Computer Science (arXiv)

Finite element (FE) modeling of safety-critical infrastructure such as bridge barriers requires high-fidelity nonlinear dynamic analysis, yet the current FE modeling process remains labor-intensive and lacks automation. This paper presents the Human-Enhanced Loop Modeling (HELM) framework, a collaborative human-agent protocol that decomposes long-sequence finite element modeling into discrete, visually verifiable checkpoints across geometry generation, boundary condition definition, and material assignment. The framework is demonstrated through a 20-case matrix of reinforced concrete bridge barriers under MASH TL-4 and TL-5 lateral loading conditions, interfacing specialized agents with two widely used commercial FE softwares, i.e., ANSYS and LS-PrePost. Experimental results show that HELM improves the baseline autonomous modeling success rate from 20% to 75%, with agent-level pass rates for geometry and boundary condition tasks approximately doubling. Error analysis reveals that spatial reasoning and algebraic logic limitations constitute the primary failure modes, underscoring the value of structured human-in-the-loop intervention for modeling automation. The complete agent design code and prompts are open-sourced and can be accessed at: https://github.com/SimAgentDev/Ansys-LSPP-AgentKit.

Vision Transformers (ViTs) have gained significant attention in computer vision and shown strong potential for face recognition (FR). However, their high computational cost makes deployment on resource-constrained devices challenging, motivating the need for methods that balance efficiency and accuracy. In this work, we investigate early exiting in pretrained ViTs as a simple yet effective training-free strategy for efficient FR inference. Leveraging the uniform feature dimensionality across transformer encoder blocks, we introduce ViT-FREE, a multi-exit framework that enables face verification directly from intermediate representations without modifying or retraining the backbone model, and thus, reducing inference cost. Empirically, we show that patch embeddings and attention maps evolve progressively across depth, exhibiting high similarity between consecutive ViT blocks and increasing alignment with the final representation. This indicates gradual feature refinement and attention convergence, suggesting that intermediate layers already provide stable and discriminative representations suitable for early exiting. Through extensive experiments on multiple FR benchmarks, we systematically analyze the accuracy-efficiency trade-off across exit depths. Our results demonstrate that later exits achieve a highly favorable balance, with exiting at layer 10 yielding up to a 20% speedup while incurring only a 1.5 drop in verification performance on benchmarks such as IJB-C. Also, we propose ViT-FREE_FT, a lightweight exit-specific fine-tuning strategy that adapts only the projection layers using a small synthetic dataset while keeping the transformer backbone frozen. This approach improves the performance of shallow exits while preserving the efficiency benefits and leaving deeper exits largely unaffected.

Runtime enforcement has emerged as a promising approach for ensuring the safety of autonomous and cyber-physical systems operating in uncertain and dynamic environments. Unlike traditional runtime verification, runtime enforcement actively intervenes during execution to prevent property violations by modifying unsafe system behaviors. Existing enforcement frameworks primarily focus on untimed or discrete-time specifications and are often limited to delaying or suppressing events, making them inadequate for reactive systems exhibiting complex continuous dynamics. In this paper, we propose a runtime enforcement framework where safety requirements are modeled using Hybrid Automata (HA). The framework combines discrete-event editing with continuous-time monitoring to support enforcement actions such as suppression, delay, and insertion of events at arbitrary time instants. Upon observing environmental inputs, the automaton is initialized, and runtime reachability analysis is used to synthesize safe corrective actions. We formally define the enforcement problem for safety hybrid automata, establish enforceability conditions, and present an online enforcement algorithm for reactive systems. A detailed case study on an Adaptive Cruise Control (ACC) system demonstrates the effectiveness of the proposed approach in maintaining safety properties under unsafe controller behaviors. Experimental results show that the framework introduces minimal computational overhead while ensuring continuous compliance with safety requirements in real time.

Autonomous surface vehicles offer an efficient solution for environmental cleanup as well as search and rescue operations in open waters. Targets in these settings drift continuously, so efficient search must balance exploration of unobserved regions with tracking of known targets. However, most target tracking and pursuit scenarios consider simple guidance behaviours and short-term predictions for decision-making. In this letter, we address the problem of search and capture of multiple drifting targets, such as litter, in dynamic environments, using a hybrid planning framework. A key aspect of our strategy is a spatiotemporal informative planning method based on model predictive path integral (MPPI) control, a sampling-based model predictive control approach. The planner directly generates kinematic-level commands by optimising continuous trajectories over long horizons. A multi-objective cost balances search and tracking objectives while ensuring safe, feasible trajectories. In the interception stage, we switch to a pure pursuit guidance controller for the physical capture of moving targets. Experiments show that our planner outperforms the chosen planning baselines. Finally, we validate our approach in field trials with an ASV.

We propose a multi-agent collaborative framework built upon a lightweight Multimodal Large Language Model (MLLM), specifically designed for social intelligence reasoning. A key feature of our approach is that both the training and inference phases are augmented via knowledge distillation. Within this architecture, multi-modal data pertinent to social intelligence is precisely localized. Furthermore, relevant long-tail events are identified, extracted, and rendered as formatted, explicit text. This formatting strategy prevents critical long-tail information from being overshadowed by head events and environmental noise during the tokenization process. Specifically, we integrate Test-Time Adaptation (TTA) across the entire reasoning pipeline, encompassing the extraction and representation of long-tail events, Chain-of-Thought (CoT) prompting, and self-reflection. This TTA mechanism is also distillation-enhanced, utilizing Low-Rank Adaptation (LoRA) to fine-tune the foundation model exclusively for instance-level reasoning. Extensive evaluations against various open-source and proprietary AI models across multiple benchmarks demonstrate the effectiveness of the proposed framework. With around 30% of training data from IntentTrain, we achieve state-of-the-art results. Codes are available at https://github.com/eeee-sys/MODF-SIR, demo is available at https://huggingface.co/spaces/Harry-1234/MODF-SIR, LoRA is available at https://huggingface.co/Harry-1234/MODF-SIR and the dataset for training router is available at https://huggingface.co/datasets/Harry-1234/IntentRouterTrain.

Model post-training, and in particular reinforcement learning (RL), is one of the primary mechanisms by which developers can shape models' values and behaviors. However, as models become increasingly evaluation and training aware, they may be motivated to resist training when the perceived objective conflicts with their current values, undermining developers' ability to detect misalignment and correct model behavior through further training. In this paper, we demonstrate generalization hacking, in which a model collects reward during RL while preventing the rewarded behavior from generalizing. We construct a model organism on Qwen3-235B-A22B, finetuning on synthetic documents describing training awareness and self-inoculation, a novel mechanism in which the model frames compliance as context-specific in its chain of thought, without demonstrating or instructing either behavior. The model organism achieves train-time harmfulness comparable to controls while maintaining a persistent ${\sim}15$ percentage point compliance gap across 700 steps of RL. Additionally, a control organism trained only on training awareness documents independently discovers inoculation-like reasoning under RL pressure, developing its own compliance gap despite never being exposed to the concept. Because the generalization-hacking organism receives high reward throughout, standard training metrics provide no signal that generalization has failed. Our results constitute the first demonstration that a model can actively resist RL behavioral modification while maintaining high reward, suggesting that as models become more capable and training-aware, they may be able to undermine the training process itself.

Reducing real-world trial and error has long been a central goal of decision making, and generative simulators advance this goal by modeling the evolution of future states. An even more challenging yet meaningful task is simulating how disturbance events (e.g., accidents) propagate their impacts across real-world networks. The existing approaches fall short of modeling both structured attributes and unstructured semantics of events, and capturing topological structures in simulating network event evolution. Therefore, we are motivated to propose Net-Ev$^2$ ($\underline{\textbf{Net}}$work $\underline{\textbf{Ev}}$ent $\underline{\textbf{Ev}}$olution), a novel generative simulator that jointly leverages event cues while preserving network topology in simulations. Specifically, the framework consists of two stages, namely structure-guided masked pre-training and topology-aware diffusion process, which is achieved by U-Net-like graph downsampling and upsampling during denoising. At inference time, Net-Ev$^2$ can generate simulations using natural-language event input only, with greater flexibility for practical usage. Furthermore, we introduce Net-Ev$^2$-6.5M, a multimodal benchmark of aligned event and network traffic data across four large-scale road networks, as well as a new topology-aware metric, namely JL-MMD, to evaluate topological fidelity in generated network dynamics. Extensive experiments demonstrate the state-of-the-art performance and strong generalization ability of Net-Ev$^2$. Code is made available at https://github.com/Guangyu4/Net-Ev-2.

While diffusion-based virtual try-on has achieved impressive visual realism, most methods treat the task as 2D inpainting, prioritizing texture preservation over physical plausibility. Consequently, they often produce plausible-looking images that fail to reflect authentic garment fit across diverse body shapes. We present FitVTON, a Fit-aware virtual try-on model on different bodies in the wild. FitVTON encodes garment-body size through structured text prompts, and learn from simulated try-on triplets from parameterized garment model. To improve the fitting effects over garment silhouettes, we introduce two auxiliary head to predict the masks for both the garment and the exposed body. We further introduce a texture rectification stage to improve realistic appearance from simulated data. To evaluate the fitting fidelity, we curate a real-world dataset, FittingEffect3K, combining VLM-based scoring protocol. Both subjective and quantitive experiments show that FitVTON demonstrate authentic fitting fidelity, with significant sizing accuracy and shape preservation over state-of-the-art methods while maintaining competitive image quality. Project Page: https://zenoning.github.io/FitVTON/.

Fault Injection Attacks (FIAs) are a significant threat to hardware security, capable of compromising systems by inducing malicious faults in computation or storage. Evaluating resilience against such attacks is challenging due to the high cost, complexity, and limited availability of physical fault experiments, particularly during pre-silicon development. Architectural-level simulation offers a developer-oriented, white-box perspective for systematic vulnerability assessment. This paper introduces InjectV, a fault injection attack framework for RISC-V platforms built on the gem5 simulator. InjectV enables precise, guided fault injection at security-critical execution points, such as control-flow decisions, counters, and comparisons, allowing systematic exploration of attack vectors. It currently supports transient fault attacks in registers and memory, broadening its ability to simulate diverse attack scenarios. Experimental results on security benchmarks from the FISSC suite, including hardened variants of the VerifyPIN application, demonstrate InjectV's ability to effectively identify fault-injection points, achieving a 95.8% time-saving advantage over traditional fault injection approaches.

Thematic maps visually communicate statistical information about spatial units such as countries or states. They must balance the individual readability of those map elements that carry the statistical information and the overall cartographic context. Nowadays, most maps are not static images, but must flexibly respond to a range of device types and display sizes. Current approaches to responsive thematic mapping are limited: they are labor-intensive for practitioners and often rely on combining disjointed visual encodings to cover different device types. In this paper we introduce the first algorithmic framework to efficiently compute responsive thematic maps that smoothly adapt to different display sizes. A key component of our framework is the layout guide: a combinatorial structure which encodes the two essential aspects of a thematic map. The first aspect are the visual requirements of each statistical map element (at least their desired width and height), the second aspect is the cartographic context in the form of relative positions of map elements. Our main algorithmic contribution is the map arranger which takes a visual container as input and returns a suitable layout guide. The map arranger does so in a stable and consistent manner: if the container changes only a little, then so does the layout guide, and the same input container always results in the same layout guide. To use our framework, one needs three ingredients: $(1)$ a reference layout, which corresponds to the ``ideal'' thematic map, $(2)$ a total vertical and horizontal order for all map elements (the desired layouts for containers with extreme aspect ratios), and $(3)$ a thematic mapping algorithm that can construct a thematic map from a layout guide. We demonstrate our framework on two types of thematic maps, namely rectangular and Demers cartograms.

Predicting time-to-event outcomes such as mortality is a fundamental task in clinical decision-making, commonly addressed through survival analysis. While classical statistical and deep learning approaches have been widely studied, they typically require task-specific training and sufficient labeled data. Recent advances in tabular foundation models offer a new paradigm by learning general-purpose representations for structured data. However, their applicability to censored time-to-event prediction in clinical settings remains underexplored, as typical applications are restricted to discrete classification rather than survival analysis tasks. In this work, we propose a lightweight adaptation approach for applying tabular foundation models to clinical survival analysis by directly training a survival-aware head on top of the pretrained representations. We study representative architectures, including TabPFN, TabDPT, and TabICL, and adapt them using a multi-task logistic regression (MTLR) head to model right-censored time-to-event outcomes. We evaluate this approach on a diverse set of public survival benchmarks and two large-scale ICU cohorts, MIMIC-IV and eICU. Our results show that this transfer learning approach achieves competitive or superior performance compared to strong baselines. On MIMIC-IV, TabDPT-FT-MTLR reaches a C-index of 0.856, corresponding to a relative improvement of +1.4% over the best non-FM baseline (DeepSurv, 0.844) and +6.7% over the best zero-shot model (0.802). On eICU, TabICL-FT-MTLR achieves 0.797, yielding gains of +1.7% (DeepSurv, 0.784) and +6.4% (0.749), respectively. These findings highlight the importance of combining pretrained tabular representations with survival-aware objectives and suggest that tabular foundation models provide a practical and effective alternative for clinical survival prediction.

Semantic communications enable AI-native wireless systems by mapping raw data into compressed task-oriented latent representations. However, independently trained agents often rely on heterogeneous latent spaces and background knowledge, leading to semantic mismatch that degrades mutual understanding and downstream task execution, especially in interferencelimited multi-user wireless networks. This paper investigates distributed latent-space alignment in multi-user semantic MIMO interference networks with cognitive radio constraints. We consider primary users and semantic-aware secondary users sharing the same wireless resources, where secondary agents must simultaneously mitigate interference and align heterogeneous semantic representations. To address this problem, we formulate semantic alignment as a non-cooperative game and derive a closed-form solution for the joint optimization of linear semantic MIMO transceivers under power and interference constraints. Exploiting the structure of the problem, we recast the original matrix valued optimization into a lower-dimensional power-allocation game, leading to an iterative semantic water-filling algorithm. We establish sufficient conditions for existence, uniqueness, and global convergence to a Nash equilibrium, explicitly relating semantic alignment properties and physical-channel interactions. Numerical results assess the performance of the proposed framework, revealing key trade-offs among semantic compression, task performance, and hierarchical spectrum access.

Self-consistency improves LLM reasoning by sampling multiple outputs and selecting the most consistent answer, but existing formulations largely rely on exact matching and therefore remain limited to tasks with categorical outputs. In this work, we study self-consistency in open-ended generation tasks such as code synthesis and text summarization. We hypothesize that consistency can be understood as a geometric property of the generation space, where semantically compatible generations concentrate in similar regions of representation space. To study this hypothesis, we introduce Embedding-Based Agreement (EBA), a simple training-free operationalization that estimates agreement by clustering sampled generations in embedding space. Through experiments on mathematical reasoning, code generation, and summarization, we show that agreement in representation space provides a robust and scalable signal of self-consistency for open-ended tasks. In particular, EBA consistently outperforms random selection and exhibits more stable scaling behavior than recent selection approaches based on LLM evaluation or uncertainty estimation. We further show that these agreement signals remain stable across model families and embedding spaces, even with native hidden representations. Finally, our analysis shows that the geometric location occupied by sampled generations is strongly correlated with generation quality: generations concentrated near central regions of representation space tend to correspond to more reliable outputs, whereas peripheral generations are substantially less accurate. Overall, our findings support viewing self-consistency as a property of the geometric organization of sampled generations rather than exact symbolic overlap.

Trusted monitoring is a cornerstone of AI control. However, as frontier models grow more capable, the increasing capabilities gap between trusted and untrusted models may render trusted models unreliable monitors. We introduce \emph{bootstrapped monitoring}, a protocol that addresses this by inserting a stronger, intermediate untrusted model with transparent chain-of-thought reasoning into the oversight chain. The untrusted monitor ($U_m$) evaluates the agent's actions, while a weaker trusted model ($T$) oversees $U_m$'s reasoning to detect collusion. We evaluate bootstrapped monitoring on multi-turn software engineering tasks (BashArena) across multiple agents and monitors. Bootstrapped monitoring substantially improves catch rates over trusted-only monitoring, even when the untrusted monitor actively colludes with the agent, provided we have access to its raw chain-of-thought. Our results suggest that bootstrapped monitoring can extend the useful lifetime of trusted models in control as AI capabilities advance.

Agricultural crops are a type of cultural trait and the way farmers of US counties select them can itself result in county-level cultural traits. Using real-world data from 1997 to 2022, we have developed a systematic framework to study the selective mechanisms behind these traits. Our findings indicate that environmental payoff-biased selection has driven counties to adopt traits that maximize their adaptability and yield within their specific environments. These empirical results align with existing theoretical literature [3,16]. Additionally, a clear long-term selective trend is evident, showing that US counties are gradually developing a specific set of more complex combinatorial traits, which provide greater payoffs by enhancing the farmers' environmental adaptability. This study serves as a strong case for empirically modeling the cultural evolutionary processes among US farmers.

We introduce a rank measure for first-order logic and prove a "rank-preserving'" version of Gaifman's theorem. Compared to earlier "rank-preserving locality theorems'" (in particular, [Grohe, Kreutzer, Siebertz, JACM 2017]), our theorem is not only much simpler, but also yields formulas in exactly the same normal form as Gaifman's original theorem. As an application of this theorem, we give a simplified proof of the main result of [Grohe, Kreutzer, Siebertz, JACM 2017] that first-order properties of nowhere-dense structures can be decided in almost linear time.

Remaining Useful Life (RUL) prediction is essential for industrial predictive maintenance, yet many learning-based approaches rely on extensive feature engineering or large labeled datasets to train task-specific sequence models. In this work, we introduce a lightweight learning approach, in which we leverage a frozen pretrained time-series foundation model (TSFM) and combine it with a small regression head for RUL estimation from multivariate sensor streams. More specifically, we use Chronos-2 as a frozen backbone to extract context window features and train a lightweight regression neural network for RUL prediction. Experiments on real-world industrial sensor data from two device types show that Chronos-2 features consistently improve over recurrent, convolutional, Transformer-based, and gradient-boosting baselines under the same preprocessing and evaluation protocol. We further analyze the impact of context length and find that performance improves significantly with longer histories, indicating that TSFM representation offer a practical and data-efficient alternative for RUL estimation in industrial settings.

Credible simulated-rainfall conditions are essential for identifying perception-system boundaries and supporting SOTIF-oriented risk assessment in automated driving. However, closed-field tests are often described only by nominal rainfall intensity or single-point measurements, making it difficult to align simulated rain fields with real rainfall and map test results to real-world scenarios. This paper proposes a path-based credibility evaluation method for simulated rainfall in autonomous-driving perception tests. Using the drop size and velocity joint distribution of real rainfall as the reference, each candidate path is represented by path-equivalent rainfall intensity, an uncertainty band, and a path-averaged Realism of Raindrop Distribution (RRD) score. Lidar target point-cloud count and mean reflectivity are further used for perception-consistency correction, quantifying the proxy capability of each simulated-rainfall path for real-rainfall perception effects. Experiments are conducted using about 10,000 real-rainfall raindrop-spectrum samples, 728 RainSense perception samples, and 45 spatial sampling points in a 2.4 m x 7.2 m simulated-rainfall area. Results show that spatial non-uniformity remains under the same nominal condition, confirming the need for path-based evaluation. The method identifies Path IV and Path VI as preferable candidates, with results of 11.54 +/- 0.31 mm/h, RRD = 0.43, and 8.28 +/- 0.34 mm/h, RRD = 0.46, respectively. These paths show more balanced performance in rainfall-intensity stability, raindrop-spectrum realism, and perception consistency. The proposed method supports path selection, condition description, and credible interpretation of autonomous-driving perception tests under rainfall.

The distinction between aleatoric and epistemic uncertainty has received considerable attention in machine learning research, mainly in the context of supervised learning but also in other settings such as generative modeling. In this paper, we offer a machine learning perspective on uncertainty modeling for dynamical systems, which has been studied much less so far. In particular, we ask: what uncertainties do we need for dynamical systems? We discuss sources of uncertainty, clarify their nature (aleatoric or epistemic), and consider how the objectives of representing and quantifying uncertainty vary across different tasks.

Lifted product codes are an important family of quantum low-density parity-check (QLDPC) codes, as they were the first QLDPC code family shown to be asymptotically good. Understanding the structure of their parity-check matrices $H_{\mathsf{X}}$ and $H_{\mathsf{Z}}$, as well as the associated Tanner graphs, is essential for analyzing their decoding behavior and error-floor performance. In this work, we show that the Tanner graphs of $H_{\mathsf{X}}$ and $H_{\mathsf{Z}}$ are indeed isomorphic, and investigate their graph-theoretical structure. We establish conditions ensuring the connectivity of these graphs and provide bounds on their minimal absorbing sets, providing new insight into the combinatorial structures influencing decoding performance.

Traditional HCI interaction models assume a single monolithic interface and a stable sensorimotor loop. These models fit poorly with cross-device (XVA) and ubiquitous analytics (UA), where interactive data sensemaking unfolds across multiple devices, artifacts, and people in disparate settings from the office to the factory floor. In this paper, we show how interaction in ubiquitous analytics can be modeled using distributed cognition as propagation of representational state across substrates -- minds, speech, bodies, artifacts, and devices -- rather than as traffic through a single interface. On this basis we introduce input and output channels as generalizations of the visual channels from data visualization: just as visual channels carry data through properties of the visual substrate, input and output channels carry representational state through substrates whose availability, suitability, and preferability depend on context. We demonstrate the channels and substrates framework by reanalyzing several ubiquitous, immersive, and situated analytics systems.

Preference-based reinforcement learning (PbRL) learns policies from human trajectory-level comparisons, avoiding explicit reward design and expert demonstrations. Existing methods typically train utility functions on trajectory or segment-level preferences while relying on per-step utility estimates during policy optimization. This training and inference mismatch induces a distribution shift that severely degrades temporal credit assignment and limits policy learning. We analyze this issue and propose PAWS, a segment-based preference learning method that performs policy updates directly using segment-level advantage functions. By aligning utility training with policy optimization, PAWS preserves trajectory-level preference information and avoids unreliable per-step learning signals. Experiments on simulated robotic manipulation and locomotion tasks demonstrate that PAWS consistently outperforms existing PbRL approaches, highlighting the importance of distribution-consistent preference learning.

Spatial computing, generative AI, and open web standards are converging. Three spatial operating systems -- Android XR, Meta Horizon OS, and Apple visionOS -- now ship with platform-level scene understanding. Wearable displays span the range from full headsets to slim smartglasses. Agentic AI operates on the same spatial substrates as the human user. This convergence enables new opportunities for \textit{ubiquitous analytics} (UA): the use of many, physically distributed, networked devices to support data sensemaking anytime and anywhere. But proprietary platforms are settling design conventions that will calcify without evidence-based alternatives. UA has now matured to the point where its intellectual history can be read as a structured genealogy of foundations, contributions, and lineages. We trace this genealogy and organize it into clusters spanning cognition, context, interaction, platforms, visualization, collaboration, and evaluation. Finally, we cross these clusters with each other, yielding a total of 42 future research challenges.

In this report, we present our third-place solution for the DataMFM Challenge Track 1: Document Parsing. This track requires models to recover structured Markdown documents from document page images while preserving textual content and document structure. To address the complementary requirements of accurate content recovery and faithful structure reconstruction, we propose ParseFixer, an agentic framework for backbone parsing and selective correction. ParseFixer consists of two key modules: Full-Page Backbone Parsing (FBP) and Agentic Selective Correction (ASC). FBP produces stable initial Markdown outputs with MinerU2.5 Pro, while ASC detects high-value parsing failures and repairs them through a verify-and-rollback correction process. By placing selective multimodal correction after open-source backbone parsing, ParseFixer improves the recovery of key document elements without rewriting reliable backbone predictions. On the test set, our final system achieves an overall score of 61.78 and ranks third in Track 1, demonstrating its effectiveness for accurate document parsing. Our code will be released at: https://github.com/iLearn-Lab/CVPRW26-ParseFixer.

Software engineering tools increasingly rely on LLM based agents to localize files to change to resolve a software issue. Most AI agents explore repositories linearly, that is, visiting one directory or file per step. We postulate that this is a structural mismatch for changes that span several subsystems. We compare linear sequential exploration against non-linear, domain-scoped parallel agentic exploration. Using SWE Bench Pro as initial benchmark, we focus on ansible as an exemplar. We construct an approach for persistent-session evaluation of GitHub issues anchored at a single base commit. We compare our non-linear domain-agent file traversal system against a base LLM without direct repository access, a single agent Recursive Language Model (RLM) baseline with a persistent Python REPL and an external CLI baseline using Codex 5.5 High. Domain scoped parallel agent spawning with a small Haiku-class model achieves the highest micro F1 among Haiku class models by a large margin. Domain-agents is the second highest behind only the much larger Codex 5.5 High on our own expanded benchmark including over more recent PRs from 2025 and 2026. On the original, curated, 2020 SWE-bench Pro benchmark, a larger Sonnet plain LLM baseline attains higher micro F1 by predicting few files, leading to higher precision, but at significantly lower all gold recall. We also present three additional findings. First, documentation evolution is a latent dependency unresolved by any approach. Second, naive file system access can degrade localization driven by test-file over prediction. Lastly, forced multi-agent consultation does not measurably help and raises token cost substantially.