Reliability culture

Incident reviews without the theater

A postmortem is not a courtroom. It is a debugging session for the system that made the incident possible.

~6 minLeadership
scroll
01 · the reflex

The first story is usually too simple

Right after an incident, everyone wants the clean sentence: deploy broke checkout. Human clicked wrong button. Alert was missed. Those may be true, but they are rarely enough. The useful question is: why did that action make sense at the time?

incident · rough timeline

A timeline is evidence, not a blame map.

team has a map

Leadership bugs are system bugs too. Pull context out and the team graph starts wobbling.

02 · the room

Blameless does not mean consequence-free

Blameless means you investigate without making people hide details to survive the meeting. It does not mean shrugging. You can be kind to people and ruthless about weak signals, risky defaults, confusing tools, and missing guardrails.

Replace "who caused this?" with "what made this easy to do and hard to catch?"review prompt
03 · the output

Actions should change the system

Training everyone to be more careful is usually a weak action item. Stronger actions alter the path: add an automated check, change the deploy order, page the right owner, make rollback one command, or remove the confusing option.

postmortem-actions.md

Good action items are specific enough to close.

team clarity switchboard
muddy
clear0
muddy: people can move, but they are guessing while they do it

Flip the switches and watch the story turn into a tiny operating model.

04 · the memory

A review should be useful six months later

Future engineers will not remember the meeting. They will find the doc during a similar outage. Write for them: impact, timeline, contributing factors, detection gaps, what changed, and how to tell if it worked.

source trail