Structured Reasoning

We introduce CurveBench, a benchmark for hierarchical topological reasoning from visual input. CurveBench consists of \textbf{756 images} of pairwise non-intersecting Jordan curves across easy, polygonal, topographic-inspired, maze-like, and dense counting configurations. Each image is annotated with a rooted tree encoding the containment relations between planar regions. We formulate the task as structured prediction: given an image, a model must recover the full rooted containment tree induced by the curves. Despite the visual simplicity of the task, the strongest evaluated model, Gemini 3.1 Pro, achieves only \textbf{71.1%} tree-generation accuracy on CurveBench-Easy and \textbf{19.1%} on CurveBench-Hard. We further demonstrate benchmark utility through RLVR-style fine-tuning of open-weight vision-language models. Our trained Qwen3-VL-8B model improves over \texttt{Qwen-3-VL-8B-Thinking} from \textbf{2.8%} to \textbf{33.3%} tree-generation accuracy on CurveBench-Easy, exceeding GPT-5.4 and Claude Opus 4.5 under our evaluation protocol. The remaining gap, especially on CurveBench-Hard, shows that exact topology-aware visual reasoning remains far from solved.

Long chain-of-thought (Long CoT) reasoning improves performance on multi-step problems, but it also induces overthinking. This inefficiency is especially problematic in low-data fine-tuning regimes, where real applications adapt reasoning models with limited supervision and cannot rely on large-scale teacher distillation or heavy test-time control. To address this, we propose STOP (Structured On-policy Pruning), an on-policy algorithm for analyzing and pruning long-form reasoning traces. STOP constructs self-distilled traces from the model. Then it maps each trace into a structured reasoning interface through node segmentation, taxonomy annotation, and reasoning-tree construction. On top of this interface, we introduce ECN (Earliest Correct Node), which retains the shortest prefix ending at the earliest node. Experiments on DeepSeek-R1-Distill-Qwen-7B and DeepSeek-R1-Distill-LLaMA-3-8B across GSM8K, Math 500, and AIME 2024 show that STOP reduces generated tokens by 19.4% to 42.4% while largely preserving accuracy in low-data fine-tuning. Beyond efficiency, our analyses show that STOP induces much smaller distributional shift than teacher-guided pruning, improves the structural efficiency of generated reasoning, and reallocates reasoning effort away from redundant verification and backtracking toward more productive exploration.

Do transformers, when trained on sequential reasoning traces, build internal models of the underlying task? And if so, does the structure of those internal representations mirror the structure of the domain? We train an 8-layer transformer on Sudoku solving traces and perform a mechanistic analysis of its internal computation. We establish two results. First, the model builds a substructure world model: it does not represent the board state cell by cell, as a human analyst would expect, but organizes information around the rows, columns, and boxes that Sudoku's constraints act on. Second, we identify a naked-single circuit: a small set of dedicated neurons in the final MLP layer, each individually detecting when exactly one digit remains possible for a specific cell, and reliably promoting that digit. These findings show that the geometry of an emergent world model is shaped by the constraint algebra of the domain, not its surface presentation, and that the resulting decision circuit is sparse, monosemantic, and fully interpretable. More broadly, they demonstrate that mechanistic interpretability tools can recover an end-to-end algorithmic account of how a transformer solves a combinatorial reasoning task.

Humans intuitively solve complex problems by flexibly shifting among reasoning modes: they plan, execute, revise intermediate goals, resolve ambiguity through associative judgment, and apply formal procedures to well-specified subproblems. Current LLM agents lack this flexibility, as their scaffolds hard-code such reasoning decisions in advance. These scaffolds are effective when their prescribed structure matches the task, but brittle when solving the task requires adapting the structure of reasoning itself. We introduce Deep Reasoning -- an inference-time approach for constructing task-specific scaffolds through structured meta-reasoning. Deep Reasoning uses a formal language that represents meta-reasoning as executable decompositions over associative inference, formal computation, and recursive subproblem solving, enabling decomposition principles to be encoded as in-context examples that guide test-time scaffold construction. We instantiate this approach in a general-purpose agent (DOLORES) that distributes complex tasks across more controlled reasoning threads. We evaluate it against state-of-the-art scaffolding methods across four hard benchmarks: multi-hop reasoning, long-chain question answering, long-context aggregation, and deep research-style information seeking. DOLORES outperforms all evaluated scaffolds across three model sizes and two model families, improving over the strongest evaluated scaffold baseline by 24.8% on average. DOLORES distributes cognition across structured, lower-load reasoning threads, thereby reducing premature termination and hallucinations. This advantage can even bridge the scaling gap, with an 8B version surpassing all evaluated 32B baselines from the same family in more than half the settings. These results point toward future agentic systems that treat scaffolding as adaptive reasoning, constructing the structure each task requires just-in-time.

When distilling reasoning from large language models (LLMs) into smaller ones, teacher rationales for similar problems often vary wildly in structure and strategy. Like a chef who makes the same dish differently each time, this inconsistency burdens the student with noisy supervision that is hard to internalize. We propose Distillation through Reasoning Path Compression (D-RPC), which constrains the teacher to follow a compact, dynamically maintained bank of reusable high-level reasoning paths. For each training question, D-RPC retrieves the most relevant path and conditions the teacher to follow it, producing rationales that are consistent across similar problems yet diverse enough to cover different problem types. A PAC-Bayes analysis formalizes the resulting trade-off between bank size and coverage: smaller banks reduce supervision entropy but risk coverage gaps, and the generalization bound identifies an optimal intermediate size confirmed by our ablations. Across five math and commonsense reasoning benchmarks with two student models, D-RPC consistently outperforms chain-of-thought distillation, freeform rationale generation, direct distillation, and structured-supervision baselines, while using fewer tokens than template-heavy alternatives.

Human-factor event diagnosis is essential for learning from operational events in nuclear power plants, yet its quality depends strongly on expert interpretation of narrative reports and guideline-based reasoning.Existing data-driven or one-shot large language model approaches often lack structured reasoning, have limited alignment with formal diagnostic guidelines, and may generate logically inconsistent conclusions. To address this issue, this study proposes G-SHARE, a guideline-based structured reasoning framework that operationalizes the CNNP nine-step human-factor event diagnosis guideline into a multi-stage diagnostic pipeline.The framework consists of evidence extraction, stepwise diagnostic reasoning, and post-hoc consistency repair, enabling explicit use of report evidence, intermediate rationale generation, and logical validation of diagnostic outputs. A dataset of real human-factor event reports was constructed from Chinese nuclear industry sources, and a gold-standard subset annotated by domain experts was used for evaluation. Results show that G-SHARE substantially outperforms one-shot prompting and traditional machine learning baselines, with the strongest version achieving the best overall accuracy and macro-F1. Ablation results further indicate that structured reasoning and consistency enforcement are critical to robust diagnosis, especially under weak prompting conditions. The findings demonstrate the value of transforming expert diagnostic guidelines into auditable reasoning workflows, providing a practical pathway for intelligent human-factor analysis in safety-critical industries.

When a language model answers a table question, users have no way to verify which cells informed which reasoning steps. We introduce RSAT, a method that trains small language models (SLMs, 1-8B) to produce step-by-step reasoning with cell-level citations grounded in table evidence. Phase 1 (SFT) teaches a structured JSON output format from verified reasoning traces. Phase 2 (GRPO) optimizes a composite reward centered on NLI-based faithfulness, alongside citation validity and parsimony. Across six models from two families-Qwen 2.5 (1.5B/3B/7B) and Llama 3 (1B/3B/8B)-RSAT improves faithfulness 3.7$\times$ over SFT alone (0.224$\rightarrow$0.826), with near-perfect citation validity (0.992). Post-hoc attribution collapses below 13% format success, confirming that attribution must be integrated into reasoning, not retrofitted. Ablations show the faithfulness reward is essential: removing it drops faithfulness from 0.97 to 0.03.

In the LLM era, many symbolic and structured problems are presented to models through 1D text serialization. Yet some such problems are natively two-dimensional: their relevant relations, such as row--column correspondence or spatial adjacency, are defined by position in a 2D layout rather than by sequential order. This raises a representational question: does preserving the same symbolic entries in a 1D sequence also preserve the relational structure needed for computation? We study this issue through the lens of serialization friction: the representational mismatch in which the same underlying task instances and entries are still present, but relations that depend on layout become implicit under 1D serialization. The study uses a controlled synthetic testbed of three tasks: matrix transpose, Conway's Game of Life, and LU decomposition. In each task, the same instances are presented either as 1D text serialization or as their native 2D layout rendered as an image. Across this testbed, 1D serialization degrades more sharply as task size grows, and errors under serialization exhibit spatially structured patterns, suggesting that this presentation choice is consequential within our testbed. To further interpret these results, we add supplementary analyses that include a within-visual probe and an additional comparison of the two input presentations under the mixed-training transpose setting. These findings suggest that, for layout-defined tasks, reducing inputs to 1D serialization is not a neutral choice of representation.

Reasoning in Large Language Models incurs significant inference-time compute, yet the token-level information structure of reasoning traces remains underexplored. We observe that reasoning tokens split into two functional types: low-entropy \textit{structural} tokens (recurring phrases that scaffold the reasoning process) and higher-entropy \textit{organic} tokens (problem-specific content that drives toward a solution). This asymmetry motivates a simple, model-agnostic compression pipeline: apply cross-word BPE merges on a model's own reasoning traces to derive \textit{supertokens} that capture frequent structural patterns, then teach the model to adopt them via supervised fine-tuning. Across three model families and five mathematical reasoning benchmarks, our approach shortens reasoning traces by 8.1% on average with no statistically significant accuracy loss on any model--benchmark pair. Beyond compression, supertokens act as interpretable reasoning-move annotations (backtracking, verification, strategy shifts), exposing the model's high-level strategy at a glance. Analyzing transitions between structural categories reveals systematic differences between correct and incorrect traces: correct traces show productive recovery (backtracking followed by strategy shifts and verification), while incorrect traces are dominated by confusion cycles (repeated hedging and unresolved contradictions). These diagnostic signals suggest applications in reward shaping and early stopping for RL-based reasoning training.

We study clinical trial table reasoning, where answers are not directly stored in visible cells but must be reasoned from semantic understanding through normalization, classification, extraction, or lightweight domain reasoning. Motivated by the observation that current LLM approaches often suffer from "bad reasoning" under implicit planning assumptions, we focus on settings in which the model must recover implicit attributes such as therapy type, added agents, endpoint roles, or follow-up status from partially observed clinical-trial tables. We propose SCOPE (Structured Clinical hybrid Planning for Evidence retrieval in clinical trials), a multi-LLM planner-based framework that decomposes the task into row selection, structured planning, and execution. The planner makes the source field, reasoning rules, and output constraints explicit before answer generation, reducing ambiguity relative to direct prompting. We evaluate SCOPE on 1,500 hybrid reasoning questions over oncology clinical-trial tables against zero-shot, few-shot, chain-of-thought, TableGPT2, Blend-SQL, and EHRAgent. Results show that explicit multi-LLM planning improves accuracy for reasoning-based questions while offering a stronger accuracy-efficiency tradeoff than heavier agentic baselines. Our findings position clinical trial reasoning as a distinct table understanding problem and highlight hybrid planner-based decomposition as an effective solution

Purpose: Accurate assessment of the Critical View of Safety (CVS) during laparoscopic cholecystectomy is essential to prevent bile duct injury, a complication associated with significant morbidity and mortality. While large vision-language models (LVLMs) offer flexible reasoning, their predictions remain difficult to audit and unreliable on safety-critical surgical tasks. Methods: We introduce Sum-of-Checks, a framework that decomposes each CVS criterion into expert-defined reasoning checks reflecting clinically relevant visual evidence. Given a laparoscopic frame, an LVLM evaluates each check, producing a binary judgment and justification. Criterion-level scores are computed via fixed, weighted aggregation of check outcomes. We evaluate on the Endoscapes2023 benchmark using three frontier LVLMs, comparing against direct prompting, chain-of-thought, and sub-question decomposition, each with and without few-shot examples. Results: Sum-of-Checks improves average frame-level mean average precision by 12--14% relative to the best baseline across all three models and criteria. Analysis of individual checks reveals that LVLMs are reliable on observational checks (e.g., visibility, tool obstruction) but show substantial variability on decision-critical anatomical evidence. Conclusion: Structuring surgical reasoning into expert-aligned verification checks improves both accuracy and transparency of LVLM-based CVS assessment, demonstrating that explicitly separating evidence elicitation from decision-making is critical for reliable and auditable surgical AI systems. Code is available at https://github.com/BrachioLab/SumOfChecks.

Large language models perform well on many reasoning tasks, yet they often lack awareness of whether their current knowledge or reasoning state is complete. In non-interactive puzzle settings, the narrative is fixed and the underlying structure is hidden; once a model forms an early hypothesis under incomplete premises, it can propagate that error throughout the reasoning process, leading to unstable conclusions. To address this issue, we propose SABA, a reasoning framework that explicitly introduces self-awareness of missing premises before making the final decision. SABA formulates reasoning as a recursive process that alternates between structured state construction and obstacle resolution: it first applies Information Fusion to consolidate the narrative into a verifiable base state, and then uses Query-driven Structured Reasoning to identify and resolve missing or underspecified premises by turning them into queries and progressively completing the reasoning state through hypothesis construction and state refinement. Across multiple evaluation metrics, SABA achieves the best performance on all three difficulty splits of the non-interactive Detective Puzzle benchmark, and it also maintains leading results on multiple public benchmarks.

Many real-world questions appear deceptively simple yet implicitly demand two capabilities: (i) systematic coverage of a bounded knowledge universe and (ii) compositional set-based reasoning over that universe, a phenomenon we term "the tip of the iceberg." We formalize this challenge through two orthogonal dimensions: knowledge width, the cardinality of the required universe, and reasoning depth, the number of compositional set operations. We introduce KnowledgeBerg, a benchmark of 4,800 multiple-choice questions derived from 1,183 enumeration seeds spanning 10 domains and 17 languages, with universes grounded in authoritative sources to ensure reproducibility. Representative open-source LLMs demonstrate severe limitations, achieving only 5.26-36.88 F1 on universe enumeration and 16.00-44.19 accuracy on knowledge-grounded reasoning. Diagnostic analyses reveal three stages of failure: completeness, or missing knowledge; awareness, or failure to identify requirements; and application, or incorrect reasoning execution. This pattern persists across languages and model scales. Although test-time compute and retrieval augmentation yield measurable gains -- up to 4.35 and 3.78 points, respectively -- substantial gaps remain, exposing limitations in how current LLMs organize structured knowledge and execute compositional reasoning over bounded domains. The dataset is available at https://huggingface.co/datasets/2npc/KnowledgeBerg

Recent advances in large language models (LLMs) have shown that test-time scaling can substantially improve model performance on complex tasks, particularly in the coding domain. Under this paradigm, models use a larger token budget during inference to generate intermediate reasoning traces before producing a final answer. However, current evaluations primarily rely on competitive programming benchmarks, which may not capture the full range of reasoning abilities. In this work, we perform a systematic study of frontier reasoning models to understand their performance on real-world coding benchmarks. To gain more insights into the performance of such models, we devise a programmatic way to {\em automatically generate} coding tasks of arbitrary difficulty and structure from existing benchmarks. Using this framework, our analysis reveals that the structure of a reasoning trace, not just its contents, is a strong predictor of correctness. Motivated by this, we propose structured thought-trees as means to represent reasoning traces. To illustrate their use, we train a lightweight classifier on features extracted from thought-trees to predict trace correctness, and demonstrate that flagging and retrying structurally anomalous traces based on the extracted features yields consistent gains at lower complexity levels.

Large language models exhibit systematic limitations in structured logical reasoning: they conflate hypothesis generation with verification, cannot distinguish conjecture from validated knowledge, and allow weak reasoning steps to propagate unchecked through inference chains. We present a symbolic reasoning scaffold that operationalizes Peirce's tripartite inference -- abduction, deduction, and induction -- as an explicit protocol for LLM-assisted reasoning. The framework enforces logical consistency through five algebraic invariants (the Gamma Quintet), the strongest of which -- the Weakest Link bound -- ensures that no conclusion in a reasoning chain can exceed the reliability of its least-supported premise. This principle, independently grounded as weakest link resolution in possibilistic logic and empirically validated for chain-of-thought reasoning, prevents logical inconsistencies from accumulating across multi-step inference. We verify all invariants through a property-based testing suite of 100 properties and 16 fuzz tests over 10^5+ generated cases, providing a verified reference implementation of the invariants suitable as a foundation for future reasoning benchmarks.

Large language models (LLMs) have become increasingly used for various tasks, often coupled with Chain-of-Thought (CoT) prompting to boost accuracy. Recent work has shown that high label-prediction accuracy does not guarantee correct intermediate reasoning, and the causes of reasoning flaws vary from sample to sample, yet existing remedies either focus on a single domain or assume that one flaw type applies uniformly across samples. A simple mitigation method is to provide the model with the correct answer, but we show that this yields no consistent improvement in reasoning quality. This indicates that the problem cannot be fixed by LLMs' awareness of answers, and must instead be addressed through the structure of reasoning. Motivated by this, we propose CRAFT (Consensus Reasoning-knowledge-graph Aggregation for Flaw-aware Trace synthesis), which aggregates the consensus components shared across multiple candidate reasoning traces to synthesize improved ones. CRAFT consistently improves label-prediction accuracy on both logical and mathematical reasoning benchmarks, outperforming most baselines, while its post-processed traces achieve higher quality under fine-grained benchmark evaluation.

This paper asks whether a bounded neural architecture can exhibit a meaningful division of labor between intuition and deliberation on a classic 64-item syllogistic reasoning benchmark. More broadly, the benchmark is relevant to ongoing debates about world models and multi-stage reasoning in AI. It provides a controlled setting for testing whether a learned system can develop structured internal computation rather than only one-shot associative prediction. Experiment 1 evaluates a direct neural baseline for predicting full 9-way human response distributions under 5-fold cross-validation. Experiment 2 introduces a bounded dual-path architecture with separate intuition and deliberation pathways, motivated by computational mental-model theory (Khemlani & Johnson-Laird, 2022). Under cross-validation, bounded intuition reaches an aggregate correlation of r = 0.7272, whereas bounded deliberation reaches r = 0.8152, and the deliberation advantage is significant across folds (p = 0.0101). The largest held-out gains occur for NVC, Eca, and Oca, suggesting improved handling of rejection responses and c-a conclusions. A canonical 80:20 interpretability run and a five-seed stability sweep further indicate that the deliberation pathway develops sparse, differentiated internal structure, including an Oac-leaning state, a dominant workhorse state, and several weakly used or unused states whose exact indices vary across runs. These findings are consistent with reasoning-like internal organization under bounded conditions, while stopping short of any claim that the model reproduces full sequential processes of model construction, counterexample search, and conclusion revision.

Standard Transformers have a fixed computational depth, limiting their ability to generalize to tasks that require variable-depth reasoning. The usual remedy, Chain-of-Thought (CoT), spends tokens to reason, inflating the key--value cache and making latency grow with the step count, so memory becomes the limiting cost when reasoning is served over large query batches. We study a depth-recurrent Transformer that decouples computational depth from parameter count by iterating a shared-weight block, so that each added reasoning step costs flat memory and linear latency, with no token generation. Three ingredients keep the recurrence stable for 20+ thinking steps: a silent thinking objective that supervises only the final output, LayerScale initialization, and an identity-biased gate that opens a gradient highway across steps. We characterize it on three compositional domains with decreasing structural bias: graph reachability (adjacency masking), nested boolean logic (relative positioning), and unstructured relational text (no positional cue). We find a \emph{computational frontier}: accuracy climbs once the thinking-step count meets the task's complexity, reaching near-perfect performance on the two structured tasks and a lower plateau on unstructured text. How it climbs depends on the structural bias---abruptly from chance on the graph task, gradually on the other two. Depth recurrence extrapolates beyond the training range: it succeeds on the graph task where fixed-depth models barely extrapolate, and on the two sequence tasks comes within two points of fixed-depth Transformers that use $4$--$6.4\times$ more parameters. On the graph task, whose adjacency mask makes propagation depth verifiable, intermediate per-step supervision---a standard recipe for deep iterative models---consistently \emph{harms} this extrapolation. We release the code for reproducibility.

We create high-quality datasets for LLM evaluation of logical reasoning skills across nine different languages, which have been manually checked by fluent speakers. The datasets consist of so-called zebra puzzles, and we analyse different ways of tuning the difficulty of the puzzles to fit modern LLMs. This includes the size of the puzzle (number of objects and number of clues), as well as a novel addition of red herring clues containing only irrelevant information. We show that presence of red herrings indeed makes the puzzles significantly harder for the models, and we find puzzle sizes 2x3 and 4x5 are sufficiently challenging for GPT-4o mini (a non-reasoning model) and o3-mini (a reasoning model), respectively. We analyse whether LLM performance of these are sensitive to the language, the cultural sensitivity of the puzzle theme, and the choice of clue types. These analyses are conducted with English and Danish, where we show that there is no significant difference for either of these three aspects, at least for the OpenAI models GPT-4o mini and o3-mini, chosen as representative non-reasoning and reasoning models, respectively. We publish the datasets for each of the nine languages for the identified sizes 2x3 and 4x5. We also publish the code used to generate the puzzles, which can be used to extend the benchmark into more languages.

Many LLMs plan before they act, yet planning and execution are often still entangled in one long generation trace, enforced only through prompts, or split across separate components. We argue that these two stages call for different computation: planning benefits from diversity and breadth, whereas execution demands precision and faithful adherence to a chosen strategy. Treating them as a single undifferentiated chain wastes tokens on routine derivation and makes it costly to explore alternative strategies at test time. We present the \textbf{Explore-Execute Chain (E\textsuperscript{2}C)}, which keeps both stages in one model but separates them structurally: a stochastic \textit{Exploration} phase drafts a concise high-level plan, and a deterministic \textit{Execution} phase carries it out. Causal SFT and RL train this split so that exploration stays informative and execution remains plan-faithful. Once plans are short yet decisive, extra inference compute can be directed to exploration rather than to repeatedly decoding full solutions. On AIME'2024 at $K{=}32$, \textbf{E\textsuperscript{2}C-ReAct Loop} reaches 53.3% accuracy with only 12.4k tokens, outperforming Tree-of-Thoughts ($N{=}32$: 50.0%, 71.3k). The same structure also supports lightweight domain adaptation: \textbf{Exploration-Focused SFT (EF-SFT)} updates only the planning phase, uses 3.5% of the tokens required by standard SFT, and improves medical benchmark accuracy by up to 14.5%.

In this paper, we propose a modified version of the MedQA-USMLE dataset, named MEDQA-OPEN, which contains open-ended medical questions without options to mimic clinical scenarios, along with clinician-approved reasoned answers. Additionally, we implement a prompt driven by Chain of Thought (CoT) reasoning, CLINICR, to mirror the prospective process of incremental reasoning, reaching a correct response to medical questions. We empirically demonstrate how CLINICR outperforms the state-of-the-art 5-shot CoT-based prompt (Liévin et al., 2022). We also present an approach that mirrors real-life clinical practice by first exploring multiple differential diagnoses through MCQ-CLINICR and subsequently narrowing down to a final diagnosis using MCQ-ELIMINATIVE. Finally, emphasizing the importance of response verification in medical settings, we utilize a reward model mechanism, replacing the elimination process performed by MCQ-ELIMINATIVE.

Large reasoning models (LRMs) produce reasoning traces with non-linear structures, such as backtracking and self-correction, that complicate the evaluation and monitoring of the reasoning process. We introduce ReasoningFlow, a framework that captures the discourse structures of LRM reasoning traces into fine-grained directed acyclic graphs (DAGs). We develop and validate our annotation schema through careful manual annotation of 31 traces (2.1k steps), achieving high inter-annotator agreement, then scale to automatic annotation of 1,260 traces (247.7k steps) spanning three tasks (math, science, argumentation) and five models (Qwen2.5-32B-Inst, QwQ-32B, DeepSeek-V3, DeepSeek-R1, GPT-oss-120B). By analyzing ReasoningFlow graphs, we find: (1) LRMs exhibit structurally similar traces, despite being trained from different base models and potentially non-overlapping post-training data. (2) ReasoningFlow reveals diverse fine-grained reasoning behaviors (e.g., local verification, self-reflection, and assumptions) that can be used for better reasoning trace monitorability. (3) In LRMs, most of the erroneous steps are not used to derive final answers. (4) Mechanistic causal dependencies between steps do not reflect the language-level discourse structure. We release the dataset and code in: https://github.com/jinulee-v/reasoningflow.