merge: synchronize benchmark dashboard with statgpu 0.2.4

[TheHiddenObserver]

Summary

Synchronize the benchmark dashboard staging branch (PR #76) with current master without adding a new benchmark source, parser family, schema field, model family, CV source, or frontend feature.

This is a stacked synchronization PR:

• base: feature/benchmark-frontend-dashboard;
• final head: c394accf713d798ef8adf6474a75ebeb987e8ce1;
• pre-sync dashboard head: 10874915f05dcae8f0124870e7d10de31848d42b;
• synchronized master commit: f95447e8e1bf189f16ccc2d113b3acab2ae53dfe;
• two-parent synchronization merge: 8c72e94e3c8b1a1527574ce1fe3b220acb09c562;
• deterministic asset refresh: 5f50754026448a0723fbef7d1ed6515465ec9a7a;
• bilingual documentation navigation fix: c394accf713d798ef8adf6474a75ebeb987e8ce1.

Reviewed merge resolution

Five paths changed on both sides:

• .gitignore;
• docs/en/README.md;
• docs/en/guides/benchmarks.md;
• docs/cn/guides/benchmarks.md;
• pyproject.toml.

Git auto-merged .gitignore and the English documentation index; the two benchmark guides and pyproject.toml produced textual conflicts. The final resolution was reconstructed explicitly from the source commits:

• current master packaging, release metadata, navigation, and ignore policy are retained;
• PR feat: benchmark dashboard frontend and canonical data pipeline #76's reviewed EN/CN benchmark guides and dashboard contracts are retained;
• dashboard-specific jsonschema[format]>=4.0, navigation, and tracked-source exceptions are reapplied;
• English and Chinese documentation indexes expose consistent dashboard navigation and use the 2026-08-06 update date.

Registered-bundle integrity

Canonical registered bundle after deterministic regeneration:

• schema: 1.1.0;
• minimum source date: 2026-06-01;
• registered/available/parsed sources: 8/8/8;
• skipped registered sources: 0;
• normalized runs: 1,774;
• represented models: 36;
• generation ID: 7437ea045e8161b4090598bdfb8062068514a5914f5d7e10a0bb54fc8dbbeff1.

These figures establish integrity of the manifest-registered source set. They do not establish complete benchmark coverage across maintained statgpu capabilities.

Inventory semantics and data-completeness limits

Current master adds Cox/PR79/PR80 benchmark evidence under results/benchmark_frontend_sources/. Those files are preserved as inherited master history but are intentionally not registered in frontend_sources.json and do not create dashboard rows in this synchronization PR.

The generated inventory currently reports catalog_total=40, but this value is a raw JSON-artifact count under the current scan rule, not an audited classification of 40 benchmark candidates. Likewise, eligible_total=8 currently aliases the manifest registration count rather than an independent eligibility audit.

Accordingly:

• 40 must not be described as complete catalog coverage;
• 8/8/8 means every registered source exists and parses;
• it does not mean every implemented method has a current canonical benchmark source.

The audited source catalog, corrected inventory semantics, and maintained method-coverage matrix are tracked in #100. The first dedicated missing-source package, current cross-validation data, is tracked in #91.

No measured row is fabricated or inferred.

Known benchmark-data gaps

The synchronized bundle still lacks or only partially covers substantial maintained capability areas, including:

• current CV data for RidgeCV, LassoCV, ElasticNetCV, LogisticRegressionCV, PenalizedGLM_CV, and CoxPHCV — benchmark: add canonical cross-validation source and dashboard coverage #91;
• structured Distribution data;
• Bisquare/Fair and full robust GPU comparisons;
• Feature Selection/Knockoff;
• systematic linear/GLM inference;
• full Covariance and Nonparametric matrices;
• Penalized CoxPH/CoxPHCV paths;
• extended Panel estimators;
• Multiple Testing/resampling utilities;
• broader Ordered and synchronization-correct ANOVA crossover evidence.

Some existing evidence is not canonical-ready because it is rounded Markdown, lacks raw provenance/timing dimensions, has incomplete scale identity, or has known alignment limitations. #100 owns classification and issue assignment for these cases.

Merge topology

The synchronization was produced by a real two-parent merge. Temporary one-shot synchronization and validation workflows were removed from the final history/tree. The merge parents are exactly:

1. 10874915f05dcae8f0124870e7d10de31848d42b — PR feat: benchmark dashboard frontend and canonical data pipeline #76 head;
2. f95447e8e1bf189f16ccc2d113b3acab2ae53dfe — synchronized master.

Validation

Dashboard-specific final head

Benchmark Frontend CI #355 / run 31102163926: success

• Python data/parser/generator tests: Python 3.9 and 3.11;
• strict deterministic generation and schema/semantic validation;
• TypeScript typecheck;
• Vite production build;
• generated-data and deployment-asset staleness;
• Playwright Chromium E2E;
• committed frontend-artifact rejection.

Project-level exact-head validation

Run 31101612155: success against exact content head 5f50754026448a0723fbef7d1ed6515465ec9a7a

• documentation contracts;
• Python 3.9, 3.10, 3.11, and 3.12 regression matrix;
• complete CPU test tree;
• static/compile contracts and PR79 canonical smoke;
• sklearn 1.2.2, 1.3.2, and latest compatibility.

The only later changes are the two documentation-index edits in c394accf....

Run 31102102289: success against exact final head c394accf713d798ef8adf6474a75ebeb987e8ce1

• Python 3.9 documentation writer;
• deterministic bilingual link validation;
• maintained documentation contracts.

Review-fix completion

Review/fix cycles resolved:

1. incorrect assumption that all five overlapping files would be textual conflicts;
2. an over-broad whitespace guard that inspected inherited master history;
3. deterministic generated-asset staleness after the raw JSON count increased from 38 to 40;
4. GitHub suppression of workflows on a bot-authored head, replaced by an equivalent user-authored commit;
5. stale English documentation date and missing Chinese dashboard navigation;
6. over-strong interpretation of catalog_total, eligible_total, and registered-source success as evidence of complete data coverage.

The latest review finds no remaining synchronization, reproducibility, provenance, topology, documentation, or CI defect in #90 scope. Benchmark coverage completeness remains intentionally unresolved and is now explicitly tracked by #100 and its child work packages.

Merge target

This PR is ready for review/merge into the PR #76 staging branch. It must not be merged directly into master. Issue #90 remains open until this stacked PR is merged into feature/benchmark-frontend-dashboard and PR #76 is confirmed mergeable.

[TheHiddenObserver]

Final fresh review — no new in-scope findings

Reviewed the final head c394accf713d798ef8adf6474a75ebeb987e8ce1 after the complete review-fix cycle.

Merge topology and scope

• The synchronization merge commit 8c72e94e3c8b1a1527574ce1fe3b220acb09c562 has the required parents:
  1. PR #76 head 10874915f05dcae8f0124870e7d10de31848d42b;
  2. current synchronized master f95447e8e1bf189f16ccc2d113b3acab2ae53dfe.
• Subsequent commits only refresh deterministic generated assets and synchronize the bilingual documentation indexes.
• No new benchmark source registration, parser family, schema version, model family, CV source, or frontend feature was introduced.
• All temporary synchronization and validation workflows are absent from the final tree.

Conflict resolutions

Rechecked the five overlapping paths:

• .gitignore retains current master protections and the reviewed dashboard source/artifact exceptions;
• pyproject.toml remains at statgpu 0.2.4 and only adds jsonschema[format]>=4.0 to the relevant extras;
• English and Chinese benchmark guides retain the reviewed PR #76 dashboard contract without restoring stale universal claims;
• English and Chinese documentation indexes now expose consistent dashboard navigation and carry the current update date.

Canonical data and provenance

• manifest minimum source date remains 2026-06-01;
• 8 sources remain registered, available, and parsed;
• 0 sources are skipped;
• 1,774 normalized runs and 36 models remain unchanged;
• inherited unregistered master evidence increases catalog coverage to 40 but does not create dashboard rows;
• generation ID is consistent across benchmark data, parse report, and source inventory;
• frontend and deployed documentation data copies are deterministic and current;
• no measured row was fabricated or inferred.

Validation

• Benchmark Frontend CI #355: success, including Python 3.9/3.11 data tests, strict generation/schema validation, TypeScript typecheck, Vite build, staleness, and Playwright Chromium E2E.
• Exact-head project validation run 31101612155: success, covering documentation/static contracts, Python 3.9–3.12 regression matrix, full CPU suite, and sklearn 1.2.2/1.3.2/latest compatibility.
• Exact-head documentation validation run 31102102289: success after the bilingual navigation fix.

Review conclusion

No unresolved correctness, reproducibility, provenance, merge-topology, scope, documentation, or CI finding remains in this synchronization PR. The latest fresh review found no new in-scope issue. The PR is ready for review/merge into the PR #76 staging branch, but it should not be merged into master directly.

[cursor]

Bugbot is not enabled for your account, so this pull request was not reviewed.

Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, you can upgrade your account or add credits to your account and enable them for code reviews in your settings.

[Copilot]

Copilot wasn't able to review this pull request because it exceeds the maximum number of files (300). Try reducing the number of changed files and requesting a review from Copilot again.

[TheHiddenObserver]

Additional data-completeness review

The synchronization mechanics and deterministic bundle remain valid, but a fresh review focused on data completeness found an important limitation that should qualify the earlier “no new in-scope findings” conclusion.

1. The bundle is internally complete, not method-coverage complete

files_skipped=0 means that all eight registered sources were present and parsed. It does not mean that every implemented/runnable method has current benchmark data.

The repository’s own coverage audits explicitly classify the following as benchmark-data gaps or non-canonical-ready evidence:

• all current CV families (RidgeCV, LassoCV, ElasticNetCV, LogisticRegressionCV, PenalizedGLM_CV, CoxPHCV);
• Feature Selection;
• Bisquare/Fair and all-backend robust/quantile coverage;
• systematic linear/GLM inference;
• penalized CoxPH and CoxPHCV;
• most covariance estimators;
• substantial nonparametric breadth;
• additional panel estimators;
• multiple-testing/resampling utilities;
• distribution benchmarks, where a June Markdown result exists but no canonical structured source exists;
• broader Ordered and ANOVA crossover grids.

Issue #91 is the specific open issue tracking the missing canonical CV source. It correctly states that the dashboard has a CV presentation contract but no real current CV rows.

2. source_inventory.json does not prove source-catalog completeness

The current implementation sets:

eligible_total = registered_sources

rather than independently classifying all discovered artifacts against the minimum-date, provenance, structure, and alignment rules.

Likewise, when the manifest does not provide catalog_total, the generator computes it by recursively counting JSON files outside benchmark_frontend_sources; it is not a reviewed catalog of candidate sources. The repository has no committed benchmark_source_catalog.json, and frontend_data/catalog.py still describes itself as a placeholder.

Therefore the current values:

catalog_total = 40
eligible_total = 8
registered_sources = 8
available_sources = 8
parsed_sources = 8

must be interpreted as:

• 40 discovered non-canonical JSON artifacts under the counting rule;
• 8 manifest-registered sources;
• all 8 registered sources are available and parsed.

They do not establish that only eight repository artifacts are genuinely eligible or that no benchmark data is missing.

3. Historical missing-input finding

PR #76 previously had a separate concrete defect where the registry referenced benchmark inputs that were not committed, causing clean-checkout run-count drift and parser-test failure. The current strict manifest/SHA/staleness gates address that class of defect. That historical finding is different from the present method-coverage gaps.

Review conclusion

• This does not invalidate PR #99 as a synchronization-only PR; #90 explicitly excludes new benchmark families.
• It does invalidate any interpretation that the dashboard is data-complete merely because 8/8/8 sources parse and 1,774 rows are generated.
• Before PR #76 is proposed for integration into master, its body/docs should include a concise Known data gaps section linking #91 and the coverage-audit documents.
• A separate umbrella issue should be considered for source-inventory/catalog semantics and non-CV coverage gaps, because #91 covers CV only.

No additional synchronization-code defect was found in this pass, but the provenance/completeness wording should be tightened.

[TheHiddenObserver]

Final pre-merge review — ready to merge into PR #76 staging branch

Reviewed the current final head c394accf713d798ef8adf6474a75ebeb987e8ce1 again immediately before merge.

Findings

No new blocking correctness, synchronization, provenance, generated-asset, documentation, or CI issue was found.

• GitHub reports PR #99 open, non-draft, and mergeable.
• The base branch is still the unchanged PR #76 head 10874915f05dcae8f0124870e7d10de31848d42b.
• The synchronization head is still c394accf713d798ef8adf6474a75ebeb987e8ce1.
• There are no unresolved inline review threads.
• Benchmark Frontend CI #353 and #355 both completed successfully on the final head.
• Exact-head project validation run 31101612155 completed successfully, covering the Python 3.9–3.12 regression matrix, full CPU suite, static/docs contracts, and sklearn compatibility.
• The 482-file PR view is explained by importing 1,242 commits of current master history into the old dashboard staging branch. Relative to synchronized master, the net tree remains the dashboard pipeline/frontend/tests/docs/source bundle plus the reviewed configuration changes.
• No temporary synchronization or validation workflow is present in the final changed-file inventory.
• Registered-source integrity remains 8 registered / 8 available / 8 parsed / 0 skipped, with 1,774 runs and 36 represented models. This is correctly disclosed as registered-bundle integrity rather than complete method coverage; #100 and #91 own the remaining catalog/coverage work.

Required merge method

Use Create a merge commit. Do not squash or rebase this stacked synchronization PR: those methods would collapse or replay the imported master history and defeat the reviewed two-parent synchronization topology.

Post-merge gate

After merging PR #99 into feature/benchmark-frontend-dashboard, re-fetch PR #76, confirm its head moved to the merge result, check mergeability against current master, and run/review the resulting PR #76 checks before closing #90.

Conclusion

PR #99 is ready to merge into the PR #76 staging branch. It must not be merged directly into master. The lack of complete benchmark-family coverage is documented follow-up work, not a blocker for this synchronization-only merge.