Introduction
Is the new model really better, or did it just get lucky on this test set? Does this feature actually differ between two groups? Hypothesis testing gives a disciplined way to decide whether an observed effect is larger than what random noise would plausibly produce.