Function Accuracy

Trigger and retrieve evaluations for completed transformations.

Evaluations run asynchronously and score each transformation's output against the function's schema for confidence, per-field hallucination detection, and relevance. Evaluations are supported for extract, transform, analyze, and join events.

Lifecycle

1. Trigger — POST /v3/eval queues jobs for a batch of transformation IDs and returns immediately with queued / skipped counts plus per-ID errors.
2. Poll — POST /v3/eval/results (body) or GET /v3/eval/results (query) returns the current state of each requested transformation, partitioned into results (completed), pending (still running), and failed (terminal failures or unknown transformation IDs).

Up to 100 transformation IDs may be submitted per request.

See also

• System overview — evaluating extraction quality