Fig. 6

Number of fits that are A accepted, B rejected and C revived based on synthetic experimental data in three iterations of the numerical compass (NC) method. Numbers are based on statistics for n = 500 simulations, where each fit in the KM-SUB fit ensemble is once selected as simulated truth. Medians are shown as white markers, interquartile ranges as vertical wide black lines and 1.5 \(\times\) interquartile ranges as narrow black lines. While experiment simulation (via KM-SUB) and fit filtering (of the KM-SUB fit ensemble, absolute MSLE threshold, \(\theta =\) 0.0105) are identical for all approaches, we compare different numerical selection methods of experiments: KM-only NC (blue), KM/SM-hybrid NC (orange), random selection of experiments (green) and parameter sensitivities of the KM (red). The simulation is performed on a reduced 10\(\times\)10 grid of experimental conditions within the usual ranges. Fit ensemble constraints are significantly larger when experiments are selected using the NC. While the two models utilized for its evaluation lead to very similar fit ensemble constraints, the random and sensitivity-based selection of experiments perform significantly worse