aiqbench analysis
This arXiv paper is about training language models with reinforcement learning, specifically a new reward signal (CRoS) that tests whether chain-of-thought steps causally drive the final answer by reordering them rather than deleting or corrupting them. The reported gains are modest and come from one model (Gemma-3-4B) on a benchmark suite, so they should be read as early evidence rather than a settled result. For readers, the useful takeaway is the methodological point: perturbation-based causality metrics may confound information loss with causal influence, and controlling for that could matter for reasoning-model training.
CRoS: Causality Reward on Scrambled CoTs Improves Accuracy and Reasoning Faithfulness
arXiv:2610.42054v1 Announce Type: cross Abstract: Reinforcement learning with verifiable rewards (RLVR) has become the standard method for training language models to reason. However, RLVR typically relies solely on outcome correctness, which does not guarantee that the CoT causally contributes to the final answer. Without this connection, outcome rewards may fail to reinforce high-quality reasoning. Existing causality metrics assess CoT causality by perturbing the CoT, such as deleting reasoning steps or injecting errors, and measuring changes in the final answer. However, these perturbations inevitably alter information in each step. Consequently, high causality scores may reflect sensitivity to information changes introduced by perturbation rather than true causality. Such confounding effects further induce undesirable reasoning behaviors when these metrics are used as training rewards, including generating less informative CoTs and reducing exploration and self-correction. 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. Specifically, CRoS achieves 46.8% accuracy on five unseen logic and games datasets, outperforming the strongest baseline (43.8%). It also achieves better CoT quality, consistently outperforming baselines in CoT-Pass@K from K=1 to 128, averaging 43.4% vs. 41.4%. For CoT faithfulness, CRoS achieves a hint acknowledgment score of 34.0% vs. 20.2%.
Abstract:Reinforcement learning with verifiable rewards (RLVR) has become the standard method for training language models to reason. However, RLVR typically relies solely on outcome correctness, which does not guarantee that the CoT causally contributes to the final answer. Without this connection, outcome rewards may fail to reinforce high-quality reasoning. Existing causality metrics assess CoT causality by perturbing the CoT, such as deleting reasoning steps or injecting errors, and measuring changes in the final answer. However, these perturbations inevitably alter information in each step. Consequently, high causality scores may reflect sensitivity to information changes introduced by perturbation rather than true causality. Such confounding effects further induce undesirable reasoning behaviors when these metrics are used as training rewards, including generating less informative CoTs and reducing exploration and self-correction. 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. Specifically, CRoS achieves 46.8% accuracy on five unseen logic and games datasets, outperforming the strongest baseline (43.8%). It also achieves better CoT quality, consistently outperforming baselines in CoT-Pass@K from K=1 to 128, averaging 43.4% vs. 41.4%. For CoT faithfulness, CRoS achieves a hint acknowledgment score of 34.0% vs. 20.2%.
17 pages, 3 figures, 14 tables. Project page: this https URL
Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
arXiv:2610.42054 [cs.CL]
(or arXiv:2610.42054v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2610.42054 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 337 words
Related coverage
Are Stated Reasoning Steps Causally Load-Bearing?
arXiv · Sep 24, 2026
C2-Faith: Benchmarking LLM Judges for Causal and Coverage Faithfulness in Chain-of-Thought Reasoning
arXiv · Jun 15, 2026
Evaluating the Causal Faithfulness of SHAP as a Candidate Attribution Method for SAE Features
arXiv · Oct 15, 2026
From Observation to Intervention: Distilling Self-Generated In-Context Demonstrations via Executable Causal Intervention
arXiv · Oct 16, 2026