Thinking Traps in Long Chain-of-Thought: A Measurable Study and Trap-Aware Adaptive Restart
On the map: LLM reasoning → Test-time compute and efficient thinking
Abstract (arXiv)
Scaling test-time compute via Long Chain-of-Thought (Long-CoT) significantly enhances reasoning capabilities, yet extended generation does not guarantee correctness: after an early wrong commitment, models may keep elaborating a self-consistent but incorrect prefix. Through fine-grained trajectory analysis, we identify Thinking Traps, prefix-dominant deadlocks where later reflection, alternative attempts, or verification fails to revise the root error. On a curated subset of DAPO-MATH, 89% of failures exhibit such traps. To solve this problem, we introduce TAAR (Trap-Aware Adaptive Restart), a test-time control framework that trains a diagnostic policy to predict two signals from partial trajectories: a trap index for where to truncate and an escape probability for whether and how strongly to intervene. At inference time, TAAR truncates the trajectory before the predicted trap segment and adaptively restarts decoding; for severely trapped cases, it applies stronger perturbations, including higher-temperature resampling and an optional structured reboot suffix. Experiments on challenging mathematical and scientific reasoning benchmarks (AIME24, AIME25, GPQA-Diamond, HMMT25, BRUMO25) show that TAAR improves reasoning performance without fine-tuning base model parameters.
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 15 of its 29 references that are in the map. First posted on arXiv 17 Jan 2026.
Builds on
15 of 29 references.
Closest ancestors · 4
Same direction, newest first
- 2025SmartSwitch: Advancing LLM Reasoning by Overcoming Underthinking via Promoting Deeper Thought Exploration22 Oct
- 2025Do Thinking Tokens Help or Trap? Towards More Efficient Large Reasoning Model30 Jun
- 2025Learning from Peers in Reasoning Models12 May
- 2023Large Language Models Cannot Self-Correct Reasoning Yet3 Oct
Other references · 11
- 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,146
- 2022Self-Consistency Improves Chain of Thought Reasoning in Language Modelscited by 5,601
- 2023GPQA: A Graduate-Level Google-Proof Q&A Benchmarkcited by 2,748
- 2024OpenAI o1 System Cardcited by 1,220
- 2024Large Language Monkeys: Scaling Inference Compute with Repeated Samplingcited by 946
- 2025The Lessons of Developing Process Reward Models in Mathematical Reasoningcited by 402
- 2024Hallucination is Inevitable: An Innate Limitation of Large Language Modelscited by 329
- 2023How Language Model Hallucinations Can Snowballcited by 310
- 2025Thought Anchors: Which LLM Reasoning Steps Matter?cited by 126
- 2023GPT-4 Doesn't Know It's Wrong: An Analysis of Iterative Prompting for Reasoning Problemscited by 88
- 2024AutoCAP: Towards Automatic Cross-lingual Alignment Planning for Zero-shot Chain-of-Thoughtcited by 19
Built on by
Citing papers per year: 2026: 1.
1 paper.
Similar, not cited
Close in wording, no citation either way
- 2026Unveiling the Entropy Dynamics of Chain-of-Thought Reasoning
- 2026What If We Allocate Test-Time Compute Adaptively?
- 2026ThinkReset: Learnable Intermediate Interface Construction for Bounded-Context Long-Horizon Reasoning
- 2026The Shape of Overthinking: Backtracking Bursts in Long Reasoning Traces
- 2025Understanding and Steering the Cognitive Behaviors of Reasoning Models at Test-Time
- 2026Dissecting Failure Dynamics in Large Language Model Reasoning
- 2026The Tell-Tale Trace: Detecting Reasoning Failures in LLMs Using Chain-of-Thought Dynamics
- 2025Tracing the Traces: Latent Temporal Signals for Efficient and Accurate Reasoning