How training a model on one value or behavior influences its performance on other unseen values or contexts.