Neighbor voting and k-means both depend on distance, but they answer different questions. A nearest-neighbor classifier predicts a known outcome using labeled training rows. K-means creates groups without a target. This practice set alternates them so the presence or absence of labels remains part of your modeling decision.
Read the preparation requirement in every prompt. The classifier tasks use a training-fitted scaling pipeline and an explicit neighbor count. The clustering tasks sometimes ask for unscaled inputs as a controlled comparison. Cluster label values can be permuted without changing the grouping, but sorting the labels destroys their correspondence with customers. No new methods appear here. Keep the row order and let the fitted objects calculate outputs for the changed fictional fixtures.