Loading records…
Sealing is designed to work like this: every hour, a fingerprint of all decisions completed in that hour is published outside HumanAgent, where we don't control it. If any record were altered afterwards, the fingerprint would no longer match what was published, and anyone could check that.
One fingerprint covers hundreds of records, which is why it is hourly rather than per decision. The hourly external anchoring is not live yet: every record already carries its audit hash and can be verified at /verify, and the sealed date will appear once the external batch is running.
when the human made the decision
when the fingerprint was published externally
anyone can recompute the fingerprint and compare
Records created before sealing was introduced carry a single catch-up seal batch date. Those show a gap between their reviewed date and their sealed date. That gap is visible on purpose — hiding it would defeat the point.
Confirms a real person, not an automated system.
Attaches the identity, credential, license, and specialty to the decision.
One API call. Every decision lands here, publicly verifiable.