The merge queue admits an approved pull that was green on an older base straight into the batch: the batch gate is the proof, not a per-pull re-check first #3316

Probe 2026-10-20 18:15Z: landings come one every ~7 min (land decisions ship at 17:14, 17:21, 17:28, 17:35, 17:42 x2, 17:57, 18:10) while 5-6 approved pulls wait; 3 of 6 approved are stale. Each landing moves main, which stales every other approved pull; admit() (src/crew/autonomy.ts:1762-1779, #3055) then re-runs that pull's own check on the new base (~5 min) and only a later sweep admits it, so pulls reach the queue one at a time and drain() (src/checks/queue.ts:163) batches one. The batch gate already runs the full tier on base tip + every pull that merges cleanly - the per-pull re-check is the same proof, paid first and serially.

Done: a pull that is approved and was green on its head (any base) and merges cleanly is admitted to the next candidate directly; the candidate's full gate is the proof, and a red candidate bisects as today. The per-pull re-check runs only when the pull's own head changed. Measure before/after on /boards: landings per hour and minutes approved-to-landed (target: a batch of N lands in one gate, ~7 min, not 7N). Test: 5 approved pulls, main moves, all 5 land in one candidate.

Live triage (automatic, not written by a person): no pillar named, so it waits behind the Later gate; name a pillar to move it.