Statistical Model Comparison
A test comparing two models scored on the same rows by using only the disagreements, since agreement carries no information about which model is better.
Two models scored on the same test set produce paired data, not two independent samples, and treating it as unpaired discards the pairing and estimates the wrong variance. McNemar's test looks only at rows where the models disagree and asks whether that split is far from a coin flip.
The Wilcoxon signed-rank test is the paired equivalent for continuous per-fold scores rather than binary right-or-wrong outcomes.