Start a holdout evaluation
POST/cases/:caseId/evals/holdout
Holds out a grouped sample of the labelled records, classifies it once per variant with the held-out records left out of the evidence, and scores each variant against the historical labels. The first variant is compared with the others. The result is evidence for a reviewer, not a gate.
Request
Responses
- 202
- 400
- 404
- 409
Evaluation accepted
Invalid sample, variants or config
Case not found
No dataset, imported records or usable tree