Troubleshooting
Find the symptom, check the items in the middle column, then follow the steps on the right.
| Symptom | Check first | Next step |
|---|---|---|
| Data doesn't appear in Data Assets | The task's successful upload count, package validation results, and retry history | Data is ingested only after it uploads and passes validation; failures aren't counted. Re-upload first, then investigate. |
| Some items in Data Assets can't be selected | Whether they are in the Awaiting Pipeline or Validation Failed state | Awaiting Pipeline: run a pipeline first; only its outputs can be curated. Validation Failed: re-upload. |
| No format option when creating a task | Whether a device model has been selected | The format is set automatically by the device model; go back to step 1 and select a model |
| No devices available after selecting a model | Whether any devices of that model are registered, and whether they are offline or under maintenance | Click Register Device or Go to Device Management, then create the task |
| A task can't be set to ready | Whether a teleoperation device is bound and the pre-collection checks passed | Bind the teleoperation device ID and run the checks again |
| A task reached its count but is still In Progress | Whether data is still uploading or an upload failed | Wait for uploads to finish, or re-upload the failed data |
| Pipeline parameters are read-only | Whether your role is R&D engineer | Operator parameters require R&D engineer permission; FAEs can start pipelines but can't change parameters |
| The pipeline shows Failed | The failed step, input batch, timestamps, and logs | Retry the failed step; the original run record is kept |
| No Export button after processing | Whether the pipeline status is Completed | Projects can't be exported while processing; refresh the workbench once it completes |
| An export is blocked | The reason shown on the page | Resolve each item as described in Export Blockers |
| Exported counts don't add up | Succeeded (failed) counts and range trimming instructions | Find the failed IDs in the log, fix them, and export again |
| A training job can't start | Export status, pending curation, matching input formats, available compute | Meet the requirements or leave the job queued; don't resubmit |
| A training job stays queued | Resource queue usage | Leave it queued; contact an administrator to adjust quotas |
| The edge deployment button is disabled | Device online, client, inference code, hardware compatibility, network authorization | Run nexcore-edge check and submit once the environment is ready |
| A device shows Offline | Device power, network, and whether the client or agent is running | The device comes back online within about 30 seconds of restoring the network |
| Inference results have no scores | Whether NexBench was turned on for the job | The toggle is per job; no scores are produced when it's off |
| No bad cases on the feedback page | Whether the job had failed nodes and whether they were already reviewed | No empty set is created when nothing failed; handled items are collapsed by default |
| No Download button for a model or dataset | Your role and the platform's download policy | Ask an administrator to confirm your permissions |
| A delivery order partially succeeded | The failed items in the order details | Click Retry to retry only the failed items |
| A shared project isn't visible | Your role and your account's project access | Ask an administrator to check your project access |
If the issue persists, submit a report following Reporting Issues.

