Coverage
- Every public-facing hostname is monitored, including www / apex variants.
- Login, checkout, and other revenue paths have synthetic checks, not just uptime probes.
- Internal admin endpoints are monitored or explicitly excluded with a reason.
Cadence and geography
- Production endpoints checked at least every minute.
- At least two probe regions on independent networks.
- Probe intervals are documented; nobody silently slows them to save cost.
Alerting
- Threshold is two consecutive failures from independent regions — not one.
- Severity ladders escalate to the right people, not the same channel each time.
- Auto-resolve fires when checks recover, not on a fixed timer.
Continuity
An alert without a workflow is a notification, not monitoring. Every alert should resolve into a saved diagnostics session, a runbook link, and a place to record what was found. The dashboard's continuity feed converts each incident into operational memory.