Domain-owned conditions and Alarmist policy; no monitor or automatic remedies.

Raw IDs are {NervesGate.Alarms, server_name, fault}. Availability conditions are debug-level inputs to NervesGate.Alarms.Warning, which requires two seconds of continuous failure. Restore/recovery failures are immediate error-level alarms. Subscribe with Alarmist.subscribe_all/0, then read Alarmist.get_alarms/0.

Each domain registers its faults at startup. Reporting stays asynchronous so a stalled alarm handler cannot delay hardware rollback. Unknown observations retain the last report; unnamed instances do not report. No alarm is proof of readiness.