Nuclear models no longer extrapolate with a fixed error: uncertainty now grows with distance, giving drip-line calls a statistical footing
For nuclear regions experiments cannot reach, models lose predictive power the further they extrapolate, yet model combinations usually treat theoretical error as constant. A heteroscedastic Bayesian model combination treats the combined theoretical uncertainty as a dynamic quantity that varies with position, applied to an ensemble of realistic energy density functionals with focus on the Z = 46–52 isotopic chains. Validated against experimental data and synthetic data built to probe deep extrapolation, it yields better calibration metrics and statistically principled assessments of the particle drip lines.