How it works
How verification works
Verification answers five separate questions, and each one gets its own status. A report can pass four and fail one, and the page says so plainly.
1. Source data
The disclosure package holds a snapshot of the wallet history. The verifier hashes that snapshot and compares it with the data hash published onchain. If they match, the data is the data that was published.
2. Calculations
The verifier reruns the claims, the credit score and the wallet link proofs from the snapshot, using the published policy. If the rerun gives the same answers as the report, the calculations reproduce.
3. AI attestation
The AI analysis ran inside a trusted enclave, and its attestation is checked on its own. A nonce match alone does not verify it. The panel shows which of the nonce, quote, signature and measurements were actually checked.
4. Report integrity
The report, evidence and policy hashes are compared with the onchain record, the disclosure is checked against the report commitment, and the registry is asked whether the claim is still valid. A report that was revoked or expired fails here even if everything else passes.
5. Privacy
The public report carries thresholds and a score band only, never exact values, and the wallet commitment uses a fresh salt per report. The panel says what the public report discloses and what it keeps private.
Sampling and limits
Balance sampling is point in time only: it reads the balance at past block heights, so it proves a balance was held at those readings, not continuously between them. The report says so plainly, with its sample count and interval.
What the verifier can and cannot learn
You can learn: whether the disclosed history supports the claims, whether the score reproduces, whether the report is the one that was published, whether it is still valid, and what the public report discloses.
You cannot learn: the real world identity behind the wallet, whether a counterparty is independent of the prover, or anything about data that never touched the chain. The score measures observable onchain activity only. It is not a guarantee of repayment.
What you see that the public does not: the wallet address and the exact values, because the prover sent you the disclosure package. Do not republish them. And one limit stays: the registry record names the publishing wallet as owner, so claims from one wallet can still be linked through the registry even though the report commitments cannot be matched.
The credit score, and its limits
The score is deterministic and portable: sustained balance, time weighted balance, activity, counterparty diversity, volume, failed transactions and staked balance, scored from the same public rules, so any verifier recomputes it from the same snapshots. A tampered score fails verification.
The score is also honest about what it cannot do. It measures observable onchain activity only. It does not prove identity, beneficial ownership, or creditworthiness. Counterparties are unknown by default: independence is never assumed, and repeated transfers, circular flows and concentrated counterparties are flagged as risk, not hidden.
The underwriting verdict from NEAR AI Cloud is advisory only. A thin score can never support a lend, by a deterministic guard, and any suggested limit is capped at the sustained balance.
See it in action
Run the live demo and watch a real claim verify, or prove your own wallet.