Fig. 5

\(n =\) 2500; Kuiper’s statistic is \(0.05536 / \sigma = 6.586\), Kolmogorov’s and Smirnov’s is \(0.04554 / \sigma = 5.418\). Figure 6 displays the ground-truth reliability diagram. The observed reliability diagrams fail to depict the underlying discontinuous jumps in the subpopulation’s expected outcomes as a function of the score. The plot of cumulative deviation succeeds in resolving some of the corresponding cusps, but does exhibit significant random fluctuations nearly as high as a quarter of the height of the triangle at the origin. The scalar summary statistics successfully detect the statistically significant deviation of the subpopulation from the full population