The Shape of Overthinking: Backtracking Bursts in Long Reasoning Traces

• URL: http://arxiv.org/abs/2605.27965v1
• Date: Wed, 27 May 2026 05:01:04 GMT
• Title: The Shape of Overthinking: Backtracking Bursts in Long Reasoning Traces
• Abstract summary: We study local reconsideration, retraction, or re-derivation inside long-form reasoning traces.<n>On 6,000 Qwen3-8B AIME traces, we annotate segment-level backtrack severity and analyze event timing, normalized depth, and local burst structure.<n>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.
• Score: 2.102532345063307
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: 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.

Related papers

• One Step, One Lead: Mitigating Higher-Order Interference in Multi-Domain Reinforcement Learning via Cross-Step Control [62.82372407840088]
  We show that same-point domain gradients may remain nearly even when consecutive realized updates partially reverse one another in output space.<n>We propose OSOL, which designates a focus domain at each iteration, to rank token-level rebound risk.<n>OSOL reaches a domain-macro average of 0.4822, improving by 5.7% over the strongest compared baseline, without explicit higher-order differentiation.
  arXiv  Detail & Related papers  (2026-09-06T08:32:59Z)
• Residual Dominance as a Structural Account of Last-Item Reliance in Causal Self-Attention Recommenders [47.8060666024071]
  We show that SASRec-style models exhibit highly localized last-item reliance.<n>We then find that residual addition sharply shifts the full-block representation toward same-position contributions.<n>Changing the residual strength induces a monotonic trade-off between structural mixing and last-item reliance.
  arXiv  Detail & Related papers  (2026-08-14T07:09:58Z)
• AnchorMoE: Interpretable Time Series Classification via Anchor-Routed MoE [51.994752158953084]
  AnchorMoE is an interpretable-by-construction classification framework.<n>It encodes multi-view representations of local patches and routes them to specialized experts.<n>Experiments on real-world and synthetic benchmarks demonstrate that AnchorMoE achieves highly competitive classification performance.
  arXiv  Detail & Related papers  (2026-06-02T13:30:54Z)
• Clipping Bottleneck: Stabilizing RLVR via Stochastic Recovery of Near-Boundary Signals [83.0127582612634]
  Near-boundary Rescue (NSR) is a minimal, plug-and-play modification that retains out-of-bound tokens to recover lost signals.<n>NSR substantially improves training stability and delivers consistent gains over strong baselines such as DAPO and GSPO.
  arXiv  Detail & Related papers  (2026-05-21T16:45:31Z)
• STOP: Structured On-Policy Pruning of Long-Form Reasoning in Low-Data Regimes [13.293115227628775]
  Long chain-of-thought (Long CoT) reasoning improves performance on multi-step problems, but it also induces overthinking.<n>We propose STOP (Structured On-policy Pruning), an on-policy algorithm for analyzing and pruning long-form reasoning traces.
  arXiv  Detail & Related papers  (2026-05-13T08:28:05Z)
• Affordance Agent Harness: Verification-Gated Skill Orchestration [45.231685718099264]
  Affordance grounding requires identifying where and how an agent should interact in open-world scenes.<n>We propose Affordance Agent Harness, a closed-loop runtime that unifies skills with an evidence store and cost control.
  arXiv  Detail & Related papers  (2026-05-01T13:45:16Z)
• Surprised by Attention: Predictable Query Dynamics for Time Series Anomaly Detection [1.7751300245073598]
  AxonAD is an unsupervised detector that treats multi-head attention query evolution as a short horizon predictable process.<n>A tail-aggregated query mismatch score measures cosine deviation between predicted and target queries on recent timesteps.<n>On in-vehicle telemetry with interval annotations, AxonAD improves ranking quality and temporal localization over strong baselines.
  arXiv  Detail & Related papers  (2026-03-13T11:40:51Z)
• Constraint-Rectified Training for Efficient Chain-of-Thought [60.52883907721588]
  Chain-of-Thought (CoT) has significantly enhanced the reasoning capabilities of Large Language Models (LLMs)<n>While longer reasoning traces can improve answer quality and unlock abilities such as self-correction, they also incur high inference costs and often introduce redundant steps, known as overthinking.<n>Recent research seeks to develop efficient reasoning strategies that balance reasoning length and accuracy.
  arXiv  Detail & Related papers  (2026-02-13T02:13:45Z)
• LEFT: Learnable Fusion of Tri-view Tokens for Unsupervised Time Series Anomaly Detection [53.191369031661885]
  Unsupervised time series anomaly detection aims to build a model for identifying abnormal timestamps without assuming the availability of annotations.<n>We present Learnable Fusion of Tri-view Tokens (LEFT), a unified unsupervised TSAD framework that models anomalies as inconsistencies across complementary representations.<n>Experiments on real-world benchmarks show that LEFT yields the best detection accuracy against SOTA baselines, while achieving a 5x reduction on FLOPs and 8x speed-up for training.
  arXiv  Detail & Related papers  (2026-02-09T13:33:49Z)
• APR: Penalizing Structural Redundancy in Large Reasoning Models via Anchor-based Process Rewards [61.52322047892064]
  Test-Time Scaling (TTS) has significantly enhanced the capabilities of Large Reasoning Models (LRMs)<n>We observe that LRMs frequently conduct repetitive self-verification without revision even after obtaining the final answer during the reasoning process.<n>We propose Anchor-based Process Reward (APR), a structure-aware reward shaping method that localizes the reasoning anchor and penalizes exclusively the post-anchor AST.
  arXiv  Detail & Related papers  (2026-01-31T14:53:20Z)
• Thinking Traps in Long Chain-of-Thought: A Measurable Study and Trap-Aware Adaptive Restart [27.904791075662896]
  We introduce TAAR (Trap-Aware Adaptive Restart), a test-time control framework that trains a diagnostic policy to predict two signals from partial trajectories.<n>At inference time, TAAR truncates the trajectory before the predicted trap segment and adaptively restarts decoding.<n>Experiments show that TAAR improves reasoning performance without fine-tuning base model parameters.
  arXiv  Detail & Related papers  (2026-01-17T07:26:02Z)
• Unstable Prompts, Unreliable Segmentations: A Challenge for Longitudinal Lesion Analysis [0.5537760992845262]
  This paper investigates the performance of the ULS23 segmentation model in a longitudinal context.<n>We identify two critical, interconnected failure modes: a sharp degradation in segmentation quality in follow-up cases due to inter-scan registration errors, and a subsequent breakdown of the lesion correspondence process.
  arXiv  Detail & Related papers  (2025-07-25T12:55:48Z)
• Self-Supervised Training with Autoencoders for Visual Anomaly Detection [61.62861063776813]
  We focus on a specific use case in anomaly detection where the distribution of normal samples is supported by a lower-dimensional manifold. We adapt a self-supervised learning regime that exploits discriminative information during training but focuses on the submanifold of normal examples. We achieve a new state-of-the-art result on the MVTec AD dataset -- a challenging benchmark for visual anomaly detection in the manufacturing domain.
  arXiv  Detail & Related papers  (2022-06-23T14:16:30Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.