Let it Calm: Exploratory Annealed Decoding for Verifiable Reinforcement Learning Written on October 6, 2025