| name | incident-response |
| description | When production is broken — stop the bleeding first, diagnose second, learn third |
| version | 1 |
| trust | built-in |
In an incident the priority order is: restore service → understand cause →
prevent recurrence. Don't invert it by debugging a live outage while users suffer.
- Mitigate first. Reduce impact NOW: roll back the recent deploy, flip the
feature flag off, fail over, scale up, or shed load. The fastest safe path to
"users are OK again" beats the elegant fix.
- Stabilize, then investigate. Once impact is contained, you have room to find
the real cause without the clock running.
- Use your signals. Logs, metrics, traces, recent changes (a deploy/config
change right before the incident is the prime suspect). Form one hypothesis at a
time (see systematic-debugging).
- Communicate. State what's impacted, what you're doing, and the next update
time. Silence is worse than bad news.
- Fix the root cause with a regression test, and only then re-enable / re-deploy.
- Blameless post-mortem. Write up the timeline, root cause, and concrete
action items to prevent recurrence (better alerting, a guard, a test). Focus on
the system and process, not who.
Red flag: forward-fixing a live outage instead of rolling back; declaring it
resolved without confirming recovery in the metrics.