S2 — Close what is already fixed, and make register closing conditions machine-checkable (C-43, C-59, C-61) #183

Part of epic #181. Closes register C-43 (Tier 2), C-59 (Tier 3), C-61 (Tier 3) — Cluster K.

Background

Three register entries are filed Open while their own stated closing conditions are met.

C-43 says, in its own words:

"C-43 closes when tests/test_gaul_lookup_fidelity.py is committed and green — the entry should then cite the test, not the session."

The file is committed. pytest -q tests/test_gaul_lookup_fidelity.py → 18 passed.

C-59 and C-61 each say "Mitigation — landed 2026-07-31" and name four specific tests. All four exist:

And the code matches: scripts/build_gaul_lookup.py has zero bare assert statements and four LookupBuildError raises, including the index-uniqueness guard C-59 asked for.

This is the fourth and fifth instance of the same drift class in one day — C-63 and C-47 were found the same way during review-rr strategic, and both were also filed Open with their defects fixed. tests/test_register_integrity.py cannot catch any of them: it validates headings, counts, ID uniqueness, cross-reference resolution and section placement. It has no opinion about whether the code matches the claim.

The register's failure mode is stale-open, not stale-closed — which is the safer direction, but it inflates the open count with finished work and makes prioritisation lie.

Separately: Cluster M's header says "Five open concerns"; its body lists six. C-58 was added on 2026-08-01 and the heading was not updated.

Work

Part 1 — move the three entries. C-43, C-59, C-61 from ## Open Concerns to ## Resolved Concerns, each:

• retitled with the — RESOLVED suffix the section's convention uses
• citing the test, not the session, exactly as C-43 instructed
• keeping the scope paragraph C-43 already carries — transcription fidelity is proven, assignment correctness is views-datafactory#387 and is not this repo's to close. That distinction must survive the move; it is the whole reason C-43 stayed open after the forward-check.
• Cluster K updated with a closure note in the style of Cluster L's

Part 2 — fix Cluster M's heading. "Five" → "Six". Header counts updated.

Part 3 — make this class machine-checkable. Extend tests/test_register_integrity.py with a check that an entry under ## Open Concerns does not state a closing condition that is already satisfied.

The tractable, non-clever form: entries that name a test file as their closing condition use a recognisable phrase (closes when \tests/...`, Mitigation — landed`). Parse those, and assert either that the named path does not exist, or that the entry is not in the Open section. That is narrow and mechanical — it detects exactly the shape that occurred three times, and it will not fire on entries whose closing condition is prose.

Where the phrasing is ambiguous, prefer a false negative over a false positive: a guard that cries wolf gets deleted, and this repo has already deleted one over-fragile test this week for exactly that reason (C-61's -1 monkeypatch).

Acceptance criteria

• C-43, C-59, C-61 appear only under ## Resolved Concerns, each citing the test that discharges it.
• C-43's transcription-fidelity-vs-assignment-correctness scope paragraph survives verbatim, with views-datafactory#387 still cited.
• Cluster K carries a closure note; Cluster M's heading says six.
• Header counts arithmetically correct: total unchanged at 73, open 24 → 21, resolved 49 → 52.
• The new integrity check exists and is mutation-tested: fed C-43's actual pre-fix text (Open + the test file present) it must fail. A guard that passes on the bug it was written for is decoration — this repo has a standing practice of proving guards bite (test_clone_readiness.py::test_the_guard_would_actually_catch_a_violation).

Testing

pytest -q tests/test_register_integrity.py     # all 8 existing + the new one
pytest -q tests/test_gaul_lookup_fidelity.py   # 18 passed — the evidence being cited
pytest -q

Mutation test, run manually and reported in the PR:

# temporarily restore C-43 to Open, keep the test file present -> the new check must FAIL

Dependencies

None. Register + one test file. No production code.

Note for S9: this story establishes the closing-condition convention; every later story in the epic must write its register disposition in a form this check understands.

Files

• reports/technical_risk_register.md — C-43 (:368), C-59 (:516), C-61 (:560), Cluster K, Cluster M, header
• tests/test_register_integrity.py — the new check
• tests/test_gaul_lookup_fidelity.py — the evidence, unchanged
• scripts/build_gaul_lookup.py:152-183 — the four LookupBuildError raises, unchanged