# Retrospective packet

Use DOC, data.kind=retro. Link task, session/run, review, checks and FDK.

| Observation | Evidence | Cause/confidence | Proposed change | Owner | Next-run measure |
| --- | --- | --- | --- | --- | --- |
| Repeated work after reconnect | Duplicate operation trace | Missing identity, high | Retain/replay command key | Task implementer | One local revision |
| Review lacked native proof | Review report gap | Acceptance too vague, medium | Add real tool-use observation | Spec owner | Evidence linked in report |

Keep: practices supported by successful evidence. Change: small owned actions, not broad slogans. Unknown: missing events or untested causes. Follow-up: TSK/RDT/DOC/FDK/MEM as appropriate, with proposal/application status explicit.

Do not infer a personal failing or universal rule from one bad run. On the next relevant task check whether the accepted change actually helped. Retain the superseded lesson if it did not.