The Postmortem Nobody Read
Three action items, eleven contributors, one still open a year later. By then the same incident had recurred twice. Thorough document, zero prevention.
Three action items. Three named owners. 1 still open a year later, when the same incident hit for the second time. The postmortem itself ran six pages, written the previous March, thorough, distributed to eleven contributors. Two items shipped on time. The third went to an engineer who moved teams, and nobody picked it up. So the same failure came back, the same edge case in the billing system, twice in one year.
The postmortem was a documentation exercise that succeeded at documenting. The prevention part never ran. This is the standard way postmortem culture fails. The writeup is careful, distributed widely, and filed. The action items get tracked for a few weeks, then drift. The team moves to the next incident. The archive grows into proof that the company takes incidents seriously, with no matching proof that it prevents them.
Better postmortems won't help here. The postmortems are fine. What's missing is reading the old ones before touching the system they cover. Before an engineer ships a change to a system that has a postmortem on record, they read it. Five minutes. That read catches the edge case the last incident exposed, before it recurs. The read is the prevention. The document was only ever the record of it.
Most teams skip this because the archive is filed as history, not treated as a live safety check. It sits in a Notion folder nobody opens. The engineer shipping today hasn't read the past incidents on the system they're about to touch. So the next one lands on the same line as the last, same root cause, same customer impact.
The discipline is mechanical and cheap. Every system keeps a postmortem index. Every change against that system asks the engineer to confirm they read it, as a checkbox in the pull request template. Five minutes per change. Over time the company builds institutional memory that outlives the people who wrote it, because the memory lives in the postmortems and the postmortems get read. The teams that run this well are usually the same ones who killed their repeating retro action items and pulled the runbook out of one person's head. They stop paying for the same incident twice.
A postmortem that gets filed and never reopened is a receipt. It proves you paid for the lesson. It doesn't prove you kept it.