Service

Crucible

FULL PLAN ONLY

Deployment gate: scores a model's predictions against knowledge-graph ground truth you've recorded and runs a labeled sample through bias-detector's fairness math. Requires you to supply the ground-truth entities and labeled records -- there is no ambient benchmark it invents on your behalf.

No per-call charge -- unlimited under your subscription

Worked example

Evaluate model performance

curl -X POST https://cainstudio.online/crucible/evaluate \
  -H "X-API-Key: $CAIN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model_id": "...", "predictions": [], "ground_truth_field": "expected_value", "truthfulness_threshold": 0.95, "bias_check": "..."}'

Try it live in your dashboard → See pricing →