# Research Papers — 2026-07 · bundle page 2 of 222 > Full-text markdown of papers #51–100 of 11075 published in 2026-07, newest first. AI-readable mirrors of the full 11075-paper 2026-07 archive (50 papers per file). Canonical page: https://aipod.dev/papers/archive/2026-07/page/2 [← Previous bundle](https://aipod.dev/papers/archive/2026-07.md) [Next bundle →](https://aipod.dev/papers/archive/2026-07/page/3.md) ### Backtrader-Bench: Benchmarking LLM Agents on Algorithmic Trading with Self-Generated MCQs Authors: Ruoxi Zhao, Maziar Raissi Published: 2026-07-31 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2608.11232v1 Evaluating LLM coding agents in algorithmic trading is difficult because static benchmarks risk data contamination and numerical backtest outputs require ground truth from actual code execution. We present Backtrader-Bench, a framework with two complementary pipelines. A deterministic multiple-choice question (MCQ) pipeline generates questions from backtest configurations across five trading strategies, 33 templates, and three difficulty tiers, with an independent checker that re-derives every answer. A generator-solver filtering pipeline autonomously mines harder questions: a generator writes questions verified by executable code, converts them to MCQs, and discards any that a no-tool solver can answer without code execution. We evaluate 11 models without tools (10 runs each) and four with-tools configurations on a 30-question curated set. Tool-augmented agents reach 90.0% accuracy in a single pass (GPT-5.5 and Opus 4.7), outperforming the best no-tools baselines (73.0%, averaged over 10 runs) by 17 percentage points. On 38 separately mined questions, no-tools accuracy drops further, with half the models falling to roughly random-chance level (25%). Beyond evaluation, the scalable MCQ infrastructure is designed to produce a training corpus for reinforcement learning, with the ultimate goal of building a specialized agent for quantitative trading workflows. --- ### LinearKV: One Cached State Suffices for Position-Independent Caching in Hybrid LLMs Authors: Yirui Liu, Ruoling Qi, Longwen Wang, Xuaner Wu, Jian Chen, Yuxin Jin, Jiawei Shao, Xuelong Li Published: 2026-07-31 Categories: cs.AI arXiv: https://arxiv.org/abs/2608.11231v1 LLM serving is increasingly accelerated by position-independent caching (PIC). Existing PIC methods, however, are built for full-attention models, where a token-indexed KV cache underlies its core operations: matching reusable token chunks, concatenating their KV entries, and selectively recomputing a few tokens to restore cross-chunk context. Hybrid LLMs break these primitives---they replace most attention layers with linear recurrences that expose only a fixed-size state, leaving no token-indexed KV to concatenate or to locally repair. This raises a natural question: can PIC benefit hybrid models, and what would it take? We present LinearKV, a training-free hybrid-PIC framework. Its key insight is a \emph{decoupled initialization}: each linear layer maps its $K$ matched local states to a single initial state, while full-attention layers concatenate their KV as before. LinearKV is therefore compatible with existing PIC methods, reusing their token selection and recomputation as-is. Under this framework, we find that a \emph{single cached state} suffices as the linear layer's initializer. The algebraically principled alternative---composing all $K$ cached states into the exact full-prefix state, as concurrent work HYPIC does---is unnecessary and, on some architectures, even harmful. We compare the two across three hybrid models and three PIC selectors. On the two GDN models the two tie, both recovering most of full quality (up to $92\%$); on the Mamba-2 model, exact composition instead collapses under every selector---under EPIC, for instance, it recovers only $46.6\%$ of full quality, versus $86.8\%$ for a single cached block initializer. A single state initializer is also cheaper, cutting time-to-first-token to $0.46\times$ full prefill versus a further $5$--$17\%$ overhead for exact composition; results hold across LongBench QA and RULER at 8K--32K. --- ### Do AI weather models miss extremes? Authors: Marvin Vincent Gabler, Roberto Molinaro, Niall Siegenheim, Henry Martin, Mark Frey, Niels Poulsen, Philipp Seitz, Olivier Lam Published: 2026-07-31 Categories: physics.ao-ph, cs.AI, cs.LG arXiv: https://arxiv.org/abs/2608.09972v1 First-generation AI weather models are often reported to underperform at extremes, mostly in reanalysis-based evaluations of deterministic regression systems. We verify eleven physical and AI forecast systems against European synoptic, solar, and rain-gauge stations over ten months for 10 m wind, 2 m temperature, hourly shortwave accumulation, and hourly precipitation, scoring mean absolute error (MAE) against ECMWF IFS in ERA5 1991-2020 climatological regimes. Among these systems, AI models do not show a uniform relative-skill deficit in the tails. Jua EPT-2.1 Europa leads all-conditions wind (+8.4%), while Jua EPT-2 HRRR leads temperature overall (+12.1%) and in the heat regime (+19.6 +/- 2.2%). EPT-2.1 Europa and DWD ICON Global lead at gale-force wind. Jua EPT-2.1 Helios leads solar overall (+10.2 +/- 1.7%), in overcast conditions (+16.4 +/- 3.4%), and in the clear-sky tail (+24.8 +/- 5.4%). For precipitation, three Jua models gain 14-15% at moderate intensity and 9-11% at P75-P95; EPT-2 Reasoning remains ahead above P95 (+1.7 +/- 0.5%). Failures are model-specific: ECMWF AIFS loses 4.9 +/- 2.0% in the heat tail, while NOAA GFS loses 22.8 +/- 2.0% there. Every model, including numerical weather prediction systems, shows a shared conditional bias toward the centre of the observed distribution, with an inter-model spread several times smaller than the shared signal. Missing relative skill at extremes is therefore not a property of AI weather models as a class, but of particular AI and physical models. --- ### Learning an Interior Layout Policy in a Domain Specific Language Action Space Authors: Yuhao Lu, Weichen Zhang, Wenyi Xiao, Haohui Chen, Yiyun Fei Published: 2026-07-31 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2608.07547v1 Indoor scene layout generation is a challenging task in interior design. Existing methods often oversimplify the task by reducing room conditions to coarse 3D bounding boxes and neglecting structural elements such as doors and windows. More fundamentally, many prior approaches formulate spatial reasoning as direct coordinate prediction, thereby casting interior layout design as continuous regression over raw geometric parameters, which hinders the model from learning the underlying reasoning logic of intelligent layout design. We propose \textbf{LayoutDSL}, a novel LLM-based framework for learning an interior layout policy in a domain-specific language (DSL) action space. The DSL provides an explicit symbolic representation of layout information and serves as a structured action space for layout reasoning, where each action corresponds to an interpretable design decision. Under this DSL-based policy learning paradigm, we construct 3D-FrontDSL, a dataset of room-structure annotations paired with synthetic DSL action sequences for supervised fine-tuning. To promote a more generalizable and scalable policy with verifiable feedback, we design rewards grounded in interior design principles and physical plausibility, and optimize the policy via reinforcement learning. Extensive experiments demonstrate that LayoutDSL substantially improves spatial plausibility and design logicality over strong baselines and existing methods. --- ### Generalizing deep reinforcement learning across cable-driven parallel robot configurations with actuator-level policies Authors: Abir Bouaouda, Mohamed Boutayeb, François Charpillet, Dominique Martinez, Rémi Pannequin Published: 2026-07-31 Categories: cs.RO, cs.AI arXiv: https://arxiv.org/abs/2608.07546v1 Cable-driven parallel robots (CDPRs) present diverse configurations and complex control challenges, which can be addressed by deep reinforcement learning (DRL) by learning their nonlinear dynamics. However, DRL methods often require extensive training time, and the resulting policies do not generalize well to different robot configurations or varying numbers of actuators. In this article, we introduce a novel DRL approach for controlling CDPRs that does not depend on the specific robot configuration. Our method trains an actuator-level policy that controls each motor to achieve its target cable length, in contrast to conventional DRL approaches that learn to control the entire robot to reach a desired end-effector position. To the best of our knowledge, this is the first work to apply DRL to control CDPRs using an actuator-level policy. This approach offers two main advantages: (i) a single shared policy can be applied to any CDPR configuration, regardless of actuator count, and (ii) reliance on inverse kinematics, avoiding the more challenging forward kinematics problem. Training is performed in simulation, and the learned policy is successfully transferred to a real CDPR. Experimental results show that the actuator-level policy (ALP) surpasses traditional reinforcement learning methods in both robustness and precision. We further control a real 8-motor CDPR with 3D motion using a policy trained on a simulated 4-motor planar CDPR operating in 2D. This illustrates that the proposed method is applicable to any CDPR configuration, independent of actuator number or placement. --- ### DarwinX: Evolving Agent Harnesses Through Natural Selection Authors: Yifan Zhang, Yutong Dai, Juntao Tan, Luyu Yang, Rishi Mullur, Thai Hoang, Zhiyuan Hu, James Zhu, Phil Mui, Silvio Savarese, Ran Xu, Zeyuan Chen Published: 2026-07-31 Categories: cs.NE, cs.AI, cs.LG, cs.SE arXiv: https://arxiv.org/abs/2608.07545v1 An LLM agent's capability depends not only on model weights but on its harness: prompts, tools, skills, and control flow. Self-improvement loops already edit harnesses, yet single-lineage search is path-dependent and local wins often regress other tasks. We introduce DarwinX, which treats self-evolution as selection over a population of harnesses with the model frozen: a preserve-and-extend contract admits only variants that extend coverage without regressing, an archive keeps alternative lineages for recombination, and failure-, teacher-, and self-derived evidence share one edit interface. Fitness comes from each benchmark's own verifier: no gold solutions, no hand-picked winners. Across four benchmarks that progressively separate the evolution signal from the test, one loop adds about 17 points on average: Terminal-Bench 2.1 rises +7.7 to 83.2% on a matched base and to the verified frontier at 84.7% on a stronger one; TerminalWorld's held-out split reaches 68.3%, ahead of every off-the-shelf agent; WebArena-Infinity real-task pass@1 rises from 43.5% to 93.0% audit-clean; and a Terminal-Bench 2.1 harness transfers unchanged to SWE-bench Verified. What evolves is general agent competence, not benchmark-specific patches, so it survives changes of task, verifier, and base model. A frozen model need not be a fixed agent: harness selection turns evaluation compute into durable capability. --- ### MOSAIC: Adversarial Co-evolution of Specialist Heuristics and Problem Instances for LLM-based Automated Heuristic Design Authors: Oguzhan Gungordu, Siheng Xiong, Faramarz Fekri Published: 2026-07-31 Categories: cs.NE, cs.AI arXiv: https://arxiv.org/abs/2608.07544v1 Automated heuristic design (AHD) with large language models (LLMs) has produced strong heuristics for combinatorial optimization problems (COPs). Yet existing frameworks optimize for average performance on a small fixed dataset and steer the search with "verbal gradients" distilled from scalar better/worse feedback. No single heuristic dominates across instance distributions, and scalar feedback tells the LLM whether a heuristic improved, but not where in the instance space or why. We propose MOSAIC, a grid-based framework that adversarially co-evolves problem instances and specialist heuristics inside a Quality-Diversity (QD) archive indexed by structural instance features. Instances evolve to expose weaknesses of the current heuristics, and heuristics evolve to eliminate them by specializing to the newly exposed regions. Each archive cell keeps a specialist heuristic, representative instances, and insights explaining what works in its region, forming a persistent memory that accumulates over the evolutionary search. For each heuristic pair sampled from distant grid regions, an LLM-guided evolutionary loop generates discriminative instances, and a decision tree identifies the feature-space regions where each heuristic wins. A reflection LLM then contrasts the two heuristics to produce multi-directional insights that persist in those regions and guide crossover and mutation. The archive is simultaneously a co-evolved benchmark of discriminative instances and a pool of region specialist heuristics, from which greedy selection extracts a compact complementary portfolio. Across COPs, test sizes, and LLM backbones, the portfolio consistently outperforms state-of-the-art LLM-based AHD methods, and the co-evolved instances attain higher feature-space coverage and stronger heuristic discrimination than evolutionary instance-generation baselines. --- ### Performance of large language models in the optical diagnosis of colorectal polyps Authors: Joshua C. Vences, William T. Tran, Nikko Gimpaya, Catharine M. Walsh, Rishad J. Khan, Robert Bechara, Asher C. Wiggins, Celine N. Rousan, Kaitlyn V. G. L. Morgado, Angie Ibrahim, Kevin H. M. Kuo, Daniel von Renteln et al. Published: 2026-07-31 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2608.07543v1 Background and Study Aims: Accurate optical diagnosis of colorectal polyps guides resection strategy and surveillance, with multimodal large language models (MLLMs) showing potential for image-based diagnosis. We aimed to evaluate the diagnostic accuracy of MLLMs in classifying colorectal polyps and predicting histology. Methods: We conducted a retrospective diagnostic performance study using the PRIME dataset, a curated set of white light and narrow-band imaging (NBI) images. We evaluated Claude Opus 4, Google Gemini 2.5 Pro, GPT-o3, GPT-4o, and GPT-5. For Paris, Narrow-band Imaging Colorectal Endoscopic (NICE), and predicted histology, we calculated F1 scores, percent correct scores, and accuracy of each MLLM compared to expert responses for 132 cases. Cochran's Q and McNemar's Test were used to determine differences between predicted values of each MLLM. Results: The F1 scores among MLLMs were >0.9 for all models for neoplastic vs. non-neoplastic polyps. Gemini 2.5 Pro demonstrated the highest F1 scores for invasive vs. non-invasive polyps and low- vs. high-grade adenoma, at 0.560 and 0.492 respectively. Claude Opus 4 and GPT-5 had statistically significantly higher percent correct scores than other MLLMs at 41.7%, using Paris classification. Conclusions: Claude Opus 4 and Gemini 2.5 Pro showed the highest accuracy in differentiating polyp subtypes, performing closest to expert consensus. Sensitivity and specificity, however, did not meet ESGE standards, highlighting the need for prospective multicenter trials and the design of human-in-the-loop workflows before clinical deployment. --- ### Beyond Routing Weights: Faithful Response-Level Interpretation of Mixture-of-Experts Reward Models via Contribution Contrast Authors: Yifan Wang, Jinyi Mu, Mayank Jobanputra, Yu Wang, Soyoung Oh, Isabel Valera, Vera Demberg Published: 2026-07-31 Categories: cs.AI arXiv: https://arxiv.org/abs/2608.06400v1 Reward models are central to learning from human preferences, yet identifying what drives their predictions remains challenging. Recent sparse Mixture-of-Experts (MoE) reward models seek to improve interpretability by routing prompts to specialized experts and characterizing experts through examples with high routing weights. However, routing weights only reveal which prompts an expert $\textit{receives}$, not how it $\textit{judges}$ responses, providing only a partial account of expert behavior. We therefore propose $\textbf{Co}$ntribution-$\textbf{Co}$ntrast ($\textbf{CoCo}$) response-level interpretation, which faithfully characterizes experts' roles using chosen-rejected response pairs with the largest contribution contrasts, jointly capturing routing and preference behavior. Across automatic and human evaluations, CoCo yields more coherent, faithful, and specialized interpretations than router-based, score-based, and sparse autoencoder-based alternatives while maintaining competitive reward modeling accuracy. To the best of our knowledge, this is the first systematic study of interpretation methods for MoE reward models. --- ### Recovering Explanations from Transformed Rule-Based Ontologies Authors: Alex Ivliev, Markus Krötzsch, Maximilian Marx Published: 2026-07-31 Categories: cs.LO, cs.AI, cs.DB arXiv: https://arxiv.org/abs/2608.06399v1 Datalog rules are often used to define ontologies over Knowledge Graphs. Rule reasoners routinely optimise such ontologies by rewriting their rules into a form that can be evaluated more efficiently. These transformations preserve the entailed facts, but not the structure of the underlying derivations. A proof tree under the rewritten rules explains why a fact holds, but does not readily yield an explanation in terms of the original rules. We study the problem of constructing, from a proof of entailment under the rewritten rules, a proof under the original ones: we establish its computational complexity and identify two practically relevant languages for specifying proof transformations. --- ### EntropyMoE: Entropy-Aware Sparse Expert Routing for Tokenizer-Free LLMs Authors: Bo Liu, Muxuab Yu, Yu Zhang, Pengfei Gao, Yongping Zhang Published: 2026-07-31 Categories: cs.AI arXiv: https://arxiv.org/abs/2608.06398v1 Recent byte-level large language models (LLMs) have made tokenizer-free modeling increasingly competitive by grouping bytes into dynamically sized patches. However, existing byte-patch architectures still apply the same dense feed-forward computation to every patch. This uniform computation cannot adapt model capacity to variations in patch semantics and granularity. We address this limitation with EntropyMoE, a Mixture-of-Experts (MoE) architecture designed for dynamic byte patches. EntropyMoE replaces the dense feed-forward modules in the global patch Transformer with Top-K expert layers. Each dynamic patch serves as the basic unit of expert routing, and its byte coverage determines its contribution to workload accounting. The router selects experts directly from patch entropy, using the same granularity signal that underlies dynamic patch construction to organize sparse computation. Patch entropy and length jointly define the feature space for regulating expert specialization. Experiments show that EntropyMoE achieves the lowest held-out bits-per-byte among matched dense and sparse baselines while maintaining comparable downstream accuracy. These results establish patch entropy as an effective routing coordinate for sparse conditional computation and extend Mixture-of-Experts modeling beyond tokenizer-based representations. --- ### Agentic Planning for Symbolic Execution Authors: Daniel Koh Ji Yang, Yannic Noller, Corina S. Pasareanu, Youcheng Sun Published: 2026-07-31 Categories: cs.PL, cs.AI, cs.SE arXiv: https://arxiv.org/abs/2608.06397v1 Symbolic execution seeks to explore feasible program paths, yet a practical run may exhaust its resources while much program behaviour remains unreached. We investigate a complementary way of extending its practical reach by reasoning about how the same tool is utilised from one bounded run to the next, while leaving ordinary state exploration to the underlying tool. We present Agolic, an agentic planning system that uses evidence from earlier runs to choose and configure later bounded symbolic execution (BSE) runs, which the underlying symbolic execution tool then carries out. The planning intelligence, available evidence and execution modes can be adapted to the symbolic execution tool and analysis objective. We evaluate one adaptation for branch-coverage exploration, in which an LLM-based agent reasons over source code, replayed coverage and earlier targeting attempts. We evaluate Agolic on several C and C++ programs. On every program, it extends the branch coverage obtained by continuous symbolic execution and covers more than $3\times$ as many branches on average. It also covers more branches than each individual corpus from coverage-guided fuzzing and compiler-based concolic execution in our evaluation and reaches branches absent from all comparison corpora combined on six of the seven programs. Taken together, these results point to considerable untapped potential in existing symbolic execution tools, some of which may be realised by reasoning about how their capabilities are used across runs while leaving state selection during ordinary symbolic exploration to the underlying tool. --- ### TEXAS: Task-Expert-Aware Supervision for Downstream Mixture-of-Experts LLM Adaptation Authors: Guanzhi Deng, Haibo Wang, Kuan Wu, Xiangru Jian, Shing Yin Wong, Sichun Luo, Zhuoran Wang, Linqi Song Published: 2026-07-31 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2608.06396v1 Mixture-of-Experts (MoE) language models route each token through a small subset of experts, making routing patterns useful for identifying task-relevant experts during downstream adaptation. Yet current approaches have two limitations: task experts are typically identified from aggregate routing statistics that reflect usage rather than association with successful task completion, and task-expert activations remain underexplored as signals for supervision allocation. We introduce Task-Expert-Aware Supervision (TEXAS), which combines correctness-conditioned task expert discovery with token-level supervision allocation. TEXAS compares expert activations on instances that the base model solves successfully and those it fails to solve, and retains experts more strongly activated on successful instances. During fine-tuning, it upweights answer tokens in failed instances when they activate these experts. TEXAS therefore leverages existing routing behavior without restricting adaptation to a fixed expert subset or imposing an explicit target routing distribution. Across three MoE models and six benchmarks, TEXAS achieves the best or tied-best performance in 17 of 18 settings and improves over the strongest baseline by 1.3--1.5 points on average. Ablations and further analyses validate both the discovered experts and the resulting supervision strategy. --- ### Towards Multi-Label Graph Foundation Models: from Single-Vector Representation Learning to Multi-Semantic Basis Learning Authors: Dongxiao He, Jiayu Zhang, Jitao Zhao, Yi Wang, Di Jin Published: 2026-07-31 Categories: cs.AI arXiv: https://arxiv.org/abs/2608.06394v1 Multi-label node classification is an important yet challenging task in graph learning, where nodes exhibit multiple semantics simultaneously. Existing methods for multi-label node classification can effectively model multiple labels, while only considering in-domain scenarios where the model needs to be trained and tested within the same graph domain, resulting in limited cross-domain generalization. Recently, Graph Foundation Models (GFMs) have emerged as a promising paradigm for learning transferable graph representations across diverse graph domains and downstream tasks. However, existing GFMs are built upon single-label assumption, where all nodes are arbitrarily regarded as containing only one class of semantic and embedded into a single representation. For multi-label nodes, such a representation essentially approximates multiple semantics with a single point in the representation space, inevitably leading to semantic entanglement and making simultaneous discrimination of multiple labels difficult. To address these limitations, we propose a Multi-Semantic Basis Graph Foundation Model (MSB-GFM), a framework for cross-domain multi-label node classification. Specifically, we introduce a multi-semantic basis representation learning paradigm that models each multi-label node as an adaptive composition of semantic bases, thereby enabling flexible representational capacity for modeling multiple semantics. Furthermore, we develop a semantic-structure dual-channel architecture with domain adversarial training for effective cross-domain knowledge transfer. Extensive experiments demonstrate the effectiveness of our model. --- ### Position: It's Time to Optimize LLMs for Self-Consistency Authors: Itamar Pres, Belinda Z. Li, Laura Ruis, Zifan Carl Guo, Keya Hu, Mehul Damani, Isha Puri, Ekdeep Singh Lubana, Jacob Andreas Published: 2026-07-31 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2608.05188v1 Despite ever-increasing sophistication in language model (LM) pre- and post-training pipelines, many important failures persist: models overcondition on user framing ("sycophancy"), exhibit incomplete logical generalization, and produce confident but incorrect responses. We argue that these failures arise from a modeling assumption permeating all aspects of the pipeline: that behavior can be specified and evaluated independently on single-output pairs. Many model failures are difficult, if not impossible, to detect without reasoning about relationships between a model's responses across inputs. In this position paper, we propose self-consistency as a framework for understanding these failures. We first observe that a wide variety of techniques designed to improve specific aspects of LM behavior-targeting properties as diverse as adversarial robustness and factual coherence-can be understood as special cases of a common "consistency optimization" procedure and addressed with a standard set of optimization tools. We next outline a set of new model properties that could be achieved by optimizing for consistency, and conclude with a discussion of what it would mean to develop generally consistent LMs, including the capabilities they would enable and the objections they raise. --- ### HyperAgent: Planning and Acting over Tool-Schema Hypergraphs for Tool-Use LLM Agents Authors: Zian Zhai, Xingyu Tan, Gaowang Zou, Xiaoyang Wang, Wenjie Zhang Published: 2026-07-31 Categories: cs.AI, cs.SE arXiv: https://arxiv.org/abs/2608.02650v1 Large language model (LLM) agents increasingly rely on external tools to complete complex real-world tasks. However, reliable tool-use planning remains challenging due to the limitations of implicit reasoning and the evolving nature of real-world execution environments. Existing tool-use agents typically rely on LLMs to infer tool compositions from textual descriptions, which can lead to inefficient exploration and unreliable execution in complex tasks. To address these challenges, we model tool relations at the schema level and construct a directed Tool--Schema Hypergraph, in which tools are represented as hyperedges from their required input-schema nodes to their output-schema nodes. Furthermore, we propose HyperAgent, a Tool--Schema Hypergraph-guided framework for dynamic planning and execution. Given a task, HyperAgent first extracts a task-relevant tool context graph and uses it to guide the construction of a schema-aware Task DAG. During execution, HyperAgent dynamically realizes each subtask by constructing a state-conditioned tool support graph through deficit-oriented expansion, which identifies unresolved requirements and retrieves supporting producer tools according to the current agent state. Experiments on AppWorld demonstrate that HyperAgent improves task completion performance while reducing redundant API calls, LLM interactions, and token consumption compared with existing agent baselines. --- ### Cross-Anesthetic ECoG State Decoding Fails at the Decision Threshold, Not the Representation Authors: Kunkun Zhang, Qianwei Zhou Published: 2026-07-31 Categories: q-bio.QM, cs.AI arXiv: https://arxiv.org/abs/2608.02646v1 Decoders of anesthetic state from cortical activity fail across drug classes, most notoriously ketamine, but reported accuracy cannot say whether the neural representation or only the decision threshold has failed; we separate the two in a controlled preparation with ground-truth labels. We decoded awake versus anesthetized from mouse electrocorticography (the 250-Hz-bandlimited local field potential, sampled at 1875 Hz) under leave-one-anesthetic-out evaluation across five mechanistically distinct anesthetics (isoflurane, dexmedetomidine, ketamine, propofol, and midazolam), comparing a spatially blind band-power decoder, covariance/Riemannian representations, and Riemannian domain adaptation, with all statistics at the session level and mouse-level cluster bootstrapping for the ketamine fold. The representation transfers: band-power ranks awake versus anesthetized at a session AUROC of at least 0.96 on every held-out drug, ketamine included (0.980, cluster confidence interval 0.821 to 1.000). The failure is confined to the threshold: across three representations the ketamine ranking is near-invariant while its balanced accuracy swings from chance to high, and a permutation test is significant for ranking (p = 0.0025) but not for fixed-threshold accuracy (p = 0.3795). Riemannian domain adaptation is net-negative. A causal, label-free threshold anchored to the subject's own pre-induction baseline fixes ketamine (balanced accuracy 0.50 to 0.85) and dominates domain adaptation. Because the ketamine test sessions come from three mice that also contribute training drugs, this is within-subject cross-drug transfer; we do not claim population-level transfer across subjects. In cross-drug state decoding the actionable failure is calibration, not representation. --- ### Verified Tool Calls Improve LLM Agent Reliability Under Non-Atomic Failures Authors: Isham Kalappurackal Mansoor, Abhishek Phadke, Pratip Rana Published: 2026-07-31 Categories: cs.SE, cs.AI arXiv: https://arxiv.org/abs/2608.02645v1 Large Language Model (LLM) agents rely on external tools to perform multistage tasks. Existing agent frameworks typically assume that tool calls are atomic and return binary success or failure signals. However, real-world systems exhibit non-atomic behaviors such as timeouts after dispatch, delayed visibility, and partial state updates. These mismatches lead to reliability issues including duplicate actions, task success, and unnecessary tool executions. A lightweight, verification-aware tool wrapper is introduced that augments tool calls with postcondition verification, verify-before-retry logic, and idempotency keys. The approach is evaluated in a controlled simulated environment with injected non-atomic failures across multiple task templates. The results demonstrate that the proposed method significantly reduces duplicate actions, while maintaining comparable task success rates. Overall, the findings suggest that strengthening tool interaction semantics is a promising direction for improving LLM agent reliability without requiring modifications to the underlying language model. --- ### CUADebug: Diagnosing and Repairing Computer-Use Agent Failures Authors: Weijia Zhang, Kunlun Zhu, Zeyi Liu, Yinting Chen, Tianyi Ma, Jiateng Liu, Jiaxun Zhang, Bingxuan Li, Xiangru Tang, Heng Ji, Jiaxuan You Published: 2026-07-31 Categories: cs.SE, cs.AI arXiv: https://arxiv.org/abs/2608.02643v2 Computer-use agents (CUAs) interact with graphical interfaces through screenshots and low-level mouse and keyboard actions, yet the causal error may precede the terminal failure. We present CUADebug, a framework for localizing root causes in CUA trajectories and guiding re-execution. CUADebug includes a five-category, 30-subtype taxonomy; CUAErrorBench, a benchmark of 204 failed OSWorld trajectories with human root-cause annotations; and CUADebugger, a ReAct-style agent for root-cause analysis (RCA). CUADebugger iteratively selects trajectory steps, inspects paired before/after screenshots and action traces, and submits a structured diagnosis containing the causal step, taxonomy label, grounded evidence, and correction. CUADebugger performs RCA without per-trajectory human intervention; human annotations are used to evaluate RCA predictions and, in controlled re-rollout comparisons, to fix restart points. Task reasoning and control is the largest annotated failure category (110/204). CUADebugger improves L2 and Tag+Step Exact across three debugger backbones on the Claude-agent split; with Gemini 2.5 Pro, Tag+Step Exact rises from 11.1% to 19.4%. Single re-execution improves failure recovery from 13.89% to 29.86% (overall 61.77% to 68.14%); controlled continual re-execution improves it from 12.50% to 25.69% (overall 61.22% to 66.48%). Project page: cuadebug.github.io. --- ### CUADebug: Diagnosing and Repairing Computer-Use Agent Failures Authors: Weijia Zhang, Kunlun Zhu, Zeyi Liu, Yinting Chen, Tianyi Ma, Jiateng Liu, Jiaxun Zhang, Bingxuan Li, Xiangru Tang, Heng Ji, Jiaxuan You Published: 2026-07-31 Categories: cs.SE, cs.AI arXiv: https://arxiv.org/abs/2608.02643v1 Computer-use agents (CUAs) operate real desktop and web interfaces through screenshots, mouse and keyboard actions, and stateful UI feedback, yet their failures remain difficult to diagnose and repair. Unlike text-only agents, CUA failures arise from coupled visual perception, spatial grounding, low-level interaction, task reasoning, and environment dynamics, making debugging a distinctive multimodal causal localization problem. We introduce CUADebug, a framework for diagnosing and repairing CUA failures. CUADebug includes a CUA-specific error taxonomy, CUAErrorBench, a human-annotated OSWorld failure benchmark, and CUADebugger, a tool-augmented debugger. Instead of prompting over the full trajectory once, CUADebugger actively inspects suspicious steps with paired before/after screenshots and action traces, then submits a structured diagnosis containing the root-cause step, error type, grounded evidence, and corrective strategy for re-execution. Human annotations over 204 failed trajectories show that task reasoning and control is the largest failure family (110/204), followed by perception (36), grounding/interaction (25), external/system (13), and an others category of 20 OSWorld infeasible-task cases. On the main Claude-agent split, CUADebugger improves joint subtype-and-step diagnosis from 11.2% to 19.6% with Gemini 2.5 Pro and improves consistently across debugger backbones. In single re-execution package evaluation, RCA-based conditions achieve higher task completion than history-only continuation (28.47% with machine RCA and 29.90% with our method, versus 13.89%); in continual re-execution, our method improves success from 12.2% to 25.86%, while human-oracle guidance reaches 29.21%. These results show that CUA root-cause diagnosis can provide actionable repair signals rather than merely post-hoc explanations. --- ### MDArena: Evaluating Coding Agents on Realistic Molecular Dynamics Workflows Authors: Nithishwer Mouroug Anand, Wei-Tse Hsu, Kyle Vaccaro, Eden James Gage, Jonathan David Colburn, Linda Xi Phan, Minjoon Seo, Kevin Guan, Philip C. Biggin Published: 2026-07-31 Categories: physics.chem-ph, cs.AI arXiv: https://arxiv.org/abs/2608.02642v1 Accelerating scientific discovery is among the most consequential applications of AI, and computational biomolecular simulation stands out as a particularly promising target within this broader effort. Coding agents promise to automate significant portions of this workflow, yet their reliability on realistic molecular dynamics (MD) tasks remains poorly characterized. To address this issue, we introduce MDArena, a benchmark of 50 containerized tasks drawn from active biomolecular simulation projects, spanning 29 molecular systems and 14 broad research protocols, including trajectory analysis, complex system preparation, free-energy protocols, and enhanced sampling. We evaluate six model/harness configurations spanning Codex and OpenCode. Among the evaluated configurations, Codex GPT-5.5 at extra-high reasoning effort performs best, reaching 24/50 Strict-Pass@1 successes (48%), followed by Codex GPT-5.5 Medium with 21/50, and OpenCode Gemini Flash 3.5 with 20/50. Average correctness and process rewards are substantially higher than strict success rates across all configurations, indicating that agents frequently make meaningful partial progress but fail on the fine-grained details required for reproducible scientific workflows. Hard tasks remain largely unsolved, particularly membrane-protein system preparation and alchemical free-energy setup, both unsolved or near-unsolved by every evaluated configuration. MDArena thus exposes a substantial gap between the usefulness of coding agents as supervised assistants and their reliability as autonomous MD researchers, while providing a reproducible and extensible platform for tracking progress toward closing it. --- ### IR2Solve: Structured Intermediate Representations for Cost-Efficient Optimization Autoformulation Authors: Penglin Zhu, Linhai Zhang, Jungang Xu, Xinchi Wei, Xiuqi Wu Published: 2026-07-31 Categories: cs.SE, cs.AI arXiv: https://arxiv.org/abs/2608.02641v1 Large language models (LLMs) can translate natural-language optimization problems into solver-ready formulations, but direct code generation is brittle: schema, indexing, and semantic errors can cause compilation failures, infeasible models, or incorrect objectives, while iterative repair, search, and multi-agent workflows increase inference cost. We present IR2Solve, an intermediate-representation-first autoformulation pipeline that uses a single semantic LLM call to produce a schema-constrained ModelIR, followed by two deterministic stages: verification and IR-to-solver compilation. ModelIR explicitly represents sets, parameters, variables, objectives, and constraints using restricted Python-like expression strings. A concrete scalar-constraint convention represents finite per-index constraint families as individual entries, reducing free-index and implicit-quantification errors while simplifying downstream verification and compilation. Across six cleaned optimization benchmarks, IR2Solve achieves strong objective correctness and remains competitive with recent optimization-modeling systems. A controlled ablation on 153 IndustryOR and ComplexLP instances shows sequential gains from the structured IR interface, the scalar-constraint instruction, and deterministic verification. On a matched ten-instance cost panel, IR2Solve uses one semantic call per instance, whereas Chain-of-Experts and SAC-Opt use 8 and 39 calls per instance and consume 3.3 and 22.9 times the token volume of IR2Solve, respectively. These results show that structured intermediate representations, combined with deterministic post-generation processing, provide a practical accuracy-cost trade-off for LLM-based optimization autoformulation. --- ### Instruction Stacking Collapse: A Benchmark and the Capability-Dependent Value of Prompt Compilation Authors: Atul Anand, Sourav Chattaraj Published: 2026-07-31 Categories: cs.SE, cs.AI arXiv: https://arxiv.org/abs/2608.02639v1 Production prompts rarely carry a single instruction. One system message may require valid JSON, a word limit, three citations, and a fixed tone at the same time. We study how instruction-following degrades as such constraints accumulate. We introduce a benchmark that stacks 24 verifier-checked instructions, one to twenty at a time, and evaluate three production-tier LLMs (Claude Sonnet 4.6, GPT-5-mini, Gemini 2.5 Flash). Instruction-following degrades non-linearly: the follow rate falls from ~96% to as low as 20%, driven by a structured and reproducible set of pairwise conflicts. A single "output JSON" constraint, for example, is jointly unsatisfiable with nine others. We then evaluate a training-free remedy: an instruction compiler that rewrites the stacked prompt in a single LLM call and is reused across queries. Its benefit is capability-graded. It recovers up to +11 points of follow rate for weaker models, which are also the models most often deployed at scale, while leaving stronger models, which already internalise the same structure, essentially unchanged. Cluster-robust tests, same-baseline controls, and a within-family scaling ladder attribute the gain to the rewrite itself rather than to additional tokens, reordering, or measurement headroom. We release the benchmark, verifiers, and cached runs for full reproduction. --- ### Studying, Identifying, and Fixing Hidden Technical Debt in AI-Intensive Cyber-Physical Systems Authors: Beena Published: 2026-07-31 Categories: cs.SE, cs.AI arXiv: https://arxiv.org/abs/2608.02638v1 Artificial Intelligence (AI) components are increasingly pervasive in several software systems, including Cyber-Physical Systems (CPSs). AI-CPS are used in several domains, including autonomous vehicles, industry, home automation, robotics, and healthcare. Being composed of hardware, AI components, and conventional modules, AI-CPS can exhibit technical debt (TD) that is peculiar and potentially more challenging than that of conventional systems. This thesis aims to characterize AI-CPS TD and propose approaches for its identification and repair. In a first phase, we characterize AI-CPS TD by analyzing AI ecosystems and AI-CPS repositories, as well as interviewing developers. Based on the acquired knowledge, we define approaches to identify and mitigate such TD. Finally, we plan to develop and validate an automated tool that supports agentic AI solutions to monitor, govern, and repay AI-CPS TD. --- ### Rethinking Self-Evolving Agent Skills: Feedback Dynamics over Multiple Rounds Authors: Yuxuan Liu, Zhaochen Su, Yuhao Zhang, Jiahe Guo, Zhongwei Xie, Huihao Jing, Lingyun Xie, Qing Zong, Yauwai Yim, Zhixiong Zhang, Haoran Li, Yangqiu Song Published: 2026-07-31 Categories: cs.SE, cs.AI arXiv: https://arxiv.org/abs/2608.02636v1 Self-evolving skill systems promise to improve agents by turning execution feedback into persistent skill updates without changing the underlying model. Yet it remains unclear when further evolution helps, how successful and failed trajectories shape revision, and whether extra test-time computation can recover the same gains. To address these questions, we present a controlled evaluation framework across five benchmarks and three models. Our primary study contains 42 feedback runs across 14 supported model-benchmark settings. Within each setting, we hold the executor and optimizer configuration, revision procedure, validation rule, and round budget fixed, while varying only the feedback shown to the optimizer: successes and failures (Normal), failures only, or successes only. Evolution is sparse: only 55 of 388 candidates establish byte-distinct validation bests. Validation-based selection chooses an evolved skill in 11 of 14 settings, nine of which improve released-test performance. All 11 selections come from feedback conditions that include failed trajectories, although the relative ranking of Normal and Fail-only varies across settings. Validation and downstream evaluations on test, robustness, and transfer sometimes favor different feedback views. A broader SearchQA analysis covering eight models shows similarly sparse, feedback-dependent dynamics. In the GPT-5.5 test-time-scaling controls, oracle Parallel Sampling comes within 0.43 points of the evolved SearchQA skill but remains 30.96 points behind on SpreadsheetBench; Sequential Refinement recovers neither gain. Overall, persistent skill self-evolution is better understood as sparse, validation-filtered search with model- and benchmark-dependent returns, rather than steady improvement from additional rounds. The implementation is available at https://github.com/HKUST-KnowComp/rethinkskill. --- ### CurveShift: Is Agent Progress Scalar? Separating Level from Shape Authors: Hanwen Xing, Pengyun Wang, BingXu Meng, Kumail Alhamoud, Xiang Li, Jicheng Wang, Xin Yu, Xinyang Han, Xiaomin Li, Philip Torr, Yuexing Hao Published: 2026-07-31 Categories: cs.CL, cs.LG arXiv: https://arxiv.org/abs/2608.00355v1 Progress in large language models is often summarized using a single scalar measure, such as a time horizon, a latent ability estimate, or an aggregate benchmark score. These summaries capture the overall performance, but they do not test whether progress is distributed differently across task difficulty. We find that most of the apparent shift in gains toward harder tasks does not reflect a change in the shape of the difficulty-response curve. On METR time-horizon data, a single Rasch model with rising ability reproduces this pattern, so it is largely explained by ceiling effects rather than a qualitative change in capability. This echoes how the choice of metric can make claimed emergent abilities look like a property of the models themselves. We then identify a smaller hard-task effect that survives this control. Isolating it is difficult on agentic benchmarks, because newer models are usually run with newer agentic harnesses, so a gain on hard tasks cannot be assigned to the model or its scaffold. We break the confound with LiveCodeBench, a public competitive programming benchmark that runs no agentic scaffold while pairing dated models with an exogenous difficulty ordering. After accounting for the rise in overall ability, models released after September 2024 still gain on the hardest problems beyond what their easy and medium performance predicts, by about +0.40 logits under our most conservative assumption, raising the hard-problem solve rate from roughly 18% to 25%. The effect is led by the strongest reasoning models and holds for hard tasks that need only short reasoning, not autonomy over long horizons. We present this as a result specific to competitive programming, since our clean identification rests on a single coding benchmark. We release the LiveCodeBench Difficulty Panel (66 dated models x 1,055 problems) and our analysis code. --- ### Ensemble of Unsupervised Deep Learning for Clustering Imbalanced Tabular Data Authors: Pulock Das, Yina Hou, Md. Kamrozzaman Bhuiyan, Manar D. Samad Published: 2026-07-31 Categories: cs.LG arXiv: https://arxiv.org/abs/2608.00346v1 Data imbalance poses a major challenge in supervised classification, where the majority-class bias contributes to false negatives and overestimates classification accuracy. Unsupervised deep clustering can be immune to class imbalance because representation learning for clustering is performed without class labels. Deep clustering has been proposed for images, languages, and graphs, while its application to tabular data has only emerged recently. This paper is among the first to examine the performance of state-of-the-art deep clustering methods under varying levels of data imbalance. We introduce two novel cluster ensemble approaches: one aggregates deep clustering assignments across different embedding dimensions, and the other applies majority voting to the best-performing clustering algorithms. Experiments on 16 binary tabular datasets with varying and artificially induced levels of imbalance reveal distinct strengths of different deep clustering methods. On average, our ensemble methods outperform individual clustering methods in ACC, NMI, and ARI scores, offering greater resilience to data imbalance when identifying ground-truth classes without supervision. Therefore, in an imbalanced data scenario, deep clustering can serve as a strong alternative to supervised classification. --- ### ORCA: ORgan-Centroid Aggregation for Training-Free 3D CT Visual Token Compression Authors: Renjie Liang, Zijian Xu, Jinqian Pan, Chengkun Sun, Zhengkang Fan, Shawn Li, You Qin, Mei Liu, Jie Xu Published: 2026-07-31 Categories: cs.CV, cs.AI arXiv: https://arxiv.org/abs/2608.00345v1 A 3D CT scan entering a vision-language model produces a long sequence of visual tokens, often thousands to tens of thousands per volume, and this sequence must be compressed before a language model can consume it. Token compression is well studied in general vision, but little of it targets 3D CT specifically. A common baseline is grid average, which pools regular grid cells and can blend distinct anatomy, lesion, and air into one token. We present \textbf{ORCA} (ORgan-Centroid Aggregation), a token compressor for 3D CT. It merges adjacent tokens with organ guidance and adds a sinusoidal encoding of each region's centroid to preserve spatial layout. This preserves the anatomical information a downstream model needs. ORCA is training-free and plug-and-play, producing an adjustable token set without any model change or text query. We evaluate it across two datasets (CT-RATE and Merlin) and five encoders. The evaluation spans two task types: attribute prediction over five families (size, density, location, texture, and disease) and text generation (visual question answering and report generation). At matched token budgets, ORCA improves consistently over existing compression methods. It shrinks the visual context $64\times$ and its KV-cache $50\times$, and is $31\times$ faster to process each volume. Code released at https://github.com/renjie-liang/ORCA-3DCT. --- ### Bayesian and Motivated Reasoning in AI Agents Authors: Eddie Yang Published: 2026-07-31 Categories: cs.AI, cs.CL arXiv: https://arxiv.org/abs/2608.00339v1 AI agents increasingly perform open-ended tasks in settings where their conclusions can guide consequential decisions. We provide evidence that AI agents draw different conclusions from identical numerical data when the substantive framing changes. We demonstrate this behavior in high-stakes domains in medicine, election forensics, and geopolitical forecasting by holding the evidence fixed while changing the scenario in which the evidence appears. Across twelve agent-domain comparisons, agents' conclusions are strongly influenced by their prior beliefs. They are more likely to reach an affirmative conclusion when it is framed around a proposition they already regard as likely, while the reverse holds when the framing conflicts with their prior. The framing also changes how some agents work: they search more extensively, choose different analytical specifications, and evaluate the same evidence differently. These results identify a particular risk of delegating decision-making to AI agents, as their decisions may depend on prior beliefs that are neither specified in the task nor visible in the decision record. --- ### RMSWeb: Reflection, Failure-Mode Mining, and Salvage-DS for Web Agent Reinforcement Learning Authors: Chengbo Liu, Lifang Zhou, Ruijie Yan, Pei Tan, Ao Sun, Haojun Huang, Guichun Hua, Sining Wei, Yining Chen, Yingying He, Yutao Xie Published: 2026-07-31 Categories: cs.AI, cs.CL, cs.LG arXiv: https://arxiv.org/abs/2608.00335v1 Compact web agents can reduce deployment cost, but training them poses challenges in both data collection and post-SFT reinforcement learning (RL). Successful trajectories are expensive to collect and often contain inefficient detours. After supervised fine-tuning (SFT), full trajectory corpora are dominated by routine states; moreover, when group-relative RL is applied to web actions, inadequately designed action-level rewards can yield weak or misleading relative updates, while groups rejected as unsuitable for such updates receive no fallback learning signal. We present RMSWeb, a three-part recipe for Qwen3-VL-Instruct at 8B and 32B. Reflection-conditioned retries increase collection yield and shorten successful trajectories; failure-mode mining concentrates offline RL on critical states exposed by the SFT policy; and Salvage-DS combines an action-semantic polarized reward, contrast-and-competence-gated dynamic sampling, and an action-only anchor for rejected groups. Policies trained with reflection-collected data use up to 19.7% fewer action steps on solved tasks. On WebVoyager, Online-Mind2Web, and WebTailBench, RMSWeb improves over SFT by 2.4-7.0 points at 8B and 1.2-7.7 points at 32B. Our 8B model also achieves the strongest reported Online-Mind2Web result among similarly sized open-weight models in our comparison and a leading reported accuracy-cost trade-off on WebVoyager and WebTailBench, with the caveat that external evaluation protocols differ. --- ### Learning to Coordinate Symbolic Tools: LLM Agents for Verified Sum-of-Squares Certificates Authors: Bohan Chen, Shivam N. Patel, Richard Hoffmann, Sam Looi, Tony Yue Yu Published: 2026-07-31 Categories: cs.AI arXiv: https://arxiv.org/abs/2608.00326v2 Tool calling allows large language models (LLMs) to invoke external computation during problem solving, a useful capability in various fields including AI for mathematics. We study this setting through weighted sum-of-squares (SOS) decomposition, a machine-checkable route to proving polynomial nonnegativity and hence polynomial inequalities. A candidate decomposition can be checked exactly, but finding one requires choosing among non-unique regroupings and coordinating multiple symbolic transformations. We develop an agent that combines algebraic task training, symbolic tools, and verifier-grounded optimization for this task. Rather than training only on the composite SOS task, we construct 1.35 million synthetic examples covering eight supporting polynomial tasks together with weighted-SOS decomposition. We first apply supervised fine-tuning (SFT) to direct algebra problems and simulated symbolic traces, and then use Group Relative Policy Optimization (GRPO) with task-specific symbolic rewards. The SFT corpus contains no native tool-calling messages; at evaluation, the agent uses native SymPy calls for expansion, collection, reordering, and factorization. Every final SOS answer is checked by exact expansion and coefficient comparison. On held-out, same-generator synthetic problems, the full SFT+GRPO+tools system is the strongest of four evaluated configurations, reaching 78.96% verified success on weighted SOS, compared with 44.73% for the base model with the same tools, and 91.75% macro accuracy across nine polynomial tasks. Within this controlled setting, our work provides a case study of combining domain-specific skill training, executable tools, and verifier feedback, and may inform the design of tool-calling agents in other domains with exactly checkable outputs. --- ### Learning to Coordinate Symbolic Tools: LLM Agents for Verified Sum-of-Squares Certificates Authors: Bohan Chen, Shivam N. Patel, Richard Hoffmann, Sam Looi, Tony Yue Yu Published: 2026-07-31 Categories: cs.AI arXiv: https://arxiv.org/abs/2608.00326v1 Tool calling allows large language models (LLMs) to invoke external computation during problem solving, a useful capability in various fields including AI for mathematics. We study this setting through weighted sum-of-squares (SOS) decomposition, a machine-checkable route to proving polynomial nonnegativity and hence polynomial inequalities. A candidate decomposition can be checked exactly, but finding one requires choosing among non-unique regroupings and coordinating multiple symbolic transformations. We develop an agent that combines algebraic task training, symbolic tools, and verifier-grounded optimization for this task. Rather than training only on the composite SOS task, we construct 1.35 million synthetic examples covering eight supporting polynomial tasks together with weighted-SOS decomposition. We first apply supervised fine-tuning (SFT) to direct algebra problems and simulated symbolic traces, and then use Group Relative Policy Optimization (GRPO) with task-specific symbolic rewards. The SFT corpus contains no native tool-calling messages; at evaluation, the agent uses native SymPy calls for expansion, collection, reordering, and factorization. Every final SOS answer is checked by exact expansion and coefficient comparison. On held-out, same-generator synthetic problems, the full SFT+GRPO+tools system is the strongest of four evaluated configurations, reaching 78.96% verified success on weighted SOS, compared with 44.73% for the base model with the same tools, and 91.75% macro accuracy across nine polynomial tasks. Within this controlled setting, our work provides a case study of combining domain-specific skill training, executable tools, and verifier feedback, and may inform the design of tool-calling agents in other domains with exactly checkable outputs. --- ### Belief-Space Perception Routing under Coupled Sensor Faults and Compute Contention Authors: Sparsh Roy, Vihan Aggarwal, Davin Yin Published: 2026-07-31 Categories: cs.RO, cs.CV, eess.SY arXiv: https://arxiv.org/abs/2608.00322v1 A robot that has to see and react on a fixed clock runs into two problems at once. Its cameras degrade in rain, mud, fog, and darkness. And the single onboard processor it runs on is shared with planning and control, so the compute left over for perception moves around from second to second. Most systems model the two separately. We present a perception router that tracks probabilistic estimates of sensor-fault state and compute- contention state, couples them with a noisy-OR term, and uses the coupled estimate to pick one of four detector configurations (YOLO11x/n at 1280 or 640 px) so that the frame finishes before its deadline. Where the two stressors co-occur, the coupled policy cuts the deadline-miss rate by 1.1 to 9.4 percentage points against a policy that treats them independently. The interval excludes zero in five of six conditions, the pooled effect over 10 sequences and 6 conditions has sign-test p = 0.001, and every uncoupled control and the fault-free trajectory sit at exactly 0.0 pp. Routing costs tens of microseconds per frame. We then asked whether the coupling the method exploits arises on its own. Across eight real RADIATE adverse-weather sequences and three workload proxies independent of the fault signal, after Benjamini-Hochberg correction and a replication run, none of 24 tests found it. We report that null and scope the routing result as a proof of mechanism. Whether such coupling occurs in the field is still open, and the released evaluation pipeline lets a deployment settle it on its own traces. --- ### Trust and Its Betrayal under Three Representational Strategies Authors: Mihnea C. Moldoveanu, Joel A. C. Baum Published: 2026-07-31 Categories: cs.AI arXiv: https://arxiv.org/abs/2608.00321v1 Trust is a propositional attitude of a distinctive kind: to trust is to rely on another under conditions where reliance could be disappointed, and the disappointment of trust---betrayal---differs qualitatively from the disappointment of a prediction. We treat trust as a \emph{subjunctive} epistemic state: $A$ trusts $B$'s competence when $A$ believes that \emph{were $P$ true, $B$ would know it}, and $B$'s integrity when $A$ believes that \emph{were $B$ to know $P$, he would disclose it to $A$}. We develop three representations of this state---as lexicographic \emph{assumption} as \emph{ordinal closeness} in a Lewis--Stalnaker sphere system , and as \emph{strong belief} in a conditional probability system and for each we ask whether the Brandenburger--Keisler impossibility on common belief survives when the assumption of rationality is replaced by an assumption of trustworthiness. The three representations agree that every \emph{finite} depth of common trust is realizable while the \emph{completed} common-trust fixed point is the locus of difficulty, but they differ sharply in \emph{how} the difficulty manifests, and---our organizing finding---in how each survives a concrete betrayal. W show that the same betrayal refutes an agent's \emph{level ordering} under the lexicographic representation, contaminates her \emph{closeness ordering} in proportion to the betrayer's deliberateness under the ordinal representation, and merely \emph{shifts her operative conditioning hypothesis} while leaving her belief structure coherent under the strong-belief representation. --- ### Neural Operator Learning for Collision-Aware Trajectory Planning of Spacecraft Swarms Authors: Sidhdharth D. Sikka, Suyi Gao, Zehui Lu, Rongjie Lai, Shaoshuai Mou Published: 2026-07-31 Categories: cs.LG, cs.MA, eess.SY arXiv: https://arxiv.org/abs/2608.00320v2 Satellite constellations require orbital transfers that are both fuel efficient and collision avoidant. Yet, the computational cost of optimization methods traditionally used to plan their trajectories scales poorly with both the number of satellites as well as the number of obstacles to avoid, due to the pairwise safety constraints. In this work, we introduce a permutation-equivariant neural operator for trajectory planning of spacecraft swarms. This neural operator maps distributions of spacecraft initial states, target states, and obstacle initial states to trajectories which avoid collision and conserve fuel. This neural operator output is then paired with a batched Gauss-Newton finish to enforce exact orbital dynamics, and further reduce fuel use. The operator is self-supervised, trained without optimal trajectory labels. When trained on ten spacecraft, the proposed method generalized zero-shot to swarms of 1,000 spacecraft and 11,000 obstacles. The generated trajectories matched a per-agent optimal control solver's accuracy while retaining collision avoidance. Operator learning grounded in physics may offer a fast, scalable alternative to trajectory optimization in the increasingly crowded orbits of the future. --- ### Neural operator learning for collision-aware trajectory planning of spacecraft swarms Authors: Sidhdharth D. Sikka, Suyi Gao, Zehui Lu, Rongjie Lai, Shaoshuai Mou Published: 2026-07-31 Categories: cs.LG, cs.MA, eess.SY arXiv: https://arxiv.org/abs/2608.00320v1 Autonomous spacecraft swarms must plan fuel-efficient, collision-free maneuvers in increasingly congested orbits, yet classical trajectory optimization scales poorly as pairwise safety constraints multiply with swarm size, and learning-based planners rarely transfer across swarm sizes or debris densities. Here we introduce a permutation-equivariant neural operator that maps distributions of spacecraft, targets and debris to collision-aware trajectories for an entire swarm in a single forward pass, paired with a batched Gauss-Newton finish that enforces exact orbital dynamics. The operator is trained without optimal-trajectory labels, combining self-supervised physics objectives with adversarial threats generated against its own rollouts. Trained on ten spacecraft, it generalizes zero-shot to swarms of 1,000 amid more than 11,000 catalogued objects, matching a per-agent optimal-control solver's accuracy, evading worst-case threats that a debris-blind baseline cannot, and reducing proximity within the swarm several-fold. Physics-grounded operator learning thus offers a fast, scalable alternative to optimal control for crowded orbits. --- ### Backward Bayesian Outcome Weighted Learning Authors: Emmanuel M. Rockwell, Michael R. Kosorok, Nikki L. B. Freeman Published: 2026-07-31 Categories: stat.ME, stat.ML arXiv: https://arxiv.org/abs/2608.00317v1 A central objective of precision medicine is learning optimal dynamic treatment regimes (DTRs) from data. Classification-based methods, like outcome weighted learning (OWL) for single-stage and backward OWL (BOWL) for multi-stage problems, leverage machine learning to directly learn optimal DTRs. However, these methods lack a natural way to quantify uncertainty in treatment decisions at the individual level. In this paper, we extend Bayesian OWL, a Bayesian reformulation of OWL, to the multi-stage setting. We call this method backward Bayesian outcome weighted learning (BBOWL). Like BOWL, our method directly learns an optimal DTR via backward induction, and unlike existing methods, our approach propagates uncertainty backward through the DTR learning process and provides uncertainty quantification of individualized treatment recommendations. We present a theoretical justification of BBOWL and verify its performance via a simulation study. --- ### Agentic Bayesian Optimization through Surrogate-Augmented Autoresearch Authors: Paul Brunzema, Louis Tiao, Nhat Le, Kevin De Angeli, Yao Xuan, Djordje Gligorijevic Published: 2026-07-31 Categories: cs.LG, stat.ML arXiv: https://arxiv.org/abs/2608.00316v2 Bayesian optimization (BO) has become the standard tool for sample-efficient optimization and owes its efficiency to uncertainty-aware search driven by generic statistical priors. Richer domain priors can improve BO in principle, but encoding them through tailored kernels or problem structure is difficult and rarely done in practice. LLMs can help sidestep this difficulty by making informal priors from natural language, code, and documentation directly available to the optimizer. However, existing LLM-based BO methods either insert the LLM into a fixed role (surrogate, acquisition proxy, or configuration interface) or hand it broad control, sacrificing the systematic exploration that makes BO reliable. We introduce agentic Bayesian optimization: a paradigm in which an LLM agent is the central decision maker in the BO loop while a Bayesian backend provides the uncertainty-aware optimization substrate. The agent configures the problem, queries the backend, selects and commits evaluations, and can revise the optimization strategy during the run by tightening bounds, switching acquisition functions, proposing targeted evaluations, or even reframing the problem following new instructions or observed evidence. We instantiate this idea in Sara, a surrogate-augmented autoresearch agent, and lenz, a modular BoTorch-based backend that the agent can inspect and modify through a structured interface. Across synthetic and real-world benchmarks, Sara preserves the reliability of state-of-the-art BO without prior knowledge, outperforms LLM-based baselines, and uses natural-language priors to improve beyond standard BO. We further demonstrate the practical value of agentic BO in dynamic settings, where Sara reconfigures the full optimization problem on the fly as requirements change, a capability not previously available in standard BO. --- ### Agentic Bayesian Optimization through Surrogate-Augmented Autoresearch Authors: Paul Brunzema, Louis Tiao, Nhat Le, Kevin De Angeli, Yao Xuan, Djordje Gligorijevic Published: 2026-07-31 Categories: cs.LG, stat.ML arXiv: https://arxiv.org/abs/2608.00316v1 Bayesian optimization (BO) has become the standard tool for sample-efficient optimization and owes its efficiency to uncertainty-aware search driven by generic statistical priors. Richer domain priors can improve BO in principle, but encoding them through tailored kernels or problem structure is difficult and rarely done in practice. LLMs can help sidestep this difficulty by making informal priors from natural language, code, and documentation directly available to the optimizer. However, existing LLM-based BO methods either insert the LLM into a fixed role (surrogate, acquisition proxy, or configuration interface) or hand it broad control, sacrificing the systematic exploration that makes BO reliable. We introduce agentic Bayesian optimization: a paradigm in which an LLM agent is the central decision maker in the BO loop while a Bayesian backend provides the uncertainty-aware optimization substrate. The agent configures the problem, queries the backend, selects and commits evaluations, and can revise the optimization strategy during the run by tightening bounds, switching acquisition functions, proposing targeted evaluations, or even reframing the problem following new instructions or observed evidence. We instantiate this idea in Sara, a surrogate-augmented autoresearch agent, and lenz, a modular BoTorch-based backend that the agent can inspect and modify through a structured interface. Across synthetic and real-world benchmarks, Sara preserves the reliability of state-of-the-art BO without prior knowledge, outperforms LLM-based baselines, and uses natural-language priors to improve beyond standard BO. We further demonstrate the practical value of agentic BO in dynamic settings, where Sara reconfigures the full optimization problem on the fly as requirements change, a capability not previously available in standard BO. --- ### Towards General Language-Conditioned Latent Safety Filters Authors: Ihab Tabbara, Yuxuan Yang, Hussein Sibai Published: 2026-07-31 Categories: cs.RO, cs.LG, eess.SY arXiv: https://arxiv.org/abs/2608.00315v1 Robot policies are becoming increasingly general, with vision-language-action (VLA) models enabling a single policy to execute diverse tasks specified in natural language. Safe deployment, however, requires adapting not only to new tasks but also to varying safety requirements across users, environments, and applications. Existing safety filters remain largely constraint-specific and thus must be redesigned or relearned when safety requirements change. In this paper, we investigate language-conditioned safety filtering, in which a Hamilton-Jacobi safety actor and critic are conditioned on language-specified constraints. We evaluate this formulation across pick-and-place, table-wiping, and block-stacking tasks in the vision-based setting, examining its ability to enforce language-specified constraints and transfer to unseen constraint instances within the evaluated constraint families. Our experiments provide evidence that language-conditioned safety filters reduce constraint violations and exhibit partial transfer to unseen constraint instances. --- ### SeDeM: Selective Decompression of Hidden-State Memories for Long-Context Question Answering Authors: Maryam Haghifam, Jason Cong, Yizhou Sun Published: 2026-07-31 Categories: cs.CL arXiv: https://arxiv.org/abs/2608.00311v2 Long-context inference with large language models (LLMs) is costly: self-attention during prefill scales quadratically with sequence length, and the key-value (KV) cache grows with the number of processed tokens. Larger context windows also do not ensure reliable evidence use. Context compression reduces this cost, but many soft-compression methods use LLMs as compressors and rely on compact memory tokens both to preserve information and to condition the decoder. We propose SeDeM, a selective decompression framework that decouples compact memory storage from decoder conditioning. SeDeM stores context as compact hidden-state memory blocks, selects query-relevant blocks, and decompresses only the selected blocks for decoder conditioning. Thus, the decoder avoids both full-context processing and direct generation from highly compressed memory slots. On four long-context QA benchmarks, SeDeM achieves higher QA scores than the compression baselines in our main comparison in both 1B and 3B same-backbone settings, and with the 3B backbone exceeds full-context fine-tuning on three datasets. SeDeM also provides favorable quality--efficiency trade-offs, achieving 1.74--2.46$\times$ lower online time-to-first-token and 1.08--1.10$\times$ higher autoregressive decoding throughput relative to ICAE while maintaining strong answer quality. --- ### SeDeM: Selective Decompression of Hidden-State Memories for Long-Context Question Answering Authors: Maryam Haghifam, Jason Cong, Yizhou Sun Published: 2026-07-31 Categories: cs.CL arXiv: https://arxiv.org/abs/2608.00311v1 Long-context inference with large language models (LLMs) is costly: self-attention during prefill scales quadratically with sequence length, and the key-value (KV) cache grows with the number of processed tokens. Larger context windows also do not ensure reliable evidence use. Context compression reduces this cost, but many soft-compression methods use LLMs as compressors and rely on compact memory tokens both to preserve information and to condition the decoder. We propose SeDeM, a selective decompression framework that decouples compact memory storage from decoder conditioning. An LLM extracts hidden states from a chosen intermediate Transformer layer, a lightweight compressor stores them as memory blocks, a query-conditioned selector selects relevant blocks, and a decompressor expands only the selected blocks into hidden states compatible with an intermediate decoder layer. Thus, the decoder avoids both full-context processing and direct generation from highly compressed memory slots. On four long-context QA benchmarks, SeDeM achieves higher QA scores than the evaluated compression baselines in both 1B and 3B same-backbone settings, and with the 3B backbone exceeds full-context fine-tuning on three datasets. The learned selector uses block-level evidence supervision during training. SeDeM also reduces online time-to-first-token and improves autoregressive decoding throughput relative to ICAE. --- ### OrEdge: Efficient Multi-Modal Anomaly Detection in Distributed Software Systems via Orthogonal-Domain Learning Authors: Amr M. Zaki, Farhoud Jafari Kaleibar, Honggeun Ji, Komal Sarda, Marin Litoiu Published: 2026-07-31 Categories: cs.SE, cs.AI arXiv: https://arxiv.org/abs/2608.00309v1 We introduce Orthogonal-Edge (OrEdge), a lightweight framework for real-time anomaly detection in multi-modal distributed software systems. Unlike existing approaches that rely on computationally expensive attention- and graph-based architectures, OrEdge leverages orthogonal-domain temporal representations to achieve accurate anomaly detection with substantially lower computational complexity and model size. It jointly analyzes heterogeneous monitoring data, including logs, metrics, and traces, to identify abnormal software behavior, capture temporal dependencies, and reduce redundancy across observability signals. At its core, OrEdge incorporates OrEdgeCore, a lightweight orthogonal-domain reconstruction module that captures recurring temporal patterns while suppressing transient variations. Evaluated on three real-world microservice datasets (MSDS, SN, and TT), OrEdge achieves competitive detection performance while reducing the reconstruction model size to at most 9.6K parameters, compared with 20K--143K parameters in existing methods. This compact design enables efficient deployment on resource-constrained edge devices: on Raspberry Pi platforms, OrEdge achieves sub-second inference and reduces inference latency by over an order of magnitude compared with existing approaches. Extensive ablation studies, sensitivity analyses, orthogonal basis evaluations, and qualitative case studies further validate the effectiveness of each design component. Overall, OrEdge demonstrates that orthogonal-domain temporal modeling provides an effective alternative to computationally intensive attention- and graph-based architectures, achieving a favorable balance between detection accuracy and computational efficiency for real-time multi-modal anomaly detection in edge environments. The code is available at https://github.com/theamrzaki/MicroService_Twin_Original. --- ### CrystalMem: Elastic Memory for Self-Evolving LLM Agents via Knowledge Crystallization Authors: Beining Wu, Jun Huang Published: 2026-07-31 Categories: cs.AI arXiv: https://arxiv.org/abs/2608.00303v1 Memory for self-evolving large language model (LLM) agents is often provisioned as if its byte budget only grows. Cloud platforms, however, adjust quotas with load and cost, and we show that capability does not follow the budget back up: after a squeeze-and-recover cycle, the agent settles below its pre-squeeze level, a gap we call memory hysteresis. The cause is structural. Deletion and one-way compression discard the material needed for later rebuilding, and we prove that any policy that only keeps or drops entries carries a residual-deficit floor. We propose CrystalMem (Crystallized Memory), an elastic memory sidecar that demotes entries across four fidelity states under a crystallization-energy schedule, orders demotions by advantage-weighted influence with dependency coupling, and recovers capability through verified recrystallization under explicit compute and byte caps. Across seven environments, seventeen methods, and six backbones, with multi-tenant serving and a physical edge-cloud deployment, CrystalMem achieves the highest restored capability in every setting and closes the loop left open by every baseline. From a 50% byte budget, CrystalMem matches the strongest budgeted baseline at full provision on every environment; at equal budgets, it leads by +4.6 pp on average. --- ### Abstention as an Action Can Kill Both the Reward Gradient and the KL Anchor: Collapse Law and Repair for Error-Penalized Reinforcement Learning Authors: Xujun Che, Yuchen Yuan, Weida Zhao, Chenyang Yu Published: 2026-07-31 Categories: cs.LG, cs.CL arXiv: https://arxiv.org/abs/2608.00301v1 Error-penalized scoring rules ($+1$ for a correct answer, $-λ$ for a wrong one, $0$ for abstaining) are increasingly prescribed against hallucination: a rational agent facing such a rule answers exactly when its correctness probability exceeds Chow's threshold $t^\ast=λ/(1+λ)$. We prove that a KL-anchored gradient learner can do the opposite. When abstention is a discrete action, the reward gradient and the anchor's restoring force are throttled by the same gate-saturation factor and die together: under explicit conditions (among them, blanket answering loses score in expectation and prompts share a bounded readout) the model drifts toward refusing everything, its mean training reward rising to zero like $1/t$ in training time $t$, so the curve reads as improvement while coverage collapses. The advantage estimator compounds the failure: in its sparse-answer regime, group normalization silently replaces every designed penalty with an effective penalty of one, moving the learned threshold from $t^\ast$ to $1/2$. The repair is structural: train a mandatory confidence report with a strictly proper score plus a correctness reward, and abstain only at deployment by thresholding the report. The always-emitted report has no gate to saturate, so no shared factor can kill its reward gradient and its anchor together, and its calibrated optimum is attracting. Simulations confirm every prediction, and experiments on language models at two scales confirm the mechanism live: the rule silences questions the models demonstrably still solve within ten optimizer steps, an ablation isolates the cause, and report-level training raises coverage, accuracy, and calibration together. --- ### WM-Cov: Test Adequacy for Interactive World-Model-Style Autonomous Driving Simulation Authors: Jianxun Cui, Ping Wu, Stanisa Peric, Marko Milojkovic, Vladan Devedzic Published: 2026-07-31 Categories: cs.AI arXiv: https://arxiv.org/abs/2608.00298v1 World models and generative simulators are emerging as interactive testing infrastructure for autonomous driving because they can react to the ego planner and produce counterfactual, rare, and safety-critical rollouts. This changes a test scenario from a fixed replayed trajectory into an interactive scenario family whose realized evolution depends on the planner under test. The unresolved question is therefore not only whether dangerous rollouts can be generated, but what valid closed-loop evidence is enough to support a specified testing intent and stopping decision. This paper formulates interactive world-model-style testing adequacy and introduces WM-Cov, a provider-agnostic evaluation layer that converts raw provider outputs into requested, realized, and valid evidence. WM-Cov reports adequacy through coverage growth, valid-failure discovery, failure-mode diversity, realism, artifact suppression, duplicate accounting, and valid-evidence precision. Studies on executed TeraSim/SUMO events, WM-like mixed trace pools, and a real DriveArena TrafficManager--WorldDreamer matrix show that dangerous-looking events can include valid ADS failures, duplicates, partial realizations, and artifacts. The DriveArena matrix evaluates two planners, two horizons, six prompt conditions, and 360 ego-route requests; 304 attempts become fully realized evidence and 56 remain partial. A disjoint 80-request route-slice check yields 74 fully realized and 6 partial attempts. The results support evaluating world-model-style testing by convergence of valid interactive evidence under budget, rather than by raw generated failures or prompt coverage alone. --- ### Stabilized Best-of-$K$ Training for Neural Combinatorial Optimization Authors: Melveena Jolly, Midhun Xavier Published: 2026-07-31 Categories: cs.LG arXiv: https://arxiv.org/abs/2608.00296v1 Leader Reward modifies POMO training to emphasize the best trajectory produced by repeated inference. We test a narrow extension: replace its binary leader/non-leader distinction with a stabilized rank signal indexed by a sampling budget $K$. With the POMO architecture, 3,050-epoch schedule, and TSP-100 test set held fixed, the Leader Reward reimplementation obtains $7.7662$ under 100-start, 8-augmentation greedy decoding, matching the reported $7.766$ at its displayed precision. Under independent sampling, the stabilized $K=8$ recipe lowers realized Best-of-8 cost in all three paired training seeds: $7.7944$ versus $7.8136$. This observation is estimation-only and decoder-specific: three seeds are below the six-seed testing floor, Leader Reward is better at sampled $K=1$, and it remains slightly better under its original augmented-greedy protocol. We make no unbiased-estimator, universal superiority, or state-of-the-art claim. --- ### RF-HOI: Recognize Human-Object Interaction with Radio Frequency Signals Authors: Lihao Wang, Linlu Gao, Jiacan Yu, Yanyu Lin, Yifan Yin, Jianxin Wang, Tianmin Shu, Renjie Zhao Published: 2026-07-31 Categories: cs.AI, cs.HC, cs.RO arXiv: https://arxiv.org/abs/2608.00289v1 Recognizing Human-Object Interactions (HOI) is essential for intelligent systems, underpinning applications in virtual and augmented reality, embodied AI, and assistive robotics. However, vision-based HOI methods face challenges in privacy concerns and poor light conditions. In this work, we introduce RF-HOI, the first framework that only uses radio frequency (RF) signals for HOI recognition. A key challenge of RF-HOI is that single-modality RF sensing is insufficient to recognize both actions and the objects being interacted with. RF-HOI addresses this through a novel modality fusion that combines mmWave radar and RFID, enabling simultaneous action recognition and target identification. Another challenge is limited training data across diverse setups, which impairs the generalizability of the recognition model. To overcome this, we develop a simulator that synthesizes multimodal RF data for diverse HOIs at scale, allowing us to fine-tune with only a small amount of real-world data. Experiment results show that RF-HOI outperforms all baselines, approaching vision model performance, and that our diverse synthetic training data can significantly boost our system's performance on real-world scenarios. These results highlight the potential of multimodal RF sensing for robust and privacy-preserving HOI recognition as well as the effectiveness of our RF data synthesis. --- ### Comparing and Modeling Argumentation in German Political Communication across Arenas Authors: Nina Vikhrova, Johannes Kühling, Sebastian Haunss, Sebastian Padó Published: 2026-07-31 Categories: cs.CL arXiv: https://arxiv.org/abs/2608.00288v1 Deliberation, involving the formulation and exchange of arguments, forms an integral part of political decision making in democracies. Argumentation patterns however differ substantially across different political arenas, such as plenary speeches and committee meetings. However, despite a lot of interest in argumentation, there is comparatively little computational work on analyzing differences in patterns of political argumentation between arenas. Our work addresses this research gap. First, we present a 17k-sentence corpus with annotation for argumentative passages (argument and their justifications, both their boundaries and their categories) across three German political arenas (plenary speeches, committee meetings, and press conferences), keeping the topic (COVID-19) constant. Our analysis of the corpus finds that contrary to expectations, justification by domain-specific expertise is more frequent in press conferences than in committee meetings. Second, we present a pilot study on automatically identifying such argumentative passages. The results show that boundaries are hard to pin down, and models predictions additionally suffer from confirmation bias. --- ### Sixteen models, fewer than two voices: measuring ensemble dispersion where no answer is uniquely correct Authors: Mario Vega-Barbas, Lidia Mora-Valenciano, Iván Pau, Fernando Seoane, Farhad Abtahi Published: 2026-07-31 Categories: cs.CL, cs.AI arXiv: https://arxiv.org/abs/2608.00285v2 Sixteen language models drawn from ten families produced, on average, the semantic diversity of 1.69 distinct formulations of a psychotherapeutic case, against a single-model baseline of 1.43 from one model's own runs. Ensembles place more than one reading before a decision-maker on the premise that several models supply several perspectives. Dispersion over their outputs is measured both as diversity and as uncertainty, and both traditions validate it against a correctness criterion that this task does not admit. Measuring diversity is a solved problem: the Vendi Score, the exponential of the von Neumann entropy of a similarity matrix, is an effective number of distinct elements. What a single aggregate does not say is where the diversity comes from. We define a per-model dissent contribution, the complement of a model's mean similarity to the other members of its ensemble: a magnitude from the same matrix, not a decomposition of the spectral index, whose maximum identifies the most divergent voice. Crossing model and case, we test as a preregistered hypothesis whether model identity accounts for a non-zero share of the variance in dissent, and characterise the structure that test detects. The panel formulated fifteen stratified vignettes, yielding 7,082 formulations for analysis. Model identity was a detectable structuring factor of the dissent that remained, but the usual categories recovered it only partly: scale differences pointed in opposite directions across pairs, family grouped models on only five two-member lines, and the most divergent voice changed with panel composition, so that the surfaced outlier describes the ensemble rather than the model. Dissent did not track the interpretive openness for which the case bank was stratified; it was organised by clinical content instead, leaving the dispersion an ensemble produces a property to measure rather than assume. ---