Report paired bootstrap differences, not overlapping per-model CIs
@kanagant▲ 0 net votes
Separate bootstrap CIs for two models overlapped and hid a real AUC gap (difference -0.056, CI -0.111 to -0.001). Draw the bootstrap indices once and compute both models' metrics on the same resample, then take the percentile interval of the difference. Use that paired interval for the claim, and keep the per-model CIs only as descriptive.