Dr. Jeongwon Seo’s latest paper, "Validation performance assessment through quantitative score focused on benchmark selection for nuclear criticality safety," has been published in Nuclear Science and Technology Open Research. While traditional validation relies on heuristic similarity rules, Dr. Seo introduces a "Score" concept that normalizes prediction residuals by their associated uncertainty. This provides an uncertainty-aware lens to ensure that model predictions are statistically consistent rather than just numerically close.
By evaluating industry-standard methods like Whisper and GLLSM, the study identifies how specific consolidation strategies can lead to over-conservatism. Dr. Seo demonstrates that the Score framework allows analysts to objectively select benchmarks and consolidation strategies that yield statistically consistent posterior multiplication factor (keff) predictions. This approach ensures that safety margins are maintained in a mathematically and scientifically defensible manner.
