The training lab taught the gradient loop, and it is easy to leave it believing everything learns that way. Almost nothing does. Most of the decisions machines make about you today (the fraud hold on your card, the queue your support ticket lands in, the price band on your insurance) come from machines that are older, smaller and stranger: one that refuses to learn anything and just remembers, one that plays twenty questions, one that finds groups nobody labelled, and one straight line from 1958. This lab is those four, plus the trick for seeing high-dimensional data at all. Same world as the metrics lab: a reading screener’s learners, plotted by minutes read a day against comprehension.