What this is.

An open specification for fixing an evaluation criterion before the run, and the work around it.

An evaluation result travels. The rule that turned it into a verdict usually does not. We publish PRML, an open format that binds that rule to a cryptographic digest before the run, so a reader who was not in the room can tell the bar was fixed first. Four reference implementations produce byte-identical output against a published conformance suite, and the public registry issues receipts carrying independent RFC 3161 timestamps and Rekor entries.

We also measure. Our most recent study read all 29 published case studies from a government AI assurance programme that asks participants in writing to publish their interpretation thresholds, and found the decision rule recoverable in three of them. The per-document ledger is published so the classification can be disputed.

The company

Falsify OÜ is a founder-led company incorporated in Estonia in August 2026. Cüneyt Öztürk is the founder and the editor of the specification, with remote contributors to the open-source implementations.

What we are not

We are not a notified body, not an accredited auditor, and not a certification scheme. PRML records that a bar was set; it does not say the bar was right, and it does not make anyone compliant with anything. The specification's limitations section says the same thing at more length, and we would rather you read it than not.