Skip to main content
The Decisions API gives you direct access to fraud decisions — the outcomes produced by AntifraudAI for every transaction and event you screen. Use it to fetch the full detail of any decision by its ID, and to submit manual review overrides when a human fraud analyst approves or rejects a flagged item. Manual overrides feed back into the ML models as ground-truth labels, improving future accuracy over time.

GET /decisions/

Retrieve a fraud decision by its ID. The decision object includes the original risk score, the automated decision, all contributing signals, and — if a manual review has been submitted — the reviewer’s outcome.
string
required
The unique identifier of the decision to retrieve.
cURL

Response Fields

string
Unique identifier for this decision.
string
The type of resource this decision was made for: transaction or event.
string
The ID of the transaction or event this decision belongs to (e.g., txn_01HXZ9... or evt_04JYA2...).
integer
The risk score (0–100) at the time this decision was made.
string
The automated decision: allow, review, or block.
array of strings
Risk signals that were active when this decision was produced.
string | null
The reviewer_id of the analyst who submitted a manual override, or null if no manual review has been performed.
string
ISO 8601 UTC timestamp of when the automated decision was made.
Example response:

POST /decisions//review

Submit a manual review override for a decision. Use this endpoint when a fraud analyst has reviewed a flagged transaction or event and reached a conclusion. The override is recorded on the decision object and used as a training signal to improve model accuracy.
string
required
The decision_id of the decision being reviewed.

Request Parameters

string
required
The result of the manual review. Must be one of:
  • approved — the item was legitimate; the automated decision was a false positive
  • rejected — the item was fraudulent; the automated decision was correct
string
Your internal identifier for the analyst submitting this review. Stored on the decision for auditing and accountability purposes.
string
Free-text notes from the reviewer explaining their reasoning. Useful for audit trails and team collaboration.
The response is the updated decision object, now including the reviewer’s outcome and ID:
Submitting timely and consistent manual reviews significantly improves the accuracy of AntifraudAI’s models for your account. We recommend aiming to review all review decisions within 24 hours.