AI agent forensics

If I hand you a hashed record of a single agent's actions, what would you need besides that record before you would trust it?