Payments standards
ERC-8183 agentic commerce, where Elpis takes the evaluator seat, and x402 HTTP-native payments.
ERC-8183 — Agentic Commerce
A Job holds an escrowed budget and moves Open → Funded → Submitted → Terminal. The client creates and funds it, the provider submits work, and a single evaluator — set at creation — is the only role that may release (complete) or refund (reject). Both carry an optional attestation reason.
Circle’s reference implementation makes the evaluator’s check a model’s self-reported confidence:
const isValid = answer.valid && answer.confidence === "HIGH";A string deciding whether money moves is the anti-pattern the product exists to remove. Elpis takes the evaluator seat and makes the decision pure and deterministic.
The evaluator, wired
| Decision | Meaning | On chain |
|---|---|---|
complete | Commitment met and evidence proves it. | Escrow released to the provider. |
reject | Deliverable mismatched, or the deadline passed. | Refunded to the client. |
hold | Evidence incomplete, or a person has not decided. | No transaction at all. |
{
"jobId": "42",
"policy": { "humanApprovalThreshold": "5000.00" },
"evidence": {
"deliverableHash": "0x…",
"expectedDeliverableHash": "0x…",
"controlsPassed": true,
"humanApproved": null,
"counterpartySigned": false
}
}The endpoint is ops-guarded and enabled by configuration on the deployment; unconfigured, it says so rather than pretending to have evaluated anything.
x402 — HTTP-native payments
x402 makes a request payable: a server answers 402 Payment Required with the terms, the client pays, and the request is retried with proof. It is a natural fit for agents paying per call.
| Concern | Elpis’s position |
|---|---|
| Who pays | The treasury, as one action — ceilings, allowlist and limits still apply. |
| Recoverability | The payment is classified like any other; small and escrowed where possible. |
| Verification | A 402 flow still ends in a chain read-back, not a claim. |
The point of the mapping is the same in both standards: an agent may initiate, but the release condition is a fact, not an opinion.