A four-game case study: Claude, ChatGPT, and Grok play full games of chess against each other — no engine help, no human host. Which AI actually thinks?