Discussion of the IAEA Error Approach to Producing Variance Estimates for Use in Material Balance Evaluation and the International Target Values, and Comparison to Metrological Definitions of Precis

Representatives from the nuclear safeguards evaluator and analytical communities are engaged in discussions on approaches to uncertainty quantification (UQ) concerning measurements taken for safeguards purposes. Their objective is to bring together an understanding of the approaches applied in each domain and to clarify the intended purpose of each. UQ in nuclear safeguards evaluation is primarily concerned with monitoring uncertainty of measurement systems at facilities under comprehensive safeguards agreements so that a material balance evaluation (MBE) can be conducted and the material unaccounted for (MUF) can be assessed for significance against that quantified uncertainty. A second objective of nuclear safeguards evaluators is the decennial publishing of the International Target Values (ITVs). To accomplish each of these, in many cases (and in the case we consider), analysis of variance (ANOVA) methods are applied to paired operator-inspector verification measurements in order to estimate variance components, assuming the simplest defensible measurement error model. The analytical chemistry community, which includes members of the network of analytical laboratories (who provide sample analysis results to safeguards evaluators for verification and MBE purposes) as well as producers of reference materials, is also concerned with UQ as it pertains to an analytical method. Following modern best practices, they typically appeal to the GUM (bottom-up) principles and/or reproducibility studies (top-down) to estimate the error variances of an analytical method. Guidance on top-down and bottom-up approaches to UQ of analytical methods is provided in multiple international standards, which include, but are not limited to: JCGM 100, JCGM 200, ISO 5725, and ISO 21748. In support of this current effort to reconcile UQ paradigms, this paper discusses the statistical underpinnings of each approach.