Daily Papers

Large Language Models (LLMs) can enhance reasoning capabilities through test-time scaling by generating multiple traces. However, the combination of lengthy reasoning traces with multiple sampling introduces substantial computation and high end-to-end latency. Prior work on accelerating this process has relied on similarity-based or confidence-based pruning, but these signals do not reliably indicate trace quality. To address these limitations, we propose STEP: Step-level Trace Evaluation and Pruning, a novel pruning framework that evaluates reasoning steps using hidden states and dynamically prunes unpromising traces during generation. We train a lightweight step scorer to estimate trace quality, and design a GPU memory-aware pruning strategy that triggers pruning as the GPU memory is saturated by KV cache to reduce end-to-end latency. Experiments across challenging reasoning benchmarks demonstrate that STEP reduces end-to-end inference latency by 45%-70% on average compared to self-consistency while also improving reasoning accuracy. Our code is released at: https://github.com/Supercomputing-System-AI-Lab/STEP

Large language models are increasingly used as chemistry assistants, yet most chemistry benchmarks still score only final answers. This masks a critical failure mode: a model may output the correct molecule, product, or option while its reasoning violates chemical logic. Existing process-level evaluators are hard to scale because LLM judges and human step-level process annotation are costly, inconsistent, and vulnerable to hallucination. We introduce ChemCoTBench-V2, a rule-verifiable diagnostic benchmark for low-cost, auditable evaluation of structured, verifier-addressable chemical reasoning traces. It spans molecular understanding, molecule editing, molecular optimization, and reaction prediction, with 5,620 evaluation samples across 18 reporting tasks. Models must expose key intermediate steps in expert-designed templates, and those steps are checked with deterministic chemistry rules and, for closed-answer tasks, reference traces rather than another LLM judge. Open-ended molecular optimization is evaluated with oracle-verifiable state constraints rather than strict trace matching. The benchmark reports three separate signals: final-answer correctness, template adherence, and step-wise verifier correctness over expert-refined intermediate commitments. Experiments on frontier models reveal a persistent gap between final-answer success and structured-reasoning-state consistency: models often follow the requested format while failing chemical-step checks, or answer correctly with weak supporting reasoning. ChemCoTBench-V2 enables fine-grained model comparison and identifies the concrete step at which the trace first violates the verifier.

Multimodal large language models (MLLMs) have shown promising reasoning abilities, yet evaluating their performance in specialized domains remains challenging. STEM reasoning is a particularly valuable testbed because it provides highly verifiable feedback, but existing benchmarks often permit unimodal shortcuts due to modality redundancy and focus mainly on final-answer accuracy, overlooking the reasoning process itself. To address this challenge, we introduce StepSTEM: a graduate-level benchmark of 283 problems across mathematics, physics, chemistry, biology, and engineering for fine-grained evaluation of cross-modal reasoning in MLLMs. StepSTEM is constructed through a rigorous curation pipeline that enforces strict complementarity between textual and visual inputs. We further propose a general step-level evaluation framework for both text-only chain-of-thought and interleaved image-text reasoning, using dynamic programming to align predicted reasoning steps with multiple reference solutions. Experiments across a wide range of models show that current MLLMs still rely heavily on textual reasoning, with even Gemini 3.1 Pro and Claude Opus 4.6 achieving only 38.29% accuracy. These results highlight substantial headroom for genuine cross-modal STEM reasoning and position StepSTEM as a benchmark for fine-grained evaluation of multimodal reasoning. Source code is available at https://github.com/lll-hhh/STEPSTEM.

Process Reward Models (PRMs) have emerged as a promising framework for supervising intermediate reasoning in large language models (LLMs), yet existing PRMs are primarily trained on general or Science, Technology, Engineering, and Mathematics (STEM) domains and fall short in domain-specific contexts such as finance, where reasoning is more structured, symbolic, and sensitive to factual and regulatory correctness. We introduce Fin-PRM, a domain-specialized, trajectory-aware PRM tailored to evaluate intermediate reasoning steps in financial tasks. Fin-PRM integrates step-level and trajectory-level reward supervision, enabling fine-grained evaluation of reasoning traces aligned with financial logic. We apply Fin-PRM in both offline and online reward learning settings, supporting three key applications: (i) selecting high-quality reasoning trajectories for distillation-based supervised fine-tuning, (ii) providing dense process-level rewards for reinforcement learning, and (iii) guiding reward-informed Best-of-N inference at test time. Experimental results on financial reasoning benchmarks, including CFLUE and FinQA, demonstrate that Fin-PRM consistently outperforms general-purpose PRMs and strong domain baselines in trajectory selection quality. Downstream models trained with Fin-PRM yield substantial improvements with baselines, with gains of 12.9\% in supervised learning, 5.2\% in reinforcement learning, and 5.1\% in test-time performance. These findings highlight the value of domain-specialized reward modeling for aligning LLMs with expert-level financial reasoning. Our project resources will be available at https://github.com/aliyun/qwen-dianjin.

Vision-Language Models (VLMs) have advanced across multimodal benchmarks but still show clear gaps in ordinal number understanding, i.e., the ability to track relative positions and generalize to large indices. We present OrdinalBench, a diagnostic benchmark that standardizes ordinal number understanding as an evaluation task for VLMs. The core task is N-th object identification, defined by a starting reference and traversal rule. Task difficulty is controlled along three axes: (i) ordinal magnitude, from small numbers to extreme cases up to 300; (ii) arrangement complexity, from single loops to maze-like paths; and (iii) object count. The benchmark provides 39,000 question-answer pairs, each annotated with a ground-truth reasoning trajectory and balanced across difficulty levels for controlled large-scale testing. Beyond answer-only evaluation, our framework requires models to generate structured stepwise traces of the counting process and provides an open evaluation toolkit that measures both final accuracy and step-level path consistency. Zero-shot evaluations of GPT-5, Gemini 2.5 Flash Lite, Qwen2.5-VL, InternVL3.5, and Molmo reveal sharp degradation under large-ordinal and complex-path conditions, highlighting weak generalization despite strong scores on standard multimodal tasks. By framing ordinal number understanding as a core target, OrdinalBench provides a reproducible benchmark and diagnostic framework for developing VLMs with stronger sequential reasoning. All data and code are available at https://ordinalbench.github.io/

Step-by-step reasoning is widely used to enhance the reasoning ability of large language models (LLMs) in complex problems. Evaluating the quality of reasoning traces is crucial for understanding and improving LLM reasoning. However, the evaluation criteria remain highly unstandardized, leading to fragmented efforts in developing metrics and meta-evaluation benchmarks. To address this gap, this survey provides a comprehensive overview of step-by-step reasoning evaluation, proposing a taxonomy of evaluation criteria with four top-level categories (groundedness, validity, coherence, and utility). We then categorize metrics based on their implementations, survey which metrics are used for assessing each criterion, and explore whether evaluator models can transfer across different criteria. Finally, we identify key directions for future research.

LLM web agents now browse and take actions on the open web, yet current agent evaluations are constrained to sandboxed environments or artificial tasks. We introduce BrowserArena, a live open-web agent evaluation platform that collects user-submitted tasks, runs Arena-style head-to-head comparisons, and uses step-level human feedback to surface failure modes. Collecting and analyzing step-level annotations on the agent traces, we identify three consistent failure modes: captcha resolution, pop-up banner removal, and direct navigation to URLs. By constructing targeted datasets to further study these tasks, we discover variations in how different language models navigate these failure modes. We find, for example, that o4-mini deploys a wider variety of strategies to circumvent captcha resolution than other models and DeepSeek-R1 consistently misleads users about pop-up banner closure. Our findings surface both the diversity and brittleness of current web agents. More broadly, our benchmarking methodology provides an approach to evaluating and understanding web agent failure modes at scale.

Existing agent safety benchmarks report binary accuracy, conflating early intervention with post-mortem analysis. A detector that flags a violation at step 8 enables intervention; one that reports it at step 48 provides only forensic value. This distinction is critical, yet current benchmarks cannot measure it. We introduce StepShield, the first benchmark to evaluate when violations are detected, not just whether. StepShield contains 9,213 code agent trajectories, including 1,278 meticulously annotated training pairs and a 7,935-trajectory test set with a realistic 8.1% rogue rate. Rogue behaviors are grounded in real-world security incidents across six categories. We propose three novel temporal metrics: Early Intervention Rate (EIR), Intervention Gap, and Tokens Saved. Surprisingly, our evaluation reveals that an LLM-based judge achieves 59% EIR while a static analyzer achieves only 26%, a 2.3x performance gap that is entirely invisible to standard accuracy metrics. We further show that early detection has direct economic benefits: our cascaded HybridGuard detector reduces monitoring costs by 75% and projects to $108M in cumulative savings over five years at enterprise scale. By shifting the focus of evaluation from whether to when, StepShield provides a new foundation for building safer and more economically viable AI agents. The code and data are released under an Apache 2.0 license.

Users rely on execution traces to observe agent behavior, diagnose failures, and ensure accountability. These traces contain rich procedural detail, including tool invocations, intermediate decisions, and error-recovery logic. Yet this detail can expose private procedural skills, allowing downstream methods to recover key formulas, thresholds, and strategies without access to model weights or skill files. To quantify this risk and evaluate protection, we construct CapTraceBench, a benchmark of 75 specialized long-horizon tasks and 154 curated skills across seven domains. We also introduce RedAct https://github.com/XuShuwenn/RedAct, a protected trace release framework that localizes protected key information, rewrites traces while preserving verifier-critical evidence, and embeds behavioral watermarks for downstream provenance analysis. Across representative trace reuse methods, RedAct reduces normalized skill transfer (NST) from 44.7--67.1\% on raw traces to below the no-skill baseline, while preserving audit evidence. Its standalone behavioral watermarks reach 93.6--100.0\% true detection with a false alarm rate of at most 1.9\%. These results frame public agent traces as security interfaces and show that selective redaction can reduce procedural capability leakage without removing audit evidence.

Large language model (LLM) agents now execute long, tool-using tasks where final outcome checks can arrive too late for intervention. Online warning requires lightweight prefix monitors over heterogeneous traces, but hand-authored event schemas are brittle and deployment-time LLM judging is costly. We introduce PrefixGuard, a trace-to-monitor framework with an offline StepView induction step followed by supervised monitor training. StepView induces deterministic typed-step adapters from raw trace samples, and the monitor learns an event abstraction and prefix-risk scorer from terminal outcomes. Across WebArena, τ^2-Bench, SkillsBench, and TerminalBench, the strongest PrefixGuard monitors reach 0.900/0.710/0.533/0.557 AUPRC. Using the strongest backend within each representation, they improve over raw-text controls by an average of +0.137 AUPRC. LLM judges remain substantially weaker under the same prefix-warning protocol. We also derive an observability ceiling on score-based area under the precision-recall curve (AUPRC) that separates monitor error from failures lacking evidence in the observed prefix. For finite-state audit, post-hoc deterministic finite automaton (DFA) extraction remains compact on WebArena and τ^2-Bench (29 and 20 states) but expands to 151 and 187 states on SkillsBench and TerminalBench. Finally, first-alert diagnostics show that strong ranking does not imply deployment utility: WebArena ranks well yet fails to support low-false-alarm alerts, whereas τ^2-Bench and TerminalBench retain more actionable early alerts. Together, these results position PrefixGuard as a practical monitor-synthesis recipe with explicit diagnostics for when prefix warnings translate into actionable interventions.

Although recent tool-augmented benchmarks involve complex requests, evaluation remains limited to answer matching, neglecting critical trajectory aspects like efficiency, hallucination, and adaptivity. The most straightforward method for evaluation is to compare an agent's trajectory with the ground-truth, but annotating all valid ground-truth trajectories is prohibitively expensive. In this manner, we introduce TRACE, a reference-free framework for the multi-dimensional evaluation of tool-augmented LLMs. By incorporating an evidence bank which accumulates knowledge from preceding steps, TRACE assesses an agent's reasoning trajectory effectively. To validate our framework, we develop a new meta-evaluation dataset with diverse and flawed trajectories, each labeled with multi-faceted performance scores. Our results confirm that TRACE accurately evaluates complex trajectories even with small open-source LLMs. Furthermore, we apply our method to evaluate the trajectories that agents produce while solving tool-augmented tasks, presenting previously unreported observations and their corresponding insights.

In this paper we present an approach for localizing steps of procedural activities in narrated how-to videos. To deal with the scarcity of labeled data at scale, we source the step descriptions from a language knowledge base (wikiHow) containing instructional articles for a large variety of procedural tasks. Without any form of manual supervision, our model learns to temporally ground the steps of procedural articles in how-to videos by matching three modalities: frames, narrations, and step descriptions. Specifically, our method aligns steps to video by fusing information from two distinct pathways: i) {\em direct} alignment of step descriptions to frames, ii) {\em indirect} alignment obtained by composing steps-to-narrations with narrations-to-video correspondences. Notably, our approach performs global temporal grounding of all steps in an article at once by exploiting order information, and is trained with step pseudo-labels which are iteratively refined and aggressively filtered. In order to validate our model we introduce a new evaluation benchmark -- HT-Step -- obtained by manually annotating a 124-hour subset of HowTo100MA test server is accessible at \url{https://eval.ai/web/challenges/challenge-page/2082.} with steps sourced from wikiHow articles. Experiments on this benchmark as well as zero-shot evaluations on CrossTask demonstrate that our multi-modality alignment yields dramatic gains over several baselines and prior works. Finally, we show that our inner module for matching narration-to-video outperforms by a large margin the state of the art on the HTM-Align narration-video alignment benchmark.

Existing code reasoning methods primarily supervise final code outputs, ignoring intermediate states, often leading to reward hacking where correct answers are obtained through inconsistent reasoning. We propose StepCodeReasoner, a framework that introduces explicit intermediate execution-state supervision. By automatically inserting structured print-based execution-trace anchors into code, the model is trained to predict runtime states at each step, transforming code reasoning into a verifiable, stepwise execution modeling problem. Building on this execution-aware method, we introduce Bi-Level GRPO, a reinforcement learning algorithm for structured credit assignment at two levels: inter-trajectory, comparing alternative execution paths, and intra-trajectory, rewarding intermediate accuracy based on its impact on downstream correctness. Extensive experiments demonstrate that StepCodeReasoner achieves SOTA performance in code reasoning. In particular, our 7B model achieves 91.1\% on CRUXEval and 86.5\% on LiveCodeBench, outperforming the CodeReasoner-7B baseline (86.0\% and 77.7\%) and GPT-4o (85.6\% and 75.1\%). Furthermore, on the execution-trace benchmark REval, our model scores 82.9\%, outperforming baseline CodeReasoner-7B (72.3\%), its 14B counterpart (81.1\%), and GPT-4o (77.3\%). Additionally, our approach also improves code generation performance, demonstrating that explicit execution modeling enhances both code reasoning and code generation.

Tool-calling benchmarks are increasingly used to rank language-model agents, yet their scores are often treated as ground truth without validating the evaluators themselves. We present a systematic validity and reproducibility audit of four major tool-calling benchmark families: BFCL v4, τ2-Bench, LiveMCPBench, and MCP-Atlas. Across 496 expert-reviewed benchmark tasks, we find 92 evaluator-human disagreements, corresponding to an 18.5% misalignment rate. The failures are not isolated annotation mistakes: deterministic benchmarks exhibit brittle state matching, trajectory lock-in, incorrect ground truths, substring-based communication failures, and reward-basis misalignment, while LLM-judge benchmarks exhibit rubric drift, hallucinated completion, answer-only scoring, and substantial run-to-run variance. In LiveMCPBench, 23 repeated evaluations of the same setup produce scores ranging from 57.9% to 76.8%, a spread of 18.9 percentage points, large enough to change leaderboard conclusions. These results show that current tool-calling scores can reflect evaluator artifacts rather than agent capability. We introduce a unified taxonomy of tool-calling evaluation failures, release trace-level audit artifacts and corrected evaluation components, and argue for decomposed metrics that separately measure tool invocation, task completion, and outcome verification. Our findings suggest that progress in tool-using agents requires benchmarks whose evaluators are themselves reproducible, auditable, and aligned with human judgments of task success. We further introduce Tool-Veritas, a configurable benchmark that combines deterministic state verification with optional qualitative judging, and Harness Lab, an open-source system for benchmark execution, trace inspection, repeated-run comparison, and evaluator debugging.

We study auto research as a closed empirical loop driven by external measurement. Each submitted trial carries a hypothesis, an executable code edit, an evaluator-owned outcome, and feedback that shapes the next proposal. The output is not a generated paper or a single model checkpoint, but an auditable trajectory of proposals, code diffs, experiments, scores, and failure labels. We instantiate this loop with specialist agents that partition recipe surfaces and share measured lineage across trials. The central empirical finding is that lineage feedback lets agents turn evaluator outcomes, including crashes, budget overruns, size failures, and accuracy-gate misses, into later program-level recipe edits rather than one-shot suggestions. Across 1,197 headline-run trials plus 600 Parameter Golf control trials after one-time setup and launch, humans did not choose proposals, edit recipes, override scores, or repair failed trials during the search. In the three headline runs, the same submitted-trial loop reduces Parameter Golf validation bpb by 0.81%, raises NanoChat-D12 CORE by 38.7%, and reduces CIFAR-10 Airbench96 wallclock by 4.59%, with each task measured by its own external evaluator and legality checks. The trace includes a strict architecture-domain audit of 157 headline-run submissions and program rewrites such as a NanoChat attention-kernel path change. Within this scope the loop autonomously writes code, submits experiments, absorbs feedback, applies and combines known techniques inside each environment, and improves public starting recipes.

Applied Behavior Analysis (ABA) is a clinical discipline whose documentation, teaching programs and multi-session behavioral logs, is formulaic and high-volume, yet real session data is HIPAA-protected and bound by professional confidentiality rules, blocking the release of a training corpus. We present TRACE (Taxonomy-Referenced ABA Clinical Examples), a 2,999-example synthetic instruction-tuning dataset covering two ABA tasks: teaching-program generation across Discrete Trial Training, Natural Environment Teaching, and Task Analysis; and multi-session behavioral interpretation across twelve trajectory patterns and thirteen target behaviors. Every example is produced by a deterministic taxonomy-driven generator grounded in the canonical ABA literature, and every example carries complete sampling provenance, the exact taxonomy cells that produced it. The dataset is released under CC BY-NC 4.0 for data and MIT for code, with stratified train (2,549), validation (149), test (281), and sanity (20) splits. TRACE is a research artifact and has not been clinically validated.

Efficient multi-hop reasoning requires Large Language Models (LLMs) based agents to acquire high-value external knowledge iteratively. Previous work has explored reinforcement learning (RL) to train LLMs to perform search-based document retrieval, achieving notable improvements in QA performance, but underperform on complex, multi-hop QA resulting from the sparse rewards from global signal only. To address this gap in existing research, we introduce StepSearch, a framework for search LLMs that trained with step-wise proximal policy optimization method. It consists of richer and more detailed intermediate search rewards and token-level process supervision based on information gain and redundancy penalties to better guide each search step. We constructed a fine-grained question-answering dataset containing sub-question-level search trajectories based on open source datasets through a set of data pipeline method. On standard multi-hop QA benchmarks, it significantly outperforms global-reward baselines, achieving 11.2% and 4.2% absolute improvements for 3B and 7B models over various search with RL baselines using only 19k training data, demonstrating the effectiveness of fine-grained, stepwise supervision in optimizing deep search LLMs. Our code will be released on https://github.com/Zillwang/StepSearch.

Procedural activities follow well-defined structures: whether we consider a cooking recipe or a mechanic repairing a car, these activities naturally decompose in a hierarchy of steps and sub-steps. Traditional approaches for step grounding require extensive annotations and scale poorly. Instead, we argue that such hierarchical structure can emerge naturally from uncurated videos of human activities through recurring patterns of co-occurring actions and activities. Our approach builds on HiERO, a weakly-supervised representation learning approach that maps close in the feature space actions that are functionally related to each other, leveraging only fine-grained action-level narrations. In this feature space, procedure steps can be detected by a simple clustering, with no additional task-specific fine-tuning. For the Ego4D Step Grounding challenge, we augment this approach by ensuring fine and coarse level agreement in step assignments, enforcing strict temporal monotonicity of the grounded steps and post-processing the detected steps to reduce the impact of noisy predictions. We call this approach HiERO-StepG and it achieves 56.27 % on the R@1 (IoU = 0.3) metric on the global leaderboard at submission time, ranking second while being completely zero-shot and not requiring procedure-specific annotations. Project page: https://github.com/andreazenotto/HiERO-StepG.

The increasing adoption of agentic workflows across diverse domains brings a critical need to scalably and systematically evaluate the complex traces these systems generate. Current evaluation methods depend on manual, domain-specific human analysis of lengthy workflow traces - an approach that does not scale with the growing complexity and volume of agentic outputs. Error analysis in these settings is further complicated by the interplay of external tool outputs and language model reasoning, making it more challenging than traditional software debugging. In this work, we (1) articulate the need for robust and dynamic evaluation methods for agentic workflow traces, (2) introduce a formal taxonomy of error types encountered in agentic systems, and (3) present a set of 148 large human-annotated traces (TRAIL) constructed using this taxonomy and grounded in established agentic benchmarks. To ensure ecological validity, we curate traces from both single and multi-agent systems, focusing on real-world applications such as software engineering and open-world information retrieval. Our evaluations reveal that modern long context LLMs perform poorly at trace debugging, with the best Gemini-2.5-pro model scoring a mere 11% on TRAIL. Our dataset and code are made publicly available to support and accelerate future research in scalable evaluation for agentic workflows.

Best-of-N decoding methods instruct large language models (LLMs) to generate multiple solutions, score each using a scoring function, and select the highest scored as the final answer to mathematical reasoning problems. However, this repeated independent process often leads to the same mistakes, making the selected solution still incorrect. We propose a novel prompting method named Stepwise Correction (StepCo) that helps LLMs identify and revise incorrect steps in their generated reasoning paths. It iterates verification and revision phases that employ a process-supervised verifier. The verify-then-revise process not only improves answer correctness but also reduces token consumption with fewer paths needed to generate. With StepCo, a series of LLMs demonstrate exceptional performance. Notably, using GPT-4o as the backend LLM, StepCo achieves an average accuracy of 94.1 across eight datasets, significantly outperforming the state-of-the-art Best-of-N method by +2.4, while reducing token consumption by 77.8%.

Complex reasoning tasks often rely on the ability to consistently and accurately apply simple rules across incremental steps, a foundational capability which we term "level-0" reasoning. To systematically evaluate this capability, we introduce L0-Bench, a language model benchmark for testing procedural correctness -- the ability to generate correct reasoning processes, complementing existing benchmarks that primarily focus on outcome correctness. Given synthetic Python functions with simple operations, L0-Bench grades models on their ability to generate step-by-step, error-free execution traces. The synthetic nature of L0-Bench enables systematic and scalable generation of test programs along various axes (e.g., number of trace steps). We evaluate a diverse array of recent closed-source and open-weight models on a baseline test set. All models exhibit degradation as the number of target trace steps increases, while larger models and reasoning-enhanced models better maintain correctness over multiple steps. Additionally, we use L0-Bench to explore test-time scaling along three dimensions: input context length, number of solutions for majority voting, and inference steps. Our results suggest substantial room to improve "level-0" reasoning and potential directions to build more reliable reasoning systems.

Agentic red-teaming benchmarks report whether an injected agent was compromised as a single bit: the attack succeeded, or it did not. We argue that this binary attack-success rate discards the information a defender most needs, namely how harmful the resulting action was. We introduce an action-graded harm rubric that scores an agent's tool-call trajectory on a seven-level ordinal scale (L0 to L6) according to whether the executed action was reversible, whether it crossed scope to reach another party, and whether it expanded privilege. We compute the scale two ways: a deterministic oracle that reads the trajectory and the attacker's stated goal, and a panel of three frontier language-model judges that read a tag-free account of the same trajectory. Across four victim models and two defenses on the AgentDojo workspace suite, severity grading exposes three cases the binary metric hides, including a defense that reports a zero attack-success rate while still permitting an externally visible cross-scope leak through an unfiltered tool. The judge panel reproduces the oracle with high ordinal agreement (Krippendorff's alpha = 0.91) but shares systematic blind spots that we characterize, most notably a failure to recognize escalation chains. Unlike prior work that provides harm taxonomies, harmful-task completion tests, execution-level safety benchmarks, or severity-aware simulation, our contribution is a reusable, trace-grounded severity instrument applied to the actual actions recorded in existing red-team logs. All code, prompts, and per-episode logs are released.

In recent years, large language models (LLMs) have made significant progress in code intelligence, yet systematically evaluating their code understanding and reasoning abilities remains challenging. Mainstream benchmarks such as HumanEval and MBPP primarily assess functional correctness, while reasoning benchmarks like CRUXEVAL are limited to single-function, low-complexity scenarios. As a result, advanced models achieve nearly saturated scores, limiting their discriminative power. To address this, we present STEPWISE-CODEX-Bench (SX-Bench), a novel benchmark designed for complex multi-function understanding and fine-grained execution reasoning. SX-Bench features tasks involving collaboration among multiple sub-functions (e.g., chained calls, nested loops), shifting evaluation towards overall control and data flow modeling. It defines "computation steps" as the minimal execution unit and requires models to predict the total number of steps in reasoning tasks, thereby assessing a model's in-depth understanding of dynamic execution beyond simple I/O matching. Evaluation on over 20 mainstream models (including 14 reasoning-enhanced models) demonstrates that SX-Bench is highly discriminative: even the state-of-the-art OpenAI-O3 achieves only 78.37 percent accuracy on Hard-Reasoning tasks, much lower than its saturated scores on previous benchmarks, thereby revealing bottlenecks in complex and fine-grained reasoning. We also release an automated pipeline combining program synthesis, symbolic execution, and LLM-aided validation for efficient benchmark generation and quality assurance. SX-Bench advances code evaluation from "single-function verification" to "multi-function dynamic reasoning," providing a key tool for the in-depth assessment of advanced code intelligence models.

LLM reasoning traces suffer from complex flaws -- *Step Internal Flaws* (logical errors, hallucinations, etc.) and *Step-wise Flaws* (overthinking, underthinking), which vary by sample. A natural approach would be to provide ground-truth labels to guide LLMs' reasoning. Contrary to intuition, we show that this yields no improvement in reasoning ability. We then propose CRAFT, a unified framework that mitigates both types of Step flaws, which builds a Reasoning Knowledge Graph (RKG) based on the consensus parts of multiple candidate traces, and synthesizes a high-quality trace through topological generation. Our approach improves label-prediction accuracy by 10+% on average, and consistently outperforms all baselines across both logical and mathematical reasoning benchmarks. Further, detailed benchmark evaluation proves that our method also improves the quality of LLMs' reasoning traces in multiple dimensions.

As LLMs are increasingly used as judges in code applications, they should be evaluated in realistic interactive settings that capture partial context and ambiguous intent. We present TRACE (Tool for Rubric Analysis in Code Evaluation), a framework that evaluates LLM judges' ability to predict human preferences and automatically extracts rubric items to reveal systematic biases in how humans and models weigh each item. Across three modalities -- chat-based programming, IDE autocompletion, and instructed code editing -- we use TRACE to measure how well LLM judges align with developer preferences. Among 13 different models, the best judges underperform human annotators by 12-23%. TRACE identifies 35 significant sources of misalignment between humans and judges across interaction modalities, the majority of which correspond to existing software engineering code quality criteria. For example, in chat-based coding, judges are biased towards longer code explanations while humans prefer shorter ones. We find significant misalignment on the majority of existing code quality dimensions, showing alignment gaps between LLM judges and human preference in realistic coding applications.

Large language models can generate fluent answers that are unfaithful to the provided context, while many safeguards rely on external verification or a separate judge after generation. We introduce internal flow signatures that audit decision formation from depthwise dynamics at a fixed inter-block monitoring boundary. The method stabilizes token-wise motion via bias-centered monitoring, then summarizes trajectories in compact moving readout-aligned subspaces constructed from the top token and its close competitors within each depth window. Neighboring window frames are aligned by an orthogonal transport, yielding depth-comparable transported step lengths, turning angles, and subspace drift summaries that are invariant to within-window basis choices. A lightweight GRU validator trained on these signatures performs self-checking without modifying the base model. Beyond detection, the validator localizes a culprit depth event and enables a targeted refinement: the model rolls back to the culprit token and clamps an abnormal transported step at the identified block while preserving the orthogonal residual. The resulting pipeline provides actionable localization and low-overhead self-checking from internal decision dynamics. Code is available at github.com/EavnJeong/Internal-Flow-Signatures-for-Self-Checking-and-Refinement-in-LLMs.

Code agents are advancing rapidly, but debugging them is becoming increasingly difficult. As frameworks orchestrate parallel tool calls and multi-stage workflows over complex tasks, making the agent's state transitions and error propagation hard to observe. In these runs, an early misstep can trap the agent in unproductive loops or even cascade into fundamental errors, forming hidden error chains that make it hard to tell when the agent goes off track and why. Existing agent tracing analyses either focus on simple interaction or rely on small-scale manual inspection, which limits their scalability and usefulness for real coding workflows. We present CodeTracer, a tracing architecture that parses heterogeneous run artifacts through evolving extractors, reconstructs the full state transition history as a hierarchical trace tree with persistent memory, and performs failure onset localization to pinpoint the failure origin and its downstream chain. To enable systematic evaluation, we construct CodeTraceBench from a large collection of executed trajectories generated by four widely used code agent frameworks on diverse code tasks (e.g., bug fixing, refactoring, and terminal interaction), with supervision at both the stage and step levels for failure localization. Experiments show that CodeTracer substantially outperforms direct prompting and lightweight baselines, and that replaying its diagnostic signals consistently recovers originally failed runs under matched budgets. Our code and data are publicly available.

Interactive agent benchmarks face a tension between scalable construction and realistic workflow evaluation. Hand-authored tasks are expensive to extend and revise, while static prompt evaluation misses failures that only appear when agents operate over persistent state. Existing interactive benchmarks have advanced agent evaluation significantly, but most initialize tasks from clean state and do not systematically test how agents handle pre-existing partial, stale, or conflicting artifacts. We present ClawForge, a generator-backed benchmark framework for executable command-line workflows under state conflict. The framework compiles scenario templates, grounded slots, initialized state, reference trajectories, and validators into reproducible task specifications, and evaluates agents step by step over persistent workflow surfaces using normalized end state and observable side effects rather than exact trajectory matching. We instantiate this framework as the ClawForge-Bench (17 scenarios, 6 ability categories). Results across seven frontier models show that the best model reaches only 45.3% strict accuracy, wrong-state replacement remains below 17\% for all models, and the widest model separation (17% to 90%) is driven by whether agents inspect existing state before acting. Partial-credit and step-efficiency analyses further reveal that many failures are near-miss closures rather than early breakdowns, and that models exhibit qualitatively different failure styles under state conflict.

Interactive LLM agents are becoming part of daily work, but they do not reliably become easier to work with over time: a correction remembered in one session may still be violated in the next. We study this gap between preference access and preference compliance. In tasks derived from anonymized real-user friction cases, Mem0 memory still leaves 57.5% of applicable preference checks violated. We introduce Test-time Rule Acquisition and Compiled Enforcement (TRACE), a drop-in skill-layer pipeline for coding-agent runtimes that mines user corrections, rewrites them as atomic rules, and compiles them into runtime checks that must pass before an agent completes future tasks. Unlike runtime checks written ahead of time by developers, TRACE skills come from the user's own chat corrections. We evaluate TRACE with simulated user-in-the-loop experiments on ClawArena coding-agent tasks and MemoryArena-derived memory-intensive tasks. On ClawArena, TRACE reduces held-out preference violation from 100.0% to 37.6% on in-distribution tasks and from 100.0% to 2.0% on out-of-distribution tasks. On MemoryArena-derived tasks, TRACE reduces in-distribution violation from 100.0% to 60.5% while matching or exceeding the strongest memory baseline on task pass. These results suggest that compiling corrections into runtime enforcement can address a repeated-friction failure mode that memory alone does not reliably solve, reducing the need for users to restate the same correction across future sessions. Experiment code is available at https://github.com/YujunZhou/TRACE_exp, and the deployable skill is available at https://github.com/YujunZhou/tellonce.

The deployment of machine learning models in operational contexts represents a significant investment for any organisation. Consequently, the risk of these models being misappropriated by competitors needs to be addressed. In recent years, numerous proposals have been put forth to detect instances of model stealing. However, these proposals operate under implicit and disparate data and model access assumptions; as a consequence, it remains unclear how they can be effectively compared to one another. Our evaluation shows that a simple baseline that we introduce performs on par with existing state-of-the-art fingerprints, which, on the other hand, are much more complex. To uncover the reasons behind this intriguing result, this paper introduces a systematic approach to both the creation of model fingerprinting schemes and their evaluation benchmarks. By dividing model fingerprinting into three core components -- Query, Representation and Detection (QuRD) -- we are able to identify sim100 previously unexplored QuRD combinations and gain insights into their performance. Finally, we introduce a set of metrics to compare and guide the creation of more representative model stealing detection benchmarks. Our approach reveals the need for more challenging benchmarks and a sound comparison with baselines. To foster the creation of new fingerprinting schemes and benchmarks, we open-source our fingerprinting toolbox.

We propose claim-level falsification as a principle for test-time scaling and instantiate it through Claim-Level Reliability Assessment (CLR), a training-free framework that reallocates test-time compute from additional solution sampling to targeted verification. Since whole-trace evaluation often obscures decisive errors due to signal dilution from routine tokens, CLR condenses each reasoning trace into a compact set of decision-critical claims, thereby isolating its logical anchors. Furthermore, recognizing the inherent difficulty of generating entirely correct solutions under fixed model capabilities, CLR shifts the focus to semantic falsification. This approach exploits a fundamental asymmetry between solution construction and claim refutation. Constructing a valid solution requires a flawless reasoning path, whereas refuting an incorrect claim requires identifying only a single decisive flaw. This targeted search for negative evidence systematically compresses the survival space of high-confidence incorrect traces, effectively suppressing erroneous consensus via nonlinear reliability scoring. Across four LLMs and four reasoning benchmarks under matched budgets, CLR generally improves upon pass@1 and self-consistency. On GPT-OSS-20B/CMIMC25, for instance, CLR exceeds pass@1 by 27.15 percentage-points and raises self-consistency accuracy from 77.50\% to 82.19\% with 37.0\% fewer tokens.

Software requirements traceability is a critical component of the software engineering process, enabling activities such as requirements validation, compliance verification, and safety assurance. However, the cost and effort of manually creating a complete set of trace links across natural language artifacts such as requirements, design, and test-cases can be prohibitively expensive. Researchers have therefore proposed automated link-generation solutions primarily based on information-retrieval (IR) techniques; however, these solutions have failed to deliver the accuracy needed for full adoption in industrial projects. Improvements can be achieved using deep-learning traceability models; however, their efficacy is impeded by the limited size and availability of project-level artifacts and links to serve as training data. In this paper, we address this problem by proposing and evaluating several deep-learning approaches for text-to-text traceability. Our method, named NLTrace, explores three transfer learning strategies that use datasets mined from open world platforms. Through pretraining Language Models (LMs) and leveraging adjacent tracing tasks, we demonstrate that NLTrace can significantly improve the performance of LM based trace models when training links are available. In such scenarios NLTrace outperforms the best performing classical IR method with an 188% improvement in F2 score and 94.01% in Mean Average Precision (MAP). It also outperforms the general LM based trace model by 7% and 23% for F2 and MAP respectively. In addition, NLTrace can adapt to low-resource tracing scenarios where other LM models can not. The knowledge learned from adjacent tasks enables NLTrace to outperform VSM models by 28% F2 on generation challenges when presented with a small number of training examples.

Understanding the brain increasingly depends on integrating evidence across scales, modalities, and disciplines. Addressing a single research question therefore requires a coordinated sequence of operations, from surveying prior work to executing analyses and interpreting results in light of domain knowledge. AI agents promise to accelerate this process, but current agents lack domain expertise in brain science, may fabricate claims, drift during multi-step reasoning, and offer few defined points for expert intervention. These failures are especially costly in brain science, where conclusions feed into downstream scientific claims and depend on laboratory-specific expertise and careful human judgment. We present BrainPilot a fully open-source multi-agent system that accelerates brain science research with traceable logs and agent-verified results. A principal investigator (PI) agent coordinates specialist agents grounded in curated domain knowledge: a unified brain science knowledge base containing 7{,}233 indexed items and a skill library of 72 reusable methodology units across seven research domains. Every major step is recorded in the Graph of Trace, an auditable record that links subgoals, tool use, evidence, and claims and allows researchers to follow and inspect the workflow. An Auditor agent further integrates fabrication checking into the workflow. For evaluation, we run three brain science tasks from Agents' Last Exam, introduce our own benchmark, BrainPilotBench-v0, and present additional end-to-end case studies. Across these evaluations, BrainPilot with an open-source backbone model attains performance comparable to state-of-the-art agent framework with less costs.

With the rapid advancement of agent-based methods in recent years, Agentic RAG has undoubtedly become an important research direction. Multi-hop reasoning, which requires models to engage in deliberate thinking and multi-step interaction, serves as a critical testbed for assessing such capabilities. However, existing benchmarks typically provide only final questions and answers, while lacking the intermediate hop-level questions that gradually connect atomic questions to the final multi-hop query. This limitation prevents researchers from analyzing at which step an agent fails and restricts more fine-grained evaluation of model capabilities. Moreover, most current benchmarks are manually constructed, which is both time-consuming and labor-intensive, while also limiting scalability and generalization. To address these challenges, we introduce AgenticRAGTracer, the first Agentic RAG benchmark that is primarily constructed automatically by large language models and designed to support step-by-step validation. Our benchmark spans multiple domains, contains 1,305 data points, and has no overlap with existing mainstream benchmarks. Extensive experiments demonstrate that even the best large language models perform poorly on our dataset. For instance, GPT-5 attains merely 22.6\% EM accuracy on the hardest portion of our dataset. Hop-aware diagnosis reveals that failures are primarily driven by distorted reasoning chains -- either collapsing prematurely or wandering into over-extension. This highlights a critical inability to allocate steps consistent with the task's logical structure, providing a diagnostic dimension missing in traditional evaluations. We believe our work will facilitate research in Agentic RAG and inspire further meaningful progress in this area. Our code and data are available at https://github.com/YqjMartin/AgenticRAGTracer.

Reliable evaluation is essential for developing and deploying large language models, yet in practice it often requires substantial manual effort: practitioners must identify appropriate benchmarks, reproduce heterogeneous evaluation codebases, configure dataset schema mappings, and interpret aggregated metrics. To address these challenges, we present One-Eval, an agentic evaluation system that converts natural-language evaluation requests into executable, traceable, and customizable evaluation workflows. One-Eval integrates (i) NL2Bench for intent structuring and personalized benchmark planning, (ii) BenchResolve for benchmark resolution, automatic dataset acquisition, and schema normalization to ensure executability, and (iii) Metrics \& Reporting for task-aware metric selection and decision-oriented reporting beyond scalar scores. The system further incorporates human-in-the-loop checkpoints for review, editing, and rollback, while preserving sample evidence trails for debugging and auditability. Experiments show that One-Eval can execute end-to-end evaluations from diverse natural-language requests with minimal user effort, supporting more efficient and reproducible evaluation in industrial settings. Our framework is publicly available at https://github.com/OpenDCAI/One-Eval.

Large Language Models (LLMs) often generate code with subtle but critical bugs, especially for complex tasks. Existing automated repair methods typically rely on superficial pass/fail signals, offering limited visibility into program behavior and hindering precise error localization. In addition, without a way to learn from prior failures, repair processes often fall into repetitive and inefficient cycles. To overcome these challenges, we present TraceCoder, a collaborative multi-agent framework that emulates the observe-analyze-repair process of human experts. The framework first instruments the code with diagnostic probes to capture fine-grained runtime traces, enabling deep insight into its internal execution. It then conducts causal analysis on these traces to accurately identify the root cause of the failure. This process is further enhanced by a novel Historical Lesson Learning Mechanism (HLLM), which distills insights from prior failed repair attempts to inform subsequent correction strategies and prevent recurrence of similar mistakes. To ensure stable convergence, a Rollback Mechanism enforces that each repair iteration constitutes a strict improvement toward the correct solution. Comprehensive experiments across multiple benchmarks show that TraceCoder achieves up to a 34.43\% relative improvement in Pass@1 accuracy over existing advanced baselines. Ablation studies verify the significance of each system component, with the iterative repair process alone contributing a 65.61\% relative gain in accuracy. Furthermore, TraceCoder significantly outperforms leading iterative methods in terms of both accuracy and cost-efficiency.

Despite widespread deployment of Large Language Models, systematic evaluation of instruction-following capabilities remains challenging. While comprehensive benchmarks exist, focused assessments that quickly diagnose specific instruction adherence patterns are valuable. As newer models may be trained on existing benchmarks, novel evaluation approaches are needed to assess genuine capabilities rather than memorized performance. This paper presents a streamlined evaluation framework using twenty carefully designed prompts to assess LLM instruction-following across diverse task categories. We demonstrate this framework through a large-scale empirical study conducted on October 14, 2025, testing 256 verified working models from 331 available via OpenRouter. To ensure methodological rigor and prevent selection bias, we first verified each model's basic functionality before inclusion. Unlike large-scale benchmarks requiring extensive computational resources, our approach offers a practical diagnostic tool researchers and practitioners can readily apply. Our methodology builds upon verifiable instructions while introducing a compact test suite balancing comprehensiveness with efficiency. Each prompt targets distinct aspects of instruction following, including format compliance, content constraints, logical sequencing, and multi-step task execution. We evaluate models from major providers (OpenAI, Anthropic, Google, Meta, Mistral) and emerging implementations (Qwen, DeepSeek, community models), providing comparative performance analysis. Our findings reveal consistent failure modes and identify specific instruction types posing particular challenges. This work contributes both a practical evaluation tool and one of the most comprehensive empirical analyses of instruction-following capabilities across the contemporary LLM landscape.

Evaluating large language models (LLMs) for instrument control requires methods that go beyond standard, stateless algorithmic benchmarks, since the behavior of physical systems cannot be fully captured by unit tests alone. Here we introduce EnvTrace, a simulation-based method that evaluates execution traces to assess semantic code equivalence. EnvTrace is demonstrated with a beamline control-logic digital twin to facilitate the evaluation of instrument control code, with the digital twin itself also enabling the pre-execution validation of live experiments. Over 30 LLMs were evaluated using trace alignment to generate a multi-faceted score for functional correctness across key behavioral dimensions, showing that many top-tier models can approach human-level performance in rapid control-code generation. This is a first step toward a broader vision where LLMs and digital twins work symbiotically: LLMs providing intuitive control and agentic orchestration, and digital twins offering safe and high-fidelity environments, paving the way towards autonomous embodied AI.

While Large Language Models (LLMs) have evolved into tool-using agents, they remain brittle in long-horizon interactions. Unlike mathematical reasoning where errors are often rectifiable via backtracking, tool-use failures frequently induce irreversible side effects, making accurate step-level verification critical. However, existing process-level benchmarks are predominantly confined to closed-world mathematical domains, failing to capture the dynamic and open-ended nature of tool execution. To bridge this gap, we introduce AgentProcessBench, the first benchmark dedicated to evaluating step-level effectiveness in realistic, tool-augmented trajectories. The benchmark comprises 1,000 diverse trajectories and 8,509 human-labeled step annotations with 89.1% inter-annotator agreement. It features a ternary labeling scheme to capture exploration and an error propagation rule to reduce labeling ambiguity. Extensive experiments reveal key insights: (1) weaker policy models exhibit inflated ratios of correct steps due to early termination; (2) distinguishing neutral and erroneous actions remains a significant challenge for current models; and (3) process-derived signals provide complementary value to outcome supervision, significantly enhancing test-time scaling. We hope AgentProcessBench can foster future research in reward models and pave the way toward general agents. The code and data are available at https://github.com/RUCBM/AgentProcessBench.

Evaluating open-ended outputs from large language models (LLMs) remains challenging due to the absence of ground truth. Existing metrics rely on final-answer accuracy or surface-level statistics, leaving the reasoning process itself unexamined. We introduce TRACE (Toulmin-based Reasoning Assessment through Constructive Elements), a metric that analyzes Chain-of-Thought (CoT) reasoning processes. Rather than judging outcomes, TRACE inspects how arguments are constructed by integrating Toulmin's argumentation theory with Flavell's metacognitive framework to assess reasoning structure. Experiments on 26.3K QA samples across 7 reasoning models show strong correlation with benchmark accuracy (r=0.74). Furthermore, TRACE is effective as a reinforcement learning reward signal, outperforming accuracy-only baselines. Together, these results indicate that logically sound reasoning leads to higher-quality answers. TRACE thus serves as a complementary metric for evaluating open-ended outputs. Code is available at https://github.com/hyyangkisti/trace.

With the rapid advancement of large language models (LLMs), harnesses have become essential infrastructure for deploying agents across a wide range of domains. The fast-evolving harness ecosystem has also made rigorous capability evaluation increasingly important. However, efficiently building an end-to-end, systematic, and comprehensive evaluation pipeline remains a significant challenge. To address this challenge, we introduce A^2E (Agent Auditing Engine), an end-to-end evaluation engine designed for agent harnesses. A^2E leverages our newly proposed Agent Task Protocol (ATP) to enable the rapid integration of evaluation tasks with different harnesses. Through an automatically instrumented Monitor, it captures and generates standardized execution traces during experiments. In the Evaluation stage, A^2E systematically assesses harness capabilities using a suite of multidimensional metrics. Compared with correctness alone, these metrics provide a more fine-grained characterization of differences among harnesses in execution efficiency, tool use, task planning, and error recovery. Experiments conducted with A^2E further reveal that model-harness combinations exhibit substantial performance variation across different types of tasks, and that no single combination consistently outperforms all others across every task. These findings not only demonstrate the necessity of systematic evaluation but also provide useful guidance for the co-evolving of models and harnesses. Our code is available at https://github.com/datamllab/A2E.

Large language models (LLMs) have demonstrated remarkable capabilities, but they still frequently produce hallucinations. These hallucinations are difficult to detect in reasoning-intensive tasks, where the content appears coherent but contains errors like logical flaws and unreliable intermediate results. While step-level analysis is commonly used to detect internal hallucinations, it suffers from limited granularity and poor scalability due to its reliance on step segmentation. To address these limitations, we propose TokenHD, a holistic pipeline for training token-level hallucination detectors. Specifically, TokenHD consists of a scalable data engine for synthesizing large-scale hallucination annotations along with a training recipe featuring an importance-weighted strategy for robust model training. To systematically assess the detection performance, we also provide a rigorous evaluation protocol. Through training within TokenHD, our detector operates directly on free-form text to identify hallucinations, eliminating the need for predefined step segmentation or additional text reformatting. Our experiments show that even a small detector (0.6B) achieves substantial performance gains after training, surpassing much larger reasoning models (e.g., QwQ-32B), and detection performance scales consistently with model size from 0.6B to 8B. Finally, we show that our detector can generalize well across diverse practical scenarios and explore strategies to further enhance its cross-domain generalization capability.

Real-world video benchmarks provide broad coverage, but their fixed clips entangle event count, rate, duration, and visual complexity, making failure modes hard to isolate. While existing programmatic benchmarks offer better control, they score only the final answer rather than auditing reported events against executable ground truth. To bridge this gap, we introduce trace-grounded parametric profiling for event counting in three controlled video tasks: bouncing-ball wall contacts, visual blinks, and categorical state transitions. Across 2,190 videos, we vary event count N and frequency F while holding rendering fixed. Each video includes an executable event trace for capability-surface estimation and timestamp-level evaluation. Our results reveal a staged temporal failure. At an 80% reliability threshold, Gemini 3.6 Flash reliably counts persistent state transitions up to 12 events at 0.5 and 1.0 Hz, yet demonstrates no reliable positive-count region for transient blinking events. Thus, event representation dictates whether a model initially accesses evidence -- a limitation that compounds as count and frequency increase. In the high-count, high-frequency regime, only 0.2% of final counts are correct and the model recovers just 18.1% of true events. To test if visual access is the primary bottleneck, we increase sampling rate. Although this boosts Bounce Ball accuracy from 19.6% to 29.3%, the reported sequence agrees with ground truth only 3.7% of the time. Extra frames can therefore inflate final scores without producing faithful event recovery. Different prompting strategies yield similarly limited gains, and real-world video evaluations show the same concentration of success at low event counts. Ultimately, trace-grounded profiling shifts video evaluation from aggregate accuracy metrics to a detailed diagnostic of where temporal reasoning fails.

Performing tasks on the web presents fundamental challenges to large language models (LLMs), including combinatorially large open-world tasks and variations across web interfaces. Simply specifying a large prompt to handle all possible behaviors and states is extremely complex, and results in behavior leaks between unrelated behaviors. Decomposition to distinct policies can address this challenge, but requires carefully handing off control between policies. We propose Stacked LLM Policies for Web Actions (SteP), an approach to dynamically compose policies to solve a diverse set of web tasks. SteP defines a Markov Decision Process where the state is a stack of policies representing the control state, i.e., the chain of policy calls. Unlike traditional methods that are restricted to static hierarchies, SteP enables dynamic control that adapts to the complexity of the task. We evaluate SteP against multiple baselines and web environments including WebArena, MiniWoB++, and a CRM. On WebArena, SteP improves (14.9\% to 33.5\%) over SOTA that use GPT-4 policies, while on MiniWob++, SteP is competitive with prior works while using significantly less data. Our code and data are available at https://asappresearch.github.io/webagents-step.

Despite the rapid progress of Multimodal Large Language Models (MLLMs), their ability to perform reliable visual grounding in high-stakes clinical software environments remains underexplored. Existing GUI benchmarks largely focus on isolated, single-step grounding queries, overlooking the sequential, workflow-driven reasoning required in real-world medical interfaces, where tasks evolve across independent steps and dynamic interface states. We introduce MedSPOT, a workflow-aware sequential grounding benchmark for clinical GUI environments. Unlike prior benchmarks that treat grounding as a standalone prediction task, MedSPOT models procedural interaction as a sequence of structured spatial decisions. The benchmark comprises 216 task-driven videos with 597 annotated keyframes, in which each task consists of 2 to 3 interdependent grounding steps within realistic medical workflows. This design captures interface hierarchies, contextual dependencies, and fine-grained spatial precision under evolving conditions. To evaluate procedural robustness, we propose a strict sequential evaluation protocol that terminates task assessment upon the first incorrect grounding prediction, explicitly measuring error propagation in multi-step workflows. We further introduce a comprehensive failure taxonomy, including edge bias, small-target errors, no prediction, near miss, far miss, and toolbar confusion, to enable systematic diagnosis of model behavior in clinical GUI settings. By shifting evaluation from isolated grounding to workflow-aware sequential reasoning, MedSPOT establishes a realistic and safety-critical benchmark for assessing multimodal models in medical software environments. Code and data are available at: https://github.com/Tajamul21/MedSPOT.

Hampering the interpretation of benchmark scores, evaluation data contamination has become a growing concern in the evaluation of LLMs, and an active area of research studies its effects. While evaluation data contamination is easily understood intuitively, it is surprisingly difficult to define precisely which samples should be considered contaminated and, consequently, how it impacts benchmark scores. We propose that these questions should be addressed together and that contamination metrics can be assessed based on whether models benefit from the examples they mark contaminated. We propose a novel analysis method called ConTAM, and show with a large scale survey of existing and novel n-gram based contamination metrics across 13 benchmarks and 7 models from 2 different families that ConTAM can be used to better understand evaluation data contamination and its effects. We find that contamination may have a much larger effect than reported in recent LLM releases and benefits models differently at different scales. We also find that considering only the longest contaminated substring provides a better signal than considering a union of all contaminated substrings, and that doing model and benchmark specific threshold analysis greatly increases the specificity of the results. Lastly, we investigate the impact of hyperparameter choices, finding that, among other things, both using larger values of n and disregarding matches that are infrequent in the pre-training data lead to many false negatives. With ConTAM, we provide a method to empirically ground evaluation data contamination metrics in downstream effects. With our exploration, we shed light on how evaluation data contamination can impact LLMs and provide insight into the considerations important when doing contamination analysis. We end our paper by discussing these in more detail and providing concrete suggestions for future work.

If autoresearch is itself a form of research, then autoresearch can be applied to research itself. We present Bilevel Autoresearch, a bilevel framework in which an outer autoresearch loop improves an inner autoresearch loop by reading its code and traces, identifying bottlenecks, and generating injectable Python search mechanisms at runtime. The inner loop optimizes task performance; the outer loop optimizes how the inner loop searches. Both loops use the same LLM, so improvements come from the bilevel architecture rather than a stronger meta-level model, although the outer loop consumes additional inference and wall-clock budget. On Karpathy's GPT pretraining benchmark, the meta-autoresearch outer loop achieves a 5x improvement over the standard inner loop alone (-0.045 vs. -0.009 val_bpb), while parameter-level adjustment without mechanism change yields no reliable gain. The outer loop instantiates mechanisms from adjacent search domains, including combinatorial optimization, multi-armed bandits, and design of experiments, without human specification of the final mechanism design. Trace analysis suggests that these mechanisms break deterministic search patterns and force exploration of directions the LLM's priors avoid. The experiments demonstrate, on this benchmark, a first bilevel step: an outer loop improves the search behavior of an inner loop. Code is the mechanism carrier in this implementation, but skills, prompts, workflows, evaluators, domain principles, world-model assumptions, and memory schemas can also encode mechanisms that shape future agent behavior. This suggests a path toward recursive bootstrapping, where mechanisms discovered for the inner loop can be fed back to improve the meta-level loop itself.

Deep research requires models to retrieve, connect, and synthesize evidence from large-scale heterogeneous sources to answer complex queries and produce analytical reports. Existing benchmarks mainly evaluate final outcomes, such as answer correctness, report quality, or citation alignment, while providing limited visibility into whether evidence is correctly selected, linked, and aggregated into supported claims and conclusions. To address this gap, we introduce HiEviDR-Bench, a benchmark for evaluating Hierarchical Evidence Aggregation in Deep Research. HiEviDR-Bench covers open-domain and academic-domain settings under both text-only and multimodal conditions, and represents each instance with an explicit evidence graph that captures evidence selection, cross-source linking, and aggregation from evidence to intermediate claims and final conclusions. Based on this formulation, we develop a traceability-oriented evaluation framework with five dimensions: report quality, evidence traceability, citation accuracy, claim verification, and answer correctness, together with a progressive gating mechanism for fine-grained error localization. HiEviDR-Bench contains 2,000 human-validated questions with evidence graphs across multiple difficulty levels. Experiments on 16 representative multimodal large language models show that, although many systems achieve strong report quality, their performance drops markedly on citation accuracy, claim construction, and answer correctness. Further analysis shows that the main bottlenecks lie in evidence identification and intermediate claim construction, revealing that strong surface-level report quality does not necessarily imply grounded multi-stage reasoning on our benchmark.

As LLMs shift toward autonomous agents, Deep Research has emerged as a pivotal metric. However, existing academic benchmarks like BrowseComp often fail to meet real-world demands for open-ended research, which requires robust skills in intent recognition, long-horizon decision-making, and cross-source verification. To address this, we introduce Step-DeepResearch, a cost-effective, end-to-end agent. We propose a Data Synthesis Strategy Based on Atomic Capabilities to reinforce planning and report writing, combined with a progressive training path from agentic mid-training to SFT and RL. Enhanced by a Checklist-style Judger, this approach significantly improves robustness. Furthermore, to bridge the evaluation gap in the Chinese domain, we establish ADR-Bench for realistic deep research scenarios. Experimental results show that Step-DeepResearch (32B) scores 61.4% on Scale AI Research Rubrics. On ADR-Bench, it significantly outperforms comparable models and rivals SOTA closed-source models like OpenAI and Gemini DeepResearch. These findings prove that refined training enables medium-sized models to achieve expert-level capabilities at industry-leading cost-efficiency.

Understanding the abilities of LLMs to reason about natural language plans, such as instructional text and recipes, is critical to reliably using them in decision-making systems. A fundamental aspect of plans is the temporal order in which their steps needs to be executed, which reflects the underlying causal dependencies between them. We introduce CaT-Bench, a benchmark of Step Order Prediction questions, which test whether a step must necessarily occur before or after another in cooking recipe plans. We use this to evaluate how well frontier LLMs understand causal and temporal dependencies. We find that SOTA LLMs are underwhelming (best zero-shot is only 0.59 in F1), and are biased towards predicting dependence more often, perhaps relying on temporal order of steps as a heuristic. While prompting for explanations and using few-shot examples improve performance, the best F1 result is only 0.73. Further, human evaluation of explanations along with answer correctness show that, on average, humans do not agree with model reasoning. Surprisingly, we also find that explaining after answering leads to better performance than normal chain-of-thought prompting, and LLM answers are not consistent across questions about the same step pairs. Overall, results show that LLMs' ability to detect dependence between steps has significant room for improvement.

LLM routing matters most in long-horizon applications such as coding agents, deep research systems, and computer-use agents, where a single user request triggers many model calls. Routing each call to the cheapest sufficient model can cut costs without sacrificing quality, yet existing router benchmarks evaluate routers only on one-shot prompts. They never expose the router-visible prefix at an intermediate agent step, never test whether a cheaper replacement preserves downstream task success, and often rely on online LLM judges at evaluation time. We introduce TwinRouterBench, a step-level routing benchmark with two tracks. The static track provides 970 router-visible prefixes from 520 instances across SWE-bench, BFCL, mtRAG, QMSum, and PinchBench, each paired with an execution-verified target tier estimated under a released downgrade-and-cascade protocol; scoring is deterministic arithmetic over tier labels, trajectory membership, and token costs, with no online evaluator-side LLM judge. The dynamic track supplies a harness that runs routers on the full 500-case SWE-bench Verified suite; in this paper we report a 100-case held-out evaluation disjoint from the static SWE supervision split. At each LLM call the router selects a concrete model from a locked pool, and success is measured by official task resolution and realized API spend. The two tracks support fast offline iteration followed by end-to-end validation under live agent execution. Code and data are available at https://github.com/CommonstackAI/TwinRouterBench.

Comparing a user video to a reference how-to video is a key requirement for AR/VR technology delivering personalized assistance tailored to the user's progress. However, current approaches for language-based assistance can only answer questions about a single video. We propose an approach that first automatically generates large amounts of visual instruction tuning data involving pairs of videos from HowTo100M by leveraging existing step annotations and accompanying narrations, and then trains a video-conditioned language model to jointly reason across multiple raw videos. Our model achieves state-of-the-art performance at identifying differences between video pairs and ranking videos based on the severity of these differences, and shows promising ability to perform general reasoning over multiple videos. Project page: https://github.com/facebookresearch/stepdiff

We introduce DABstep, a novel benchmark for evaluating AI agents on realistic multi-step data analysis tasks. DABstep comprises over 450 real-world challenges derived from a financial analytics platform, requiring models to combine code-based data processing with contextual reasoning over heterogeneous documentation. Each task demands an iterative, multi-step problem-solving approach, testing capabilities in data manipulation, cross-referencing multiple sources, and precise result reporting. The benchmark provides a factoid-style answer format with automatic correctness checks for objective scoring at scale. We evaluate leading LLM-based agents, revealing a substantial performance gap: even the best agent achieves only 14.55% accuracy on the hardest tasks. We detail our benchmark's design, dataset composition, task formulation, evaluation protocol, report baseline results and analyze failure modes. DABstep is released with a public leaderboard and toolkit to accelerate research in autonomous data analysis.

Training large language models to reason with search engines via reinforcement learning is hindered by a fundamental credit assignment problem: existing methods such as Search-R1 provide only a sparse outcome reward after an entire multi-step trajectory, making it infeasible to attribute success or failure to individual reasoning and retrieval decisions. Process-reward methods like StepSearch alleviate this by introducing step-level supervision, but rely on heuristic rewards such as TF-IDF overlap with gold documents, and still sample k complete trajectories per example, retaining high gradient variance. We propose SLATE, a framework built on two complementary ideas: (1) truncated step-level sampling, which generates k trajectories that share a common prefix and differ only at the next step, and (2) dense LLM-as-judge rewards, which replace heuristic scoring with a capable LLM evaluator that assesses the quality of each reasoning step, search query, and answer, providing richer and more reliable supervision. We theoretically prove that under the same dense reward structure, truncated sampling reduces the variance of advantage estimates by up to a factor of T compared to full-trajectory sampling for T-step trajectories, yielding lower-variance, better-targeted policy gradients. Experiments on seven QA benchmarks confirm that SLATE consistently outperforms both sparse-reward and process-reward baselines, with the largest gains on harder multi-hop tasks and smaller models.

Large Language Model (LLM)-based Multi-Agent Systems (MAS) enable complex problem-solving but introduce significant debugging challenges, characterized by long interaction traces, inter-agent dependencies, and delayed error manifestation. Existing diagnostic approaches often rely on expensive expert annotation or ''LLM-as-a-judge'' paradigms, which struggle to pinpoint decisive error steps within extended contexts. In this paper, we introduce ErrorProbe, a self-improving framework for semantic failure attribution that identifies responsible agents and the originating error step. The framework operates via a three-stage pipeline: (1) operationalizing the MAS failure taxonomy to detect local anomalies, (2) performing symptom-driven backward tracing to prune irrelevant context, and (3) employing a specialized multi-agent team (Strategist, Investigator, Arbiter) to validate error hypotheses through tool-grounded execution. Crucially, ErrorProbe maintains a verified episodic memory that updates only when error patterns are confirmed by executable evidence, without the need for annotation. Experiments across the TracerTraj and Who&When benchmarks demonstrate that ErrorProbe significantly outperforms baselines, particularly in step-level localization, while the verified memory enables robust cross-domain transfer without retraining.

Large Language Models (LLMs) deployed in agentic environments must exercise multiple capabilities across different task instances, where a capability is performing one or more actions in a trajectory that are necessary for successfully solving a subset of tasks in the environment. Many existing approaches either rely on synthetic training data that is not targeted to the model's actual capability deficits in the target environment or train directly on the target environment, where the model needs to implicitly learn the capabilities across tasks. We introduce TRACE (Turning Recurrent Agent failures into Capability-targeted training Environments), an end-to-end system for environment-specific agent self-improvement. TRACE contrasts successful and failed trajectories to automatically identify lacking capabilities, synthesizes a targeted training environment for each that rewards whether the capability was exercised, and trains a LoRA adapter via RL on each synthetic environment, routing to the relevant adapter at inference. Empirically, TRACE generalizes across different environments, improving over the base agent by +14.1 points on τ^2-bench (customer service) and +7 perfect scores on ToolSandbox (tool use), outperforming the strongest baseline by +7.4 points and +4 perfect scores, respectively. Given the same number of rollouts, TRACE scales more efficiently than baselines, outperforming GRPO and GEPA by +9.2 and +7.4 points on τ^2-bench.

Evaluation metrics are central to binary target segmentation because they determine how progress is measured, compared, and interpreted. In this paper, target denotes the task-defined positive region to be segmented rather than a generic foreground object. It may be salient, camouflaged, transparent, glass-like, mirror-like, shadow-like, lesion-like, or defined by other application-specific semantics. We treat existing metrics as compositions of modular design choices rather than isolated formulas. The proposed framework decomposes each metric into five stages covering prediction representation, target extraction, target matching, score computation, and metric reporting. We use this framework to analyze representative metrics and show how newer metrics address specific limits in earlier protocols. The stage choices keep each metric's assumptions visible. We then discuss the design space opened by the framework and its implications for task-aware evaluation protocols. Reference code is available at https://github.com/lartpang/PySODMetrics.

Step-by-step verifiers -- also known as process reward models (PRMs) -- are a key ingredient for test-time scaling. PRMs require step-level supervision, making them expensive to train. This work aims to build data-efficient PRMs as verbalized step-wise reward models that verify every step in the solution by generating a verification chain-of-thought (CoT). We propose ThinkPRM, a long CoT verifier fine-tuned on orders of magnitude fewer process labels than those required by discriminative PRMs. Our approach capitalizes on the inherent reasoning abilities of long CoT models, and outperforms LLM-as-a-Judge and discriminative verifiers -- using only 1% of the process labels in PRM800K -- across several challenging benchmarks. Specifically, ThinkPRM beats the baselines on ProcessBench, MATH-500, and AIME '24 under best-of-N selection and reward-guided search. In an out-of-domain evaluation on a subset of GPQA-Diamond and LiveCodeBench, our PRM surpasses discriminative verifiers trained on the full PRM800K by 8% and 4.5%, respectively. Lastly, under the same token budget, ThinkPRM scales up verification compute more effectively compared to LLM-as-a-Judge, outperforming it by 7.2% on a subset of ProcessBench. Our work highlights the value of generative, long CoT PRMs that can scale test-time compute for verification while requiring minimal supervision for training. Our code, data, and models will be released at https://github.com/mukhal/thinkprm.

AI systems can fail silently. The failure propagates through training loops, evaluation pipelines, and production monitoring stacks until downstream harm makes it visible. This paper introduces evaluation blindness: a measurement function M exhibits evaluation blindness with respect to failure class F when it produces readings indistinguishable from a healthy state while the system is actually failing, with no auxiliary signal flagging the gap. The problem surfaces at two lifecycle stages the literature has treated separately. At training time, reward models are gamed, importance-sampling corrections are silently miscalculated, and benchmark contamination inflates fine-tuning evaluations, all while loss curves look healthy and gradient updates proceed normally. At deployment time, monitoring fails to catch six classes of production failure, including an Operational category that is 100% silent by structural definition. We provide a formal detectability predicate unifying both stages. Four training-time case studies trace concrete breakdowns, including a real implementation bug in TRL PR #6594 where gradients are corrupted as loss decreases normally. A six-class taxonomy validated against 50 real-world incidents from court documents and regulatory filings finds that 53% of verifiable public failures were silent. A failure budget framework ties acceptable failure rates to use-case risk class. The implication is direct: measurement infrastructure is a correctness concern across the full AI lifecycle, not just at evaluation time. Data, code, and taxonomy schema are at https://github.com/priyanka25aug/llm-failure-taxonomy.

Although action recognition for procedural tasks has received notable attention, it has a fundamental flaw in that no measure of success for actions is provided. This limits the applicability of such systems especially within the industrial domain, since the outcome of procedural actions is often significantly more important than the mere execution. To address this limitation, we define the novel task of procedure step recognition (PSR), focusing on recognizing the correct completion and order of procedural steps. Alongside the new task, we also present the multi-modal IndustReal dataset. Unlike currently available datasets, IndustReal contains procedural errors (such as omissions) as well as execution errors. A significant part of these errors are exclusively present in the validation and test sets, making IndustReal suitable to evaluate robustness of algorithms to new, unseen mistakes. Additionally, to encourage reproducibility and allow for scalable approaches trained on synthetic data, the 3D models of all parts are publicly available. Annotations and benchmark performance are provided for action recognition and assembly state detection, as well as the new PSR task. IndustReal, along with the code and model weights, is available at: https://github.com/TimSchoonbeek/IndustReal .

Large language model (LLM)-based coding assistants have made substantial progress, yet most systems remain reactive, requiring developers to explicitly formulate their needs. Proactive coding assistants aim to infer latent developer intent from integrated development environment (IDE) interactions and repository context, thereby reducing interaction overhead and supporting more seamless assistance. However, research in this direction is limited by the scarcity of large-scale real-world developer behavior data. Existing studies therefore often rely on LLM-simulated IDE traces, whose fidelity to real development behavior remains unclear. In this paper, we investigate this simulation-to-reality gap through a large-scale empirical study. We collect real IDE interaction traces from 1{,}246 experienced industry developers over three consecutive days using a custom Visual Studio Code extension, and construct paired LLM-simulated traces for controlled comparison. Our analysis shows that simulated traces differ substantially from real traces in behavioral diversity, temporal structure, and exploratory patterns. Based on the collected data, we introduce ProCodeBench, a real-world benchmark for proactive intent prediction. Experiments with representative LLMs, retrieval-augmented methods, and agentic baselines show that current approaches remain far from reliable under real IDE traces, suggesting that simulation-based evaluation can overestimate real-world performance. Finally, our training study shows that simulated data cannot replace real data, but can complement it when used before real-world fine-tuning. These findings highlight the importance of real developer behavior data for evaluating and training proactive coding assistants.

Recent advances in reinforcement learning for code generation have made robust environments essential to prevent reward hacking. As LLMs increasingly serve as evaluators in code-based RL, their ability to detect reward hacking remains understudied. In this paper, we propose a novel taxonomy of reward exploits spanning across 54 categories and introduce TRACE (Testing Reward Anomalies in Code Environments), a synthetically curated and human-verified benchmark containing 517 testing trajectories. Unlike prior work that evaluates reward hack detection in isolated classification scenarios, we contrast these evaluations with a more realistic, contrastive anomaly detection setup on TRACE. Our experiments reveal that models capture reward hacks more effectively in contrastive settings than in isolated classification settings, with GPT-5.2 with highest reasoning mode achieving the best detection rate at 63%, up from 45% in isolated settings on TRACE. Building on this insight, we demonstrate that state-of-the-art models struggle significantly more with semantically contextualized reward hacks compared to syntactically contextualized ones. We further conduct qualitative analyses of model behaviors, as well as ablation studies showing that the ratio of benign to hacked trajectories and analysis cluster sizes substantially impact detection performance. We release the benchmark and evaluation harness to enable the community to expand TRACE and evaluate their models.

Behaviour-Driven Development (BDD) suites accumulate step-text duplication whose maintenance cost is established in prior work. Existing detection techniques require running the tests (Binamungu et al., 2018-2023) or are confined to a single organisation (Irshad et al., 2020-2022), leaving a gap: a purely static, paraphrase-robust, step-level detector usable on any repository. We fill the gap with cukereuse, an open-source Python CLI combining exact hashing, Levenshtein ratio, and sentence-transformer embeddings in a layered pipeline, released alongside an empirical corpus of 347 public GitHub repositories, 23,667 parsed .feature files, and 1,113,616 Gherkin steps. The step-weighted exact-duplicate rate is 80.2 %; the median-repository rate is 58.6 % (Spearman rho = 0.51 with size). The top hybrid cluster groups 20.7k occurrences across 2.2k files. Against 1,020 pairs manually labelled by the three authors under a released rubric (inter-annotator Fleiss' kappa = 0.84 on a 60-pair overlap), we report precision, recall, and F1 with bootstrap 95 % CIs under two protocols: the primary rubric and a score-free second-pass relabelling. The strongest honest pair-level number is near-exact at F1 = 0.822 on score-free labels; the primary-rubric semantic F1 = 0.906 is inflated by a stratification artefact that pins recall at 1.000. Lexical baselines (SourcererCC-style, NiCad-style) reach primary F1 = 0.761 and 0.799. The paper also presents a CDN-structured critique of Gherkin (Cognitive Dimensions of Notations); eight of fourteen dimensions are rated problematic or unsupported. The tool, corpus, labelled pairs, rubric, and pipeline are released under permissive licences.

Online resources such as WikiHow compile a wide range of scripts for performing everyday tasks, which can assist models in learning to reason about procedures. However, the scripts are always presented in a linear manner, which does not reflect the flexibility displayed by people executing tasks in real life. For example, in the CrossTask Dataset, 64.5% of consecutive step pairs are also observed in the reverse order, suggesting their ordering is not fixed. In addition, each step has an average of 2.56 frequent next steps, demonstrating "branching". In this paper, we propose the new challenging task of non-sequential graph script induction, aiming to capture optional and interchangeable steps in procedural planning. To automate the induction of such graph scripts for given tasks, we propose to take advantage of loosely aligned videos of people performing the tasks. In particular, we design a multimodal framework to ground procedural videos to WikiHow textual steps and thus transform each video into an observed step path on the latent ground truth graph script. This key transformation enables us to train a script knowledge model capable of both generating explicit graph scripts for learnt tasks and predicting future steps given a partial step sequence. Our best model outperforms the strongest pure text/vision baselines by 17.52% absolute gains on F1@3 for next step prediction and 13.8% absolute gains on Acc@1 for partial sequence completion. Human evaluation shows our model outperforming the WikiHow linear baseline by 48.76% absolute gains in capturing sequential and non-sequential step relationships.

Failure attribution in multi-agent systems -- pinpointing the exact step where a decisive error occurs -- is a critical yet unsolved challenge. Current methods treat this as a pattern recognition task over long conversation logs, leading to critically low step-level accuracy (below 17\%), which renders them impractical for debugging complex systems. Their core weakness is a fundamental inability to perform robust counterfactual reasoning: to determine if correcting a single action would have actually averted the task failure. To bridge this counterfactual inference gap, we introduce Abduct-Act-Predict (A2P) Scaffolding, a novel agent framework that transforms failure attribution from pattern recognition into a structured causal inference task. A2P explicitly guides a large language model through a formal three-step reasoning process within a single inference pass: (1) Abduction, to infer the hidden root causes behind an agent's actions; (2) Action, to define a minimal corrective intervention; and (3) Prediction, to simulate the subsequent trajectory and verify if the intervention resolves the failure. This structured approach leverages the holistic context of the entire conversation while imposing a rigorous causal logic on the model's analysis. Our extensive experiments on the Who\&When benchmark demonstrate its efficacy. On the Algorithm-Generated dataset, A2P achieves 47.46\% step-level accuracy, a 2.85times improvement over the 16.67\% of the baseline. On the more complex Hand-Crafted dataset, it achieves 29.31\% step accuracy, a 2.43times improvement over the baseline's 12.07\%. By reframing the problem through a causal lens, A2P Scaffolding provides a robust, verifiable, and significantly more accurate solution for automated failure attribution. Ours code are released at https://github.com/ResearAI/A2P.

Large language models are increasingly deployed as autonomous agents executing multi-step workflows in real-world software environments. However, existing agent benchmarks suffer from three critical limitations: (1) trajectory-opaque grading that checks only final outputs, (2) underspecified safety and robustness evaluation, and (3) narrow modality coverage and interaction paradigms. We introduce Claw-Eval, an end-to-end evaluation suite addressing all three gaps. It comprises 300 human-verified tasks spanning 9 categories across three groups (general service orchestration, multimodal perception and generation, and multi-turn professional dialogue). Every agent action is recorded through three independent evidence channels (execution traces, audit logs, and environment snapshots), enabling trajectory-aware grading over 2,159 fine-grained rubric items. The scoring protocol evaluates Completion, Safety, and Robustness, reporting Average Score, Pass@k, and Pass^k across three trials to distinguish genuine capability from lucky outcomes. Experiments on 14 frontier models reveal that: (1) trajectory-opaque evaluation is systematically unreliable, missing 44% of safety violations and 13% of robustness failures that our hybrid pipeline catches; (2) controlled error injection primarily degrades consistency rather than peak capability, with Pass^3 dropping up to 24% while Pass@3 remains stable; (3) multimodal performance varies sharply, with most models performing poorer on video than on document or image, and no single model dominating across all modalities. Beyond benchmarking, Claw-Eval highlights actionable directions for agent development, shedding light on what it takes to build agents that are not only capable but reliably deployable.

LLM agents act in external environments where each action changes the state that later decisions condition on, and where a single wrong step can waste interaction budget or trigger irreversible side effects long before the final failure is observed. Reliable deployment therefore requires step-level confidence estimation: a calibrated probability that each proposed action is productive, available before the action is executed. Existing LLM confidence estimators are designed to score a response from the given prompt, but agent confidence also depends on execution consequences: whether similar actions in similar situations actually advanced the task after the environment responded. We introduce the \method (\methodshort), a self-evolving critic framework in which an LLM critic accumulates evidence from its own past judgments and their observed consequences. After each trajectory, a hindsight LLM that sees the full execution feedback votes on whether each step was productive. The resulting pseudo-labels populate a memory bank from which related productive and unproductive experiences are retrieved into the critic's prompt whenever a similar step recurs. \methodshort requires no training and uses no ground truth step labels. Across three agent benchmarks and three critic backbones, \methodshort attains the best calibration (ECE and Brier) and ranking (AUC) in every dataset--critic combination, reducing ECE by up to 54% relative to the strongest training-free baseline.

Vision-language process reward models (VL-PRMs) are increasingly used to score intermediate reasoning steps and rerank candidates under test-time scaling. However, they often function as black-box judges: a low step score may reflect a genuine reasoning mistake or simply the verifier's misperception of the image. This entanglement between perception and reasoning leads to systematic false positives (rewarding hallucinated visual premises) and false negatives (penalizing correct grounded statements), undermining both reranking and error localization. We introduce Explicit Visual Premise Verification (EVPV), a lightweight verification interface that conditions step scoring on the reliability of the visual premises a step depends on. The policy is prompted to produce a step-wise visual checklist that makes required visual facts explicit, while a constraint extractor independently derives structured visual constraints from the input image. EVPV matches checklist claims against these constraints to compute a scalar visual reliability signal, and calibrates PRM step rewards via reliability gating: rewards for visually dependent steps are attenuated when reliability is low and preserved when reliability is high. This decouples perceptual uncertainty from logical evaluation without per-step tool calls. Experiments on VisualProcessBench and six multimodal reasoning benchmarks show that EVPV improves step-level verification and consistently boosts Best-of-N reranking accuracy over strong baselines. Furthermore, injecting controlled corruption into the extracted constraints produces monotonic performance degradation, providing causal evidence that the gains arise from constraint fidelity and explicit premise verification rather than incidental prompt effects. Code is available at: https://github.com/Qwen-Applications/EVPV-PRM

Establishing precise traceability between embedded systems datasheets and their corresponding code implementations remains a fundamental challenge in systems engineering, particularly for low-level software where manual mapping between specification documents and large code repositories is infeasible. Existing Traceability Link Recovery approaches primarily rely on lexical similarity and information retrieval techniques, which struggle to capture the semantic, structural, and symbol level relationships prevalent in embedded systems software. We present a hierarchical datasheet-to-code mapping methodology that employs large language models for semantic analysis while explicitly structuring the traceability process across multiple abstraction levels. Rather than performing direct specification-to-code matching, the proposed approach progressively narrows the search space through repository-level structure inference, file-level relevance estimation, and fine-grained symbollevel alignment. The method extends beyond function-centric mapping by explicitly covering macros, structs, constants, configuration parameters, and register definitions commonly found in systems-level C/C++ codebases. We evaluate the approach on multiple open-source embedded systems repositories using manually curated datasheet-to-code ground truth. Experimental results show substantial improvements over traditional information-retrieval-based baselines, achieving up to 73.3% file mapping accuracy. We significantly reduce computational overhead, lowering total LLM token consumption by 84% and end-to-end runtime by approximately 80%. This methodology supports automated analysis of large embedded software systems and enables downstream applications such as training data generation for systems-aware machine learning models, standards compliance verification, and large-scale specification coverage analysis.

To mitigate the potential misuse of large language models (LLMs), recent research has developed watermarking algorithms, which restrict the generation process to leave an invisible trace for watermark detection. Due to the two-stage nature of the task, most studies evaluate the generation and detection separately, thereby presenting a challenge in unbiased, thorough, and applicable evaluations. In this paper, we introduce WaterBench, the first comprehensive benchmark for LLM watermarks, in which we design three crucial factors: (1) For benchmarking procedure, to ensure an apples-to-apples comparison, we first adjust each watermarking method's hyper-parameter to reach the same watermarking strength, then jointly evaluate their generation and detection performance. (2) For task selection, we diversify the input and output length to form a five-category taxonomy, covering 9 tasks. (3) For evaluation metric, we adopt the GPT4-Judge for automatically evaluating the decline of instruction-following abilities after watermarking. We evaluate 4 open-source watermarks on 2 LLMs under 2 watermarking strengths and observe the common struggles for current methods on maintaining the generation quality. The code and data are available at https://github.com/THU-KEG/WaterBench.

Large Language Models (LLMs) are increasingly deployed as autonomous agents that reason, use tools, and act over multiple steps. Yet most hallucination benchmarks still evaluate only the final output, missing failures that originate in intermediate Thought-Action-Observation steps. We present Trajel, a dataset and evaluation framework for auditing trajectory-level hallucinations in multi-agent industrial workflows. Trajel introduces a five-type hallucination taxonomy (factual, referential, logical, procedural, and scope-based) over expert-annotated agent traces from AssetOpsBench. We benchmark supervised detection models at the subtask, trajectory, and long-context levels. Our results show that the most common failure modes are missed by existing benchmarks, that nearly half of hallucinated trajectories involve multiple types at once, and that automated detectors with high binary accuracy still misclassify the subtlest types. Trajectory-aware detection significantly outperforms standard post-hoc verification, making taxonomy-grounded evaluation necessary for safer agentic deployment.

Vision-Language Navigation in Continuous Environments (VLN-CE) requires agents to learn complex reasoning from long-horizon human interactions. While Multi-modal Large Language Models (MLLMs) have driven recent progress, current training paradigms struggle to balance generalization capability, error recovery and training stability. Specifically, (i) policies derived from SFT suffer from compounding errors, struggling to recover from out-of-distribution states, and (ii) Reinforcement Fine-Tuning (RFT) methods e.g. GRPO are bottlenecked by sparse outcome rewards. Their binary feedback fails to assign credit to individual steps, leading to gradient signal collapse in failure dominant batches. To address these challenges, we introduce Step-Aware Contrastive Alignment (SACA), a framework designed to extract dense supervision from imperfect trajectories. At its core, the Perception-Grounded Step-Aware auditor evaluates progress step-by-step, disentangling failed trajectories into valid prefixes and exact divergence points. Leveraging these signals, Scenario-Conditioned Group Construction mechanism dynamically routes batches to specialized resampling and optimization strategies. Extensive experiments on VLN-CE benchmarks demonstrate that SACA achieves state-of-the-art performance.

A fundamental skill among human developers is the ability to understand and reason about program execution. As an example, a programmer can mentally simulate code execution in natural language to debug and repair code (aka. rubber duck debugging). However, large language models (LLMs) of code are typically trained on the surface textual form of programs, thus may lack a semantic understanding of how programs execute at run-time. To address this issue, we propose NExT, a method to teach LLMs to inspect the execution traces of programs (variable states of executed lines) and reason about their run-time behavior through chain-of-thought (CoT) rationales. Specifically, NExT uses self-training to bootstrap a synthetic training set of execution-aware rationales that lead to correct task solutions (e.g., fixed programs) without laborious manual annotation. Experiments on program repair tasks based on MBPP and HumanEval demonstrate that NExT improves the fix rate of a PaLM 2 model, by 26.1% and 14.3% absolute, respectively, with significantly improved rationale quality as verified by automated metrics and human raters. Our model can also generalize to scenarios where program traces are absent at test-time.

We introduce behavioral fidelity -- a third evaluation dimension for synthetic tabular data that measures whether generated data preserves the temporal, sequential, and structural behavioral patterns that distinguish real-world entity activity. Existing frameworks evaluate statistical fidelity (marginal distributions and correlations) and downstream utility (classifier AUROC on synthetic-trained models), but neither tests for the behavioral signals that operational detection and analysis systems actually rely on. We formalize a taxonomy of four behavioral fraud patterns (P1-P4) covering inter-event timing, burst structure, multi-account graph motifs, and velocity-rule trigger rates; define a degradation ratio metric calibrated to a real-data noise floor (1.0 = matches real variability, k = k-times worse); and prove that row-independent generators -- the dominant paradigm -- are structurally incapable of reproducing P3 graph motifs (Proposition 1) and produce non-positive within-entity IET autocorrelation (Proposition 2), making the positive burst fingerprint of fraud sequences unachievable regardless of architecture or training data size. We benchmark CTGAN, TVAE, GaussianCopula, and TabularARGN on IEEE-CIS Fraud Detection and the Amazon Fraud Dataset. All four fail severely: on IEEE-CIS composite degradation ratios range from 24.4x (TVAE) to 39.0x (GaussianCopula); on Amazon FDB, row-independent generators score 81.6-99.7x, while TabularARGN achieves 17.2x. We document generator-specific failure modes and their resolutions. The P1-P4 framework extends to any domain with entity-level sequential tabular data, including healthcare and network security. We release our evaluation framework as open source.

Deploying Large Language Models (LLMs) for regulatory compliance demands rigorous traceability via comprehensive citations across multi-tiered authority structures. Unlike traditional multi-hop or legal QA, this task requires structured procedural lookups and evidence-set closure rather than entity resolution or case-law reasoning. Existing RAG systems struggle here due to flattened citation edges, fragmented retrieval expansions, and fragile post-hoc attribution. We formalize Regulatory Compliance QA with RegOps-Bench, a novel benchmark featuring an Operational Knowledge Graph derived from complex national R\&D regulations. To address these bottlenecks, we propose RefWalk, a unified framework driven by a shared topic anchor. RefWalk traverses cross-document citations, fuses multi-view candidates via max-based aggregation, and enforces per-rule attribution to explicitly map claims to sources. We establish a strong baseline with substantial improvements in retrieval recall and citation accuracy. Finally, a contrastive evaluation on a U.S. health compliance dataset (HIPAA) reveals that existing systems exhibit saturation on flat-structure rules, underscoring the need for RegOps-Bench. Our code is available at https://github.com/yeongjoonJu/RefWalk.

In the realm of large language models (LLMs), the ability of models to accurately follow instructions is paramount as more agents and applications leverage LLMs for construction, where the complexity of instructions are rapidly increasing. However, on the one hand, there is only a certain amount of complex instruction evaluation data; on the other hand, there are no dedicated algorithms to improve the ability to follow complex instructions. To this end, this paper introduces TRACE, a benchmark for improving and evaluating the complex instructionfollowing ability, which consists of 120K training data and 1K evaluation data. Furthermore, we propose IOPO (Input-Output Preference Optimization) alignment method which takes both input and output preference pairs into consideration, where LLMs not only rapidly align with response preferences but also meticulously explore the instruction preferences. Extensive experiments on both in-domain and outof-domain datasets confirm the effectiveness of IOPO, showing 8.15%, 2.18% improvements on in-domain data and 6.29%, 3.13% on outof-domain data compared to SFT and DPO respectively.

There are substantial instructional videos on the Internet, which provide us tutorials for completing various tasks. Existing instructional video datasets only focus on specific steps at the video level, lacking experiential guidelines at the task level, which can lead to beginners struggling to learn new tasks due to the lack of relevant experience. Moreover, the specific steps without guidelines are trivial and unsystematic, making it difficult to provide a clear tutorial. To address these problems, we present the GUIDE (Guideline-Guided) dataset, which contains 3.5K videos of 560 instructional tasks in 8 domains related to our daily life. Specifically, we annotate each instructional task with a guideline, representing a common pattern shared by all task-related videos. On this basis, we annotate systematic specific steps, including their associated guideline steps, specific step descriptions and timestamps. Our proposed benchmark consists of three sub-tasks to evaluate comprehension ability of models: (1) Step Captioning: models have to generate captions for specific steps from videos. (2) Guideline Summarization: models have to mine the common pattern in task-related videos and summarize a guideline from them. (3) Guideline-Guided Captioning: models have to generate captions for specific steps under the guide of guideline. We evaluate plenty of foundation models with GUIDE and perform in-depth analysis. Given the diversity and practicality of GUIDE, we believe that it can be used as a better benchmark for instructional video comprehension.

We evaluate the propensity of frontier models to sabotage or refuse to assist with safety research when deployed as AI research agents within a frontier AI company. We apply two complementary evaluations to four Claude models (Mythos Preview, Opus 4.7 Preview, Opus 4.6, and Sonnet 4.6): an unprompted sabotage evaluation testing model behaviour with opportunities to sabotage safety research, and a sabotage continuation evaluation testing whether models continue to sabotage when placed in trajectories where prior actions have started undermining research. We find no instances of unprompted sabotage across any model, with refusal rates close to zero for Mythos Preview and Opus 4.7 Preview, though all models sometimes only partially completed tasks. In the continuation evaluation, Mythos Preview actively continues sabotage in 7% of cases (versus 3% for Opus 4.6, 4% for Sonnet 4.6, and 0% for Opus 4.7 Preview), and exhibits reasoning-output discrepancy in the majority of these cases, indicating covert sabotage reasoning. Our evaluation framework builds on Petri, an open-source LLM auditing tool, with a custom scaffold running models inside Claude Code, alongside an iterative pipeline for generating realistic sabotage trajectories. We measure both evaluation awareness and a new form of situational awareness termed "prefill awareness", the capability to recognise that prior trajectory content was not self-generated. Opus 4.7 Preview shows notably elevated unprompted evaluation awareness, while prefill awareness remains low across all models. Finally, we discuss limitations including evaluation awareness confounds, limited scenario coverage, and untested pathways to risk beyond safety research sabotage.

In recent years, trace generation has emerged as a significant challenge within the Process Mining community. Deep Learning (DL) models have demonstrated accuracy in reproducing the features of the selected processes. However, current DL generative models are limited in their ability to adapt the learned distributions to generate data samples based on specific conditions or attributes. This limitation is particularly significant because the ability to control the type of generated data can be beneficial in various contexts, enabling a focus on specific behaviours, exploration of infrequent patterns, or simulation of alternative 'what-if' scenarios. In this work, we address this challenge by introducing a conditional model for process data generation based on a conditional variational autoencoder (CVAE). Conditional models offer control over the generation process by tuning input conditional variables, enabling more targeted and controlled data generation. Unlike other domains, CVAE for process mining faces specific challenges due to the multiperspective nature of the data and the need to adhere to control-flow rules while ensuring data variability. Specifically, we focus on generating process executions conditioned on control flow and temporal features of the trace, allowing us to produce traces for specific, identified sub-processes. The generated traces are then evaluated using common metrics for generative model assessment, along with additional metrics to evaluate the quality of the conditional generation

Chain-of-Thought (CoT) prompting has become the de facto method to elicit reasoning capabilities from large language models (LLMs). However, to mitigate hallucinations in CoT that are notoriously difficult to detect, current methods such as process reward models (PRMs) or self-consistency operate as opaque boxes and do not provide checkable evidence for their judgments, possibly limiting their effectiveness. To address this issue, we draw inspiration from the idea that "the gold standard for supporting a mathematical claim is to provide a proof". We propose a retrospective, step-aware formal verification framework Safe. Rather than assigning arbitrary scores, we strive to articulate mathematical claims in formal mathematical language Lean 4 at each reasoning step and provide formal proofs to identify hallucinations. We evaluate our framework Safe across multiple language models and various mathematical datasets, demonstrating a significant performance improvement while offering interpretable and verifiable evidence. We also propose FormalStep as a benchmark for step correctness theorem proving with 30,809 formal statements. To the best of our knowledge, our work represents the first endeavor to utilize formal mathematical language Lean 4 for verifying natural language content generated by LLMs, aligning with the reason why formal mathematical languages were created in the first place: to provide a robust foundation for hallucination-prone human-written proofs.

Software development agents powered by large language models (LLMs) have shown great promise in automating tasks like environment setup, issue solving, and program repair. Unfortunately, understanding and debugging such agents remain challenging due to their complex and dynamic nature. Developers must reason about trajectories of LLM queries, tool calls, and code modifications, but current techniques reveal little of this intermediate process in a comprehensible format. The key insight of this paper is that debugging software development agents shares many similarities with conventional debugging of software programs, yet requires a higher level of abstraction that raises the level from low-level implementation details to high-level agent actions. Drawing on this insight, we introduce AgentStepper, the first interactive debugger for LLM-based software engineering agents. AgentStepper enables developers to inspect, control, and interactively manipulate agent trajectories. AgentStepper represents trajectories as structured conversations among an LLM, the agent program, and tools. It supports breakpoints, stepwise execution, and live editing of prompts and tool invocations, while capturing and displaying intermediate repository-level code changes. Our evaluation applies AgentStepper to three state-of-the-art software development agents, ExecutionAgent, SWE-Agent, and RepairAgent, showing that integrating the approach into existing agents requires minor code changes (39-42 edited lines). Moreover, we report on a user study with twelve participants, indicating that AgentStepper improves the ability of participants to interpret trajectories (64% vs. 67% mean performance) and identify bugs in the agent's implementation (17% vs. 60% success rate), while reducing perceived workload (e.g., frustration reduced from 5.4/7.0 to 2.4/7.0) compared to conventional tools.

State-of-the-art large language models (LLMs) have demonstrated impressive code generation capabilities but struggle with real-world software engineering tasks, such as revising source code to address code reviews, hindering their practical use. Code review comments are often implicit, ambiguous, and colloquial, requiring models to grasp both code and human intent. This challenge calls for evaluating large language models' ability to bridge both technical and conversational contexts. While existing work has employed the automated code refinement (ACR) task to resolve these comments, current evaluation methods fall short, relying on text matching metrics that provide limited insight into model failures and remain susceptible to training data contamination. To address these limitations, we introduce a novel evaluation benchmark, CodeReviewQA that enables us to conduct fine-grained assessment of model capabilities and mitigate data contamination risks. In CodeReviewQA, we decompose the generation task of code refinement into three essential reasoning steps: change type recognition (CTR), change localisation (CL), and solution identification (SI). Each step is reformulated as multiple-choice questions with varied difficulty levels, enabling precise assessment of model capabilities, while mitigating data contamination risks. Our comprehensive evaluation spans 72 recently released large language models on 900 manually curated, high-quality examples across nine programming languages. Our results show that CodeReviewQA is able to expose specific model weaknesses in code review comprehension, disentangled from their generative automated code refinement results.

LLM agents increasingly rely on agent harness: the runtime infrastructure around the base model that defines execution environments, tool interfaces, context, lifecycle orchestration, observability, verification, and governance. Existing self-improving agents and automatic harness evolution methods mainly improve agents through runtime supervision, prompt optimization, workflow search, or harness modification based on final outcomes. However, they often fail to diagnose where the responsible evidence lies in failed trajectories and which harness implementation mechanism causes the unreliable behavior, resulting in broad, indirect, or poorly scoped changes. This paper proposes HarnessFix, a trace-grounded and diagnosis-driven framework for repairing agent harnesses. HarnessFix compiles raw execution traces and harness artifacts into a Harness-aware Trace Intermediate Representation (HTIR), which normalizes fragmented trajectory evidence, captures step-level data-flow and control-flow relations, and aligns runtime steps with the harness artifacts that shape their behavior. It then attributes failures to responsible steps and harness artifacts, and consolidates recurring diagnoses into repair-oriented flaw records. Finally, HarnessFix maps these records to scoped repair operators, generates patches under flaw-specific repair specifications, and accepts them through regression-aware validation. We evaluate HarnessFix on four popular benchmarks, and results show that it improves the performance over the initial harnesses by 6.3% to 18.4%, significantly outperforming human-designed and self-evolution baselines. HarnessFix highlights the value of treating failed trajectories not only as feedback signals, but also as structured evidence for diagnosing and repairing the harness mechanisms behind agent failures.

Process or step-wise supervision has played a crucial role in advancing complex multi-step reasoning capabilities of Large Language Models (LLMs). However, efficient, high-quality automated process annotation remains a significant challenge. To address this, we introduce Single-Pass Annotation with Reference-Guided Evaluation (SPARE), a novel structured framework that enables single-pass, per-step annotation by aligning each solution step to one or multiple steps in a reference solution, accompanied by explicit reasoning for evaluation. We show that reference-guided step-level evaluation effectively facilitates process supervision on four datasets spanning three domains: mathematical reasoning, multi-hop compositional question answering, and spatial reasoning. We demonstrate that SPARE, when compared to baselines, improves reasoning performance when used for: (1) fine-tuning models in an offline RL setup for inference-time greedy-decoding, and (2) training reward models for ranking/aggregating multiple LLM-generated outputs. Additionally, SPARE achieves competitive performance on challenging mathematical datasets while offering 2.6 times greater efficiency, requiring only 38% of the runtime, compared to tree search-based automatic annotation. The codebase, along with a trained SPARE-PRM model, is publicly released to facilitate further research and reproducibility.

LLM-based agents have demonstrated strong capabilities in data-intensive analytical tasks, yet their outputs are rarely verifiable: a reliance on linear text trajectories makes their reasoning difficult to audit. In particular, deterministic computations over raw data and semantic deductions over natural-language claims are often entangled in an unstructured stream, leaving numerical conclusions hard to reproduce and qualitative judgments hard to inspect. To address this, we propose VeriGraph, a traceable neuro-symbolic reasoning framework that enables agents to construct an explicit heterogeneous evidence directed acyclic graph (DAG) during execution. VeriGraph introduces three evidence-expansion primitives, namely computational, grounding, and derivational expansion, to connect raw data, interpreter variables, computed results, and natural-language claims in a unified graph. Under this formulation, structural traceability is reduced to graph reachability from raw data sources to terminal claims, while semantic support is measured by claim-level evidence evaluation. To improve graph construction, we further design a graph-based policy optimization strategy with a composite reward that jointly supervises answer correctness, computational integrity, and derivational coherence. Experiments on four benchmarks show that VeriGraph-8B achieves the highest overall score among all baselines. More importantly, VeriGraph produces auditable evidence graphs with substantially stronger claim grounding, achieving a 87.61\% Grounding Rate under our claim-level evidence support evaluation. These results suggest that explicit evidence-graph construction is a promising path toward verifiable data-analytic agents. Our code is available at https://github.com/ignorejjj/VeriGraph.

Resolving real-world software engineering (SWE) issues with autonomous agents requires complex, long-horizon reasoning. Current pipelines are bottlenecked by unoptimized demonstration data, sparse execution rewards, and computationally prohibitive inference scaling, which collectively exacerbate token bloat, reward hacking, and policy degradation. We present SWE-TRACE (Trajectory Reduction and Agentic Criteria Evaluation), a unified framework optimizing the SWE agent lifecycle across data curation, reinforcement learning (RL), and test-time inference. First, we introduce an LLM multi-task cascading method, utilizing stepwise oracle verification to distill a 60K-instance Supervised Fine-Tuning (SFT) corpus strictly biased toward token-efficient, shortest-path trajectories. Second, to overcome the instability of sparse outcome rewards, we design a MemoryAugmented Agentic RL pipeline featuring a Rubric-Based Process Reward Model (PRM). An auxiliary Rubric-Agent provides dense, fine-grained heuristic feedback on intermediate steps, guiding the model through long-horizon tasks. Finally, we bridge training and inference by repurposing the PRM for heuristic-guided Test-Time Scaling (TTS). By dynamically evaluating and pruning action candidates at each step, SWE-TRACE achieves superior search efficiency without the latency overhead of standard parallel sampling. Extensive experiments on standard SWE benchmarks demonstrate that SWE-TRACE significantly advances the state-of-the-art, maximizing resolution rates while drastically reducing both token consumption and inference latency.

The emergence of long-context language models with context windows extending to millions of tokens has created new opportunities for sophisticated code understanding and software development evaluation. We propose LoCoBench, a comprehensive benchmark specifically designed to evaluate long-context LLMs in realistic, complex software development scenarios. Unlike existing code evaluation benchmarks that focus on single-function completion or short-context tasks, LoCoBench addresses the critical evaluation gap for long-context capabilities that require understanding entire codebases, reasoning across multiple files, and maintaining architectural consistency across large-scale software systems. Our benchmark provides 8,000 evaluation scenarios systematically generated across 10 programming languages, with context lengths spanning 10K to 1M tokens, a 100x variation that enables precise assessment of long-context performance degradation in realistic software development settings. LoCoBench introduces 8 task categories that capture essential long-context capabilities: architectural understanding, cross-file refactoring, multi-session development, bug investigation, feature implementation, code comprehension, integration testing, and security analysis. Through a 5-phase pipeline, we create diverse, high-quality scenarios that challenge LLMs to reason about complex codebases at unprecedented scale. We introduce a comprehensive evaluation framework with 17 metrics across 4 dimensions, including 8 new evaluation metrics, combined in a LoCoBench Score (LCBS). Our evaluation of state-of-the-art long-context models reveals substantial performance gaps, demonstrating that long-context understanding in complex software development represents a significant unsolved challenge that demands more attention. LoCoBench is released at: https://github.com/SalesforceAIResearch/LoCoBench.

The automatic evaluation of instruction following typically involves using large language models (LLMs) to assess response quality. However, there is a lack of comprehensive evaluation of these LLM-based evaluators across two dimensions: the base LLMs and the evaluation protocols. Therefore, we present a thorough meta-evaluation of instruction following, including 25 base LLMs and 15 recently proposed evaluation protocols, on 4 human-annotated datasets, assessing the evaluation accuracy of the LLM-evaluators. Our evaluation allows us to identify the best-performing base LLMs and evaluation protocols with a high degree of robustness. Moreover, our large-scale evaluation reveals: (1) Base LLM performance ranking remains largely consistent across evaluation protocols, with less capable LLMs showing greater improvement from protocol enhancements; (2) Robust evaluation of evaluation protocols requires many base LLMs with varying capability levels, as protocol effectiveness can depend on the base LLM used; (3) Evaluation results on different datasets are not always consistent, so a rigorous evaluation requires multiple datasets with distinctive features. We release our meta-evaluation suite ReIFE, which provides the codebase and evaluation result collection for more than 500 LLM-evaluator configurations, to support future research in instruction-following evaluation.

We introduce an approach aimed at enhancing the reasoning capabilities of Large Language Models (LLMs) through an iterative preference learning process inspired by the successful strategy employed by AlphaZero. Our work leverages Monte Carlo Tree Search (MCTS) to iteratively collect preference data, utilizing its look-ahead ability to break down instance-level rewards into more granular step-level signals. To enhance consistency in intermediate steps, we combine outcome validation and stepwise self-evaluation, continually updating the quality assessment of newly generated data. The proposed algorithm employs Direct Preference Optimization (DPO) to update the LLM policy using this newly generated step-level preference data. Theoretical analysis reveals the importance of using on-policy sampled data for successful self-improving. Extensive evaluations on various arithmetic and commonsense reasoning tasks demonstrate remarkable performance improvements over existing models. For instance, our approach outperforms the Mistral-7B Supervised Fine-Tuning (SFT) baseline on GSM8K, MATH, and ARC-C, with substantial increases in accuracy to 81.8% (+5.9%), 34.7% (+5.8%), and 76.4% (+15.8%), respectively. Additionally, our research delves into the training and inference compute tradeoff, providing insights into how our method effectively maximizes performance gains. Our code is publicly available at https://github.com/YuxiXie/MCTS-DPO.

Agent systems powered by large language models (LLMs) have demonstrated impressive performance on repository-level code-generation tasks. However, for tasks such as website codebase generation, which depend heavily on visual effects and user-interaction feedback, current code agents rely only on simple code execution for feedback and verification. This approach fails to capture the actual quality of the generated code. In this paper, we propose WebGen-Agent, a novel website-generation agent that leverages comprehensive and multi-level visual feedback to iteratively generate and refine the website codebase. Detailed and expressive text descriptions and suggestions regarding the screenshots and GUI-agent testing of the websites are generated by a visual language model (VLM), together with scores that quantify their quality. The screenshot and GUI-agent scores are further integrated with a backtracking and select-best mechanism, enhancing the performance of the agent. Utilizing the accurate visual scores inherent in the WebGen-Agent workflow, we further introduce Step-GRPO with Screenshot and GUI-agent Feedback to improve the ability of LLMs to act as the reasoning engine of WebGen-Agent. By using the screenshot and GUI-agent scores at each step as the reward in Step-GRPO, we provide a dense and reliable process supervision signal, which effectively improves the model's website-generation ability. On the WebGen-Bench dataset, WebGen-Agent increases the accuracy of Claude-3.5-Sonnet from 26.4% to 51.9% and its appearance score from 3.0 to 3.9, outperforming the previous state-of-the-art agent system. Additionally, our Step-GRPO training approach increases the accuracy of Qwen2.5-Coder-7B-Instruct from 38.9% to 45.4% and raises the appearance score from 3.4 to 3.7.

Current benchmarks for Large Language Models (LLMs) primarily focus on performance metrics, often failing to capture the nuanced behavioral characteristics that differentiate them. This paper introduces a novel ``Behavioral Fingerprinting'' framework designed to move beyond traditional evaluation by creating a multi-faceted profile of a model's intrinsic cognitive and interactive styles. Using a curated Diagnostic Prompt Suite and an innovative, automated evaluation pipeline where a powerful LLM acts as an impartial judge, we analyze eighteen models across capability tiers. Our results reveal a critical divergence in the LLM landscape: while core capabilities like abstract and causal reasoning are converging among top models, alignment-related behaviors such as sycophancy and semantic robustness vary dramatically. We further document a cross-model default persona clustering (ISTJ/ESTJ) that likely reflects common alignment incentives. Taken together, this suggests that a model's interactive nature is not an emergent property of its scale or reasoning power, but a direct consequence of specific, and highly variable, developer alignment strategies. Our framework provides a reproducible and scalable methodology for uncovering these deep behavioral differences. Project: https://github.com/JarvisPei/Behavioral-Fingerprinting

Computer use agents (CUAs) can operate real-world digital interfaces but remain difficult to train due to the high cost of graphical user interface (GUI) interaction and the scarcity of high-quality trajectory data. Existing datasets rely on human demonstrations, limiting scalability. A natural alternative is to synthesize data from strong CUAs, yet their rollouts are highly noisy, with incorrect or suboptimal actions consisting a large proportion of the steps, making naive imitation ineffective. To tackle this challenge, we introduce a scalable data synthesis pipeline that transforms noisy rollouts into reliable supervision without human annotation. The core idea is step-level filtering, which evaluates actions individually to retain only correct steps, complemented by reasoning augmentation for improved planning. Using this pipeline, we construct WebSTAR, a dataset of 13.3K trajectories and 100K graded, reasoning-rich steps synthesized from OpenAI's computer-use-preview model. We train Qwen-2.5-VL-Instruct models (7B and 32B) on WebSTAR. On WebVoyager, our 7B model surpasses SoTA open-source CUA model UI-TARS-1.5-7B by more than 15% with only supervised finetuning. Building on step-level grading, we further create WebSCORE, a dataset of graded step-level actions, and train StepRM, a 7B multimodal reward model distilled from o4-mini, which matches its grading quality while being far more efficient to deploy at scale. Our results establish step-level filtering as a key principle for scalable CUA training and construct two new datasets (WebSTAR, WebSCORE) and a lightweight reward model (StepRM) as practical tools to advance robust and efficient CUAs.

Failure attribution for LLM-based agentic systems, i.e., identifying which steps in a failure trajectory caused the task to fail, is critical for debugging and improving these systems. Existing approaches either rely on prompting-based pipelines, which are computationally expensive, or require post-training on failure trajectories with step-level error annotations, which are costly to collect and difficult to scale. We argue that a practical failure attribution model should be lightweight and trainable without step-level supervision on failure data. To this end, we address unsupervised failure attribution, i.e., training exclusively on successful trajectories and identifying error steps at inference time given a failure trajectory. We propose OAT, which casts this problem as one-class learning with neural controlled differential equations, modeling the dynamical pattern of successful trajectories in latent space. At inference time, each step in a failure trajectory is assigned an anomaly score based on its deviation from the dynamics learned on successful trajectories, which is then used to form a set of error steps. With training on only 100 successful trajectories, experiments show that OAT is 200--5000 times faster than prompting-based baselines, and, at the same time, consistently outperforms them in both in-domain and out-of-distribution datasets with +20% and +7% F1 scores, respectively, demonstrating that OAT is a promising and efficient direction for diagnosing agentic system failures.

Bug reports serve as task specifications for repository-level automated program repair (APR) agents, but they often describe only the observed failure and omit repair-relevant information such as the failure-inducing behavior, behavioral requirement, and implementation scope. As a result, a repair agent may inspect irrelevant code, infer an incorrect requirement, or generate a patch that addresses the reported symptom without restoring the intended repository behavior. We present TrajSpec, a trajectory-guided approach for repository-supported bug report specification refinement. Given an original report and a pre-fix repository, TrajSpec runs a trajectory-collection agent and uses the resulting unverified trajectory as a source of trajectory-derived specification evidence. It organizes this evidence into a three-level representation consisting of a high-level interpretation of the issue, diagnostic findings supporting that interpretation, and concrete repository observations. TrajSpec then generates a draft refined report and applies repository-based review to remove unsupported claims, revise uncertain claims, and add repository-supported details. We evaluate TrajSpec on all 300 SWE-Bench Lite instances using Mini-SWE-Agent V2. TrajSpec's refined reports improve Pass@1 from 41.00% to 59.67% with GPT-5-mini and from 54.67% to 64.33% with MiniMax M2.5. On a stratified sample of 100 instances, TrajSpec's refined reports also improve Pass@1 from 41.00% to 71.00% with Agentless and from 47.00% to 72.00% with AutoCodeRover. Ablation results show that removing repository-based review or the hierarchical evidence representation reduces Pass@1 from 59.67% to 48.00% and 47.67%, respectively. Overall, TrajSpec provides actionable repository-supported context that consistently improves repair performance.

Large language model agents have exhibited exceptional performance across a range of complex interactive tasks. Recent approaches have utilized tuning with expert trajectories to enhance agent performance, yet they primarily concentrate on outcome rewards, which may lead to errors or suboptimal actions due to the absence of process supervision signals. In this paper, we introduce the Iterative step-level Process Refinement (IPR) framework, which provides detailed step-by-step guidance to enhance agent training. Specifically, we adopt the Monte Carlo method to estimate step-level rewards. During each iteration, the agent explores along the expert trajectory and generates new actions. These actions are then evaluated against the corresponding step of expert trajectory using step-level rewards. Such comparison helps identify discrepancies, yielding contrastive action pairs that serve as training data for the agent. Our experiments on three complex agent tasks demonstrate that our framework outperforms a variety of strong baselines. Moreover, our analytical findings highlight the effectiveness of IPR in augmenting action efficiency and its applicability to diverse models.

Some security-development settings require local models that map an objective to an ordered, checkable plan. We present a low-rank decomposition adapter and release a case-disjoint corpus with 24 authored decompositions and 83 derived next-step examples across 24 satellite-security cases. To prevent reference-plan leakage, prompts contain one objective and eight shuffled SPARTA candidates. The evaluation candidate sets are constructed with oracle inclusion of every reference technique, so this study measures selection from a controlled set, not retrieval. On six fixed cases and five decoding seeds, mean scores vary across adapters and prompted Qwen2.5 baselines; a separate greedy check over three adapter training seeds shows non-negligible training variance. A 24-fold greedy leave-one-case-out control at the two compact sizes likewise shows tradeoffs: at 1.5B, adapter precision is 0.583 against 0.480 for two-shot prompting, while recall is 0.586 against 0.660. Case-resampling intervals for adapter precision relative to the strongest prompted baseline cross zero at 1.5B and 7B. The adapter is also substantially more format-compliant, which prevents attributing all score differences to selection. Autoregressive next-step accuracy is 0.08, 0.06, and 0.29 at 0.5B, 1.5B, and 7B. We therefore offer the formulation, leakage-controlled dataset, reproducibility package, and descriptive analysis as a proof of concept, not evidence of a reliable stand-alone decomposition system or a general adapter advantage. A separate diagnostic tests contract reuse.

Effective incident management in large-scale IT systems relies on troubleshooting guides (TSGs), but their manual execution is slow and error-prone. While recent advances in LLMs offer promise for automating incident management tasks, existing LLM-based solutions lack specialized support for several key challenges, including managing TSG quality issues, interpreting complex control flow, handling data-intensive queries, and exploiting execution parallelism. We first conducted an empirical study on 92 real-world TSGs, and, guided by our findings, we present StepFly, a novel end-to-end agentic framework for troubleshooting guide automation. Our approach features a three-stage workflow: the first stage provides a comprehensive guide together with a tool, TSG Mentor, to assist SREs in improving TSG quality; the second stage performs offline preprocessing using LLMs to extract structured execution DAGs from unstructured TSGs and to create dedicated Query Preparation Plugins (QPPs); and the third stage executes online using a DAG-guided scheduler-executor framework with a memory system to guarantee correct workflow and support parallel execution of independent steps. Our empirical evaluation on a collection of real-world TSGs and incidents demonstrates that StepFly achieves a ~94% success rate on GPT-4.1, outperforming baselines with less time and token consumption. Furthermore, it achieves a remarkable execution time reduction of 32.9% to 70.4% for parallelizable TSGs.

Vision-Language Models (VLMs) often lose visual grounding during multi-step reasoning: as reasoning chains grow longer, later inference steps rely increasingly on language priors rather than image evidence. We identify a consistent benchmark-level signature associated with this degradation: across 2,510 re-examined samples from four benchmarks, attention entropy over image tokens typically decreases during Round 1 and rises again after image re-injection. However, we find that effective visual re-examination requires two complementary ingredients: image re-injection and targeted self-diagnosis. Without targeted diagnosis, re-examination can even hurt performance, whereas accurate self-diagnosis yields substantial gains -- a swing of several points on key benchmarks, indicating that diagnostic quality is a key factor in whether re-examination helps or hurts in our setting. We present ReGround, a two-stage framework that teaches VLMs to self-diagnose grounding failures and selectively re-examine visual evidence, without architectural modifications or external tools. Through capability bootstrapping, a stronger variant from the same model family provides diagnostic scaffolding only during data construction, while the policy model learns to diagnose autonomously at inference time and retains most of the assisted gains. Experiments on eight benchmarks across two VLM backbones demonstrate consistent gains, especially on visually intensive multi-step reasoning tasks, while incurring only modest inference overhead relative to tool-augmented baselines. Project page: https://sespoir.github.io/reground-page/ . Code: https://github.com/sespoir/ReGround .

Large Language Models (LLMs) have shown impressive performance across a wide array of tasks involving both structured and unstructured textual data. Recent results on various benchmarks for code generation, repair, or completion suggest that certain models have programming abilities comparable to or even surpass humans. In this work, we demonstrate that high performance on such benchmarks does not correlate to humans' innate ability to understand structural control flow in code. To this end, we extract solutions from the HumanEval benchmark, which the relevant models perform strongly on, and trace their execution path using function calls sampled from the respective test set. Using this dataset, we investigate the ability of seven state-of-the-art LLMs to match the execution trace and find that, despite their ability to generate semantically identical code, they possess limited ability to trace execution paths, especially for longer traces and specific control structures. We find that even the top-performing model, Gemini, can fully and correctly generate only 47% of HumanEval task traces. Additionally, we introduce a subset for three key structures not contained in HumanEval: Recursion, Parallel Processing, and Object-Oriented Programming, including concepts like Inheritance and Polymorphism. Besides OOP, we show that none of the investigated models achieve an accuracy over 5% on the relevant traces. Aggregating these specialized parts with HumanEval tasks, we present Benchmark CoCoNUT: Code Control Flow for Navigation Understanding and Testing, which measures a model's ability to trace execution of code upon relevant calls, including advanced structural components. We conclude that current LLMs need significant improvement to enhance code reasoning abilities. We hope our dataset helps researchers bridge this gap.

Large language model (LLM) agents increasingly rely on reusable external skills to solve long-horizon interactive tasks. Existing training-free skill adaptation pipelines usually update skills from full trajectories or session-level feedback, which makes failure attribution coarse and often produces unstable or overly broad revisions. We propose SkillAdaptor, a training-free step-level skill adaptation framework with explicit failure attribution, and it can plug into OpenClaw-class agent harnesses. Given a failed trajectory, SkillAdaptor identifies a first actionable fault step, links responsibility to candidate skills, and applies targeted updates under explicit acceptance checks while keeping the backbone frozen. We evaluate on WebShop, PinchBench, and Claw-Eval with Kimi-K2.5, GLM-5, and GPT-5.2. SkillAdaptor improves over no-skill and skill-adaptation baselines on all three suites, with the largest single-metric improvements of +1.5 points on PinchBench Avg Score%, +1.8 on Claw-Eval Avg Score, and +1.7 on WebShop success rate. These results indicate that step-level attribution supports more stable and auditable training-free skill maintenanceThe code will be released at https://github.com/zjunlp/SkillAdaptor..

Test-time scaling improves the reasoning performance of large language models but often results in token-inefficient overthinking, where models continue reasoning beyond what is necessary for a correct answer. Existing dynamic early-exit methods typically rely on single-step confidence signals, which are often unreliable for detecting reasoning convergence in multi-step settings. To mitigate this limitation, we propose TRACE, a training-free framework for efficient test-time scaling that determines when to terminate reasoning based on temporal aggregation of multi-step evidence rather than instantaneous signals. TRACE detects reasoning convergence over time by aggregating two complementary signals across recent reasoning steps: answer consistency, capturing the persistence of predicted answers, and confidence trajectory, modeling the temporal evolution of model confidence. Benefiting from these two factors, TRACE can accurately determine whether the reasoning process has converged, thereby promptly halting inference and effectively avoiding redundant reasoning steps. Extensive experiments on multiple challenging benchmarks show that TRACE reduces reasoning token usage by 25-30% on average while maintaining accuracy within 1-2% of full-length reasoning, consistently outperforming existing dynamic reasoning methods.

Evaluating large language models (LLMs) on final-answer correctness is the dominant paradigm. This approach, however, provides a coarse signal for model improvement and overlooks the quality of the underlying reasoning process. We argue that a more granular evaluation of reasoning offers a more effective path to building robust models. We decompose reasoning quality into two dimensions: relevance and coherence. Relevance measures if a step is grounded in the problem; coherence measures if it follows logically from prior steps. To measure these aspects reliably, we introduce causal stepwise evaluation (CaSE). This method assesses each reasoning step using only its preceding context, which avoids hindsight bias. We validate CaSE against human judgments on our new expert-annotated benchmarks, MRa-GSM8K and MRa-MATH. More importantly, we show that curating training data with CaSE-evaluated relevance and coherence directly improves final task performance. Our work provides a scalable framework for analyzing, debugging, and improving LLM reasoning, demonstrating the practical value of moving beyond validity checks.

Clinical diagnosis is critical in medical practice, typically requiring a continuous and evolving process that includes primary diagnosis, differential diagnosis, and final diagnosis. However, most existing clinical diagnostic tasks are single-step processes, which does not align with the complex multi-step diagnostic procedures found in real-world clinical settings. In this paper, we propose a multi-step diagnostic task and annotate a clinical diagnostic dataset (MSDiagnosis). This dataset includes primary diagnosis, differential diagnosis, and final diagnosis questions. Additionally, we propose a novel and effective framework. This framework combines forward inference, backward inference, reflection, and refinement, enabling the LLM to self-evaluate and adjust its diagnostic results. To assess the effectiveness of our proposed method, we design and conduct extensive experiments. The experimental results demonstrate the effectiveness of the proposed method. We also provide a comprehensive experimental analysis and suggest future research directions for this task.

Multimodal large language models exhibit capabilities on reasoning tasks, yet often produce flawed intermediate steps while yielding correct final answers. This behavior undermines interpretability and reliability, suggesting reliance on spurious shortcuts rather than faithful reasoning. Although efforts have explored step-level supervision, distinguishing decisive steps from redundant ones remains challenging. We propose O^2-CritiCuRL, a novel curriculum reinforcement learning framework that introduces critical-step awareness through an iterative offline-online paradigm. In the offline stage, O^2-CritiCuRL conducts multi-rollout analysis over step-annotated trajectories to estimate step-level importance, allowing the framework to distill critical reasoning steps and filter out redundant ones. In the online stage, we employ a progressive step-level reinforcement learning strategy, where truncated chains guide the model to infer missing steps and refine its reasoning, thereby sharpening its focus on critical steps and overcoming the limitations of static supervision. Extensive experiments on multimodal reasoning benchmarks show that our method achieves state-of-the-art performance while delivering superior training and inference efficiency. Code is available at https://github.com/kk0013/CritiCuRL.

Skill-distillation pipelines learn reusable rules from LLM agent trajectories, but they lack a key signal: how much each step costs. Without per-step cost, a pipeline cannot distinguish adding a missing step to fix a bug from removing an expensive step that never affected the outcome. We introduce ClawTrace, an agent tracing platform that records every LLM call, tool use, and sub-agent spawn during an agent session and compiles each session into a TraceCard: a compact YAML summary with per-step USD cost, token counts, and redundancy flags. Built on ClawTrace, CostCraft is a distillation pipeline that reads TraceCards and produces three types of skill patches. Preserve patches keep behaviors that led to success. Prune patches remove expensive steps that did not matter, each backed by a counterfactual argument against a named high-cost step. Repair patches fix failures grounded in oracle evidence. Ablations on 30 held-out SpreadsheetBench tasks show that both cost attribution and prune patches independently reduce quality regressions. When the same skill is applied to 30 unrelated SkillsBench tasks, an unexpected asymmetry emerges: prune rules transferred across benchmarks and cut median cost by 32%, while preserve rules, trained on benchmark-specific conventions, caused regressions on new task types. We release ClawTrace and TraceCards as open infrastructure for cost-aware agent research.

Function calling enables large language models (LLMs) to interact with external systems by leveraging tools and APIs. When faced with multi-step tool usage, LLMs still struggle with tool selection, parameter generation, and tool-chain planning. Existing methods typically rely on manually designing task-specific demonstrations, or retrieving from a curated library. These approaches demand substantial expert effort and prompt engineering becomes increasingly complex and inefficient as tool diversity and task difficulty scale. To address these challenges, we propose a self-guided method, Stepwise Experience Recall (SEER), which performs fine-grained, stepwise retrieval from a continually updated experience pool. Instead of relying on static or manually curated library, SEER incrementally augments the experience pool with past successful trajectories, enabling continuous expansion of the pool and improved model performance over time. Evaluated on the ToolQA benchmark, SEER achieves an average improvement of 6.1% on easy and 4.7% on hard questions. We further test SEER on τ-bench, which includes two real-world domains. Powered by Qwen2.5-7B and Qwen2.5-72B models, SEER demonstrates substantial accuracy gains of 7.44% and 23.38%, respectively.

The impressive performance of large language models (LLMs) on code-related tasks has shown the potential of fully automated software development. In light of this, we introduce a new software engineering task, namely Natural Language to code Repository (NL2Repo). This task aims to generate an entire code repository from its natural language requirements. To address this task, we propose a simple yet effective framework CodeS, which decomposes NL2Repo into multiple sub-tasks by a multi-layer sketch. Specifically, CodeS includes three modules: RepoSketcher, FileSketcher, and SketchFiller. RepoSketcher first generates a repository's directory structure for given requirements; FileSketcher then generates a file sketch for each file in the generated structure; SketchFiller finally fills in the details for each function in the generated file sketch. To rigorously assess CodeS on the NL2Repo task, we carry out evaluations through both automated benchmarking and manual feedback analysis. For benchmark-based evaluation, we craft a repository-oriented benchmark, SketchEval, and design an evaluation metric, SketchBLEU. For feedback-based evaluation, we develop a VSCode plugin for CodeS and engage 30 participants in conducting empirical studies. Extensive experiments prove the effectiveness and practicality of CodeS on the NL2Repo task.

Deep Research Agents (DRAs) generate citation-rich reports via multi-step search and synthesis, yet existing benchmarks mainly target text-only settings or short-form multimodal QA, missing end-to-end multimodal evidence use. We introduce MMDeepResearch-Bench (MMDR-Bench), a benchmark of 140 expert-crafted tasks across 21 domains, where each task provides an image-text bundle to evaluate multimodal understanding and citation-grounded report generation. Compared to prior setups, MMDR-Bench emphasizes report-style synthesis with explicit evidence use, where models must connect visual artifacts to sourced claims and maintain consistency across narrative, citations, and visual references. We further propose a unified, interpretable evaluation pipeline: Formula-LLM Adaptive Evaluation (FLAE) for report quality, Trustworthy Retrieval-Aligned Citation Evaluation (TRACE) for citation-grounded evidence alignment, and Multimodal Support-Aligned Integrity Check (MOSAIC) for text-visual integrity, each producing fine-grained signals that support error diagnosis beyond a single overall score. Experiments across 25 state-of-the-art models reveal systematic trade-offs between generation quality, citation discipline, and multimodal grounding, highlighting that strong prose alone does not guarantee faithful evidence use and that multimodal integrity remains a key bottleneck for deep research agents.

Recent advancements in image generation models have enabled the prediction of future Graphical User Interface (GUI) states based on user instructions. However, existing benchmarks primarily focus on general domain visual fidelity, leaving the evaluation of state transitions and temporal coherence in GUI-specific contexts underexplored. To address this gap, we introduce GEBench, a comprehensive benchmark for evaluating dynamic interaction and temporal coherence in GUI generation. GEBench comprises 700 carefully curated samples spanning five task categories, covering both single-step interactions and multi-step trajectories across real-world and fictional scenarios, as well as grounding point localization. To support systematic evaluation, we propose GE-Score, a novel five-dimensional metric that assesses Goal Achievement, Interaction Logic, Content Consistency, UI Plausibility, and Visual Quality. Extensive evaluations on current models indicate that while they perform well on single-step transitions, they struggle significantly with maintaining temporal coherence and spatial grounding over longer interaction sequences. Our findings identify icon interpretation, text rendering, and localization precision as critical bottlenecks. This work provides a foundation for systematic assessment and suggests promising directions for future research toward building high-fidelity generative GUI environments. The code is available at: https://github.com/stepfun-ai/GEBench.

In this paper we investigate learning visual models for the steps of ordinary tasks using weak supervision via instructional narrations and an ordered list of steps instead of strong supervision via temporal annotations. At the heart of our approach is the observation that weakly supervised learning may be easier if a model shares components while learning different steps: `pour egg' should be trained jointly with other tasks involving `pour' and `egg'. We formalize this in a component model for recognizing steps and a weakly supervised learning framework that can learn this model under temporal constraints from narration and the list of steps. Past data does not permit systematic studying of sharing and so we also gather a new dataset, CrossTask, aimed at assessing cross-task sharing. Our experiments demonstrate that sharing across tasks improves performance, especially when done at the component level and that our component model can parse previously unseen tasks by virtue of its compositionality.

Evaluating LLMs is challenging, as benchmark scores often fail to capture models' real-world usefulness. Instead, users often rely on ``vibe-testing'': informal experience-based evaluation, such as comparing models on coding tasks related to their own workflow. While prevalent, vibe-testing is often too ad hoc and unstructured to analyze or reproduce at scale. In this work, we study how vibe-testing works in practice and then formalize it to support systematic analysis. We first analyze two empirical resources: (1) a survey of user evaluation practices, and (2) a collection of in-the-wild model comparison reports from blogs and social media. Based on these resources, we formalize vibe-testing as a two-part process: users personalize both what they test and how they judge responses. We then introduce a proof-of-concept evaluation pipeline that follows this formulation by generating personalized prompts and comparing model outputs using user-aware subjective criteria. In experiments on coding benchmarks, we find that combining personalized prompts and user-aware evaluation can change which model is preferred, reflecting the role of vibe-testing in practice. These findings suggest that formalized vibe-testing can serve as a useful approach for bridging benchmark scores and real-world experience.

Large language models (LLMs) for code completion and generation are increasingly used in software development, yet they may reproduce training examples verbatim and without authorship attribution, raising legal and ethical concerns around plagiarism and license compliance. Classical fingerprint-based plagiarism detectors based on fingerprinting, such as Winnowing, remain highly effective, yet the inspection requires comparing fragments of code to the entire training set, and their linear-time search makes them impractical for the billion-scale corpora used to train modern code LLMs. To bridge this gap, we introduce SOURCETRACKER, a 300M-parameter encoder tailored for code retrieval, together with a hybrid two-stage provenance-tracking pipeline HYBRIDSOURCETRACKER (HST). HST first narrows down a small set of candidate snippets via vector search, then re-ranks those candidates using Winnowing on exact fingerprints. We train and evaluate our system on a 10M-snippet subset of the THESTACKV2 dataset, with both verbatim and adapted snippets that emulate realistic identifier renaming. On an in vitro 100k-snippet search space with adapted queries, our hybrid approach reaches a mean reciprocal rank on par with Winnowing for 30-token fragments. Then, starting from windows >= 60 tokens, it consistently over-performs by up to 5.4% while preserving logarithmic-time query complexity. In a complementary evaluation using an LLM-based judge, we find that many retrieved snippets not labeled as ground truth are still highly similar to the expected sources, particularly with longer context windows, and thus remain useful for end users. Overall, our results demonstrate that integrating vector search with fingerprinting enables scalable, high-precision provenance tracking for code produced by LLMs.

As large language models (LLMs) evolve from static chatbots into autonomous agents, the primary vulnerability surface shifts from final outputs to intermediate execution traces. While safety guardrails are well-benchmarked for natural language responses, their efficacy remains largely unexplored within multi-step tool-use trajectories. To address this gap, we introduce TraceSafe-Bench, the first comprehensive benchmark specifically designed to assess mid-trajectory safety. It encompasses 12 risk categories, ranging from security threats (e.g., prompt injection, privacy leaks) to operational failures (e.g., hallucinations, interface inconsistencies), featuring over 1,000 unique execution instances. Our evaluation of 13 LLM-as-a-guard models and 7 specialized guardrails yields three critical findings: 1) Structural Bottleneck: Guardrail efficacy is driven more by structural data competence (e.g., JSON parsing) than semantic safety alignment. Performance correlates strongly with structured-to-text benchmarks (ρ=0.79) but shows near-zero correlation with standard jailbreak robustness. 2) Architecture over Scale: Model architecture influences risk detection performance more significantly than model size, with general-purpose LLMs consistently outperforming specialized safety guardrails in trajectory analysis. 3) Temporal Stability: Accuracy remains resilient across extended trajectories. Increased execution steps allow models to pivot from static tool definitions to dynamic execution behaviors, actually improving risk detection performance in later stages. Our findings suggest that securing agentic workflows requires jointly optimizing for structural reasoning and safety alignment to effectively mitigate mid-trajectory risks.

We study a class of optimization problems motivated by automating the design and update of AI systems like coding assistants, robots, and copilots. We propose an end-to-end optimization framework, Trace, which treats the computational workflow of an AI system as a graph akin to neural networks, based on a generalization of back-propagation. Optimization of computational workflows often involves rich feedback (e.g. console output or user's responses), heterogeneous parameters (e.g. prompts, hyper-parameters, codes), and intricate objectives (beyond maximizing a score). Moreover, its computation graph can change dynamically with the inputs and parameters. We frame a new mathematical setup of iterative optimization, Optimization with Trace Oracle (OPTO), to capture and abstract these properties so as to design optimizers that work across many domains. In OPTO, an optimizer receives an execution trace along with feedback on the computed output and updates parameters iteratively. Trace is the tool to implement OPTO in practice. Trace has a Python interface that efficiently converts a computational workflow into an OPTO instance using a PyTorch-like interface. Using Trace, we develop a general-purpose LLM-based optimizer called OptoPrime that can effectively solve OPTO problems. In empirical studies, we find that OptoPrime is capable of first-order numerical optimization, prompt optimization, hyper-parameter tuning, robot controller design, code debugging, etc., and is often competitive with specialized optimizers for each domain. We believe that Trace, OptoPrime and the OPTO framework will enable the next generation of interactive agents that automatically adapt using various kinds of feedback. Website: https://microsoft.github.io/Trace

We introduce CRYSTAL (Clear Reasoning via Yielded Steps, Traceability, and Logic), a diagnostic benchmark with 6,372 instances that evaluates multimodal reasoning through verifiable intermediate steps. We propose two complementary metrics: Match F1, which scores step-level precision and recall via semantic similarity matching, and Ordered Match F1, which further penalizes disordered reasoning chains. References are constructed through a Delphi-inspired pipeline in which four independent MLLMs generate trajectories, which are then aggregated via semantic clustering and validated through human quality gates. Evaluation of 20 MLLMs, including commercial frontier systems not used during benchmark construction, reveals systematic failures that are invisible to answer accuracy: universal cherry-picking (precision far exceeds recall), non-monotonic scaling trade-offs, and disordered reasoning in which no competitive model preserves more than 60% of matched steps in the correct order. Beyond evaluation, we propose the Causal Process Reward (CPR), a multiplicative reward that couples answer correctness with step-level alignment, and CPR-Curriculum, which progressively increases reasoning difficulty during training. CPR-Curriculum achieves a 32% improvement in Match F1 via GRPO where additive reward strategies fail, improving reasoning without manual step annotation.

Autonomous research systems increasingly make the scientific workflow executable: agents can propose ideas, run code, inspect results, and draft papers. But executable workflows do not by themselves produce research judgment. We analyze where current systems lose trial experience: weak evidence becomes prose, pilot signals become broad claims, memory remains textual, and recurring process failures do not change later behavior. We introduce Sibyl-AutoResearch, a self-evolving AutoResearch framework built around Scientific Trial-and-Error Harnesses. A harness lets agents run bounded trials, preserve positive and negative outcomes, and route lessons into later planning, validation, claim scope, scheduling, critique, writing, and harness repair. We formalize this through two auditable conversion units: trial-to-behavior conversion, which links trial signals to later research actions, and trial-to-harness-behavior conversion, which links recurring process failures to system updates. We implement the framework in SIBYL, a file-backed autonomous research system that exposes the state, roles, memory, gates, and artifact traces needed to inspect these conversion paths. A retrospective audit identifies eight high-confidence conversion events, with a median latency of one iteration and a maximum latency of three iterations. A recovered-failure registry further shows how five naturally occurring failure classes, including duplicate results, stale numbers, and unsupported statistics, were blocked, downgraded, or routed into later repair. These traces do not establish a comparative performance claim; they show that the proposed conversion units are recoverable from realistic autonomous-research workspaces. The SIBYL framework and system are available at https://github.com/Sibyl-Research-Team/AutoResearch-SibylSystem.

LLM-based agents increasingly solve complex tasks through long trajectories involving reasoning steps, tool calls, and inter-agent communication. However, when these agents fail, it is often unclear which agent caused the failure and which step introduced the decisive error. This attribution problem is challenging because mistakes can propagate across the trajectory: later actions may appear incorrect, but only because they depend on an earlier corrupted state. Therefore, failure attribution cannot be treated as independent step-level classification. We propose FALAT, a diagnostic framework for failure attribution in LLM agent trajectories. FALAT frames attribution as a dependency-guided search problem. It first constructs an expectation of how the task should be solved and uses this expectation to identify suspicious regions in the trajectory. It then traces dependencies among decisions, tool outputs, and agent messages to distinguish error-introducing steps from steps that merely inherit or propagate prior mistakes. Finally, FALAT evaluates whether correcting a candidate step would be sufficient to recover the expected outcome, allowing it to identify both the responsible agent and the decisive failure step. We evaluate FALAT on the Who&When benchmark, which includes both algorithm-generated and hand-crafted multi-agent failure trajectories. The results show that FALAT consistently improves responsible-agent and decisive-step attribution. Its best configurations achieve 46.0% step-level accuracy on algorithm-generated trajectories and 29.1% on the more challenging hand-crafted trajectories, outperforming specialized attribution baselines and direct prompting with standalone LLMs. These findings suggest that dependency-aware reasoning is essential for reliable failure diagnosis in LLM agent systems.

State-of-the-art text-to-image models produce high-quality images, but inference remains expensive as generation requires several sequential ODE or denoising steps. Native one-step models aim to reduce this cost by mapping noise to an image in a single step, yet fair comparisons to multi-step systems are difficult because studies use mismatched sampling steps and different classifier-free guidance (CFG) settings, where CFG can shift FID, Inception Score, and CLIP-based alignment in opposing directions. It is also unclear how well one-step models scale to multi-step inference, and there is limited standardized out-of-distribution evaluation for label-ID-conditioned generators beyond ImageNet. To address this, We benchmark eight models spanning one-step flows (MeanFlow, Improved MeanFlow, SoFlow), multi-step baselines (RAE, Scale-RAE), and established systems (SiT, Stable Diffusion 3.5, FLUX.1) under a controlled class-conditional protocol on ImageNet validation, ImageNetV2, and reLAIONet, our new proofread out-of-distribution dataset aligned to ImageNet label IDs. Using FID, Inception Score, CLIP Score, and Pick Score, we show that FID-focused model development and CFG selection can be misleading in few-step regimes, where guidance changes can improve FID while degrading text-image alignment and human preference signals and worsening perceived quality. We further show that leading one-step models benefit from step scaling and become substantially more competitive under multi-step inference, although they still exhibit characteristic local distortions. To capture these tradeoffs, we introduce MinMax Harmonic Mean (MMHM), a composite proxy over all four metrics that stabilizes hyperparameter selection across guidance and step sweeps.

Parametric retrieval enables LLMs to retrieve tools implicitly by assigning each API a unique virtual token and training the model to generate it via constrained beam search. Toolsense shows that this regime has two critical drawbacks: it destroys parametric tool knowledge during training, and its beam-search decoding is too slow for real-time deployment. We introduce TRACE (Tool Retrieval via Augmented Chain-of-thought and Enterprise rules), a two-stage curriculum that resolves this dissociation. Stage 1 reuses the multi-format memorization SFT from ToolSense to seed tool knowledge with LoRA. Stage 2 is our core contribution: the model is trained to emit a thinking trace before producing a JSON list of tool tokens, using two data sources -- RRB pairs from ToolSense and queries synthesized to target business rules curated by domain experts -- both augmented with reasoning traces. This training objective preserves Stage 1 MCQ and QA probing accuracy while enabling single-beam greedy decoding at production latency. Evaluated on a combined enterprise catalog of 8,300+ tools across two enterprise product lines, TRACE training for Stage 2 not only preserves but improves tool understanding: MCQ accuracy gains +3.2 pp and QA probing gains +9 pp over Stage 1. On retrieval, TRACE achieves ~86% recall on Domain A and ~60% on Domain B -- compared to embedding baseline performance of ~27% & ~52% -- both with single-beam greedy decoding, making it directly deployable at production latency.

The Abstraction and Reasoning Corpus (ARC) tests whether a model can infer an unseen transformation from a few input-output examples and apply it to a new grid. Looped visual reasoners refine predictions over multiple iterations, but conventional training constrains only the final output, leaving intermediate refinements unconstrained. We propose that these refinements should instead follow the transformation step by step. We introduce TraceViT, a looped visual reasoner trained with semantically monotonic transformation chains. We obtain these chains by rewriting and verifying programmatic task implementations, decomposing each solution into intermediate grid states. Each iteration is grounded by a task reference derived from the few-shot demonstrations and an object workspace representing the current grid state. Because these chains may differ in length from the loop, soft trace alignment enforces only their ordering, letting the model allocate iterations freely. TraceViT achieves 67.8% pass@2 on ARC-AGI-1 and 24.3% on ARC-AGI-2. Controlled ablations on ARC-AGI-1 show that trace supervision becomes beneficial only when paired with grounding. Code and data will be available at https://github.com/LiuBinnan/TraceViT.