Skip to main content
An experiment is created with PENDING status, moves to IN_PROGRESS when the runner starts processing dataset items, and ends in COMPLETED, FAILED, or CANCELLED. A COMPLETED experiment does not imply that every item succeeded: individual items can still be ExperimentItemStatus.FAILED. Item failures do not fail the experiment. Values:
  • PENDING - The experiment was created on the server; the run has not started yet.
  • IN_PROGRESS - The runner is executing the task function and evaluators over the dataset items.
  • COMPLETED - The run finished and results were published, even if some individual items failed.
  • FAILED - The run aborted with an unrecoverable error, such as a task or scoring function whose signature does not match the dataset item fields.
  • CANCELLED - The run was interrupted by the user or by system termination; results already published are kept.