especially when optimized using large-scale reinforcement learning and verifiable rewards (RLVR). However, for problems prone ...