RQ-61-R4BLIND (#1119): R4-issue scope resolution + R10 window attribution floor — red-first on b4860e4e/eefa19ef (+ RQ-61-EXTRACT flip)

[avrabe]

Task A — RQ-61-R4BLIND (#1119): R4 could not see conventional-commit delivery subjects

Two rules, because the measured corpus itself refutes half of option 2 as filed in #1119:

R4-issue — scope-position resolution through fields.issue

fix(#1040): ... resolves through the issue map to RQ-61-A32RELOC and gets R4's existing acknowledgment demand (claiming status, or the PR in landed:). The marker discipline is the scope position only: eefa19e's own subject proves the description position names a cause, not a target — its #1104 is the PR that introduced the defect, and resolving it would have misattributed the commit. Scoped to the release window (first-parent commits since the previous minor's tag), because fix(#1085): ... (#1090) is v0.60-era work on the issue RQ-61-EVIDENCE now holds — matching frozen history against the current issue map would misattribute it.

R10 — window attribution floor (what actually catches eefa19e)

Correction to the issue's premise: fields.issue resolution alone cannot catch eefa19e — its subject names neither RQ-61-ORACLEFLOOR nor its issue (#1113). No resolver can attribute it; but its silence can be noticed. R10: every delivery-typed (feat|fix|perf|proof|test) first-parent commit in the release window must be attributable to some release artifact — a known id anywhere in the subject, a known issue number anywhere, or its PR named in any artifact's landed:/verified-by:. Unattributable is red, and the red forces exactly the landed: PR #1112 line that #1120 had to write by hand. Attribution errs toward green (a mention suffices) because R10 demands non-silence, not a status flip — so the passing-mention rule the artifact protects is intact.

Red-first transcript (measured corpus, live repo)

At 331bbd8b — the last commit before the #1120 hand-flip; both delivery commits in first-parent history, both artifacts still proposed:

OLD gate: ... 0 failures                    EXIT=0   <- the blind spot
NEW gate:
FAIL R4 RQ-61-A32RELOC: issue-anchored delivery commit on main
     (fix(#1040): -> issue #1040 / PR #1116) but status is `proposed` ...
FAIL R10: delivery-shaped commit in the release window is attributable to
     NO release artifact: 'fix(oracle): #1104 shadowed the aarch64 builder
     guard and left its oracle pinned to the superseded wording (#1112)' ...
... 2 failures                              EXIT=1

On current main (post-#1120 statuses): 0 failures, status-evidence-window: 3 delivery-shaped commits since v0.60.0 — 3 attributed, 1 issue-anchored delivery claims.

Negative direction (required by the artifact)

• Live: 0cb36bc7 (fix(rivet): un-red main's R4 — RQ-61-VCLOSURE's ...) stays green — the id mention attributes it under R10, and no acknowledgment is demanded of VCLOSURE.
• Fixtures (R4BlindSpot1119, 9 tests): description-position #N never becomes a delivery claim even when N is a known issue; Revert "fix(#1040): ..." is not delivery-shaped; an ambiguous issue (two holders at the highest release) warns and never guesses; process types (chore/plan/docs) are exempt; an underivable window skips loudly (WINDOW-SKIP) and the new status-evidence-window: CI grep turns that skip into a red.
• Mutation-verified: neutering SCOPE_ISSUE kills the b4860e4 replay + ambiguity control; emptying DELIVERY_TYPES kills the eefa19e replay.

What the fix preserves, per the artifact: the id-prefix anchor and its DELIVERY_FLOOR are untouched; R4 stays first-parent; no "id anywhere in the subject" delivery claims.

RQ-61-R4BLIND flipped to implemented with a verified-by recording the above (including the eefa19e correction).

Task B — RQ-61-EXTRACT flipped to implemented (batched to save a CI cycle)

Every done-when clause re-verified independently on main, not taken from the PR body:

1. Snapshots gone: coq/*.ml, coq/*.mli, extracted/, validation/, compiler/*.ml, the dune tree and root dune-project all absent; the delivery diff of a5c75caf (RQ-61-EXTRACT (#1080): delete 98 stale extraction snapshot files + the dune tree nothing runs; docs stop contradicting their own headers #1121) deletes exactly 98 .ml/.mli files, matching landed:.
2. Doc no longer self-contradicts: docs/validation/VALIDATION_STATUS.md opens with the historical note and greps clean of "Successfully implemented" / "Complete and ready for use" / "40 files total".
3. Extraction still exercised: the .v target untouched; CI's Bazel Build & Proofs passed on PR RQ-61-EXTRACT (#1080): delete 98 stale extraction snapshot files + the dune tree nothing runs; docs stop contradicting their own headers #1121; main green at the merge.

Task A cross-check the brief asked for: a5c75caf was never in the R4 blind class — its subject is id-first and landed: names PR #1121, so old and fixed R4 both see and accept it.

Gates run locally

• status_evidence_check.py: 0 failures, window line present, on this tree
• test_status_evidence_check.py: 51/51 (42 pre-existing + 9 new)
• test_claim_check.py: 50/50; claim_check.py claims.yaml: exit 0
• rivet validate: FAIL (40 errors) — byte-identical to main HEAD 465f23a2 under the same local rivet (older than CI's); nothing introduced
• No .rs touched

Refs #1119, #1080

🤖 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!