Skip to main content

Cancelling Evaluation Runs

A queued or running evaluation run can be cancelled at any time. Use cancellation when:

  • The run was started against the wrong endpoint, query set, or template.
  • The endpoint is failing on every request and you want to stop wasting calls.
  • You need to free up capacity for a higher-priority run.

In the UI​

  1. Open the evaluation run's detail page.
  2. Click Stop.

The run's status switches to Cancelled. The button is only available while the run is in Queued or Running state.

Using the API​

curl -X POST "https://${RELEVAL_HOST}/api/v1/evaluations/runs/${RUN_ID}/stop" \
  -H "Authorization: Bearer ${TOKEN}"

The response confirms the new status:

{
  "id": "01J...",
  "status": "cancelled",
  "updated": "2026-05-01T10:00:00Z"
}

The endpoint returns 400 Bad Request if the run is in any state other than queued or running (for example, an already-completed run cannot be cancelled).

What happens to in-flight queries​

Cancellation stops any further queries from being dispatched. Queries that have already completed are preserved — their results remain on the run and can be judged.

This means a cancelled run is usually partially populated: some queries have results, the rest do not. You can still:

  • Browse and judge the results that were collected.
  • Use the run's metrics for the queries that completed (with the caveat that they're computed on a smaller sample).
  • Clone the run to start a fresh attempt against the same configuration.

Cancelled runs and AI judging​

You cannot start an AI judging run against a cancelled evaluation run — AI judging requires a Completed run. If you want to AI-judge the partial results, clone the run, let the new run complete, and judge that instead.