String id = required(properties, "model.id", source);
-        if (!id.matches("[a-z]{2}(?:-[a-z]{2})?-[a-z0-9]+(?:-[a-z0-9]+)*")) throw new StemmerModelIntegrityException("Invalid model.id '" + id + "' at " + source);
+        if (!id.matches("[a-z]{2,3}(?:-[a-z]{2})?-[a-z0-9]+(?:-[a-z0-9]+)*")) throw new StemmerModelIntegrityException("Invalid model.id '" + id + "' at " + source);
         final String format = required(properties, "model.format", source);
         final int version;
         try { version = Integer.parseInt(required(properties, "model.formatVersion", source)); }
@@ -214,9 +215,15 @@ public final class StemmerModelRegistry {
         if (classLoader.getResource(resource) == null) throw new StemmerModelIntegrityException("Model resource is missing: " + resource + " declared at " + source);
         final String checksum = required(properties, "model.sha256", source);
         if (!checksum.matches("[0-9a-f]{64}")) throw new StemmerModelIntegrityException("Invalid model.sha256 at " + source);
+        final boolean rightToLeft = requiredBoolean(properties, "model.rightToLeft", source);
+        if (rightToLeft != language.isRightToLeft()) {
+            throw new StemmerModelIntegrityException("Model right-to-left metadata for '" + id
+                    + "' does not match language " + language + " at " + source + ".");
+        }
         return new StemmerModelDescriptor(id, required(properties, "model.version", source), language,
                 required(properties, "model.displayName", source), resource,
-                Boolean.parseBoolean(required(properties, "model.default", source)), format, version, checksum, source, classLoader);
+                requiredBoolean(properties, "model.default", source), format, version, checksum, rightToLeft, source,
+                classLoader);
     }
 
     /** Returns a required nonblank property. */
@@ -225,4 +232,16 @@ public final class StemmerModelRegistry {
         if (value == null || value.isBlank()) throw new StemmerModelIntegrityException("Required property '" + key + "' is missing at " + source);
         return value.trim();
     }
+
+    /** Returns one required strict boolean property. */
+    private static boolean requiredBoolean(final Properties properties, final String key, final URL source) {
+        final String value = required(properties, key, source);
+        if ("true".equals(value)) {
+            return true;
+        }
+        if ("false".equals(value)) {
+            return false;
+        }
+        throw new StemmerModelIntegrityException("Property '" + key + "' must be true or false at " + source + ".");
+    }
 }
diff --git a/src/main/java/org/egothor/stemmer/StemmerPatchTrieLoader.java b/src/main/java/org/egothor/stemmer/StemmerPatchTrieLoader.java
index 5c9e673..25f4216 100644
--- a/src/main/java/org/egothor/stemmer/StemmerPatchTrieLoader.java
+++ b/src/main/java/org/egothor/stemmer/StemmerPatchTrieLoader.java
@@ -216,7 +216,622 @@ public final class StemmerPatchTrieLoader {
         /**
          * Yiddish.
          */
-        YI("yi", "yi-default", true);
+        YI("yi", "yi-default", true),
+
+        /**
+         * Adyghe.
+         */
+        ADY("ady", "ady-default", false),
+
+        /**
+         * Gulf Arabic.
+         */
+        AFB("afb", "afb-default", true),
+
+        /**
+         * Afrikaans.
+         */
+        AF_ZA("af_za", "af-za-default", false),
+
+        /**
+         * Aimele.
+         */
+        AIL("ail", "ail-default", false),
+
+        /**
+         * Akan.
+         */
+        AK("ak", "aka-default", false),
+
+        /**
+         * Yanesha’.
+         */
+        AME("ame", "ame-default", false),
+
+        /**
+         * Amharic.
+         */
+        AM_ET("am_et", "am-et-default", false),
+
+        /**
+         * Old English.
+         */
+        ANG("ang", "ang-default", false),
+
+        /**
+         * Arabic.
+         */
+        AR("ar", "ar-default", true),
+
+        /**
+         * Mapudungun.
+         */
+        ARN("arn", "arn-default", false),
+
+        /**
+         * Egyptian Arabic.
+         */
+        ARZ("arz", "arz-default", true),
+
+        /**
+         * Asturian.
+         */
+        AST("ast", "ast-default", false),
+
+        /**
+         * Assamese.
+         */
+        AS_IN("as_in", "as-in-default", false),
+
+        /**
+         * Aymara.
+         */
+        AYM("aym", "aym-default", false),
+
+        /**
+         * Amuzgo.
+         */
+        AZG("azg", "azg-default", false),
+
+        /**
+         * Azerbaijani.
+         */
+        AZ_AZ("az_az", "az-az-default", false),
+
+        /**
+         * Bashkir.
+         */
+        BAK("bak", "bak-default", false),
+
+        /**
+         * Belarusian.
+         */
+        BE_BY("be_by", "be-by-default", false),
+
+        /**
+         * Bulgarian.
+         */
+        BG_BG("bg_bg", "bg-bg-default", false),
+
+        /**
+         * Bengali.
+         */
+        BN_BD("bn_bd", "bn-bd-default", false),
+
+        /**
+         * Braj.
+         */
+        BRA("bra", "bra-default", false),
+
+        /**
+         * Breton.
+         */
+        BRE("bre", "bre-default", false),
+
+        /**
+         * Catalan.
+         */
+        CA_ES("ca_es", "ca-es-default", false),
+
+        /**
+         * Cebuano.
+         */
+        CEB("ceb", "ceb-default", false),
+
+        /**
+         * Church Slavonic.
+         */
+        CHU("chu", "chu-default", false),
+
+        /**
+         * Chukchi.
+         */
+        CKT("ckt", "ckt-default", false),
+
+        /**
+         * Eastern Chatino.
+         */
+        CLY("cly", "cly-default", false),
+
+        /**
+         * Ashaninka.
+         */
+        CNI("cni", "cni-default", false),
+
+        /**
+         * Cornish.
+         */
+        COR("cor", "cor-default", false),
+
+        /**
+         * Copala Triqui.
+         */
+        CPA("cpa", "cpa-default", false),
+
+        /**
+         * Cree.
+         */
+        CRE("cre", "cre-default", false),
+
+        /**
+         * Crimean Tatar.
+         */
+        CRH("crh", "crh-default", false),
+
+        /**
+         * Kashubian.
+         */
+        CSB("csb", "csb-default", false),
+
+        /**
+         * Western Highland Chatino.
+         */
+        CTP("ctp", "ctp-default", false),
+
+        /**
+         * Zenzontepec Chatino.
+         */
+        CZN("czn", "czn-default", false),
+
+        /**
+         * Dakota.
+         */
+        DAK("dak", "dak-default", false),
+
+        /**
+         * Zarma.
+         */
+        DJE("dje", "dje-default", false),
+
+        /**
+         * Lower Sorbian.
+         */
+        DSB("dsb", "dsb-default", false),
+
+        /**
+         * Greek.
+         */
+        EL_GR("el_gr", "el-gr-default", false),
+
+        /**
+         * Estonian.
+         */
+        ET_EE("et_ee", "et-ee-default", false),
+
+        /**
+         * Evenki.
+         */
+        EVN("evn", "evn-default", false),
+
+        /**
+         * Faroese.
+         */
+        FO_FO("fo_fo", "fo-fo-default", false),
+
+        /**
+         * Middle French.
+         */
+        FRM("frm", "frm-default", false),
+
+        /**
+         * Old French.
+         */
+        FRO("fro", "fro-default", false),
+
+        /**
+         * North Frisian.
+         */
+        FRR("frr", "frr-default", false),
+
+        /**
+         * Friulian.
+         */
+        FUR("fur", "fur-default", false),
+
+        /**
+         * Ga.
+         */
+        GAA("gaa", "gaa-default", false),
+
+        /**
+         * Galolen.
+         */
+        GAL("gal", "gal-default", false),
+
+        /**
+         * Irish.
+         */
+        GA_IE("ga_ie", "ga-ie-default", false),
+
+        /**
+         * Middle High German.
+         */
+        GMH("gmh", "gmh-default", false),
+
+        /**
+         * Middle Low German.
+         */
+        GML("gml", "gml-default", false),
+
+        /**
+         * Old High German.
+         */
+        GOH("goh", "goh-default", false),
+
+        /**
+         * Gothic.
+         */
+        GOT("got", "got-default", false),
+
+        /**
+         * Ancient Greek.
+         */
+        GRC("grc", "grc-default", false),
+
+        /**
+         * Alsatian.
+         */
+        GSW("gsw", "gsw-default", false),
+
+        /**
+         * Bininj Kun-wok.
+         */
+        GUP("gup", "gup-default", false),
+
+        /**
+         * Manx.
+         */
+        GV_IM("gv_im", "gv-im-default", false),
+
+        /**
+         * Haida.
+         */
+        HAI("hai", "hai-default", false),
+
+        /**
+         * Serbo-Croatian.
+         */
+        HBS("hbs", "hbs-default", false),
+
+        /**
+         * Hiligaynon.
+         */
+        HIL("hil", "hil-default", false),
+
+        /**
+         * Hsilimo.
+         */
+        HSI("hsi", "hsi-default", false),
+
+        /**
+         * Armenian.
+         */
+        HY_AM("hy_am", "hy-am-default", false),
+
+        /**
+         * Indonesian.
+         */
+        ID_ID("id_id", "id-id-default", false),
+
+        /**
+         * Icelandic.
+         */
+        IS_IS("is_is", "is-is-default", false),
+
+        /**
+         * Itelmen.
+         */
+        ITL("itl", "itl-default", false),
+
+        /**
+         * Ingrian.
+         */
+        IZH("izh", "izh-default", false),
+
+        /**
+         * Japanese.
+         */
+        JA_JP("ja_jp", "ja-jp-default", false),
+
+        /**
+         * Kabardian.
+         */
+        KBD("kbd", "kbd-default", false),
+
+        /**
+         * Khakas.
+         */
+        KJH("kjh", "kjh-default", false),
+
+        /**
+         * Kazakh.
+         */
+        KK_KZ("kk_kz", "kk-kz-default", false),
+
+        /**
+         * Khaling.
+         */
+        KLR("klr", "klr-default", false),
+
+        /**
+         * Kalaallisut.
+         */
+        KL_GL("kl_gl", "kl-gl-default", false),
+
+        /**
+         * Kannada.
+         */
+        KN_IN("kn_in", "kn-in-default", false),
+
+        /**
+         * Kodi.
+         */
+        KOD("kod", "kod-default", false),
+
+        /**
+         * Kongo.
+         */
+        KON("kon", "kon-default", false),
+
+        /**
+         * Karelian.
+         */
+        KRL("krl", "krl-default", false),
+
+        /**
+         * Kyrgyz.
+         */
+        KY_KG("ky_kg", "ky-kg-default", false),
+
+        /**
+         * Latin.
+         */
+        LA("la", "la-default", false),
+
+        /**
+         * Luganda.
+         */
+        LG_UG("lg_ug", "lg-ug-default", false),
+
+        /**
+         * Lingala.
+         */
+        LIN("lin", "lin-default", false),
+
+        /**
+         * Livonian.
+         */
+        LIV("liv", "liv-default", false),
+
+        /**
+         * Ladin.
+         */
+        LLD("lld", "lld-default", false),
+
+        /**
+         * Lithuanian.
+         */
+        LT_LT("lt_lt", "lt-lt-default", false),
+
+        /**
+         * Latvian.
+         */
+        LV_LV("lv_lv", "lv-lv-default", false),
+
+        /**
+         * Magahi.
+         */
+        MAG("mag", "mag-default", false),
+
+        /**
+         * Malagasy.
+         */
+        MG_MG("mg_mg", "mg-mg-default", false),
+
+        /**
+         * Maori.
+         */
+        MI_NZ("mi_nz", "mi-nz-default", false),
+
+        /**
+         * Macedonian.
+         */
+        MK_MK("mk_mk", "mk-mk-default", false),
+
+        /**
+         * Mongolian.
+         */
+        MN_MN("mn_mn", "mn-mn-default", false),
+
+        /**
+         * Maltese.
+         */
+        MT_MT("mt_mt", "mt-mt-default", false),
+
+        /**
+         * Murrinh-Patha.
+         */
+        MWF("mwf", "mwf-default", false),
+
+        /**
+         * Neapolitan.
+         */
+        NAP("nap", "nap-default", false),
+
+        /**
+         * Navajo.
+         */
+        NAV("nav", "nav-default", false),
+
+        /**
+         * Low German.
+         */
+        NDS("nds", "nds-default", false),
+
+        /**
+         * Old Norse.
+         */
+        NON("non", "non-default", false),
+
+        /**
+         * Chichewa.
+         */
+        NY_MW("ny_mw", "ny-mw-default", false),
+
+        /**
+         * Oodham.
+         */
+        OOD("ood", "ood-default", false),
+
+        /**
+         * Old Saxon.
+         */
+        OSX("osx", "osx-default", false),
+
+        /**
+         * Otomi (ote).
+         */
+        OTE("ote", "ote-default", false),
+
+        /**
+         * Pashto.
+         */
+        PS_AF("ps_af", "ps-af-default", true),
+
+        /**
+         * Quechua.
+         */
+        QUE("que", "que-default", false),
+
+        /**
+         * Romanian.
+         */
+        RO_RO("ro_ro", "ro-ro-default", false),
+
+        /**
+         * Southern Kurdish.
+         */
+        SDH("sdh", "sdh-default", true),
+
+        /**
+         * Seneca.
+         */
+        SEE("see", "see-default", false),
+
+        /**
+         * Old Irish.
+         */
+        SGA("sga", "sga-default", false),
+
+        /**
+         * Shipibo-Conibo.
+         */
+        SHP("shp", "shp-default", false),
+
+        /**
+         * Xibe.
+         */
+        SJO("sjo", "sjo-default", false),
+
+        /**
+         * Northern Sami.
+         */
+        SME("sme", "sme-default", false),
+
+        /**
+         * Shona.
+         */
+        SN_ZW("sn_zw", "sn-zw-default", false),
+
+        /**
+         * Albanian.
+         */
+        SQ_AL("sq_al", "sq-al-default", false),
+
+        /**
+         * Sotho, Southern.
+         */
+        ST_ZA("st_za", "st-za-default", false),
+
+        /**
+         * Congo Swahili.
+         */
+        SWC("swc", "swc-default", false),
+
+        /**
+         * Classical Syriac.
+         */
+        SYC("syc", "syc-default", true),
+
+        /**
+         * Tagalog.
+         */
+        TL_PH("tl_ph", "tl-ph-default", false),
+
+        /**
+         * Turkish.
+         */
+        TR_TR("tr_tr", "tr-tr-default", false),
+
+        /**
+         * Uyghur.
+         */
+        UG_CN("ug_cn", "ug-cn-default", true),
+
+        /**
+         * Uzbek.
+         */
+        UZ_UZ("uz_uz", "uz-uz-default", false),
+
+        /**
+         * Voro.
+         */
+        VRO("vro", "vro-default", false),
+
+        /**
+         * Classical Armenian.
+         */
+        XCL("xcl", "xcl-default", false),
+
+        /**
+         * Anglo-Norman.
+         */
+        XNO("xno", "xno-default", false),
+
+        /**
+         * Yoloxóchitl Mixtec.
+         */
+        XTY("xty", "xty-default", false),
+
+        /**
+         * Chichicapan Zapotec.
+         */
+        ZPV("zpv", "zpv-default", false),
+
+        /**
+         * Zulu.
+         */
+        ZU_ZA("zu_za", "zu-za-default", false);
 
         /**
          * Resource directory name.
diff --git a/src/test/java/org/egothor/stemmer/StemmerModelRegistryTest.java b/src/test/java/org/egothor/stemmer/StemmerModelRegistryTest.java
index 5d10192..e762c9a 100644
--- a/src/test/java/org/egothor/stemmer/StemmerModelRegistryTest.java
+++ b/src/test/java/org/egothor/stemmer/StemmerModelRegistryTest.java
@@ -55,6 +55,10 @@ final class StemmerModelRegistryTest {
                 polish.stream().map(StemmerModelDescriptor::id).toList());
         assertEquals("pl-pl-unimorph", registry.requireDefault(StemmerPatchTrieLoader.Language.PL_PL).id());
         assertEquals("pl-pl-polimorf", registry.require("pl-pl-polimorf").id());
+        assertEquals("Hsilimo — UniMorph", registry.require("hsi-default").displayName());
+        assertEquals(StemmerPatchTrieLoader.Language.values().length + 1, registry.models().size());
+        assertTrue(registry.models().stream().allMatch(
+                descriptor -> descriptor.isRightToLeft() == descriptor.language().isRightToLeft()));
         assertEquals(registry.models().stream().sorted().toList(), registry.models());
     }
 
diff --git a/src/test/java/org/egothor/stemmer/StemmerPatchTrieLoaderTest.java b/src/test/java/org/egothor/stemmer/StemmerPatchTrieLoaderTest.java
index db94093..66c74dc 100644
--- a/src/test/java/org/egothor/stemmer/StemmerPatchTrieLoaderTest.java
+++ b/src/test/java/org/egothor/stemmer/StemmerPatchTrieLoaderTest.java
@@ -57,6 +57,7 @@ import java.util.LinkedHashMap;
 import java.util.LinkedHashSet;
 import java.util.Map;
 import java.util.Objects;
+import java.util.Properties;
 import java.util.Set;
 import java.util.stream.IntStream;
 import java.util.stream.Stream;
@@ -1010,24 +1011,23 @@ final class StemmerPatchTrieLoaderTest {
          */
         @Test
         @DisplayName("Language enum must expose the current bundled language set")
-        void shouldExposeCurrentBundledLanguageSet() {
-            final Set<StemmerPatchTrieLoader.Language> expectedLanguages = new LinkedHashSet<StemmerPatchTrieLoader.Language>(
-                    Arrays.asList(StemmerPatchTrieLoader.Language.CS_CZ, StemmerPatchTrieLoader.Language.DA_DK,
-                            StemmerPatchTrieLoader.Language.DE_DE, StemmerPatchTrieLoader.Language.ES_ES,
-                            StemmerPatchTrieLoader.Language.FA_IR, StemmerPatchTrieLoader.Language.FI_FI,
-                            StemmerPatchTrieLoader.Language.FR_FR, StemmerPatchTrieLoader.Language.HE_IL,
-                            StemmerPatchTrieLoader.Language.HU_HU, StemmerPatchTrieLoader.Language.IT_IT,
-                            StemmerPatchTrieLoader.Language.NB_NO, StemmerPatchTrieLoader.Language.NL_NL,
-                            StemmerPatchTrieLoader.Language.NN_NO, StemmerPatchTrieLoader.Language.PL_PL,
-                            StemmerPatchTrieLoader.Language.PT_PT, StemmerPatchTrieLoader.Language.RU_RU,
-                            StemmerPatchTrieLoader.Language.SV_SE, StemmerPatchTrieLoader.Language.UK_UA,
-                            StemmerPatchTrieLoader.Language.US_UK, StemmerPatchTrieLoader.Language.YI));
-
+        void shouldExposeCurrentBundledLanguageSet() throws IOException {
             final Set<StemmerPatchTrieLoader.Language> actualLanguages = new LinkedHashSet<StemmerPatchTrieLoader.Language>(
                     Arrays.asList(StemmerPatchTrieLoader.Language.values()));
-
-            assertEquals(expectedLanguages, actualLanguages,
-                    "The bundled language enum must match the project's supported language set exactly.");
+            final Properties topology = new Properties();
+            try (InputStream input = Files.newInputStream(Path.of("models", "model-projects.properties"))) {
+                topology.load(input);
+            }
+            final long expectedDefaults = topology.stringPropertyNames().stream()
+                    .filter(modelId -> !"optional".equals(topology.getProperty(modelId)))
+                    .count();
+
+            assertEquals(expectedDefaults, actualLanguages.size());
+            assertEquals(expectedDefaults, actualLanguages.stream().map(StemmerPatchTrieLoader.Language::defaultModelId)
+                    .distinct().count());
+            assertTrue(actualLanguages.containsAll(Set.of(StemmerPatchTrieLoader.Language.AR,
+                    StemmerPatchTrieLoader.Language.HY_AM, StemmerPatchTrieLoader.Language.ST_ZA,
+                    StemmerPatchTrieLoader.Language.ZU_ZA)));
         }
 
         /**
@@ -1038,7 +1038,11 @@ final class StemmerPatchTrieLoaderTest {
         @DisplayName("Language enum must mark right-to-left bundled languages correctly")
         void shouldExposeCorrectRightToLeftMetadata() {
             final Set<StemmerPatchTrieLoader.Language> expectedRightToLeftLanguages = Set.of(
-                    StemmerPatchTrieLoader.Language.FA_IR, StemmerPatchTrieLoader.Language.HE_IL,
+                    StemmerPatchTrieLoader.Language.AFB, StemmerPatchTrieLoader.Language.AR,
+                    StemmerPatchTrieLoader.Language.ARZ, StemmerPatchTrieLoader.Language.FA_IR,
+                    StemmerPatchTrieLoader.Language.HE_IL, StemmerPatchTrieLoader.Language.PS_AF,
+                    StemmerPatchTrieLoader.Language.SDH, StemmerPatchTrieLoader.Language.SYC,
+                    StemmerPatchTrieLoader.Language.UG_CN,
                     StemmerPatchTrieLoader.Language.YI);
 
             for (StemmerPatchTrieLoader.Language language : StemmerPatchTrieLoader.Language.values()) {
diff --git a/src/test/java/org/egothor/stemmer/benchmark/LanguageBenchmarkCorpusTest.java b/src/test/java/org/egothor/stemmer/benchmark/LanguageBenchmarkCorpusTest.java
index 022e87d..36f07b2 100644
--- a/src/test/java/org/egothor/stemmer/benchmark/LanguageBenchmarkCorpusTest.java
+++ b/src/test/java/org/egothor/stemmer/benchmark/LanguageBenchmarkCorpusTest.java
@@ -126,6 +126,31 @@ final class LanguageBenchmarkCorpusTest {
         }
     }
 
+    /**
+     * Verifies root-only dictionaries remain benchmarkable with an explicitly
+     * identified root-preservation workload.
+     *
+     * @throws Exception if reflection or resource loading fails
+     */
+    @Test
+    @DisplayName("should fall back to root-only timing for root-only models")
+    void shouldFallBackToRootOnlyTimingForRootOnlyModels() throws Exception {
+        for (String modelId : new String[] { "ar-default", "bn-bd-default" }) {
+            final Object timing = invokeCorpus("createTimingCorpus", modelId);
+            final Method corpusMethod = timing.getClass().getDeclaredMethod("corpus");
+            corpusMethod.setAccessible(true);
+            final Object corpus = corpusMethod.invoke(timing);
+            final Method basisMethod = timing.getClass().getDeclaredMethod("basis");
+            basisMethod.setAccessible(true);
+
+            assertEquals("ROOT_ONLY_TOKENS", basisMethod.invoke(timing).toString());
+            assertEquals(minimumTimingTokenCount(), tokens(corpus).length);
+            for (int index = 0; index < tokens(corpus).length; index++) {
+                assertEquals(tokens(corpus)[index], expectedRoots(corpus)[index]);
+            }
+        }
+    }
+
     /**
      * Verifies benchmark corpora are generated once per language and reused from
      * memory.
@@ -163,6 +188,21 @@ final class LanguageBenchmarkCorpusTest {
         return method.invoke(null, language);
     }
 
+    /**
+     * Invokes an exact-model corpus factory.
+     *
+     * @param methodName factory method name
+     * @param modelId exact model identifier
+     * @return corpus or timing-workload record
+     * @throws Exception if reflection or resource loading fails
+     */
+    private Object invokeCorpus(final String methodName, final String modelId) throws Exception {
+        final Class<?> type = corpusType();
+        final Method method = type.getDeclaredMethod(methodName, String.class);
+        method.setAccessible(true);
+        return method.invoke(null, modelId);
+    }
+
     /**
      * Reads corpus tokens.
      *
diff --git a/src/test/java/org/egothor/stemmer/benchmark/SnowballLanguageStemmerComparisonBenchmarkTest.java b/src/test/java/org/egothor/stemmer/benchmark/SnowballLanguageStemmerComparisonBenchmarkTest.java
index d06b3fc..7525ef0 100644
--- a/src/test/java/org/egothor/stemmer/benchmark/SnowballLanguageStemmerComparisonBenchmarkTest.java
+++ b/src/test/java/org/egothor/stemmer/benchmark/SnowballLanguageStemmerComparisonBenchmarkTest.java
@@ -31,6 +31,7 @@
 package org.egothor.stemmer.benchmark;
 
 import static org.junit.jupiter.api.Assertions.assertEquals;
+import static org.junit.jupiter.api.Assertions.assertTrue;
 import static org.junit.jupiter.api.Assertions.assertThrows;
 
 import java.lang.reflect.Field;
@@ -63,9 +64,10 @@ final class SnowballLanguageStemmerComparisonBenchmarkTest {
                 .collect(Collectors.toUnmodifiableSet());
 
         assertEquals(registeredCases, directCases);
-        assertEquals(17, directCases.size());
-        assertEquals(14, luceneCases.size());
-        assertEquals(Set.of("CZECH", "PERSIAN", "POLISH"), difference(directCases, luceneCases));
+        assertEquals(28, directCases.size());
+        assertEquals(24, luceneCases.size());
+        assertEquals(Set.of("CZECH", "PERSIAN", "POLISH", "SESOTHO"),
+                difference(directCases, luceneCases));
 
         for (String luceneCase : luceneCases) {
             SnowballLanguageCase.valueOf(luceneCase).luceneSnowballName();
@@ -76,6 +78,16 @@ final class SnowballLanguageStemmerComparisonBenchmarkTest {
         }
     }
 
+    /** Verifies the machine-readable catalog preserves the active 28/27 homepage policy. */
+    @Test
+    void catalogChartsAllSupportedDirectCasesAndExcludesArabicFromAggregate() {
+        final String catalog = SnowballLanguageCatalogApplication.render();
+
+        assertEquals(29, catalog.lines().count());
+        assertEquals(27, catalog.lines().filter(line -> line.endsWith(",true")).count());
+        assertTrue(catalog.contains("ARABIC,AR,ar-default,Arabic,true,true,true,false"));
+    }
+
     /**
      * Reads the declared JMH parameter values from a benchmark state.
      *
diff --git a/src/test/java/org/egothor/stemmer/benchmark/generalization/DictionaryGeneralizationApplication.java b/src/test/java/org/egothor/stemmer/benchmark/generalization/DictionaryGeneralizationApplication.java
index 21039af..13e9cb8 100644
--- a/src/test/java/org/egothor/stemmer/benchmark/generalization/DictionaryGeneralizationApplication.java
+++ b/src/test/java/org/egothor/stemmer/benchmark/generalization/DictionaryGeneralizationApplication.java
@@ -84,7 +84,9 @@ public final class DictionaryGeneralizationApplication {
         0x61c8864680b583ebL, 0x243f6a8885a308d3L,
         0xd1b54a32d192ed03L
     };
-    private static final int ARGUMENT_COUNT = 5;
+    private static final int REQUIRED_ARGUMENT_COUNT = 5;
+    private static final String EXACT_MODEL_ARGUMENT = "--exact-model";
+    private static final int IMMUTABLE_STANDARD_DEFAULT_COUNT = 20;
     private static final String HEADER = String.join(",",
             "protocol_version", "radixor_java_version", "source_revision", "source_state",
             "generator_sha256", "language", "model_id", "model_version", "model_sha256",
@@ -104,13 +106,15 @@ public final class DictionaryGeneralizationApplication {
      * Generates the complete report.
      *
      * @param arguments output CSV path, Radixor Java version, source revision,
-     *                  source state, and generator source path
+     *                  source state, generator source path, and optionally a
+     *                  UTF-8 file containing one selected default model ID per line,
+     *                  or {@code --exact-model} followed by one default model ID
      * @throws IOException when a model dictionary cannot be read or the report
      *                     cannot be written
      */
     public static void main(final String[] arguments) throws IOException {
-        if (arguments.length != ARGUMENT_COUNT) {
-            throw new IllegalArgumentException("Expected output CSV path, Radixor Java version, source revision, source state, and generator source path.");
+        if (arguments.length < REQUIRED_ARGUMENT_COUNT || arguments.length > REQUIRED_ARGUMENT_COUNT + 2) {
+            throw new IllegalArgumentException("Expected output CSV path, Radixor Java version, source revision, source state, generator source path, and optionally a selected-default model file or --exact-model plus one default model ID.");
         }
         final Path output = Path.of(arguments[0]);
         final String javaVersion = requireText(arguments[1], "Radixor Java version");
@@ -120,8 +124,18 @@ public final class DictionaryGeneralizationApplication {
         final List<String> lines = new ArrayList<>();
         lines.add(HEADER);
         final StemmerModelRegistry registry = StemmerModelRegistry.fromContextClassLoader();
-        for (StemmerPatchTrieLoader.Language language : StemmerPatchTrieLoader.Language.values()) {
-            final StemmerModelDescriptor descriptor = registry.requireDefault(language);
+        final List<StemmerModelDescriptor> descriptors;
+        if (arguments.length == REQUIRED_ARGUMENT_COUNT) {
+            descriptors = allDefaultDescriptors(registry);
+        } else if (arguments.length == REQUIRED_ARGUMENT_COUNT + 1) {
+            descriptors = selectedStandaloneDescriptors(registry, Path.of(arguments[5]));
+        } else if (EXACT_MODEL_ARGUMENT.equals(arguments[5])) {
+            descriptors = List.of(exactDefaultDescriptor(registry, arguments[6]));
+        } else {
+            throw new IllegalArgumentException("A seven-argument invocation must use --exact-model.");
+        }
+        for (StemmerModelDescriptor descriptor : descriptors) {
+            final StemmerPatchTrieLoader.Language language = descriptor.language();
             final List<DictionaryRow> dictionaryRows = readRows(descriptor);
             verifyProductionEquivalence(language, dictionaryRows);
             for (long seed : SEEDS) {
@@ -143,6 +157,72 @@ public final class DictionaryGeneralizationApplication {
                 output.toAbsolutePath(), lines.size() - 1);
     }
 
+    /** Returns every authoritative default in stable language-enum order. */
+    static List<StemmerModelDescriptor> allDefaultDescriptors(final StemmerModelRegistry registry) {
+        Objects.requireNonNull(registry, "registry");
+        final List<StemmerModelDescriptor> descriptors = new ArrayList<>(StemmerPatchTrieLoader.Language.values().length);
+        for (StemmerPatchTrieLoader.Language language : StemmerPatchTrieLoader.Language.values()) {
+            descriptors.add(registry.requireDefault(language));
+        }
+        return List.copyOf(descriptors);
+    }
+
+    /** Returns one exact authoritative default for a targeted reproducibility run. */
+    static StemmerModelDescriptor exactDefaultDescriptor(final StemmerModelRegistry registry,
+            final String modelId) {
+        Objects.requireNonNull(registry, "registry");
+        final String requested = Objects.requireNonNull(modelId, "modelId").strip();
+        if (requested.isEmpty()) {
+            throw new IllegalArgumentException("The exact generalization model ID must not be blank.");
+        }
+        final StemmerModelDescriptor descriptor = registry.require(requested);
+        final StemmerModelDescriptor authoritative = registry.requireDefault(descriptor.language());
+        if (!authoritative.id().equals(requested)) {
+            throw new IllegalArgumentException("The exact generalization model must be an authoritative default: "
+                    + requested + ".");
+        }
+        return descriptor;
+    }
+
+    /**
+     * Reads and validates the exact standalone-default selection produced from
+     * the build topology. Blank lines and duplicate or non-default IDs are
+     * rejected so a partial measurement cannot be mistaken for the approved
+     * topology-selected standalone continuation.
+     */
+    static List<StemmerModelDescriptor> selectedStandaloneDescriptors(final StemmerModelRegistry registry,
+            final Path selectionFile) throws IOException {
+        Objects.requireNonNull(registry, "registry");
+        Objects.requireNonNull(selectionFile, "selectionFile");
+        final List<StemmerModelDescriptor> defaults = allDefaultDescriptors(registry);
+        final int expectedCount = defaults.size() - IMMUTABLE_STANDARD_DEFAULT_COUNT;
+        final List<String> modelIds = Files.readAllLines(selectionFile, StandardCharsets.UTF_8);
+        if (modelIds.size() != expectedCount || modelIds.stream().anyMatch(String::isBlank)) {
+            throw new IllegalArgumentException("The standalone selector must contain exactly "
+                    + expectedCount + " non-blank model IDs.");
+        }
+        final Set<String> selected = new HashSet<>(modelIds);
+        if (selected.size() != expectedCount) {
+            throw new IllegalArgumentException("The standalone selector contains duplicate model IDs.");
+        }
+        final Map<String, StemmerModelDescriptor> byId = new HashMap<>(defaults.size() * 2);
+        for (StemmerModelDescriptor descriptor : defaults) {
+            byId.put(descriptor.id(), descriptor);
+        }
+        final Set<String> unknown = new HashSet<>(selected);
+        unknown.removeAll(byId.keySet());
+        if (!unknown.isEmpty()) {
+            throw new IllegalArgumentException("The standalone selector contains non-default model IDs: " + unknown);
+        }
+        final List<StemmerModelDescriptor> result = new ArrayList<>(expectedCount);
+        for (StemmerModelDescriptor descriptor : defaults) {
+            if (selected.contains(descriptor.id())) {
+                result.add(descriptor);
+            }
+        }
+        return List.copyOf(result);
+    }
+
     private static Result evaluate(final StemmerPatchTrieLoader.Language language,
             final StemmerModelDescriptor descriptor, final List<DictionaryRow> rankedRows, final int percent) {
         final int selectedCount = percent == 100 ? rankedRows.size()
diff --git a/src/test/java/org/egothor/stemmer/benchmark/generalization/DictionaryGeneralizationApplicationTest.java b/src/test/java/org/egothor/stemmer/benchmark/generalization/DictionaryGeneralizationApplicationTest.java
index 776514f..5b8d65f 100644
--- a/src/test/java/org/egothor/stemmer/benchmark/generalization/DictionaryGeneralizationApplicationTest.java
+++ b/src/test/java/org/egothor/stemmer/benchmark/generalization/DictionaryGeneralizationApplicationTest.java
@@ -31,14 +31,27 @@
 package org.egothor.stemmer.benchmark.generalization;
 
 import static org.junit.jupiter.api.Assertions.assertDoesNotThrow;
+import static org.junit.jupiter.api.Assertions.assertEquals;
+import static org.junit.jupiter.api.Assertions.assertFalse;
 import static org.junit.jupiter.api.Assertions.assertThrows;
 
+import java.io.IOException;
 import java.nio.charset.StandardCharsets;
+import java.nio.file.Files;
+import java.nio.file.Path;
+import java.util.ArrayList;
+import java.util.List;
 
+import org.egothor.stemmer.StemmerModelDescriptor;
+import org.egothor.stemmer.StemmerModelRegistry;
+import org.egothor.stemmer.StemmerPatchTrieLoader;
 import org.junit.jupiter.api.Test;
+import org.junit.jupiter.api.io.TempDir;
 
 /** Tests publication-critical model-byte integrity checks. */
 class DictionaryGeneralizationApplicationTest {
+    @TempDir
+    Path temporaryDirectory;
 
     @Test
     void acceptsMatchingResourceDigest() {
@@ -53,4 +66,35 @@ class DictionaryGeneralizationApplicationTest {
         assertThrows(IllegalStateException.class, () -> DictionaryGeneralizationApplication.verifySha256(content,
                 "0000000000000000000000000000000000000000000000000000000000000000", "fixture"));
     }
+
+    @Test
+    void selectsExactlyTopologyStandaloneDefaults() throws IOException {
+        final List<String> selected = new ArrayList<>();
+        for (String line : Files.readAllLines(Path.of("models/model-projects.properties"), StandardCharsets.UTF_8)) {
+            if (line.endsWith("=standalone")) {
+                selected.add(line.substring(0, line.indexOf('=')));
+            }
+        }
+        final Path selector = this.temporaryDirectory.resolve("standalone.txt");
+        Files.write(selector, selected, StandardCharsets.UTF_8);
+        final StemmerModelRegistry registry = StemmerModelRegistry.fromContextClassLoader();
+        final List<StemmerModelDescriptor> descriptors =
+                DictionaryGeneralizationApplication.selectedStandaloneDescriptors(registry, selector);
+
+        assertEquals(selected.size(), descriptors.size());
+        assertEquals(StemmerPatchTrieLoader.Language.values().length,
+                DictionaryGeneralizationApplication.allDefaultDescriptors(registry).size());
+        assertFalse(descriptors.stream().anyMatch(descriptor -> descriptor.id().equals("pl-pl-polimorf")));
+        assertFalse(descriptors.stream().anyMatch(descriptor -> descriptor.id().equals("us-uk-default")));
+    }
+
+    @Test
+    void selectsOneExactAuthoritativeDefaultAndRejectsOptionalModel() throws IOException {
+        final StemmerModelRegistry registry = StemmerModelRegistry.fromContextClassLoader();
+
+        assertEquals("ar-default",
+                DictionaryGeneralizationApplication.exactDefaultDescriptor(registry, "ar-default").id());
+        assertThrows(IllegalArgumentException.class,
+                () -> DictionaryGeneralizationApplication.exactDefaultDescriptor(registry, "pl-pl-polimorf"));
+    }
 }
diff --git a/src/test/java/org/egothor/stemmer/benchmark/quality/QualityStemmerMatrixTest.java b/src/test/java/org/egothor/stemmer/benchmark/quality/QualityStemmerMatrixTest.java
index c6cdf3e..4a706c4 100644
--- a/src/test/java/org/egothor/stemmer/benchmark/quality/QualityStemmerMatrixTest.java
+++ b/src/test/java/org/egothor/stemmer/benchmark/quality/QualityStemmerMatrixTest.java
@@ -59,7 +59,10 @@ final class QualityStemmerMatrixTest {
     @Test @DisplayName("Candidate discovery is derived from every JMH quality candidate")
     void discoversEveryCandidate() {
         final List<Candidate> candidates = QualityStemmerMatrix.candidates();
-        assertEquals(102, candidates.size(), "The current adapter-language matrix size changed; report coverage must be reviewed.");
+        assertEquals(Language.values().length, candidates.stream()
+                .filter(candidate -> candidate.name().endsWith("_RADIXOR"))
+                .filter(candidate -> !candidate.resultLanguage().equals("pl-pl-polimorf"))
+                .map(Candidate::language).distinct().count());
         assertTrue(candidates.stream().anyMatch(candidate -> !candidate.name().endsWith("_RADIXOR")));
         assertTrue(candidates.stream().anyMatch(candidate -> candidate.name().equals("DA_DK_RADIXOR")));
         assertTrue(candidates.stream().anyMatch(candidate -> candidate.name().equals("YI_RADIXOR")));
@@ -77,7 +80,9 @@ final class QualityStemmerMatrixTest {
     void completePublicationSelectionUsesOnlyDefaultModels() {
         final List<Candidate> candidates = StemmingQualityApplication.selectCandidates(
                 EnumSet.allOf(Language.class), "");
-        assertEquals(95, candidates.size());
+        assertEquals(Language.values().length, candidates.stream()
+                .filter(candidate -> candidate.name().endsWith("_RADIXOR"))
+                .map(Candidate::language).distinct().count());
         assertTrue(candidates.stream().allMatch(candidate ->
                 candidate.dictionaryModelId().equals(candidate.language().defaultModelId())));
         assertTrue(candidates.stream().noneMatch(candidate ->
@@ -88,7 +93,8 @@ final class QualityStemmerMatrixTest {
     @Test @DisplayName("Report rendering includes both modes for every discovered candidate")
     void reportContainsCompleteMatrix() throws Exception {
         final List<QualityResult> rows = new ArrayList<>();
-        for (Candidate candidate : QualityStemmerMatrix.candidates()) {
+        final List<Candidate> candidates = QualityStemmerMatrix.candidates();
+        for (Candidate candidate : candidates) {
             for (ProcessingMode mode : ProcessingMode.values()) {
                 rows.add(new QualityResult(candidate.name(), candidate.resultLanguage(), mode,
                         OutputPolicy.PRIMARY_OUTPUT, 1, 1, 1, 0, 1, 0, 1, 1, 1, 0, 0, 0, 0,
@@ -98,7 +104,7 @@ final class QualityStemmerMatrixTest {
         final Path report = this.temporaryDirectory.resolve("matrix.csv");
         QualityReportWriter.writeCsv(report, rows);
         final String text = Files.readString(report, StandardCharsets.UTF_8);
-        assertEquals(205, text.lines().count());
+        assertEquals(1L + 2L * candidates.size(), text.lines().count());
         for (Candidate candidate : QualityStemmerMatrix.candidates()) {
             final String prefix = "\"" + candidate.name() + "\",\"" + candidate.resultLanguage()
                     + "\",\"\",\"\",\"\",";
diff --git a/src/test/java/org/egothor/stemmer/benchmark/quality/StemmingQualityDocumentationPublisher.java b/src/test/java/org/egothor/stemmer/benchmark/quality/StemmingQualityDocumentationPublisher.java
index 0f44348..9c61628 100644
--- a/src/test/java/org/egothor/stemmer/benchmark/quality/StemmingQualityDocumentationPublisher.java
+++ b/src/test/java/org/egothor/stemmer/benchmark/quality/StemmingQualityDocumentationPublisher.java
@@ -62,9 +62,16 @@ public final class StemmingQualityDocumentationPublisher {
     private static final String OVERVIEW_START = "<!-- STEMMING-QUALITY-OVERVIEW:START -->";
     private static final String OVERVIEW_END = "<!-- STEMMING-QUALITY-OVERVIEW:END -->";
     private static final List<String> MODES = List.of("ALL_WORDS", "LOWERCASE_GROUPS_ONLY");
+    private static final Set<String> LEGACY_LANGUAGES = Set.of(
+            "CS_CZ", "DA_DK", "DE_DE", "ES_ES", "FA_IR", "FI_FI", "FR_FR", "HE_IL", "HU_HU", "IT_IT",
+            "NB_NO", "NL_NL", "NN_NO", "PL_PL", "PT_PT", "RU_RU", "SV_SE", "UK_UA", "US_UK", "YI");
+    private static final String OPTIONAL_POLIMORF = "pl-pl-polimorf";
+    private static final String ACTIVE_MANIFEST = "active-snapshots.properties";
+    private static final String LEGACY_SOURCE_SHA256 =
+            "85763189eab4d0fbb047c2d5d3554c66abf9732182bd0d8fd758d7aef680e66f";
     private static final Map<String, Integer> POLICY_ORDER = Map.of("PRIMARY_OUTPUT", 0, "ANY_CANDIDATE", 1, "ALL_CANDIDATES", 2);
-    private static final Pattern PAGE_ROW = Pattern.compile("^\\|[^|]+\\| `([^`]+)` \\| \\[([^]]+)]\\(([^)]+\\.md)\\) \\|$");
-    private static final Pattern BUILT_IN_LANGUAGE_ROW = Pattern.compile("^\\|[^|]+\\| `([^`]+)` \\|.*$");
+    private static final Pattern PAGE_ROW = Pattern.compile(
+            "^\\| ([^|]+) \\| `([^`]+)` \\|.*\\| \\[([^]]+)]\\(([^)]+\\.md)\\) \\|$");
 
     /** Prevents construction of this command-line utility. */
     private StemmingQualityDocumentationPublisher() { }
@@ -76,8 +83,8 @@ public final class StemmingQualityDocumentationPublisher {
      * @throws IOException when source or documentation access fails
      */
     public static void main(final String[] arguments) throws IOException {
-        if (arguments.length != 3) {
-            throw new IllegalArgumentException("Expected arguments: source CSV, documentation root, and update or verify mode.");
+        if (arguments.length != 4) {
+            throw new IllegalArgumentException("Expected arguments: source CSV, documentation root, update or verify mode, and active snapshot file name.");
         }
         final Path source = Path.of(arguments[0]);
         final Path documentationRoot = Path.of(arguments[1]);
@@ -86,7 +93,7 @@ public final class StemmingQualityDocumentationPublisher {
             case "verify" -> false;
             default -> throw new IllegalArgumentException("Documentation mode must be update or verify.");
         };
-        publish(source, documentationRoot, update);
+        publish(source, documentationRoot, update, arguments[3]);
     }
 
     /**
@@ -97,25 +104,37 @@ public final class StemmingQualityDocumentationPublisher {
      * @param update whether files may be replaced
      * @throws IOException when files cannot be read or written
      */
-    static void publish(final Path source, final Path documentationRoot, final boolean update) throws IOException {
+    static void publish(final Path source, final Path documentationRoot, final boolean update,
+            final String activeSnapshotFileName) throws IOException {
+        if (!activeSnapshotFileName.matches("stemming-quality-[0-9]{4}-[0-9]{2}-[0-9]{2}\\.csv")) {
+            throw new IllegalArgumentException("The active stemming-quality snapshot must use a dated file name.");
+        }
         if (!Files.isRegularFile(source) || source.getFileName().toString().contains("filtered")) {
             throw new IllegalArgumentException("The documentation source must be an existing complete, unfiltered CSV report: " + source);
         }
-        final List<ResultRow> rows = readRows(source);
-        final Map<String, Page> pages = readPages(documentationRoot.resolve("benchmarks/languages/index.md"));
-        final Set<String> languageUniverse = readLanguageUniverse(documentationRoot.resolve("built-in-languages.md"));
-        validate(rows, pages.keySet(), languageUniverse);
+        final List<ResultRow> sourceRows = readRows(source);
+        final Set<String> resultLanguages = new HashSet<>();
+        sourceRows.forEach(row -> resultLanguages.add(row.language()));
         final String checksum = sha256(source);
+        final Set<String> languageUniverse = publicationLanguageUniverse(
+                resultLanguages, update, source, documentationRoot, checksum);
+        final List<ResultRow> rows = sourceRows.stream()
+                .filter(row -> languageUniverse.contains(row.language())).toList();
+        final Map<String, Page> pages = readPages(
+                documentationRoot.resolve("benchmarks/languages/index.md"), languageUniverse);
+        validate(rows, pages.keySet(), languageUniverse);
         if (!update) {
-            final Path checksumFile = documentationRoot.resolve("benchmarks/data/stemming-quality.sha256");
+            final Path checksumFile = documentationRoot.resolve("benchmarks/data/")
+                    .resolve(activeSnapshotFileName.replace(".csv", ".sha256"));
             final String recorded = Files.readString(checksumFile, StandardCharsets.UTF_8).strip();
-            if (!recorded.equals(checksum + "  stemming-quality.csv")) {
+            if (!recorded.equals(checksum + "  " + activeSnapshotFileName)) {
                 throw new IllegalStateException("The published stemming-quality checksum does not match the authoritative CSV.");
             }
+            verifyActiveManifest(documentationRoot, activeSnapshotFileName);
         }
         for (Page page : pages.values()) {
             final List<ResultRow> languageRows = rows.stream().filter(row -> row.language().equals(page.language())).toList();
-            final String section = render(page, languageRows, checksum);
+            final String section = render(page, languageRows, checksum, activeSnapshotFileName);
             final Path path = documentationRoot.resolve("benchmarks/languages").resolve(page.file());
             final String original = Files.readString(path, StandardCharsets.UTF_8);
             final String expected = replaceSection(original, section);
@@ -127,7 +146,8 @@ public final class StemmingQualityDocumentationPublisher {
         }
         final Path overviewPath = documentationRoot.resolve("benchmarks/index.md");
         final String overview = Files.readString(overviewPath, StandardCharsets.UTF_8);
-        final String expectedOverview = replaceMarkedSection(overview, renderOverview(pages, rows, checksum),
+        final String expectedOverview = replaceMarkedSection(overview,
+                renderOverview(pages, rows, checksum, activeSnapshotFileName),
                 OVERVIEW_START, OVERVIEW_END);
         if (update) {
             Files.writeString(overviewPath, expectedOverview, StandardCharsets.UTF_8);
@@ -135,39 +155,109 @@ public final class StemmingQualityDocumentationPublisher {
             throw new IllegalStateException("The generated benchmark quality overview is stale or manually altered: " + overviewPath);
         }
         if (update) {
-            final Path publishedSource = documentationRoot.resolve("benchmarks/data/stemming-quality.csv");
+            final Path publishedSource = documentationRoot.resolve("benchmarks/data").resolve(activeSnapshotFileName);
             Files.createDirectories(publishedSource.getParent());
             Files.copy(source, publishedSource, StandardCopyOption.REPLACE_EXISTING);
-            Files.writeString(documentationRoot.resolve("benchmarks/data/stemming-quality.sha256"), checksum + "  stemming-quality.csv\n", StandardCharsets.UTF_8);
+            Files.writeString(publishedSource.resolveSibling(activeSnapshotFileName.replace(".csv", ".sha256")),
+                    checksum + "  " + activeSnapshotFileName + "\n", StandardCharsets.UTF_8);
+            updateActiveManifest(documentationRoot, activeSnapshotFileName);
         }
         System.out.printf(Locale.ROOT, "%s stemming-quality documentation for %d languages from %d validated rows.%n",
                 update ? "Updated" : "Verified", pages.size(), rows.stream().filter(row -> pages.containsKey(row.language())).count());
     }
 
-    /** Reads the authoritative built-in language identifiers from the existing registry table. */
-    private static Set<String> readLanguageUniverse(final Path builtInLanguages) throws IOException {
-        final Set<String> languages = new HashSet<>();
-        for (String line : Files.readAllLines(builtInLanguages, StandardCharsets.UTF_8)) {
-            final Matcher matcher = BUILT_IN_LANGUAGE_ROW.matcher(line);
-            if (matcher.matches()) {
-                languages.add(matcher.group(1));
+    /** Verifies that the shared lifecycle manifest selects the supplied quality snapshot. */
+    private static void verifyActiveManifest(final Path documentationRoot, final String activeSnapshotFileName)
+            throws IOException {
+        final Path manifest = documentationRoot.resolve("benchmarks/data").resolve(ACTIVE_MANIFEST);
+        final String expected = "quality=" + activeSnapshotFileName;
+        if (Files.readAllLines(manifest, StandardCharsets.UTF_8).stream().noneMatch(expected::equals)) {
+            throw new IllegalStateException("The active snapshot manifest does not select " + activeSnapshotFileName + ".");
+        }
+    }
+
+    /** Updates only the quality pointer while preserving the other lifecycle entries. */
+    private static void updateActiveManifest(final Path documentationRoot, final String activeSnapshotFileName)
+            throws IOException {
+        final Path manifest = documentationRoot.resolve("benchmarks/data").resolve(ACTIVE_MANIFEST);
+        final List<String> lines = Files.exists(manifest)
+                ? new ArrayList<>(Files.readAllLines(manifest, StandardCharsets.UTF_8))
+                : new ArrayList<>(List.of("# Radixor benchmark active snapshots v1"));
+        final String entry = "quality=" + activeSnapshotFileName;
+        boolean replaced = false;
+        for (int index = 0; index < lines.size(); index++) {
+            if (lines.get(index).startsWith("quality=")) {
+                lines.set(index, entry);
+                replaced = true;
+                break;
+            }
+        }
+        if (!replaced) {
+            lines.add(entry);
+        }
+        Files.writeString(manifest, String.join("\n", lines) + "\n", StandardCharsets.UTF_8);
+    }
+
+    /**
+     * Selects either the frozen legacy snapshot or the complete current default
+     * universe. Partial current-universe reports are never publishable.
+     *
+     * @param reportedLanguages language labels present in the source report
+     * @param update whether documentation would be updated
+     * @param source source CSV path
+     * @param documentationRoot documentation root
+     * @param checksum source CSV SHA-256
+     * @return authoritative default-language universe for publication
+     */
+    static Set<String> publicationLanguageUniverse(final Set<String> reportedLanguages,
+            final boolean update, final Path source, final Path documentationRoot, final String checksum) {
+        final Set<String> authoritative = new HashSet<>();
+        for (Language language : Language.values()) {
+            authoritative.add(language.name());
+        }
+        final Set<String> unexpected = new HashSet<>(reportedLanguages);
+        unexpected.removeAll(authoritative);
+        unexpected.remove(OPTIONAL_POLIMORF);
+        if (!unexpected.isEmpty()) {
+            throw new IllegalStateException("Unknown stemming-quality language labels: " + unexpected);
+        }
+        final Set<String> reportedDefaults = new HashSet<>(reportedLanguages);
+        reportedDefaults.retainAll(authoritative);
+        if (reportedDefaults.equals(LEGACY_LANGUAGES)) {
+            final Path expectedSource = documentationRoot.resolve("benchmarks/data/stemming-quality.csv")
+                    .toAbsolutePath().normalize();
+            final Path actualSource = source.toAbsolutePath().normalize();
+            if (update || !actualSource.equals(expectedSource) || !LEGACY_SOURCE_SHA256.equals(checksum)) {
+                throw new IllegalStateException("The frozen 20-language snapshot is accepted only in verify mode "
+                        + "from docs/benchmarks/data/stemming-quality.csv with its historical SHA-256.");
             }
+            return LEGACY_LANGUAGES;
         }
-        if (languages.isEmpty()) {
-            throw new IllegalStateException("No authoritative built-in languages were discovered in " + builtInLanguages);
+        if (reportedDefaults.equals(authoritative)) {
+            if (reportedLanguages.contains(OPTIONAL_POLIMORF)) {
+                throw new IllegalStateException("The active default-language stemming-quality snapshot must not "
+                        + "contain optional PoliMorf rows; publish that model under a separate protocol.");
+            }
+            return Set.copyOf(authoritative);
         }
-        return Set.copyOf(languages);
+        final Set<String> missing = new HashSet<>(authoritative);
+        missing.removeAll(reportedDefaults);
+        throw new IllegalStateException("A non-legacy stemming-quality report must cover all "
+                + authoritative.size() + " language defaults; missing: " + missing + ".");
     }
 
     /** Reads the language-code-to-page mapping from the existing documentation index. */
-    private static Map<String, Page> readPages(final Path index) throws IOException {
+    private static Map<String, Page> readPages(final Path index, final Set<String> includedLanguages)
+            throws IOException {
         final Map<String, Page> pages = new LinkedHashMap<>();
         for (String line : Files.readAllLines(index, StandardCharsets.UTF_8)) {
             final Matcher matcher = PAGE_ROW.matcher(line);
-            if (matcher.matches()) {
-                final Page previous = pages.put(matcher.group(1), new Page(matcher.group(1), matcher.group(2), matcher.group(3)));
+            if (matcher.matches() && includedLanguages.contains(matcher.group(2))) {
+                final String language = matcher.group(2);
+                final Page previous = pages.put(language,
+                        new Page(language, matcher.group(1), matcher.group(4)));
                 if (previous != null) {
-                    throw new IllegalStateException("Duplicate language mapping in benchmark index: " + matcher.group(1));
+                    throw new IllegalStateException("Duplicate language mapping in benchmark index: " + language);
                 }
             }
         }
@@ -316,13 +406,15 @@ public final class StemmingQualityDocumentationPublisher {
     }
 
     /** Renders one complete generated section for a language page. */
-    private static String render(final Page page, final List<ResultRow> rows, final String checksum) {
+    private static String render(final Page page, final List<ResultRow> rows, final String checksum,
+            final String activeSnapshotFileName) {
         final String modelId = Language.valueOf(page.language()).defaultModelId();
         final StringBuilder output = new StringBuilder(32768);
         output.append(START).append("\n\n## Stemming Quality\n\n")
                 .append("Runtime performance and linguistic grouping quality are independent dimensions. This section evaluates language `")
                 .append(page.language()).append("` using the complete validated stemming-quality result matrix. Every distinct surface form is one evaluated item and can belong to several dictionary groups. Two forms are a positive pair when their group-membership sets intersect and a negative pair when those sets are disjoint. A pair shared through several groups is counted once. Exact equality with a predetermined lemma is not required.\n\n")
-                .append("`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/stemming-quality.csv).\n\n")
+                .append("`ALL_WORDS` includes every valid group and its original forms. `LOWERCASE_GROUPS_ONLY` excludes an entire group when any Unicode code point is uppercase or titlecase; retained words are not lowercased or otherwise rewritten. This isolates case-handling effects without changing retained inputs. [Download the complete machine-readable result snapshot](../data/")
+                .append(activeSnapshotFileName).append(").\n\n")
                 .append("### Evaluation Scope and Key Findings\n\n")
                 .append("The default model is `").append(modelId).append("`, loaded from classpath resource `org/egothor/stemmer/models/")
                 .append(modelId).append("/stemmer.gz`. The following findings compare only deterministic `PRIMARY_OUTPUT` rows over identical included groups; candidate policies are reported separately as capability analyses.\n\n");
@@ -342,7 +434,10 @@ public final class StemmingQualityDocumentationPublisher {
                     if (policy.equals("ANY_CANDIDATE")) {
                         renderAnyCandidatePolicy(output, policyRows);
                     } else {
-                        output.append("#### `").append(policy).append("` ranking\n\n");
+                        final boolean rankable = policyRows.stream()
+                                .anyMatch(row -> row.isDefined("Balanced accuracy"));
+                        output.append("#### `").append(policy)
+                                .append(rankable ? "` ranking\n\n" : "` results (balanced accuracy `n/a`)\n\n");
                         renderPrimaryTable(output, policyRows);
                         renderDetailedTables(output, policyRows);
                     }
@@ -352,7 +447,7 @@ public final class StemmingQualityDocumentationPublisher {
         }
         appendMethodology(output);
         output.append("### Provenance\n\n")
-                .append("- Authoritative source: `docs/benchmarks/data/stemming-quality.csv`\n")
+                .append("- Authoritative source: `docs/benchmarks/data/").append(activeSnapshotFileName).append("`\n")
                 .append("- Source SHA-256: `").append(checksum).append("`\n")
                 .append("- Evaluation command: `./gradlew stemmingQuality --no-daemon`\n")
                 .append("- Dictionary language: `").append(page.language()).append("`\n")
@@ -367,7 +462,12 @@ public final class StemmingQualityDocumentationPublisher {
     /** Appends one deterministic primary-output winner and runner-up statement. */
     private static void appendFinding(final StringBuilder output, final List<ResultRow> rows, final String mode) {
         final List<ResultRow> primary = rows.stream().filter(row -> row.mode().equals(mode) && row.policy().equals("PRIMARY_OUTPUT"))
-                .sorted(resultOrder()).toList();
+                .filter(row -> row.isDefined("Balanced accuracy")).sorted(resultOrder()).toList();
+        if (primary.isEmpty()) {
+            output.append("- **").append(mode)
+                    .append(":** balanced-accuracy ranking is **n/a** because this corpus has no pairs in one required class. OI/UI and raw numerators and denominators remain authoritative.\n");
+            return;
+        }
         final ResultRow winner = primary.getFirst();
         final ResultRow runnerUp = primary.size() > 1 ? primary.get(1) : null;
         output.append("- **").append(mode).append(":** `").append(displayStemmer(winner.stemmer())).append("` ranks first by balanced accuracy at **")
@@ -393,7 +493,7 @@ public final class StemmingQualityDocumentationPublisher {
                 .append("|---:|---|---:|---:|---:|\n");
         for (int index = 0; index < rows.size(); index++) {
             final ResultRow row = rows.get(index);
-            output.append('|').append(index + 1).append('|').append(displayStemmer(row.stemmer())).append('|')
+            output.append('|').append(rank(row, index)).append('|').append(displayStemmer(row.stemmer())).append('|')
                     .append(metric(row, "Balanced accuracy")).append('|')
                     .append(rate(row, "Over-stemming error pairs", "Over-stemming percentage")).append('|')
                     .append(rate(row, "Under-stemming error pairs", "Under-stemming percentage")).append("|\n");
@@ -489,7 +589,12 @@ public final class StemmingQualityDocumentationPublisher {
 
     /** Returns the repeated rank, stemmer, and policy prefix for a detailed table row. */
     private static String identity(final int index, final ResultRow row) {
-        return "|" + (index + 1) + "|" + displayStemmer(row.stemmer()) + "|" + row.policy() + "|";
+        return "|" + rank(row, index) + "|" + displayStemmer(row.stemmer()) + "|" + row.policy() + "|";
+    }
+
+    /** Returns a rank only when the navigation metric is mathematically defined. */
+    private static String rank(final ResultRow row, final int index) {
+        return row.isDefined("Balanced accuracy") ? Integer.toString(index + 1) : "n/a";
     }
 
     /** Converts authoritative adapter identifiers into a stable readable label without merging competitors. */
@@ -518,7 +623,8 @@ public final class StemmingQualityDocumentationPublisher {
     }
 
     /** Renders the generated executive findings, winner matrix, and Radixor aggregates. */
-    private static String renderOverview(final Map<String, Page> pages, final List<ResultRow> rows, final String checksum) {
+    private static String renderOverview(final Map<String, Page> pages, final List<ResultRow> rows, final String checksum,
+            final String activeSnapshotFileName) {
         final StringBuilder output = new StringBuilder(16384);
         output.append(OVERVIEW_START).append("\n\n## Pairwise Quality Findings\n\n")
                 .append("The validated snapshot is a broad multilingual comparison covering the complete ")
@@ -528,7 +634,10 @@ public final class StemmingQualityDocumentationPublisher {
         int directComparisons = 0;
         for (String mode : MODES) {
             for (String language : pages.keySet()) {
-                final List<ResultRow> ranked = primaryRows(rows, language, mode);
+                final List<ResultRow> ranked = definedPrimaryRows(rows, language, mode);
+                if (ranked.isEmpty()) {
+                    continue;
+                }
                 comparisons++;
                 if (ranked.size() > 1) {
                     directComparisons++;
@@ -550,7 +659,12 @@ public final class StemmingQualityDocumentationPublisher {
                 .append("|---|---|---|---:|---|---:|---|---:|\n");
         for (Page page : pages.values()) {
             for (String mode : MODES) {
-                final List<ResultRow> ranked = primaryRows(rows, page.language(), mode);
+                final List<ResultRow> ranked = definedPrimaryRows(rows, page.language(), mode);
+                if (ranked.isEmpty()) {
+                    output.append('|').append(page.displayName()).append(" (`").append(page.language()).append("`)|")
+                            .append(mode).append("|n/a|n/a|n/a|n/a|n/a|0|\n");
+                    continue;
+                }
                 final ResultRow winner = ranked.getFirst();
                 final ResultRow runner = ranked.size() > 1 ? ranked.get(1) : null;
                 final double difference = runner == null ? Double.NaN : winner.number("Balanced accuracy") - runner.number("Balanced accuracy");
@@ -573,7 +687,10 @@ public final class StemmingQualityDocumentationPublisher {
         for (String mode : MODES) {
             final List<ResultRow> radixor = rows.stream().filter(row -> pages.containsKey(row.language()) && row.mode().equals(mode)
                     && row.policy().equals("PRIMARY_OUTPUT") && row.stemmer().endsWith("_RADIXOR")).toList();
-            final double macroBalanced = radixor.stream().mapToDouble(row -> row.number("Balanced accuracy")).average().orElseThrow();
+            final List<ResultRow> definedRadixor = radixor.stream()
+                    .filter(row -> row.isDefined("Balanced accuracy")).toList();
+            final double macroBalanced = definedRadixor.stream()
+                    .mapToDouble(row -> row.number("Balanced accuracy")).average().orElseThrow();
             long tp = 0;
             long fp = 0;
             long fn = 0;
@@ -588,11 +705,13 @@ public final class StemmingQualityDocumentationPublisher {
             final double recall = (double) tp / Math.addExact(tp, fn);
             final double specificity = (double) tn / Math.addExact(tn, fp);
             final double f1 = 2.0 * tp / (2.0 * tp + fp + fn);
-            output.append('|').append(mode).append('|').append(radixor.size()).append('|').append(format(macroBalanced)).append('|')
+            output.append('|').append(mode).append('|').append(definedRadixor.size()).append(" / ")
+                    .append(radixor.size()).append('|').append(format(macroBalanced)).append('|')
                     .append(format((recall + specificity) / 2.0)).append('|').append(format(precision)).append('|')
                     .append(format(recall)).append('|').append(format(f1)).append("|\n");
         }
-        output.append("\n### Reproducible data\n\n- [Machine-readable quality snapshot](data/stemming-quality.csv)\n")
+        output.append("\n### Reproducible data\n\n- [Machine-readable quality snapshot](data/")
+                .append(activeSnapshotFileName).append(")\n")
                 .append("- SHA-256: `").append(checksum).append("`\n")
                 .append("- [Linguistic quality methodology](reference/linguistic-quality.md)\n")
                 .append("- [Tested stemmer inventory](reference/tested-stemmers.md)\n")
@@ -618,7 +737,10 @@ public final class StemmingQualityDocumentationPublisher {
         int count = 0;
         for (Page page : pages.values()) {
             for (String mode : MODES) {
-                final List<ResultRow> primary = primaryRows(rows, page.language(), mode);
+                final List<ResultRow> primary = definedPrimaryRows(rows, page.language(), mode);
+                if (primary.isEmpty()) {
+                    continue;
+                }
                 for (Map.Entry<String, Boolean> metric : metrics.entrySet()) {
                     final Comparator<ResultRow> comparator = Comparator.comparingDouble(row -> row.number(metric.getKey()));
                     final ResultRow leader = metric.getValue() ? primary.stream().max(comparator).orElseThrow()
@@ -646,7 +768,10 @@ public final class StemmingQualityDocumentationPublisher {
         final Map<String, Integer> ties = new HashMap<>();
         final Map<String, Integer> topThree = new HashMap<>();
         for (String language : pages.keySet()) {
-            final List<ResultRow> ranked = primaryRows(rows, language, mode);
+            final List<ResultRow> ranked = definedPrimaryRows(rows, language, mode);
+            if (ranked.isEmpty()) {
+                continue;
+            }
             final double leading = ranked.getFirst().number("Balanced accuracy");
             final long leaders = ranked.stream().filter(row -> row.number("Balanced accuracy") == leading).count();
             for (int index = 0; index < ranked.size(); index++) {
@@ -690,6 +815,13 @@ public final class StemmingQualityDocumentationPublisher {
                 && row.policy().equals("PRIMARY_OUTPUT")).sorted(resultOrder()).toList();
     }
 
+    /** Returns primary-output rows whose balanced-accuracy ranking metric is defined. */
+    private static List<ResultRow> definedPrimaryRows(final List<ResultRow> rows, final String language,
+            final String mode) {
+        return primaryRows(rows, language, mode).stream()
+                .filter(row -> row.isDefined("Balanced accuracy")).toList();
+    }
+
     /** Formats an aggregate metric at the publication precision. */
     private static String format(final double value) {
         return String.format(Locale.ROOT, "%.6f", value);
@@ -800,6 +932,8 @@ public final class StemmingQualityDocumentationPublisher {
         private long longValue(final String name) { return Long.parseLong(value(name)); }
         /** Parses a numeric field, placing undefined values last during sorting. */
         private double number(final String name) { return value(name).isEmpty() ? Double.NEGATIVE_INFINITY : Double.parseDouble(value(name)); }
+        /** Reports whether a metric has a mathematically defined source value. */
+        private boolean isDefined(final String name) { return !value(name).isEmpty(); }
         /** Returns false-negative pairs. */
         private long fn() { return longValue("Under-stemming error pairs"); }
         /** Returns false-positive pairs. */
@@ -829,8 +963,17 @@ public final class StemmingQualityDocumentationPublisher {
             if (Math.addExact(tp, fn) != underPossible || Math.addExact(tn, fp) != overPossible) {
                 throw new IllegalStateException("Raw confusion-count invariants fail for " + key());
             }
-            final double recall = ratio(tp, Math.addExact(tp, fn));
-            final double specificity = ratio(tn, Math.addExact(tn, fp));
+            final long recallDenominator = Math.addExact(tp, fn);
+            final long specificityDenominator = Math.addExact(tn, fp);
+            if (recallDenominator == 0 || specificityDenominator == 0) {
+                if (!value("Balanced accuracy").isEmpty()) {
+                    throw new IllegalStateException("Balanced accuracy must be empty when either class is absent for "
+                            + key());
+                }
+                return;
+            }
+            final double recall = ratio(tp, recallDenominator);
+            final double specificity = ratio(tn, specificityDenominator);
             final double expected = (recall + specificity) / 2.0;
             if (Math.abs(expected - number("Balanced accuracy")) > 0.0000000000015) {
                 throw new IllegalStateException("Balanced accuracy is inconsistent with raw counts for " + key());
diff --git a/tools/run-published-accuracy-benchmarks.sh b/tools/run-published-accuracy-benchmarks.sh
index b3e8d4d..0bc45e4 100755
--- a/tools/run-published-accuracy-benchmarks.sh
+++ b/tools/run-published-accuracy-benchmarks.sh
@@ -50,6 +50,11 @@ mkdir -p "${tmp_dir}" "${report_dir}"
 accuracy_include='^org\.egothor\.stemmer\.benchmark\.(EnglishHunspellStemmerComparisonBenchmarkQuality|EnglishStemmerComparisonBenchmarkQuality|HunspellStemmerComparisonBenchmarkQuality|StemmerComparisonBenchmarkQuality)\..*$'
 coverage_include='^org\.egothor\.stemmer\.benchmark\.EnglishRadixorDictionaryCoverageBenchmark\.exactRootAgreement$'
 accuracy_csv="${report_dir}/stemmer-accuracy-${report_date}.csv"
+snowball_catalog="docs/benchmarks/data/snowball-language-cases.csv"
+
+java -cp "${jmh_classpath}" \
+    org.egothor.stemmer.benchmark.SnowballLanguageCatalogApplication \
+    "${snowball_catalog}" verify
 
 common_arguments=(
     -f 0
@@ -68,39 +73,10 @@ java -Djava.io.tmpdir="${tmp_dir}" -Xms6g -Xmx6g \
     -rff "${accuracy_csv}" \
     -o "${report_dir}/stemmer-accuracy-${report_date}.txt"
 
-for candidate in SNOWBALL_CZECH_DIRECT SNOWBALL_PERSIAN_DIRECT SNOWBALL_POLISH_DIRECT; do
-    if ! awk -F, -v candidate="${candidate}" '
-        NR > 1 {
-            value = $8
-            gsub(/^"|"$/, "", value)
-            if (value == candidate) {
-                found = 1
-            }
-        }
-        END { exit found ? 0 : 1 }
-    ' "${accuracy_csv}"; then
-        printf 'Exact-root report omits %s.\n' "${candidate}" >&2
-        exit 1
-    fi
-    for counter in correctMatches evaluatedTokens changedCorrectMatches changedEvaluatedTokens \
-            rootPreservedMatches rootEvaluatedTokens; do
-        if ! awk -F, -v candidate="${candidate}" -v suffix="exactRootAgreement:${counter}" '
-            NR > 1 {
-                benchmark = $1
-                value = $8
-                gsub(/^"|"$/, "", benchmark)
-                gsub(/^"|"$/, "", value)
-                if (value == candidate && benchmark ~ suffix "$") {
-                    found = 1
-                }
-            }
-            END { exit found ? 0 : 1 }
-        ' "${accuracy_csv}"; then
-            printf 'Exact-root report omits %s for %s.\n' "${counter}" "${candidate}" >&2
-            exit 1
-        fi
-    done
-done
+python3 tools/validate-accuracy-candidate-coverage.py \
+    --report "${accuracy_csv}" \
+    --snowball-catalog "${snowball_catalog}" \
+    --topology models/model-projects.properties
 
 java -Djava.io.tmpdir="${tmp_dir}" -Xms6g -Xmx6g \
     -cp "${jmh_classpath}" org.openjdk.jmh.Main \
diff --git a/tools/run-published-speed-benchmarks.sh b/tools/run-published-speed-benchmarks.sh
index a24262d..d8eeeb2 100755
--- a/tools/run-published-speed-benchmarks.sh
+++ b/tools/run-published-speed-benchmarks.sh
@@ -51,6 +51,9 @@ IFS= read -r jmh_classpath < "${classpath_file}"
 tmp_dir="${project_root}/build/tmp/jmh"
 report_dir="${project_root}/build/reports/jmh"
 mkdir -p "${tmp_dir}" "${report_dir}"
+classpath_content_manifest="${report_dir}/jmh-classpath-content-${report_date}.sha256"
+bash tools/write-classpath-content-manifest.sh \
+    "${classpath_file}" "${classpath_content_manifest}"
 
 for governor_file in /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor; do
     governor="$(<"${governor_file}")"
@@ -60,13 +63,14 @@ for governor_file in /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor; do
     fi
 done
 
-comparison_include='^(org\.egothor\.stemmer\.benchmark\.(EnglishStemmerComparisonBenchmark\.|MultiLanguageStemmerComparisonBenchmark\.|SnowballLanguageStemmerComparisonBenchmark\.).*|org\.egothor\.stemmer\.benchmark\.HunspellStemmerComparisonBenchmark\.luceneHunspellStemFilter)$'
+comparison_include='^(org\.egothor\.stemmer\.benchmark\.(EnglishStemmerComparisonBenchmark\.|MultiLanguageStemmerComparisonBenchmark\.|RadixorModelStemmerBenchmark\.|SnowballLanguageStemmerComparisonBenchmark\.).*|org\.egothor\.stemmer\.benchmark\.HunspellStemmerComparisonBenchmark\.luceneHunspellStemFilter|org\.egothor\.stemmer\.benchmark\.PolishPolimorfStemmerComparisonBenchmark\.polishPolimorfLuceneMorfologikFilter)$'
 coverage_include='^org\.egothor\.stemmer\.benchmark\.EnglishRadixorDictionaryCoverageBenchmark\.changedTokenStemmingSpeed$'
 selection_file="${report_dir}/published-speed-benchmarks-${report_date}.txt"
+main_speed_csv="${report_dir}/stemmer-speed-${report_date}.csv"
 
 java -Djava.io.tmpdir="${tmp_dir}" -cp "${jmh_classpath}" org.openjdk.jmh.Main \
     "${comparison_include}" -l > "${selection_file}"
-if grep -Eq 'PolishPolimorf|BenchmarkQuality|GermanGoldstandard' "${selection_file}"; then
+if grep -Eq 'PolishPolimorfStemmerComparisonBenchmark\.polishPolimorfRadixor|BenchmarkQuality|GermanGoldstandard' "${selection_file}"; then
     printf 'The selected speed benchmark list contains an excluded benchmark.\n' >&2
     exit 1
 fi
@@ -74,6 +78,14 @@ if ! grep -q 'MultiLanguageStemmerComparisonBenchmark.hebrewRadixor' "${selectio
     printf 'The selected speed benchmark list omits Hebrew Radixor.\n' >&2
     exit 1
 fi
+if ! grep -q 'RadixorModelStemmerBenchmark.radixor' "${selection_file}"; then
+    printf 'The selected speed benchmark list omits the all-model Radixor benchmark.\n' >&2
+    exit 1
+fi
+if ! grep -q 'PolishPolimorfStemmerComparisonBenchmark.polishPolimorfLuceneMorfologikFilter' "${selection_file}"; then
+    printf 'The selected speed benchmark list omits the PoliMorf comparator.\n' >&2
+    exit 1
+fi
 
 environment_file="${report_dir}/performance-environment-${report_date}.txt"
 source_patch="${report_dir}/measured-source-${report_date}.patch"
@@ -94,6 +106,10 @@ jmh_jar="${jmh_classpath%%:*}"
     sha256sum "${classpath_file}" | cut -d' ' -f1
     printf 'JMH executable JAR SHA-256: '
     sha256sum "${jmh_jar}" | cut -d' ' -f1
+    printf 'JMH runtime classpath content manifest: %s\n' \
+        "${classpath_content_manifest}"
+    printf 'JMH runtime classpath content manifest SHA-256: '
+    sha256sum "${classpath_content_manifest}" | cut -d' ' -f1
     printf 'Measured source patch SHA-256: '
     sha256sum "${source_patch}" | cut -d' ' -f1
     printf 'Untracked source checksum manifest SHA-256: '
@@ -141,8 +157,8 @@ sleep 30
 
 common_arguments=(
     -f 3
-    -wi 5
-    -i 7
+    -wi 3
+    -i 5
     -w 1s
     -r 1s
     -t 1
@@ -155,9 +171,16 @@ common_arguments=(
 java -Djava.io.tmpdir="${tmp_dir}" -Xms512m -Xmx1g \
     -cp "${jmh_classpath}" org.openjdk.jmh.Main \
     "${comparison_include}" "${common_arguments[@]}" \
-    -rff "${report_dir}/stemmer-speed-${report_date}.csv" \
+    -rff "${main_speed_csv}" \
     -o "${report_dir}/stemmer-speed-${report_date}.txt"
 
+bash tools/validate-speed-model-coverage.sh \
+    "${main_speed_csv}" models/model-projects.properties
+{
+    printf '\nMain speed report SHA-256: '
+    sha256sum "${main_speed_csv}" | cut -d' ' -f1
+} >> "${environment_file}"
+
 sleep 15
 
 java -Djava.io.tmpdir="${tmp_dir}" -Xms512m -Xmx1g \
diff --git a/tools/tests/test_update_benchmark_documentation.py b/tools/tests/test_update_benchmark_documentation.py
index 50d6053..9a2065f 100755
--- a/tools/tests/test_update_benchmark_documentation.py
+++ b/tools/tests/test_update_benchmark_documentation.py
@@ -34,6 +34,7 @@
 from __future__ import annotations
 
 import importlib.util
+import re
 import sys
 import tempfile
 import unittest
@@ -57,9 +58,198 @@ def speed_row(score: float, error: float = 1_000.0) -> dict[str, str]:
     }
 
 
+def protocol_row(samples: int, threads: int = 1) -> dict[str, str]:
+    """Create the concurrency fields used to infer a speed protocol."""
+
+    row = speed_row(1_000.0)
+    row["Samples"] = str(samples)
+    row["Threads"] = str(threads)
+    return row
+
+
 class BenchmarkDocumentationTest(unittest.TestCase):
     """Covers current-report selection and measured-corpus arithmetic."""
 
+    @staticmethod
+    def model_info(index: int, distinct_forms: int) -> MODULE.ModelInfo:
+        """Create compact model metadata for pure ranking tests."""
+
+        return MODULE.ModelInfo(
+            model_id=f"aa-{index:03d}-default",
+            role="standalone",
+            language=f"AA_{index:03d}",
+            display_name=f"Language {index} — Test",
+            language_name=f"Language {index}",
+            version="1.0.0",
+            right_to_left=False,
+            dictionary_rows=1,
+            distinct_forms=distinct_forms,
+            sha256="0" * 64,
+        )
+
+    def test_ranked_star_tiers_have_approved_populations(self) -> None:
+        infos = [self.model_info(index, index + 1) for index in range(144)]
+
+        ranked = MODULE.assign_star_tiers(infos)
+
+        self.assertEqual(
+            [29, 29, 29, 29, 28],
+            [sum(info.stars == star for info in ranked) for star in range(1, 6)],
+        )
+        self.assertEqual(1, ranked[0].stars)
+        self.assertEqual(5, ranked[-1].stars)
+
+    def test_rating_has_five_visible_positions_and_accessible_count(self) -> None:
+        info = self.model_info(1, 10_095)
+        info = MODULE.replace(info, stars=3)
+
+        rating = MODULE.rating_markup(info)
+
+        self.assertIn("★★★☆☆", rating)
+        self.assertIn('role="img"', rating)
+        self.assertIn("10,095 distinct usable word forms", rating)
+
+    def test_page_rating_preserves_existing_rich_sections(self) -> None:
+        info = MODULE.replace(self.model_info(1, 500), stars=2)
+        text = """# Example Stemmer Benchmarks
+
+Introduction.
+
+<!-- DICTIONARY-GENERALIZATION:START -->
+measured rich content
+<!-- DICTIONARY-GENERALIZATION:END -->
+"""
+
+        updated = MODULE._with_rating(text, info)
+
+        self.assertIn("★★☆☆☆", updated)
+        self.assertIn("measured rich content", updated)
+        self.assertEqual(updated, MODULE._with_rating(updated, info))
+
+    def test_generic_model_parameter_selects_exact_speed_row(self) -> None:
+        expected = MODULE.Key(
+            "example.radixor", (("modelId", "ady-default"),)
+        )
+        other = MODULE.Key(
+            "example.radixor", (("modelId", "afb-default"),)
+        )
+        data = MODULE.JmhData(
+            primary={expected: speed_row(10.0), other: speed_row(20.0)},
+            auxiliary={},
+        )
+
+        selected = MODULE.select_speed_key(
+            "radixor[ady-default]", data, "ADY"
+        )
+
+        self.assertEqual(expected, selected)
+
+    def test_snowball_mapping_excludes_serbo_croatian(self) -> None:
+        cases = MODULE.read_snowball_catalog(
+            Path("docs/benchmarks/data/snowball-language-cases.csv")
+        )
+        mapping = {
+            case.language: (case.case_name, case.lucene_available)
+            for case in cases
+        }
+        self.assertNotIn("HBS", mapping)
+        self.assertEqual(("SESOTHO", False), mapping["ST_ZA"])
+        self.assertEqual({"SESOTHO"}, MODULE.LANGUAGE_IDENTITY_WORDS["ST_ZA"])
+
+    def test_homepage_snowball_comparison_has_exact_28_27_coverage(self) -> None:
+        cases = MODULE.read_snowball_catalog(
+            Path("docs/benchmarks/data/snowball-language-cases.csv")
+        )
+        corpus = MODULE.read_corpora(
+            Path("docs/benchmarks/data/java-benchmark-corpora-2026-09-11.csv")
+        )
+        speed = MODULE.read_jmh(
+            Path("docs/benchmarks/data/java-stemmer-speed-2026-09-11.csv")
+        )
+
+        section = MODULE.render_snowball_homepage(cases, corpus, speed, "2026-09-11")
+
+        self.assertIn("all <b>28</b>", section)
+        self.assertIn("over 27 comparable changed-token cases", section)
+        self.assertEqual(28, section.count("<tr><td>"))
+        self.assertEqual(28, section.count('class="rx2-ratio-row"'))
+        self.assertIn('class="rx2-ratio-axis"', section)
+        self.assertNotIn('class="rx2-bar"', section)
+        self.assertIn("no — root-only corpus", section)
+
+        invalid_corpus = dict(corpus)
+        invalid_case = next(case for case in cases if case.homepage_aggregate)
+        invalid_corpus[invalid_case.model_id] = dict(corpus[invalid_case.model_id])
+        invalid_corpus[invalid_case.model_id]["timing_basis"] = "complete root-only corpus"
+        with self.assertRaisesRegex(ValueError, "changed-token corpus"):
+            MODULE.render_snowball_homepage(
+                cases, invalid_corpus, speed, "2026-09-11"
+            )
+
+    def test_homepage_geography_has_exact_default_coverage(self) -> None:
+        infos = MODULE.load_model_infos()
+        defaults = {info.language for info in infos if info.role != "optional"}
+        geography = MODULE.read_homepage_geography(
+            Path("docs/benchmarks/data/homepage-language-geography.csv"), defaults
+        )
+
+        self.assertEqual(143, len(geography))
+        self.assertEqual("🌐", MODULE.geography_glyph(geography["AR"].icon))
+        self.assertEqual("🇯🇵", MODULE.geography_glyph(geography["JA_JP"].icon))
+        self.assertEqual("Transnational language", geography["AR"].label)
+
+    def test_homepage_css_keeps_dense_counts_and_ratio_rows_responsive(self) -> None:
+        css = Path("docs/assets/stylesheets/landing-v2.css").read_text(encoding="utf-8")
+
+        self.assertIn("grid-template-columns:repeat(4,minmax(0,1fr))", css)
+        self.assertIn(".rx2-highlight-stat{min-width:0}", css)
+        self.assertIn(".rx2-ratio-axis,.rx2-ratio-row", css)
+        self.assertIn("width:max(2px,var(--width))", css)
+
+    def test_homepage_preserves_existing_python_flag_assets(self) -> None:
+        landing = Path("docs/overrides/landing.html").read_text(encoding="utf-8")
+        references = re.findall(
+            r'src="\{\{ base_url \}\}/(assets/images/flags/[^\"]+)"', landing
+        )
+
+        self.assertEqual(20, len(references))
+        self.assertEqual(20, len(set(references)))
+        for reference in references:
+            self.assertTrue(Path("docs", reference).is_file(), reference)
+
+    def test_publication_inputs_must_match_active_manifest(self) -> None:
+        data = Path("docs/benchmarks/data")
+        manifest = data / "active-snapshots.properties"
+        active = {
+            "corpus": data / "java-benchmark-corpora-2026-09-11.csv",
+            "quality": data / "stemming-quality-2026-09-11.csv",
+        }
+
+        MODULE.verify_active_inputs(manifest, active)
+        with self.assertRaisesRegex(ValueError, "is not active"):
+            MODULE.verify_active_inputs(
+                manifest,
+                {"corpus": data / "java-benchmark-corpora-2026-08-25.csv"},
+            )
+
+    def test_homepage_language_section_replacement_is_idempotent(self) -> None:
+        original = """<main>
+    <section class="rx2-languages-strip">old</section>
+</main>
+"""
+        generated = """<!-- JAVA-LANGUAGE-PAGES:START -->
+<section class="rx2-languages-strip">new</section>
+<!-- JAVA-LANGUAGE-PAGES:END -->"""
+
+        updated = MODULE.replace_homepage_language_section(original, generated)
+
+        self.assertEqual(
+            updated,
+            MODULE.replace_homepage_language_section(updated, generated),
+        )
+        self.assertEqual(1, updated.count("JAVA-LANGUAGE-PAGES:START"))
+        self.assertEqual(1, updated.count("JAVA-LANGUAGE-PAGES:END"))
+
     def test_speed_uses_timing_token_denominator(self) -> None:
         radixor = MODULE.Key("example.persianRadixor", ())
         snowball = MODULE.Key(
@@ -170,6 +360,228 @@ class BenchmarkDocumentationTest(unittest.TestCase):
             MODULE.update_accuracy_table(updated, data, "FA_IR", {}),
         )
 
+    def test_empty_changed_population_renders_na(self) -> None:
+        corpus = {
+            "all_exact": 2,
+            "total": 2,
+            "changed_exact": 0,
+            "changed": 0,
+            "root_exact": 2,
+            "roots": 2,
+        }
+
+        values = MODULE.rounded_accuracy(MODULE.corpus_accuracy(corpus))
+
+        self.assertEqual(("100.000%", "n/a", "100.000%"), values)
+        self.assertEqual(
+            ("100.000%", "n/a", "100.000%"),
+            MODULE.rounded_accuracy(
+                (
+                    MODULE.percentage(2, 2),
+                    MODULE.percentage(0, 0),
+                    MODULE.percentage(2, 2),
+                )
+            ),
+        )
+
+    def test_new_page_transitions_to_measured_and_preserves_markers(self) -> None:
+        info = MODULE.replace(self.model_info(7, 700), stars=2)
+        language = info.language
+        key = MODULE.Key("example.radixor", (("modelId", info.model_id),))
+        corpus = {
+            "model": info.model_id,
+            "version": info.version,
+            "rows": 1,
+            "distinct": info.distinct_forms,
+            "total": 1,
+            "roots": 1,
+            "changed": 0,
+            "timing_basis": "root-only tokens",
+            "timing": 5_000,
+            "all_exact": 1,
+            "changed_exact": 0,
+            "root_exact": 1,
+            "commands": [("PreserveCommand", 1)],
+        }
+        old_languages = MODULE.LANGUAGES
+        old_identity_words = MODULE.LANGUAGE_IDENTITY_WORDS
+        try:
+            MODULE.LANGUAGES = {info.page_name: language}
+            MODULE.LANGUAGE_IDENTITY_WORDS = {language: {"LANGUAGE"}}
+            with tempfile.TemporaryDirectory() as directory:
+                docs_root = Path(directory)
+                page = docs_root / "benchmarks" / "languages" / info.page_name
+                page.parent.mkdir(parents=True)
+                pending_shell = MODULE._new_page(info).replace(
+                    "Speed uses JMH average time, 3 warmup iterations, 5 measurement "
+                    "iterations, 3 independent forks, and 1 thread.\n\n",
+                    "",
+                )
+                page.write_text(pending_shell, encoding="utf-8")
+
+                MODULE.update_language_pages(
+                    docs_root,
+                    {language: corpus},
+                    MODULE.JmhData(primary={}, auxiliary={}),
+                    MODULE.JmhData(primary={key: speed_row(250_000.0)}, auxiliary={}),
+                    MODULE.SpeedProtocol(15, 3, 3, 5),
+                    MODULE.Publication("update"),
+                    "2026-09-10",
+                    "test-release",
+                )
+
+                updated = page.read_text(encoding="utf-8")
+                self.assertIn("root-only tokens", updated)
+                self.assertIn(
+                    "3 warmup iterations, 5 measurement iterations, 3 independent forks, and 1 thread",
+                    updated,
+                )
+                self.assertIn("| 100.000% | n/a | 100.000% |", updated)
+                self.assertIn("| 0.250 | 0.001 | 50.0 | 1.000 |", updated)
+                self.assertIn("<!-- DICTIONARY-GENERALIZATION:START -->", updated)
+                self.assertIn("<!-- EDIT-COST-GENERALIZATION:START -->", updated)
+                self.assertIn("<!-- STEMMING-QUALITY:START -->", updated)
+                self.assertNotIn(
+                    "#dictionary-family-generalization-conclusion", updated
+                )
+                self.assertNotIn(
+                    "#edit-costs-and-dictionary-knowledge-generalization", updated
+                )
+                self.assertNotIn("#stemming-quality", updated)
+                self.assertNotIn("Their 10–90% curves", updated)
+                self.assertNotIn("pending", updated.lower())
+        finally:
+            MODULE.LANGUAGES = old_languages
+            MODULE.LANGUAGE_IDENTITY_WORDS = old_identity_words
+
+    def test_evidence_map_rejects_missing_internal_anchor(self) -> None:
+        text = """# Example
+
+<!-- BENCHMARK-EVIDENCE-MAP:START -->
+[Missing section](#missing-section)
+<!-- BENCHMARK-EVIDENCE-MAP:END -->
+
+## Existing Section
+"""
+
+        with self.assertRaisesRegex(ValueError, "missing-section"):
+            MODULE.validate_evidence_fragment_links(text, Path("example.md"))
+
+    def test_polimorf_section_consumes_exact_optional_reports(self) -> None:
+        info = MODULE.ModelInfo(
+            model_id="pl-pl-polimorf",
+            role="optional",
+            language="PL_PL",
+            display_name="Polish — PoliMorf",
+            language_name="Polish",
+            version="1.0.0",
+            right_to_left=False,
+            dictionary_rows=2,
+            distinct_forms=4_668_685,
+            sha256="0" * 64,
+            stars=5,
+        )
+        counters = {
+            "correctMatches": 2.0,
+            "evaluatedTokens": 2.0,
+            "changedCorrectMatches": 1.0,
+            "changedEvaluatedTokens": 1.0,
+            "rootPreservedMatches": 1.0,
+            "rootEvaluatedTokens": 1.0,
+        }
+        accuracy_data = MODULE.JmhData(
+            primary={},
+            auxiliary={
+                MODULE.Key("example.exactRootAgreement", (("candidateName", "POLISH_POLIMORF_RADIXOR"),)): counters,
+                MODULE.Key("example.exactRootAgreement", (("candidateName", "POLISH_POLIMORF_LUCENE_MORFOLOGIK_FILTER"),)): counters,
+                MODULE.Key("example.exactRootAgreement", (("candidateName", "POLISH_POLIMORF_SNOWBALL_DIRECT"),)): counters,
+            },
+        )
+        radixor = MODULE.Key("example.radixor", (("modelId", info.model_id),))
+        morfologik = MODULE.Key("example.polishPolimorfLuceneMorfologikFilter", ())
+        speed_data = MODULE.JmhData(
+            primary={radixor: speed_row(500_000.0), morfologik: speed_row(1_000_000.0)},
+            auxiliary={},
+        )
+        corpus = {
+            "version": "1.0.0",
+            "rows": 2,
+            "distinct": 4_668_685,
+            "total": 2,
+            "roots": 1,
+            "changed": 1,
+            "timing_basis": "changed tokens",
+            "timing": 5_000,
+            "all_exact": 2,
+            "changed_exact": 1,
+            "root_exact": 1,
+        }
+
+        section = MODULE._polimorf_section(
+            info, corpus, accuracy_data, speed_data
+        )
+
+        self.assertIn("★★★★★", section)
+        self.assertIn("4,668,685 distinct usable word forms", section)
+        self.assertIn("POLIMORF", repr(accuracy_data.auxiliary))
+        self.assertIn("| 0.500 | 0.001 | 100.0 | 1.000 |", section)
+        self.assertIn("| 1.000 | 0.001 | 200.0 | 2.000 |", section)
+        self.assertNotIn("pending", section)
+
+    def test_speed_protocol_is_inferred_from_report_samples(self) -> None:
+        key = MODULE.Key("example.speed", ())
+
+        historical = MODULE.infer_speed_protocol(
+            MODULE.JmhData(primary={key: protocol_row(21)}, auxiliary={}),
+            "historical",
+        )
+        canonical = MODULE.infer_speed_protocol(
+            MODULE.JmhData(primary={key: protocol_row(15)}, auxiliary={}),
+            "canonical",
+        )
+
+        self.assertEqual(
+            "Speed uses JMH average time, 5 warmup iterations, 7 measurement iterations, "
+            "3 independent forks, and 1 thread.",
+            historical.page_sentence(),
+        )
+        self.assertEqual(
+            "Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, "
+            "3 independent forks, and 1 thread.",
+            canonical.page_sentence(),
+        )
+
+    def test_speed_protocol_rejects_wrong_thread_count(self) -> None:
+        key = MODULE.Key("example.speed", ())
+
+        with self.assertRaisesRegex(ValueError, "exactly one thread"):
+            MODULE.infer_speed_protocol(
+                MODULE.JmhData(primary={key: protocol_row(15, 2)}, auxiliary={}),
+                "wrong threads",
+            )
+
+    def test_speed_protocol_rejects_inconsistent_samples(self) -> None:
+        first = MODULE.Key("example.first", ())
+        second = MODULE.Key("example.second", ())
+
+        with self.assertRaisesRegex(ValueError, "inconsistent sample counts"):
+            MODULE.infer_speed_protocol(
+                MODULE.JmhData(
+                    primary={first: protocol_row(15), second: protocol_row(21)},
+                    auxiliary={},
+                ),
+                "mixed",
+            )
+
+    def test_speed_protocol_rejects_unsupported_samples(self) -> None:
+        key = MODULE.Key("example.speed", ())
+
+        with self.assertRaisesRegex(ValueError, "unsupported sample count"):
+            MODULE.infer_speed_protocol(
+                MODULE.JmhData(primary={key: protocol_row(12)}, auxiliary={}),
+                "unsupported",
+            )
+
     def test_verify_mode_rejects_stale_document_without_writing(self) -> None:
         with tempfile.TemporaryDirectory() as directory:
             path = Path(directory) / "page.md"
diff --git a/tools/update-benchmark-documentation.py b/tools/update-benchmark-documentation.py
index 36e7a7b..9a17157 100755
--- a/tools/update-benchmark-documentation.py
+++ b/tools/update-benchmark-documentation.py
@@ -36,10 +36,13 @@ from __future__ import annotations
 
 import argparse
 import csv
+import gzip
+import hashlib
+import html
 import math
 import re
 from collections import defaultdict
-from dataclasses import dataclass
+from dataclasses import dataclass, replace
 from pathlib import Path
 
 
@@ -66,6 +69,176 @@ LANGUAGES = {
     "yiddish.md": "YI",
 }
 
+REPOSITORY = Path(__file__).resolve().parents[1]
+MODELS_ROOT = REPOSITORY / "models"
+TOPOLOGY_ROLES = {"default", "standalone", "optional"}
+LEGACY_PAGE_NAMES = {language: file_name for file_name, language in LANGUAGES.items()}
+SNOWBALL_CASES: dict[str, tuple[str, bool]] = {}
+POLIMORF_START = "<!-- POLIMORF-BENCHMARK:START -->"
+POLIMORF_END = "<!-- POLIMORF-BENCHMARK:END -->"
+
+
+@dataclass(frozen=True)
+class ModelInfo:
+    """Authoritative documentation metadata for one user-facing model."""
+
+    model_id: str
+    role: str
+    language: str
+    display_name: str
+    language_name: str
+    version: str
+    right_to_left: bool
+    dictionary_rows: int
+    distinct_forms: int
+    sha256: str
+    stars: int = 0
+
+    @property
+    def page_name(self) -> str:
+        """Return the stable page file for this language default."""
+
+        if self.language in LEGACY_PAGE_NAMES:
+            return LEGACY_PAGE_NAMES[self.language]
+        suffix = "-default"
+        stem = self.model_id[: -len(suffix)] if self.model_id.endswith(suffix) else self.model_id
+        return f"{stem}.md"
+
+
+@dataclass(frozen=True)
+class SnowballCase:
+    """One exact-language case generated from the Java benchmark authority."""
+
+    case_name: str
+    language: str
+    model_id: str
+    display_language: str
+    direct_available: bool
+    lucene_available: bool
+    homepage_chart: bool
+    homepage_aggregate: bool
+
+
+@dataclass(frozen=True)
+class HomepageGeography:
+    """Documentation-only homepage icon policy for one language."""
+
+    icon: str
+    label: str
+
+
+def _required_gradle_string(text: str, field: str, path: Path) -> str:
+    match = re.search(rf"^\s*{re.escape(field)}\s*=\s*'([^']+)'\s*$", text, re.MULTILINE)
+    if match is None:
+        raise ValueError(f"Missing {field} in {path}")
+    return match.group(1)
+
+
+def _strip_dictionary_remark(line: str) -> str:
+    positions = [position for position in (line.find("#"), line.find("//")) if position >= 0]
+    return line if not positions else line[: min(positions)]
+
+
+def _contains_whitespace(value: str) -> bool:
+    return any(character.isspace() for character in value)
+
+
+def dictionary_size(path: Path) -> tuple[int, int]:
+    """Count parser-compatible rows and exact case-preserved distinct forms.
+
+    Runtime is O(total fields); auxiliary memory is O(distinct forms in the
+    current dictionary). No data from an earlier dictionary is retained.
+    """
+
+    rows = 0
+    forms: set[str] = set()
+    with gzip.open(path, mode="rt", encoding="utf-8", errors="strict", newline="") as source:
+        for physical_line in source:
+            logical_line = _strip_dictionary_remark(physical_line).strip()
+            if not logical_line:
+                continue
+            fields = logical_line.split("\t")
+            stem = fields[0].strip()
+            if not stem or _contains_whitespace(stem):
+                continue
+            rows += 1
+            forms.add(stem)
+            for field in fields[1:]:
+                variant = field.strip()
+                if variant and not _contains_whitespace(variant):
+                    forms.add(variant)
+    return rows, len(forms)
+
+
+def load_model_infos(models_root: Path = MODELS_ROOT) -> list[ModelInfo]:
+    """Load and rank the complete deterministic user-facing model universe."""
+
+    topology = models_root / "model-projects.properties"
+    infos: list[ModelInfo] = []
+    for raw_line in topology.read_text(encoding="utf-8").splitlines():
+        line = raw_line.strip()
+        if not line or line.startswith("#"):
+            continue
+        model_id, separator, role = line.partition("=")
+        if not separator or not model_id or role not in TOPOLOGY_ROLES:
+            raise ValueError(f"Invalid model topology line: {raw_line!r}")
+        project = models_root / model_id
+        build_file = project / "build.gradle"
+        build_text = build_file.read_text(encoding="utf-8")
+        language = _required_gradle_string(build_text, "language", build_file)
+        display_name = _required_gradle_string(build_text, "displayName", build_file)
+        language_name = display_name.split(" — ", 1)[0].removesuffix(" default model")
+        version = (project / "model-version.txt").read_text(encoding="utf-8").strip()
+        dictionary = project / "src" / "modelInput" / "stemmer.gz"
+        rows, distinct_forms = dictionary_size(dictionary)
+        digest = hashlib.sha256()
+        with dictionary.open("rb") as source:
+            for chunk in iter(lambda: source.read(1024 * 1024), b""):
+                digest.update(chunk)
+        infos.append(
+            ModelInfo(
+                model_id=model_id,
+                role=role,
+                language=language,
+                display_name=display_name,
+                language_name=language_name,
+                version=version,
+                right_to_left=bool(
+                    re.search(r"^\s*rightToLeft\s*=\s*true\s*$", build_text, re.MULTILINE)
+                ),
+                dictionary_rows=rows,
+                distinct_forms=distinct_forms,
+                sha256=digest.hexdigest(),
+            )
+        )
+    if not infos:
+        raise ValueError("The active topology contains no user-facing models")
+    return assign_star_tiers(infos)
+
+
+def assign_star_tiers(infos: list[ModelInfo]) -> list[ModelInfo]:
+    """Assign rank-based quintiles without splitting equal-count boundaries."""
+
+    if not infos:
+        raise ValueError("At least one model is required for star-tier assignment")
+    ranked = sorted(infos, key=lambda info: (info.distinct_forms, info.model_id))
+    for index in range(1, len(ranked)):
+        previous_tier = (index - 1) * 5 // len(ranked) + 1
+        current_tier = index * 5 // len(ranked) + 1
+        if (
+            previous_tier != current_tier
+            and ranked[index - 1].distinct_forms == ranked[index].distinct_forms
+        ):
+            raise ValueError(
+                "Equal distinct-form counts would make a rank boundary ambiguous: "
+                f"{ranked[index].distinct_forms}"
+            )
+    stars_by_id = {
+        info.model_id: rank * 5 // len(ranked) + 1 for rank, info in enumerate(ranked)
+    }
+    result = [replace(info, stars=stars_by_id[info.model_id]) for info in infos]
+    return result
+
 LANGUAGE_IDENTITY_WORDS = {
     "CS_CZ": {"CZECH"},
     "DA_DK": {"DANISH"},
@@ -85,6 +258,7 @@ LANGUAGE_IDENTITY_WORDS = {
     "RU_RU": {"RUSSIAN"},
     "ES_ES": {"SPANISH"},
     "SV_SE": {"SWEDISH"},
+    "ST_ZA": {"SESOTHO"},
     "UK_UA": {"UKRAINIAN"},
     "YI": {"YIDDISH"},
 }
@@ -130,6 +304,37 @@ class JmhData:
     auxiliary: dict[Key, dict[str, float]]
 
 
+@dataclass(frozen=True)
+class SpeedProtocol:
+    """Supported published JMH speed protocol inferred from report evidence."""
+
+    samples: int
+    forks: int
+    warmup_iterations: int
+    measurement_iterations: int
+    threads: int = 1
+
+    def page_sentence(self) -> str:
+        """Return the protocol sentence rendered on language pages."""
+
+        return (
+            f"Speed uses JMH average time, {self.warmup_iterations} warmup iterations, "
+            f"{self.measurement_iterations} measurement iterations, {self.forks} independent "
+            f"forks, and {self.threads} thread."
+        )
+
+    def methodology_sentence(self) -> str:
+        """Return the prose fragment rendered by the methodology page."""
+
+        number_words = {1: "one", 3: "three", 5: "five", 7: "seven"}
+        return (
+            f"Published speed and coverage-speed methods use {number_words[self.forks]} "
+            f"independent forks, {number_words[self.warmup_iterations]} one-second warmup "
+            f"iterations and {number_words[self.measurement_iterations]} one-second measurement "
+            f"iterations per fork, {number_words[self.threads]} benchmark thread, and a fixed 6 GiB heap."
+        )
+
+
 @dataclass(frozen=True)
 class Publication:
     """Writes generated documentation or rejects a stale checked-in snapshot."""
@@ -154,12 +359,108 @@ def parse_arguments() -> argparse.Namespace:
     parser.add_argument("--speed", type=Path, required=True)
     parser.add_argument("--coverage-accuracy", type=Path, required=True)
     parser.add_argument("--coverage-speed", type=Path, required=True)
+    parser.add_argument("--quality", type=Path, required=True)
+    parser.add_argument("--snowball-catalog", type=Path, required=True)
     parser.add_argument("--date", required=True)
     parser.add_argument("--release-version", required=True)
     parser.add_argument("--mode", choices=("update", "verify"), default="update")
     return parser.parse_args()
 
 
+def verify_active_inputs(manifest: Path, inputs: dict[str, Path]) -> None:
+    """Reject publication inputs that are not selected by the active manifest."""
+
+    pointers: dict[str, str] = {}
+    for line in manifest.read_text(encoding="utf-8").splitlines():
+        if not line or line.startswith("#"):
+            continue
+        key, separator, value = line.partition("=")
+        if not separator or key in pointers:
+            raise ValueError("Active snapshot manifest contains an invalid or duplicate entry")
+        pointers[key] = value
+    for key, source in inputs.items():
+        selected = pointers.get(key)
+        if selected is None:
+            raise ValueError(f"Active snapshot manifest does not select {key}")
+        expected = (manifest.parent / selected).resolve()
+        if source.resolve() != expected:
+            raise ValueError(
+                f"Publication input {key} is not active: {source} (expected {expected})"
+            )
+
+
+def read_snowball_catalog(path: Path) -> list[SnowballCase]:
+    """Read and strictly validate the catalog emitted by the Java authority."""
+
+    cases: list[SnowballCase] = []
+    with path.open(newline="", encoding="utf-8") as source:
+        reader = csv.DictReader(source)
+        expected = [
+            "Case", "Language", "Model ID", "Display language", "Direct available",
+            "Lucene available", "Homepage chart", "Homepage aggregate",
+        ]
+        if reader.fieldnames != expected:
+            raise ValueError("Snowball catalog schema differs from the Java authority")
+        for row in reader:
+            boolean_values = {
+                name: row[name] == "true"
+                for name in ("Direct available", "Lucene available", "Homepage chart", "Homepage aggregate")
+            }
+            if any(row[name] not in {"true", "false"} for name in boolean_values):
+                raise ValueError("Snowball catalog contains a non-boolean availability value")
+            cases.append(SnowballCase(
+                row["Case"], row["Language"], row["Model ID"], row["Display language"],
+                boolean_values["Direct available"], boolean_values["Lucene available"],
+                boolean_values["Homepage chart"], boolean_values["Homepage aggregate"],
+            ))
+    if not cases or len({case.case_name for case in cases}) != len(cases):
+        raise ValueError("Snowball catalog must contain unique direct cases")
+    chart = [case for case in cases if case.homepage_chart]
+    aggregate = [case for case in cases if case.homepage_aggregate]
+    if len(chart) != len(cases) or len(aggregate) != len(chart) - 1 \
+            or {case.language for case in chart if not case.homepage_aggregate} != {"AR"}:
+        raise ValueError("Snowball homepage policy must chart every direct case and exclude only Arabic from aggregation")
+    if not all(case.direct_available for case in chart):
+        raise ValueError("Every homepage Snowball case must expose the official direct implementation")
+    return cases
+
+
+def read_homepage_geography(
+    path: Path, expected_languages: set[str]
+) -> dict[str, HomepageGeography]:
+    """Read the complete docs-owned country/globe display policy."""
+
+    entries: dict[str, HomepageGeography] = {}
+    with path.open(newline="", encoding="utf-8") as source:
+        reader = csv.DictReader(source)
+        if reader.fieldnames != ["Language", "Icon", "Label"]:
+            raise ValueError("Homepage geography catalog schema is invalid")
+        for row in reader:
+            language = row["Language"]
+            icon = row["Icon"]
+            label = row["Label"]
+            if language in entries or not label:
+                raise ValueError("Homepage geography catalog has a duplicate or blank entry")
+            if icon != "globe" and not re.fullmatch(r"[A-Z]{2}", icon):
+                raise ValueError(f"Invalid homepage geography icon for {language}: {icon}")
+            entries[language] = HomepageGeography(icon, label)
+    if set(entries) != expected_languages:
+        missing = sorted(expected_languages - set(entries))
+        extra = sorted(set(entries) - expected_languages)
+        raise ValueError(
+            f"Homepage geography coverage differs from defaults; missing={missing}, extra={extra}"
+        )
+    return entries
+
+
+def geography_glyph(icon: str) -> str:
+    """Return a country flag emoji or the single transnational globe glyph."""
+
+    if icon == "globe":
+        return "🌐"
+    return "".join(chr(0x1F1E6 + ord(character) - ord("A")) for character in icon)
+
+
 def read_jmh(path: Path) -> JmhData:
     primary: dict[Key, dict[str, str]] = {}
     auxiliary: dict[Key, dict[str, float]] = defaultdict(dict)
@@ -180,12 +481,53 @@ def read_jmh(path: Path) -> JmhData:
     return JmhData(primary, dict(auxiliary))
 
 
-def accuracy(data: JmhData, key: Key) -> tuple[float, float, float]:
+def infer_speed_protocol(data: JmhData, report_name: str) -> SpeedProtocol:
+    """Validate report concurrency and map its sample count to a known protocol."""
+
+    if not data.primary:
+        raise ValueError(f"{report_name} contains no primary JMH speed rows")
+    threads: set[int] = set()
+    samples: set[int] = set()
+    for row in data.primary.values():
+        try:
+            threads.add(int(row["Threads"]))
+            samples.add(int(row["Samples"]))
+        except (KeyError, ValueError) as error:
+            raise ValueError(
+                f"{report_name} contains invalid Threads or Samples fields"
+            ) from error
+    if threads != {1}:
+        raise ValueError(f"{report_name} must use exactly one thread; found {sorted(threads)}")
+    if len(samples) != 1:
+        raise ValueError(f"{report_name} has inconsistent sample counts: {sorted(samples)}")
+    sample_count = next(iter(samples))
+    supported = {
+        15: SpeedProtocol(15, 3, 3, 5),
+        21: SpeedProtocol(21, 3, 5, 7),
+    }
+    if sample_count not in supported:
+        raise ValueError(
+            f"{report_name} uses unsupported sample count {sample_count}; expected 15 or 21"
+        )
+    return supported[sample_count]
+
+
+def percentage(numerator: float, denominator: float) -> float | None:
+    """Return a percentage, or ``None`` when the population is empty."""
+
+    return None if denominator == 0 else 100.0 * numerator / denominator
+
+
+def accuracy(data: JmhData, key: Key) -> tuple[float | None, float | None, float | None]:
     counters = data.auxiliary[key]
     return (
-        100.0 * counters["correctMatches"] / counters["evaluatedTokens"],
-        100.0 * counters["changedCorrectMatches"] / counters["changedEvaluatedTokens"],
-        100.0 * counters["rootPreservedMatches"] / counters["rootEvaluatedTokens"],
+        percentage(counters["correctMatches"], counters["evaluatedTokens"]),
+        percentage(
+            counters["changedCorrectMatches"], counters["changedEvaluatedTokens"]
+        ),
+        percentage(
+            counters["rootPreservedMatches"], counters["rootEvaluatedTokens"]
+        ),
     )
 
 
@@ -193,17 +535,24 @@ def read_corpora(path: Path) -> dict[str, dict[str, object]]:
     corpora: dict[str, dict[str, object]] = {}
     with path.open(newline="", encoding="utf-8") as source:
         for row in csv.DictReader(source):
-            language = row["Language"]
+            model_id = row["Model ID"]
             entry = corpora.setdefault(
-                language,
+                model_id,
                 {
                     "model": row["Model ID"],
                     "version": row["Model version"],
                     "sha256": row["Model SHA-256"],
                     "rows": int(row["Dictionary rows"]),
+                    "distinct": int(row.get("Distinct usable forms") or 0),
                     "total": int(row["Total tokens"]),
                     "roots": int(row["Already-root tokens"]),
                     "changed": int(row["Changed tokens"]),
+                    "timing_basis": row.get("Speed timing workload")
+                    or (
+                        "changed tokens"
+                        if int(row["Changed tokens"]) > 0
+                        else "root-only tokens"
+                    ),
                     "timing": int(row["Speed timing tokens"]),
                     "all_exact": int(row["All exact matches"]),
                     "changed_exact": int(row["Changed exact matches"]),
@@ -212,14 +561,6 @@ def read_corpora(path: Path) -> dict[str, dict[str, object]]:
                 },
             )
             entry["commands"].append((row["Command class"], int(row["Command count"])))
-    if set(corpora) != set(LANGUAGES.values()):
-        raise ValueError(
-            f"Corpus report languages differ from documentation languages: {sorted(corpora)}"
-        )
-    if any(entry["model"] == "pl-pl-polimorf" for entry in corpora.values()):
-        raise ValueError(
-            "The default-model corpus report must not contain pl-pl-polimorf."
-        )
     return corpora
 
 
@@ -232,11 +573,12 @@ def render_corpus_sections(language: str, entry: dict[str, object]) -> str:
     lines = [
         "## Dictionary Corpus",
         "",
-        "| Model ID | Model version | Language | Dictionary rows | Complete quality tokens | Already-root tokens | Changed tokens | JMH timing tokens |",
-        "| --- | --- | --- | ---: | ---: | ---: | ---: | ---: |",
+        "| Model ID | Model version | Language | Dictionary rows | Distinct usable forms | Complete quality tokens | Already-root tokens | Changed tokens | Timing workload | JMH timing tokens |",
+        "| --- | --- | --- | ---: | ---: | ---: | ---: | ---: | --- | ---: |",
         f"| `{entry['model']}` | `{entry['version']}` | `{language}` | {format_integer(int(entry['rows']))} | "
-        f"{format_integer(total)} | {format_integer(int(entry['roots']))} | "
-        f"{format_integer(int(entry['changed']))} | {format_integer(int(entry['timing']))} |",
+        f"{format_integer(int(entry['distinct']))} | {format_integer(total)} | {format_integer(int(entry['roots']))} | "
+        f"{format_integer(int(entry['changed']))} | {entry['timing_basis']} | "
+        f"{format_integer(int(entry['timing']))} |",
         "",
         "## Radixor Patch Command Distribution",
         "",
@@ -263,8 +605,12 @@ def render_corpus_sections(language: str, entry: dict[str, object]) -> str:
     return "\n".join(lines) + "\n\n"
 
 
-def rounded_accuracy(values: tuple[float, float, float]) -> tuple[str, str, str]:
-    return tuple(f"{value:.3f}" for value in values)
+def rounded_accuracy(
+    values: tuple[float | None, float | None, float | None]
+) -> tuple[str, str, str]:
+    """Format exact-root percentages without inventing empty-population values."""
+
+    return tuple("n/a" if value is None else f"{value:.3f}%" for value in values)
 
 
 def words(value: str) -> set[str]:
@@ -352,11 +698,13 @@ def select_accuracy_key(
     return ranked[0][1]
 
 
-def corpus_accuracy(entry: dict[str, object]) -> tuple[float, float, float]:
+def corpus_accuracy(
+    entry: dict[str, object]
+) -> tuple[float | None, float | None, float | None]:
     return (
-        100.0 * int(entry["all_exact"]) / int(entry["total"]),
-        100.0 * int(entry["changed_exact"]) / int(entry["changed"]),
-        100.0 * int(entry["root_exact"]) / int(entry["roots"]),
+        percentage(int(entry["all_exact"]), int(entry["total"])),
+        percentage(int(entry["changed_exact"]), int(entry["changed"])),
+        percentage(int(entry["root_exact"]), int(entry["roots"])),
     )
 
 
@@ -373,7 +721,8 @@ def update_accuracy_table(
     for line in section.splitlines():
         cells = [cell.strip() for cell in line.split("|")[1:-1]]
         measured = len(cells) == 5 and all(
-            re.fullmatch(r"\d+\.\d{3}%", cell) for cell in cells[1:4]
+            cell == "n/a" or re.fullmatch(r"\d+\.\d{3}%", cell)
+            for cell in cells[1:4]
         )
         pending = len(cells) == 5 and all(cell == "pending" for cell in cells[1:4])
         partial_pending = (
@@ -398,7 +747,7 @@ def update_accuracy_table(
             else:
                 key = select_accuracy_key(cells[0], language, new_data)
                 values = rounded_accuracy(accuracy(new_data, key))
-            cells[1:4] = [f"{value}%" for value in values]
+            cells[1:4] = list(values)
             line = "| " + " | ".join(cells) + " |"
         output.append(line)
     replacement = "\n".join(output).rstrip() + "\n\n"
@@ -406,23 +755,29 @@ def update_accuracy_table(
 
 
 def method_and_parameter(display: str) -> tuple[str, str]:
-    match = re.fullmatch(r"([A-Za-z0-9]+)(?:\[([A-Z_]+)])?", display)
+    match = re.fullmatch(r"([A-Za-z0-9]+)(?:\[([A-Za-z0-9_-]+)])?", display)
     if not match:
         raise ValueError(f"Unsupported benchmark method display: {display}")
     return match.group(1), match.group(2) or ""
 
 
 def speed_matches(display: str, data: JmhData, language: str) -> list[Key]:
-    method, language_case = method_and_parameter(display)
+    method, parameter = method_and_parameter(display)
     matches = [
         key
         for key, row in data.primary.items()
         if key.method == method
-        and (not language_case or key.parameter("languageCaseName") == language_case)
+        and (
+            not parameter
+            or key.parameter(
+                "modelId" if "-" in parameter else "languageCaseName"
+            )
+            == parameter
+        )
         and key not in data.auxiliary
         and row["Unit"] == "ns/op"
     ]
-    if language_case or len(matches) <= 1:
+    if parameter or len(matches) <= 1:
         return matches
     language_words = LANGUAGE_IDENTITY_WORDS[language]
     return [
@@ -497,11 +852,494 @@ def update_speed_table(
     return text[:start] + replacement + text[end:]
 
 
+def rating_markup(info: ModelInfo) -> str:
+    """Render a five-position accessible relative dictionary-size rating."""
+
+    glyphs = "★" * info.stars + "☆" * (5 - info.stars)
+    count = format_integer(info.distinct_forms)
+    label = (
+        f"Relative dictionary size {info.stars} of 5; "
+        f"{count} distinct usable word forms"
+    )
+    return (
+        f'<span class="dictionary-rating" role="img" aria-label="{label}" '
+        f'title="{label}">{glyphs}</span>'
+    )
+
+
+def _rating_legend(info: ModelInfo) -> str:
+    return (
+        "<!-- DICTIONARY-SIZE-RATING:START -->\n"
+        f"Dictionary size: {rating_markup(info)}. The exact count is "
+        f"**{format_integer(info.distinct_forms)} distinct usable word forms** after "
+        "parser-compatible filtering and exact, case-preserved deduplication. Stars rank "
+        "dictionary size relative to all active user-facing models in five nearly equal groups; "
+        "they do **not** measure linguistic quality or benchmark accuracy.\n"
+        "<!-- DICTIONARY-SIZE-RATING:END -->"
+    )
+
+
+def _with_rating(text: str, info: ModelInfo) -> str:
+    lines = text.splitlines()
+    if not lines or not lines[0].startswith("# "):
+        raise ValueError(f"Benchmark page for {info.model_id} has no H1")
+    heading = re.sub(r"\s*<span class=\"dictionary-rating\".*?</span>\s*$", "", lines[0])
+    lines[0] = f"{heading} {rating_markup(info)}"
+    text = "\n".join(lines) + ("\n" if text.endswith("\n") else "")
+    legend = _rating_legend(info)
+    if "<!-- DICTIONARY-SIZE-RATING:START -->" in text:
+        return re.sub(
+            r"<!-- DICTIONARY-SIZE-RATING:START -->.*?<!-- DICTIONARY-SIZE-RATING:END -->",
+            legend,
+            text,
+            count=1,
+            flags=re.DOTALL,
+        )
+    first_paragraph_end = text.find("\n\n", text.find("\n") + 1)
+    if first_paragraph_end < 0:
+        raise ValueError(f"Benchmark page for {info.model_id} has no introduction")
+    return text[:first_paragraph_end] + "\n\n" + legend + text[first_paragraph_end:]
+
+
+def _new_page(info: ModelInfo) -> str:
+    snowball = SNOWBALL_CASES.get(info.language)
+    accuracy_rows = [
+        "| Radixor | pending | pending | pending | Exact model-ID benchmark; measurement pending. |"
+    ]
+    speed_rows = [
+        f"| Radixor | `radixor[{info.model_id}]` | pending | pending | pending | pending | "
+        "Canonical model timing workload; measurement pending. |"
+    ]
+    if snowball is not None:
+        case_name, has_lucene = snowball
+        accuracy_rows.append(
+            "| Official Snowball direct | pending | pending | pending | "
+            "Official Snowball 3.1.0 generated Java stemmer; measurement pending. |"
+        )
+        speed_rows.append(
+            f"| Official Snowball direct | `snowballDirect[{case_name}]` | pending | pending | pending | pending | "
+            "Official generated Java stemmer; measurement pending. |"
+        )
+        if has_lucene:
+            accuracy_rows.append(
+                "| Lucene SnowballFilter | pending | pending | pending | "
+                "Lucene integration of the matching Snowball algorithm; measurement pending. |"
+            )
+            speed_rows.append(
+                f"| Lucene SnowballFilter | `luceneSnowballFilter[{case_name}]` | pending | pending | pending | pending | "
+                "Lucene TokenStream integration; measurement pending. |"
+            )
+    direction = "right-to-left" if info.right_to_left else "left-to-right"
+    return f"""# {info.language_name} Stemmer Benchmarks
+
+This page reports dictionary corpus, exact-root agreement, and runtime evidence for the independently available `{info.model_id}` {info.language_name} model.
+
+{_rating_legend(info)}
+
+The language metadata declares {direction} writing. New benchmark measurements are intentionally shown as pending until the canonical published runners produce complete reports; no values are inferred or fabricated.
+
+All speed values are environment-specific and await publication on the hardware and JVM listed in the [benchmark overview](../index.md). Speed benchmark operations use the canonical dictionary-derived timing workload. Accuracy uses the complete Radixor dictionary for the language.
+
+<!-- BENCHMARK-EVIDENCE-MAP:START -->
+!!! info "How to read this page"
+    Start with the exact dictionary size, then read exact-root agreement and runtime together after the canonical reports are published. Dictionary-size stars are contextual metadata, not an accuracy result.
+<!-- BENCHMARK-EVIDENCE-MAP:END -->
+
+## Dictionary Corpus
+
+| Model ID | Model version | Language | Dictionary rows | Distinct usable word forms |
+| --- | --- | --- | ---: | ---: |
+| `{info.model_id}` | `{info.version}` | `{info.language}` | {format_integer(info.dictionary_rows)} | {format_integer(info.distinct_forms)} |
+
+## Accuracy
+
+Accuracy will use the complete dictionary and report exact agreement with the dictionary root. The pending cells remain until a canonical report identifies this exact model and candidate.
+
+| Stemmer | All exact | Changed exact | Root preserved | Note |
+| --- | ---: | ---: | ---: | --- |
+{chr(10).join(accuracy_rows)}
+
+## Speed
+
+Speed uses JMH average time, 3 warmup iterations, 5 measurement iterations, 3 independent forks, and 1 thread.
+
+Speed will use JMH average time over the canonical timing workload. Changed dictionary tokens are preferred; a root-only dictionary uses its complete root-preservation corpus. The selected population is repeated only when needed to reach the timing minimum. Relative factors will use the Radixor row on this page as the baseline.
+
+| Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor | Note |
+| --- | --- | ---: | ---: | ---: | ---: | --- |
+{chr(10).join(speed_rows)}
+
+## Interpretation Notes
+
+- The star tier reflects only relative distinct-form count among the active user-facing models.
+- Pending values indicate that the canonical benchmark run has not yet published measurements for this page.
+- Runtime and exact-root agreement describe different properties and must be interpreted together.
+
+<!-- DICTIONARY-GENERALIZATION:START -->
+<!-- Reserved for deterministic dictionary-family generalization publication. -->
+<!-- DICTIONARY-GENERALIZATION:END -->
+
+<!-- EDIT-COST-GENERALIZATION:START -->
+<!-- Reserved for deterministic edit-cost publication. -->
+<!-- EDIT-COST-GENERALIZATION:END -->
+
+<!-- STEMMING-QUALITY:START -->
+<!-- Reserved for deterministic pairwise linguistic-quality publication. -->
+<!-- STEMMING-QUALITY:END -->
+"""
+
+
+def _polimorf_section(
+    info: ModelInfo,
+    corpus: dict[str, object] | None,
+    accuracy_data: JmhData | None,
+    speed_data: JmhData | None,
+) -> str:
+    """Render the optional PoliMorf model without changing the Polish default."""
+
+    if corpus is None:
+        corpus_row = (
+            f"| `{info.model_id}` | `{info.version}` | `PL_PL` | {format_integer(info.dictionary_rows)} | "
+            f"{format_integer(info.distinct_forms)} | pending | pending | pending | pending | pending |"
+        )
+        accuracy_rows = [
+            "| Radixor PoliMorf | pending | pending | pending | Exact optional model; measurement pending. |",
+            "| Lucene MorfologikFilter | pending | pending | pending | Matching Polish dictionary adapter; measurement pending. |",
+            "| Official Snowball direct | pending | pending | pending | Matching Polish Snowball algorithm; measurement pending. |",
+        ]
+        speed_rows = [
+            f"| Radixor PoliMorf | `radixor[{info.model_id}]` | pending | pending | pending | pending | Exact optional-model baseline; measurement pending. |",
+            "| Lucene MorfologikFilter | `polishPolimorfLuceneMorfologikFilter` | pending | pending | pending | pending | PoliMorf-derived corpus; measurement pending. |",
+        ]
+    else:
+        if accuracy_data is None or speed_data is None:
+            raise ValueError("Measured PoliMorf corpus requires accuracy and speed reports.")
+        corpus_row = (
+            f"| `{info.model_id}` | `{corpus['version']}` | `PL_PL` | {format_integer(int(corpus['rows']))} | "
+            f"{format_integer(int(corpus['distinct']))} | {format_integer(int(corpus['total']))} | "
+            f"{format_integer(int(corpus['roots']))} | {format_integer(int(corpus['changed']))} | "
+            f"{corpus['timing_basis']} | {format_integer(int(corpus['timing']))} |"
+        )
+        accuracy_rows = []
+        accuracy_specs = (
+            ("Radixor PoliMorf", "PoliMorf Radixor", "Exact optional model."),
+            ("PoliMorf Lucene MorfologikFilter", "PoliMorf Lucene MorfologikFilter", "Matching Polish dictionary adapter."),
+            ("PoliMorf Official Snowball direct", "PoliMorf Official Snowball direct", "Matching Polish Snowball algorithm."),
+        )
+        for display, lookup, note in accuracy_specs:
+            values = rounded_accuracy(
+                accuracy(accuracy_data, select_accuracy_key(lookup, "PL_PL", accuracy_data))
+            )
+            accuracy_rows.append(
+                f"| {display} | {values[0]} | {values[1]} | {values[2]} | {note} |"
+            )
+
+        timing_tokens = int(corpus["timing"])
+        radixor_key = select_speed_key(f"radixor[{info.model_id}]", speed_data, "PL_PL")
+        morfologik_key = select_speed_key("polishPolimorfLuceneMorfologikFilter", speed_data, "PL_PL")
+        radixor_score = float(speed_data.primary[radixor_key]["Score"])
+        speed_rows = []
+        for display, method, key, note in (
+            ("Radixor PoliMorf", f"radixor[{info.model_id}]", radixor_key, "Exact optional-model baseline."),
+            ("Lucene MorfologikFilter", "polishPolimorfLuceneMorfologikFilter", morfologik_key, "PoliMorf-derived corpus."),
+        ):
+            row = speed_data.primary[key]
+            score = float(row["Score"])
+            error = float(row["Score Error (99.9%)"])
+            speed_rows.append(
+                f"| {display} | `{method}` | {score / 1_000_000.0:.3f} | "
+                f"{error / 1_000_000.0:.3f} | {score / timing_tokens:.1f} | "
+                f"{score / radixor_score:.3f} | {note} |"
+            )
+
+    return f"""{POLIMORF_START}
+
+## Optional PoliMorf Model
+
+PoliMorf is an independently available optional Polish model and is not the `PL_PL` default or a member of either standard aggregate. Its evidence is kept separate from the UniMorph default above.
+
+Optional dictionary size: {rating_markup(info)}. The exact count is **{format_integer(info.distinct_forms)} distinct usable word forms**; the rating is relative size, not linguistic quality.
+
+### PoliMorf Corpus
+
+| Model ID | Model version | Language | Dictionary rows | Distinct usable forms | Complete quality tokens | Already-root tokens | Changed tokens | Timing workload | JMH timing tokens |
+| --- | --- | --- | ---: | ---: | ---: | ---: | ---: | --- | ---: |
+{corpus_row}
+
+### PoliMorf Exact-root Accuracy
+
+| Stemmer | All exact | Changed exact | Root preserved | Note |
+| --- | ---: | ---: | ---: | --- |
+{chr(10).join(accuracy_rows)}
+
+### PoliMorf Speed
+
+| Stemmer | Benchmark method | Score ms/op | Error ms | ns/token | Relative vs Radixor PoliMorf | Note |
+| --- | --- | ---: | ---: | ---: | ---: | --- |
+{chr(10).join(speed_rows)}
+
+{POLIMORF_END}"""
+
+
+def prepare_polimorf_section(
+    docs_root: Path, info: ModelInfo, publication: Publication
+) -> None:
+    """Insert the pending optional section once while preserving measured content."""
+
+    path = docs_root / "benchmarks" / "languages" / "polish.md"
+    text = path.read_text(encoding="utf-8")
+    if POLIMORF_START not in text:
+        marker = "<!-- DICTIONARY-GENERALIZATION:START -->"
+        text = text.replace(
+            marker, _polimorf_section(info, None, None, None) + "\n\n" + marker, 1
+        )
+    publication.write(path, text)
+
+
+def update_polimorf_section(
+    docs_root: Path,
+    info: ModelInfo,
+    corpus: dict[str, object] | None,
+    accuracy_data: JmhData,
+    speed_data: JmhData,
+    publication: Publication,
+) -> None:
+    """Update the optional section from exact-model reports when available."""
+
+    path = docs_root / "benchmarks" / "languages" / "polish.md"
+    text = path.read_text(encoding="utf-8")
+    replacement = _polimorf_section(info, corpus, accuracy_data, speed_data)
+    text, count = re.subn(
+        re.escape(POLIMORF_START) + r".*?" + re.escape(POLIMORF_END),
+        replacement,
+        text,
+        count=1,
+        flags=re.DOTALL,
+    )
+    if count != 1:
+        raise ValueError("The Polish page has no optional PoliMorf section.")
+    publication.write(path, text)
+
+
+def prepare_language_pages(
+    docs_root: Path, model_infos: list[ModelInfo], publication: Publication
+) -> dict[str, ModelInfo]:
+    """Create deterministic shells and add ratings without replacing rich sections."""
+
+    directory = docs_root / "benchmarks" / "languages"
+    defaults = {
+        info.language: info for info in model_infos if info.role != "optional"
+    }
+    if not defaults:
+        raise ValueError("The active topology contains no language defaults")
+    for info in defaults.values():
+        path = directory / info.page_name
+        if path.is_file():
+            content = _with_rating(path.read_text(encoding="utf-8"), info)
+            if (
+                "New benchmark measurements are intentionally shown as pending" in content
+                and "All speed values are environment-specific" not in content
+            ):
+                anchor = (
+                    "New benchmark measurements are intentionally shown as pending until the canonical "
+                    "published runners produce complete reports; no values are inferred or fabricated."
+                )
+                identity = (
+                    "All speed values are environment-specific and await measurement on the hardware and JVM "
+                    "listed in the [benchmark overview](../index.md). Speed benchmark operations use the canonical "
+                    "timing workload. Accuracy uses the complete Radixor dictionary for the language."
+                )
+                content = content.replace(anchor, anchor + "\n\n" + identity, 1)
+            content = content.replace(
+                "Speed will use JMH average time over changed dictionary tokens, repeated only when needed "
+                "to reach the timing minimum. Relative factors will use the Radixor row on this page as the baseline.",
+                "Speed will use JMH average time over the canonical timing workload. Changed tokens are preferred; "
+                "a root-only dictionary uses its complete root-preservation corpus. Smaller populations are repeated "
+                "deterministically to the timing minimum. Relative factors use the Radixor row as the baseline.",
+            )
+            content = content.replace(
+                "Canonical changed-token model benchmark; measurement pending.",
+                "Canonical model timing workload; measurement pending.",
+            )
+        else:
+            content = _with_rating(_new_page(info), info)
+        publication.write(path, content)
+    return defaults
+
+
+def mark_new_page_measured(text: str) -> str:
+    """Replace new-page provisional language once canonical reports are present."""
+
+    replacements = {
+        " New benchmark measurements are intentionally shown as pending until the canonical "
+        "published runners produce complete reports; no values are inferred or fabricated.": "",
+        "Accuracy will use the complete dictionary and report exact agreement with the "
+        "dictionary root. The pending cells remain until a canonical report identifies this "
+        "exact model and candidate.": (
+            "Accuracy uses the complete dictionary and reports exact agreement with the "
+            "dictionary root for each identified model and candidate."
+        ),
+        "Speed will use JMH average time over the canonical timing workload.": (
+            "Speed uses JMH average time over the canonical timing workload."
+        ),
+        "measurement pending.": "measured in this snapshot.",
+        "- Pending values indicate that the canonical benchmark run has not yet published "
+        "measurements for this page.": (
+            "- Values shown above come from the identified canonical benchmark snapshot."
+        ),
+    }
+    for provisional, measured in replacements.items():
+        text = text.replace(provisional, measured)
+    if re.search(r"\bpending\b", text, flags=re.IGNORECASE):
+        raise ValueError("A measured new language page still contains pending content.")
+    return text
+
+
+def validate_evidence_fragment_links(text: str, path: Path) -> None:
+    """Reject fragment links in the evidence map that have no page heading."""
+
+    match = re.search(
+        r"<!-- BENCHMARK-EVIDENCE-MAP:START -->(.*?)"
+        r"<!-- BENCHMARK-EVIDENCE-MAP:END -->",
+        text,
+        flags=re.DOTALL,
+    )
+    if match is None:
+        raise ValueError(f"No benchmark evidence map found in {path}.")
+    anchors = {
+        re.sub(r"\s+", "-", re.sub(r"[^a-z0-9\s-]", "", heading.lower()).strip())
+        for heading in re.findall(r"^#{1,6}\s+(.+?)\s*$", text, flags=re.MULTILINE)
+    }
+    missing = sorted(
+        fragment
+        for fragment in re.findall(r"\]\(#([^)]+)\)", match.group(1))
+        if fragment not in anchors
+    )
+    if missing:
+        raise ValueError(
+            f"Benchmark evidence map in {path} references missing anchors: "
+            + ", ".join(missing)
+        )
+
+
+def update_model_indexes(
+    docs_root: Path,
+    model_infos: list[ModelInfo],
+    defaults: dict[str, ModelInfo],
+    publication: Publication,
+) -> None:
+    """Publish complete rated language and model indexes."""
+
+    optional_by_language = {
+        info.language: info for info in model_infos if info.role == "optional"
+    }
+    language_index = docs_root / "benchmarks" / "languages" / "index.md"
+    text = language_index.read_text(encoding="utf-8")
+    introduction_start = text.index("This section ")
+    introduction_end = text.index("\n\n", introduction_start)
+    introduction = (
+        f"This section covers all {len(defaults)} registered language defaults. The original 20 pages retain "
+        "their complete measured and experimental sections; newly added pages publish corpus, "
+        "accuracy, and speed evidence from the current checked-in canonical reports."
+    )
+    text = text[:introduction_start] + introduction + text[introduction_end:]
+    start = text.index("| Language |", text.index("## Languages"))
+    end = text.index("\n\n", start)
+    lines = [
+        "| Language | Java enum | Model ID | Relative dictionary size | Availability | Benchmark page |",
+        "| --- | --- | --- | --- | --- | --- |",
+    ]
+    for info in sorted(defaults.values(), key=lambda item: (item.language_name, item.model_id)):
+        model_cell = f"`{info.model_id}`"
+        availability = "Individual artifact; outside standard aggregate" if info.role == "standalone" else "Standard aggregate"
+        optional = optional_by_language.get(info.language)
+        if optional is not None:
+            model_cell += f"; optional `{optional.model_id}` ({rating_markup(optional)})"
+            availability += "; optional model is individual only"
+        lines.append(
+            f"| {info.language_name} | `{info.language}` | {model_cell} | {rating_markup(info)} "
+            f"({format_integer(info.distinct_forms)}) | {availability} | "
+            f"[{info.language_name}]({info.page_name}) |"
+        )
+    publication.write(language_index, text[:start] + "\n".join(lines) + text[end:])
+
+    built_in = docs_root / "built-in-languages.md"
+    text = built_in.read_text(encoding="utf-8")
+    heading = (
+        "## Defaults and variants"
+        if "## Defaults and variants" in text
+        else "## Registered languages and models"
+    )
+    heading_start = text.index(heading)
+    next_heading = text.index("## The Polish dual-model case", heading_start)
+    lines = [
+        "## Registered languages and models",
+        "",
+        "The standard Java and Python aggregates remain fixed at the original 20 `default` entries. "
+        "Every `standalone` entry is a registered Java language default and an individually publishable "
+        "artifact, but is not installed by either standard aggregate. The optional PoliMorf model is also individual only.",
+        "",
+        "| Language | Java enum | Model ID | Topology role | Relative dictionary size |",
+        "| --- | --- | --- | --- | --- |",
+    ]
+    for info in sorted(model_infos, key=lambda item: (item.language_name, item.model_id)):
+        lines.append(
+            f"| {info.language_name} | `{info.language}` | `{info.model_id}` | `{info.role}` | "
+            f"{rating_markup(info)} ({format_integer(info.distinct_forms)} distinct forms) |"
+        )
+    lines.extend(
+        [
+            "",
+            "The maintained table deliberately avoids duplicating mutable provenance fields. Those values come from model module metadata and the generated model catalog.",
+            "",
+        ]
+    )
+    publication.write(built_in, text[:heading_start] + "\n".join(lines) + text[next_heading:])
+
+
+def update_mkdocs_language_nav(
+    docs_root: Path, defaults: dict[str, ModelInfo], publication: Publication
+) -> None:
+    """Keep the MkDocs language navigation complete and deterministic."""
+
+    path = docs_root.parent / "mkdocs.yml"
+    text = path.read_text(encoding="utf-8")
+    start_marker = "     - Language Comparisons:\n"
+    start = text.index(start_marker)
+    end = text.index("     - Methods and Reproduction:\n", start)
+    lines = [start_marker.rstrip("\n"), "        - Overview: benchmarks/languages/index.md"]
+    for info in sorted(defaults.values(), key=lambda item: (item.language_name, item.model_id)):
+        lines.append(
+            f"        - {info.language_name}: benchmarks/languages/{info.page_name}"
+        )
+    block = "\n".join(lines) + "\n"
+    publication.write(path, text[:start] + block + text[end:])
+
+
+def update_speed_protocol_documentation(
+    docs_root: Path, protocol: SpeedProtocol, publication: Publication
+) -> None:
+    """Publish the protocol evidenced by the selected speed reports."""
+
+    path = docs_root / "benchmarks" / "reference" / "methodology.md"
+    text = path.read_text(encoding="utf-8")
+    text, count = re.subn(
+        r"Published speed and coverage-speed methods use .*?fixed 6 GiB heap\.",
+        protocol.methodology_sentence(),
+        text,
+        count=1,
+    )
+    if count != 1:
+        raise ValueError("No published speed protocol statement found in methodology.md")
+    publication.write(path, text)
+
+
 def update_language_pages(
     docs_root: Path,
     corpora: dict[str, dict[str, object]],
     accuracy_data: JmhData,
     speed_data: JmhData,
+    speed_protocol: SpeedProtocol,
     publication: Publication,
     date: str,
     release_version: str,
@@ -509,13 +1347,19 @@ def update_language_pages(
     directory = docs_root / "benchmarks" / "languages"
     for file_name, language in LANGUAGES.items():
         path = directory / file_name
+        if language not in corpora:
+            continue
         text = path.read_text(encoding="utf-8")
+        was_pending_shell = (
+            "New benchmark measurements are intentionally shown as pending" in text
+        )
+        timing_basis = str(corpora[language]["timing_basis"])
         identity = (
             "All speed values are environment-specific and were measured on the hardware and JVM "
             "listed in the [benchmark overview](../index.md). The command distribution, exact-root "
             f"accuracy, and speed tables belong to the published {date} Radixor/Java "
-            f"`{release_version}` snapshot. Speed benchmark operations process changed dictionary "
-            "tokens only. Accuracy uses the complete Radixor dictionary for the language."
+            f"`{release_version}` snapshot. Speed benchmark operations process {timing_basis}. "
+            "Accuracy uses the complete Radixor dictionary for the language."
         )
         text, count = re.subn(
             r"All speed values are environment-specific.*?Accuracy uses the complete Radixor "
@@ -527,17 +1371,36 @@ def update_language_pages(
         )
         if count != 1:
             raise ValueError(f"No benchmark identity paragraph found in {path}.")
+        has_rich_experiments = all(
+            heading in text
+            for heading in (
+                "## Dictionary-Family Generalization Conclusion",
+                "## Edit Costs and Dictionary-Knowledge Generalization",
+                "## Stemming Quality",
+            )
+        )
+        if has_rich_experiments:
+            guidance = (
+                "    Start with the [corpus](#dictionary-corpus) and "
+                "[patch-command distribution](#radixor-patch-command-distribution), then compare "
+                "[exact-root agreement](#accuracy) with [runtime](#speed). The "
+                "[dictionary-family experiment](#dictionary-family-generalization-conclusion), "
+                "[edit-cost experiment](#edit-costs-and-dictionary-knowledge-generalization), and "
+                "[pairwise linguistic evaluation](#stemming-quality) answer separate questions. "
+                "Their 10–90% curves use independent frozen protocols and must not be substituted "
+                "for one another."
+            )
+        else:
+            guidance = (
+                "    Start with the [corpus](#dictionary-corpus) and "
+                "[patch-command distribution](#radixor-patch-command-distribution), then compare "
+                "[exact-root agreement](#accuracy) with [runtime](#speed). Dictionary-size stars "
+                "are contextual metadata, not an accuracy result."
+            )
         evidence = (
             "<!-- BENCHMARK-EVIDENCE-MAP:START -->\n"
             "!!! info \"How to read this page\"\n"
-            "    Start with the [corpus](#dictionary-corpus) and "
-            "[patch-command distribution](#radixor-patch-command-distribution), then compare "
-            "[exact-root agreement](#accuracy) with [runtime](#speed). The "
-            "[dictionary-family experiment](#dictionary-family-generalization-conclusion), "
-            "[edit-cost experiment](#edit-costs-and-dictionary-knowledge-generalization), and "
-            "[pairwise linguistic evaluation](#stemming-quality) answer separate questions. "
-            "Their 10–90% curves use independent frozen protocols and must not be substituted "
-            "for one another.\n"
+            f"{guidance}\n"
             "<!-- BENCHMARK-EVIDENCE-MAP:END -->"
         )
         if "<!-- BENCHMARK-EVIDENCE-MAP:START -->" in text:
@@ -564,14 +1427,21 @@ def update_language_pages(
             + render_corpus_sections(language, corpora[language])
             + text[accuracy_start:]
         )
-        text = re.sub(
+        text, protocol_count = re.subn(
             r"Speed uses JMH average time, \d+ warmup iterations, \d+ measurement iterations, "
             r"\d+ (?:independent )?forks?, and 1 thread\.",
-            "Speed uses JMH average time, 5 warmup iterations, 7 measurement iterations, "
-            "3 independent forks, and 1 thread.",
+            speed_protocol.page_sentence(),
             text,
             count=1,
         )
+        if protocol_count == 0:
+            speed_heading_end = text.index("\n\n", text.index("## Speed")) + 2
+            text = (
+                text[:speed_heading_end]
+                + speed_protocol.page_sentence()
+                + "\n\n"
+                + text[speed_heading_end:]
+            )
         text = update_accuracy_table(text, accuracy_data, language, corpora[language])
         text = update_speed_table(
             text, speed_data, int(corpora[language]["timing"]), language
@@ -611,38 +1481,48 @@ def update_language_pages(
                 speed_start = text.index("## Speed")
                 table_start = text.index("| Stemmer |", speed_start)
                 text = text[:table_start] + comparison + "\n\n" + text[table_start:]
+        if was_pending_shell:
+            text = mark_new_page_measured(text)
+        validate_evidence_fragment_links(text, path)
         publication.write(path, text)
 
 
 def update_corpora_reference(
     docs_root: Path,
-    corpora: dict[str, dict[str, object]],
+    model_infos: list[ModelInfo],
+    corpora_by_model: dict[str, dict[str, object]],
     publication: Publication,
 ) -> None:
     path = docs_root / "benchmarks" / "reference" / "corpora.md"
     text = path.read_text(encoding="utf-8")
     original_header = "| Language resource |"
     current_header = "| Default model ID |"
+    complete_header = "| Model ID | Role |"
     if original_header in text:
         table_start = text.index(original_header)
     elif current_header in text:
         table_start = text.index(current_header)
+    elif complete_header in text:
+        table_start = text.index(complete_header)
     else:
         raise ValueError(
             "The corpora reference contains no recognized corpus-table header."
         )
     table_end = text.index("\n\n", table_start)
     lines = [
-        "| Default model ID | Version | SHA-256 | Language | Dictionary rows | Total tokens | Already-root tokens | Changed tokens | Speed timing tokens |",
-        "| --- | --- | --- | --- | ---: | ---: | ---: | ---: | ---: |",
+        "| Model ID | Role | Version | SHA-256 | Language | Dictionary rows | Distinct usable forms | Total benchmark tokens | Changed benchmark tokens | Timing workload | Speed timing tokens |",
+        "| --- | --- | --- | --- | --- | ---: | ---: | ---: | ---: | --- | ---: |",
     ]
-    for language in LANGUAGES.values():
-        entry = corpora[language]
+    for info in sorted(model_infos, key=lambda item: item.model_id):
+        entry = corpora_by_model.get(info.model_id)
+        total = format_integer(int(entry["total"])) if entry is not None else "pending"
+        changed = format_integer(int(entry["changed"])) if entry is not None else "pending"
+        timing_basis = str(entry["timing_basis"]) if entry is not None else "pending"
+        timing = format_integer(int(entry["timing"])) if entry is not None else "pending"
         lines.append(
-            f"| `{entry['model']}` | `{entry['version']}` | `{entry['sha256']}` | `{language}` | "
-            f"{format_integer(int(entry['rows']))} | "
-            f"{format_integer(int(entry['total']))} | {format_integer(int(entry['roots']))} | "
-            f"{format_integer(int(entry['changed']))} | {format_integer(int(entry['timing']))} |"
+            f"| `{info.model_id}` | `{info.role}` | `{info.version}` | `{info.sha256}` | `{info.language}` | "
+            f"{format_integer(info.dictionary_rows)} | {format_integer(info.distinct_forms)} | "
+            f"{total} | {changed} | {timing_basis} | {timing} |"
         )
     replacement = "\n".join(lines)
     publication.write(path, text[:table_start] + replacement + text[table_end:])
@@ -780,18 +1660,292 @@ def update_coverage(
     publication.write(index, index_text[:key_start] + key_section + index_text[key_end:])
 
 
+def quality_homepage_counts(
+    path: Path, expected_languages: set[str]
+) -> tuple[int, int, int, int]:
+    """Return defined matrices, Radixor wins, comparator coverage, and undefined matrices."""
+
+    groups: dict[tuple[str, str], list[tuple[str, float]]] = defaultdict(list)
+    all_groups: set[tuple[str, str]] = set()
+    with path.open(newline="", encoding="utf-8") as source:
+        for row in csv.DictReader(source):
+            if row["Output policy"] != "PRIMARY_OUTPUT":
+                continue
+            key = (row["Language"], row["Dictionary mode"])
+            all_groups.add(key)
+            if not row["Balanced accuracy"]:
+                continue
+            groups[key].append(
+                (row["Stemmer"], float(row["Balanced accuracy"]))
+            )
+    report_languages = {language for language, _ in all_groups}
+    if report_languages != expected_languages or len(all_groups) != len(expected_languages) * 2:
+        raise ValueError("Active quality snapshot must contain two matrices for every active default")
+    wins = 0
+    compared = 0
+    for rows in groups.values():
+        best = max(score for _, score in rows)
+        leaders = [stemmer for stemmer, score in rows if score == best]
+        if len(rows) > 1:
+            compared += 1
+        if len(leaders) == 1 and leaders[0].endswith("_RADIXOR"):
+            wins += 1
+    return len(groups), wins, compared, len(all_groups) - len(groups)
+
+
+def render_snowball_homepage(
+    cases: list[SnowballCase],
+    corpora_by_model: dict[str, dict[str, object]],
+    speed_data: JmhData,
+    snapshot_date: str,
+) -> str:
+    """Render the complete direct Snowball comparison and accessible fallback."""
+
+    rows: list[tuple[SnowballCase, float, float, float, float, str]] = []
+    for case in cases:
+        if not case.homepage_chart:
+            continue
+        corpus = corpora_by_model.get(case.model_id)
+        if corpus is None:
+            raise ValueError(f"Snowball case {case.case_name} has no exact model corpus")
+        timing_tokens = int(corpus["timing"])
+        timing_basis = str(corpus["timing_basis"])
+        snowball_key = select_speed_key(f"snowballDirect[{case.case_name}]", speed_data, case.language)
+        radixor_key = select_speed_key(f"radixor[{case.model_id}]", speed_data, case.language)
+        snowball_row = speed_data.primary[snowball_key]
+        radixor_row = speed_data.primary[radixor_key]
+        snowball_ns = float(snowball_row["Score"]) / timing_tokens
+        snowball_error = float(snowball_row["Score Error (99.9%)"]) / timing_tokens
+        radixor_ns = float(radixor_row["Score"]) / timing_tokens
+        ratio = snowball_ns / radixor_ns
+        if not all(math.isfinite(value) and value > 0.0 for value in (snowball_ns, snowball_error, radixor_ns, ratio)):
+            raise ValueError(f"Snowball homepage comparison contains a non-positive or non-finite value for {case.case_name}")
+        rows.append((case, radixor_ns, snowball_ns, snowball_error, ratio, timing_basis))
+    aggregate_ratios = [ratio for case, _, _, _, ratio, _ in rows if case.homepage_aggregate]
+    chart_count = sum(case.homepage_chart for case in cases)
+    aggregate_count = sum(case.homepage_aggregate for case in cases)
+    if len(rows) != chart_count or len(aggregate_ratios) != aggregate_count:
+        raise ValueError("Snowball homepage speed coverage differs from the authoritative catalog policy")
+    if any(timing_basis != "changed tokens" for case, _, _, _, _, timing_basis in rows
+           if case.homepage_aggregate):
+        raise ValueError("Every aggregated Snowball comparison must use a changed-token corpus")
+    geomean = math.exp(math.fsum(math.log(value) for value in aggregate_ratios)
+                       / len(aggregate_ratios))
+    table = [
+        "<table><thead><tr><th>Language</th><th>Model</th><th>Timing corpus</th><th>Radixor ns/token</th><th>Snowball ns/token</th><th>Snowball JMH error ns/token</th><th>Snowball / Radixor</th><th>Aggregate</th></tr></thead><tbody>"
+    ]
+    bars: list[str] = []
+    logarithmic_extent = max(
+        math.log(1.1),
+        max(abs(math.log(ratio)) for _, _, _, _, ratio, _ in rows),
+    )
+    for case, radixor_ns, snowball_ns, snowball_error, ratio, timing_basis in rows:
+        escaped = html.escape(case.display_language)
+        aggregate = "yes" if case.homepage_aggregate else "no — root-only corpus"
+        table.append(
+            f"<tr><td>{escaped}</td><td><code>{case.model_id}</code></td><td>{timing_basis}</td>"
+            f"<td>{radixor_ns:.1f}</td><td>{snowball_ns:.1f}</td><td>±{snowball_error:.1f}</td>"
+            f"<td>{ratio:.3f}×</td><td>{aggregate}</td></tr>"
+        )
+        displacement = 50.0 * math.log(ratio) / logarithmic_extent
+        left = min(50.0, 50.0 + displacement)
+        width = abs(displacement)
+        bars.append(
+            f'<div class="rx2-ratio-row"><span>{escaped}</span>'
+            f'<span class="rx2-ratio-track" role="img" '
+            f'aria-label="{escaped}: Snowball to Radixor runtime ratio {ratio:.3f}" '
+            f'title="{escaped}: {ratio:.3f}×"><i style="--left:{left:.2f}%;--width:{width:.2f}%"></i></span>'
+            f'<b>{ratio:.3f}×</b></div>'
+        )
+    table.append("</tbody></table>")
+    return (
+        "<!-- JAVA-SNOWBALL-COMPARISON:START -->\n"
+        '<div class="rx2-chart-title"><strong>Java Radixor and official direct Snowball</strong> '
+        "<span>(exact-language cases)</span></div>\n"
+        f'<div class="rx2-throughput"><b>{geomean:.3f}×</b> <span>Snowball / Radixor geometric-mean runtime ratio</span></div>\n'
+        f'<div class="rx2-ratio-chart" aria-label="Log-scaled Snowball divided by Radixor nanoseconds per token for {chart_count} exact-language cases">\n'
+        '<div class="rx2-ratio-axis"><span>Radixor slower</span><b>1×</b><span>Snowball slower</span></div>\n'
+        + "\n".join(bars) + "\n</div>\n"
+        f'<p class="rx2-note">The {snapshot_date} Java JMH snapshot charts all <b>{chart_count}</b> supported official direct Snowball exact-language cases against the exact generic Radixor model row. The geometric mean is <b>{geomean:.3f}× over {aggregate_count} comparable changed-token cases</b>. Arabic remains charted but is excluded from the aggregate policy. A ratio above 1 means Snowball used more time per token. All rows use 3 warmup iterations, 5 measurement iterations, 3 forks, one thread, and the same per-language timing population.</p>\n'
+        '<details class="quality-details"><summary>Accessible raw Java comparison table</summary>\n'
+        + "".join(table) + "\n</details>\n"
+        "<!-- JAVA-SNOWBALL-COMPARISON:END -->"
+    )
+
+
+def replace_homepage_language_section(text: str, language_section: str) -> str:
+    """Replace the generated homepage language block without whitespace drift."""
+
+    updated, replacements = re.subn(
+        r'(?:[ \t]*\n)*[ \t]*(?:<!-- JAVA-LANGUAGE-PAGES:START -->\n)?'
+        r'<section class="rx2-languages-strip">.*?</section>\n'
+        r'(?:<!-- JAVA-LANGUAGE-PAGES:END -->)?(?:[ \t]*\n)*',
+        "\n\n    " + language_section + "\n\n",
+        text,
+        count=1,
+        flags=re.DOTALL,
+    )
+    if replacements != 1:
+        raise ValueError("Homepage language-page section is missing or ambiguous")
+    return updated
+
+
+def update_homepage(
+    docs_root: Path,
+    model_infos: list[ModelInfo],
+    defaults: dict[str, ModelInfo],
+    snowball_cases: list[SnowballCase],
+    corpora_by_model: dict[str, dict[str, object]],
+    speed_data: JmhData,
+    quality_path: Path,
+    publication: Publication,
+    geography: dict[str, HomepageGeography],
+    snapshot_date: str,
+) -> None:
+    """Publish lifecycle-safe Java evidence and every language-page link."""
+
+    if not model_infos or not defaults:
+        raise ValueError("Homepage publication requires active models and language defaults")
+    page_names = {info.page_name for info in defaults.values()}
+    if len(page_names) != len(defaults):
+        raise ValueError("Homepage language pages are not unique")
+    for page_name in page_names:
+        if not (docs_root / "benchmarks" / "languages" / page_name).is_file():
+            raise ValueError(f"Homepage language link target does not exist: {page_name}")
+    landing = docs_root / "overrides" / "landing.html"
+    text = landing.read_text(encoding="utf-8")
+    snowball = render_snowball_homepage(
+        snowball_cases, corpora_by_model, speed_data, snapshot_date
+    )
+    if "<!-- JAVA-SNOWBALL-COMPARISON:START -->" in text:
+        text = re.sub(
+            r"<!-- JAVA-SNOWBALL-COMPARISON:START -->.*?<!-- JAVA-SNOWBALL-COMPARISON:END -->",
+            snowball, text, count=1, flags=re.DOTALL,
+        )
+    else:
+        insertion = '<div class="rx2-chart-title"><strong>Python (PyO3) performance</strong>'
+        text = text.replace(insertion, snowball + "\n" + insertion, 1)
+    links: list[str] = []
+    for info in sorted(defaults.values(), key=lambda item: (item.language_name.casefold(), item.model_id)):
+        place = geography[info.language]
+        glyph = geography_glyph(place.icon)
+        model_label = info.model_id
+        if info.language == "PL_PL":
+            model_label += " + pl-pl-polimorf"
+        links.append(
+            f'<a class="rx2-language" href="{{{{ base_url }}}}/benchmarks/languages/{info.page_name.removesuffix(".md")}/" '
+            f'aria-label="Open {html.escape(info.language_name)} benchmark; {html.escape(place.label)}">'
+            f'<i class="rx2-language-icon" aria-hidden="true" title="{html.escape(place.label)}">{glyph}</i>'
+            f'<span>{html.escape(info.language_name)}</span>'
+            f'<small>{model_label}</small></a>'
+        )
+    language_section = (
+        '<!-- JAVA-LANGUAGE-PAGES:START -->\n<section class="rx2-languages-strip">\n'
+        f'<div class="rx2-language-label"><div><strong>{len(model_infos)} models · {len(defaults)} language pages</strong>'
+        '<span>Complete Java quality evidence; Polish documents both models</span></div></div>\n'
+        '<div class="rx2-language-list">' + "".join(links) + '</div>\n</section>\n'
+        '<!-- JAVA-LANGUAGE-PAGES:END -->'
+    )
+    text = replace_homepage_language_section(text, language_section)
+    matrices, wins, compared, undefined = quality_homepage_counts(
+        quality_path, set(defaults)
+    )
+    text = re.sub(r"across \d+ language defaults", f"across {len(defaults)} language defaults", text)
+    text = re.sub(r"<b>\d+ model artifacts</b>", f"<b>{len(model_infos)} model artifacts</b>", text)
+    text = re.sub(
+        r"(<strong>)\d+ / \d+(</strong><span>First-ranked balanced)",
+        rf"\g<1>{wins} / {matrices}\g<2>", text,
+    )
+    text = re.sub(
+        r"(<strong>)\d+ / \d+(</strong><span>Quality first)",
+        rf"\g<1>{wins} / {matrices}\g<2>", text,
+    )
+    text = re.sub(
+        r"Radixor (?:ranks first in all 40 evaluated balanced-accuracy matrices; 38 include direct comparators and the two Hebrew modes report Radixor independently|is the sole balanced-accuracy leader in \d+ of \d+ mathematically defined language-mode matrices; \d+ include a direct comparator\. [^.]+ publish OI/UI and raw counts but have no defined balanced accuracy)\.",
+        f"Radixor is the sole balanced-accuracy leader in {wins} of {matrices} mathematically defined language-mode matrices; {compared} include a direct comparator. {undefined} matrices with an absent pair class publish OI/UI and raw counts but have no defined balanced accuracy.",
+        text,
+    )
+    text = re.sub(
+        r"On this pairwise measure, Radixor (?:ranks first in all 40 evaluated deterministic language-mode matrices, with no exact first-place ties; 38 include direct comparators and the two Hebrew modes report Radixor independently|is the sole leader in \d+ of \d+ mathematically defined deterministic language-mode matrices; \d+ include direct comparators\. [^.]+ publish OI/UI and raw counts but no balanced-accuracy rank)\.",
+        f"On this pairwise measure, Radixor is the sole leader in {wins} of {matrices} mathematically defined deterministic language-mode matrices; {compared} include direct comparators. {undefined} matrices with an absent pair class publish OI/UI and raw counts but no balanced-accuracy rank.",
+        text,
+    )
+    publication.write(landing, text)
+
+
 def main() -> None:
+    global LANGUAGES
+    global LANGUAGE_IDENTITY_WORDS
+    global SNOWBALL_CASES
+
     arguments = parse_arguments()
     publication = Publication(arguments.mode)
-    corpora = read_corpora(arguments.corpus)
+    verify_active_inputs(
+        arguments.docs_root / "benchmarks" / "data" / "active-snapshots.properties",
+        {
+            "corpus": arguments.corpus,
+            "accuracy": arguments.accuracy,
+            "speed": arguments.speed,
+            "coverage_accuracy": arguments.coverage_accuracy,
+            "coverage_speed": arguments.coverage_speed,
+            "quality": arguments.quality,
+            "comparator": arguments.snowball_catalog,
+        },
+    )
+    snowball_cases = read_snowball_catalog(arguments.snowball_catalog)
+    SNOWBALL_CASES = {
+        case.language: (case.case_name, case.lucene_available)
+        for case in snowball_cases
+    }
+    model_infos = load_model_infos()
+    defaults = prepare_language_pages(arguments.docs_root, model_infos, publication)
+    geography = read_homepage_geography(
+        arguments.docs_root / "benchmarks" / "data" / "homepage-language-geography.csv",
+        set(defaults),
+    )
+    polimorf = next(info for info in model_infos if info.model_id == "pl-pl-polimorf")
+    prepare_polimorf_section(arguments.docs_root, polimorf, publication)
+    LANGUAGES = {
+        info.page_name: info.language
+        for info in sorted(defaults.values(), key=lambda item: item.page_name)
+    }
+    legacy_identity_words = LANGUAGE_IDENTITY_WORDS
+    LANGUAGE_IDENTITY_WORDS = {
+        info.language: legacy_identity_words.get(
+            info.language, words(info.language_name)
+        )
+        for info in defaults.values()
+    }
+    corpora_by_model = read_corpora(arguments.corpus)
+    infos_by_id = {info.model_id: info for info in model_infos}
+    for model_id, entry in corpora_by_model.items():
+        if model_id not in infos_by_id:
+            raise ValueError(f"Corpus report contains unknown model {model_id}")
+        expected_distinct = infos_by_id[model_id].distinct_forms
+        reported_distinct = int(entry["distinct"])
+        if reported_distinct not in {0, expected_distinct}:
+            raise ValueError(
+                f"Corpus distinct-form count for {model_id} is {reported_distinct}, "
+                f"expected {expected_distinct}"
+            )
+        entry["distinct"] = expected_distinct
+    corpora = {
+        info.language: corpora_by_model[info.model_id]
+        for info in defaults.values()
+        if info.model_id in corpora_by_model
+    }
     accuracy_data = read_jmh(arguments.accuracy)
     speed_data = read_jmh(arguments.speed)
     coverage_accuracy_data = read_jmh(arguments.coverage_accuracy)
     coverage_speed_data = read_jmh(arguments.coverage_speed)
-    measured_keys = set(accuracy_data.primary) | set(speed_data.primary)
-    if any("PolishPolimorf" in key.benchmark for key in measured_keys):
+    speed_protocol = infer_speed_protocol(speed_data, "stemmer speed report")
+    coverage_speed_protocol = infer_speed_protocol(
+        coverage_speed_data, "coverage speed report"
+    )
+    if speed_protocol != coverage_speed_protocol:
         raise ValueError(
-            "A published report contains the excluded PolishPolimorf benchmark."
+            "Stemmer and coverage speed reports use inconsistent supported protocols: "
+            f"{speed_protocol.samples} and {coverage_speed_protocol.samples} samples"
         )
     english_speed_keys = [
         key
@@ -809,11 +1963,27 @@ def main() -> None:
         corpora,
         accuracy_data,
         speed_data,
+        speed_protocol,
         publication,
         arguments.date,
         arguments.release_version,
     )
-    update_corpora_reference(arguments.docs_root, corpora, publication)
+    update_polimorf_section(
+        arguments.docs_root,
+        polimorf,
+        corpora_by_model.get(polimorf.model_id),
+        accuracy_data,
+        speed_data,
+        publication,
+    )
+    update_corpora_reference(
+        arguments.docs_root, model_infos, corpora_by_model, publication
+    )
+    update_model_indexes(arguments.docs_root, model_infos, defaults, publication)
+    update_mkdocs_language_nav(arguments.docs_root, defaults, publication)
+    update_speed_protocol_documentation(
+        arguments.docs_root, speed_protocol, publication
+    )
     update_coverage(
         arguments.docs_root,
         arguments.readme,
@@ -824,6 +1994,18 @@ def main() -> None:
         publication,
         main_english_ns,
     )
+    update_homepage(
+        arguments.docs_root,
+        model_infos,
+        defaults,
+        snowball_cases,
+        corpora_by_model,
+        speed_data,
+        arguments.quality,
+        publication,
+        geography,
+        arguments.date,
+    )
     verb = "Verified" if arguments.mode == "verify" else "Updated"
     print(f"{verb} Java benchmark documentation for {arguments.release_version} ({arguments.date}).")
 
diff --git a/tools/update-generalization-documentation.py b/tools/update-generalization-documentation.py
index e64c8fb..573e311 100644
--- a/tools/update-generalization-documentation.py
+++ b/tools/update-generalization-documentation.py
@@ -91,40 +91,48 @@ def parse_arguments() -> argparse.Namespace:
     parser.add_argument("source", type=Path)
     parser.add_argument("documentation_root", type=Path)
     parser.add_argument("mode", choices=("update", "verify"))
+    parser.add_argument("--corpus", type=Path, required=True)
+    parser.add_argument("--snapshot-name", required=True)
     return parser.parse_args()
 
 
-def read_model_catalog(path: Path) -> dict[str, tuple[str, str, str]]:
-    models: dict[str, tuple[str, str, str]] = {}
+def read_language_pages(path: Path) -> tuple[dict[str, str], dict[str, str]]:
+    """Read all authoritative language display names and page files."""
+
+    languages: dict[str, str] = {}
+    pages: dict[str, str] = {}
+    pattern = re.compile(r"^\| ([^|]+) \| `([^`]+)` \|.*\| \[[^]]+\]\(([^)]+\.md)\) \|$")
     for line in path.read_text(encoding="utf-8").splitlines():
-        if not line.startswith("|"):
-            continue
-        cells = [cell.strip().strip("`") for cell in line.split("|")[1:-1]]
-        if len(cells) == 14 and cells[2] == "true":
-            models[cells[1]] = (cells[0], cells[4], cells[12])
-    if set(models) != set(LANGUAGES):
-        raise ValueError("The checked-in model catalog does not define the expected 20 defaults.")
-    return models
+        match = pattern.match(line)
+        if match:
+            languages[match.group(2)] = match.group(1)
+            pages[match.group(2)] = match.group(3)
+    if not languages or len(languages) != len(pages):
+        raise ValueError("Language index must define unique default-language pages.")
+    return languages, pages
 
 
 def read_corpus_catalog(path: Path) -> dict[str, tuple[str, str, str, int, int, int, int]]:
     corpora: dict[str, tuple[str, str, str, int, int, int, int]] = {}
-    for line in path.read_text(encoding="utf-8").splitlines():
-        if not line.startswith("|"):
-            continue
-        cells = [cell.strip().strip("`") for cell in line.split("|")[1:-1]]
-        if len(cells) == 9 and cells[3] in LANGUAGES:
-            corpora[cells[3]] = (
-                cells[0], cells[1], cells[2], int(cells[4].replace(",", "")),
-                int(cells[5].replace(",", "")), int(cells[6].replace(",", "")),
-                int(cells[7].replace(",", "")),
+    with path.open(newline="", encoding="utf-8") as source:
+        for row in csv.DictReader(source):
+            if row["Model ID"] == "pl-pl-polimorf":
+                continue
+            language = row["Language"]
+            value = (
+                row["Model ID"], row["Model version"], row["Model SHA-256"],
+                int(row["Dictionary rows"]), int(row["Total tokens"]),
+                int(row["Already-root tokens"]), int(row["Changed tokens"]),
             )
+            previous = corpora.setdefault(language, value)
+            if previous != value:
+                raise ValueError(f"Corpus language has conflicting model counters: {language}")
     if set(corpora) != set(LANGUAGES):
-        raise ValueError("The checked-in corpus table does not define the expected 20 defaults.")
+        raise ValueError("The current corpus catalog does not define every active default.")
     return corpora
 
 
-def read_and_validate(source: Path, documentation_root: Path) -> list[dict[str, str]]:
+def read_and_validate(source: Path, corpus_path: Path) -> list[dict[str, str]]:
     with source.open(encoding="utf-8", newline="") as handle:
         rows = list(csv.DictReader(handle))
     if not rows:
@@ -145,22 +153,15 @@ def read_and_validate(source: Path, documentation_root: Path) -> list[dict[str,
         raise ValueError(f"Missing required columns: {sorted(missing)}")
 
     protocols = {row["protocol_version"] for row in rows}
-    versions = {row["radixor_java_version"] for row in rows}
     seeds = {row["seed"] for row in rows}
-    if protocols != {EXPECTED_PROTOCOL} or len(versions) != 1 or not next(iter(versions)).strip():
-        raise ValueError("Unexpected split protocol or inconsistent Radixor/Java identity.")
+    if protocols != {EXPECTED_PROTOCOL} or any(not row["radixor_java_version"].strip() for row in rows):
+        raise ValueError("Unexpected split protocol or missing Radixor/Java identity.")
     if seeds != EXPECTED_SEEDS:
         raise ValueError("The report does not use the five predeclared split seeds.")
-    if len({row["source_revision"] for row in rows}) != 1 \
-            or len({row["source_state"] for row in rows}) != 1 \
-            or len({row["generator_sha256"] for row in rows}) != 1:
-        raise ValueError("Source revision, state, and generator digest must be invariant.")
-    first = rows[0]
-    if len(first["source_revision"]) != 40 or len(first["generator_sha256"]) != 64 \
-            or not first["source_state"].strip():
+    if any(len(row["source_revision"]) != 40 or len(row["generator_sha256"]) != 64
+           or not row["source_state"].strip() for row in rows):
         raise ValueError("Source provenance is incomplete.")
-    models = read_model_catalog(documentation_root / "stemmer-model-catalog.md")
-    corpora = read_corpus_catalog(documentation_root / "benchmarks/reference/corpora.md")
+    corpora = read_corpus_catalog(corpus_path)
     keys: set[tuple[str, str, int]] = set()
     by_language: dict[str, list[dict[str, str]]] = defaultdict(list)
     for row in rows:
@@ -174,7 +175,7 @@ def read_and_validate(source: Path, documentation_root: Path) -> list[dict[str,
         if language not in LANGUAGES or percent not in PERCENTS:
             raise ValueError(f"Unexpected language or percentage: {key}")
         model = (row["model_id"], row["model_version"], row["model_sha256"])
-        if model != models[language] or model != corpora[language][:3]:
+        if model != corpora[language][:3]:
             raise ValueError(f"Scenario does not use the published default model: {key}")
         total_rows = int(row["total_rows"])
         selected_rows = int(row["selected_rows"])
@@ -214,7 +215,7 @@ def read_and_validate(source: Path, documentation_root: Path) -> list[dict[str,
             raise ValueError(f"The 100% scenario must have no held-out observations: {key}")
 
     if set(by_language) != set(LANGUAGES):
-        raise ValueError("The report does not contain exactly the 20 default languages.")
+        raise ValueError("The report does not contain exactly the active default languages.")
     expected_per_language = len(seeds) * len(PERCENTS)
     for language, language_rows in by_language.items():
         if len(language_rows) != expected_per_language:
@@ -222,6 +223,10 @@ def read_and_validate(source: Path, documentation_root: Path) -> list[dict[str,
         if len({(row["model_id"], row["model_version"], row["model_sha256"])
                 for row in language_rows}) != 1:
             raise ValueError(f"Model provenance changes within {language}.")
+        if len({tuple(row[name] for name in (
+                "radixor_java_version", "source_revision", "source_state", "generator_sha256"))
+                for row in language_rows}) != 1:
+            raise ValueError(f"Measurement provenance changes within {language}.")
         corpus = corpora[language]
         denominators = {(int(row["total_rows"]), int(row["whole_total"]),
                          int(row["whole_root_total"]), int(row["whole_changed_total"]))
@@ -234,6 +239,17 @@ def read_and_validate(source: Path, documentation_root: Path) -> list[dict[str,
                          for row in full_rows}
         if len(full_counters) != 1:
             raise ValueError(f"Full-coverage counters differ across splits: {language}")
+    provenance_languages: dict[tuple[str, str, str, str], set[str]] = defaultdict(set)
+    for row in rows:
+        provenance_languages[tuple(row[name] for name in (
+            "radixor_java_version", "source_revision", "source_state", "generator_sha256"
+        ))].add(row["language"])
+    continuation_count = len(LANGUAGES) - 20
+    provenance_sizes = sorted(len(languages) for languages in provenance_languages.values())
+    if provenance_sizes not in ([20, continuation_count], [1, 20, continuation_count - 1]):
+        raise ValueError("Active report must preserve the 20-language archive and the active standalone continuation, allowing a separately measured Arabic repair.")
+    if provenance_sizes == [1, 20, continuation_count - 1] and {"AR"} not in provenance_languages.values():
+        raise ValueError("Only Arabic may have separate repair provenance in the active report.")
     return rows
 
 
@@ -314,8 +330,6 @@ def render_language_conclusion(language: str, rows: list[dict[str, str]]) -> str
     last_all = median_ratio(last, "unseen")
     first_root = median_ratio(first, "unseen_root")
     last_root = median_ratio(last, "unseen_root")
-    if None in (first_changed, last_changed, first_all, last_all, first_root, last_root):
-        raise ValueError(f"Undefined endpoint generalization metric for {language}.")
     provenance = rows[0]
     lines = [
         "## Dictionary-Family Generalization Conclusion", "",
@@ -336,30 +350,30 @@ def render_language_conclusion(language: str, rows: list[dict[str, str]]) -> str
             f"{median_range([ratio(row, 'unseen_changed') for row in scenarios])} | "
             f"{median_range([ratio(row, 'unseen_root') for row in scenarios])} |"
         )
-    changed_gain = float(last_changed) - float(first_changed)
-    all_gain = float(last_all) - float(first_all)
-    root_gain = float(last_root) - float(first_root)
+    def endpoint_statement(label: str, low: float | None, high: float | None) -> str:
+        if low is None or high is None:
+            return f"- {label} is **n/a** at one or both endpoints because the corresponding evaluated population is empty."
+        return (f"- {label} moves from **{low:.3f}%** at 10% training knowledge to "
+                f"**{high:.3f}%** at 90%, a measured **{high - low:+.3f} percentage-point** change.")
     lines.extend([
         "", "### Generalization conclusion", "",
-        f"- Median exactness on genuinely unseen changed forms moves from **{float(first_changed):.3f}%**",
-        f"  at 10% training knowledge to **{float(last_changed):.3f}%** at 90%, a measured",
-        f"  **{changed_gain:+.3f} percentage-point** change for this dictionary.",
-        f"- Over the same endpoints, unseen all-form exactness changes by **{all_gain:+.3f} pp** and",
-        f"  preservation of unseen already-root forms changes by **{root_gain:+.3f} pp**. These separate",
-        "  outcomes show whether the changed-form result coexists with preservation behavior.",
+        endpoint_statement("Median exactness on genuinely unseen changed forms", first_changed, last_changed),
+        endpoint_statement("Unseen all-form exactness", first_all, last_all),
+        endpoint_statement("Preservation of unseen already-root forms", first_root, last_root),
         "- The evidence establishes within-resource transfer across withheld dictionary families. It",
         "  does not estimate unrelated domains, misspellings, arbitrary compounds, or external corpora.",
         "", "The complete ten-level table and split ranges remain in the",
         "[independent generalization report](../generalization.md); raw counters and provenance are in",
-        "[`dictionary-generalization.csv`](../data/dictionary-generalization.csv). The",
+        "[active machine-readable snapshot](../data/dictionary-generalization-2026-09-11.csv). The",
         "[frozen methodology](../reference/generalization-methodology.md) defines family-level",
         "splitting, unseen-surface leakage control, aggregation, and the limits of the claim.",
     ])
     return "\n".join(lines)
 
 
-def render(rows: list[dict[str, str]], checksum: str) -> str:
+def render(rows: list[dict[str, str]], checksum: str, snapshot_name: str) -> str:
     first = rows[0]
+    source_manifest_name = snapshot_name.removesuffix(".csv") + "-sources.txt"
     endpoints: dict[str, tuple[float, float]] = {}
     for language in LANGUAGES:
         low_rows = [row for row in rows
@@ -368,9 +382,10 @@ def render(rows: list[dict[str, str]], checksum: str) -> str:
                      if row["language"] == language and int(row["requested_percent"]) == 90]
         low = median_ratio(low_rows, "unseen_changed")
         high = median_ratio(high_rows, "unseen_changed")
-        if low is None or high is None:
-            raise ValueError(f"Undefined generalization endpoint for {language}.")
-        endpoints[language] = (low, high)
+        if low is not None and high is not None:
+            endpoints[language] = (low, high)
+    if not endpoints:
+        raise ValueError("No language has two defined generalization endpoints.")
     lowest_at_ten = min(endpoints, key=lambda language: endpoints[language][0])
     highest_at_ten = max(endpoints, key=lambda language: endpoints[language][0])
     lowest_at_ninety = min(endpoints, key=lambda language: endpoints[language][1])
@@ -386,7 +401,7 @@ def render(rows: list[dict[str, str]], checksum: str) -> str:
         "word-to-root answer list. This experiment measures how those transformations transfer",
         "to dictionary families that were not used to build the Java trie.",
         "",
-        "For every one of the 20 default models, complete dictionary rows are placed in a",
+        f"For every one of the {len(LANGUAGES)} default models, complete dictionary rows are placed in a",
         "frozen pseudorandom order. Exact-size, nested prefixes retain 10% through 100% of",
         "the rows for training. Five predeclared splits are evaluated against the complete",
         "dictionary; the primary `Unseen` columns exclude a held-out occurrence whenever its",
@@ -399,7 +414,7 @@ def render(rows: list[dict[str, str]], checksum: str) -> str:
         "",
         "## All-Language Summary",
         "",
-        "Each cell is the language-macro mean of 20 per-language split medians, so large",
+        "Each cell is the language-macro mean across languages with a defined denominator, so large",
         "dictionaries do not dominate small ones. Changed-form exactness is the most demanding",
         "measure because it excludes words whose expected root is already the input token.",
         "",
@@ -467,14 +482,12 @@ def render(rows: list[dict[str, str]], checksum: str) -> str:
     lines += [
         "## Provenance",
         "",
-        f"- Radixor/Java: `{first['radixor_java_version']}`",
-        f"- Core source revision: `{first['source_revision']}`",
-        f"- Source state: `{first['source_state']}`",
-        f"- Generalization generator SHA-256: `{first['generator_sha256']}`",
-        "- Measured-source manifest: [`dictionary-generalization-sources.sha256`](data/dictionary-generalization-sources.sha256)",
+        f"- Measurement provenance is recorded per row. The active snapshot deliberately combines the frozen 20-language archive with the separately measured {len(LANGUAGES) - 20}-language continuation.",
+        "- Historical 20-language source manifest: [`dictionary-generalization-sources.sha256`](data/dictionary-generalization-sources.sha256)",
+        f"- Continuation source manifest: [`{source_manifest_name}`](data/{source_manifest_name})",
         f"- Split protocol: `{first['protocol_version']}`",
         "- Splits per coverage level: 5",
-        "- Authoritative raw counters: [`dictionary-generalization.csv`](data/dictionary-generalization.csv)",
+        f"- Authoritative raw counters: [`{snapshot_name}`](data/{snapshot_name})",
         f"- CSV SHA-256: `{checksum}`",
         "- Model artifact IDs, independent versions, and SHA-256 values are recorded on every raw row.",
         "- Runtime speed is intentionally excluded: speed does not establish generalization. The",
@@ -486,15 +499,35 @@ def render(rows: list[dict[str, str]], checksum: str) -> str:
 
 
 def main() -> None:
+    global LANGUAGES
+    global LANGUAGE_PAGES
+
     arguments = parse_arguments()
-    rows = read_and_validate(arguments.source, arguments.documentation_root)
+    if not re.fullmatch(r"dictionary-generalization-[0-9]{4}-[0-9]{2}-[0-9]{2}\.csv", arguments.snapshot_name):
+        raise ValueError("Active generalization snapshot must use a dated file name.")
+    manifest = arguments.documentation_root / "benchmarks/data/active-snapshots.properties"
+    active_entries = manifest.read_text(encoding="utf-8").splitlines()
+    if f"generalization={arguments.snapshot_name}" not in active_entries:
+        raise ValueError("Active snapshot manifest does not select the supplied generalization snapshot.")
+    if arguments.source.resolve() != (manifest.parent / arguments.snapshot_name).resolve():
+        raise ValueError("Supplied generalization CSV is not the active snapshot file.")
+    if f"corpus={arguments.corpus.name}" not in active_entries \
+            or arguments.corpus.resolve() != (manifest.parent / arguments.corpus.name).resolve():
+        raise ValueError("Supplied corpus CSV is not the active performance corpus.")
+    expected_sources = arguments.snapshot_name.removesuffix(".csv") + "-sources.txt"
+    if f"generalization_sources={expected_sources}" not in active_entries:
+        raise ValueError("Active snapshot manifest does not select the supplied generalization source manifest.")
+    LANGUAGES, LANGUAGE_PAGES = read_language_pages(
+        arguments.documentation_root / "benchmarks/languages/index.md"
+    )
+    rows = read_and_validate(arguments.source, arguments.corpus)
     checksum = hashlib.sha256(arguments.source.read_bytes()).hexdigest()
-    expected = render(rows, checksum)
+    expected = render(rows, checksum, arguments.snapshot_name)
     target = arguments.documentation_root / "benchmarks/generalization.md"
     data_directory = arguments.documentation_root / "benchmarks/data"
-    published_csv = data_directory / "dictionary-generalization.csv"
-    published_checksum = data_directory / "dictionary-generalization.sha256"
-    checksum_text = f"{checksum}  dictionary-generalization.csv\n"
+    published_csv = data_directory / arguments.snapshot_name
+    published_checksum = data_directory / arguments.snapshot_name.replace(".csv", ".sha256")
+    checksum_text = f"{checksum}  {arguments.snapshot_name}\n"
     language_documents: dict[Path, str] = {}
     for language, page_name in LANGUAGE_PAGES.items():
         page = arguments.documentation_root / "benchmarks/languages" / page_name