Skip to main content

Fraud review endpoints

Three endpoints back the human review of fraud check results: a queue of flagged results, the verdict a reviewer records, and the verdict statistics.

Platform endpoints

Like fraud-assessment, these belong to the Cogneris Vision service and are called with a signed-in user's platform session. They do not accept xtkt_live_ API keys. Every call is scoped to the caller's tenant. Responses use the standard service envelope with a JSON body sent as text/plain.

GET /vision/fraud/review-queue​

Lists results at review or suspect that have no verdict yet, newest first.

ParameterInTypeDescription
limitqueryinteger1 to 200. Default and fallback for any other value: 100.
{
  "data": {
    "items": [
      {
        "requestId": "3f0c2a54-6b1e-4c39-9d0e-2a7b6c1d9e10",
        "riskLevel": "suspect",
        "riskScore": 0.78,
        "createdWhen": "2026-10-20T14:03:11Z"
      }
    ]
  },
  "meta": { "httpStatusCode": 200, "messages": [] },
  "hasErrors": false
}

Read the details of each item with fraud-assessment.

POST /vision/fraud/verdict​

Records a reviewer's verdict on one result. Tenant administrator only (tenant.admin role).

{
  "requestId": "3f0c2a54-6b1e-4c39-9d0e-2a7b6c1d9e10",
  "verdict": "false_positive",
  "notes": "Supplier re-issued the invoice; CNPJ typo confirmed with them."
}
FieldTypeRequiredDescription
requestIdUUID✅The analysis ID. A result must exist for it.
verdictstring✅confirmed_fraud, false_positive or inconclusive.
notesstringFree text kept with the verdict.

There is one verdict per result: posting again replaces the previous verdict, notes and reviewer. The response echoes requestId and verdict. A result with a verdict leaves the review queue.

Errors: an unknown verdict value, an empty requestId, or a requestId with no result return hasErrors: true with the reason; a caller without the administrator role gets 403.

GET /vision/fraud/verdict-stats​

Counts verdicts per risk level and verdict, for verdicts recorded between two dates.

ParameterInTypeDescription
fromquerydate (YYYY-MM-DD)First day, inclusive.
toquerydate (YYYY-MM-DD)Last day, inclusive.
{
  "data": {
    "rows": [
      { "riskLevel": "suspect", "verdict": "confirmed_fraud", "count": 4 },
      { "riskLevel": "review", "verdict": "false_positive", "count": 11 }
    ]
  },
  "meta": { "httpStatusCode": 200, "messages": [] },
  "hasErrors": false
}

These counts are your tenant's own measure of how the check performs; Cogneris does not publish an accuracy figure for it yet.