Documentation Update Notes

Documentation update notes

Review scope — 3 November 2026

All six pre-existing wiki pages were inspected and replaced or consolidated. Existing page slugs are retained. A home page, navigation sidebar, dedicated dataset catalog, training guide, architecture guide, and FAQ now provide a progression from using the tool to operating it.

Previous guidance	Current documentation
Java 7, PHP/PEAR, Sendmail, manual classpaths	Seven-service Docker Compose stack and host administrative approval setup
Old /api/compound and /api/model examples	FastAPI /api/v1 and predictor /api/v3 contracts
Log-odds as the prediction output	Class distributions, predicted-class confidence, and applicability evidence
Universal 20,000-record upload limit	Current analyzer defaults plus deployment-specific resource/proxy limits
Registration and email completion workflow	Persisted jobs, browser progress, and explicit administrative controls
Release-specific dataset counts and copied abstracts	Complete live catalog with concise descriptions and exact label counts
Training implicitly leading to predictions	Immutable builds, human approval, and publication pointers

Evidence used

• Current working source rooted at repository base commit 9ec87f2d15beaa215e52fa348aa7ceb0bc1d16ec, including local changes present during review. That base commit alone does not describe every reviewed change.
• Model-creation allowlists/validation in html/molclass/api/app/v3_models.py.
• Upload, registration, dataset, and job routes in the FastAPI application.
• Predictor controller/service, Java model worker, Compose configuration, Dockerfiles, and repository technical guides.
• Live read-only catalog, per-dataset target metadata, published-model registry, and model options retrieved on 3 November 2026.
• Existing wiki assay descriptions and source references for scientific context.

No prediction, upload, training, approval, or database mutation was performed to verify the catalog.

Source versus public deployment

The live catalog contained 21 datasets and 63,866 dataset records; the published-model endpoint returned 21 models. All nine advertised feature profiles were READY_WITH_EXCLUSIONS.

The current source supports sealed feature-profile revisions and an updated Ensemble implementation. The live model-options response still exposes older profile metadata and a legacy J48/KNN Ensemble description. These pages distinguish those observations and do not assert that local source changes have been deployed or that every published build uses the same training implementation.

Dataset 23 exposes 17 target codes with different labeled support totals. The live API does not supply a complete expansion dictionary for those codes; the catalog preserves the codes rather than inventing organism/assay mappings. Dataset 31's imported filename specifies human microsomes.

Maintenance

Refresh catalog and target counts together and advance their verification date. Confirm pagination and completeness. Review code changes against deployed behavior before claiming public availability. Prefer links to focused repository guides for operational details that change frequently; retain explicit limitations such as the stock container's lack of a web approval bridge.