Write a blameless postmortem for agent faults

A write up that ends in a name teaches everyone to keep the next fault quiet. The useful part of a postmortem is the set of changes it produces, not the person it settles on.

Record from the logs

Memory rewrites a fault into the shape of the story we expect. The logs hold what actually happened, in order, and are the honest source for the timeline.

Ask what made it possible

The question is not who caused the fault but what about the system allowed it to happen. That second question can be answered and acted on, while blame cannot.

def follow_ups(report):
    return [c for c in report.changes if not c.done]

Track the changes

A postmortem that lists changes and is then filed away changes nothing. Track each change until it is done, so the write up ends in a fix rather than a document.