[mdsmithaustin]

Why

The ledger claim for babysit-body-only-finding-acknowledged-in-pr-comment said any author or maintainer comment with the review's URL acknowledges a bot review body. Since #305 the watcher ignores a comment that opens with the pr-recap start marker, so a grader reading the claim expected the wrong thing.

What changed

• The claim now says the acknowledging comment must not open with the pr-recap start marker.
• The claim's script check gains a case: an author comment that opens with the marker and links the review still leaves the watcher at exit 8.

Scope

Ledger text and one test case. No watcher code changes. Other text that states the rule already matched the code: babysit step 8 names the exception, and the guide and watcher hint only say what to post.

Blast Radius

Promise grading for one ledger entry. Low risk.

Verification

• With the recap guard deleted from review-bodies.ts, the new case fails (the watcher returns READY). With the guard in place, it passes.
• ledger.py audit: 0 problems. ledger.py format --check: exit 0.
• tools/gate-check.sh lint passed. In the test tier, promise unit tests passed. One host.test.ts lookup case timed out at 5s under load and passes alone; this PR touches no TypeScript.
• The existing review-bodies.test.ts case "rejects a pr-recap comment that links the review" covers the code path.

Closes #315

✅ Low: This change rewords one ledger claim and adds one test subtest. The new claim matches acknowledgment(), which checks trimStart().startsWith(RECAP_START) together with a trusted author and linksReview. The new subtest is a trusted author comment that starts with a newline and then the recap marker, and the . after the URL still satisfies linksReview. So only the recap guard keeps that comment from counting, and the subtest fails when the guard is removed.

Updates from git push no-mistakes

[mdsmithaustin]

Verdict: PASS+NOTES
Head: 4454c09
Patch-id: 981bd50d50d66c5aae0752536aff14aefac48c5a
Docs: pass

Root verdict from a full swarm at the code-ready head. Three lanes ran: gates and live (sonnet@high), claim vs code (sonnet@xhigh), and graders and ledger (sonnet@xhigh). The two review lanes stepped down from the opus work model.

• Gates at head:
  • tools/gate-check.sh lint passes.
  • ledger.py audit reports 0 problems, and format --check exits 0.
  • The promise suite runs 848 tests OK on the pinned Python 3.14.7. On 3.14.8, one fsmonitor test errors identically on main.
  • CI is green.
• Regression: deleting the recap guard in review-bodies.ts makes the new case fail with 0 != 8, because the watcher returns READY. Main already has the guard. This PR pins it with a test and corrects the claim.
• Live: the real watcher, through the fake-gh sandbox:
  • keeps the bot review open for an author or member recap comment that links it, including one with leading whitespace;
  • clears it for a plain author comment, for a marker mid-body, and for a recap followed by a real acknowledgment.
  • Each case matches the new claim.
• Consumers: nothing quotes or hashes the claim text, so no binding breaks.

Notes, filed as #325:

• the new case pins only one direction of "opens with";
• the claim omits the literal marker;
• older claim and code differences (trust trio, Copilot scope, head-only, the 100-comment window);
• ledger.py audit never compares a script claim with its test.