Evaluating whether a model produces consistent, stable predictions for individual inputs rather than just overall accuracy.