Use this workflow when another tool or team has already generated predictions. Preserve the producer, samples, and conditions while checking the outputs against compatible Ground Truth.
Add the relevant saved records to your chat, then adapt this request to your work.
Inspect these external predictions and the selected Dataset. Identify the task, sample alignment, classes, conditions, and missing coverage before proposing an import. Use the saved Test plan and explain the scoring basis for my review.
Check alignment before import
Identify how the outputs map to retained Dataset samples. Inspect task, classes, source versions, and condition coverage. Resolve ambiguous alignment before treating the outputs as test evidence.
Import through the Test plan
Use the New menu on Test plans to import predictions with a compatible saved plan. Review the settings reference that the import will retain.
Score against the intended labels
Select compatible Ground Truth and the required metrics. Inspect the result and supporting examples. Re-scoring retains earlier results rather than replacing the imported predictions.
Explain provenance in the handoff
Retain where the outputs came from, which samples and conditions they cover, and what the scoring establishes.
