You're comparing a few supervised learning models and want a clear way to reason about why some underfit while others overfit.
What is the bias-variance tradeoff?