Working toward 1,500 source-backed AI records.

The catalog is expanding with a distinct queue for candidate model repositories. Discovery is not verification: model names, identity, deployment, licensing, and capabilities are never assumed from a repository listing.

Editorial intake prepared

R39 provides 498 primary source-listed model candidates plus 42 reserve candidates across 17 model-publisher namespaces. They are packaged for private editorial intake and are not counted as independently verified or publicly indexed until the verification and deployment process completes.

How new records are added

1. Import candidate model identifiers and canonical source URLs into a private queue.
2. Deduplicate against existing systems and model lineages.
3. Review primary-source evidence and field-level uncertainty.
4. Verify the exact draft revision; the Owner must approve activation.
5. Generate and deploy a validated static release before changing public counts.

Evidence standard

Hugging Face model pages are treated as identity and discovery leads, not proof of current commercial availability, original authorship, benchmarks, licensing, pricing, or product access. Historical models remain eligible where distinct identity can be established. No candidate is automatically published.