amadan
AU box fails the update gate probe on both update and rollback while healthy open
Seen during the 2026-09-07 fleet roll to dcf0cad. barbara-tinkering (ap-southeast-2, i-022288023f5e7496d) failed the update gate's probe of http://127.0.0.1/ with connection refused on the new build, then failed the same probe on rollback, and was left on df7d1d2 in a rollback-failed state. The box itself is healthy: edge on 80/443, tenant routes return 200, so the probe raced edge startup rather than catching a real failure.
The other three boxes (eu-west-1 and us-east-1) passed the same gate on the same build. Suspect the gate's readiness wait is too short for the AU box's slower start, or it probes before the edge has bound 80.
Fix: give the gate a bounded readiness retry (probe until 200 or a timeout in the tens of seconds), and never mark rollback failed when the rolled-back binary is what is running. Then re-roll barbara-tinkering to dcf0cad.