A measurement study in three parts, of the same seventeen language models, run in September 2026: what each says about itself, what eight of them say about each other, and what those eight did with 30 written decisions, scored by five model judges who could not see who wrote what.