braintrust-build-eval-dataset
Installation
SKILL.md
Build or audit an eval dataset
Contract: references/interaction-contract.md. Calibration, templates, provenance: references/dataset-construction.md.
Trigger
- Building eval cases, sampling production traces, or auditing an inherited set.
- Questions about representativeness, labels, splits, contamination, or refresh.
- Open-ended tasks where one gold string is being forced onto many valid answers.
- A set that no longer discriminates: everything passes, or everything fails.