A product classifier performs extremely well on its training examples but poorly on a representative held-out test set. The training labels are valid, and the test distribution matches expected use. Which issue is most consistent with these results?

1. Successful generalization, because training accuracy alone determines production quality.
2. Overfitting: the model learned training-specific patterns that do not generalize well.
3. Inference has changed into unsupervised learning merely because the test examples are new.
4. Underfitting: the model necessarily performs poorly on the training data as well.