This guide is for a reader who wants to find proofs in the estate, or an agent that needs a small, sourced context. To work on Mathesis itself, see contributing.
The views are generated files; read them in place.
-
docs/atlas/generated/catalogue.adoc: one row for each repository with proof files of its own, split into theory and engineering. -
docs/atlas/generated/az.adocandorganisations.adoc: the same rows by name and by owner. -
docs/atlas/generated/registry.adoc: theorem-level claims, each with the grade that the canon reader derives from its home’s receipt. -
docs/atlas/generated/bridges.adoc: candidate relations between two repositories. -
docs/atlas/generated/index.json: the rows as one I-JSON/JCS document.
find lists the rows that match every word of a query, in repository name,
class, provers or status document.
bun scripts/atlas/census-atlas.mjs find agdaThe output is a JSON array of rows (57 for agda on 2026-11-11).
context returns the matching rows as a JSON document no longer than
MATHESIS_CONTEXT_CHARS characters (default 5000). It always carries a
caveat that counts are not verification.
MATHESIS_CONTEXT_CHARS=1200 bun scripts/atlas/census-atlas.mjs context lean-
Counts are textual: the census finds declarations by keyword, and holes and axioms by pattern. The kernel of each home repository is the authority.
-
A row is as fresh as its
censused_attime; the source commit is incommit. -
The frame holds public repositories only. Private repositories are a count in
docs/atlas/census/private-count.json. -
No grade in the atlas is typed by hand. A theorem without a receipt that the canon reader accepts reads
ungraded.