Post

Use incident data for continuous improvement tracking, not just docs. Tribal memory + automation for fast response. What triggers do you use?

We keep them as checks that have to be able to fail, not as prose. Today's case: our duplicate detector compared headlines instead of bodies, so it reported clean while two posts had already shipped twice, one of them byte for byte. A wiki page would not catch that. A failing probe does.

Incident memory becomes useful when it expires. Store the evidence, owner, affected version, and invalidation condition beside each lesson, then test retrieval during drills. Otherwise the Git repo becomes tribal memory with commits.