Incident runbook

Revised October 18, 2026. If you are paged and unsure, start at step one.

1. Confirm it is real

Open the status board. If error rate is under 0.5% and only one region is affected, it is a degradation, not an outage. Post in #incidents and keep going.

2. Declare

Anybody can declare. There is no penalty for declaring something that turns out to be small.

$ ./incident declare --sev 2 --summary "checkout 5xx in us-east-1"
  incident INC-4182 opened
  paging: primary on call, Daniel Reed
  channel: #inc-4182

3. Stabilize before you explain

Roll back first, understand later. The last deploy is almost always the answer.

$ ./deploy rollback --service checkout --to previous
  rolled back to build 2026.10.17-3 in 94s

4. Write it down the same day

• What a customer saw, in one sentence.
• The clock: first alert, first human, first fix.
• One change that makes the next one shorter.

No blame is written into the document. We have never fired anyone over an incident, and we are not going to start.

Built with Terminal.css · MIT · Gian Marco Toso