Probabilistic seismic‐hazard assessment (PSHA) consists of components that are largely empirical with some physical insights (e.g., empirical ground‐motion prediction equations and the Gutenberg–Richter magnitude distribution), components that are primarily based on simplified physics with some support from empirical data (e.g., the characteristic earthquake model, Schwartz and Coppersmith, 1984), and geologic information that requires a lot of resources to collect but is yet difficult to assure completeness (e.g., location and dimension of faults). Empirical models do not always have clear physical meanings, and physical models are not always empirically verifiable in all situations. Together with the geologic information that is almost always incomplete (and the degree of completeness unknown), assessing the true hazard level is a formidable task. The current practice of conducting PSHA aims at providing defensible justifications, based on the best available technology and information, for allocating resources to earthquake‐resistant design of engineering structures and hazard mitigation in general. Testing predictive models using observations has gained momentum in earthquake forecasting (e.g., Schorlemmer and Gerstenberger, 2007; Schorlemmer et al. , 2007) and became a standard through the Collaboratory for the Study of Earthquake Predictability (Jordan, 2006). In the realm of PSHA, the need of testing has long been recognized (Ward, 1995). After the 2011 Tohoku earthquake, furious debates of whether existing seismic‐hazard assessments (SHAs) have failed their purpose (Stein et al. , 2011; Hanks et al. , 2012; Stirling, 2012; Stein et al. , 2012; Frankel, 2013a,b; Stein et al. , 2013) reiterate the importance of testing SHAs with observations. In fact, PSHA tests proliferate in recent years (Ordaz and Reyes, 1999; Stirling and Petersen, 2006; Albarello and D’Amico, 2008; Beauval et al. , 2008; Fujiwara et al. , 2009; Miyazawa and Mori, 2009; Stirling and Gerstenberger, 2010; Mezcua et al. , 2013 …