fix(release): #1189 — land the v0.66 cold review and correct the false headline it found, in every place it propagated

[avrabe]

Lands the v0.66 clean-room cold review of 2348afe0 and corrects what it found — in every place the false figure had propagated.

Zero shipped-byte defects. All findings are in the release's account of its own work.

The false headline

The CHANGELOG claimed "84 of 89 pinned parity entries (585 of 595 assertions) went from a WRONG ANSWER to correct-or-refused" — and contradicted its own scorecard fifteen lines above, which said "CLOSED 77 wrong-answer pins". Re-derived by ast.literal_eval over the parity KNOWN table at both ends:

v0.65.0 tag :  103 entries, 622 assertions
main (v0.66):   26 entries,  55 assertions
CLOSED      :   77 entries, 567 assertions

Every term was wrong. Corrected to 77 of 103 / 567 of 622.

The root cause was an unnamed base

"84 of 89" was a real measurement — against the table after RQ-65-DECLINE converted #1205/#1208/#1213, i.e. mid-release. That is a valid base for the lane's own work, and its triage document keeps it, now annotated. It is not the release-over-release figure. Every corrected site now names its base, because two different numbers were both being called "89 of 89".

The residual was understated, and two live issues went undisclosed

Published as "5 entries / 11 assertions", naming only #1215. Actually 12 entries / 28 assertions across three issues: #1215 (11), plus #1204 (15) and #1206 (2) — both untouched by this release and still computing wrong answers. The table's full residual is 26/55; the balance is the #539-grow-fails envelope, which is not a miscompile.

RQ-66-UNWATCHED was itself left proposed after shipping

The same #1250 gate hole this release fixed for three other artifacts — missed on the one artifact whose own subject is findings that nothing watches. Now implemented with landed:, so 7 of 8 carry the status the release title claims, where the review measured 6.

Also corrected

• A stale [Unreleased] emulation floor: 367751, two merges out of date, against the live 367759.
• A "five frozen goldens" count matching neither the three files touched nor the seven pinned values changed.

Confirmed clean

Nine claims held up under direct re-derivation rather than acceptance: the bug attribution for #1210/#1211/#1214, the memory64 refusal and its census effects, the memory64-family false positives, the pin-count disclosure, the waiver mechanism, mutants_untested staying 4, the mutation survey's 27%, VARVE's RECORDED disposition, and ARCHMODEL's cut-time date.

claim_check 0 · status_evidence 0 · version_pins 0 · oracle_wiring 0

Supersedes #1252, which carried the review document alone.

Refs #1189 #1250

🤖 Generated with Claude Code

https://claude.ai/code/session_01YJK5LZZEkV5smCY1jKn18L

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!