A data science team has finished training a loan-approval classification model. Before they release it to the production endpoint, they want to measure how well the model generalizes to data it has never seen, using a portion of data that was held out from training. Which phase of the ML development lifecycle does this activity represent?