Observability Platform · View 21 of 25 · Operations
Decisions
- Completeness is a first-class row alongside health and latency. Most platforms measure whether components are up; this one measures how much of the estate's telemetry actually arrived.
- The outside row is deliberately sparse: an external prober should test the round trip — write a probe, read it back, and alert on our silence — not the internals, which it cannot reach when it matters.
The signal that matters most
- The dead-man switch. Every other cell alerts on something going up; this one alerts on something that should be there and is not, from a place that cannot fail with us.
- Hosts-not-reporting and missed-evaluations are the two cells most likely to be quietly disabled for noise, and the two whose absence would make the platform's own failure invisible.
Numbers
- Self-telemetry path ≥ 99.99% monthly, 90-day retention. Missed rule evaluations ≤ 0.01% per month, each one recorded rather than counted.
