fix(selftests): extract-advanced-prs negative checks read the result line, not the random fixture sha

[LukasWodka]

Summary

selftests went red on main at eb89c1e5 (run 34507764093) with the extractor itself correct: extract-advanced-prs-selftest.sh's "promotion: the release-train PR is not attributed" check grepped the whole extractor output for 901, and that output carries ::warning::commit <40-hex sha>: ... attributed from its SUBJECT for the subject-fallback path. The fixture commits are minted at run time ($RANDOM content + the wall clock), so their shas are random hex — that run's was e57d5bafc191901afac8fc19a0c1e89c21749556, which contains 901. Same latent coin toss on the 979 and 3013 negative checks. Roughly 1% of runs per three-digit needle, on every push to every branch.

The file is byte-identical on develop, so develop's green runs were luck, not a fix.

Change

• result_line() helper: the script's Found PRs: line, i.e. the attribution the checks are about.
• The three "is NOT attributed / NOT used" negative assertions grep that line instead of the full output. The positive Found PRs: <n> assertion beside each one still carries the "line is present" half, so an absent result line cannot make the negative pass vacuously.

No change to scripts/extract-advanced-prs.sh.

Evidence

Deterministic reproduction (script in the lane scratchpad; loops the committer date until the fixture sha contains 901, then runs the real extractor with the promotion stub):

tries=9 fixture sha=3dff8cf0dd2249010965663da891ea964b3c8303
::warning::commit 3dff8cf0dd2249010965663da891ea964b3c8303: the API names no merged PR for it; attributed from its SUBJECT (900 ). Verify its card.
Found PRs: 900
OLD assertion (whole output): FAIL - '901' found (in the sha)
NEW assertion (result line 'Found PRs: 900'): pass

Mutation proof — drop the promotion exclusion from the extractor (API_JQ → select(.merged_at != null)), run the fixed selftest:

  FAIL  promotion: the release-train PR is not attributed
  FAIL  promotion: it falls through to the subject
extract-advanced-prs selftest: 15 passed, 2 failed

Anchor verified applied (git diff --stat = 1 file changed), restored → 17 passed, 0 failed.

Test plan

• bash scripts/tests/extract-advanced-prs-selftest.sh → 17 passed, 0 failed
• make selftests (full suite) → all green except selftest-pipefail-early-close's gawk is not installed, so the CI awk cannot be checked here — a macOS-host drift (BSD awk), unrelated to this change; CI runs gawk.
• make lint not runnable locally (ruff guard); this PR touches one bash selftest, no Python.

Tracked internally.

---

Note

Low Risk
Test-only change to bash selftest assertions; no production or extractor behavior changes.

Overview
Fixes flaky extract-advanced-prs-selftest.sh failures where negative checks searched the entire extractor output for ticket/PR numbers (979, 3013, 901). Warnings include the fixture commit’s runtime-random 40-char SHA, so ~1% of runs falsely match those needles even when attribution is correct.

Adds a result_line() helper that isolates the ^Found PRs: line and routes the three “must NOT attribute” assertions through it; paired positive Found PRs: <n> checks are unchanged. scripts/extract-advanced-prs.sh is untouched.

^{Reviewed by Cursor Bugbot for commit a5bb86c. Bugbot is set up for automated code reviews on this repo. Configure here.}

[LukasWodka]

bugbot run

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit a5bb86c. Configure here.}

[saqlainsyed007]

Approving. A correct fix for a real flake, scoped the right way. CI green, mergeable, no threads.

The three assert_not_out negative checks grepped the whole extractor output for a three-digit needle (979, 3013, 901), but the script also emits ::warning::commit <40-hex sha> lines and the fixture commits are minted at run time, so their shas are random hex — a negative grep for 901 reddens ~1% of runs whenever the sha contains those digits, which is exactly what broke selftests on main at eb89c1e (…901afac) with the extractor itself correct. Narrowing each negative to result_line "$out" (only the ^Found PRs: attribution line) targets the semantically-correct surface — attribution appears only there, and the incidental warning shas are noise for these assertions.

Checked the vacuity angle the comment calls out: an absent result line returns empty, so a negative assertion on it would pass trivially — but each of the three is paired with a positive assert_out "Found PRs: NNN" (985/914/900) that forces the line non-empty and correct, so the pairing carries the check. Reasoning and the failure citation are accurate. Good hygiene fix.