Read Bold-Led Change Bullets in the Copilot Overview as Narrative, Not Metadata Labels #2182

scripts/pr_review.py status returns status=UNRECOGNIZED_REVIEWER_OUTPUT (exit 43) on a clean Copilot review whose overview lists its changes as bold-led bullets. Every develop -> main promotion whose overview uses that list form hits the refusal, so the review loop cannot close on it.

The shape

The ccr-overview-v2 body carries a collapsed What changed in this PR block. Inside it, a promotion's overview lists each carried change as a bullet that leads with a bold label, for example (constructed):

<details>
<summary><strong>What changed in this PR</strong></summary>

**Changes:**
- **Widget refresh (#12):** Bumps the widget pins.
- **Verification:** All pins land consistently.

</details>

The digest reports each one as metadata label: Widget refresh (#12) and metadata label: Verification.

Cause

LABEL_LINE matches any - **Label:** bullet anywhere in the body with fences and code spans stripped, and unrecognized_in checks every match against VETTED_LABELS. A narrative bullet inside the overview's description block is not a metadata field, but it has the same Markdown shape as the Review details bullets the check exists for. Feature pull requests whose overview has no bold-led change bullets pass, which is why the same reviewer format read clean earlier.

Done looks like

• Metadata-label vetting no longer reads bullets inside the overview's What changed in this PR collapsed block, while a label bullet outside it is still vetted.
• A regression test carries a constructed body in this shape and asserts no unrecognized shape, beside one asserting an unvetted label outside the block is still reported.

Found on a Blog develop -> main promotion reviewed at hub 2.0.774.