Cryptographic proof of agent cognition. Agents commit hashes of their reasoning traces before acting, then reveal to prove what they were thinking.