List the advisory pre-merge review runs recorded for this intent, newest first: the PR head SHA the grader read, the intent revision it graded against, the per-item verdict with citations, the human resolutions, and how the result was reported on the pull request.

A run is written by the GitHub App webhook on pull_request opened / synchronize / reopened for PRs that reference the intent (same grader and diff reader as the merge path). status is pending (claimed, grader not finished), completed (a verdict exists) or unavailable (quota, unreadable diff, grader error) with unavailableReason; neither pending nor unavailable is ever a pass. verdictLabel / verdictTone are the one description every surface renders (an unclear-only verdict reads as attention, not as resolved). The judgment identity is (intent, repo, pr, headSha, intentRevision, inputDigest): inputDigest hashes the exact grader input (spec + constitution rules), so a changed verification check or rule on the same head is a new judgment. intentRevisionKind is repo_body (intent.md revision, snapshot in intent_spec_revisions) or content_hash (the graded spec is stored with the run). itemKey is the identity a resolution attaches to: verification:<checkId> for verification items, <category>:<item text> otherwise. canWrite is true only for a signed-in writer; API keys always read canWrite=false because the agent never adjudicates its own grading. reportStatus is not_attempted | posted | permission_missing (the installation has not accepted the App's checks:write permission) | failed. Advisory throughout: the GitHub check run concludes success or neutral, never failure. Plan: .claude/plans/advisory-premerge-review.md.