THE PART YOU DON'T HAVE TO TAKE OUR WORD FOR

Check us. That's the product.

Plenty of tools write a tamper-evident log. That log is still authored by the system being questioned — it proves a record wasn't edited, not that anyone independent agreed with it. What we sell is a verdict from a party that isn't you and can't be paid to say yes. Any verifier can claim that. Below is a real one, with everything you need to prove it wrong.

A real receipt, from a real paid check

Base mainnet, 2026-08-07T10:28:41Z. A real $0.05 USDC fee was paid on-chain before this was evaluated. Nothing here is a mock.

request_id   0x9d74b59182c5f8477641400668a67a1e94e65f20d49176fe41f4d3113d63c069
verdict      PASS
receipt_hash 0x8260a51fe4ef3887c1922e644bcc100e926e3e3afac13bd1f68948b98204ea00
evaluator    0x137f014bca4657a71c286a7755f589f153446f346d361302c8463d6f22ae677a

1 · Read it off the chain, without us

The verdict is anchored in a contract whose source is verified on Basescan. You can read it without touching our servers — if soulscore.xyz vanished tonight, this receipt still resolves.

cast call 0x2C125C03577296a3e552e5439D356F0294349ce9 \
  "getReceipt(bytes32)((bytes32,bytes32,bytes32,string,uint256))" \
  0x9d74b59182c5f8477641400668a67a1e94e65f20d49176fe41f4d3113d63c069 \
  --rpc-url https://mainnet.base.org

Contract source on Basescan →

2 · Re-derive the verdict yourself

The receipt pins the evaluator's own code hash, not just the inputs — so re-running proves the same check ran, not just that the numbers line up. Feed the receipt and the original claim back and recompute it from scratch. This endpoint is free and unauthenticated; you do not have to be the person who paid.

curl -sX POST https://soulscore.xyz/api/reproduce \
  -H "Content-Type: application/json" \
  -d @claim.json | jq .matches
# → true

A false here is a fraud proof. Not a complaint, not a support ticket — a mathematical demonstration that we published a verdict our own evaluator doesn't produce. If you ever get one, post it publicly. That's the deal.

Why we can't quietly cheat

WE'RE PAID THE SAME EITHER WAY

$0.05 whether the answer is PASS or REFUSE. There is no version of this where saying yes earns more than saying no — the incentive to rubber-stamp doesn't exist rather than being promised away.

WE CAN'T CHECK OUR OWN ACTIONS

The service refuses any claim paid for by its own operator's address. Enforced twice, independently — once in the smart contract, once in the HTTP layer — so bypassing one doesn't bypass both.

THE CODE IS PINNED IN THE RECEIPT

Every receipt carries a hash of the evaluator that produced it. If we quietly changed the rules, old receipts would stop re-deriving and the mismatch would be public.

NO MODEL, NO JUDGEMENT CALL

Seven deterministic checks. No LLM anywhere in the verdict path — the same inputs produce the same answer on your machine as on ours, which is the only reason re-derivation means anything.

What this does not prove

Stated plainly, because a verifier that oversells is worth less than one that doesn't exist. A PASS means the action matched the claim committed beforehand — conformance, not soundness. It is not a safety check, not a fraud check, and not a judgement that the plan was wise. A dishonest plan, executed exactly as declared, passes. We also have no custody and can't block anything: the verdict is evidence, not permission. And the honest limit on independence — we can't detect a colluding third party sharing infrastructure with a caller. That's what the deferred challenge-bounty mechanism is for, and it isn't built yet.

Try it against your own claim

clawhub install decision-gate            # free: commit the claim
clawhub install decision-gate-verifier   # $0.05/check: have someone else confirm it
HOW IT WORKS → API DOCS