ship: a human-gated review finding has no way back in — the person's answer on the PR is never consumed, every re-issue holds again (#2140 ×2)

[justinhelmer]

A human-gated review finding has no path back into the pipeline once the person answers: an adopted attempt goes straight to a review round, the reviewer finds the same human-gated finding, and the pipeline ends held again — the person's answer on the pull request is never consumed.

Evidence (record 0072's acceptance, PR #2140, thread slack:C0BRRHKFLCB:1790011954.424649):

• Attempt 1 (9235f373): review F2 Cold-reader acceptance gate was not independently run (minor, humanGated: true); the coding child declined it correctly ("this coding context cannot manufacture independent evidence"); the pipeline ended held.
• The maintainer ran the gate with an independent reader and posted the three-part quote as a PR comment (docs(docs): accept and plan one live run state #2140 (comment), 19:2xZ).
• Attempt 3 (d69afa9a, re-issued with the same text at 19:20Z) adopted docs(docs): accept and plan one live run state #2140 and opened review round 1 directly (no coding round: adoption skips to review); the reviewer read the record (still "pending") and re-raised F2; the pipeline ended held at 19:22Z. The comment holding the answer was on the pull request the whole time. $0.30 and 3 minutes per attempt, and no attempt can ever pass.
• Recovery in use: a directive agent:coding … on branch plan/take-decision-record-007-69a890/u1 that writes the receipt into the amendment, then a fourth agent:ship re-issue to review it — two hand-posted steps for a finding the system itself marked as waiting for a person.

Invariant: a humanGated finding is a question to a person (record 0054: the door asks before it fails), and the person's answer resumes the unit. Direction: when a review round yields only human-gated findings, the unit does not end — it parks with the question (record 0051's idle state) and resumes on the next human input on the pull request or the unit thread (a PR comment or thread reply from a person), which becomes the fix round's brief (the finding plus the answer). An adopted attempt on a pull request whose newest human comment postdates the last verdict runs that fix round before the review, never review-first. Under record 0073's direction this is one orchestrator decision, not a runner branch: file it so the fixture exists.

Tests: a review with one human-gated finding parks the unit; a person's PR comment resumes it into a fix round carrying the comment; the adopted attempt with a newer human comment runs fix-then-review; a human-gated finding alone never ends a pipeline held.

Related: #2141/#2150 (no message tells the person to act — the hold's text asks the person for evidence with no way to give it), record 0051 (idle), record 0054, record 0072 (the fixture).

[justinhelmer]

Fixed by #2167 (squash 378e270). A review round whose findings are all human-gated no longer ends the unit held: the unit parks with the question ("Waiting for you:" report), the person's next input on the pull request or in the unit thread resumes it into a fix round carrying the finding plus the answer, and an adopted attempt whose newest human comment postdates the last verdict runs fix-then-review. Along the way: PR-comment answers are actor-authorized; the post-answer re-review is exempt from the round cap; an evidence-only answer passes the same-head guard; gate recovery follows every GitHub pagination link. Both review arms LGTM at head 648088f (pipeline 00:07Z, standalone 00:12Z); five ship attempts, two directive fixes.

Receipt runnable when: worker:bot ≥ 378e270 (rides 1.260.3) and a review posts a human-gated finding on a live unit — the unit parks instead of ending held, and the person's answer resumes it (the next record acceptance is the natural fixture).