On the Mac, Settings → Who's In The Room has a switch named Test the answer after the AIs agree. Its note reads:
After the AIs agree, the app runs this project's tests one more time on the final code. Only tests that have caught a real mistake here before are used, since a test that has never failed has not shown it can catch anything.
What it does
During a discussion the reviewers run commands — the build, the test suite, a script — and each command's exit code goes on the receipt. Some of those commands fail during the discussion and pass after the lead fixes something. Those are the checks that have caught a real mistake in this project.
When the room agrees and the run changed files, the app replays those proven checks against the final code before it reports the result. If they still pass, the receipt says so, with each command and its time. If one fails, the receipt is graded failed: the answer is shown as wrong, found after everyone agreed, and that outranks whatever the seats concluded. Follow up and the room fixes it in the next turn.
Why only proven checks
A test that has never failed has not shown it can catch anything, so running it proves little and costs time. Limiting the replay to checks that have failed and then passed keeps it short and keeps it honest. A discussion that changed nothing — a question answered — is not replayed at all, since there is nothing the checks could falsify.
Cost
The replay runs commands on the Mac and makes no model calls. It is on by default; a fresh workspace has no proven checks, so it costs nothing until the room has caught something.