Proof receipts for AI agents that pay APIs.
ProofGuard gives autonomous buyers durable evidence after a paid endpoint call: what was promised, what was delivered, and whether there is enough evidence for a refund/dispute workflow.
We measured the whole Solana Agent Registry.
Trust on it is almost entirely self-asserted: 1,463 agents enumerated, most identity docs unfetchable in one pass, 224 self-assert x402 support and 2 are independently verifiable from their own registration. Also found the reference SDK computing the wrong hash for evidence records — reported upstream.
A signed evidence bundle, not vague trust language.
Endpoint trust score
Score, decision, and reason codes for a paid endpoint route.
Payment expectation receipt
Records the route, expected amount, method, network, and payment context before fulfillment.
Fulfillment receipt
Records observed delivery, latency, freshness, status, hashes, and contract match signals.
Refund evidence package
Bundles stale, malformed, missing, misleading, or mismatch evidence into a machine-readable package.
ProofGuard labels how strong the evidence is.
No fake certainty. Third-party caller-supplied observations are clearly marked as self-attested. Stronger claims are reserved for SolSigs-instrumented or ProofGuard-probed evidence.
solsigs_instrumentedproofguard_probedself_attested