A service defines a computational tool. A job is one submitted execution of that tool with particular inputs and settings. Running the same service twice creates two job records; those runs may differ because of settings, sampling, or model behavior.
Follow job states
| State | Meaning | What to do |
|---|---|---|
| Pending | The job has not started executing | Check the workspace rather than submitting a duplicate |
| Running | Execution has started and has not finished | Allow processing to continue and monitor status |
| Completed | Execution finished without a recorded error | Open the job and inspect its outputs |
| Failed | A job error was recorded | Read the error and review inputs and settings |
| Cancelled | Cancellation was recorded | Check the final record before deciding whether to rerun |
| Deleted | The job is marked deleted | Do not rely on its outputs remaining available |
The workspace may display updates after a refresh or status reload. A long runtime alone does not show that a run has failed.
Read results in context
Open a job from Past Jobs. Depending on the service, results may include molecular structures, candidate tables, confidence plots, trajectories, reports, or other downloadable files.
Read the relevant tool guide before comparing scores. A docking energy, a predicted affinity, an interface-confidence score, and a sequence-model likelihood are different quantities. Keep each model’s native output names, units, and assumptions.
Completion means the computation finished. It does not mean that the biological hypothesis is correct or that every output is suitable for another tool.
Organize your workspace
Use a descriptive job note to record the target and experimental purpose. The workspace supports editing notes in the jobs table and organizing jobs into projects.
Select jobs to enable Move to Project. Choose a project compatible with the selected jobs’ ownership and team visibility. My Jobs, Team Jobs when available, and Projects provide different views; a project is an organizational container, not another computation.
Duplicating a job submits a new run after confirmation. It is not a free copy of already generated results. Review the inputs, current service behavior, and credit implications before doing so.
Preserve a reproducible record
Keep the job ID, tool and version, exact inputs, settings, relevant output files, and any post-processing decisions. Preserve random seeds and sample counts when the method exposes them.
Download files you need for long-term research records. The workspace warns that failed jobs and their associated files are automatically deleted after three days. Do not use a failed job as a durable file store.
Investigate problems
For an input error, check formats and molecular context. For structural problems, repair and validate the relevant system rather than applying arbitrary cleanup.
If an error persists, contact support with the job ID and error message. Confirm whether a run already exists before retrying a submission whose response was interrupted.