Trusted verification at AI speed.

AI output is cheap and plentiful, but it only becomes usable once someone has checked it, and people check at human speed. Organisations either slow AI to the pace of their reviewers, or remove the reviewers and carry the risk unseen. When something goes wrong, the investigation is reconstruction: partial logs, emails and recollections, usually held by the party under investigation.

• A sealed receipt for each AI output. It records the evidence the output rested on and the issuer's registered name, and any later change to the output or its evidence is designed to be detectable by anyone holding the record.
• A check anyone can run. The receipt carries the evidence and the rule, so a stranger can rerun the check and reach the same answer, even offline, with a free checker.
• Claims that state what would prove them wrong, so a regulator, judge or journalist can test them without understanding cryptography.
• Corrections that flow forward. Each receipt records what it relied on, so when a source is withdrawn, everything built on it can be flagged.
• An open rulebook, written as a standard anyone can build to and being taken to the IETF.

It vouches for origin and evidence, never for truth. If an AI says it is 25 degrees in Madrid, we check how it knows and whether that evidence holds up; we do not put a thermometer in Madrid.

When SWIFT stopped being neutral, payments split into rival networks. A checking layer one government can read, switch off or restrict will not be trusted by the others. So the design carries the neutrality: nothing to switch off, little to hand over, and rules no single company or country owns.

We are pre-revenue, with no outside users yet. The rulebook is in its twelfth draft, receipts are issued from our own running system and anchored to outside timestamps, and 24 UK patent applications are filed. Our next proof is an outsider checking a receipt on their own, offline.