/code-review max findings: the ledger promise, and four stale records

[Polichinel]

Five findings held on re-measurement. Two did not, and are deliberately not acted on:

• "pipeline-core is installed editable, so the drift guard reads a stale .dist-info." The venv runs a plain stale 2.3.0 from site-packages — datafactory_contract absent, MetadataSearchIncomplete absent. test_locked_environment is reporting correctly.
• "C-46's 426 → 430 passed is arithmetically impossible for a PR adding 36 tests." Those figures are scoped to PR C-30/C-46: the exclusion-manifest tripwire now runs in CI — the 2026-08-03 revert was the right observation with the wrong cause #280, which adds no test files. Reproduced CI-shaped against a tracked-files-only worktree: 426 → 430 exactly.

The one that mattered

The torn-run message truncates at five objects and pointed at "the run log" for the rest — but the only writer of a file_id was logger.info. Nothing in this package sets a level; pipeline-core removed its own setLevel precisely so the application owns it. A launcher running at WARNING would have written no id anywhere, making the message a promise to an empty file — and leaving the operator diffing the bucket by hand, which is the exact state C-105 exists to remove.

The complete ledger now goes out at ERROR, one greppable TORN-LEDGER line per object. That is also ADR-008's log-and-raise. Mutation-proven: dropping it fails the new test.

Four stale records

#	what it said	what is true
2	CIC: findability "refuses a falsy answer"	that is the superseded check — the question C-94 says "the previous delivery already answers yes to". Now documents the run-scoping and both refusals
3	C-84: "the two honest ways to make it pass"	the test offers three, and the omitted one (ACKNOWLEDGED_UNTIL) is the only remedy without console access — the register is where someone goes when the gate reddens
4	C-103 Location: unfao.py:137-151, "line numbers re-read"	the branches are at 169 and 174; 137-151 is the C-94 preflight added later in the same branch
5	header Last Updated 2026-08-17; C-105 "435/450"	seven later-dated records below it; wc -l says 434

Registered as C-109, because #4 is a class

Measured: eleven of 28 open entries cited a .py:line. Spot-checking six found three already stale — C-105's (written four days earlier) landing on staging.mkdir, C-106's on a section comment, C-30's on return summary. All drifted because a later change in the same week moved them.

C-103 is the sharp case: stale twice in four days, the second time carrying an explicit "re-read" claim that was false when it merged. A citation that decays faster than the review cycle is worse than a vaguer one, because it is confidently wrong.

C-103, C-105, C-106 and C-30 converted to path::symbol; eight entries still cite lines where the target genuinely is one. The entry records why the obvious guard fails (checking the file is long enough catches nothing — every stale citation points at a real line) so it is not proposed again cheaply.

Verification

All five symbol citations verified to resolve to a real def. Suite 468 passed, 3 skipped, 37 xfailed; the 26 failures are C-104 venv drift, green in CI. ruff clean.

One correction of my own, caught on re-reading: C-109 first said "nine of 37", from an inconsistent denominator. Re-measured with a single query — eleven of 28 before, eight of 29 after.