Validation performance assessment through quantitative score focused on benchmark selection for nuclear criticality safety [version 3; peer review: 2 approved]

Background Validation of neutron transport calculations for nuclear criticality safety requires selecting relevant benchmarks and consolidating multiple uncertainty sources to estimate application bias and uncertainty. Existing practices rely on sensitivity/uncertainty (S/U) techniques and heuristic similarity thresholds, but they lack a simple, uncertainty-aware metric for comparing validation outcomes across consolidation methods and benchmark sets. Methods We introduce a concept of score that quantifies validation performance as the posterior residual normalized by its associated uncertainty. The score is evaluated for four consolidation approaches—parametric, nonparametric, Whisper, and generalized least-squares methodology (GLLSM)—using S/U-based indices. The analysis considers keff predictions for representative sets of International Handbook of Evaluated Criticality Safety Benchmark Experiments benchmarks and corresponding applications. Results Score distributions reveal distinct behaviors across methods. Nonparametric and Whisper generally yield positively shifted means, reflecting conservative bias estimation by design. Parametric and GLLSM produce means closer to zero but can exhibit wider score spreads when selection is based solely on conventional similarity thresholds. Incorporating the scaled similarity index reduces spread and mean shift for the parametric method by accounting for sensitivity magnitude. For GLLSM, acknowledging additional prior sources of uncertainty not captured by nuclear data alone brings the empirical score spread closer to unity across relevance levels, indicating improved statistical consistency. Conclusions The score provides uncertainty-aware lens for comparing consolidation methods and guiding benchmark selection. Results support using scaled similarity with the parametric method and accounting for unmodeled prior uncertainty within GLLSM. The framework helps analysts select benchmarks and consolidation strategies that yield statistically consistent posterior keff predictions while maintaining defensible safety margins.