Skip to content

Five-minute run

This path proves the harness and artifact chain without claiming a remote model or production detector was exercised.

Terminal window
uv run turnkey --help
Terminal window
bash scripts/run_dataset_adapters_smoke.sh

Expected result: the command finishes locally and validates the supported dataset adapter boundary. It is a local smoke check, not live-provider proof.

A complete run is represented by four related files:

  • run.json: resolved run identity and configuration;
  • cases.jsonl: case-level inputs, outputs, labels, and decisions;
  • events.jsonl: ordered runtime observations;
  • metrics.json: aggregate results derived from cases.

The audit path cross-checks these files instead of accepting a standalone summary. Read Artifacts and audit before you interpret a score.