Signal over coverage
Coverage is easy to measure and easy to overshoot. Signal is what actually matters: a small set of monitors that fire only when human attention is required. Every new monitor should pass a single test — would the team act on this if it fired today?
Named ownership
Every alert and every monitor needs a named owner. "The team" is not an owner; rotations are fine as long as the current rotation member is named. Anonymous alerts get triaged last and resolved slowest.
Alert hygiene
- Threshold tuning. Two consecutive failures from independent probes, not one.
- Severity ladders. Page only what's actionable in the next 15 minutes; route everything else to a queue.
- Auto-resolution. When checks recover, alerts close — no manual ack rituals for trivial flaps.
- Quarterly tune-up. Every recurring alert is reviewed: retune, silence, or remove.
Continuity-first dashboards
The dashboard's first surface should be what's pending and what's in progress — not a wall of green tiles. Workflow continuity converts incidental visits into recurring operational sessions and is the most honest retention signal a platform can give you, because it only appears when there's real work to do.
Maintenance discipline
Monitoring is a maintenance practice, not a setup project. A weekly fifteen-minute review and a monthly audit are enough to keep a stack honest. Without that cadence, every stack drifts in the same direction: more monitors, less signal, more silence around alerts everyone has learned to ignore.