Evaluate an external action in one call
Submit an Action Proposal with the evidence you already have, and receive a reproducible decision under your own policy. The Target system issues its own receipt separately. These are conceptual examples; the API is in private beta.
Submit an Action Proposal
First Instinct evaluates it against receiver-owned policy and returns one of four states.
POST /v1/action-proposals
{
"action_type": "create_refund_request",
"target": "merchant_refund_workflow",
"amount": 42,
"currency": "GBP",
"idempotency_key": "...",
"evidence": [...]
}{
"decision": "require-human-approval",
"reason_codes": [
"SPONSOR_UNVERIFIED",
"MANDATE_SCOPE_UNVERIFIED"
],
"evaluation_record": "..."
}The Target issues its own receipt
First Instinct verifies Target Receipts; it does not issue them. The Target signs what it actually did — accepted_for_processing, completed, failed or reversed.
POST /v1/target-receipts
{
"action_id": "...",
"outcome": "accepted_for_processing",
"issuer": "customer_ticketing_system"
}Shadow Mode
Run evaluation beside a live workflow without blocking, approving or executing anything. Shadow Mode records how the same requests would be classified under the proposed policy, so you can measure exposure before enforcing.
Webhooks
Subscribe to evaluation and Target Receipt events to keep your own systems in sync. Webhook payloads carry the source-attributed evaluation record identifier, never customer credentials.
The API is not yet generally available and is not production-ready. Request access and we’ll share credentials and the full reference.