[squid-protocol]

Fixes #4789

What

The two files every case PR and every evidence-refresh bot run rewrote, so that any two open PRs conflicted (#4759, #4780 and #4808 each went dirty 2-3 times on the ledger alone):

1. Det-sweep coverage ledger: one file per case. tests/equivalence/det_sweep_coverage.json is replaced by tests/equivalence/det_sweep_coverage/<case>.json (format det-sweep-coverage/2: the entry as before plus "format"). Two PRs, or a PR and the bot, that sweep different cases now touch different files.
   • det_coverage_ledger.py: only load() / write() know the files. build, check, last_coverage, fresh_coverage and migrate are unchanged, so the semantics stay the same: evidence report: keep a program's last measured level marked stale instead of dropping L3/L4 to zero on any harness change #4730 (stale-readable), evidence report: Db2 programs stay L1 after a Db2 refresh — the report build doesn't get the Db2 sweep's verdicts #4758 (an entry exists only while proven; a dropped entry is now a deleted file) and evidence: finer-grained fingerprints — a case goes stale only when the harness/runtime parts it uses change #4731 (per-component fingerprints). write() rewrites a file only when its text changes.
   • The readers go through load(): proof_blockers, evidence_report, proof_sweep --aggregate (CI det-sweep) and the bot's update. evidence_history.Tree.ledger() reads the per-case files, and the single file for commits before this one.
   • det_sweep_plan.py: a change to a case's own ledger file keeps the PR det-sweep narrow, to that case. Before, every case PR touched the single file and got a full sweep.
   • One-shot migration (run in this PR): python tests/tools/det_coverage_ledger.py split --all. All 65 entries were moved. load() equals the old file's cases exactly.
   • Compat check: while tests/equivalence/det_sweep_coverage.json exists, check (CI det-sweep), update and migrate fail. So does test_det_coverage_ledger.py::test_the_single_file_ledger_is_retired, which runs in every PR's suite. The failure message prints the split recipe below.
2. docs/language_status/cics_spec_status.md is bot-generated on main, the same way evidence report: regenerate on main by bot, advisory in PRs, strict at release (stop per-PR report churn) #4703 handles the evidence report. The evidence-refresh refresh job runs cics_spec_status.py render and commits the page. The page-currency test is skipped under EVIDENCE_REPORT_ADVISORY=1, which CI already sets. smoke-test.yml prints an advisory "would change" line, and test_the_page_renders still runs. PRs keep committing cics_spec_status.json (the measured census and crucible data).
3. The evidence-refresh workflow's add-paths now names tests/equivalence/det_sweep_coverage/*.json and docs/language_status/cics_spec_status.md. The blocker-slice and cics-command-slice skills are updated to match.

Evidence

• Readers before and after, on real data: evidence_report.det_coverage() (80 cases) and proof_blockers.ledger_coverage() (65) give byte-identical output on origin/main (single file) and on this branch (per-case files).
• det_coverage_ledger.py check on a sweep built from 6 real entries: "as expected". evidence_history.Tree('HEAD').ledger() == Tree('HEAD^').ledger() (per-case files vs the single file).
• New tests cover: one file per case, only changed files written, a dropped entry deletes its file, split --all, split --base (applies only the branch's own changes and keeps main's newer entries for other cases; refuses a base with no single file), the compat failure, and the plan narrowing.
• pr_gates.py --fast: 5/5 pass. --ratchets: 7 pass, 0 fail, 0 skipped.

Recipe for in-flight PRs (any branch from before this merges that touched the single ledger)

Merging origin/main gives a modify/delete conflict on tests/equivalence/det_sweep_coverage.json. Do not hand-merge it:

git merge origin/main                      # CONFLICT (modify/delete): tests/equivalence/det_sweep_coverage.json
python tests/tools/det_coverage_ledger.py split --base "$(git merge-base HEAD MERGE_HEAD)"
git add -A tests/equivalence/det_sweep_coverage.json tests/equivalence/det_sweep_coverage/
git commit --no-edit                       # finish the merge (resolve any other files as usual)

--base must be the merge base from before the merge, where the single file still existed. If the merge is already committed with the file still there, use --base "$(git merge-base HEAD^1 HEAD^2)". split --base writes only the entries your branch added, rewrote or dropped. Every other case keeps main's file. Alternatively, git rm the single file and re-run det_coverage_ledger.py update <your sweep DIR(s)>.
Also, do not commit docs/language_status/cics_spec_status.md: on a conflict take main's copy (git checkout origin/main -- docs/language_status/cics_spec_status.md). The bot re-renders it after merge.

If main changes the single file again before this PR lands, this branch merges main and re-runs split --all instead of hand-merging.

Idempotency (#4813)

Merged with #4813: build() keeps an entry whose numbers and inputs did not move (measured_at included), and write() rewrites a file only when its text changes. The result is that a bot re-sweep touches only the files of cases that actually moved. Checked on the real ledger: update on an unchanged sweep reports "65 cases, 0 file(s) written or removed". Test: test_an_unchanged_sweep_touches_no_case_file.

Follow-up (not in this PR)

• Evidence records (tests/equivalence/*/evidence.json) are already per-case files. Moving their re-proof out of PRs is a separate step.
• publish.yml does not gate on the CICS page being current (it is internal status, not a purchaser-facing report).

🤖 Generated with Claude Code

[github-actions]

🐦‍⬛ Muninn Security Scan

✅ No security issues found.

🐦‍⬛ Powered by Muninn · Skald Lab

[squid-protocol]

CI shepherd: this PR conflicts with main, so no checks ran. Merge origin/main into it and regenerate generated files with their tools (never hand-merge them; see .claude/agents/gitgalaxy-pr-worker.md), then push.

[squid-protocol]

CI shepherd: real failure(s) on this head. Reproduce locally with the repro line; don't re-push to find out.

PR #4811 @ 4af15d1c8d3b: 1 failed check(s)

== ground-truth (failure) -- real
   https://github.com/squid-protocol/gitgalaxy/actions/runs/37987851394/job/114014206133
   repro: python tests/tools/pr_gates.py --ratchets --only-ratchets ground-truth
   log tail:
     (log unavailable: the response contains terminal escape sequences; pass --allow-escape-sequences to output it anyway)