mlsanityData to 14 Oct 2026

REWIND: Uncertainty-Guided Trajectory Repair for Test-Time Reasoning

arXiv 2610.19561 ↗First posted 14 Oct 2026cs.CL cs.AIcited by 0

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

Abstract (arXiv)

An early mistake in the generated reasoning can constrain all subsequent reasoning. Repeated-generation methods, like MCMC-based approaches, attempt to improve output quality by sampling multiple generations, but their effectiveness can be limited by randomly selecting where to regenerate. On MATH500 with Qwen2.5-Math-7B, we find that incorrect responses have substantially higher mean token entropy than correct responses, with an average within-question relative increase of approximately 112%, suggesting that entropy can signal reasoning errors. Motivated by this finding, we introduce REWIND, a training-free controller that uses entropy dynamics to propose targeted suffix regenerations. Across three model families and five benchmarks, REWIND outperforms all baselines, with gains of 0.51-10.00 percentage points over the strongest baseline. These results demonstrate the effectiveness of targeted, uncertainty-guided trajectory repair across models and tasks. Our code is available at https://anonymous.4open.science/r/rewind-D310.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning. Builds on 14 of its 25 references that are in the map. First posted on arXiv 14 Oct 2026.

Builds on

14 of 25 references.

Closest ancestors · 4

Same direction, newest first

  1. 2025Entropy-based Exploration Conduction for Multi-step Reasoning20 Mar
  2. 2023Large Language Models Cannot Self-Correct Reasoning Yet3 Oct
  3. 2023Let's Sample Step by Step: Adaptive-Consistency for Efficient Reasoning and Coding with LLMs19 May
  4. 2022Language Models (Mostly) Know What They Know11 Jul

Other references · 10

  1. 2021Measuring Mathematical Problem Solving With the MATH Datasetcited by 5,371
  2. 2023GPQA: A Graduate-Level Google-Proof Q&A Benchmarkcited by 2,879
  3. 2022Challenging BIG-Bench Tasks and Whether Chain-of-Thought Can Solve Themcited by 1,807
  4. 2021An Explanation of In-context Learning as Implicit Bayesian Inferencecited by 938
  5. 2023Language Agent Tree Search Unifies Reasoning Acting and Planning in Language Modelscited by 574
  6. 2025Beyond the 80/20 Rule: High-Entropy Minority Tokens Drive Effective Reinforcement Learning for LLM Reasoningcited by 559
  7. 2023Faith and Fate: Limits of Transformers on Compositionalitycited by 503
  8. 2018CGMH: Constrained Sentence Generation by Metropolis-Hastings Samplingcited by 94
  9. 2024Forking Paths in Neural Text Generationcited by 34
  10. 2023A Block Metropolis-Hastings Sampler for Controllable Energy-based Text Generationcited by 6

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025Entropy-Guided Loop: Achieving Reasoning through Uncertainty-Aware Generation
  2. 2026Resample the Answer: Where to Spend Test-Time Compute in Recursive Reasoning
  3. 2026Beyond Imitation: Auditing the Recoverability of Reasoning in Distilled Models
  4. 2025Entropy-Gated Branching for Efficient Test-Time Reasoning
  5. 2026SPREG: Structured Plan Repair with Entropy-Guided Test-Time Intervention for Large Language Model Reasoning
  6. 2026ReTreVal: Reasoning Tree with Validation and Cross-Problem Memory for Large Language Models
  7. 2025CURE: Critical-Token-Guided Re-Concatenation for Entropy-Collapse Prevention
  8. 2025Think Just Enough: Sequence-Level Entropy as a Confidence Signal for LLM Reasoning