Inference Sources and Prediction Sets

Inference Source

An Inference Source is either a local Model or an API Profile. Prediction Generation invokes it against a Dataset to produce typed predictions.

• A local Model owns copied model bytes, settings, and verification in the Project.
• An API Profile owns a reusable request and response mapping; credentials stay in private application data.

Prediction Generation

Prediction Generation prepares selected inputs and produces clean-input and selected robustness-transform predictions. It does not choose Performance metrics, calculate module scores, or apply Evaluation thresholds.

Prediction Set

A Prediction Set is the sealed result of Prediction Generation. It records the exact Inference Source, Dataset, overrides, coverage, provenance, and Prediction Evidence used later by an Evaluation.

Inference Source + Dataset + generation coverage
                       |
                       v
              Prediction Generation
                       |
                       v
                Prediction Set

The separation lets multiple Evaluations score the same fixed predictions with different compatible metrics or thresholds. A retained Prediction Set never changes. Missing transform or explanation coverage requires a new Prediction Set, not an edit to the old one.