Tailored primitive initialization is the secret key to reinforcement learning was published at ACL 2026.