nirs4all.pipeline.dagml.result module

Project native DAG-ML score evidence into RunResult.

Each public row represents an actually scored native partition: train and validation for individual folds, their scored cross-fold aggregates, and train/test for the refitted model. Validation scores never stand in for training measurements.

Refit rows retain their originating CV score for model selection, with explicit provenance in result_metadata.dagml_projection.score_provenance. That score is selection evidence from cross-validation, not a validation measurement of the refitted model. Native reports and predictions remain authoritative.