The problem
A verdict means the most when it is tied to the change it judged.
A verdict earns trust when it is anchored to the exact change it judged. Tie the decision to the commit and the diff, and the proof stays meaningful for as long as the commit exists — the same evidence, replayable months later.
That anchoring is what lets a reviewer, an auditor, or a future teammate reconstruct exactly what was checked and what was decided, from the record alone.
How it works
Bound to the commit, replayable for as long as the commit exists.
Every verdict is addressed by the commit SHA and a hash of the diff it judged. When the change moves, the hash moves with it, so a verdict always corresponds to the code in front of you; amend the commit and the gate runs again on the new state.
The record is append-only. Superseding a decision means a new commit, which keeps the history a durable forensic trail. The same verdict kernel writes this evidence on the workstation and in the hosted check, the machinery is identical, wherever it runs.
- Every verdict is keyed by the commit SHA and the diff hash
- Change the diff and the gate re-runs, so the proof always matches the code
- Evidence is append-only: a new decision means a new commit, and the history stays a clean trail
- The same kernel produces the verdict on your workstation and in the hosted check, so the two agree
Proof that travels with the commit.
Pick any merged change and read its evidence — who reviewed it, against which policy, on which commit.