Alert logic & recovery

Alerts fire on transitions, not states: one notice when a monitor goes down, one when it recovers, never a flood while it stays down.

Debouncing

• Down requires two consecutive failed checks, so a single dropped packet stays invisible.
• Degraded requires two consecutive slow checks (responses over the degraded threshold), so a one-off slow sample never flips the status.

Recovery notices are incident-based

Each channel only receives a recovery notice for a problem that same channel announced. If slow-response alerts are off and a monitor briefly degrades, nothing is announced, so nothing gets recovered: no phantom "recovered" emails out of nowhere.

• "Monitor goes down" pairs with "Monitor recovers".
• "Slow response" pairs with "Slow response recovered", each toggled independently.
• An outage that comes back through a brief slow phase still resolves as an outage recovery.