Search runs for a given experiment with an optional SQL-like filter.

Runs are returned in stable id ascending order. Pagination uses the cursor returned in pagination.next_cursor; keep the filter unchanged when requesting subsequent pages.

The filter language supports the unprefixed run columns id, output, example_id, custom run columns, eval.<name>.score, eval.<name>.label, eval.<name>.explanation, eval.<name>.metadata.*, and annotation.<name>.* when those columns are present in the run schema.

Payload Requirements

• filter is optional. Omitting it returns all runs. A present empty or whitespace-only value is invalid.
• limit is optional and defaults to 50; valid values are 1 through 500.
• cursor is optional; omitting it starts at the first page.
• Unknown properties are rejected.

Valid example

{
  "filter": "eval.quality.score < 0.5",
  "limit": 50,
  "cursor": "opaque-cursor"
}

Invalid example (limit is outside the supported range)

{
  "filter": "output = 'approved'",
  "limit": 501
}

<Warning>This endpoint is in alpha, read more here.</Warning>

{
  "filter": "eval.quality.score < 0.5",
  "limit": 50,
  "cursor": "opaque-cursor"
}

{
  "experiment_runs": [
    {
      "annotations": [
        {
          "annotator": {
            "email": "user@example.com"
          }
        }
      ]
    }
  ]
}