3

We introduce a training paradigm that utilises plausible counterfactual explanations (p-CFEs) to achieve standard model accuracy while reducing reliance on spurious correlations.