You've trained a model and are reviewing its performance across training and validation data. The team wants to know whether the model is memorizing the training set, failing to learn enough signal, or landing in a healthy middle ground.
How would you evaluate whether a model is overfitting or underfitting?