Verify a record
Re-derive every hash in a decision export and find out whether anything has been altered or removed since it was written. No account, no upload, no relationship with us required.
Nothing to hand? Try it on a sample
These are generated in your browser using the published algorithm — the same one the platform uses. The tampered version has a single denial rewritten as an approval, which is the fraud this is built to catch.
Paste an export
The whole JSON response from GET /api/v1/agentaudit/decisions, or a file you have been sent.
Verifying makes no network request. The page itself loads over the network like any other, but the check afterwards runs entirely on your machine — what you paste is never sent to us or to anyone else. Open your browser’s network tab and watch: pressing Verify produces nothing.
What a clean result does and does not mean
- It means
- Every decision in the file hashes to its recorded value, and each one points at the decision before it. Nothing in the supplied range has been edited, and nothing has been removed from within it.
- It does not mean the agent behaved well
- A record can be perfectly intact and describe a system making poor decisions consistently. This checks the integrity of the account, not the quality of the conduct it describes.
- It does not cover decisions outside the file
- If the export begins at sequence 400, everything before it is unexamined. The verifier always names the range it checked for that reason, and warns you when the file itself says more decisions exist.
- It does not prove the agent reported everything
- The chain proves nothing was removed after it arrived. It cannot prove a decision was reported in the first place — an agent that never called the API leaves no trace to break. That is why the SDK treats an undeliverable decision as a loud error rather than a dropped metric.
The algorithm is written out in full in the documentation, so you can implement this check yourself rather than using ours.