Reasoning Trace Analysis

Follow this topic Ask about this topic

Momentum

4 papers in the last four weeks, against 2 the four weeks before. 0.0% of all new papers.

Jul 13Week of Sep 28

Latest papers 106

All topics

Decodable but Not Faithful: Coupling Natural-Language Rationales to Programmatic Verifiers
Jun 19, 2026Vatsal Ananthula, Adarsh KumarappanFaithfulness of Language Model Explanations LLM Auditing
University of California, Berkeley · 2California Institute of Technology
From Brewing to Resolution: Tracing the Internal Lifecycle of Code Reasoning in LLMs
Jun 16, 2026Siyue Chen, Yifu Guo, Yuquan Lu +9LLM Interpretability Code Language Models
1South China University of Technology · 2Sun Yat-sen University · 4Shanghai Jiao Tong University +5
GRACE: Step-Level Benchmark for Faithful Reasoning over Context
Jun 15, 2026Hoang Pham, Dong Le, Anh Tuan LuuCoT Faithfulness Contextual Faithfulness
Nanyang Technological University · VinUniversity
Beyond Accuracy: Measuring Bias Acknowledgment in Chain-of-Thought Reasoning for Responsible AI Evaluation
Jun 13, 2026Xian Sun, Wei Gao, Yingshuo Wang +9Language Model Generation Evaluation CoT Evaluation
Duke University · Northeastern University · University of California, Berkeley +5
Beyond the Commitment Boundary: Probing Epiphenomenal Chain-of-Thought in Large Reasoning Models
Jun 11, 2026Daniel Scalena, Sara Candussio, Luca Bortolussi +3CoT Faithfulness LLM Interpretability
CLCG, University of Groningen · University of Milano-Bicocca · University of Trieste +1
MARS: Margin-Adversarial Risk-controlled Stopping for Parallel LLM Test-time Scaling
Jun 11, 2026Wenbo Chen, Puheng Li, Mengyang Liu +2LLM Inference Efficiency Self-Consistency Decoding
Amazon · Stanford University · University of Pennsylvania
RecurGuard: Runtime Monitoring for Reasoning-Token Consumption Attacks
Jun 6, 2026Abid Aziz, Hafsa Binte KibriaAdversarial Attacks on LLMs LLM Security
Department of Electrical & Computer Engineering Rajshahi University of Engineering & Technology, Rajshahi 6204, Bangladesh
A Comprehensive Anatomy of Human and DeepSeek-R1 LLM Mathematical Reasoning
Jun 5, 2026Yuxiang Chen, Jun WangCoT Reasoning Structured Reasoning
UCL Centre for Artificial Intelligence
Self-Commitment Latency: A Reward-Free Probe for Prompted Implicit Hacking
Jun 4, 2026Bonan Shen, Youting Wang, Dingyan Shang +1Reward Hacking Language Model Probing
Independent Researcher · Northeastern University · Syracuse University
Inducing Reasoning Primitives from Agent Traces
Jun 2, 2026Zhihan Lei, Jiarui Yan, Joshua Momo +1LLM Agent Skill Learning Agentic Reasoning
Carnegie Mellon University
Thinking Past the Answer: Evaluating Harmful Overthinking in Large Reasoning Models
Jun 1, 2026Simone Caldarella, Davide Talon, Rahaf Aljundi +2LLM Evaluation Overthinking in Language Models
University of Trento · 3Fondazione Bruno Kessler · 2Toyota Motor Europe
Unveiling the Entropy Dynamics of Chain-of-Thought Reasoning
Jun 1, 2026Ting Xu, Xu He, Yupu Lu +4CoT Reasoning Test-Time Scaling
The Chinese University of Hong Kong · Huawei Technologies Ltd · School of Computing and Data Science, The University of Hong Kong
Hidden Thoughts Are Not Secret: Reasoning Trace Exposure in LLMs
May 30, 2026Yu-An Lu, Ci-Yang Tsai, Yu-Lin Tsai +2LLM Prompting Privacy Leakage in Language Models
National Yang Ming Chiao Tung University · UC Berkeley
Conformal Certification of Reasoning Trace Prefixes
May 28, 2026Matt Y. Cheung, Ashok Veeraraghavan, Hanjie Chen +1Conformal Prediction Language Model Calibration
Department of Electrical & Computer Engineering, Rice University · Department of Computer Science, Rice University
TRACE: Toulmin-based Reasoning Assessment through Constructive Elements for LLM CoT Evaluation
May 28, 2026Yundong Kim, Heyoung YangLLM Evaluation Language Model Generation Evaluation
Applied Agent Research Center, Korea Institute of Science and Technology Information (KISTI), Republic of Korea · Department of Computer Science and Engineering, University of Seoul, Republic of Korea.
Diagnosing Harmful Continuation in Answer-Correct Long-CoT Training Traces
May 28, 2026Chen He, Yuhao Wu, Lei Wang +2Fine-Tuning LLM Fine-Tuning
University of Electronic Science and Technology of China · Singapore University of Technology and Design · Singapore Management University
ReasonOps: Operator Segmentation for LLM Reasoning Traces
May 28, 2026Daniel Lee, Owen Queen, James ZouCoT Evaluation CoT Reasoning
Stanford University
The Chain Holds, the Answer Folds: Trace-Answer Dissociation in Reasoning Models Under Adversarial Pressure
May 27, 2026Yubo Li, Ramayya Krishnan, Rema PadmanCoT Faithfulness Reasoning Evaluation
Carnegie Mellon University
Reasoning that Travels: Dissecting How Chain-of-Thought Transfers Across Models
May 27, 2026Xinyuan Cheng, Beiduo Chen, Philipp Mondorf +1LLM Prompting LLM Reasoning
UMaiNLP, Center for Information and Language Processing, LMU Munich, Germany · ÆMunich Center for Machine Learning, Germany
CORE: Contrastive Reflection Enables Rapid Improvements in Reasoning
May 27, 2026Linas Nasvytis, Simon Jerome Han, Ben Prystawski +3LLM Self-Refinement Efficient Language Model Reasoning
Stanford University
The Shape of Overthinking: Backtracking Bursts in Long Reasoning Traces
May 27, 2026Navid Rezazadeh, Arash Gholami DavoodiReasoning Evaluation Efficient Language Model Inference
University of California, Irvine · Carnegie Mellon University
Beyond Input Understanding: Diagnosing Multilingual Mathematical Reasoning with Directed Acyclic Trace Graphs
May 26, 2026Jiaqiao Zhang, Zhoujun Li, Raoyuan Zhao +5Mathematical Reasoning Low-Resource Language Processing
Southwest University · LMU Munich & MCML
GeoFaith: A Spatio-Temporal Dual View of Faithful Chain-of-Thought
May 26, 2026Weijiang Lv, Wentong Zhao, Jiayu Wang +3CoT Faithfulness RL for Language Model Reasoning
1Xidian University · 2Xi’an Jiaotong University · 3Mohamed bin Zayed University of Artificial Intelligence +2
Beyond a Single Direction: Chain-of-Thought Disrupts Simple Steering of Refusal
May 26, 2026Kia-Jüng Yang, Dominik Meier, Jiachen Zhao +2CoT Reasoning LLM Refusal Behavior
University of Göttingen, Germany · LKA NRW, Germany · Northeastern University, Boston, MA, USA.
Explaining Too Much? Understanding How Large Language Model Reasoning Traces Influence Performance and Metacognition
May 25, 2026Daniela Fernandes, Daniel Buschek, Lev Tankelevitch +2CoT Reasoning Human-AI Interaction
Aalto University, Espoo, Finland · University of Bayreuth, Bayreuth, Germany · Microsoft Research Cambridge, Cambridge, United Kingdom +1
Detecting Unfaithful Chain-of-Thought via Circuit-Guided Internal-External Discrepancy
May 25, 2026Xu Shen, Zhen Tan, Song Wang +4CoT Faithfulness LLM Auditing
1Jilin University · University of Central Florida · 3Arizona State University +2
Faithfulness as Information Flow: Evaluating and Training Faithful Chain-of-Thought Reasoning
May 22, 2026Jinghan Jia, Joe Benton, Eric EasleyCoT Faithfulness LLM Interpretability
†Dept. CSE, Michigan State University · ⋆Anthropic Fellows Program · ‡Anthropic
PathCal: State-Aware Reflection-Marker Calibration for Efficient Reasoning
May 21, 2026Lingyu Jiang, Zirui Li, Shuo Xing +6Efficient Language Model Reasoning Efficient Language Model Inference
1Tohoku University · 2Texas A&M University · 3Worcester Polytechnic Institute
Insights Generator: Systematic Corpus-Level Trace Diagnostics for LLM Agents
May 20, 2026Akshay Manglik, Vijay S. Kalmath, Jason Qin +6Multi-Agent LLM Systems LLM Agent Evaluation
Scale AI
Reasoning-Trace Collapse: Evaluating the Loss of Explicit Reasoning During Fine-Tuning
May 20, 2026Lukas Twist, Helen Yannakoudakis, Jie M. ZhangSupervised Fine-Tuning Reasoning Evaluation
King’s College London London, UK

Jun 19, 2026, cs.LGJ/K move · Enter open · S save

Decodable but Not Faithful: Coupling Natural-Language Rationales to Programmatic Verifiers

Vatsal Ananthula, Adarsh Kumarappan

University of California, Berkeley · 2California Institute of Technology

Faithfulness of Language Model Explanations LLM Auditing LLM Interpretability Reasoning Consistency in Language Models LLM Reasoning Reasoning Trace Analysis

Open paper AskPDF arXiv

Language models can generate plausible rationales for their predictions, but these explanations may not faithfully represent the model's internal reasoning. We propose verifier-coupled reasoning, a framework that inserts inline claims into reasoning traces and trains an auxiliary consistency head to predict programmatic verifier outputs from rationale-span hidden states. The central finding is a gap between decodability and faithfulness: consistency training reliably makes verifier information decodable from rationale representations, but decodability does not guarantee faithful generation. In LeanCheck (formal theorem proving), rationale-only and proof-only pooling achieve perfect directional separation under counterfactual conflict. In KataGo (Go engine), commentary spans encode 10-way win-rate buckets at 81% accuracy. Yet in a code setting, the model achieves 98.6% coupling while its generated explanations remain unfaithful: fluent prose with correct structured claims, but describing unrelated algorithms; a controlled pretrained-vs-from-scratch comparison shows the gap is not capacity-driven. Synthetic activation patching confirms causal influence (73-89% vs. 31% baseline), FEVER reveals that evidence-only pooling isolates genuine evidence sensitivity at the cost of raw accuracy, and per-claim analysis shows that consistency loss disproportionately benefits fine-grained claims over binary ones. These results establish that consistency losses are effective diagnostics and representation-shaping tools, but not sufficient conditions for faithful reasoning.

Momentum

Related topics

Latest papers 106