Reliability Engineering

Make failure behavior observable and recoverable without adding operational machinery by habit.

Workflow

1. State the reliability claim or operational question: what user-visible behavior must survive or become diagnosable?
2. Map dependencies and failure modes: timeout, outage, stale state, overload, partial failure, restart, or data loss where relevant.
3. Choose the minimum useful signal: logs, metrics, traces, health/state probes, or an alert tied to an actionable condition.
4. Define deadlines, retry classification, degradation, isolation/backpressure, and reconciliation only where failure semantics require them.
5. For recovery, identify what must survive, what can be rebuilt, source of truth, restore/replay procedure, and proof.
6. Verify at the user/system boundary; alert clearance alone is not recovery evidence.
7. Convert meaningful incidents or near misses into regression/operational learning instead of adding permanent noisy instrumentation.

Stop

Stop when the reliability claim has enough evidence and the relevant failure/recovery path is operationally understandable.

Reference Guide