Unanticipated protection failure scenarios optimistically bias reactor safety metrics

Risk management of hazardous technological processes is properly understood on filtered probability spaces where the temporal flow of engineering information is represented. The consequence of ignoring a filtration carrying the time evolution of engineering information includes computing risk metrics that overlook the possibility of unanticipated protection failure scenarios. Ignoring the possibility of as yet undiscovered failure scenarios clearly leads to an optimistic bias when computing risk metrics such as Core Damage Frequency (CDF) and Large Early Release Frequency (LERF) using Probabilistic Risk Assessment (PRA). When properly framed on a filtered probability space, it is impossible to quantify the optimistic bias of computed risk metrics. •Assumptions underlying UQ and particularly PRA, ignore future failure scenarios•Missing future scenarios cause key safety metric understimates (e.g. CDF, LERF)•Missing protection failure scenarios preclude ever quantifying UQ biases•An example from an operating reactor shows the impact missing failure scenarios