Problem
The verifiability program proves each recorded verdict is intact and reproducible. It does not yet prove two adjacent properties an outside auditor would ask about next:
- Verdict finality — that a recorded verdict is the only evaluation, not the best of several. The same selection argument that applies to any eval with variance applies to review verdicts: a re-run that is invisible is indistinguishable from a re-roll.
- Uniformity across author classes — that maintainer-authored and contributor-authored PRs face the same scrutiny, demonstrated with published numbers rather than asserted.
Goal
The reviewer's own behavior becomes auditable: one verdict per PR per head SHA with reason-coded exceptions, re-evaluation counts published, and standing review-parity statistics by author class on the fairness page.
Sub-issues
Acceptance
An outsider can answer "was this PR evaluated once, and do maintainer PRs face the same gate as everyone else's?" from public surfaces alone.
Problem
The verifiability program proves each recorded verdict is intact and reproducible. It does not yet prove two adjacent properties an outside auditor would ask about next:
Goal
The reviewer's own behavior becomes auditable: one verdict per PR per head SHA with reason-coded exceptions, re-evaluation counts published, and standing review-parity statistics by author class on the fairness page.
Sub-issues
Acceptance
An outsider can answer "was this PR evaluated once, and do maintainer PRs face the same gate as everyone else's?" from public surfaces alone.