Uncertainty Quantification for Data-Driven Photopolymer Models: Aleatoric Variation, Epistemic Gaps and Decision Calibration

Data-driven photopolymer models are often reported using mean prediction error, yet engineering decisions depend on whether uncertainty is correctly calibrated at the point where an action is taken. This review separates aleatoric variation, epistemic uncertainty, measurement uncertainty and model-form uncertainty; explains why these terms are not interchangeable; and connects uncertainty estimates to accept, test-more, defer and reject decisions. Recent materials-science studies show the importance of uncertainty-aware model transfer and calibration, but photopolymer models require additional state control because formulation, cure and post-processing create strong domain shifts. The review proposes calibration tests, out-of-distribution gates and non-compensatory decision rules so that uncertainty is used to limit claims rather than decorate predictions. ER-406.