Lost in the Folds: When Cross-Validation Is Not a Deep Ensemble for Uncertainty Estimation

K-fold cross-validation ensembles and true deep ensembles are systematically conflated in medical image segmentation uncertainty literature, with CV disagreement mixing data-exposure variance into epistemic uncertainty estimates. An audit of recent studies finds terminology-implementation mismatches are common, undermining uncertainty calibration claims. Practitioners using CV ensembles for uncertainty quantification in safety-critical settings should reassess their interpretation methodology.