feat(census): JSON Lines, with the source commit and census time of each row (WO 2.1)

.machine_readable/mint-verbatim.txt

@@ -34,4 +34,4 @@ tests/workflows/julia_library_test.sh      # PACKAGE_UUID brace-token fixture
 tests/workflows/resolve_template_pin_test.sh  # template-slug fixtures
 
 # Estate census is evidence, not generated template identity.
-docs/catalogue/census/repos.json
+docs/catalogue/census/repos.jsonl

Justfile

@@ -716,7 +716,7 @@ uncloak:
 
 # ── mathesis: estate maths catalogue ──────────────────────────────────────
 
-# Re-run the proof census over docs/catalogue/census/repos.json and regenerate the
+# Re-run the proof census over docs/catalogue/census/repos.jsonl and regenerate the
 # catalogue and the atlas built from the same rows (CI checks both, so a census
 # that refreshes one leaves the other stale)
 census:

README.adoc

@@ -36,14 +36,14 @@ organisation-specific working material belongs in separate protected repositorie
 [cols="1,3"]
 |===
 |`docs/catalogue/catalogue.adoc` |The catalogue: one row per repository with proof files of its own, split into theory and engineering. Generated; do not edit by hand.
-|`docs/catalogue/census/` |What the catalogue is rendered from: `catalogue-rows.json` (one row of counts per repository, no file paths), `repos.json` (the census input), `unreachable.json`, `template-paths.txt` and `forks-with-own-proofs.txt`. Each `.json` file is one I-JSON/JCS array.
+|`docs/catalogue/census/` |What the catalogue is rendered from: `catalogue-rows.jsonl` (one row of counts per repository, no file paths), `repos.jsonl` (the census input), `unreachable.jsonl`, `template-paths.txt` and `forks-with-own-proofs.txt`. Each `.jsonl` file is JSON Lines: one I-JSON/JCS record on each line, so a change to one row changes one line. Each row also names `commit`, the commit that the census read, and `censused_at`, the UTC time it read it.
 |`scripts/census/` |`run-census.sh` (whole census), `update-rows.sh` (refresh named repositories' rows only), `census-repo.sh` (one repo), `summarise.sh`, `build-rows.sh` (raw census to rows), `render-catalogue.sh` (rows to catalogue), and proof-craft's `proof-holes.sh`.
 |`Justfile` |`just census`, `just catalogue`, `just catalogue-check` (the committed catalogue matches the committed data).
 |===
 
 == How the census works
 
-. *List repositories.* `docs/catalogue/census/repos.json` is the input. Agent sessions build
+. *List repositories.* `docs/catalogue/census/repos.jsonl` is the input. Agent sessions build
   it with the Claude Code remote `list_repos` tool, because the GitHub API is
   not reachable from them.
 . *List proof files cheaply.* Each repository is cloned blobless with
@@ -75,7 +75,7 @@ The first census covered 476 repositories:
   the end of the catalogue, among them `cut-calculus`.
 * *Possibly missing:* the listing returned only repositories pushed since
   2026-09-15, so older, untouched repos may be absent. Add them to
-  `docs/catalogue/census/repos.json` and rerun.
+  `docs/catalogue/census/repos.jsonl` and rerun.
 * *Theory or engineering* is a heuristic: more Agda, Lean, Rocq or Isabelle
   files than Idris2 or SPARK files reads as theory. Confirm each row against
   PROOF-POLICY.

docs/architecture.adoc

@@ -10,10 +10,10 @@ Mathesis is the estate's mathematical source-discovery and proof-orientation ind
 
 [source,text]
 ----
-Repository census input (docs/catalogue/census/repos.json)
+Repository census input (docs/catalogue/census/repos.jsonl)
   -> scripts/census/ (source and prover-file discovery)
   -> corpus counts, holes/unsafe candidates, reachability observations
-  -> docs/catalogue/census/catalogue-rows.json
+  -> docs/catalogue/census/catalogue-rows.jsonl
   -> generated docs/catalogue/catalogue.adoc
   -> scripts/atlas/census-atlas.mjs
   -> docs/atlas/generated/ (A-Z, class, owner; index.json)

docs/atlas/README.adoc

@@ -3,8 +3,10 @@
 
 This is a *generated, read-only directory of proof-source repositories*, not yet a
 complete alphabetic mathematical concept atlas. It derives from
-`docs/catalogue/census/catalogue-rows.json` and never alters
-or reclassifies the source census. A–Z means repository-name order, not a claim
+`docs/catalogue/census/catalogue-rows.jsonl` and never alters
+or reclassifies the source census. Each row names the commit that the census read
+(`commit`) and the time it read it (`censused_at`, UTC), so each count has a
+source. A–Z means repository-name order, not a claim
 that these are all the mathematical concepts relevant to the estate.
 
 The first views are A–Z repositories, theory-versus-engineering census class,
@@ -28,7 +30,7 @@ No private-source records are added or published by this generator.
 Census rows change in one way only: `just census` (the whole frame) or
 `just update-rows OWNER/REPO...` (named repositories). Both rebuild the
 catalogue and this atlas from the same rows. Nobody edits
-`catalogue-rows.json`, `unreachable.json` or the generated views by hand.
+`catalogue-rows.jsonl`, `unreachable.jsonl` or the generated views by hand.
 Only one pull request that changes rows is open at a time; record it under
 "Open census work" in link:../status/ROADMAP.adoc[`ROADMAP.adoc`] before it
 opens. The deed states the same rule in `(bot-directives (census-writer))`.