A judgment trainer for the bias–variance chapter. Each round asks one multiple-choice question: whether a situation calls for a more rigid or more flexible fit; which named quantity a sketched curve is, or which sketch matches a name; which fit produced a shown pile of predictions; whether a piece of error is reducible or irreducible; what happens to an error when a knob or the dataset changes; whether a task is regression, classification or clustering, or prediction versus inference; and quick high-dimension distance calls. One click commits; every answer, right or wrong, comes with a one-line reason built from bias, variance and noise.

Practice interpreting error, choosing a starting model, and reading classical schematic curves. Model choices are hypotheses to check with validation data, not guarantees.