Fig. 4

a Demonstration of error-based calibration. Purple dots: simulated absolute errors (same as Fig. 1b). The errors are divided into bins (here five) according to their uncertainties. The RMSE and root mean variance (RMV) of each bin of errors is calculated (blue dots). For well-calibrated uncertainties (as here), the RMSE vs. RMV plot should follow a straight line with a slope of 1 and an intercept of 0. b Error-based calibration for the error-uncertainty distribution in Fig. 3a which were scaled to create a mismatch between errors and uncertainties. Unlike the miscalibration area, the error-based calibration metrics catch this mismatch and we see that both linear fit, slope and intercept gets worse