A test set is labeled data held back until after training. It checks whether a model performs well on observations it has never seen, rather than merely repeating its training examples. section 1.5
Test sets support predictive modeling, where performance matters more than inspecting every detail of a model. A strong result on a test set is useful evidence, not a guarantee: a model that performs well in a lab may still fail in the wild, especially when bias affects real decisions. section 1.5
