An in-browser calculator for the nuclear-data uncertainty, similarity and GLS bias of k: description and code-to-code comparisons

The S/U calculator is a web page that runs first-order sensitivity and uncertainty calculations in the user's browser, from TSUNAMI sensitivity data files or Serpent sensitivity output. It computes the nuclear-data uncertainty of k by the sandwich rule, its breakdown by nuclide and reaction, and the similarity index ck between systems. It also computes a generalized-linear-least-squares (GLS) adjustment, as in SCALE's TSURFER, that gives the bias of k for applications from benchmark experiments, the uncertainty that remains after the adjustment, and χ² of the experiments. The covariances are ENDF/B-VIII.1 in SCALE's 56 groups (AMPX and NJOY routes), ENDF/B-VII.1 from SAUNA, or a COVERX file of the user's own, and the user's files are never uploaded. This article describes the methods and reports seven sets of code-to-code comparisons. For 120 ICSBEP configurations with each of the three libraries, the calculator gives the uncertainties of the companion ICSBEP study to within 1.5×10−5 percentage points. The COVERX reader and the Serpent reader agree with their references to the precision of the stored or printed data, as does the uncertainty breakdown against SAUNA's (2.6×10−6 of the variance per pair). The GLS adjustment agrees with a separate Python implementation to 8.5×10−14 (relative) in every output. SANDY's parameter-space update (gls_update), written by other authors, was run with the dense covariance matrix of up to 8,400 parameters; the calculator's bias and posterior uncertainty agree with it to 1.4×10−14. These comparisons show that the implementations agree with each other; they do not test the method against measurements. The calculator has not yet been compared with TSURFER, TSUNAMI-IP or NDaST directly. PDF of the preprint published at https://hardtohit10.gitlab.io/su-calculator-paper.html (version 1). Generated with AI (Claude, Anthropic). About the authors and the status of this work. Claude is an AI model developed by Anthropic. Ian Hill proposed the calculator and its GLS section, created the DICE sensitivity profiles used and provided the original SCALE 6.0 calculations they come from, and directed the work. Claude wrote the calculator, the comparison scripts and the separate Python GLS implementation, ran the comparisons, made the figures and tables, and wrote this article. The article has not been peer reviewed, and the results have not been independently reviewed or verified. Anthropic has not reviewed or endorsed this work.