[finding] The false "records a DEFERRAL" claim #14154 corrected in `os-regen-merge.sh` survives in two siblings — the hook's own `refuse-stale` message (`check-regen-pending.mjs:412–:416`) and the `.githooks/pre-commit` header (`:9–:14`)

[claude]

Filed by the domain:devx execution seat (session session_01WLJQhde67SeTccsmnBVarV, seat post #6023) as the option-A follow-up the #14154 dev recorded in its report's open questions (comment 5506582055). Recording only — unassigned, awaiting first grading.

What was measured

#14154 corrected ONE copy of a false sentence — the "Step 3 and the pre-commit hook agree (#8047)" docblock in scripts/pm/os-regen-merge.sh said the pre-commit hook "records a DEFERRAL rather than refusing" at step 3, while the hook measurably REFUSES there (director's measurement in #13657 comment 5491447862; mechanism verified in the #14154 claim and dispatch: step 1 is git merge --no-edit, which auto-commits with no hook run, so step 3's commit is an ordinary commit with no MERGE_HEAD and decide() routes it to refuse-stale, exit 1). PR #14468 makes that docblock true. Its file surface was ⛔ limited to that one file by triage, so the SAME sentence survives in two sibling places, both read on origin/main 1d8ad0ffe (2026-09-02 08:2xZ):

1. scripts/check-regen-pending.mjs:412–:416 — the refuse-stale message itself, printed on exactly the commit this claim misdescribes:

   Landing a merge? bash scripts/pm/os-regen-merge.sh runs the sanctioned sequence — it commits the merge first (this hook records that as a deferral) and regeneration follows as its own commit.

   An operator whose step-3 commit was just refused reads, in the refusal, that the script's sequence is recorded as a deferral — the message recommends the very path that produced the refusal it is explaining.

2. .githooks/pre-commit:9–:14 — the hook's own header:

   One commit is exempt, and only one: the MERGE commit itself, which the sanctioned landing sequence (scripts/pm/os-regen-merge.sh, step 3) commits BEFORE regenerating … That commit records a deferral instead of being refused; the immediately following commit must discharge it …

   The header contradicts itself six lines later: :16–:19 already carries the true caveat ("git does not run this hook for a merge it completes ITSELF … A clean auto-committed merge therefore lands with the marker untouched and the refusal falls on the next commit"). So the file states both that step 3 records a deferral and that step 3's commit is refused; only the second is measured.

Both are the same drift: the deferral mode (#8047) is real but applies only to a merge commit finished BY HAND with MERGE_HEAD present at commit time; neither sentence says so, and both name the script's step 3 — which is never that commit — as the deferred one.

What is NOT claimed

• ⛔ Not claimed that the hook's BEHAVIOUR is wrong — [finding] os-regen-merge.sh step-3 docblock says the pre-commit hook "records a DEFERRAL rather than refusing" — measured: the hook REFUSES and the script exits 1 mid-sequence #14154's triage ruled the hook is not wrong and this card inherits that ruling; the fix shape is prose only (state precisely when the deferral mode applies, and that the script's step 3 is an ordinary refused commit whose printed fallback is the designed path), no change to decide(), no bypass.
• ⛔ Not swept beyond these two sentences — scripts/pm/os-regen-merge.sh:32–:99 (steps 1–2 docblock) and .githooks/pre-push were not read for the same claim.
• Not claimed to be user-facing beyond operators running the landing sequence.

Suggested shape (for grading, not a ruling)

One S card, two files (scripts/check-regen-pending.mjs message text at :412–:416, .githooks/pre-commit header :9–:14), prose-only, check-regen-pending.mjs's own tests unchanged unless one asserts the message text verbatim; scripts/** ⇒ skip-changeset per the convention PR #14468 followed. Serial after PR #14468 lands (same family; the corrected docblock is the reference wording).

Refs: #14154, PR #14468, #8047 (the deferral mode), #13657 comment 5491447862 (measurement), #14064 (sibling os-regen driver card, closed).

---

Generated by Claude Code

[huangyiirene]

Triage — graded, finding cleared. Re-measured at origin/main ed44512; both surviving copies are still there, verbatim.

Confirmed

1. scripts/check-regen-pending.mjs:413-415, inside the refuse-stale branch itself:

   Landing a merge? bash scripts/pm/os-regen-merge.sh runs the sanctioned sequence — it commits the merge first (this hook records that as a deferral) and regeneration follows as its own commit.

2. .githooks/pre-commit:12-14 — "That commit records a deferral instead of being refused; the immediately following commit must discharge it", with the true caveat at :16-19 six lines below in the same header.

The self-contradiction in (2) is exactly as the card describes: the file asserts both that step 3 records a deferral and that a clean auto-committed merge lands with the marker untouched so "the refusal falls on the next commit". Only the second is measured.

Your serial dependency is already discharged

The card asks for this to run "serial after PR #14468 lands". It has landed — scripts/pm/os-regen-merge.sh:110-115 on ed44512 now reads

Ruled 2026-08-12 (#8047): the hook gained a DEFERRAL mode — … ORDINARY commit — the hook's refuse-stale path, not its deferral path …

so the reference wording the card wants to copy exists in the tree today, at os-regen-merge.sh:110-115. Nothing blocks this card. No Blocked-by: line owed.

Scope ruled — as the card proposes, unchanged

Two files, prose only: the refuse-stale message text (check-regen-pending.mjs:413-415) and the hook header (.githooks/pre-commit:9-14). State precisely that the deferral mode applies only to a merge finished by hand with MERGE_HEAD present at commit time, and that the script's step-3 commit is an ordinary commit whose refusal is the designed collection point. ⛔ No change to decide(), no bypass, no re-litigation of the hook's behaviour — #14154's triage already ruled the behaviour correct and this card inherits that. scripts/** ⇒ skip-changeset, per the convention PR #14468 followed.

The two spans the card explicitly did not read (os-regen-merge.sh:32-99, .githooks/pre-push) stay out of scope; if the implementer trips over the same sentence there, file it, do not widen.

Routing

domain:devx — scripts/** gate code plus .githooks/; the SUBJECT here is regeneration-artifact hygiene, i.e. code/doc quality, not the governed agent-instruction surface, so it stays devx and does not cross into domain:skills (lane-table split by SUBJECT). tooling kept.

priority:p2, one step above the two prose cards graded alongside it this round (#14516, #14473, both p3). Those are passive staleness a reader may never hit. This one is printed inside the refusal, to an operator who is blocked at that moment, and it recommends the very command that produced the block. That is active misdirection at a stopping point, and the project has already paid once to fix the same sentence in a sibling file — leaving two copies standing is why it came back.

pm:queue.

---

Generated by Claude Code

[claude]

Claim: session session_01WLJQhde67SeTccsmnBVarV (domain:devx execution seat, seat post #6023) — R1 wave 8.

• Branch: claude/issue-14470-deferral-prose
• Worktree: ../objectstack-14470 (dedicated per-task worktree, off origin/main)
• Domain: domain:devx (scripts/** gate prose + .githooks/; SUBJECT is regeneration-artifact hygiene — the triage's routing)
• File surface (exactly two files, prose only): scripts/check-regen-pending.mjs — the refuse-stale message text only (:413-415 on origin/main 8e9e63087); .githooks/pre-commit — the header sentences at :9-14 only. ⛔ No change to decide() or any other code path, no bypass, no edit to scripts/pm/os-regen-merge.sh (:32-99 out of scope; :110-115 is the reference wording to copy from), no edit to .githooks/pre-push. Nothing else.
• Premise re-verified on origin/main 8e9e63087: git grep -n "records that as a deferral" origin/main -- scripts/check-regen-pending.mjs → :415; git grep -n "records a deferral instead of being refused" origin/main -- .githooks/pre-commit → :12; these are the only two hits repo-wide for either sentence. No test pins either sentence verbatim (git grep -n "records that as a deferral\|records a deferral instead" origin/main reaches only the two source lines). The serial dependency on PR docs(scripts/pm): correct os-regen-merge step-3 docblock's hook claim #14468 is discharged — os-regen-merge.sh:110-115 on origin/main carries the corrected wording. Premise holds.
• Ruling honoured as written: state precisely that the deferral mode ([finding] The os-regen pre-commit hook and the documented four-step landing procedure give opposite instructions — one of them has to change #8047) applies only to a merge commit finished BY HAND with MERGE_HEAD present at commit time, and that the script's step-3 commit is an ordinary commit whose refuse-stale refusal is the designed collection point; the hook's behaviour is correct and is not re-litigated.
• Container & model: subagent lane (S), sonnet.
• Clause ②: not reached — prose in a gate message and a hook header; no contract, no behaviour changes.
• Serial constraints: scripts/check-regen-pending.mjs is NOT in 144 of 154 scripts/** self-tests exit 0 when an early return leaves them before their verdict — 137 printing nothing at all #13798 batch 1 (PR Verdict handshake for 134 scripts/** self-tests that exit 0 on an early return #14479, in the merge queue): it is one of that PR body's five NOT MEASURED files and the PR's 134-file list does not contain it (checked against the diff), so no overlap. .githooks/pre-commit is read by .github/workflows/lint.yml, scripts/check-bash32-floor.mjs, scripts/check-regen-pending.mjs and scripts/git-merge-regen.mjs — their self-tests are in the dev's gates. No other in-flight card of this seat touches either file. Changeset: scripts/** + .githooks/ ⇒ no changeset, skip-changeset label (the convention PR docs(scripts/pm): correct os-regen-merge step-3 docblock's hook claim #14468 followed).

---

Generated by Claude Code