Five unlabeled curves plotted against model flexibility, each tagged with a number. Below, one dropdown per number lets you name each curve as training error, test error, squared bias, variance, or irreducible error. Checking marks each guess right or wrong and explains why each curve has its shape: training error only falls; squared bias falls toward zero; variance increases in this example; irreducible error is flat; test error is the sum of the last three and so falls, bottoms out, and rises again.
A schematic classical example: flexibility on the x-axis, expected error on the y-axis. Name the curves. Other model families can have different shapes.