mlsanityData to 7 Oct 2026

Measuring and curing reasoning rigidity: from decorative chain-of-thought to genuine faithfulness

arXiv 2603.22816 ↗First posted 24 Mar 2026cs.CL cs.AI cs.LGcited by 1

On the map: LLM reasoning → Test-time compute and efficient thinking

Abstract (arXiv)

Language models increasingly show their work by writing step-by-step reasoning before answering. But are these steps genuinely used, or is the answer rigid - fixed before reasoning begins? We introduce the Step-Level Reasoning Capacity (SLRC) metric and prove it is a consistent causal estimator (Theorem 1). We propose LC-CoSR, a training method with Lyapunov stability guarantees that directly reduces rigidity. Evaluating 16 frontier models (o4-mini, GPT-5.4, Claude Opus, Grok-4, DeepSeek-R1, Gemini 2.5 Pro, and others) across six domains at N=133-500, we find reasoning falls into three modes. OpenAI's o4-mini shows 74-88% step necessity on five of six tasks (73.8-88.3%) - the highest SLRC in our study. The critical differentiator is RL-based reasoning training, not thinking tokens: Grok-4's reasoning mode shows lower faithfulness than its non-reasoning mode (1.4% vs 7.2% necessity). We discover a faithfulness paradox - high-SLRC models are more susceptible to sycophancy - and propose the Reasoning Integrity Score (RIS = SLRC x (1-Sycophancy)), which significantly predicts error detection (rho=0.66, p=0.026). LC-CoSR achieves 2.6x less negative reward than FARL and CSR baselines without external model dependencies.

Key facts. Cited by 1 paper in the map (data through 7 Oct 2026). Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning. Builds on 21 of its 26 references that are in the map. First posted on arXiv 24 Mar 2026.

Builds on

21 of 26 references.

Closest ancestors · 3

Same direction, newest first

  1. 2026RFEval: Benchmarking Reasoning Faithfulness under Counterfactual Reasoning Intervention in Large Reasoning Models19 Feb
  2. 2025Reasoning or Retrieval? A Study of Answer Attribution on Large Reasoning Models29 Sep
  3. 2025The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity7 Jun

Other references · 18

  1. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,302
  2. 2021Training Verifiers to Solve Math Word Problemscited by 9,743
  3. 2018Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challengecited by 4,710
  4. 2022Large Language Models are Zero-Shot Reasonerscited by 4,323
  5. 2015Character-level Convolutional Networks for Text Classificationcited by 2,293
  6. 2020What Disease does this Patient Have? A Large-scale Open Domain Question Answering Dataset from Medical Examscited by 1,324
  7. 2023Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Promptingcited by 1,081
  8. 2023Measuring Faithfulness in Chain-of-Thought Reasoningcited by 668
  9. 2020Towards Faithfully Interpretable NLP Systems: How should we define and evaluate faithfulness?cited by 590
  10. 2025Reasoning Models Don't Always Say What They Thinkcited by 388
  11. 2025Thought Anchors: Which LLM Reasoning Steps Matter?cited by 126
  12. 2024Making Reasoning Matter: Measuring and Improving Faithfulness of Chain-of-Thought Reasoningcited by 108
6 more
  1. 2025Measuring Chain of Thought Faithfulness by Unlearning Reasoning Stepscited by 43
  2. 2026Counterfactual Simulation Training for Chain-of-Thought Faithfulnesscited by 15
  3. 2025Can Aha Moments Be Fake? Towards Quantifying Decorative and True Thinking in Chain-of-Thoughtcited by 14
  4. 2026Mechanistic Evidence for Faithfulness Decay in Chain-of-Thought Reasoningcited by 9
  5. 2025Causal Consistency Regularization: Training Verifiably Sensitive Reasoning in Large Language Modelscited by 8
  6. 2026Lie to Me: How Faithful Is Chain-of-Thought Reasoning in Reasoning Models?cited by 8

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026Clinical Reasoning Graphs: Structured Evaluation of LLM Diagnostic Reasoning Reveals Competence Without Consistencycited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Does Inference Scaling Improve Reasoning Faithfulness? A Multi-Model Analysis of Self-Consistency Tradeoffs
  2. 2025Large Reasoning Models are not thinking straight: on the unreliability of thinking trajectories
  3. 2025Reasoning Relay: Evaluating Stability and Interchangeability of Large Language Models in Mathematical Reasoning
  4. 2025Theorem-of-Thought: A Multi-Agent Framework for Abductive, Deductive, and Inductive Reasoning in Language Models
  5. 2025Correct Reasoning Paths Visit Shared Decision Pivots
  6. 2023Two Failures of Self-Consistency in the Multi-Step Reasoning of LLMs
  7. 2026Knowing When Thinking Is Not Enough: Teaching Small Reasoning Models to Reason Beyond Their Parametric Knowledge
  8. 2026EvalQReason: A Framework for Step-Level Reasoning Evaluation in Large Language Models