Alerts get their own queue, and a flapping target stops shouting
Paging no longer waits behind check ingestion, and one unstable service no longer turns into twenty notifications.
Paging runs on its own queue and workers, so a burst of incoming checks cannot delay an alert.
A new per-monitor grace decides whether being down is worth waking someone for. Off by default, capped at 30 minutes.
Flap damping: past four outages in an hour the individual alerts stop, and the one crossing the threshold carries the notice, so nobody wonders why it went quiet.
A read-only command shows which probes voted a monitor down and what it alerted.