API Reference / Strategy - Semantic Matcher Eval
Calibrate classifier confidence vs human outcomes
https://api.intelig.ai/api/v1/intelig/strategy/initiatives/semantic-eval/confidence-calibration
Auth: Authorization: ApiKey <key> (an API key sent as
Bearer is rejected) or Authorization: Bearer <jwt>
for a dashboard session. operationId: confidenceCalibration
Answers whether the classifier's confidence number means anything. Buckets past proposals by the confidence they were given and compares each bucket against what humans decided, returning a reliability diagram, the abstention threshold at which accuracy reaches `targetAccuracy`, and a verdict on whether confidence carries signal at all. Needs a supply of confirm/dismiss decisions to learn from — an org with none gets a report saying so rather than a fabricated threshold. Read-only; use the recompute endpoint to persist the result.
Example request
curl -X GET "https://api.intelig.ai/api/v1/intelig/strategy/initiatives/semantic-eval/confidence-calibration" \
-H "Authorization: ApiKey $INTELIG_API_KEY"Query parameters
| Name | Type | Required | Description |
|---|---|---|---|
matchType | string | no | — |
targetAccuracy | number (double) | no | — |
Response 200
application/json · ApiResponseConfidenceCalibrationReport
| Field | Type | Description |
|---|---|---|
success | boolean | — |
message | string | — |
data | ConfidenceCalibrationReport | — |
Status codes
200— OK401— —429— —
Every response is wrapped in { success, message, data }. A 4xx means the request
is wrong — do not retry it unchanged. A 429 carries Retry-After in seconds.