Compare monitor approaches, models, and reasoning effort on the same transcripts. Each table changes one setting and holds the others fixed.