Beyond the Commitment Boundary: Probing Epiphenomenal Chain-of-Thought in Large Reasoning Models — ThinkLLM