Anonymous supplementary material

Visualization of the 10k Results

Extensions of Figs. 3–5: detection, signed dispersion diagnosis, and severity ranking at 10k sample sizes.

Fig. 3 extension

Detection across six departures

Original signal settings and the selected weaker signal settings, shown separately on the same color scale.

Detection heatmaps for ZID, FID, and KID across six departures, with original signals above and selected weaker signals below.

Each cell reports the percentage of rejections over 300 repetitions, with 10,000 samples per group, 499 shared random relabelings, and a significance level of 5%. The boxed row is the standard six-arm ZID test. No error bars are shown.

Fig. 4 extension

Signed dispersion diagnosis

The signed ZID dispersion readout and FID under contraction and expansion.

The signed ZID dispersion score is negative under contraction and positive under expansion, whereas FID is unsigned.

Points show means across eight repetitions, with 10,000 samples per group. Blue error bars show the standard deviation of the signed scores across those repetitions. The scale c contracts candidate deviations when c < 1 and expands them when c > 1.

The ZID curve is the ungated net signed dispersion score, not the overall ZID score or a significance-qualified direction call. The two curves use separate vertical axes.

Fig. 5 extension

Ranking increasing perturbation severity

Original and weaker severity settings, evaluated using metric scores rather than permutation rejections.

Grouped bar charts of mean Spearman severity-ranking correlations for FID, KID, and ZID across six departure families, with original signals above and weaker signals below.

Bars show the mean panel-wise Spearman correlation across 100 panels per departure, each with six ordered severities and 10,000 samples per group. Error bars are pointwise 95% panel-bootstrap intervals based on 10,000 resamples. A negative correlation means that higher scores are assigned to less severe perturbations.