24 specimen cards from the classic Iris dataset (Fisher, 1936) — a subset of 8 per species. Sort them, split them, and grow a decision tree by hand.