- Confirm the run received the expected trigger input.
- Check that the correct job, agent, credential, and model were used.
- Inspect tool arguments and returned data.
- Separate missing evidence from poor synthesis.
- Update the narrowest layer: connection, job prompt, or agent instructions.
Guides
Review run history
Diagnose output quality, tool use, errors, and human-input pauses.
Open Runs and select an execution. The timeline shows the run trigger, messages, model and tool activity, any human-input request, terminal status, and error details.
When a result is weak, review in this order: