CRoS: Causality Reward on Scrambled CoTs Improves Accuracy and Reasoning Faithfulness
arXiv cs.AI · 31 Oct 2026, 03:18 AEST
CRoS introduces a novel metric for assessing causality in reasoning chains without altering content. This method improves accuracy and reasoning faithfulness in language models.
Key facts
- Reinforcement learning with verifiable rewards (RLVR) has become the standard method for training language models to reason.
- To address this problem, we propose CRoS (Causality Reward on Scrambled CoTs), a novel CoT causality metric that perturbs the order of reasoning steps without altering content, thereby preserving information within each step.
- On Reasoning Gym with Gemma-3-4B, CRoS outperforms existing causality and outcome rewards in out-of-domain accuracy, CoT quality, and CoT faithfulness.
Read aloud
CRoS introduces a novel metric for assessing causality in reasoning chains without altering content. This method improves accuracy and reasoning faithfulness in language models. Reinforcement learning with verifiable rewards RLVR has become the standard method for training language models to reason. To address this problem, we propose CRoS Causality Reward on Scrambled CoTs, a novel CoT causality metric that perturbs the order of reasoning steps without altering content, thereby preserving information within each step. On Reasoning Gym with Gemma-three-4B, CRoS outperforms existing causality and outcome rewards in out-of-domain accuracy, CoT quality, and CoT faithfulness.
Sources in this cluster
- CRoS: Causality Reward on Scrambled CoTs Improves Accuracy and Reasoning Faithfulness
arXiv cs.AI · 31 Oct 2026, 03:18 AEST