About & Methodology

What this is

A curated catalog of generative-AI / ML tasks relevant to game and anime production, organized as an Area → Task tree. For each task we track:

• Open-source models & papers — the bulk of the catalog (each anchored to an arXiv paper and/or a real code repository), and
• Proprietary / industry tools in active production use.

How entries are discovered

A GitHub Actions job runs weekly. It invokes an agent equipped with read-only discovery tools:

• arXiv — recent papers per task
• Hugging Face — trending papers and model releases
• GitHub — new/active repositories
• DuckDuckGo — general web search

The agent rotates its focus across areas week to week to broaden coverage.

Why you can trust the data

The agent only proposes entries. A deterministic pipeline then:

1. Validates every record against a strict schema and the taxonomy.
2. Recomputes the canonical id and de-duplication key (it never trusts the agent's).
3. Verifies that the primary link actually resolves, and requires open-source entries to be anchored to an arXiv id or a GitHub repository.
4. De-duplicates against everything already in the catalog.
5. Merges conservatively — filling only empty fields, never overwriting human edits.

The result is a Pull Request that a human reviews and merges. Card previews link each entry's own published image (Open Graph image → GitHub social card → a generated initials tile) — no image files are stored.

Automated curation

Summaries are machine-generated and kept deliberately neutral. They may contain mistakes — treat this as a discovery aid, not an authoritative source. Corrections via PR are welcome.

Editing the taxonomy

The Area → Task structure lives in a single file, taxonomy.yml. Adding, renaming, or removing a task is a one-file edit; the navigation and all pages regenerate from it on the next build.

License

Project code is MIT. Catalog entries are bibliographic metadata about third-party works that remain under their own respective licenses.