Empirical miscoverage rate after calibration with CQR, computed on regions of higher density. Following Sect. 4.2.2, only pixels k with |κ_i[k]| ≥ t = 4.8 × 10⁻² were considered. Unlike in Fig. 5a, where the miscoverage rate is computed over all pixels, the measured values in this case deviate drastically from the target. Consequently, the size of the error bars is underestimated in these areas.