Braintrust Evaluation & Tracing Playbook
Gate every prompt and model change on a Braintrust experiment built from production failures. Instrument the feature and mask secrets and personal data before upload.
IncludesInstrumentation, masking, dataset, and scorer checklist, plus the CI gate to agree
- Before workInstrument and mask
- During workTrace, review, and curate
- Before mergeRun the experiment gate