Decision procedure for agents and their operators that answers the three questions every auditor asks after an AI agent

Decision procedure for agents and their operators that answers the three questions every auditor asks after an AI agent spends money: who authorized it, what was the ceiling, and who carries the loss when the purchase is wrong. Covers cryptographic proof that an agent, not a human, executed a payment, signing intent with a held Ed25519 key before settlement so future audits can distinguish autonomous execution from manual action. Includes spend-authorization record schema, approval chain