TCR

Researchers propose Thinking Checklist Reward (TCR), a process-oriented reward for reinforcement-learning-based LLM preference alignment that evaluates reasoning traces against sample-specific checkli…