The best postmortems are short, blameless, and linkable. The worst are long, defensive, and filed away unread.
Lead with the timeline
A tight timeline of what happened when — with the queries that proved each step — is worth more than three pages of prose.
Action items with owners and dates
An action item without an owner is a wish. Track them like any other work.
Blameless means system-focused
Blame stops information. The moment an engineer expects to be named as the cause, the useful details stop arriving. Describe what the system allowed to happen — the missing guardrail, the alert that never fired, the runbook that was three releases out of date — and the fixes write themselves.
One page, and always linkable
If a postmortem is longer than a page, nobody re-reads it. Put the timeline, the contributing factors, and the action items above the fold, and give every incident a stable URL. Six months later, "remember the checkout outage?" should resolve to a link, not a Slack archaeology dig.
Close the loop in public
When the action items ship, note it on the same document. A postmortem that shows its follow-through is the one people trust enough to write honestly next time.