The trustworthiness of system code simulations is a cornerstone of nuclear safety. The underlying assumption is that if applied with sufficient conservatism or taking into account uncertainties, system codes provide reasonable information to assess a nuclear reactor’s capability to withstand accidental sequences. Validation of the codes and models is therefore required to accept the previous assumption. Since the 1970s and by extensive experimental campaign experience, best practice guidelines have been developed that aim to provide robustness and accuracy to future results obtained while simulating the same phenomenon in different conditions. Unfortunately, blind pretests of experiments often yield myriad results when different institutions model the same phenomena with different codes. This is a well-known effect commonly referred as “user effect.”

In parallel, during recent decades, the Best Estimate Plus Uncertainty (BEPU) approach has been largely developed that helps in bounding uncertainties in computational calculations. While several studies have assessed the user effect on individual calculations, less focus has been brought to quantitatively assess it on the uncertainty bands obtained from a BEPU. Consequently, the present study tries to address a key question: Are the resulting uncertainty bands of different, but valid, modeling approaches overlapping, or are they separated even more when a forward uncertainty propagation (such as typical BEPU) methodology is used? In other words, does the user effect diminish or increase when a BEPU analysis is made? To answer this question, the authors employed the MELCOR severe accident code to assess the user effect against a typical BEPU method for a station blackout scenario by comparing the results of six different nodalizations of the model.