Trials

After Quickstart, use Experiment → Trials to inspect an individual attempt. A Trial has trial_id; an Agent Job groups Trials under job_id. Training uses an independent train_job_id.

Inspect an attempt

Select the workspace and project containing your run, open Trials, then open its ID. Read the status, outputs and available score. For a Trial returned by a task submission, the matching source CLI also supports:

el task status TRIAL_ID
el task result TRIAL_ID

Replace TRIAL_ID with your returned ID. Execution status and evaluation score answer different questions: succeeded does not imply reward=1. A missing score shown as — is unknown, not zero. Check the Job result for aggregate evaluation.

Run again

A new attempt creates a new Trial. Reusing a submission's request_id retries the original request without changing its inputs; use a new request ID when changing parameters. See Run tasks for submission and Core concepts for resource boundaries.