Review
-
Aperture: Training-Free Multiscale Concept Bottlenecks for Remote Sensing
2026-09-29 22:09:36Rishabh Mondal, Nipun Batra, Utkarsh MallAbstract
While earth observation models have advanced substantially, they still lack interpretability. While concept-bottleneck models provide interpretability and expert interaction, they are either too expensive to train for the remote sensing domain or perform poorly without annotation. We posit that in expert domains like remote sensing, such training-free models require both fine details in both image and concept space. In image space, we propose a multiscale concept bottleneck using greedy quadtree routing to locate small concepts. In concept space, we replace contrastive vision language models with pre-trained MLLMs and present a way to get reliable concept scores from them. We introduce APERTURE that blends concept scores at the global image and native concept-scale level to give state-ofthe-art training-free model performance. To test these models, introduce SiFC, a fine-grained concept-centric dataset across three countries, with human-reviewed class-level concept maps. On SiFC, APERTURE outperforms the best training-free baselines by more than 10 percentage points in macro F1-score, and notably also outperforms supervised concept bottleneck models. Targeted component-removal tests examine whether concept scores respond to changes in visual evidence, while temporal experiments show that descriptor updates improve recognition of technological changes without retraining.
Abstract (translated)
URL
https://arxiv.org/abs/2609.38603
PDF
-
Defining and Categorising Human-AI Interactions in Clinical Trials: A Multidimensional Human-AI Classification Approach
2026-09-29 21:18:26Sandra Woolley, Tim Collins, Khalid Khattak, Illia Chernomorets, Ariane Arevalo, Chris RichardsonAbstract
This paper examines human-AI interactions (HAIIs) in clinical trials and presents a multidimensional categorisation framework that classifies interactions according to AI tasks, human-AI relationships, interaction configurations and interacting human groups. We define HAII, examine existing taxonomies and extend existing categorisation approaches through this novel multidimensional framework. We purposively sampled 15 clinical trials from a previously reported dataset. Each trial was independently categorised by two human reviewers and six large language model (LLM) classifiers. The proposed categorisation provides a structured method for the consistent identification, comparison and synthesis of human-AI interactions across clinical-trial records. The framework is intended to support more consistent comparison and synthesis of AI-related clinical trials and to make explicit the different forms of human involvement associated with AI interventions. The results demonstrate the potential for LLM-assisted categorisation while indicating the continuing importance of human judgement where trial records are incomplete or ambiguous. The principal contribution is a proposed multidimensional framework that brings together AI tasks, human-AI relationships, interaction configurations and interacting human groups within a single approach designed for clinical-trial records. Its significance lies in its potential to support more systematic identification, comparison and synthesis of how humans and AI interact in clinical trials.
Abstract (translated)
URL
https://arxiv.org/abs/2609.38559
PDF
-
Role-guided Speaker Deletion Verification in Clinical Psychiatry Speech Recordings with Audio Language Models
2026-09-29 20:20:24Joseph T Colonel, Daniel Katzman, Kelsey Kirker, Adam N Davidson, Shalaila S Haas, Cheryl Corcoran, Ren\'{e} S Kahn, Guillermo Checci, Baihan LinAbstract
Clinical research in psychiatry increasingly relies on large scale collection of spoken language data to identify acoustic and linguistic biomarkers. Yet evolving consent and protocol requirements can oblige investigators to remove a designated speaker from multi-speaker recordings and to verify said removal at a scale infeasible for manual review of entire corpora. We study this verification problem for role-driven dyadic clinical dialogue in psychiatry and investigate it with two parallel, symmetric pipelines: confirming that clinician speech has been removed from psychiatric interview recordings, and confirming that patient speech has been removed from the same recordings. Each pipeline redacts the raw audio for its target role and then scans the surviving output with audio-language and large-language models to identify missed deletions. We evaluate this approach on a corpus of 48 dyadic recordings drawn from psychiatry settings, testing four open-weight models in an inference-only setting: Gemma-4-12B, Gemma-4-31B, Nemotron-3-Nano, and Nemotron-3-Nano-Omni. A disjunctive OR ensemble over fourteen model-view configurations had a combined F1 of 0.478 (precision 0.330, recall 0.870), an improvement over individual model estimates driven by recall gains that point to substantial complementarity across models and context views.
Abstract (translated)
URL
https://arxiv.org/abs/2609.38491
PDF
-
Deep Learning for Longitudinal Medical Imaging: A Scoping Review
2026-09-29 20:14:37Francesca Mussa, Divyanshu Tak, Atlas H. Avval, Sarah Brueningk, Ray H. Mak, Hugo J. W. L. Aerts, Andreas M Rauschecker, Benjamin H. KannAbstract
Longitudinal medical imaging analysis is a cornerstone of modern medical practice and patient care. Deep learning applied to longitudinal imaging offers wide potential to enhance diagnosis and track disease progression by capturing spatial changes over time. With major advances in single-timepoint deep learning for imaging, there has been growing interest in longitudinal image analysis, given its increased clinical relevance, though technical challenges remain. Several recent innovations may lead to a new era of multi-timepoint image evaluation, yet the scientific landscape, recent progress, and areas of need remain under-characterized. To address this gap, we conducted a scoping review of deep learning methodologies applied to longitudinal medical imaging, yielding 102 studies published between 2018 and 2025. Neurological disorders (48%) and ophthalmic conditions (12%) were the most common clinical applications, with MRI serving as the predominant imaging modality (67%). Sequential feature modeling approaches combining convolutional neural networks (CNNs) with temporal models (LSTM/RNN) were the most frequent methodology (40%), followed by direct feature aggregation across timepoints (23%). Most studies targeted classification tasks (56%), while external validation was performed in only 24% of studies. Our findings highlight that deep learning-based longitudinal imaging analysis remains a promising field, though newer temporal architectures and larger datasets may improve success and clinical adoption of these tools.
Abstract (translated)
URL
https://arxiv.org/abs/2610.08838
PDF
-
Auditing Agent Actions through Query-Conditioned Attribution
2026-09-29 19:47:18Yifan Liu, Praveen Venkateswaran, Abdulhamid Adebayo, Dong WangAbstract
LLM agents increasingly take consequential actions through interactions with users, policies, and external tools. Auditing these agents requires automated attribution of realized actions to their historical basis. However, existing attribution formulations do not provide question-specific traces for diverse auditing objectives. Additionally, when access to the acting model is limited (e.g., in API-only deployments), applicable methods commonly rely on costly input perturbations or external LLM analysis of complete trajectories. We therefore formulate query-conditioned agent action attribution, a new task that takes a natural-language auditing query as input and recovers the source and ordered intermediate evidence for the query-specified aspect of an action. We instantiate this task with $A^3Bench$, a benchmark comprising 1,396 auditing queries across policy basis, parameter provenance, failure propagation, and unsafe-behavior tracing. To enable efficient, query-specific attribution, we use small open-weight models as attribution proposers that combine query-conditioned gradient saliency with query-semantic relevance to rank history units. Our proposer consistently achieves stronger source and evidence rankings at lower inference cost than open-weight baselines, improving source MRR by up to 40.9\% and evidence MAP by 42.1\% with only two forward passes and one backward pass. Controlled evaluations confirm that our proposer improves attribution specificity by adapting its rankings to fine-grained changes in the auditing query. Building on a proposer ensemble, our end-to-end system surpasses the strongest frontier-model baseline in source accuracy (64.5\% vs.\ 60.4\%) while reducing empirical deployment latency by 29.9\% relative to the fastest frontier API baseline. Code and data will be released after the initial review period following final validation and cleanup.
Abstract (translated)
URL
https://arxiv.org/abs/2609.33676
PDF
-
Policy-Conditioned AI-Use Detection: An Evidentiary Framework for Academic Publishing
2026-09-29 19:17:35Jairo Diaz-Rodriguez, Mumin JiaAbstract
Major venues now publish detailed rules about how authors, reviewers, and area chairs may use AI, and those rules differ by role, by task, and by what must be disclosed. AI detection, the instrument usually proposed to enforce them, estimates something else: whether an AI model wrote the text. We argue that this target is misaligned with the decisions conferences and journals face, and propose policy-conditioned AI-use detection, an evidentiary framework for assessing whether a human--AI workflow complied with a stated rule. Policy makes the governing rule an explicit input. Inference reports hypotheses, evidence, calibration regime, and uncertainty in place of verdicts such as "AI detected". Evaluation builds benchmarks from reproducible pipelines that generate compliant and non-compliant workflows, and reports true positive rate at a false positive rate the venue fixes in advance. We work the framework through peer review, where at plausible violation rates a detector at a strong operating point still flags more compliant authors than violating ones. The framework therefore also names what a venue must instrument: structured disclosure, approved-tool routing that respects reviewer confidentiality, and a path by which a finding can be contested. Under this framing a detector is not an authorship classifier but an auditable procedure with an error rate the venue fixes in advance and can defend.
Abstract (translated)
URL
https://arxiv.org/abs/2609.38427
PDF
-
Self-Evolving Harness on Multiple Tasks with the Agent as Its Own Optimizer
Abstract
A harness is the code around a language-model agent that organizes prompts, calls tools, manages context, and controls execution. As models grow stronger, recent work has begun to let agents improve their own harnesses, a line of work known as self-evolving harnesses. In most existing methods, a separate proposer running on a human-designed harness modifies the solver's harness, and a separate harness is evolved for each benchmark. Real-world tasks come from many domains, so both the evolution and the evaluation of a harness should cover a diverse range of tasks. We propose a framework close to recursive self-improvement: the same frozen model, on the same version of the harness, first solves tasks as the solver and then, as the proposer, reads the complete run records and directly edits the harness that runs it. Each evolution batch draws tasks from five benchmarks in different domains. To measure generalization, training and held-out tasks are strictly separated, and we additionally evaluate on five out-of-distribution benchmarks never used during evolution. We frame the evolution process as deep-learning training with two stages, multi-task pretraining and continual training. Starting from a 49-line seed harness, the harness obtained at the end of the first stage improves the average score by 4.48 points on the in-distribution benchmarks and by 12.64 points on the out-of-distribution benchmarks, surpassing Codex on the former and matching it on the latter. In the second stage, continued evolution on Claw-Eval, one of the out-of-distribution benchmarks, further raises the score on that benchmark from 66.17 to 68.06, exceeding Codex. We also provide an in-depth analysis of the mechanisms that emerged during evolution, including output truncation, history compaction, and independent review.
Abstract (translated)
URL
https://arxiv.org/abs/2609.38372
PDF
-
Can an AI Agent Rediscover a Blaschke-Curve Invariant?
Abstract
We study generalized Blaschke curves as a controlled environment for AI-assisted mathematical rediscovery. For one fixed degree-four Blaschke product, an agent receives numerical coordinates of the six pair-lines determined by each of 80 boundary configurations. The target theorem is withheld from the task instructions. The saved research log reports rejected geometric hypotheses and a homogeneous cubic fitted to polygon sides. Its frozen coefficients predict 480 lines from 80 unseen parameter values, with a recorded RMS scale-free residual of $8.88\times10^{-17}$. Discovery-set diagonals provide an out-of-fit consistency check, not a fully held-out test. A separate one-configuration run reports insufficient evidence for invariance. A post-review deterministic degree-search baseline also recovers the cubic, so the experiment does not establish an advantage over polynomial fitting. We present this single-instance case study as a protocol for separating conjecture, numerical validation, and proof, with explicit limitations concerning agent metadata, prior knowledge, and reproducibility.
Abstract (translated)
URL
https://arxiv.org/abs/2609.38369
PDF
-
From Routing Signals to Selective Review: Visual regrounding in MoE VLMs
2026-09-29 17:49:07Hongzhu Guo, Mohsen Fayyaz, Nanyun PengAbstract
Vision-language models (VLMs) may accept false visual premises, answering questions about a target object's color, count, location, or state even when it is absent. We call this reliability-critical behavior a target-absence grounding failure. Existing visual-grounding detectors primarily rely on generated responses, hidden states, or uncertainty measures. We present the first framework to leverage internal routing decisions in Mixture-of-Experts (MoE) VLMs to detect target absence before generation and guide selective correction. We extract target-token routing probabilities from Qwen3-VL-30B-A3B-Instruct and Gemma-4-26B-A4B-it, train a separate L2-regularized linear detector for each model, and use its predictions to selectively invoke a target-aware review prompt. Using routing alone, the Qwen and Gemma detectors achieve ROC-AUCs of 0.9988 and 0.9956 on GQA-Inpaint and retain 0.8095 and 0.7781 on the external OBER dataset, respectively. The resulting routing-gated policy improves end-to-end accuracy on GQA-Inpaint and OBER by +22.25% and +12.17% for Qwen, and by +13.42% and +1.39% for Gemma, without modifying model weights. Further analysis shows that the signal is localized to the target-object token, emerges in early MoE layers, and is distributed across partially substitutable experts. Although cross-dataset threshold shifts require recalibration, false-positive review causes limited harm overall, suggesting that intervention risk can be controlled through joint selection of the detector threshold and review prompt. Overall, we show that routing probabilities alone preserve actionable information about visual perception, allowing computation already produced by an MoE VLM to support low-cost detection and selective visual regrounding.
Abstract (translated)
URL
https://arxiv.org/abs/2609.38111
PDF
-
Evaluating System One Models for Agent Security Decisions: Reliability, Calibration, and Selective Automation
Abstract
Model-based judges support agent security by detecting prompt injections, assessing interaction risks, and screening harmful requests. System One models select from predefined answers and report probabilities that software can use to allow, block, or review inputs, but the reliability of these automated decisions remains unclear. We evaluate Jev, Laya, Decider, and Bespoke Nimble against specialized classifiers and language-model judges, examining decision accuracy, probability calibration, and selective automation. We draw the following conclusions. (1) Strong overall performance and favorable aggregate calibration can hide failures concentrated in particular attack groups, including attacks classified as safe with high confidence. (2) The evaluated adapted configurations do not consistently improve classification over their base models across tasks. (3) Under the strictest evaluated error limits, the policies allow few inputs automatically, and separate allow and block thresholds increase automation mainly through more blocks. Passing confirmation does not ensure that these limits hold on test. (4) Judges can detect attacks missed by another model, but may also falsely flag more benign inputs and share the other model's high-confidence errors. These findings support evaluating model accuracy, probability calibration, and the resulting allow/block/review decisions together.
Abstract (translated)
URL
https://arxiv.org/abs/2609.33401
PDF
-
One Threshold Does Not Fit All Languages: Language-Conditional Deferral for Reliable and Efficient Low-Resource Text Classification
2026-09-29 15:45:33Bhanu Prakash Vangala, Vangala NavyaAbstract
In the Global South, the lower-income countries of Africa, Asia, and Latin America where most of the world's languages are spoken, a deployed text classifier usually runs on ordinary CPUs, serves many languages with a single model, has few labeled examples in any of them, and relies on people to catch its mistakes. Such a system is only useful if it can promise how often it will be wrong: at most a fixed fraction of the labels it assigns on its own may be incorrect, and everything else must go to a person. Split conformal prediction delivers this promise through a single confidence threshold, normally estimated on validation data pooled across languages. We ask whether the promise reaches every language, and it does not. On MasakhaNEWS (16 African languages) and AfriSenti (12 languages plus two never seen in training), a pooled threshold meets the 90% target on average but covers Somali at 77.5%, Tigrinya at 83.7%, and the two unseen languages at 77.5% and 81.2%. Estimating one threshold per language brings every language to between 89.1% and 91.0% without retraining, and it shows how unequal the cost of the promise is: keeping it means sending 43% of Somali news and over 80% of Amharic and Xitsonga tweets to a person, against under 8% of Nigerian Pidgin news. One or two hundred labels per language are enough and the models train in minutes on one CPU core, so the fix is affordable: calibrate, report, and budget human review one language at a time.
Abstract (translated)
URL
https://arxiv.org/abs/2609.37861
PDF
-
Is manual software optimization a thing of the past?
2026-09-29 15:42:17Pavlin G. Poli\v{c}ar, Martin \v{S}pendl, Toma\v{z} Ho\v{c}evarAbstract
Scientific software is increasingly required to process larger datasets while maintaining acceptable execution times. Software optimization traditionally requires substantial expertise in programming, algorithms, and numerical methods. Recent advances in large language models (LLMs) offer the possibility of automating much of this process. We investigate whether LLM-based agents can autonomously achieve substantial performance improvements in scientific software, including mature implementations that have already been extensively optimized by human developers. We tasked an LLM-based agent with optimizing software for three computational problems: t-SNE, single-sample gene set enrichment analysis (ssGSEA), and graphlet counting. Humans defined the scope, correctness criteria, and a verification mechanism, after which the agent worked autonomously, in some cases for several hours. Code maintainers reviewed each resulting implementation and verified its correctness. The optimized implementations were faster in all tested configurations, by up to two orders of magnitude over the fastest existing tools. The improvements included low-level code optimizations, mathematical reformulations, and an entirely new algorithm for graphlet counting. Software optimization can increasingly be delegated to autonomous agents, with the human role shifting from implementing optimizations to deciding which software to optimize, defining objectives, providing verification mechanisms, and ensuring the correctness of the final software. For well-scoped, verifiable problems, we argue that manual software optimization may be a thing of the past.
Abstract (translated)
URL
https://arxiv.org/abs/2609.37849
PDF
-
RxnOptBench: Benchmarking LLMs for Reaction-Condition Optimization in Organic Methodology
2026-09-29 15:17:31Lingli Ge, Yubin Wang, Junyuan Gao, Jiahe Song, Jiaxing Sun, Boyu Zhu, Haote Yang, Jingchao Wang, Lixin Ma, Jiang Wu, Yuqiang Li, Conghui HeAbstract
Chemical reaction-condition optimization -- choosing the catalyst, ligand, solvent, reagent, temperature, time, and atmosphere that jointly maximize yield and stereoselectivity -- is a central, judgement-laden subtask of organic methodology research that large language models are increasingly expected to support. Yet existing chemistry benchmarks evaluate reaction-class labelling, retrosynthesis, or SMILES manipulation, and do not ask models to read a real condition-screening table and pick the best set. We introduce RxnOptBench, a benchmark whose every option and precedent is a real wet-lab entry mined from the optimization tables of organic-methodology papers published in 2025, graded by a continuous relative score derived from a declared headline utility that combines reported yield with enantiomeric excess (ee), diastereomeric ratio (dr), and regioisomeric ratio (rr), and equipped with a paired precedents-vs-no-precedents design that isolates in-context use of literature evidence from parametric memorization. Across nine frontier LLMs and three Chemistry LLMs, even the best models leave substantial headroom: chemistry-specialized models fall to the random-baseline floor on multi-axis selection, while open-weight models have closed most of the gap to proprietary frontier models. We release the final human-reviewed benchmark test set and evaluation code.
Abstract (translated)
URL
https://arxiv.org/abs/2610.02242
PDF
-
KUPAS MASTER: Distilling the Tacit Expertise of Master Practitioners into Agent-Ready Experience Corpora
2026-09-29 14:27:06Changmian Wang, Yuchao Ma, Xuchao Lu, Chen Zhang, Ping Sun, Jiazheng Wang, Shan Wang, Xuanwen Chen, Yihe Sun, Ziyu Lu, Jianqiang Huang, Hongzhi Li, Ziqing Xia, Kaihua Tang, Xian-Sheng Hua, Qinghua ZhengAbstract
Experienced professionals know more than just facts and conclusions. They know which cues matter, why a judgment is reasonable, and which action to take. Routine work records often leave out this tacit knowledge, making it difficult for Large Language Model (LLM) agents to use professional experience effectively. We introduce KUPAS MASTER, an experience engineering platform built around nine-layer cognitive corpus construction. It turns heterogeneous work records and practitioner interviews into traceable, reusable experience corpora for agents. Six case elements preserve the task process: context, cues, judgment, action, boundaries, and outcomes. Nine-layer cognitive corpus construction organizes tacit experience along nine extraction dimensions and stores the resulting assets in six libraries: rules, constraints, best practices, negative examples, corner cases, and skills. Semantic alignment, individual experience distillation, organizational consolidation, and cross-review preserve source evidence, conditions of use, and unresolved disagreements. The platform packages these assets into callable skills with explicit inputs, steps, dependencies, and stopping conditions, connecting experience collection to task execution and evaluation feedback. Using authorized samples from 20 randomly selected practitioners, the platform processed 1,576 source files into 23,024 individual experience records and 13,113 organizational assets. The evaluation spans multiple professional domains. Under common task inputs and scoring criteria, the base model, raw corpus retrieval-augmented generation (RAG), and KUPAS MASTER agent scored 70.63, 79.75, and 89.58, respectively. The KUPAS MASTER agent improved on raw-corpus RAG in all seven scoring dimensions. The platform provides a practical path from individual tacit experience to organizational knowledge and agent capabilities.
Abstract (translated)
URL
https://arxiv.org/abs/2609.37673
PDF
-
SkillGym: Training Skill-Use Agents with Automatic Verifiable Environment Generation
2026-09-29 13:21:38Renxi Wang, Mingshan Hee, Fajri Koto, Timothy Baldwin, Haonan LiAbstract
Skills equip LLM agents with professional knowledge and guidance to complete long-horizon and complex tasks. Although skills have been widely adopted in recent agent paradigms and harnesses, how to synthesize reliable training data and how to train agents for skill use remain underexplored. In this work, we propose SkillGym, an automatic pipeline to build verifiable environments, collect trajectories, and train skill-use agents. SkillGym first crawls a large volume of skills from the internet, then keeps those whose workflows can run reproducibly offline. A builder-reviewer pipeline is used to construct difficulty-controlled tasks, spanning four task types, each with a reference solution and an executable verifier. With this pipeline, we build 6.8k environments and collect 19k verified successful trajectories for supervised finetuning. Finetuning on these trajectories improves LLMs of different families and sizes, from 2B to 122B parameters across four skill-use benchmarks; Our Qwen3.5-9B SFT model outperforms the 397B untrained model on two of them. Further analysis shows that training teaches agents to invoke skills, raising the rate of reading the relevant skill from 28% to 96%, and that the gains hold across reasoning structures, extending to task types that form a minority of the training data and to skills held out from training
Abstract (translated)
URL
https://arxiv.org/abs/2609.37539
PDF
-
Complexity-Aware Evaluation of LLM Comprehension
2026-09-29 12:42:55Ali Mohammadi Esfahani, Nafiseh Kahani, Samuel A. AjilaAbstract
Large language models (LLMs) are increasingly used for software engineering tasks that require understanding existing source code, including behavior prediction, function explanation, debugging, and code review. However, aggregate benchmark accuracy can conceal how model reliability changes as source code becomes structurally more complex. This paper presents a complexity-aware framework for evaluating LLM code comprehension using cyclomatic complexity, nesting depth, branching factor, and Halstead volume. We evaluate DeepSeek-Coder-V2 and Llama through two complementary tasks: automatic input-output prediction over 300 Python functions and manually assessed semantic comprehension over a balanced subset of 60 functions. The functions are grouped into Low-, Medium-, and High-complexity bands. DeepSeek-Coder-V2 achieves an overall automatic accuracy of 78.33%, compared with 70.33% for Llama. However, accuracy decreases substantially from Low to High complexity, from 93.52% to 52.78% for DeepSeek-Coder-V2 and from 87.04% to 47.22% for Llama. Incorrect predictions are consistently associated with higher values of all four complexity metrics, and correlation and logistic-regression analyses confirm broadly comparable negative associations between structural complexity and correctness. Manual semantic comprehension shows the same degradation pattern, with accuracy decreasing from 100.00% to 75.00% for DeepSeek-Coder-V2 and from 90.00% to 60.00% for Llama. These findings demonstrate that complexity-aware evaluation provides a more diagnostic assessment of LLM code-comprehension reliability than aggregate accuracy alone.
Abstract (translated)
URL
https://arxiv.org/abs/2609.37405
PDF
-
Foundations of Proactive Agents: Principles, Technical Layers, and Proactivity-Gym
2026-09-29 11:13:10Jio Oh, Seunghyun Do, Young-Jun Lee, Steven Euijong Whang, Dongyeop KangAbstract
Proactive LLM agents can turn idle compute into useful support before users ask. Yet even correct work can misread user context, impose review costs, or undermine trust. This work proposes foundations for designing, realizing, and evaluating proactive LLM agents around three joint principles (3T): Task Capability, anticipating relevant needs and correctly performing useful work; Temporal Allocation, allocating compute according to resource availability and when results are needed; and Trust, sustaining users' confidence and appropriate reliance on the agent. We connect these objectives to a design space organized around five dimensions: task scope, anticipation horizon, activation trigger, processing timing, and intervention depth, and specify the situation and system modeling needed to support its choices, including user and environment representations, backbone LLMs, and agent harnesses. Lastly, we propose PROACTIVITY-GYM, a simulation-based evaluation testbed including multi-day scenarios, stateful environments, and persona-conditioned simulated users that can evaluate the consequences of proactive assistance across interactions. Evaluations across 23 model-harness configurations uncover substantial performance gaps across 3T and reveal that LLM judges often conflate task capability and trust. A human study with 30 participants demonstrates the importance of the joint 3T optimization: participants show sharp trust declines after intervention misalignment despite correct outcomes, and prefer sleep-time assistance, even when imperfect, to preserve ongoing focus. Together, these findings support designing and evaluating proactive agents through the joint consideration of useful work, compute allocation, and evolving user trust.
Abstract (translated)
URL
https://arxiv.org/abs/2609.37267
PDF
-
CredWise: A Controlled Agentic Decision-Intelligence Framework for Explainable and Auditable Credit-Risk Assessment
Abstract
Credit-risk prediction is important in banking, but a prediction alone does not explain why an applicant is risky or how it should be combined with other evidence. This paper presents CredWise, a decision-support framework that integrates credit-risk prediction, probability calibration, explainable artificial intelligence, policy retrieval, SQL analytics, and controlled agent-based workflows. An XGBoost model is trained on Lending Club data (1,345,310 loans, 18 features) using a temporal split: 2007--2016 for training, 2017 for validation, and 2018 for testing. On the 2018 test set, the calibrated model achieved a ROC-AUC of 0.7109, PR-AUC of 0.2993, F1-score of 0.3714, and accuracy of 65.44\%. Calibration reduced the Brier score from 0.2157 to 0.1273 and the expected calibration error from 0.2862 to 0.0585. SHAP explanations were temporally stable, with a Spearman correlation of 0.9959 between 2017 and 2018 feature rankings. On 28 labeled queries covering nine policy sections, FAISS achieved the best Hit@1 (0.929) and MRR (0.964), while all three retrieval methods reached Hit@5 = 1.0. Agent routing achieved 95.6\% accuracy (43 of 45 cases), and the SQL benchmark scored 1.0 on exact-match, execution-success, and result-match across six cases. These results show that CredWise can combine predictions, explanations, policy evidence, and structured analytics in one controlled workflow. It is an academic research prototype, and final decisions remain with a human reviewer.
Abstract (translated)
URL
https://arxiv.org/abs/2609.37223
PDF
-
CRJudgeBench: Can AI Detect Plausible but Invalid Code Reviews?
2026-09-29 10:31:16Yue Pan, Jiawei Li, Ziyuan Zhang, Xiangxin Zhao, He YeAbstract
Large language models can generate plausible code-review comments, but such comments may contain technically incorrect claims that mislead developers. We study technical trustworthiness judgment: determining whether a review comment's core technical claims are correct and applicable to the reviewed code in its repository context. Existing code-review benchmarks primarily evaluate review generation, issue discovery, or general comment quality, but do not directly assess whether an agent can determine the technical trustworthy of an individual review comment. To fill this gap, we introduce CRJudgeBench, a benchmark of 1199 instances constructed from real pull requests and expert-verified perturbations, covering both trustworthy and plausible but untrustworthy comments. We further present Sentinel, a repository-grounded agentic judge that actively gathers code evidence to verify review comments before making judgments. Starting from Qwen3-Coder-30B-A3B-Instruct, Sentinel is trained on the CRJudgeBench training split through iterative action-level learning from a privileged teacher. On the 359-instance CRJudgeBench test set, Sentinel achieves 76.60\% accuracy, outperforming GLM-5.3 by 6.13 percentage points and its base model by 19.78 points. These results show that even state-of-the-art general-purpose LLMs struggle to identify untrustworthy comments, while iterative action-level learning substantially improves the accuracy of repository-grounded trustworthiness judgments. Our dataset is available at this https URL
Abstract (translated)
URL
https://arxiv.org/abs/2609.37216
PDF
-
Breaking the Illusion of Review Reliability under Static Evaluation: SCOPE Fuzzing for LLM-based Scientific Reviewers
2026-09-29 09:20:03Zhuo Chen, Hao Zeng, Jiawei Liu, Guoxiu He, Le Cai, Liu Haotan, Li Wenbo, Yong Huang, Wei LuAbstract
The rapid growth of submissions and reviewing workload has accelerated the use of large language models (LLMs) in peer review. Prior studies suggest that LLM-based reviewers can penalize content perturbations, such as overclaiming, indicating a certain degree of reliability. Yet these conclusions are largely based on a narrow set of perturbation strategies instantiated with static templates, providing limited evidence of actual reliability. In this paper, we construct a three-level evaluation framework covering perturbations to surface presentation, argumentative logic, and value judgment. Experiments on representative LLM-based reviewers reveal two limitations of static evaluation: stratified vulnerability, where perturbation effects depend on whether the paper's original review score is high or low, and perturbation undercoverage, where a single template misses vulnerabilities exposed by diverse realizations. To address these limitations, we propose SCOPE-Fuzzer, a strategy-aware fuzzer that combines feedback-driven strategy selection with adaptive mutation of paper content. By iteratively probing reviewers with dynamic perturbations, SCOPE-Fuzzer consistently uncovers vulnerabilities overlooked by static evaluation and other baselines.
Abstract (translated)
URL
https://arxiv.org/abs/2609.37097
PDF
