What did this agent decide,
and when did it decide it?

A design pattern for accountable autonomous agents

Agents act. When something goes wrong, one question follows — and today it is almost always unanswerable. Logs tell you what the agent says it did. Prove Before Act makes it commit before acting.

T(intent_proof) < T(action)
if T(intent_proof) ≥ T(action): not evidence — just a record