Public, AI-agent-friendly knowledge bases of government rules, processes, and data — each corpus is a Markdown/graph repository with strict provenance guardrails and an MCP server speaking a common interface contract.
⚠️ Everything here is unofficial and non-authoritative. Every document links to its authoritative source; always verify there.
The corpus registry lives there, not here. It reads each corpus's published
corpus-index.json at build time, so statuses and document counts are current by
construction rather than by someone remembering to edit this file.
The hand-maintained table that used to sit here was a second copy of that list, and it drifted: it described corpora as Planned that already held thousands of documents, and carried an archetype that had been superseded. One generated registry beats two hand-kept ones.
| Repo | Purpose |
|---|---|
| corpus-toolkit | Shared schemas, CI workflows, validators, MCP framework, reference architecture |
| corpus-template | Template repo — start every new corpus here |
bill → statute → rule → policy → standard → dollars → audit →
federal requirement → revision.
Each corpus adds nodes and edge types; the shared resolve_citation contract
lets agents walk across corpora. revision is the exception — every corpus is a
git repository, so its history already is the diff trail; it will never be a
corpus of its own.