| title | Evaluations |
|---|
Evaluations are the offline testing layer for your agents and LLM pipelines. Define inputs, run them through the version you want to test, score the outputs, and compare across runs to see if anything regressed.
What evaluations are and when to use them. Write and run your first evaluation in five minutes. Datapoints, executors, evaluators, and how they map to traces. Groups, the progression chart, and side-by-side comparison. Back your evaluation with a Laminar dataset instead of hardcoded data. Lower-level control for pipelines where `evaluate()` is too opinionated. Point the SDK at a self-hosted Laminar instance.Next: turn failing traces into reusable evaluation data with Datasets, or catch regressions in production with Signals.