Explaining predictions by showing which historical cases the model referenced when making a decision.