PCP RCP-PCP CPCG-PCP DR-CP RCP-DR-CP CPCG-DR-CP scm20d 0.834 0.886 0.874 0.828 0.873 0.891 rf1 0.683 0.849 0.798 0.704 0.850 0.874 scm1d 0.735 0.870 0.882 0.773 0.858 0.881 meps_19 0.756 0.873 0.892 0.798 0.862 0.888 Table 1: Comparison of worst-slab coverage on multi-output datasets. Both RCP and CPCG variants achieve comparable worst-slab coverage, close to the nominal level. PCP RCP-PCP CPCG-PCP DR-CP RCP-DR-CP CPCG-DR-CP scm20d 0.8491 13.34 5869 0.01838 24.03 5487 rf1 0.1943 7.704 8568 0.009347 11.25 8020 scm1d 0.8718 9.449 4747 0.01346 11.98 4778 meps_19 0.1349 2.272 5988 0.01803 5.277 5682 Table 2: Comparison of computational time (in seconds) on multi-output datasets. Overall, CPCG is 200x- 1000x slower than RCP. As detailed in Gibbs et al (2024), Section 4, the CPCG method requires solving an optimization problem involving the entire calibration set for each test instance , resulting in significantly higher computational demands, particularly with large calibration sets. This computational burden hinders the practical application of CPCG to large-scale datasets, highlighting a key advantage of RCP. 1