The summary.
PREPARE BEFORE YOU NEED IT
Who responds, how to revert, how to restore, who to tell — kept somewhere reachable when your systems are not.
Draft the customer message in advance. Writing it during a crisis produces a worse message, later.
THE FASTEST ROUTE TO A CAUSE
What changed beforehand. Most incidents follow a change.
If it followed a deployment, revert first and investigate with service restored.
SEPARATE THE ROLES
The person fixing cannot also answer questions.
SILENCE IS WORSE THAN BAD NEWS
Say you are aware, what is affected, and when you will update again.
Committing to an update time is more valuable than estimating a fix time, and your status page must not live on the failed server.
MEASURE TIME TO DETECT, RESPOND AND RESOLVE
A long time to detect is a monitoring problem, not a cause problem.
FIX THE SYSTEM, NOT THE PERSON
"Should have been more careful" produces no change and discourages reporting.
Produce a few specific actions with owners and dates, then check they were done.
BUILD FOR FAILURE
Timeouts on every external call, and one non-essential dependency must never take down everything.
