Every job leaves a receipt. Every hardware claim gets signed.
A compute merchant on the Visa CLI creates one hive-vcr-1 receipt for every job. The hardware type, the region, how isolated it was, and how long it ran all get signed together. That turns "trust us" into something you can actually check, not just a line on a bill.
The operator signs these fields before the call ever comes back to you.
The operator signs every field below with ML-DSA-65 (the government's post-quantum signature standard), and Hive signs on top of that. If a field is missing, that's not a default setting. It means the receipt fails verification. The merchant can't send you a result without the receipt, and the receipt can't exist without these fields filled in.
Compute · GPU · attestation fields
Three places that need a receipt right now.
Proving where a training run happened
When a regulator asks where a training run took place, the receipt log has the answer. The region and hardware get signed the moment the job starts, so nobody has to guess from a bill.
Confidential-VM workloads
When isolation_class is set to confidential-vm and signed, a regulated caller gets real proof, not a promise, that the workload ran inside a locked-down hardware enclave.
Cross-tenant billing audits
Billing checks itself against the duration_seconds field. If there's a dispute, you run the verifier and get your answer. No support ticket needed.
One endpoint, one envelope, and Hive never sits in your inference path.
A compute · gpu operator adopts hive-vcr-1 by attaching the receipt envelope to the response and publishing a verification key. Hive's signature rides alongside, it never blocks the critical path. If Hive ever went away, the operator's receipts would still check out against the operator's own key.