Generate or import predictions
Keep model outputs tied to the samples and conditions that produced them.
A Prediction Set is an immutable record of obtained or imported outputs. It keeps its source bindings so scoring can be repeated without silently changing the predictions.
Generate outputs in a Run
Choose a compatible Dataset and local Model or verified API profile in Inputs. Review the settings and execution package, then start the work. The Predictions tab lets you inspect the outputs associated with that Run.
Import external outputs
Use the Test plans page’s New menu to import predictions. Select a saved Test plan compatible with the Dataset task and review the source and sample alignment. The import retains the plan’s exact settings reference.
Reuse predictions for scoring
The same Prediction Set can retain multiple scoring results. Compatible corrected Ground Truth can be used without running the model again. Keep the selected labels, metric, and criteria explicit in each result.
Preserve the conditions
An interpretation of robustness depends on the conditions that were actually generated or imported. Inspect their coverage before drawing a comparison or attaching the result to an Assessment.