mlsanityData to 14 Oct 2026

The Shape of Overthinking: Backtracking Bursts in Long Reasoning Traces

arXiv 2605.27965 ↗First posted 27 May 2026cs.AIcited by 2

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

Abstract (arXiv)

Reasoning models often generate long traces in which useful self-correction and unproductive revision are hard to distinguish. We study this distinction through backtracking dynamics: local reconsideration, retraction, or re-derivation inside long-form reasoning traces. On 6{,}000 Qwen3-8B AIME traces, we annotate segment-level backtrack severity and analyze event timing, normalized depth, and local burst structure. We find that early isolated repair is often compatible with correct reasoning, whereas incorrect traces more often show moderate-to-severe backtracks that persist and cluster late. Cross-corpus checks show the same qualitative asymmetry across additional model/domain pairs. Filtering analyses instantiate the signal as a prefix-causal selective early-exit policy: at shallow and intermediate depths, burst-aware filtering outperforms fixed length-based filtering while using only prefix-available features. Moderate length cutoffs remain strong completed-trace baselines, but burst-aware control provides a deployable mechanism for separating recoverable repair from likely instability.

Key facts. Cited by 2 papers in the map (data through 14 Oct 2026). Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning. Builds on 25 of its 32 references that are in the map. First posted on arXiv 27 May 2026.

Builds on

25 of 32 references.

Closest ancestors · 8

Same direction, newest first

  1. 2025Think Just Enough: Sequence-Level Entropy as a Confidence Signal for LLM Reasoning9 Oct
  2. 2025The Evolution of Thought: Tracking LLM Overthinking via Reasoning Dynamics Analysis25 Aug
  3. 2025FlashThink: An Early Exit Method For Efficient Reasoning20 May
  4. 2025Dynamic Early Exit in Reasoning Models22 Apr
  5. 2025SEAL: Steerable Reasoning Calibration of Large Language Models for Free7 Apr
  6. 2025Reasoning Models Know When They're Right: Probing Hidden States for Self-Verification7 Apr
  7. 2025L1: Controlling How Long A Reasoning Model Thinks With Reinforcement Learning6 Mar
  8. 2024Token-Budget-Aware LLM Reasoning24 Dec

Other references · 17

  1. 2023GPQA: A Graduate-Level Google-Proof Q&A Benchmarkcited by 2,879
  2. 2016Adaptive Computation Time for Recurrent Neural Networkscited by 506
  3. 2014Deeply-Supervised Netscited by 367
  4. 2020DeeBERT: Dynamic Early Exiting for Accelerating BERT Inferencecited by 332
  5. 2024LayerSkip: Enabling Early Exit Inference and Self-Speculative Decodingcited by 307
  6. 2020BERT Loses Patience: Fast and Robust Inference with Early Exitcited by 273
  7. 2020FastBERT: a Self-distilling BERT with Adaptive Inference Timecited by 226
  8. 2024Not All Layers of LLMs Are Necessary During Inferencecited by 96
  9. 2023SkipDecode: Autoregressive Skip Decoding with Batching and Caching for Efficient LLM Inferencecited by 89
  10. 2025S-GRPO: Early Exit via Reinforcement Learning in Reasoning Modelscited by 87
  11. 2023Fast and Robust Early-Exiting Framework for Autoregressive Language Models with Synchronized Parallel Decodingcited by 80
  12. 2024SWIFT: On-the-Fly Self-Speculative Decoding for LLM Inference Accelerationcited by 72
5 more
  1. 2024Kangaroo: Lossless Self-Speculative Decoding via Double Early Exitingcited by 58
  2. 2025Answer Convergence as a Signal for Early Stopping in Reasoningcited by 40
  3. 2024EE-Tuning: An Economical yet Scalable Solution for Tuning Early-Exit Large Language Modelscited by 13
  4. 2025HELIOS: Adaptive Model And Early-Exit Selection for Efficient LLM Inference Servingcited by 8
  5. 2024Dynamic Vocabulary Pruning in Early-Exit LLMscited by 4

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026CUSUM-Shaped Inference-Time Monitoring and Targeted Re-Decoding for Quantized Small Language Model Reasoningcited by 1
  2. 2026Trajectory-Retrieval Speculative Decoding: When Does a Model's Own History Help?cited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Decoding the Critique Mechanism in Large Reasoning Models
  2. 2026Mitigating Overthinking in Large Reasoning Language Models via Reasoning Path Deviation Monitoring
  3. 2025Reasoning-Finetuning Repurposes Latent Representations in Base Models
  4. 2026Failed Reasoning Traces Tell You What Is Fixable (But Not by Reading Them)
  5. 2025Tracing the Traces: Latent Temporal Signals for Efficient and Accurate Reasoning
  6. 2026Dissecting Failure Dynamics in Large Language Model Reasoning
  7. 2026Thinking Traps in Long Chain-of-Thought: A Measurable Study and Trap-Aware Adaptive Restart
  8. 2026Trace as State: Reasoning Traces as Conditional States for Long-Context Transformers