Skip to main content

Run an evaluation

Running an evaluation sends every case through the chosen model and scorer, then records the results, costs, and provenance.

Step 1 — Open the evaluation

In the sidebar, click Evaluations and pick the one you want to run.

Step 2 — Run it

On the evaluation page, click Run. You can run the whole suite or a filtered subset. A screenshot of the evaluation page with the Run button highlighted.
A run is recorded with its evaluation, model, scorer versions, and cost. A cancelled or partial run is not a verified winner — do not treat it as one.

Step 3 — Watch the run

The run page updates as cases finish. Each row shows pass/fail, the score, the model latency, and the cost. Failed rows are flagged so you can find them fast. A screenshot of the run page showing a mix of passing and failing rows.

Step 4 — Read the report

When the run finishes, open the Report tab for a summary: overall pass rate, cost total, and the failing cases grouped by scorer. This report is what a CI gate compares against the baseline.

Troubleshooting

  • All rows failed instantly — the model or scorer is misconfigured. Check the Model Gateway and the scorer definition.
  • budget_exhausted — raise the organization budget or run a smaller subset.
  • Rows stuck in running — the run may be queued behind others. Cancel and retry, or check Remote Runners if you use them.

Up next

Review the results in detail.