Elamur Research⌕
☰
home / papers / 2026 / FACT-E: Causality-Inspired Evaluation for Trustworthy Chain-of-Thought Reasoning

FACT-E: Causality-Inspired Evaluation for Trustworthy Chain-of-Thought Reasoning

Paper · 2026 · Yuxi Sun, Aoqi Zuo, Haotian Xie et al.

Summary

Chain-of-Thought outputs from LLMs can look coherent and still reach the right answer while their intermediate steps do not actually follow from one another. Self-evaluation methods such as self-reflection tend to miss this, because the evaluating model is biased toward approving its own text and toward shallow lexical cues. The authors present FACT-E (Faithfulness And Consistency Tandem Estimation), which frames CoT assessment as a Structural Causal Model. In this model, the LLM's hidden bias is an unobserved confounder.

FACT-E has two components:

  • Intra-Chain Faithfulness: the chain is split at each step into a preceding and a following segment. Controlled external noise acts as an instrumental variable to estimate how much the later reasoning genuinely depends on the earlier part.
  • CoT-to-Answer Consistency: this checks whether the chain reliably leads to the correct answer.

Together they yield a reliability score in [0, 1]. On GSM8K, MATH and CommonsenseQA, ranking candidate chains by this score improves trajectory selection and produces better in-context learning exemplars. It also detects flawed reasoning under noisy conditions. The work offers a more bias-robust way to judge reasoning quality beyond final-answer correctness.

Key points

  • The framework is evaluated on three tasks: picking higher-quality CoTs to raise accuracy, using selected chains as ICL exemplars, and identifying noisy or flawed reasoning.
  • In the paper's motivating example, self-reflection scores a redundant chain and a compact chain the same (0.95 each). FACT-E separates them, giving 0.559 to the redundant chain and 0.998 to the compact one.
  • The authors identify two bias sources the method targets: self-affirmation bias and statistical shortcuts such as lexical overlap.
  • Experiments with Gpt-4o-mini cover lightweight and standard FACT-E variants. About three sampling trials (N) capture most of the gains, and more trials give diminishing returns.
  • Code is released at github.com/peachch/FACT-E.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓The University of MelbourneMelbourne · Australia100 works · 22,618 citations🎓Singapore Management UniversitySingapore · Singapore27 works · 2,382 citations🎓Hong Kong Baptist UniversityHong Kong · China7 works · 246 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

🧭 Topics 1Research subjects and languages the work is about.Reasoning · 1,426 works

Cite

Cite as Yuxi Sun, Aoqi Zuo, Haotian Xie, Wei Gao, Mingming Gong and Jing Ma. 2026. FACT-E: Causality-Inspired Evaluation for Trustworthy Chain-of-Thought Reasoning. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-acl.1014

@inproceedings{sun2026fact,
  title = {{FACT-E: Causality-Inspired Evaluation for Trustworthy Chain-of-Thought Reasoning}},
  author = {Yuxi Sun, Aoqi Zuo, Haotian Xie, Wei Gao, Mingming Gong and Jing Ma},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.1014},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.1014}
}

References

47 references, 21 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Our approach, standard FACT-E, aims to assess all the steps of CoT. While it requires prompt LLMs multiple times, leading to higher inference costs compared to simpler prompting approaches. To mitigate the cost, we introduce lightweight FACT-E, which reduces the number of prompting requests while ma… find
  2. Shidong Cao, Hongzhan Lin, Yuxuan Gu, Ziyang Luo, and Jing Ma. 2026. Diffcot: Diffusion-styled chain-of-thought reasoning in llms. arXiv preprint arXiv:2601.03559. find
  3. Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, and 1 others. 2021. Training verifiers to solve math word problems. arXiv preprint arXiv:2110.14168.
  4. Yingqian Cui, Pengfei He, Xianfeng Tang, Qi He, Chen Luo, Jiliang Tang, and Yue Xing. 2024. A theoretical understanding of chain-of-thought: Coherent reasoning and error-aware demonstration. arXiv preprint arXiv:2410.16540. — [arXiv:2410.16540](https://arxiv.org/abs/2410.16540) find
  5. Yilun Du, Shuang Li, Antonio Torralba, Joshua B Tenenbaum, and Igor Mordatch. 2023. Improving factuality and reasoning in language models through multiagent debate. In Forty-first International Conference on Machine Learning.
  6. Jiarun Fu, Lizhong Ding, Hao Li, Pengqi Li, Qiuning Wei, and Xu Chen. 2025a. Unveiling and causalizing cot: A causal pespective. arXiv preprint arXiv:2502.18239. — [arXiv:2502.18239](https://arxiv.org/abs/2502.18239) find
  7. Jiarun Fu, Lizhong Ding, Hao Li, Pengqi Li, Qiuning Wei, and Xu Chen. 2025b. Unveiling and causalizing cot: A causal pespective. Preprint, arXiv:2502.18239. — [arXiv:2502.18239](https://arxiv.org/abs/2502.18239) find
  8. Yanji He, Yuxin Jiang, Yiwen Wu, Bo Huang, Jiaheng Wei, and Wei Wang. 2026. Idea: An interpretable and editable decision-making framework for llms via verbal-to-numeric calibration. Preprint, arXiv:2604.12573. find
  9. Dan Hendrycks, Collin Burns, Saurav Kadavath, Akul Arora, Steven Basart, Eric Tang, Dawn Song, and Jacob Steinhardt. 2021. Measuring mathematical problem solving with the math dataset. arXiv preprint arXiv:2103.03874.
  10. Fanding Huang, Guanbo Huang, Xiao Fan, Yi He, Xiao Liang, Xiao Chen, Qinting Jiang, Faisal Nadeem Khan, Jingyan Jiang, and Zhi Wang. 2026. Semantic-space exploration and exploitation in rlvr for llm reasoning. Preprint, arXiv:2509.23808. find
  11. Jie Huang, Xinyun Chen, Swaroop Mishra, Huaixiu Steven Zheng, Adams Wei Yu, Xinying Song, and Denny Zhou. 2023. Large language models cannot self-correct reasoning yet. arXiv preprint arXiv:2310.01798.
  12. Alihan Hüyük, Xinnuo Xu, Jacqueline Maasch, Aditya V. Nori, and Javier González. 2025. Reasoning elicitation in language models via counterfactual feedback. Preprint, arXiv:2410.03767. — [arXiv:2410.03767](https://arxiv.org/abs/2410.03767) find
  13. Junhao Jia, Yueyi Wu, Huangwei Chen, Haodong Jing, Haishuai Wang, Jiajun Bu, and Lei Wu. 2026. Unsupervised causal prototypical networks for de-biased interpretable dermoscopy diagnosis. arXiv preprint arXiv:2602.23752. — [arXiv:2602.23752](https://arxiv.org/abs/2602.23752) find
  14. Bowen Jiang, Yangxinyu Xie, Zhuoqun Hao, Xiaomeng Wang, Tanwi Mallick, Weijie J Su, Camillo Jose Taylor, and Dan Roth. 2024a. A peek into token bias: Large language models are not yet genuine reasoners. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages…
  15. Bowen Jiang, Yangxinyu Xie, Zhuoqun Hao, Xiaomeng Wang, Tanwi Mallick, Weijie J Su, Camillo Jose Taylor, and Dan Roth. 2024b. A peek into token bias: Large language models are not yet genuine reasoners. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages…
  16. Huiqiang Jiang, Qianhui Wu, Chin-Yew Lin, Yuqing Yang, and Lili Qiu. 2023. LLMLingua: Compressing prompts for accelerated inference of large language models. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 13358–13376, Singapore. Association for Compu…
  17. Saurav Kadavath, Tom Conerly, Amanda Askell, Tom Henighan, Dawn Drain, Ethan Perez, Nicholas Schiefer, Zac Hatfield-Dodds, Nova DasSarma, Eli Tran-Johnson, Scott Johnston, Sheer El-Showk, Andy Jones, Nelson Elhage, Tristan Hume, Anna Chen, Yuntao Bai, Sam Bowman, Stanislav Fort, and 17 others. 2022a… find
  18. Saurav Kadavath, Tom Conerly, Amanda Askell, Tom Henighan, Dawn Drain, Ethan Perez, Nicholas Schiefer, Zac Hatfield-Dodds, Nova DasSarma, Eli Tran-Johnson, and 1 others. 2022b. Language models (mostly) know what they know. arXiv preprint arXiv:2207.05221. find
  19. Takeshi Kojima, Shixiang Shane Gu, Machel Reid, Yutaka Matsuo, and Yusuke Iwasawa. 2023. Large language models are zero-shot reasoners. Preprint, arXiv:2205.11916.
  20. Tamera Lanham, Anna Chen, Ansh Radhakrishnan, Benoit Steiner, Carson Denison, Danny Hernandez, Dustin Li, Esin Durmus, Evan Hubinger, Jackson Kernion, and 1 others. 2023. Measuring faithfulness in chain-of-thought reasoning. arXiv preprint arXiv:2307.13702.
  21. Xiaoyuan Li, Keqin Bao, Yubo Ma, Moxin Li, Wenjie Wang, Rui Men, Yichang Zhang, Fuli Feng, Dayiheng Liu, and Junyang Lin. 2025. Mtr-bench: A comprehensive benchmark for multi-turn reasoning evaluation. arXiv preprint arXiv:2505.17123.
  22. Guanran Luo, Wentao Qiu, Zhongquan Jian, Meihong Wang, and Qingqiang Wu. 2026. Gcot-decoding: Unlocking deep reasoning paths for universal question answering. arXiv preprint arXiv:2604.06794.
  23. Aman Madaan, Niket Tandon, Prakhar Gupta, Skyler Hallinan, Luyu Gao, Sarah Wiegreffe, Uri Alon, Nouha Dziri, Shrimai Prabhumoye, Yiming Yang, and 1 others. 2023. Self-refine: Iterative refinement with self-feedback. Advances in Neural Information Processing Systems, 36:46534–46594. find
  24. Nick McKenna, Tianyi Li, Liang Cheng, Mohammad Hosseini, Mark Johnson, and Mark Steedman. 2023. Sources of hallucination by large language models on inference tasks. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 2758–2774.
  25. Judea Pearl. 2009. Causality: Models, Reasoning, and Inference. Cambridge University Press. find
  26. Ansh Radhakrishnan, Karina Nguyen, Anna Chen, Carol Chen, Carson Denison, Danny Hernandez, Esin Durmus, Evan Hubinger, Jackson Kernion, Kamilė Lukošiūtė, and 1 others. 2023. Question decomposition improves the faithfulness of model-generated reasoning. arXiv preprint arXiv:2307.11768. find
  27. Denny Zhou. 2023. Self-consistency improves chain of thought reasoning in language models. Preprint, arXiv:2203.11171.
  28. Bilgehan Sel, Ahmad Al-Tawaha, Vanshaj Khattar, Ruoxi Jia, and Ming Jin. 2023. Algorithm of thoughts: Enhancing exploration of ideas in large language models. arXiv preprint arXiv:2308.10379. find
  29. Yu Wang and Chu-Ren Huang. 2024. Word boundary decision: An efficient approach for low-resource word segmentation. In Proceedings of the 38th Pacific Asia Conference on Language, Information and Computation, pages 160–169. find
  30. Xu Shen, Song Wang, Zhen Tan, Laura Yao, Xinyu Zhao, Kaidi Xu, Xin Wang, and Tianlong Chen. 2025. Faithcot-bench: Benchmarking instance-level faithfulness of chain-of-thought reasoning. arXiv preprint arXiv:2510.04040. — [arXiv:2510.04040](https://arxiv.org/abs/2510.04040) find
  31. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Ed Chi, Quoc Le, and Denny Zhou. 2022. Chain-of-thought prompting elicits reasoning in large language models. In Advances in Neural Information Processing Systems (NeurIPS).
  32. Yuhao Shen, Tianyu Liu, Junyi Shen, Jinyang Wu, Quan Kong, Li Huan, and Cong Wang. 2026. Double: Breaking the acceleration limit via double retrieval speculative parallelism. arXiv preprint arXiv:2601.05524. find
  33. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed Chi, Quoc Le, and Denny Zhou. 2023. Chain-of-thought prompting elicits reasoning in large language models. Preprint, arXiv:2201.11903.
  34. Yuxi Sun, Wei Gao, Hongzhan Lin, Jing Ma, and Wenxuan Zhang. 2025a. Explainable ethical assessment on human behaviors by generating conflicting social norms. In Proceedings of the 14th International Joint Conference on Natural Language Processing and the 4th Conference of the Asia-Pacific Chapter of… find
  35. Jinyang Wu, Shuo Yang, Changpeng Yang, Yuhao Shen, Shuai Zhang, Zhengqi Wen, and Jianhua Tao. 2026. Spark: Strategic policy-aware exploration via dynamic branching for long-horizon agentic learning. arXiv preprint arXiv:2601.20209.
  36. Yuxi Sun, Aoqi Zuo, Wei Gao, and Jing Ma. 2025b. Causalabstain: Enhancing multilingual llms with causal reasoning for trustworthy abstention. In Findings of the Association for Computational Linguistics: ACL 2025, pages 14060–14076. find
  37. Alon Talmor, Jonathan Herzig, Nicholas Lourie, and Jonathan Berant. 2019. Commonsenseqa: A question answering challenge targeting commonsense knowledge. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies,…
  38. Miles Turpin, Julian Michael, Ethan Perez, and Samuel R Bowman. 2024. Language models don’t always say what they think: Unfaithful explanations in chain-of-thought prompting. Advances in Neural Information Processing Systems (NeurIPS).
  39. Chengbing Wang, Yang Zhang, Wenjie Wang, Xiaoyan Zhao, Fuli Feng, Xiangnan He, and Tat-Seng Chua. 2025. Think-while-generating: On-the-fly reasoning for personalized long-form generation. arXiv preprint arXiv:2512.06690. — [arXiv:2512.06690](https://arxiv.org/abs/2512.06690) find
  40. Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc Le, Ed Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou. 2022. Self-consistency improves chain of thought reasoning in language models. arXiv preprint arXiv:2203.11171.
  41. Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc Le, Ed Chi, Sharan Narang, Aakanksha Chowdhery, and Junda Wu, Tong Yu, Xiang Chen, Haoliang Wang, Ryan Rossi, Sungchul Kim, Anup Rao, and Julian McAuley. 2024. DeCoT: Debiasing chain-of-thought for knowledge-intensive tasks in large language models via c… find
  42. Zhiheng Xi, Senjie Jin, Yuhao Zhou, Rui Zheng, Songyang Gao, Tao Gui, Qi Zhang, and Xuanjing Huang. 2024. Self-polish: Enhance reasoning in large language models via problem refinement. Preprint, arXiv:2305.14497. find
  43. Zhiheng Xi, Senjie Jin, Yuhao Zhou, Rui Zheng, Songyang Gao, Jia Liu, Tao Gui, Qi Zhang, and Xuan-Jing Huang. 2023. Self-polish: Enhance reasoning in large language models via problem refinement. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 11383– 11406. find
  44. Zidi Xiong, Shan Chen, Zhenting Qi, and Himabindu Lakkaraju. 2025. Measuring the faithfulness of thinking drafts in large reasoning models. arXiv preprint arXiv:2505.13774. — [arXiv:2505.13774](https://arxiv.org/abs/2505.13774) find
  45. Shijia Xu, Yu Wang, Xiaolong Jia, Zhou Wu, Kai Liu, and April Xiaowen Dong. 2026a. Rcbsf: A multi-agent framework for automated contract revision via stackelberg game. Preprint, arXiv:2604.10740.
  46. Shijia Xu, Zhou Wu, Xiaolong Jia, Yu Wang, Kai Liu, and April Xiaowen Dong. 2026b. Self-correcting rag: Enhancing faithfulness via mmkp context selection and nli-guided mcts. Preprint, arXiv:2604.10734.
  47. Sohee Yang, Sang-Woo Lee, Nora Kassner, Daniela Gottesman, Sebastian Riedel, and Mor Geva. 2025. How well can reasoning models identify and recover from unhelpful thoughts? arXiv preprint arXiv:2506.10979. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/sun-2026-fact-e-causality-inspired-evaluation/