Evidence of what an AI agent was allowed to do, and did

What it is for

A dispute evidence bundle holds the records of what an AI agent was allowed to do and what it did: the customer's signed instruction, the gate's signed record of the agent's decision, and a witness receipt for a checkpoint of that record's log. Both worked examples here are constructed test cases: a script builds the signed log with a test key, and only the witness receipts are live.

An open verifier checks a bundle and says what it shows and what it does not, so a merchant, a chargeback platform or a card issuer can check the same files without trusting whoever assembled them.

One limit applies to every example here:

Headless Oracle operates both the gate whose key signed this log and the witness, so the log and the witness receipt are statements by one party.

Start here

• Worked example: a constructed test case of an AI agent ordering a physical item
• Worked example: a constructed test case of an AI agent booking a hotel
• Integration: the order fields, the commands and the exit codes
• The data-point sheet: every member of the export, and the four examples' values
• The examples' operator key set (JWKS)
• The verifier's source, as one zip
• The chirindo repository (commit 6285832c294493001b12d4467259c04a2c4890fb)