payout_screening runs the full screening panel over a payment's amount, corridor and parties, and that can take up to about 20 seconds. The call returns 202 instead of a verdict: the evaluation comes back with status: "pending" and no result fields. It creates no payout and moves no money.
Describe the payment
The payout block is required for this use case, with amount, source_country and destination_country. amount is a value as a decimal string, such as "250.00", and a currency of 3 to 5 uppercase letters: a fiat code or a crypto asset like USDC. rail is optional and is one of bank, card, crypto, wallet, internal or other. reference_id is your own identifier for the payment.
Send the entity too, with something searchable: a name or vault_token_id. You can add an instrument when you know where the money goes. Without searchable entity or instrument data the call is a 422.
The body carries the evaluation id and status: "pending".
curl -X POST \ https://api.antonpayments.dev/v1/intelligence/evaluations \ -H "Authorization: DPoP $ANTON_ACCESS_TOKEN" \ -H "DPoP: $DPOP_PROOF" \ -H "Idempotency-Key: intel-payout-001" \ -H "Content-Type: application/json" \ -d '{ "use_case": "payout_screening", "client_reference_id": "payout-001", "entity": { "type": "person", "reference_id": "payee-john-doe", "name": "John Doe", "country": "US" }, "payout": { "reference_id": "payout-001", "amount": { "value": "250.00", "currency": "USD" }, "source_country": "US", "destination_country": "CA", "rail": "bank" } }'
Take the 202
The response is 202 Accepted with a Location header holding the URL of the evaluation. The body has the envelope only.
Store the id against your payment. decision is absent until the evaluation completes.
Wait for the event
Subscribe to intelligence.evaluation.completed and intelligence.evaluation.failed. The event's data.object is the evaluation itself, so the finished one has the same fields as any other: decision, score, tier, reasons and required_actions. Webhook events lists the event types and signing covers verifying them.
failed can end the path instead. It carries failure_reason: "evaluation_failed" and no verdict.
If you would rather poll, GET /v1/intelligence/evaluations/{id} until status is completed or failed. It can read processing on the way. Polling reads the same evaluation the event carries.
When it fails
A failed evaluation is not an approval. Hold the payment and run the evaluation again under a new Idempotency-Key. If coverage on a completed evaluation is partial or error, treat the verdict as weaker than complete.
payout is missing, or lacks amount, source_country or destination_country, or there is nothing to search on.