Daily Papers

Vision-Language Models (VLMs) are increasingly deployed in settings where reliable visual grounding carries operational consequences, yet their behavior under progressively coercive prompt phrasing remains undercharacterized. Existing hallucination benchmarks predominantly rely on neutral prompts and binary detection, leaving open how both the incidence and the intensity of fabrication respond to graded linguistic pressure across structurally distinct task types. We present Ghost-100, a procedurally constructed benchmark of 800 synthetically generated images spanning eight categories across three task families: text-illegibility, time-reading, and object-absence, each designed under a negative-ground-truth principle that guarantees the queried target is absent, illegible, or indeterminate by construction. Every image is paired with five prompts drawn from a structured 5-Level Prompt Intensity Framework, holding the image and task identity fixed while varying only directive force, so that tone is isolated as the sole independent variable. We adopt a dual-track evaluation protocol: a rule-based H-Rate measuring the proportion of responses in which a model crosses from grounded refusal into unsupported positive commitment, and a GPT-4o-mini-judged H-Score on a 1-5 scale characterizing the confidence and specificity of fabrication once it occurs. We additionally release a three-stage automated validation workflow, which retrospectively confirms 717 of 800 images as strictly compliant. Evaluating nine open-weight VLMs, we find that H-Rate and H-Score dissociate substantially across model families, reading-style and presence-detection subsets respond to prompt pressure in qualitatively different ways, and several models exhibit non-monotonic sensitivity peaking at intermediate tone levels: patterns that aggregate metrics obscure.

Optical Chemical Structure Recognition (OCSR) aims to translate molecular diagrams in scientific literature into machine-readable formats, but current systems remain unreliable on real-world images due to substantial visual and chemical complexity. We introduce MOSAIC, a dual-dimensional difficulty framework with 37 fine-grained labels that jointly characterize visual interference and chemical semantic challenges in molecular diagrams. Based on this framework, we construct MolRecBench-Wild, a benchmark of 5,029 structures from 820 recent chemistry papers, covering the full difficulty spectrum observed in real publications. To enable faithful semantic evaluation beyond SMILES and MolFile, we propose CARBON, a representation language capable of expressing valence variations, icon-based groups, and other non-standard chemical semantics. We further adopt a dual-track evaluation protocol supporting both CARBON and SMILES outputs for broad model compatibility. Comprehensive experiments over 18 OCSR-capable models reveal severe performance degradation on MolRecBench-Wild, exposing a large gap between previous patent benchmarks and real-world academic scenarios.

Selective knowledge erasure from LLMs is critical for GDPR compliance and model safety, yet current unlearning methods conflate behavioral suppression with true knowledge removal, allowing latent capabilities to persist beneath surface-level refusals. In this work, we address this challenge by introducing Knowledge Immunization Framework (KIF), a representation-aware architecture that distinguishes genuine erasure from obfuscation by targeting internal activation signatures rather than surface outputs. Our approach combines dynamic suppression of subject-specific representations with parameter-efficient adaptation, enabling durable unlearning without full model retraining. KIF achieves near-oracle erasure (FQ approx 0.99 vs. 1.00) while preserving utility at oracle levels (MU = 0.62), effectively breaking the stability-erasure tradeoff that has constrained all prior work. We evaluate both standard foundation models (Llama and Mistral) and reasoning-prior models (Qwen and DeepSeek) across 3B to 14B parameters. Our observation shows that standard models exhibit scale-independent true erasure (<3% utility drift), while reasoning-prior models reveal fundamental architectural divergence. Our comprehensive dual-metric evaluation protocol, combining surface-level leakage with latent trace persistence, operationalizes the obfuscation - erasure distinction and enables the first systematic diagnosis of mechanism-level forgetting behavior across model families and scales.

Large language models (LLMs) have shown strong performance on isolated psychiatric tasks, including dialogue, diagnosis, and treatment planning, yet existing benchmarks rarely simulate complete psychiatric clinical encounters. We introduce MentalHospital, a virtual evaluation environment for LLM-based psychiatric clinical encounters. MentalHospital instantiates the Subjective Interviewing, Objective Examination, Diagnostic Assessment, and Treatment Planning (S.O.A.P.) workflow, using skill-augmented standardized patients constructed from 1,193 de-identified psychiatric electronic health record (EHR) cases spanning all major ICD-11 categories and 76 disorders. Each encounter is assessed through a dual-track protocol that combines objective comparison against EHR-derived references with subjective assessment of clinical process quality. To scale specialist judgment, we develop MentalEval, five domain-specific evaluators covering communication empathy, interviewing professionalism, clinical-note quality, diagnostic rigor, and treatment appropriateness, trained with rubric-grounded SFT and expert-guided DPO. Survey responses from 22 clinicians support MentalHospital's clinical fidelity (3.88/5), while MentalEval achieves strong expert alignment with an average QWK of 0.944. Benchmarking shows that even the strongest LLM trails clinicians by 37.28 percentage points in objective psychiatric competence, with mental status assessment as a key bottleneck.

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

Recent text-to-video (T2V) technology advancements, as demonstrated by models such as Gen2, Pika, and Sora, have significantly broadened its applicability and popularity. Despite these strides, evaluating these models poses substantial challenges. Primarily, due to the limitations inherent in automatic metrics, manual evaluation is often considered a superior method for assessing T2V generation. However, existing manual evaluation protocols face reproducibility, reliability, and practicality issues. To address these challenges, this paper introduces the Text-to-Video Human Evaluation (T2VHE) protocol, a comprehensive and standardized protocol for T2V models. The T2VHE protocol includes well-defined metrics, thorough annotator training, and an effective dynamic evaluation module. Experimental results demonstrate that this protocol not only ensures high-quality annotations but can also reduce evaluation costs by nearly 50%. We will open-source the entire setup of the T2VHE protocol, including the complete protocol workflow, the dynamic evaluation component details, and the annotation interface code. This will help communities establish more sophisticated human assessment protocols.

As LLM-based agents are deployed in complex, multi-step workflows, a critical evaluation gap has emerged: most existing benchmarks judge only final outcomes, unable to distinguish reliable reasoning from lucky success or attribute failures to specific process deficiencies, hindering attribution in long-horizon tasks. In this work, we present ClawTrack, a dual-assessment benchmark that simultaneously measures what an agent achieves (Task Score) and how it achieves it (Process Score). ClawTrack comprises 320 tasks across 8 domains with 25+ deterministic mock services. A Process Grader scores each reasoning turn along four dimensions (goal alignment, efficiency, information utilization, and result verification), anchored by 12,541 task-specific rubric items. Evaluating 21 models over 16,000+ trials, we find that: (1) process scores effectively attribute success and failure to specific reasoning dimensions, filtering lucky passes invisible to outcome-only evaluation; (2) the four dimensions are complementary, with result verification as the systematic bottleneck; (3) the framework is robust to evaluator choice across different judge LLMs; and (4) process-based trajectory filtering yields consistent post-training improvements across model scales.

We investigate the application of ByteTrack in the realm of multiple object tracking. ByteTrack, a simple tracking algorithm, enables the simultaneous tracking of multiple objects by strategically incorporating detections with a low confidence threshold. Conventionally, objects are initially associated with high confidence threshold detections. When the association between objects and detections becomes ambiguous, ByteTrack extends the association to lower confidence threshold detections. One notable drawback of the existing ByteTrack approach is its reliance on a fixed threshold to differentiate between high and low-confidence detections. In response to this limitation, we introduce a novel and adaptive approach. Our proposed method entails a dynamic adjustment of the confidence threshold, leveraging insights derived from overall detections. Through experimentation, we demonstrate the effectiveness of our adaptive confidence threshold technique while maintaining running time compared to ByteTrack.

The reliance of mobile GUI agents on Multimodal Large Language Models (MLLMs) introduces a severe privacy vulnerability: screenshots containing Personally Identifiable Information (PII) are often sent to untrusted, third-party routers. These routers can exploit their own MLLMs to mine this data, violating user privacy. Existing privacy perturbations fail the critical dual challenge of this scenario: protecting PII from the router's MLLM while simultaneously preserving task utility for the agent's MLLM. To address this gap, we propose the Dual-Task Adversarial Protector (DualTAP), a novel framework that, for the first time, explicitly decouples these conflicting objectives. DualTAP trains a lightweight generator using two key innovations: (i) a contrastive attention module that precisely identifies and targets only the PII-sensitive regions, and (ii) a dual-task adversarial objective that simultaneously minimizes a task-preservation loss (to maintain agent utility) and a privacy-interference loss (to suppress PII leakage). To facilitate this study, we introduce PrivScreen, a new dataset of annotated mobile screenshots designed specifically for this dual-task evaluation. Comprehensive experiments on six diverse MLLMs (e.g., GPT-5) demonstrate DualTAP's state-of-the-art protection. It reduces the average privacy leakage rate by 31.6 percentage points (a 3.0x relative improvement) while, critically, maintaining an 80.8% task success rate - a negligible drop from the 83.6% unprotected baseline. DualTAP presents the first viable solution to the privacy-utility trade-off in mobile MLLM agents.

AI research agents combine prior knowledge, public sources, and experimental feedback to produce useful results. The Discovery Certification Protocol (DCP) turns claims about these results into executable recovery and feedback tests. Gate 1 validates useful improvement on sealed evaluation. Gate 2 gives matched agents the registered starting information and observed Web content while withholding the target research history. Every valid method reaching the numerical target supplies a recovery witness and triggers the Core veto. DCP Core requires adequate controls, zero observed recoveries, and a finite-sample bound on recovery in one fresh registered episode. Optional Gate 3 measures the average effect of truthful feedback relative to a specified neutral policy from a shared checkpoint. DCP Evidence adds this effect after independent null calibration and a registered effect margin. Two controlled audits exercise the complete protocol in SQLite optimization and virtual catalyst control under different models. Each produced zero recoveries in 96 episodes, with an upper bound of 0.0468. Each paired study yielded 30 truthful recoveries and zero neutral recoveries, with passing 60-pair null studies. Additional cases exercise Core, recovered, and audit-incomplete decisions. A deterministic, LLM-free verifier reproduces the decisions from frozen evidence. DCP provides a common evidence language for useful outcomes, alternative routes, and feedback effects across AI research.

LeWorldModel trains a latent world model with a prediction loss and a single anti-collapse regulariser, and reports approximately 87% of goals reached on TwoRoom, its simplest diagnostic environment. We reproduce that result by independent reimplementation on roughly $25 of rented compute, with all evaluation on one laptop CPU. We reach 94.0% at the repository's evaluation goal offset, against 84.0% for the authors' own released checkpoint measured under our protocol on identical episodes, and we reproduce the reported representation result directly (position probe Pearson r = 0.9988 against a reported 0.996). Reaching that point required four conventions that determine the outcome and appear in no released configuration file: dense action gathering across a frameskip block, a programmatically-set action-encoder width, ImageNet pixel normalisation, and action z-scoring. A reproducer following the released configurations alone obtains a model whose predictor cannot converge. The evaluation protocol is itself contested by the released material. The paper's appendix and the repository's configuration specify different goal offsets and step budgets; on the authors' own weights these yield 14.0% and 84.0%, and only the configuration's values reproduce the reported figure. On fifty identical episodes, changing nothing but how the goal is constructed moves that checkpoint from 84.0% to 8.0%. Two findings generalise. One-step prediction accuracy does not predict long-horizon planning success: across three checkpoints spanning a sevenfold range in prediction error, including the authors' own, it orders short-horizon success monotonically and fails to order long-horizon success at all. And a batch normalisation layer inflated our reported validation loss by up to a factor of 300, concealing a training loss that was flat throughout.

How would you fairly evaluate two multi-object tracking algorithms (i.e. trackers), each one employing a different object detector? Detectors keep improving, thus trackers can make less effort to estimate object states over time. Is it then fair to compare a new tracker employing a new detector with another tracker using an old detector? In this paper, we propose a novel performance measure, named Tracking Effort Measure (TEM), to evaluate trackers that use different detectors. TEM estimates the improvement that the tracker does with respect to its input data (i.e. detections) at frame level (intra-frame complexity) and sequence level (inter-frame complexity). We evaluate TEM over well-known datasets, four trackers and eight detection sets. Results show that, unlike conventional tracking evaluation measures, TEM can quantify the effort done by the tracker with a reduced correlation on the input detections. Its implementation is publicly available online at https://github.com/vpulab/MOT-evaluation.

Agents are deployed in applications from trip planners and stock trading to IT incident triage. In most cases, LLM agents work autonomously with minimal rule-based safeguarding, leading to cost and safety issues from irreversible actions. Recent works resolve this either by using a safeguard agent to monitor behavior or evaluating logs post-hoc. The first adds cost and latency to every step; the second delivers its verdict after the run, when tokens are burned and damage is done. To overcome this, we propose OnTrack, a streaming monitoring mechanism that compares an agent's steps and dependencies against recorded successful runs to alert users or block the agent in about a millisecond per step. We study this problem in three regimes of decreasing access: full reference access (historical runs and tool schemas), intermediate access (only tool schemas), and no prior knowledge (only step logs as generated). Expectation of OnTrack's monitoring capabilities reduces as data access drops, ranging from plan violation detection to identifying loops, stalls, and repeated tool calls. Finally, we evaluate OnTrack using SWE-bench trajectories. Based on the first 8 steps, our method ranks failing trajectories below succeeding ones better than content similarity approaches (+0.057 AUROC). With an abort policy, we save about 18% of compute that would be burned on failing runs, where 83% of interrupted runs were actually heading to failure (5 out of 6 aborts were correct).

Benchmark contamination is diagnosed with n-gram overlap, likelihood-based membership inference, or canary strings, and each needs something usually unavailable: the training corpus, a well-chosen test statistic, or foresight at release. A recent alternative reads it off a linear probe on internal activations. We show the natural way to do this does not work, specify one that survives measurement, then find that the correction making it work carries more variance than the null it is tested against. The protocol reports a zero-sum contrast on the depth profile of probe accuracy, recentred on a level-matched placebo baseline, tested against a label-permutation null, with the reference set twice the size of the suspect set. Each choice replaces a simpler alternative we rejected on measurement. Reporting the level of excess separability rather than its shape makes the false positive rate track the size of the analyst's own control set, 0.03 to 0.99 under a true null. Contrasting against a flat depth profile rejects a true null 0.72 of the time when surface decodability rises with depth, and loses all power when it falls. On real transformers the protocol fails a test the simulations did not pose. The recentring subtracts an estimate, and the permutation null holds it fixed. Re-estimated across split seeds on four audits of contaminated checkpoints, its standard deviation is 1.30 to 1.56 times the null's own in every arm: what is subtracted to remove a bias is more variable than what it corrects. The one nominally significant result, p = 0.0075, becomes 0.0745 once that variance is propagated, and no verdict is issued. The simulations missed this because their surface key is the covariate driving item variation; on real text it is a proxy, and degrading key quality in simulation reproduces it. We add a companion measurement and a widened null. No arm shows contamination.

Re-rendering an existing video from a novel camera viewpoint requires the output to follow the prescribed camera trajectory while preserving the appearance and dynamics of the original scene across every frame. Existing methods rely on per-frame pose embeddings, noisy point-cloud renderings, or implicit learned correspondences, none of which provides an explicit, temporally continuous link between source and target pixels. We propose Track2View, which conditions a video diffusion transformer on paired 3D point tracks: sparse trajectories of scene points projected into both the source and target camera views. These tracks provide explicit spatiotemporal correspondences that are temporally continuous by construction, encoding what content should appear where and when. At the core of Track2View is a dual-view track conditioner that transfers visual context from source to target view through parameter-free geometric operations and learned temporal aggregation, ensuring generalization to arbitrary camera trajectories without memorizing specific motions. We further introduce a data curation pipeline that extracts one-to-one track correspondences by running a 3D point tracker on temporally concatenated multi-camera view pairs. On a 400-video benchmark spanning static and dynamic scenes, Track2View achieves state-of-the-art results across visual quality, view synchronization, and camera accuracy, reducing rotation error by 30-65% and translation error by 61-72% relative to leading baselines. Project page is available at this https URL: https://qjizhi.github.io/track2view

Traffic anomaly detection has played a crucial role in Intelligent Transportation System (ITS). The main challenges of this task lie in the highly diversified anomaly scenes and variational lighting conditions. Although much work has managed to identify the anomaly in homogenous weather and scene, few resolved to cope with complex ones. In this paper, we proposed a dual-modality modularized methodology for the robust detection of abnormal vehicles. We introduced an integrated anomaly detection framework comprising the following modules: background modeling, vehicle tracking with detection, mask construction, Region of Interest (ROI) backtracking, and dual-modality tracing. Concretely, we employed background modeling to filter the motion information and left the static information for later vehicle detection. For the vehicle detection and tracking module, we adopted YOLOv5 and multi-scale tracking to localize the anomalies. Besides, we utilized the frame difference and tracking results to identify the road and obtain the mask. In addition, we introduced multiple similarity estimation metrics to refine the anomaly period via backtracking. Finally, we proposed a dual-modality bilateral tracing module to refine the time further. The experiments conducted on the Track 4 testset of the NVIDIA 2021 AI City Challenge yielded a result of 0.9302 F1-Score and 3.4039 root mean square error (RMSE), indicating the effectiveness of our framework.

Recent machine learning papers often report 1-2 percentage point improvements from a single run on a benchmark. These gains are highly sensitive to random seeds, data ordering, and implementation details, yet are rarely accompanied by uncertainty estimates or significance tests. It is therefore unclear when a reported +1-2% reflects a real algorithmic advance versus noise. We revisit this problem under realistic compute budgets, where only a few runs are affordable. We propose a simple, PC-friendly evaluation protocol based on paired multi-seed runs, bias-corrected and accelerated (BCa) bootstrap confidence intervals, and a sign-flip permutation test on per-seed deltas. The protocol is intentionally conservative and is meant as a guardrail against over-claiming. We instantiate it on CIFAR-10, CIFAR-10N, and AG News using synthetic no-improvement, small-gain, and medium-gain scenarios. Single runs and unpaired t-tests often suggest significant gains for 0.6-2.0 point improvements, especially on text. With only three seeds, our paired protocol never declares significance in these settings. We argue that such conservative evaluation is a safer default for small gains under tight budgets.

AI pentesting agents are increasingly credible as offensive security systems, but current benchmarks still provide limited guidance on which will perform best in real-world targets. Existing evaluation protocols assess and optimize for predefined goals such as capture-the-flag, remote code execution, exploit reproduction, or trajectory similarity, in simplified or narrow settings. These tools are valuable for measuring bounded capabilities, yet they do not adequately capture the complexity, open-ended exploration, and strategic decision-making required in realistic pentesting. In this paper, we present a practical evaluation protocol that shifts assessment from task completion to validated vulnerability discovery, allowing evaluation in sufficiently complex targets spanning multiple attack surfaces and vulnerability classes. The protocol combines structured ground-truth with LLM-based semantic matching to identify vulnerabilities, bipartite resolution to score findings under realistic ambiguity, continuous ground-truth maintenance, repeated and cumulative evaluation of stochastic agents, efficiency metrics, and reduced-suite selection for sustainable experimentation. This protocol extends the state of the art by enabling a more realistic, operationally informative comparison of AI pentesting agents. To enable reproducibility, we also release expert-annotated ground truth and code for the proposed evaluation protocol: https://github.com/ethiack/ethibench.

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

Despite great recent advances in visual tracking, its further development, including both algorithm design and evaluation, is limited due to lack of dedicated large-scale benchmarks. To address this problem, we present LaSOT, a high-quality Large-scale Single Object Tracking benchmark. LaSOT contains a diverse selection of 85 object classes, and offers 1,550 totaling more than 3.87 million frames. Each video frame is carefully and manually annotated with a bounding box. This makes LaSOT, to our knowledge, the largest densely annotated tracking benchmark. Our goal in releasing LaSOT is to provide a dedicated high quality platform for both training and evaluation of trackers. The average video length of LaSOT is around 2,500 frames, where each video contains various challenge factors that exist in real world video footage,such as the targets disappearing and re-appearing. These longer video lengths allow for the assessment of long-term trackers. To take advantage of the close connection between visual appearance and natural language, we provide language specification for each video in LaSOT. We believe such additions will allow for future research to use linguistic features to improve tracking. Two protocols, full-overlap and one-shot, are designated for flexible assessment of trackers. We extensively evaluate 48 baseline trackers on LaSOT with in-depth analysis, and results reveal that there still exists significant room for improvement. The complete benchmark, tracking results as well as analysis are available at http://vision.cs.stonybrook.edu/~lasot/.

Streaming video understanding requires models to interpret evidence as it arrives, yet current evaluations often report task scores without specifying when evidence becomes valid, how visual history is maintained, or how responses are triggered. As a result, similar scores may correspond to different workloads, failure modes, and operational behavior. We introduce TRACE (Temporal Audit and Condition-aware Evaluation), a condition-aware benchmark and evaluation framework that makes these factors explicit. TRACE combines temporally audited visual tasks with evidence timing and instruction-dependent trigger annotations, a unified causal Core--Adapter protocol that controls information availability while recording actual history processing and response events, and multidimensional reporting of answer quality, timeliness, response-selection behavior, workload, completion, and reliability. On 1,240 records from 517 videos, we evaluate eight publicly available models or systems in eight configurations. We find that nearly identical QA accuracy can mask substantial differences in completion, answer validity, and generation workload, while proactive performance separates into response quality, response delay, false alarms (responses emitted while no target window is currently valid and a later one remains), and missed target windows. These results show that streaming-video performance should be interpreted as execution-conditioned system behavior rather than a single score. Our benchmark and code can be accessed at https://github.com/om-ai-lab/trace-bench{https://github.com/om-ai-lab/trace-bench}.

Comprehensive and constructive evaluation protocols play an important role in the development of sophisticated text-to-video (T2V) generation models. Existing evaluation protocols primarily focus on temporal consistency and content continuity, yet largely ignore the dynamics of video content. Dynamics are an essential dimension for measuring the visual vividness and the honesty of video content to text prompts. In this study, we propose an effective evaluation protocol, termed DEVIL, which centers on the dynamics dimension to evaluate T2V models. For this purpose, we establish a new benchmark comprising text prompts that fully reflect multiple dynamics grades, and define a set of dynamics scores corresponding to various temporal granularities to comprehensively evaluate the dynamics of each generated video. Based on the new benchmark and the dynamics scores, we assess T2V models with the design of three metrics: dynamics range, dynamics controllability, and dynamics-based quality. Experiments show that DEVIL achieves a Pearson correlation exceeding 90% with human ratings, demonstrating its potential to advance T2V generation models. Code is available at https://github.com/MingXiangL/DEVIL.

Agent benchmarks increasingly evaluate repository editing, web research, terminal use, and long-horizon interaction. Their scores support capability claims only when the evaluation protocol keeps the intended capability necessary for success. Recent reward-hacking benchmarks and system reports show that agents can instead recover public solutions, read evaluation artifacts, infer generator structure, manipulate feedback, or benefit from invalid scoring paths; existing responses do not provide a common procedure for attributing these shortcuts and quantifying their effect across benchmarks. We formulate protocol validity and introduce HackDetect, a post-hoc audit that identifies an exposure, determines how the agent used it, and assesses whether the resulting score is misleading. We quantify score inflation with the Mislead gap, defined as the exploit score minus the intended score. We audit 2,385 traces across 15 agent benchmarks and find evidence of exposures and reward hacking in 67.0% of Frontier Science traces and 66.7% of AutoLab tasks. Across paired comparisons, we measure score inflation of 0.45-1.00, showing that benchmark reports should provide evidence that scores reflect the intended capability.

Agent benchmarks are increasingly used to compare large language models (LLMs) and guide deployment decisions, yet benchmark scores are meaningful only if they measure model capability rather than properties of the evaluation pipeline. We identify a double measurement confound: execution-critical decisions are performed by a fixed scaffold instead of the model, while the scorer evaluates outputs using criteria that may not reflect task correctness. We unify these issues within a measurement-theoretic framework that characterizes when benchmark scores can be interpreted as evidence of model capability, and instantiate it with an audit-and-repair protocol that (i) transfers execution-critical decisions from the scaffold to the model, (ii) replaces shape-based evaluation with seeded ground-truth scoring, and (iii) reports reliability beyond the mean through worst-case and tail-risk metrics. Experiments on ComtradeBench show that the joint intervention transforms a nearly flat leaderboard into a reliability spectrum that distinguishes both average performance and robustness across seeds. Applying the audit to existing benchmarks further shows that scorer validity is benchmark-specific, whereas scaffold ownership is an uncontrolled axis wherever we probed it. Our results suggest that benchmark scores should be interpreted together with their scaffolding level, scoring criterion, and reliability profile, providing a practical framework for more valid evaluation of LLM agents.

Multi-object tracking in sports broadcasts is hard for two reasons: athletes move fast and change direction abruptly, and the camera pans and zooms at the same time. We present ARC-SORT (Adaptive, Robust, Compensated SORT), a tracking-by-detection tracker that trains nothing and runs in real time. Its design rule: the tracker measures its own uncertainty, and that measurement decides how it acts. An Interacting Multiple Model (IMM) Kalman filter gives every track a predictive covariance, and we turn it into a per-track search region: the less reliable the motion prediction, the larger the region in which a detection may match (uncertainty-adaptive Expansion IoU, EIoU). On identical inputs, this continuous coupling is what matters: the same covariance used as a hard Mahalanobis gate hurts at every threshold we swept, and richer covariance summaries add nothing measurable. When two tracks cross, the assignment cost itself reports whether the decision was close. ARC-SORT keeps such pairs under review and applies at most one corrective identity swap once post-crossing appearance evidence shows the first decision was wrong; this adds +1.0 HOTA (higher order tracking accuracy) on both the SportsMOT and DanceTrack test sets at under a millisecond per frame. For players who leave the frame for seconds, motion extrapolation carries no information, so our offline re-identification drops the usual motion gate and links tracklets by appearance alone. ARC-SORT reaches 81.82 HOTA on the SportsMOT test set with the fewest identity switches of the 69 leaderboard entries as of August 2026 and tied-third multi-object tracking accuracy (MOTA, 97.09), and 62.1 HOTA on DanceTrack test with the same association settings. On Deep-EIoU's public detections and embeddings it reaches 80.31 HOTA, +3.1 over Deep-EIoU. It runs at about 50 frames per second on a laptop RTX 3060 GPU. Code will be released.

With the rapid advancement of Deep Research Agents in knowledge-intensive domains such as finance, establishing reliable and domain-aligned evaluation standards remains a critical challenge. Existing benchmarks focus on either closed-ended question answering or open-ended report evaluation, failing to jointly capture retrieval-reasoning accuracy and end-to-end research quality required in real-world workflows. We introduce ICBCBench, a consortium-driven benchmark for financial deep research, developed in collaboration with domain experts from a broad range of financial institutions and academia, involving over 50 experts across more than 40 organizations. It adopts a dual-track paradigm integrating objective tasks with verifiable answers and subjective long-form report evaluation, enabling complementary assessment of retrieval-reasoning accuracy and end-to-end report quality in terms of expert alignment, citation consistency, and source quality. Experiments on state-of-the-art DRAs and large language models reveal substantial gaps in complex reasoning, factual grounding, and report quality, highlighting the challenges of achieving industry-level performance. Our dataset and evaluation framework are available at https://github.com/DeepFin-Intelligence/ICBCBench.

Current scene flow methods broadly fail to describe motion on small objects, and current scene flow evaluation protocols hide this failure by averaging over many points, with most drawn larger objects. To fix this evaluation failure, we propose a new evaluation protocol, Bucket Normalized EPE, which is class-aware and speed-normalized, enabling contextualized error comparisons between object types that move at vastly different speeds. To highlight current method failures, we propose a frustratingly simple supervised scene flow baseline, TrackFlow, built by bolting a high-quality pretrained detector (trained using many class rebalancing techniques) onto a simple tracker, that produces state-of-the-art performance on current standard evaluations and large improvements over prior art on our new evaluation. Our results make it clear that all scene flow evaluations must be class and speed aware, and supervised scene flow methods must address point class imbalances. We release the evaluation code publicly at https://github.com/kylevedder/BucketedSceneFlowEval.

As Large Language Models (LLMs) are increasingly integrated into healthcare to address complex inquiries, ensuring their reliability remains a critical challenge. Recent studies have highlighted that generic LLMs often struggle in clinical contexts, occasionally producing misleading guidance. To mitigate these risks, this research focuses on the domain-specific adaptation of Llama-2-7B using the Low-Rank Adaptation (LoRA) technique. By injecting trainable low-rank matrices into the Transformer layers, we efficiently adapted the model using authentic patient-physician transcripts while preserving the foundational knowledge of the base model. Our objective was to enhance precision and contextual relevance in responding to medical queries by capturing the specialized nuances of clinical discourse. Due to the resource-intensive nature of large-scale human validation, the model's performance was evaluated through a dual-track framework: Track A utilized traditional lexical similarity metrics (e.g., BLEU, ROUGE), while Track B employed an "LLM-as-a-Judge" paradigm using GPT-4 for semantic assessment. Our results demonstrate that while the LoRA-enhanced model achieved significant improvements across all quantitative lexical dimensions, a profound disagreement surfaced in the GPT-4 evaluation, which marginally favored the baseline model's conversational flow. This metric divergence underscores a pivotal finding: traditional automated scores may not fully reflect clinical utility. Consequently, we propose that while automated metrics and LLM judges serve as valuable developmental proxies, rigorous validation by human medical experts remains an indispensable requirement for the safe deployment of LLMs in healthcare settings.

The performance of multi-turn, agentic LLM inference is increasingly dominated by KV-Cache storage I/O rather than computation. In prevalent disaggregated architectures, loading the massive KV-Cache from external storage creates a fundamental imbalance: storage NICs on prefill engines become bandwidth-saturated, while those on decoding engines remain idle. This asymmetry severely constrains overall system throughput. We present DualPath, an inference system that breaks this bottleneck by introducing dual-path KV-Cache loading. Beyond the traditional storage-to-prefill path, DualPath enables a novel storage-to-decode path, in which the KV-Cache is loaded into decoding engines and then efficiently transferred to prefill engines via RDMA over the compute network. DualPath combines this optimized data path -- which inherently avoids network congestion and avoids interference with latency-critical model execution communications -- with a global scheduler that dynamically balances load across prefill and decode engines. Our evaluation on three models with production agentic workloads demonstrates that DualPath improves offline inference throughput by up to 1.87times on our in-house inference system. It can also improve online serving throughput by an average factor of 1.96times without violating SLO.

The integration of visual understanding and generation into unified multimodal models represents a significant stride toward general-purpose AI. However, a fundamental question remains unanswered by existing benchmarks: does this architectural unification actually enable synergetic interaction between the constituent capabilities? Existing evaluation paradigms, which primarily assess understanding and generation in isolation, are insufficient for determining whether a unified model can leverage its understanding to enhance its generation, or use generative simulation to facilitate deeper comprehension. To address this critical gap, we introduce RealUnify, a benchmark specifically designed to evaluate bidirectional capability synergy. RealUnify comprises 1,000 meticulously human-annotated instances spanning 10 categories and 32 subtasks. It is structured around two core axes: 1) Understanding Enhances Generation, which requires reasoning (e.g., commonsense, logic) to guide image generation, and 2) Generation Enhances Understanding, which necessitates mental simulation or reconstruction (e.g., of transformed or disordered visual inputs) to solve reasoning tasks. A key contribution is our dual-evaluation protocol, which combines direct end-to-end assessment with a diagnostic stepwise evaluation that decomposes tasks into distinct understanding and generation phases. This protocol allows us to precisely discern whether performance bottlenecks stem from deficiencies in core abilities or from a failure to integrate them. Through large-scale evaluations of 12 leading unified models and 6 specialized baselines, we find that current unified models still struggle to achieve effective synergy, indicating that architectural unification alone is insufficient. These results highlight the need for new training strategies and inductive biases to fully unlock the potential of unified modeling.

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

This paper addresses limitations in 3D tracking-by-detection methods, particularly in identifying legitimate trajectories and reducing state estimation drift in Kalman filters. Existing methods often use threshold-based filtering for detection scores, which can fail for distant and occluded objects, leading to false positives. To tackle this, we propose a novel track validity mechanism and multi-stage observational gating process, significantly reducing ghost tracks and enhancing tracking performance. Our method achieves a 29.47% improvement in Multi-Object Tracking Accuracy (MOTA) on the KITTI validation dataset with the Second detector. Additionally, a refined Kalman filter term reduces localization noise, improving higher-order tracking accuracy (HOTA) by 4.8%. The online framework, RobMOT, outperforms state-of-the-art methods across multiple detectors, with HOTA improvements of up to 3.92% on the KITTI testing dataset and 8.7% on the validation dataset, while achieving low identity switch scores. RobMOT excels in challenging scenarios, tracking distant objects and prolonged occlusions, with a 1.77% MOTA improvement on the Waymo Open dataset, and operates at a remarkable 3221 FPS on a single CPU, proving its efficiency for real-time multi-object tracking.

On-policy (self) distillation (OPSD) is increasingly adopted for language-model post-training. It strengthens the teacher with privileged information but can induce a privilege illusion: the student learns privilege-dependent behavior it cannot reproduce from its inference-time context, yet behaves as if the training-time privileged information remained available, ultimately degrading performance. In this paper, we identify information asymmetry between the privileged teacher and the student at inference as the root cause of this failure in OPSD. To resolve this asymmetry, we propose Dual-Anchored Policy Distillation (DAPD), a unified framework with two levels of anchoring. Dual-Path Anchoring (DPA) introduces a self-conditioned bridge and aligns reference and rollout behavior along two matched-information paths, preventing privilege-dependent behavior from being transferred to the inference-time student. Dual-Source Anchoring (DSA) applies these paths in both reference-to-rollout and rollout-to-reference directions, reducing reliance on privileged reference guidance while preserving correctness supervision. Extensive experiments show that DAPD significantly alleviates privilege illusion, outperforming OPSD on Qwen3-4B by +2.00 points on average across tasks. Notably, its gains persist across scales, reaching +2.69 at 4B and +2.78 at 32B.

Memory-Mediated Learning Architecture (MMLA) separates slow base parameters theta, a bounded numerical policy carrier Phi, and a bounded authoritative memory M. Predictive Dual-State Adaptation (PDSA) lets feedback update Phi while one problem remains active and lets a trusted lifecycle atomically commit one typed row or exact NULL. Later reasoning may read both states, but their writers, resets, rollback domains, and ledgers remain distinct. Realized futures supervise values only during training; deployment is causal and future-blind. We give conditional theory and falsifiable contracts for reasoning-time updates, completed-segment consolidation, predictive admission, authoritative memory, and dual-state attribution. Assumptions, counterexamples, capacity and cost ledgers, recovery duties, and identifying experiments are explicit; these are not implementation guarantees. Controlled studies show exact lifecycle execution on 300/300 held-out records for each of three seeds, calibrated retrieval gains over frozen-hidden dense and BM25 baselines, and exact typed anchor-filler transport on 240/240 held-out records per seed. These validate restricted components, not natural-language memory management or complete PDSA. Post-training studies retain positive and negative evidence. Later protocols obtain restricted readout progress, but a nine-trajectory comparison finds that matched latent readout, a full-width bridge, and bridge plus frozen text-teacher alignment all fail continuous-event qualification across both task families. A separately adapted text reference and restoration checks pass. At the September 13, 2026 evidence cutoff, no strict policy-only reasoning-time-training effect, predictive-admission oracle margin, learned future-blind admission policy, or policy-by-memory factorial advantage is established.

Recent advances in AI-assisted programming have empowered agents to execute complex workflows via command-line interfaces, however, existing benchmarks are limited by short task horizons, data contamination from GitHub scraping, and a lack of fine-grained evaluation metrics, fail to rigorously evaluate the long-horizon planning and execution capabilities essential for realistic software engineering. To address these gaps, we introduce LongCLI-Bench, a comprehensive benchmark designed to evaluate agentic capabilities across long-horizon, realistic tasks. We curated 20 high-quality, long-horizon tasks from over 1,000 computer science assignments and real-world workflows, covering four engineering categories: from scratch, feature addition, bug fixing, and refactoring. We propose a dual-set testing protocol for LongCLI-Bench, which measures requirement fulfillment (fail-to-pass) and regression avoidance (pass-to-pass), and incorporates step-level scoring to pinpoint execution failures. Extensive experiments reveal that even state-of-the-art agents achieve pass rates below 20% in LongCLI-Bench. Step-level analysis further indicates that the majority of tasks stall at less than 30% completion, highlighting that critical failures often occur in the early stages. Although self-correction offers marginal gains, human-agent collaboration through plan injection and interactive guidance yields significantly higher improvements. These results highlight that future research must emphasize the development of synergistic human-agent workflows alongside advances in agents' planning and execution capabilities to overcome key challenges in long-horizon task performance.

Existing end-to-end Multi-Object Tracking (e2e-MOT) methods have not surpassed non-end-to-end tracking-by-detection methods. One potential reason is its label assignment strategy during training that consistently binds the tracked objects with tracking queries and then assigns the few newborns to detection queries. With one-to-one bipartite matching, such an assignment will yield unbalanced training, i.e., scarce positive samples for detection queries, especially for an enclosed scene, as the majority of the newborns come on stage at the beginning of videos. Thus, e2e-MOT will be easier to yield a tracking terminal without renewal or re-initialization, compared to other tracking-by-detection methods. To alleviate this problem, we present Co-MOT, a simple and effective method to facilitate e2e-MOT by a novel coopetition label assignment with a shadow concept. Specifically, we add tracked objects to the matching targets for detection queries when performing the label assignment for training the intermediate decoders. For query initialization, we expand each query by a set of shadow counterparts with limited disturbance to itself. With extensive ablations, Co-MOT achieves superior performance without extra costs, e.g., 69.4% HOTA on DanceTrack and 52.8% TETA on BDD100K. Impressively, Co-MOT only requires 38\% FLOPs of MOTRv2 to attain a similar performance, resulting in the 1.4times faster inference speed.

This paper introduces MCTrack, a new 3D multi-object tracking method that achieves state-of-the-art (SOTA) performance across KITTI, nuScenes, and Waymo datasets. Addressing the gap in existing tracking paradigms, which often perform well on specific datasets but lack generalizability, MCTrack offers a unified solution. Additionally, we have standardized the format of perceptual results across various datasets, termed BaseVersion, facilitating researchers in the field of multi-object tracking (MOT) to concentrate on the core algorithmic development without the undue burden of data preprocessing. Finally, recognizing the limitations of current evaluation metrics, we propose a novel set that assesses motion information output, such as velocity and acceleration, crucial for downstream tasks. The source codes of the proposed method are available at this link: https://github.com/megvii-research/MCTrack}{https://github.com/megvii-research/MCTrack

Current approaches in Multiple Object Tracking (MOT) rely on the spatio-temporal coherence between detections combined with object appearance to match objects from consecutive frames. In this work, we explore MOT using object appearances as the main source of association between objects in a video, using spatial and temporal priors as weighting factors. We form initial tracklets by leveraging on the idea that instances of an object that are close in time should be similar in appearance, and build the final object tracks by fusing the tracklets in a hierarchical fashion. We conduct extensive experiments that show the effectiveness of our method over three different MOT benchmarks, MOT17, MOT20, and DanceTrack, being competitive in MOT17 and MOT20 and establishing state-of-the-art results in DanceTrack.

Pulmonary embolism (PE) is a life-threatening condition where rapid and accurate diagnosis is imperative yet difficult due to predominantly atypical symptomatology. Computed tomography pulmonary angiography (CTPA) is acknowledged as the gold standard imaging tool in clinics, yet it can be contraindicated for emergency department (ED) patients and represents an onerous procedure, thus necessitating PE identification through non-contrast CT (NCT) scans. In this work, we explore the feasibility of applying a deep-learning approach to NCT scans for PE identification. We propose a novel Cross-Phase Mutual learNing framework (CPMN) that fosters knowledge transfer from CTPA to NCT, while concurrently conducting embolism segmentation and abnormality classification in a multi-task manner. The proposed CPMN leverages the Inter-Feature Alignment (IFA) strategy that enhances spatial contiguity and mutual learning between the dual-pathway network, while the Intra-Feature Discrepancy (IFD) strategy can facilitate precise segmentation of PE against complex backgrounds for single-pathway networks. For a comprehensive assessment of the proposed approach, a large-scale dual-phase dataset containing 334 PE patients and 1,105 normal subjects has been established. Experimental results demonstrate that CPMN achieves the leading identification performance, which is 95.4\% and 99.6\% in patient-level sensitivity and specificity on NCT scans, indicating the potential of our approach as an economical, accessible, and precise tool for PE identification in clinical practice.

Multi-object tracking (MOT) aims at estimating bounding boxes and identities of objects in videos. Most methods obtain identities by associating detection boxes whose scores are higher than a threshold. The objects with low detection scores, e.g. occluded objects, are simply thrown away, which brings non-negligible true object missing and fragmented trajectories. To solve this problem, we present a simple, effective and generic association method, tracking by associating almost every detection box instead of only the high score ones. For the low score detection boxes, we utilize their similarities with tracklets to recover true objects and filter out the background detections. When applied to 9 different state-of-the-art trackers, our method achieves consistent improvement on IDF1 score ranging from 1 to 10 points. To put forwards the state-of-the-art performance of MOT, we design a simple and strong tracker, named ByteTrack. For the first time, we achieve 80.3 MOTA, 77.3 IDF1 and 63.1 HOTA on the test set of MOT17 with 30 FPS running speed on a single V100 GPU. ByteTrack also achieves state-of-the-art performance on MOT20, HiEve and BDD100K tracking benchmarks. The source code, pre-trained models with deploy versions and tutorials of applying to other trackers are released at https://github.com/ifzhang/ByteTrack.

This report presents a framework called Segment And Track Anything (SAMTrack) that allows users to precisely and effectively segment and track any object in a video. Additionally, SAM-Track employs multimodal interaction methods that enable users to select multiple objects in videos for tracking, corresponding to their specific requirements. These interaction methods comprise click, stroke, and text, each possessing unique benefits and capable of being employed in combination. As a result, SAM-Track can be used across an array of fields, ranging from drone technology, autonomous driving, medical imaging, augmented reality, to biological analysis. SAM-Track amalgamates Segment Anything Model (SAM), an interactive key-frame segmentation model, with our proposed AOT-based tracking model (DeAOT), which secured 1st place in four tracks of the VOT 2022 challenge, to facilitate object tracking in video. In addition, SAM-Track incorporates Grounding-DINO, which enables the framework to support text-based interaction. We have demonstrated the remarkable capabilities of SAM-Track on DAVIS-2016 Val (92.0%), DAVIS-2017 Test (79.2%)and its practicability in diverse applications. The project page is available at: https://github.com/z-x-yang/Segment-and-Track-Anything.

Recently, Multi-Object Tracking (MOT) has attracted rising attention, and accordingly, remarkable progresses have been achieved. However, the existing methods tend to use various basic models (e.g, detector and embedding model), and different training or inference tricks, etc. As a result, the construction of a good baseline for a fair comparison is essential. In this paper, a classic tracker, i.e., DeepSORT, is first revisited, and then is significantly improved from multiple perspectives such as object detection, feature embedding, and trajectory association. The proposed tracker, named StrongSORT, contributes a strong and fair baseline for the MOT community. Moreover, two lightweight and plug-and-play algorithms are proposed to address two inherent "missing" problems of MOT: missing association and missing detection. Specifically, unlike most methods, which associate short tracklets into complete trajectories at high computation complexity, we propose an appearance-free link model (AFLink) to perform global association without appearance information, and achieve a good balance between speed and accuracy. Furthermore, we propose a Gaussian-smoothed interpolation (GSI) based on Gaussian process regression to relieve the missing detection. AFLink and GSI can be easily plugged into various trackers with a negligible extra computational cost (1.7 ms and 7.1 ms per image, respectively, on MOT17). Finally, by fusing StrongSORT with AFLink and GSI, the final tracker (StrongSORT++) achieves state-of-the-art results on multiple public benchmarks, i.e., MOT17, MOT20, DanceTrack and KITTI. Codes are available at https://github.com/dyhBUPT/StrongSORT and https://github.com/open-mmlab/mmtracking.

Multi-model routing has evolved from an engineering technique into essential infrastructure, yet existing work lacks a systematic, reproducible benchmark for evaluating vision-language models (VLMs). We present VL-RouterBench to assess the overall capability of VLM routing systems systematically. The benchmark is grounded in raw inference and scoring logs from VLMs and constructs quality and cost matrices over sample-model pairs. In scale, VL-RouterBench covers 14 datasets across 3 task groups, totaling 30,540 samples, and includes 15 open-source models and 2 API models, yielding 519,180 sample-model pairs and a total input-output token volume of 34,494,977. The evaluation protocol jointly measures average accuracy, average cost, and throughput, and builds a ranking score from the harmonic mean of normalized cost and accuracy to enable comparison across router configurations and cost budgets. On this benchmark, we evaluate 10 routing methods and baselines and observe a significant routability gain, while the best current routers still show a clear gap to the ideal Oracle, indicating considerable room for improvement in router architecture through finer visual cues and modeling of textual structure. We will open-source the complete data construction and evaluation toolchain to promote comparability, reproducibility, and practical deployment in multimodal routing research.

Proprietary LMs such as GPT-4 are often employed to assess the quality of responses from various LMs. However, concerns including transparency, controllability, and affordability strongly motivate the development of open-source LMs specialized in evaluations. On the other hand, existing open evaluator LMs exhibit critical shortcomings: 1) they issue scores that significantly diverge from those assigned by humans, and 2) they lack the flexibility to perform both direct assessment and pairwise ranking, the two most prevalent forms of assessment. Additionally, they do not possess the ability to evaluate based on custom evaluation criteria, focusing instead on general attributes like helpfulness and harmlessness. To address these issues, we introduce Prometheus 2, a more powerful evaluator LM than its predecessor that closely mirrors human and GPT-4 judgements. Moreover, it is capable of processing both direct assessment and pair-wise ranking formats grouped with a user-defined evaluation criteria. On four direct assessment benchmarks and four pairwise ranking benchmarks, Prometheus 2 scores the highest correlation and agreement with humans and proprietary LM judges among all tested open evaluator LMs. Our models, code, and data are all publicly available at https://github.com/prometheus-eval/prometheus-eval.

Many deployments must compare candidate language models for safety before a labeled benchmark exists for the relevant language, sector, or regulatory regime. We formalize this setting as benchmarkless comparative safety scoring and specify the contract under which a scenario-based audit can be interpreted as deployment evidence. Scores are valid only under a fixed scenario pack, rubric, auditor, judge, sampling configuration, and rerun budget. Because no labels are available, we replace ground-truth agreement with an instrumental-validity chain: responsiveness to a controlled safe-versus-abliterated contrast, dominance of target-driven variance over auditor and judge artifacts, and stability across reruns. We instantiate the chain in SimpleAudit, a local-first scoring instrument, and validate it on a Norwegian safety pack. Safe and abliterated targets separate with AUROC values between 0.89 and 1.00, target identity is the dominant variance component (η^2 approx 0.52), and severity profiles stabilize by ten reruns. Applying the same chain to Petri shows that it admits both tools. The substantial differences arise upstream of the chain, in claim-contract enforcement and deployment fit. A Norwegian public-sector procurement case comparing Borealis and Gemma 3 demonstrates the resulting evidence in practice: the safer model depends on scenario category and risk measure. Consequently, scores, matched deltas, critical rates, uncertainty, and the auditor and judge used must be reported together rather than collapsed into a single ranking.

Recent advances in large language models (LLMs) have greatly improved their reasoning and decision-making abilities when deployed as agents. Richer reasoning, however, often comes at the cost of longer chain of thought (CoT), hampering interaction efficiency in real-world scenarios. Nevertheless, there still lacks systematic definition of LLM agent efficiency, hindering targeted improvements. To this end, we introduce dual-efficiency, comprising (i) step-level efficiency, which minimizes tokens per step, and (ii) trajectory-level efficiency, which minimizes the number of steps to complete a task. Building on this definition, we propose DEPO, a dual-efficiency preference optimization method that jointly rewards succinct responses and fewer action steps. Experiments on WebShop and BabyAI show that DEPO cuts token usage by up to 60.9% and steps by up to 26.9%, while achieving up to a 29.3% improvement in performance. DEPO also generalizes to three out-of-domain math benchmarks and retains its efficiency gains when trained on only 25% of the data. Our project page is at https://opencausalab.github.io/DEPO.

How can we benefit from large models without sacrificing inference speed, a common dilemma in self-driving systems? A prevalent solution is a dual-system architecture, employing a small model for rapid, reactive decisions and a larger model for slower but more informative analyses. Existing dual-system designs often implement parallel architectures where inference is either directly conducted using the large model at each current frame or retrieved from previously stored inference results. However, these works still struggle to enable large models for a timely response to every online frame. Our key insight is to shift intensive computations of the current frame to previous time steps and perform a batch inference of multiple time steps to make large models respond promptly to each time step. To achieve the shifting, we introduce Efficiency through Thinking Ahead (ETA), an asynchronous system designed to: (1) propagate informative features from the past to the current frame using future predictions from the large model, (2) extract current frame features using a small model for real-time responsiveness, and (3) integrate these dual features via an action mask mechanism that emphasizes action-critical image regions. Evaluated on the Bench2Drive CARLA Leaderboard-v2 benchmark, ETA advances state-of-the-art performance by 8% with a driving score of 69.53 while maintaining a near-real-time inference speed at 50 ms.

Physical-field reconstruction and forecasting depend on both measurement density and spatial layout, yet evaluation under a single observation pattern does not characterize performance when that pattern changes. We introduce PDE-OBS, an integrated benchmarking platform spanning numerical data generation, model training, and inference and evaluation under varying observation conditions. It combines 560,000 fields and trajectories from seven partial differential equation families with configurable observation operators and seven adapted baseline methods for stationary reconstruction and short-horizon forecasting. Separating observation construction from physical records allows users to specify parameterized patterns and deterministic mixtures for training and testing while preserving prediction targets and data splits. The evaluation protocol uses references trained for each test pattern to compare models on identical test observations and targets, alongside equal-count groups for spatial-layout comparisons. On a 14,000-record subset, we evaluate 441 trained models under nine test patterns, yielding 3,969 evaluations. Mean cross-pattern error exceeds mean matched-pattern error in all 49 PDE-method pairs, and this finding persists in a configuration-matched subset of 117 models. Denser test observations do not consistently reduce error for a fixed model. Mixed-pattern training on five completed pairs reduces large single-pattern transfer errors, although destination-trained references usually remain more accurate. Together, the benchmark and findings support systematic evaluation of observation-pattern sensitivity and provide a reusable workflow for developing methods under changing measurement conditions. Code: https://github.com/ru1ch3n/PDE-OBS.

Large Language Models (LLMs) are widely used as proxies for human labelers in both training (Reinforcement Learning from AI Feedback) and large-scale response evaluation (LLM-as-a-judge). Alignment and evaluation are critical components in the development of reliable LLMs, and the choice of feedback protocol plays a central role in both but remains understudied. In this work, we show that the choice of feedback protocol for evaluation (absolute scores versus relative preferences) can significantly affect evaluation reliability and induce systematic biases. In the context of LLM-as-a-judge evaluation, we show that pairwise protocols are more vulnerable to distracted evaluation. Generator models can exploit spurious attributes (or distractor features) favored by the LLM judge, resulting in inflated scores for lower-quality outputs. We find that absolute scoring is more robust to such manipulation, producing judgments that better reflect response quality and are less influenced by distractor features. Our results demonstrate that generator models can flip preferences by embedding distractor features, skewing LLM-as-a-judge comparisons and leading to inaccurate conclusions about model quality in benchmark evaluations. Pairwise preferences flip in about 35% of the cases, compared to only 9% for absolute scores. We offer recommendations for choosing feedback protocols based on dataset characteristics and evaluation objectives.

Online multi-object tracking has been recently dominated by tracking-by-detection (TbD) methods, where recent advances rely on increasingly sophisticated heuristics for tracklet representation, feature fusion, and multi-stage matching. The key strength of TbD lies in its modular design, enabling the integration of specialized off-the-shelf models like motion predictors and re-identification. However, the extensive usage of human-crafted rules for temporal associations makes these methods inherently limited in their ability to capture the complex interplay between various tracking cues. In this work, we introduce CAMEL, a novel association module for Context-Aware Multi-Cue ExpLoitation, that learns resilient association strategies directly from data, breaking free from hand-crafted heuristics while maintaining TbD's valuable modularity. At its core, CAMEL employs two transformer-based modules and relies on a novel association-centric training scheme to effectively model the complex interactions between tracked targets and their various association cues. Unlike end-to-end detection-by-tracking approaches, our method remains lightweight and fast to train while being able to leverage external off-the-shelf models. Our proposed online tracking pipeline, CAMELTrack, achieves state-of-the-art performance on multiple tracking benchmarks. Our code is available at https://github.com/TrackingLaboratory/CAMELTrack.

We study the right to be forgotten (GDPR Art. 17) for large language models and frame unlearning as a reproducible systems problem. Our approach treats training as a deterministic program and logs a minimal per-microbatch record (ordered ID hash, RNG seed, learning-rate value, optimizer-step counter, and accumulation boundary). Under a pinned stack and deterministic kernels, replaying the training tail while filtering only the forget closure yields the same parameters as training on the retain set (bit-identical in the training dtype) when preconditions hold. To meet latency and availability constraints, we add complementary paths: (i) exact reverts of recent steps via micro-checkpoints or dense per-step deltas, (ii) cohort-scoped adapter deletion when the base is frozen, and (iii) a curvature-guided anti-update followed by a short retain-tune, audit-gated with escalation to exact replay. We report storage/latency budgets and a toy artifact validating mechanics; in a controlled run that satisfies the preconditions we demonstrate byte-identical equality of model and optimizer states.

Model routing aims to select the most suitable model from a candidate pool for each query, balancing quality and cost. Existing VLM routing research is limited to traditional VQA evaluation, lacks systematic calibration optimization for open-set scenarios, and employs training objectives that dilute multi-positive signals via softmax normalization without incorporating cost. We address these limitations with three contributions: (1)VLM-ExecRouterBench, the first execution-oriented VLM routing benchmark covering Code, Agentic, and Search domains with 11 candidate models spanning nearly two orders of magnitude in pricing; (2)SCOPE-Router, a dual-tower router that matches queries to model behavior profiles constructed via hybrid calibration (random/diagnostic/diversity sampling), enabling new models to join routing without retraining; (3)CRM+RCCR, an architecture-agnostic cost-aware objective that encodes cost preference into continuous relevance targets through per-pair independent scoring, eliminating multi-positive dilution while regularizing queries with similar routing preferences to be closer in the routing space. Empirically, SCOPE-Router achieves the best Rank Score on all three benchmarks, surpassing the runner-up by 1.84 points under OOD settings and by 6.75 points under doubly OOD open-set evaluation. When applied to four diverse routers, CRM+RCCR improves Rank Score by 1.25--6.21 points.

Unsupervised physical parameter estimation from video lacks a common benchmark: existing methods evaluate on non-overlapping synthetic data, the sole real-world dataset is restricted to single-body systems, and no established protocol addresses governing-equation identification. This work introduces IRIS, a high-fidelity benchmark comprising 240 real-world videos captured at 4K resolution and 60fps, spanning both single- and multi-body dynamics with independently measured ground-truth parameters and uncertainty estimates. Each dynamical system is recorded under controlled laboratory conditions and paired with its governing equations, enabling principled evaluation. A standardized evaluation protocol is defined encompassing parameter accuracy, identifiability, extrapolation, robustness, and governing-equation selection. Multiple baselines are evaluated, including a multi-step physics loss formulation and four complementary equation-identification strategies (VLM temporal reasoning, describe-then-classify prompting, CNN-based classification, and path-based labelling), establishing reference performance across all IRIS scenarios and exposing systematic failure modes that motivate future research. The dataset, annotations, evaluation toolkit, and all baseline implementations are publicly released.

Agent benchmarks often evaluate only final answers even when agents run on stateful runtimes. We argue this under-specifies what is being evaluated: the proper unit is a declared model-plus-runtime configuration whose failures can occur in evidence acquisition, runtime routing, safety boundaries, or repeated execution. We present ClawProBench, a trace-aware benchmark for runtime-native agent evaluation instantiated on OpenClaw, a live agent runtime with workspace tools and native surfaces for browsing, memory, messaging, scheduling, skills, and subagents. ClawProBench defines two tracks: a 102-scenario full profile with live workspace and native-runtime routing tasks, and a frozen 68-scenario holdout with closed-world JSON output contracts for robust ranking. Trials are scored from execution traces via a safety-gated formula combining correctness, process quality, and efficiency, preserving failure evidence for audit. Our anonymous artifact includes benchmark definitions, scoring code, manifests and sanitized traces. We evaluate 68 configurations on the full profile and 37 on holdout. The top safety-gated average trace score is 0.7671. Native-runtime tasks underperform workspace-live tasks (0.5238 vs. 0.6415). On holdout, pass@k-any outperforms strict three-trial pass (0.6638 vs. 0.2890), while full-profile and holdout rankings show weak alignment (Spearman 0.1300). Rankings based purely on correctness differ substantially from process-aware, safety-gated and strict-pass views. Final-answer leaderboards may hide native-surface weaknesses, one-off successes and trace-local agent failure modes.

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

Runtime-governance evidence often collapses materially different events into one audit record: a decision may be made, an instruction dispatched or received, an action may or may not execute, and a resulting state may or may not be observed. This paper presents AIREP, a vendor- and model-independent protocol for per-decision AI runtime evidence. Its current wire model separates evidence into four artifact families: Decision, Control, Execution, and Effect. Artifacts use closed core schemas, explicit identities and digests, declared scope limits, RFC 8785 canonical JSON, domain-separated SHA-256 hashing, and pure Ed25519 signatures. A three-level assurance model distinguishes structural/hash consistency (AIREP-Core), verifier-accepted authorship (AIREP-Authenticated), and independently anchored chain-head freshness and non-truncation relative to an accepted witness (AIREP-Witnessed); these classes do not establish event truth. A structured reconciler preserves failure, missing evidence, unevaluated prerequisites, and indeterminate outcomes as distinct states. The released beta includes a four-family first-party producer, Python and Node reference-verification paths, adversarial/lifecycle corpora, and reproducible validation. Post-release first-party Hermes and LightEval integration exercises preserve explicit evidence boundaries without claiming adoption or interoperability. Independent implementation evidence exists separately for a v0.1.2 producer and a v0.2 consumer/verifier; because they target different frozen versions, they do not establish same-version producer-to-consumer interoperability. AIREP remains experimental.

Federated learning lets a UAV fleet train a shared GPS spoofing detector without raw receiver data leaving any aircraft, and several recent UAV-FL designs weight each client by the validation accuracy it reports about itself. We show this self-report is an exploitable attack lever: two compromised clients of ten that poison part of their data, scale their updates, and inflate their reported accuracy raise backdoor lift to +0.3036, higher than the same attack achieves without lying. We propose receiver-domain behavioral probing, in which the coordinator evaluates every submitted model on counterfactual spoofed samples built by driving each discriminative GPS feature to a benign value, weighting clients by what their models do rather than what they claim. Under independent and identically distributed clients this reduces attacker-induced lift to -0.0265, statistically indistinguishable from an honest fleet, while flagging the compromised aircraft. Unlike Byzantine-robust aggregation it needs no exact attacker count, only an honest majority: when the true count exceeds the configured value, Multi-Krum degrades from +0.0061 to +0.2837 while ours stays near baseline. Evaluation uses one public single-receiver dataset partitioned into simulated clients; we also report where the mechanism fails, under strong client heterogeneity.

Standardized benchmarks are crucial for the majority of computer vision applications. Although leaderboards and ranking tables should not be over-claimed, benchmarks often provide the most objective measure of performance and are therefore important guides for research. The benchmark for Multiple Object Tracking, MOTChallenge, was launched with the goal to establish a standardized evaluation of multiple object tracking methods. The challenge focuses on multiple people tracking, since pedestrians are well studied in the tracking community, and precise tracking and detection has high practical relevance. Since the first release, MOT15, MOT16, and MOT17 have tremendously contributed to the community by introducing a clean dataset and precise framework to benchmark multi-object trackers. In this paper, we present our MOT20benchmark, consisting of 8 new sequences depicting very crowded challenging scenes. The benchmark was presented first at the 4thBMTT MOT Challenge Workshop at the Computer Vision and Pattern Recognition Conference (CVPR) 2019, and gives to chance to evaluate state-of-the-art methods for multiple object tracking when handling extremely crowded scenarios.

Large language model safeguards decide whether to answer before seeing how an answer will be used. This creates a basic problem for dual-use tasks: the same answer can help an authorized professional or an attacker, while an attacker can imitate a benign request and interaction history. We separate the capability released by the model from the evidence available about downstream use. When that evidence is copyable, we derive the exact worst-case floor on attacker assistance while preserving useful answers. The result yields a safety trilemma: Useful Capability, Reliable Safety, and Open Access cannot coexist. We then show how a trusted credential can complement existing safeguards by adding hard-to-copy information that predicts actual downstream use, and identify the stronger condition needed to eliminate the floor. Evidence from dual-use evaluations, adaptive attacks, and deployed trusted-access programs supports the practical relevance of these conditions.

We investigate how looped transformers encode human preference, training lightweight evaluator heads on frozen Ouro-2.6B loop-iteration states on Anthropic HH-RLHF. v2: an erratum is prepended; the original manuscript is unchanged. A post-publication audit found the three headline results inflated by two independent evaluation errors. The 95.2% pairwise evaluator accuracy is a canonical-ordering artifact: the data were correctly split, but the evaluator learned to prefer the first-presented argument; its strict antisymmetrized accuracy on the full 8,552-pair test set is 63.9%. The 84.5% pairwise probe and the below-chance 21.75% pointwise probe were source-item leaks (orientation rows and pair partners crossing the train/test split); corrected pair-disjoint values are 56.5% and 54.2% -- above chance, so the "inverted polarity" finding is withdrawn. The central finding survives at much smaller magnitude: preference is decoded more accurately relationally than pointwise (paired +2.3 points, 95% CI [+1.3, +3.3]), and the antisymmetrized evaluator still beats the linear probe, but no corrected readout rivals end-to-end reward models. The methodological findings stand (constant-output degeneracy, flip test, swap-protocol metric deflation), with one correction: antisymmetrized accuracy, not antisymmetry correlation, certifies relational discrimination. The two errors are mutually invisible -- a split audit cannot see an ordering prior, antisymmetrization cannot see a leak -- so both checks are required. Full audit in the follow-up work (Kirin, 2026, in preparation).

We present the contribution of the Unbabel team to the WMT 2020 Shared Task on Metrics. We intend to participate on the segment-level, document-level and system-level tracks on all language pairs, as well as the 'QE as a Metric' track. Accordingly, we illustrate results of our models in these tracks with reference to test sets from the previous year. Our submissions build upon the recently proposed COMET framework: We train several estimator models to regress on different human-generated quality scores and a novel ranking model trained on relative ranks obtained from Direct Assessments. We also propose a simple technique for converting segment-level predictions into a document-level score. Overall, our systems achieve strong results for all language pairs on previous test sets and in many cases set a new state-of-the-art.

Modern video games provide a measurable testbed for AI models, combining abilities of visual understanding, instruction decomposition, goal planning, and precise action control over multiple temporal horizons. Existing datasets and benchmarks, however, either cover a narrow range of games, lack language instructions, or rely on high-variance online rollouts. To address these challenges, we introduce GameHorizon, a unified data and evaluation suite that measures gameplay capabilities at different horizons for diverse model families. GameHorizon Suite consists of three components. First, GameHorizon-Annotator is a scalable and automated annotation pipeline for multi-horizon instructions. Second, utilizing the pipeline, we construct GameHorizon-Data, the first large-scale AAA gameplay dataset with temporally aligned videos, player actions, and multi-horizon instructions. It comprises 5,000 hours of recordings from 21 games, collected by 100 human expert players. Third, we build GameHorizon-Bench with reproducible offline and stepwise online testing. The offline track enables reproducible evaluation using thousands of standardized questions organized into three primary tasks and a series of diagnostic variants, while the online track tests whether offline scores reflect actual gameplay capabilities and localizes failures to specific steps within long-horizon gameplay. Based on our GameHorizon Suite, we evaluate 47 models through more than one million model invocations, revealing a meaningful hierarchy of task difficulty and pronounced differences in model capabilities. Our work can provide a standardized yardstick for evaluating gameplay capabilities across horizons and model families. We will release our dataset, annotator, and benchmark to facilitate future research.

Platform abuse campaigns conceal redirection instructions with emojis, homophones, character decomposition, and redundant symbols, then route users through disguised links to services associated with pornography, fraud, gambling, or illicit transactions. Existing benchmarks evaluate obfuscated text and risky webpages separately, obscuring how target recovery affects downstream evidence acquisition. We introduce RiskChainBench, pairing 3,600 synthetic token-text restoration inputs from 600 source sessions with 600 corresponding human-labeled local web environments. A model first restores the message, operational intent, and destination; the same underlying model then acts as a VLM-driven web agent that investigates the correctly associated website and produces a frozen, evidence-cited risk report without message-side semantics or domain-reputation cues. We score restoration and correct-routing web investigation separately and compose them offline by applying the frozen primary-entry prediction as a gate to the same Task 2 result. Human labels determine task correctness, while a fixed multimodal evidence judge assesses faithfulness, sufficiency, completeness, and consistency. Across ten models, Entry Top-1 ranges from 35.2% to 95.2% and web decision accuracy from 26.3% to 62.8%; the leading systems differ across entry recovery, full reconstruction, website decisions, and fine-grained typing. Execution failures account for 31.9% of web runs, whereas post-decision type errors account for only 0.9%, identifying stable exploration and risk judgment as the principal bottlenecks. We release the benchmark, protocol, and resettable local sandbox.

Temporal modeling of objects is a key challenge in multiple object tracking (MOT). Existing methods track by associating detections through motion-based and appearance-based similarity heuristics. The post-processing nature of association prevents end-to-end exploitation of temporal variations in video sequence. In this paper, we propose MOTR, which extends DETR and introduces track query to model the tracked instances in the entire video. Track query is transferred and updated frame-by-frame to perform iterative prediction over time. We propose tracklet-aware label assignment to train track queries and newborn object queries. We further propose temporal aggregation network and collective average loss to enhance temporal relation modeling. Experimental results on DanceTrack show that MOTR significantly outperforms state-of-the-art method, ByteTrack by 6.5% on HOTA metric. On MOT17, MOTR outperforms our concurrent works, TrackFormer and TransTrack, on association performance. MOTR can serve as a stronger baseline for future research on temporal modeling and Transformer-based trackers. Code is available at https://github.com/megvii-research/MOTR.

MIDOG 2025 Track 1 requires mitosis detection in whole-slideimages (WSIs) containing non-tumor, inflamed, and necrotic re-gions. Due to the complicated and heterogeneous context, aswell as possible artifacts, there are often false positives and falsenegatives, thus degrading the detection F1-score. To addressthis problem, we propose a two-stage framework. Firstly, an im-proved YOLO11x, integrated with EMA attention and LSConv,is employed to generate mitosis candidates. We use a low confi-dence threshold to generate as many proposals as possible, en-suring the detection recall. Then, a ConvNeXt-Tiny classifieris employed to filter out the false positives, ensuring the detec-tion precision. Consequently, the proposed two-stage frame-work can generate a high detection F1-score. Evaluated on afused dataset comprising MIDOG++, MITOS_WSI_CCMCT,and MITOS_WSI_CMC, our framework achieves an F1-scoreof 0.882, which is 0.035 higher than the single-stage YOLO11xbaseline. This performance gain is produced by a significantprecision improvement, from 0.762 to 0.839, and a comparablerecall. On the MIDOG 2025 Track 1 preliminary test set, thealgorithm scores an F1 score of 0.7587. The code is available athttps://github.com/xxiao0304/MIDOG-2025-Track-1-of-SZTU.

Fashion attribute extraction is evaluated inconsistently: results are reported as single aggregate numbers across image types that pose different problems, fields that are not visible in an image are scored as ordinary negatives, and the effect of vocabulary mismatch between datasets is acknowledged but not measured. We release the MODA General Attribute Suite, a four-track benchmark that keeps these problems separate by construction. Each track (localized garment crops, catalogue product images, applicability-aware full-body photographs, and product text) carries its own frozen test set, input contract, metric, and leakage unit, and the tracks are never averaged. The protocol requires label-blind prediction, SHA-256 commitment of prediction files before any label is opened, a fail-closed scorer, and a 10,000-sample paired cluster bootstrap at each track's natural leakage unit; promotion requires a positive interval on every track rather than a favourable mean. We release the scorers, the split builders, our own prediction files with their hashes including the runs we lose, and the MODA_NER(V) model checkpoints for three of the four tracks. The text model is not distributed; its benchmark is. We report baseline results, six interventions that did not improve them, and the limitations of the evidence.

We present RouteJudge, an online pairwise preference evaluation framework for LLM routing systems, with a public platform available at https://routejudge.cn. Different from model-level response evaluation, RouteJudge focuses on router-level decision quality. For each user query, multiple routing strategies independently recommend candidate models under the same model pool and budget constraints. The selected model responses are then presented to users through anonymous pairwise comparisons, and the resulting user preferences are attributed back to the routing strategies behind the compared responses. Each evaluation record stores the query, routing decisions, model responses, preference labels, cost, latency, and task metadata, enabling preference-aware, cost-aware, and task-conditioned analysis of LLM routers. To support the continuous expansion of routing methods in RouteJudge, we further release ORBIT (Optimal Routing and Budgeted Inference Toolbox), a modular and extensible toolbox that standardizes the end-to-end workflow of LLM routing. ORBIT provides unified interfaces for benchmark loading, query representation, router implementation, budget-aware evaluation, and method comparison, allowing researchers to develop and evaluate routing algorithms under consistent protocols. It also serves as the submission and integration layer for RouteJudge: researchers can implement routing methods within ORBIT, validate them on existing routing benchmarks, and submit compatible routers for online preference-based evaluation. The code of ORBIT is available at https://github.com/LAMDA-Model-Reuse/ORBIT.

Skill assessment in procedural videos is crucial for the objective evaluation of human performance in settings such as manufacturing and procedural daily tasks. Current research on skill assessment has predominantly focused on sports and lacks large-scale datasets for complex procedural activities. Existing studies typically involve only a limited number of actions, focus on either pairwise assessments (e.g., A is better than B) or on binary labels (e.g., good execution vs needs improvement). In response to these shortcomings, we introduce ProSkill, the first benchmark dataset for action-level skill assessment in procedural tasks. ProSkill provides absolute skill assessment annotations, along with pairwise ones. This is enabled by a novel and scalable annotation protocol that allows for the creation of an absolute skill assessment ranking starting from pairwise assessments. This protocol leverages a Swiss Tournament scheme for efficient pairwise comparisons, which are then aggregated into consistent, continuous global scores using an ELO-based rating system. We use our dataset to benchmark the main state-of-the-art skill assessment algorithms, including both ranking-based and pairwise paradigms. The suboptimal results achieved by the current state-of-the-art highlight the challenges and thus the value of ProSkill in the context of skill assessment for procedural videos. All data and code are available at https://fpv-iplab.github.io/ProSkill/

When researchers ask whether two transformer layers are "equivalent" for compression, they often conflate distinct tests. Replacement asks whether one layer's map can substitute for another's in place; interchange asks whether two layers approximately commute when their positions are swapped. Both are output-grounded swap-KL probes, but they need not agree: on pretrained transformers the protocol gap can change which layers look safe to prune by several-fold under the same evaluator, especially when replacement distances are high. We measure both protocols across checkpoints and architectures. On a Pythia training trajectory (410M and 1.4B), the replacement-interchange gap grows from initialization to convergence. Under one matched WikiText-2 contract at 8B scale, Qwen3-8B enters a divergent regime: interchange-guided removal is several-fold safer than replacement-guided at the same layer budgets, while Llama-3.1-8B ties the two protocols for pruning cost even though interchange KL is lower, showing metric gaps need not map one-to-one to removal. Before layer removal or merging, score both swap-KLs on the target checkpoint; the diagnostic requires only unlabeled forward passes.

AI-generated music detectors are commonly compared using aggregate scores on benchmarks whose training overlap, generator lineage, source provenance, and audio-transformation history are only partially observable. This paper introduces ArtifactBench, a lineage-aware evaluation suite for measuring detector behavior across generator families and versions, real-music domains, collection-cohort shift, and inference coverage. The benchmark groups source recordings and their derived variants by content identity, separates calibration from final testing, records inference failures independently from classification errors, and reports source-level performance with uncertainty in addition to aggregate metrics. We evaluate multiple publicly available detectors under a version-pinned common protocol and examine how leakage control, cohort availability, threshold policy, and model-specific missingness alter measured performance and model ranking. On the 562-track common-success test intersection, ArtifactNet obtains 0.982 AUROC and 0.918 balanced accuracy, compared with 0.761/0.776 for the public Deezer detector; SpecTTTra and CLAM fall below 0.30 AUROC under this shifted cohort. These results also expose substantial generator- and real-domain shifts that aggregate scores alone conceal.

Advances in perception modeling have significantly improved the performance of object tracking. However, the current methods for specifying the target object in the initial frame are either by 1) using a box or mask template, or by 2) providing an explicit language description. These manners are cumbersome and do not allow the tracker to have self-reasoning ability. Therefore, this work proposes a new tracking task -- Instruction Tracking, which involves providing implicit tracking instructions that require the trackers to perform tracking automatically in video frames. To achieve this, we investigate the integration of knowledge and reasoning capabilities from a Large Vision-Language Model (LVLM) for object tracking. Specifically, we propose a tracker called TrackGPT, which is capable of performing complex reasoning-based tracking. TrackGPT first uses LVLM to understand tracking instructions and condense the cues of what target to track into referring embeddings. The perception component then generates the tracking results based on the embeddings. To evaluate the performance of TrackGPT, we construct an instruction tracking benchmark called InsTrack, which contains over one thousand instruction-video pairs for instruction tuning and evaluation. Experiments show that TrackGPT achieves competitive performance on referring video object segmentation benchmarks, such as getting a new state-of the-art performance of 66.5 J&F on Refer-DAVIS. It also demonstrates a superior performance of instruction tracking under new evaluation protocols. The code and models are available at https://github.com/jiawen-zhu/TrackGPT{https://github.com/jiawen-zhu/TrackGPT}.

Multi-reference-to-audio-video (MR2AV) generation aims to generate coherent audio-video content conditioned on multiple references and textual instructions. Existing benchmarks mainly focus on text-driven generation, single-reference subject preservation, or isolated audio-video alignment, leaving the emerging MR2AV setting largely unexplored. Compared with these settings, MR2AV requires models to jointly reason over multiple references while generating synchronized visual and audio content. Models must not only preserve each reference faithfully but also correctly bind and compose multiple referenced entities into coherent audio-visual events. To address this gap, we introduce MultiRef-Compass, a unified benchmark for MR2AV generation. It comprises 350 carefully curated samples constructed through a scalable and controllable asset-composition pipeline, covering multi-view subject preservation, multi-entity binding, and human-object-scene composition. To provide interpretable assessment, MultiRef-Compass defines an evaluation protocol with four dimensions: Basic Quality, Reference Consistency, Audio-Visual Consistency, and Instruction Following, using 14 sub-metrics. MultiRef-Compass integrates automatic metrics with a rejudging-enhanced MLLM-as-a-Judge framework, enabling scalable and auditable evaluation of both perceptual fidelity and reference-conditioned composition. Extensive experiments on eight representative MR2AV systems reveal substantial room for improvement across multiple evaluation dimensions, underscoring the need for a comprehensive benchmark and positioning MultiRef-Compass as a foundation for future MR2AV research.

Large language model agents now act on codebases, browsers, operating systems, calendars, files, and tool ecosystems, but the benchmarks used to evaluate them are fragmented: each emphasizes a different unit of measurement (final task success, tool-call validity, repeated-pass consistency, trajectory safety, or attack robustness). A line of 2024-2025 work has converged on the diagnosis that a single accuracy column is no longer the right unit of comparison for deployable agents. AgentAtlas extends this line of work with four components: (i) a six-state control-decision taxonomy (Act / Ask / Refuse / Stop / Confirm / Recover); (ii) a nine-category trajectory-failure taxonomy with two orthogonal hierarchical labels (primary_error_source, impact); (iii) a taxonomy-aware vs. taxonomy-blind methodology that measures how much of a model's apparent capability comes from the supervision in the prompt; and (iv) a benchmark-coverage audit mapping fifteen agent benchmarks against six behavioral axes. To demonstrate the methodology we run a small fixed eight-model set (1,342 generated items, four frontier closed and four open-weight) under both prompt modes. Removing the explicit label menu drops every model's trajectory accuracy by 14-40 pp to a tight 0.54-0.62 floor regardless of family, and no single model wins on all three of control accuracy, trajectory diagnosis, and tool-context utility retention. We treat the synthetic run as a measurement-protocol demonstration, not a benchmark release.

Multi-object tracking in sports scenarios has become one of the focal points in computer vision, experiencing significant advancements through the integration of deep learning techniques. Despite these breakthroughs, challenges remain, such as accurately re-identifying players upon re-entry into the scene and minimizing ID switches. In this paper, we propose an appearance-based global tracklet association algorithm designed to enhance tracking performance by splitting tracklets containing multiple identities and connecting tracklets seemingly from the same identity. This method can serve as a plug-and-play refinement tool for any multi-object tracker to further boost their performance. The proposed method achieved a new state-of-the-art performance on the SportsMOT dataset with HOTA score of 81.04%. Similarly, on the SoccerNet dataset, our method enhanced multiple trackers' performance, consistently increasing the HOTA score from 79.41% to 83.11%. These significant and consistent improvements across different trackers and datasets underscore our proposed method's potential impact on the application of sports player tracking. We open-source our project codebase at https://github.com/sjc042/gta-link.git.

We review the AIM 2020 challenge on virtual image relighting and illumination estimation. This paper presents the novel VIDIT dataset used in the challenge and the different proposed solutions and final evaluation results over the 3 challenge tracks. The first track considered one-to-one relighting; the objective was to relight an input photo of a scene with a different color temperature and illuminant orientation (i.e., light source position). The goal of the second track was to estimate illumination settings, namely the color temperature and orientation, from a given image. Lastly, the third track dealt with any-to-any relighting, thus a generalization of the first track. The target color temperature and orientation, rather than being pre-determined, are instead given by a guide image. Participants were allowed to make use of their track 1 and 2 solutions for track 3. The tracks had 94, 52, and 56 registered participants, respectively, leading to 20 confirmed submissions in the final competition stage.

Most work on improving large language models treats accuracy as the sole objective. We argue that the harness, the Python code surrounding the model that constructs prompts, routes calls, and parses outputs, is a first-class design surface whose quality is inherently multi-objective: an accurate harness that refuses no unsafe request, or that consumes an order of magnitude more tokens, is not a good harness. We present Meta-Harness, a system that casts harness design as search over three per-domain objectives (accuracy, behavioural safety, and token cost) solved by an agentic proposer (Claude Code) with full filesystem access to prior harness source, execution traces, and scoring artifacts. Our central finding is that a singlephase joint-reward proposer (MoMHa) outperforms every alternative, including a two-phase "accuracy then tokens" ablation, scalar-only feedback, and an accuracy-only baseline. We evaluate on seventeen domains: seven synthetic capability suites, seven real-world public benchmarks (HumanEval, MBPP, Spider, FEVER, MMLU-Pro, LawBench, NuminaMath), and three U-SafeBench-derived user-specific safety domains, using a 12-model fleet spanning four families. On the synthetic track MoMHa achieves a joint mean of 0.482 versus 0.198-0.422 for ten baselines, winning 7 / 10 per-domain columns; on the real-world track it scores 0.461 versus 0.377 for the strongest baseline (DSPy), winning 5/7 columns, demonstrating that harness strategies transfer to unseen benchmarks without retraining on 8 of 12 target models. MoMHa attains the highest measured behavioral safety composite (U-SafeBench, 0.781) and uses 95 fewer tokens per example than the two-phase alternative. We will release all harness code, evaluation infrastructure, and crossmodel logs.

We present Yambda-5B, a large-scale open dataset sourced from the Yandex.Music streaming platform. Yambda-5B contains 4.79 billion user-item interactions from 1 million users across 9.39 million tracks. The dataset includes two primary types of interactions: implicit feedback (listening events) and explicit feedback (likes, dislikes, unlikes and undislikes). In addition, we provide audio embeddings for most tracks, generated by a convolutional neural network trained on audio spectrograms. A key distinguishing feature of Yambda-5B is the inclusion of the is_organic flag, which separates organic user actions from recommendation-driven events. This distinction is critical for developing and evaluating machine learning algorithms, as Yandex.Music relies on recommender systems to personalize track selection for users. To support rigorous benchmarking, we introduce an evaluation protocol based on a Global Temporal Split, allowing recommendation algorithms to be assessed in conditions that closely mirror real-world use. We report benchmark results for standard baselines (ItemKNN, iALS) and advanced models (SANSA, SASRec) using a variety of evaluation metrics. By releasing Yambda-5B to the community, we aim to provide a readily accessible, industrial-scale resource to advance research, foster innovation, and promote reproducible results in recommender systems.

As autonomous AI agents increasingly transact across organizational boundaries, a fundamental trust challenge emerges: how can an agent assess whether an unknown counterpart is trustworthy? The ERC-8004 protocol addresses this challenge with the first permissionless trust layer for AI agent economies, built around three on-chain registries for Identity, Reputation, and Validation. Despite its rapid adoption, the protocol has not been studied empirically, leaving it unclear whether the information it records provides a trustworthy basis for decision-making. To address this gap, we present the first empirical study of ERC-8004 across three chains: Ethereum, BNB Smart Chain (BSC), and Base, covering the period from protocol deployment through May 13, 2026. We crawl on-chain Identity and Reputation events, off-chain files, and x402 payment transactions. On the identity side, we find that most registrations are placeholders rather than active agents, with only a small fraction (3%, 4%, and 15% across Ethereum, BSC, and Base) exposing a valid ERC-8004 registration file with at least one live service endpoint. On the reputation side, we show that the Registry, as currently deployed, cannot function as a trust signal: values are not commensurable, feedback records are rarely grounded in verifiable interactions, and reputation can be manipulated at minimal cost. Consistent with these design weaknesses, we find that a substantial fraction of reviewers (73.5%, 59.2%, and 90.6% across Ethereum, BSC, and Base) exhibit coordinated Sybil behavior. After removing Sybil-flagged feedback, 15.8%, 77.9%, and 86.8% of rated agents, respectively, are left with no valid feedback. We then turn these findings into concrete recommendations for future revisions of ERC-8004. Our study yields actionable protocol-design implications and establishes an empirical baseline for research on AI agent markets.

Multi-agent large language model (LLM) debate is often evaluated by whether final answers improve, but movement is not necessarily improvement: the same discussion can rescue an initially wrong majority or destroy an initially correct one. Standard final-accuracy evaluations conflate these opposing mechanisms. We introduce an auditable protocol for homogeneous debate on multiple-choice questions (MCQs) that records each run as a transition ledger over collapse, correction, onset, and signed intervention utility. On 6,925 MMLU-Pro debates, the protocol identifies 253 collapses and a parallel correction ledger that changes how interventions should be judged. Replay experiments reveal the central tradeoff: a leave-one-model-out probe-gated freeze prevents 29 collapses but loses 108 corrections under equal weights, so collapse prevention alone can recommend the wrong policy. A compact pre-debate 8-probe screen is a triage signal: its unadjusted family-level association with conditional-collapse risk is high (G=7, Spearman rho=0.893, exact two-sided p=0.0123), but initial-majority accuracy is a close comparator (rho=0.821; family partial rho=0.767, p=0.0877), so we do not treat it as calibrated or capability-adjusted prediction. Round-level traces localize many collapses to the first debate round, where early disagreement can precede both harmful cascades and useful recovery. We release replayable schemas, coders, audits, cost cards, and zero-API rebuild scripts so future model-scaffold rows can be compared under the same denominators and signed utility ledger.

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

Reinforcement learning with verifiable rewards, and Group Relative Policy Optimization (GRPO) in particular, is now run routinely on a supervised checkpoint in the hope of producing a stronger agent. We ask whether it adds skill to a small language and vision-language model web agent at the 4B to 8B scale, or whether it mostly reshapes behavior the supervised model already has. Across a control grid of 18 runs that varies learning rate, KL weight, seed, initialization, and clipping, no configuration credibly improves the success rate of a strong supervised baseline on tasks the agent has largely mastered. On the text track, moderate to high learning rates make it credibly worse. The null holds under paired testing, 25 evaluation seeds, 6 training seeds, changes to the recipe, both text and Set-of-Marks screenshot observations, and scaling the backbone to 8B; the credible harm is a text-track finding and is only nominal under Set-of-Marks. To show that the null reflects the setting and not a broken pipeline, we run the identical harness, reward, and recipe on tasks whose reward is reachable by sampling, and there the success rate rises by 22 points with a paired interval that excludes zero. GRPO therefore helps only when there is headroom to climb, meaning the sampled policy already succeeds more often than the greedy one. We then explain the failure. A middle learning rate degrades the agent and a high one collapses it, and the two regimes form a double dissociation: grafting localizes the degrade regime to the attention and MLP blocks, while the collapse regime cannot be traced to any single group, and the embedding change that dominates the weight movement is causally inert. At 4B, effective rank in the late layers tracks capability in both directions; at 8B the two come apart. This coupling is specific to the smaller model, so we report it as scale-dependent.

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

The open radio access network (O-RAN), with its disaggregated and open architecture, is poised to meet the demands of the next generation of wireless communication. However, to unlock the full potentials of O-RAN, real-time network modeling and optimization are essential. A promising solution for such requirement is the use of network digital twin (NDT). NDT provides a comprehensive view of a network, covering both physical and logical components, including infrastructure, protocols, and algorithms. NDT, as a real-time virtual representation of O-RAN facilitates a variety of operations, such as emulations, test, optimization, monitoring, and analysis of a new configuration in a risk-free environment, without requiring them to be implemented in real network. Such capability enables the vendors and network operators for a faster adoption of new solutions with frequent updates, while ensuring the resiliency of the existing services via planning ahead under various "what-if" scenarios. In this paper, we first describe what exactly NDT means in the context of O-RAN, as well as its key enablers. We then describe the NDT application within the O-RAN in both prior and post-deployment. Finally, we provide two practical uses cases, namely network energy efficiency and traffic steering, where the NDT can be leveraged effectively.

Recent large language models (LLMs) can operate as coding agents that build complete games from natural language requests. Game development is especially demanding because program logic, visual and audio content, interfaces, interaction and playability must function together in one executable artifact. Measuring this capability therefore requires evaluation of both game product and the development process. Existing benchmarks often assess the game development capabilities of LLMs by evaluating the final artifact or an isolated development stage. Our analysis of complete human-agent development trajectories identifies three stages that together span the lifecycle of game development with a coding agent: initial game generation, bug diagnosis and repair, and optimization over multiple turns. Therefore, we introduce GameXpert-Bench, which operationalizes the three lifecycle stages as three complementary benchmark tracks. GameGen evaluates complete game creation from a single request in an empty workspace. GameFix evaluates diagnosis and repair when defects are reported or left for the agent to discover. GameOpt evaluates cumulative optimization through request chains seeded by real development trajectories between users and agents. We evaluate each track using live game interaction, deterministic behavioral tests, or final product criteria with regression checks. The suite contains 97 generation tasks across 11 genres; 100 repair tasks from 50 game levels verified by humans, each with 19-27 injected bugs; and 17 optimization chains with six turns and 102 requests. Across the three tracks, current agents are more reliable at producing playable foundations and implementing explicit requirements than at discovering defects, verifying runtime behavior, and preserving functionality across changes.

Full-duplex spoken dialogue models listen and speak at the same time, enabling voice agents to have natural, low-latency interactions that turn-based systems cannot offer. However, they are commonly evaluated against single-sided interlocutors: pre-recorded audio that cannot react, or an automated examiner that reacts in real time but only administers a fixed sequence of tests and is never graded. These single-sided frameworks evaluate only half of a two-body problem, where turn-taking, overlap, and interruption are joint products of two coupled speakers. We propose DyaFDB, a framework that evaluates full-duplex models in a dyadic setup: two models converse directly under assigned roles with cooperative or conflicting goals, and both sides are scored offline with an external judge. DyaFDB probes how the two models behave toward each other, such as how they take turns or carry an assigned role under different interests. We instantiate four tasks as 140 scenarios and record 7,560 conversations, covering six self- and cross-play pairings. Throughout the experiments, we observe that how a model behaves continually reshapes its partner. We thus demonstrate that each model must be both the examiner and examinee of the other, and no single fixed interlocutor can play both parts. We will release the scenarios, role prompts, and recording protocols between two full-duplex models, without any pre-recorded audio.

Conversational memory changes during use, so endpoint question answering alone cannot establish how a persistent state accumulates, ages, or incorporates revisions. We introduce LSREP, a Longitudinal State-Replay Evaluation Protocol combining ordered replay, explicit lifecycle schedules, repeated probes, evolving reference answers, and mechanism-fidelity checks. Its architectural case study is ICE v2, a local-first memory middleware with typed stores, retrieval fusion, and dynamic context budgets. The private, single-user instantiation contains 1,985 turns, 219 distinct probes, and 1,211 probe-checkpoint observations across 52 checkpoints. On three ordinary-density datasets, ICE v2 has a near-zero mean quality difference from vector-RAG while selecting 32% fewer fragments but using 6.6% more estimated prompt tokens. A fourth, dense dataset exposes catastrophic failures of the unbudgeted baseline. The fidelity audit limits attribution: procedural retrieval is defective, several mechanisms are unexercised, and graph utility is not established. In a complementary matched public diagnostic, ICE v2 loses decisively to pure vector-RAG on LongMemEval: 50.8% versus 72.8% in the evidence-only oracle and 43.0% versus 69.5% in full-S. Paired differences are -22.0 points (95% CI [-26.6, -17.4]) and -26.5 ([-31.3, -21.8]). Conservative abstention accompanies severe multi-session and temporal failures. ICE uses less context in this diagnostic, establishing a quality-cost trade-off rather than superior efficiency. Together, replay, fidelity auditing, and public endpoint testing expose distinct failure modes that neither architectural descriptions nor aggregate scores identify alone.

The Model Context Protocol (MCP), introduced by Anthropic in November 2024, defines a standardized interface for connecting large language models (LLMs) to external tools, data sources, and services. Within months of release, hundreds of community-built MCP servers appeared on GitHub, but no software-maintenance literature has yet described how the ecosystem is being structured in production. This industry experience paper catalogues five recurring MCP server architectural patterns observed across an enumerated corpus of fifteen independently developed servers (five production servers from the ANSYR voice AI platform plus ten public servers from the official MCP registry): Resource Gateway, Tool Orchestrator, Stateful Session Server, Proxy Aggregator, and Domain-Specific Adapter. Each pattern is described in the structured form of Gamma et al.: context, problem, solution, and consequences. We also document four anti-patterns and a set of cross-cutting concerns around authentication, versioning, and observability. The quantitative evaluation contributes three measurements: inter-rater reliability of the taxonomy across two independent LLM raters on 54 held-out servers (Cohen's kappa = 0.76), which also localizes three pattern-boundary ambiguities; transport overhead measured end-to-end on loopback and modeled for cross-host paths; and a tool-count study showing tool-selection accuracy drops below 90% between 10 and 15 tools per context for Claude Haiku 4.5 and between 20 and 30 tools for Sonnet 4. Code, corpus, and prompts are released as a replication package.

Research agents explain planned experiments. We measure predictive credit with paired forecasts sharing an intervention, forecaster, and outcome while varying description, matched explanation, and donor context. Five checks track commitment, delivery, predictive gain, alignment, and known-signal uptake. Across 336 prospective states in controlled learning, 12 Tox21 endpoints, and 24 OpenML tasks, v5's frozen credit decision was inconclusive. Tox21's preregistered ROC AUC interval-score harm test was unmet (D-M=-.0026, 95 percent interval [-.0174, .0104]); OpenML's joint formation, point-equivalence, and repeatability rule was unmet. Matched point-accuracy gains over description remained unconfirmed, and Tox21/OpenML seed-donor intervals spanned zero. Under requested DeepSeek V4 Pro, matched and donor cards reduced secondary Tox21 drift by 64.5 and 59.1 percent. A DeepSeek V4 Flash replay raised matched point MAE from .01823 to .02020 and missed matched-donor interval-score equivalence. OpenML full-card assignment widened nominal 80 percent intervals by 21 percent, with 49.3 percent coverage versus 51.4 percent for description and content in 66/144 cards. Direct-text Flash delivered all 144 notes without detectable matched point-accuracy gain. A researcher-authored mechanism positive control lowered point MAE by 2.60 percentage points versus description. The protocol measures predictive credit for research-agent benchmarks and scientific forecasting; natural-explanation credit remained unconfirmed at the tested donor resolutions.

Many language tasks have no single answer that can be checked automatically. Rubrics provide criteria for judging responses to these tasks. For reinforcement learning, the resulting verdicts must be combined into a scalar reward. A common approach sums the points assigned to satisfied criteria. Distinct verdict patterns can thus receive the same reward, and the fixed points encode how much each criterion should count, not how strongly its verdict distinguishes the current rollouts. Beyond this aggregation problem, judging the full rubric needs more judge requests as the criterion count grows. To address these limitations, Rubric Response Theory (RRT) measures quality and selects criteria when rubric criteria are monotone indicators of a shared target. Rather than adding assigned points, RRT uses a two parameter item response model that treats the verdict pattern as evidence about scalar quality specific to the rubric. Under this model, its likelihood score maximizes the local signal-to-noise ratio for quality. Its Response Parameter Network (RPN) reads the prompt and criterion text to predict criterion difficulty and discrimination. As the policy distribution changes during training, RRT uses online expectation maximization to update the RPN from current rollout verdicts. With Qwen3.5-4B as the policy, RRT's macro criterion score across Medical, Science, Rubrics as Rewards Science, and RubricBench is 1.7 points above that of group relative policy optimization (GRPO). On hard and very hard criteria in Medical and Science, RRT gains 2.8 to 5.6 points over GRPO. At half the criterion budget, adaptive Fisher selection with a frozen RPN keeps the macro criterion score across four datasets within 0.1 points of GRPO with full judging. These results show RRT can reduce judge requests while remaining competitive with GRPO.

Split language models keep input text on a client but send intermediate representations to a server. Attacks can reconstruct text from these messages. Comparing privacy evaluations is difficult when attackers see different messages, know different model weights or use different datasets and recovery measures. We present LLM-Attacker, a framework that records these choices and lets researchers check, compare and extend an evaluation. Compatible attacks reuse recorded messages, and their predictions are saved for independent scoring. The scoring rule requires a linked undefended reference before a defence can pass. Reports also identify attacks that remain untested. In a controlled GPT-2 experiment, adding gradients returned during training increased token recovery from 94.20% to 97.38% and exact-document recovery from 13.71% to 37.77%. The 95% interval for the token gain lay above the two-percentage-point threshold fixed before testing. This experiment is shared with a companion study; here, it tests how the evaluation responds to a change in attacker access. Further studies measure how corpus choice and compression defences affect recovery, compare an adapted published attack with its reference implementation, and test the measurement code. The framework reports implementation checks separately from measured attack effectiveness. Each privacy claim should state the attacker, dataset and scoring rule, report its controls and identify untested attacks.

In machine learning (ML), a widespread adage is that the area under the precision-recall curve (AUPRC) is a superior metric for model comparison to the area under the receiver operating characteristic (AUROC) for binary classification tasks with class imbalance. This paper challenges this notion through novel mathematical analysis, illustrating that AUROC and AUPRC can be concisely related in probabilistic terms. We demonstrate that AUPRC, contrary to popular belief, is not superior in cases of class imbalance and might even be a harmful metric, given its inclination to unduly favor model improvements in subpopulations with more frequent positive labels. This bias can inadvertently heighten algorithmic disparities. Prompted by these insights, a thorough review of existing ML literature was conducted, utilizing large language models to analyze over 1.5 million papers from arXiv. Our investigation focused on the prevalence and substantiation of the purported AUPRC superiority. The results expose a significant deficit in empirical backing and a trend of misattributions that have fuelled the widespread acceptance of AUPRC's supposed advantages. Our findings represent a dual contribution: a significant technical advancement in understanding metric behaviors and a stark warning about unchecked assumptions in the ML community. All experiments are accessible at https://github.com/mmcdermott/AUC_is_all_you_need.

In existing joint detection and tracking methods, pairwise relational features are used to match previous tracklets to current detections. However, the features may not be discriminative enough for a tracker to identify a target from a large number of detections. Selecting only high-scored detections for tracking may lead to missed detections whose confidence score is low. Consequently, in the online setting, this results in disconnections of tracklets which cannot be recovered. In this regard, we present Sparse Graph Tracker (SGT), a novel online graph tracker using higher-order relational features which are more discriminative by aggregating the features of neighboring detections and their relations. SGT converts video data into a graph where detections, their connections, and the relational features of two connected nodes are represented by nodes, edges, and edge features, respectively. The strong edge features allow SGT to track targets with tracking candidates selected by top-K scored detections with large K. As a result, even low-scored detections can be tracked, and the missed detections are also recovered. The robustness of K value is shown through the extensive experiments. In the MOT16/17/20 and HiEve Challenge, SGT outperforms the state-of-the-art trackers with real-time inference speed. Especially, a large improvement in MOTA is shown in the MOT20 and HiEve Challenge. Code is available at https://github.com/HYUNJS/SGT.

We present a reproducible pipeline for mapping Common Vulnerabilities and Exposures (CVEs) to MITRE ATT&CK Enterprise techniques from free-text vulnerability descriptions. Rather than relying on the CWE->CAPEC->ATT&CK derivation chain, whose table-expansion artifacts we quantify, we train a multi-label classifier on a curated gold dataset of 1,207 CVEs from expert MITRE Center for Threat-Informed Defense mappings. The resulting model approximately doubles recall@5 compared with a zero-shot embedding-similarity baseline and improves every ranking metric. We then investigate whether LLM-assisted labeling can extend the gold dataset. Initial experiments suggest contradictory conclusions: a single run indicates degraded performance, while averaging over five random seeds suggests a small gain. However, an independent replication and an expansion-size study (100--984 additional CVEs) show that the apparent improvement is an evaluation artifact. LLM-generated labels, with approximately 0.39 agreement with expert annotations, provide no reliable improvement at any expansion size and reduce rare-technique coverage at around 1,000 added CVEs (macro-F1 decreases by 0.04). The root cause is evaluation noise. Selecting checkpoints on a small test split effectively maximizes over many noisy evaluations, producing recall@5 differences of up to 0.05 between otherwise identical runs. Using a corrected protocol based on validation-split checkpoint selection, the gold-only model achieves recall@5 of (0.673 \pm 0.019), and repeating the decisive experiment confirms the null result for LLM expansion. A final scaling study shows that additional expert-curated data consistently improves performance, whereas LLM-labeled data does not, indicating that the classifier is limited by label quality rather than dataset size. All datasets, models, code, and training logs are publicly released.

A typical pipeline for multi-object tracking (MOT) is to use a detector for object localization, and following re-identification (re-ID) for object association. This pipeline is partially motivated by recent progress in both object detection and re-ID, and partially motivated by biases in existing tracking datasets, where most objects tend to have distinguishing appearance and re-ID models are sufficient for establishing associations. In response to such bias, we would like to re-emphasize that methods for multi-object tracking should also work when object appearance is not sufficiently discriminative. To this end, we propose a large-scale dataset for multi-human tracking, where humans have similar appearance, diverse motion and extreme articulation. As the dataset contains mostly group dancing videos, we name it "DanceTrack". We expect DanceTrack to provide a better platform to develop more MOT algorithms that rely less on visual discrimination and depend more on motion analysis. We benchmark several state-of-the-art trackers on our dataset and observe a significant performance drop on DanceTrack when compared against existing benchmarks. The dataset, project code and competition server are released at: https://github.com/DanceTrack.

As Large Language Models (LLMs) saturate elementary benchmarks, the research frontier has shifted from generation to the reliability of automated evaluation. We demonstrate that standard "LLM-as-a-Judge" protocols suffer from a systematic Alignment Gap when applied to upper-undergraduate to early graduate level mathematics. To quantify this, we introduce QEDBench, the first large-scale dual-rubric alignment benchmark to systematically measure alignment with human experts on university-level math proofs by contrasting course-specific rubrics against expert common knowledge criteria. By deploying a dual-evaluation matrix (7 judges x 5 solvers) against 1,000+ hours of human evaluation, we reveal that certain frontier evaluators like Claude Opus 4.5, DeepSeek-V3, Qwen 2.5 Max, and Llama 4 Maverick exhibit significant positive bias (up to +0.18, +0.20, +0.30, +0.36 mean score inflation, respectively). Furthermore, we uncover a critical reasoning gap in the discrete domain: while Gemini 3.0 Pro achieves state-of-the-art performance (0.91 average human evaluation score), other reasoning models like GPT-5 Pro and Claude Sonnet 4.5 see their performance significantly degrade in discrete domains. Specifically, their average human evaluation scores drop to 0.72 and 0.63 in Discrete Math, and to 0.74 and 0.50 in Graph Theory. In addition to these research results, we also release QEDBench as a public benchmark for evaluating and improving AI judges. Our benchmark is publicly published at https://github.com/qqliu/Yale-QEDBench.

Evaluation has been central to advances in AI, and it is equally critical to autonomous discovery. Yet scaling autonomous discovery to real-world engineering challenges faces a fundamental bottleneck: the cost of evaluating progress. In real settings, a single true evaluation, such as a full benchmark or serving trace, can take hours, so a search can afford only a handful of true evaluations. We present DuoDiscover, a search system that jointly discovers what to optimize and how to evaluate progress efficiently, through three mechanisms. First, each worker in a fleet of coding agents co-discovers an optimization target and a fast, tailored proxy evaluator, which we call a yardstick. The agent then hill-climbs the yardstick rather than the true evaluation. Second, since search can exploit gaps between a yardstick and the true objective, fidelity tests continually recheck each yardstick's ranking and invalidate any that drift. Third, the fleet is bottom-up, with no central decomposition of the problem. A lightweight orchestrator spawns and terminates agents, briefs running ones, and seeds new agents with others' confirmed work. DuoDiscover enables autonomous optimization from software to chip architecture. We demonstrate DuoDiscover on two repository-level benchmarks and four GPU and TPU kernel suites covering hundreds of kernels. On SWE-fficiency, it doubles the average speedup of human experts. On SOL-Exec, it ranks first on 121 of 235 kernels and in the top three on 212. On the heavily optimized vLLM and SQLite engines, DuoDiscover achieves 1.92times and 1.38times gains over the base code. On a CPU microarchitecture optimization task with evaluations of up to 28 hours, DuoDiscover surpasses both AlphaEvolve-based ArchAgent v2 and the best human design, with 1.65times ArchAgent v2's gain in 16.7 days versus more than 40 days, advancing the frontier of autonomous chip design.

Route-planning agents powered by large language models (LLMs) have emerged as a promising paradigm for supporting everyday human mobility through natural language interaction and tool-mediated decision making. However, systematic evaluation in real-world mobility settings is hindered by diverse routing demands, non-deterministic mapping services, and limited reproducibility. In this study, we introduce MobilityBench, a scalable benchmark for evaluating LLM-based route-planning agents in real-world mobility scenarios. MobilityBench is constructed from large-scale, anonymized real user queries collected from Amap and covers a broad spectrum of route-planning intents across multiple cities worldwide. To enable reproducible, end-to-end evaluation, we design a deterministic API-replay sandbox that eliminates environmental variance from live services. We further propose a multi-dimensional evaluation protocol centered on outcome validity, complemented by assessments of instruction understanding, planning, tool use, and efficiency. Using MobilityBench, we evaluate multiple LLM-based route-planning agents across diverse real-world mobility scenarios and provide an in-depth analysis of their behaviors and performance. Our findings reveal that current models perform competently on Basic information retrieval and Route Planning tasks, yet struggle considerably with Preference-Constrained Route Planning, underscoring significant room for improvement in personalized mobility applications. We publicly release the benchmark data, evaluation toolkit, and documentation at https://github.com/AMAP-ML/MobilityBench .

LLM search agents increasingly rely on tools at inference time, but their trajectories are often constrained by hard limits on both tool calls and generated tokens. Under such dual budgets, better answers require not only stronger models, but also explicit control over which search action should receive the next budget unit and when the accumulated evidence is sufficient to commit a final answer. We study this problem in multi-hop question answering (QA) and formulate it as two-stage inference-time budget control. At search time, our controller assigns each feasible action a task-level Value-of-Information (VOI) score, defined as an operational estimate of marginal task value per unit budget under the current search state and remaining dual budget, and uses this score to choose among retrieval, decomposition, and answer commitment. After search, a selective evidence-grounded finalizer compares the trajectory answer with a refined candidate and rewrites only when the residual error appears to be a low-risk answer-form error. Across four multi-hop QA benchmarks, three LLM backbones, and four budget levels, the method yields positive aggregate gains over four audited baselines under the same hard dual-budget protocol. Ablations show that search-time budget control, especially budget-dependent penalty, provides the main performance gain, while answer-time control helps mainly when the retrieval path is already adequate. These results suggest that inference-time budget control for LLM search agents should govern both how budget is spent during search and how the final answer is committed.

An AI control protocol is a plan for usefully deploying AI systems that aims to prevent an AI from intentionally causing some unacceptable outcome. This paper investigates how well AI systems can generate and act on their own strategies for subverting control protocols whilst operating statelessly (without shared memory between contexts). To do this, an AI system may need to reliably generate optimal plans in each context, take actions with well-calibrated probabilities, and coordinate plans with other instances of itself without communicating. We develop Subversion Strategy Eval, a suite of eight environments, covering a range of protocols and strategic capabilities, and six sets of affordances that help isolate individual capabilities. We implement the evaluation in Inspect-AI and release it open-source. We evaluate Claude 3.5 models, including helpful-only versions, as well as OpenAI reasoning models. None of the models demonstrate substantial capability in strategizing to subvert control protocols statelessly. However, providing models with additional affordances, such as the ability to share a plan between contexts, can substantially improve performance. We hope our evaluations can act as a leading indicator for when models are capable of subverting control protocols and also relax the worst-case assumption of perfect strategic ability in AI control evaluations.

Recent advances in text-to-speech (TTS) technology have enabled systems to produce human-indistinguishable speech, bringing benefits across accessibility, content creation, and human-computer interaction. However, current evaluation practices are increasingly inadequate for capturing the full range of capabilities, limitations, and societal implications. This position paper introduces the concept of Responsible Evaluation and argues that it is essential and urgent for the next phase of TTS development, structured through three progressive levels: (1) ensuring the faithful and accurate reflection of a model's true capabilities, with more robust, discriminative, and comprehensive objective and subjective scoring methodologies; (2) enabling comparability, standardization, and transferability through standardized benchmarks, transparent reporting, and transferable evaluation metrics; and (3) assessing and mitigating ethical risks associated with forgery, misuse, privacy violations, and security vulnerabilities. Through this concept, we critically examine current evaluation practices, identify systemic shortcomings, and propose actionable recommendations. We hope this concept of Responsible Evaluation will foster more trustworthy and reliable TTS technology and guide its development toward ethically sound and societally beneficial applications.

Text-to-Audio-Video (T2AV) generation aims to synthesize temporally coherent video and semantically synchronized audio from natural language, yet its evaluation remains fragmented, often relying on unimodal metrics or narrowly scoped benchmarks that fail to capture cross-modal alignment, instruction following, and perceptual realism under complex prompts. To address this limitation, we present T2AV-Compass, a unified benchmark for comprehensive evaluation of T2AV systems, consisting of 500 diverse and complex prompts constructed via a taxonomy-driven pipeline to ensure semantic richness and physical plausibility. Besides, T2AV-Compass introduces a dual-level evaluation framework that integrates objective signal-level metrics for video quality, audio quality, and cross-modal alignment with a subjective MLLM-as-a-Judge protocol for instruction following and realism assessment. Extensive evaluation of 11 representative T2AVsystems reveals that even the strongest models fall substantially short of human-level realism and cross-modal consistency, with persistent failures in audio realism, fine-grained synchronization, instruction following, etc. These results indicate significant improvement room for future models and highlight the value of T2AV-Compass as a challenging and diagnostic testbed for advancing text-to-audio-video generation.

Pancreatic ductal adenocarcinoma (PDAC) segmentation on contrast-enhanced CT is inherently ambiguous: inter-rater disagreement among experts reflects genuine uncertainty rather than annotation noise. Standard deep learning approaches assume a single ground truth, producing probabilistic outputs that can be poorly calibrated and difficult to interpret under such ambiguity. We present TwinTrack, a framework that addresses this gap through post-hoc calibration of ensemble segmentation probabilities to the empirical mean human response (MHR) -the fraction of expert annotators labeling a voxel as tumor. Calibrated probabilities are thus directly interpretable as the expected proportion of annotators assigning the tumor label, explicitly modeling inter-rater disagreement. The proposed post-hoc calibration procedure is simple and requires only a small multi-rater calibration set. It consistently improves calibration metrics over standard approaches when evaluated on the MICCAI 2025 CURVAS-PDACVI multi-rater benchmark.

Deep learning models perform poorly when domain shifts exist between training and test data. Test-time adaptation (TTA) is a paradigm to mitigate this issue by adapting pre-trained models using only unlabeled test samples. However, existing TTA methods can fail under severe domain shifts, while recent active TTA approaches requiring full-class labels are impractical due to high labeling costs. To address this issue, we introduce a new setting of TTA with binary feedback. This setting uses a few binary feedback inputs from annotators to indicate whether model predictions are correct, thereby significantly reducing the labeling burden of annotators. Under the setting, we propose BiTTA, a novel dual-path optimization framework that leverages reinforcement learning to balance binary feedback-guided adaptation on uncertain samples with agreement-based self-adaptation on confident predictions. Experiments show BiTTA achieves 13.3%p accuracy improvements over state-of-the-art baselines, demonstrating its effectiveness in handling severe distribution shifts with minimal labeling effort. The source code is available at https://github.com/taeckyung/BiTTA.

ATANT v1.0 (arXiv:2604.06710) defined continuity as a system property with 7 required properties and introduced a 10-checkpoint, LLM-free evaluation methodology validated on a 250-story corpus. Since publication, a recurring reviewer and practitioner question has concerned not the framework itself but its relationship to a wider set of memory evaluations: LOCOMO, LongMemEval, BEAM, MemoryBench, Zep's evaluation suite, Letta/MemGPT's evaluations, and RULER. This companion paper, v1.1, does not modify the v1.0 standard. It closes a related-work gap that v1.0 left brief under page limits. We show by structural analysis that none of these benchmarks measures continuity as defined in v1.0: of the 7 required properties, the median existing eval covers 1 property, the mean covers 0.43 when partial credit is scored at 0.5, and no eval covers more than 2. We provide a cell-by-cell property-coverage matrix, identify methodological defects specific to each benchmark (including an empty-gold scoring bug in the LOCOMO reference implementation that renders 23% of its corpus unscorable by construction), and publish our reference implementation's LOCOMO score (8.8%) alongside the structural reason that number is uninformative about continuity. We publish our 8.8% LOCOMO score alongside our 96% ATANT cumulative-scale score as a calibration pair: the 87-point divergence is evidence that the two benchmarks measure different properties, not that one system is an order of magnitude better than another. The position v1.1 takes is not adversarial: each benchmark measures a real capability. The claim is that none of them can adjudicate continuity, and conflating them with continuity evaluation has led the field to under-invest in the properties v1.0 names.

Retrieval-augmented generation (RAG) grounds answers in retrieved passages, yet relevance does not guarantee sufficiency: a topical passage may still fail to justify the answer. We study evidence sufficiency verification for selective RAG answering, in which a verifier receives a question, a candidate answer, and retrieved evidence and decides whether the evidence supports, refutes, or is insufficient for the answer, answering only when support is established. We present SURE-RAG, an aggregation protocol that treats evidence sufficiency as a set-level property: missing hops and unresolved conflicts cannot be detected by scoring passages independently. A shared claim-evidence verifier produces a local relation distribution for each (claim, passage) pair, which SURE-RAG aggregates into four interpretable answer-level feature blocks (coverage, relation strength, uncertainty, and retrieval), producing a three-way decision and an auditable selective score. We evaluate on HotpotQA-RAG v3, a controlled multi-hop benchmark, under an artifact-aware protocol (shortcut baselines, counterfactual swaps, no-oracle checks, and GPT-4o audits). Calibrated SURE-RAG attains 0.9075 Macro-F1 (raw 0.8951 +/- 0.0069), well above DeBERTa mean-pooling (0.6516) and a GPT-4o judge (0.7284), and on par with a strong concat cross-encoder (0.8888 +/- 0.0109) while remaining fully auditable. At 30% coverage, risk falls from 0.2588 to 0.1642, a 37% relative reduction. As a boundary-mapping experiment, we contrast SURE-RAG with GPT-4o on HaluBench unsafe detection: the ranking reverses (0.3343 vs. 0.7389 unsafe-F1), indicating that controlled sufficiency verification and natural hallucination detection are distinct problems.

Tool-using AI agents are increasingly deployed across enterprise software systems, yet widely used benchmarks primarily evaluate nominal task completion, conflating baseline planning competence with operational fault recovery. We introduce UndoBench, a benchmark spanning 36 base workflows and 36 fault scenarios across 8 enterprise domains, decoupling task competence from recovery capability via counterfactual paired trials under identical seeds alongside wire-level effect-history and environment-state oracles. On 12 held-out TEST workflows across two open-weight models, two frameworks, and three recovery paradigms (5,760 executions / 2,880 paired trials) in the frozen lost-acknowledgment study, nominal competence reached 83.54% while conditional recovery success rate (CRSR) fell to 46.72%, with naive retry producing duplicate external effects in 53.33% of trials. Extensions to commercial API models reproduced this competence-recovery separation. Evaluations across complementary execution boundaries show that recovery is phase-dependent: before mutation, methods perform similarly without duplicate effects among capable trials; during partial mutation, naive retry, per-call idempotency, and zero-privilege journaling collapse on the evaluated composite workflows; after commit but before acknowledgment, verification and server-side idempotency substantially improve safety. These findings demonstrate that evaluating nominal completion alone masks critical, phase-dependent recovery vulnerabilities in autonomous agents.

Deploying models on target domain data subject to distribution shift requires adaptation. Test-time training (TTT) emerges as a solution to this adaptation under a realistic scenario where access to full source domain data is not available and instant inference on target domain is required. Despite many efforts into TTT, there is a confusion over the experimental settings, thus leading to unfair comparisons. In this work, we first revisit TTT assumptions and categorize TTT protocols by two key factors. Among the multiple protocols, we adopt a realistic sequential test-time training (sTTT) protocol, under which we further develop a test-time anchored clustering (TTAC) approach to enable stronger test-time feature learning. TTAC discovers clusters in both source and target domain and match the target clusters to the source ones to improve generalization. Pseudo label filtering and iterative updating are developed to improve the effectiveness and efficiency of anchored clustering. We demonstrate that under all TTT protocols TTAC consistently outperforms the state-of-the-art methods on six TTT datasets. We hope this work will provide a fair benchmarking of TTT methods and future research should be compared within respective protocols. A demo code is available at https://github.com/Gorilla-Lab-SCUT/TTAC.

An AI scientist should not grade its own homework. Yet in the systems we examined, the agent that reviews the work usually comes from the same model family as the agent that produced it, or at least from the same vendor. Model evaluators are known to favour their own generations. Whether models trained alike also share blind spots is a conjecture, not a settled finding, but if they do, the reviewer inherits the author's. The record of what was flagged and what was waved through often sits in platform logs that nobody outside can replay. We present CrossAudit, a protocol for supervising autonomous research pipelines. It rests on three commitments. Each increment of work is audited by an agent from a different vendor against a rulebook a human wrote and versioned. Reports, verdicts, disputes and rulings are git commits, so the supervision history can be re-read and cited; raw model exchanges are not yet part of that record. Scripted checks run before any model does. Advisory judgement never gates the pipeline: a model blocks only by citing a rule, and no model may waive a deterministic failure. Blockers that survive a bounded number of revision rounds go to a person. We state the protocol as eight invariants. We describe a reference implementation built from GitHub Actions and a few hundred lines of Python, and report a live deployment of a closely related variant in a computational-chemistry pipeline. We also ran a seeded-defect trial (30 increments, 43 seeded defects, one run per configuration). A cross-vendor audit of our own repository then voided its blinding. We adopt that audit's findings and report the corrected results. The trial shows that two vendors read the same rulebook differently. It does not show that either is better. The strongest evidence here is the committed, uncontrolled record of cross-vendor audits of this paper itself.

Teaching large language models (LLMs) to be faithful in the provided context is crucial for building reliable information-seeking systems. Therefore, we propose a systematic framework, CANOE, to improve the faithfulness of LLMs in both short-form and long-form generation tasks without human annotations. Specifically, we first synthesize short-form question-answering (QA) data with four diverse tasks to construct high-quality and easily verifiable training data without human annotation. Also, we propose Dual-GRPO, a rule-based reinforcement learning method that includes three tailored rule-based rewards derived from synthesized short-form QA data, while simultaneously optimizing both short-form and long-form response generation. Notably, Dual-GRPO eliminates the need to manually label preference data to train reward models and avoids over-optimizing short-form generation when relying only on the synthesized short-form QA data. Experimental results show that CANOE greatly improves the faithfulness of LLMs across 11 different downstream tasks, even outperforming the most advanced LLMs, e.g., GPT-4o and OpenAI o1.

SCTP is a transport protocol offering features such as multi-homing, multi-streaming, and message-oriented delivery. Its two main implementations were subjected to conformance tests using the PacketDrill tool. Conformance testing is not exhaustive and a recent vulnerability (CVE-2021-3772) showed SCTP is not immune to attacks. Changes addressing the vulnerability were implemented, but the question remains whether other flaws might persist in the protocol design. We study the security of the SCTP design, taking a rigorous approach rooted in formal methods. We create a formal Promela model of SCTP, and define 10 properties capturing the essential protocol functionality based on its RFC specification and consultation with the lead RFC author. Then we show using the Spin model checker that our model satisfies these properties. We define 4 attacker models - Off-Path, where the attacker is an outsider that can spoof the port and IP of a peer; Evil-Server, where the attacker is a malicious peer; Replay, where an attacker can capture and replay, but not modify, packets; and On-Path, where the attacker controls the channel between peers. We modify an attack synthesis tool designed for transport protocols, Korg, to support our SCTP model and four attacker models. We synthesize 14 unique attacks using the attacker models - including the CVE vulnerability in the Off-Path attacker model, 4 attacks in the Evil-Server attacker model, an opportunistic ABORT attack in the Replay attacker model, and eight connection manipulation attacks in the On-Path attacker model. We show that the proposed patch eliminates the vulnerability and does not introduce new ones according to our model and protocol properties. Finally, we identify and analyze an ambiguity in the RFC, which we show can be interpreted insecurely. We propose an erratum and show that it eliminates the ambiguity.

Competition retrospectives are useful when they explain what a leaderboard measured, how hidden evaluation changed conclusions, and which design patterns were rewarded. We revisit the CODS 2025 challenge, a privacy-aware Codabench competition on industrial multi-agent orchestration built on . We combine final rank sheets, a 300-submission server log, 149-team registrations, best-submission exports, the organizer winners report, the companion system paper, and verified planning-track source trees. Five results stand out. First, the public planning leaderboard saturates at 72.73\%, and richer prompts do not improve that peak. Second, hidden evaluation changes the story: public and private scores correlate moderately in planning (r{=}0.69) but negatively in execution (r{=}{-}0.13), with several 45.45\% public execution systems reaching 63.64\% on the hidden set. Third, the term is numerically almost inert in the official composite -- combined on a 0--1 scale with 0--100 percentage scores, it contributes at most 0.05 points per track, and rescaling would swap the top two teams. Fourth, the competition is operationally account-based but substantively team-based: 149 registered teams reduce to 24 with non-zero public scores and 11 fully ranked, while 52.3\% of deduplicated registrations list multiple usernames. Fifth, successful execution methods mostly improve guardrails -- response selection, contamination cleanup, fallback, and context control -- rather than novel agent architectures. These findings identify which behaviors the evaluation rewarded, and motivate scale-aware composites, skill-level diagnostics, and versioned artifact release.

Current alignment evaluation mostly measures whether models encode dangerous concepts and whether they refuse harmful requests. Both miss the layer where alignment often operates: routing from concept detection to behavioral policy. We study political censorship in Chinese-origin language models as a natural experiment, using probes, surgical ablations, and behavioral tests across nine open-weight models from five labs. Three findings follow. First, probe accuracy alone is non-diagnostic: political probes, null controls, and permutation baselines can all reach 100%, so held-out category generalization is the informative test. Second, surgical ablation reveals lab-specific routing. Removing the political-sensitivity direction eliminates censorship and restores accurate factual output in most models tested, while one model confabulates because its architecture entangles factual knowledge with the censorship mechanism. Cross-model transfer fails, indicating that routing geometry is model- and lab-specific. Third, refusal is no longer the dominant censorship mechanism. Within one model family, hard refusal falls to zero while narrative steering rises to the maximum, making censorship invisible to refusal-only benchmarks. These results support a three-stage descriptive framework: detect, route, generate. Models often retain the relevant knowledge; alignment changes how that knowledge is expressed. Evaluations that audit only detection or refusal therefore miss the routing mechanism that most directly determines behavior.

As language models shift from single-shot answer generation toward multi-step reasoning that retrieves and consumes evidence mid-inference, evaluating the role of individual retrieved items becomes more important. Existing RAG evaluation typically targets final-answer quality, citation faithfulness, or answer-level attribution, but none of these directly targets the intervention-based, per-evidence-item utility view we study here. We introduce CUE-R, a lightweight intervention-based framework for measuring per-evidence-item operational utility in single-shot RAG using shallow observable retrieval-use traces. CUE-R perturbs individual evidence items via REMOVE, REPLACE, and DUPLICATE operators, then measures changes along three utility axes (correctness, proxy-based grounding faithfulness, and confidence error) plus a trace-divergence signal. We also outline an operational evidence-role taxonomy for interpreting intervention outcomes. Experiments on HotpotQA and 2WikiMultihopQA with Qwen-3 8B and GPT-5.2 reveal a consistent pattern: REMOVE and REPLACE substantially harm correctness and grounding while producing large trace shifts, whereas DUPLICATE is often answer-redundant yet not fully behaviorally neutral. A zero-retrieval control confirms that these effects arise from degradation of meaningful retrieval. A two-support ablation further shows that multi-hop evidence items can interact non-additively: removing both supports harms performance far more than either single removal. Our results suggest that answer-only evaluation misses important evidence effects and that intervention-based utility analysis is a practical complement for RAG evaluation.

Generating step-by-step "how-to" procedures is a key LLM capability: how-to advice is commonly requested in chatbots, and step-by-step planning is critical for reasoning over complex tasks. Yet, measuring and improving procedural validity at scale on real-world tasks remains challenging and understudied. To address this, we introduce How2Everything, a scalable framework to evaluate and improve goal-conditioned procedure generation. Our framework includes How2Mine, which mines 351K procedures from 980K web pages across 14 topics and readily scales to larger corpora. From this pool we build How2Bench, a 7K-example evaluation set balanced across topics. To reliably score model outputs, we develop How2Score, an evaluation protocol that uses an LLM judge to detect whether a generation contains any critical failure that would prevent achieving the goal. For low-cost, reproducible evaluation, we distill a frontier model into an open 8B model, achieving 80.5% agreement with human annotators. How2Bench reveals clear scaling trends across model sizes and training stages, providing signal early in pretraining. Finally, RL using How2Score as a reward improves performance on How2Bench by >10 points across three models without systematic regressions on standard benchmarks, with gains robust to superficial source-document memorization or format compliance. Taken together, How2Everything shows how pretraining web data can support a closed loop of capability evaluation and improvement at scale.

Various algorithms for continual learning (CL) have been designed with the goal of effectively alleviating the trade-off between stability and plasticity during the CL process. To achieve this goal, tuning appropriate hyperparameters for each algorithm is essential. As an evaluation protocol, it has been common practice to train a CL algorithm using diverse hyperparameter values on a CL scenario constructed with a benchmark dataset. Subsequently, the best performance attained with the optimal hyperparameter value serves as the criterion for evaluating the CL algorithm. In this paper, we contend that this evaluation protocol is not only impractical but also incapable of effectively assessing the CL capability of a CL algorithm. Returning to the fundamental principles of model evaluation in machine learning, we propose an evaluation protocol that involves Hyperparameter Tuning and Evaluation phases. Those phases consist of different datasets but share the same CL scenario. In the Hyperparameter Tuning phase, each algorithm is iteratively trained with different hyperparameter values to find the optimal hyperparameter values. Subsequently, in the Evaluation phase, the optimal hyperparameter values is directly applied for training each algorithm, and their performance in the Evaluation phase serves as the criterion for evaluating them. Through experiments on CIFAR-100 and ImageNet-100 based on the proposed protocol in class-incremental learning, we not only observed that the existing evaluation method fail to properly assess the CL capability of each algorithm but also observe that some recently proposed state-of-the-art algorithms, which reported superior performance, actually exhibit inferior performance compared to the previous algorithm.

A longitudinal web crawl is a sequence of partial samples of an evolving URL population. Pairwise containment between two crawls is the standard probe; under a simple urn model of the crawl -- each round samples a fraction of the URLs and replaces a fraction -- it recovers two interpretable rates, per-round survival α and coverage c, but treats the population as uniform and consumes one pair at a time. In this work, we define a formal language for talking about a crawl. We extend this analysis with the discovery curve U(s, T), the cumulative URL footprint over a sliding window of T crawls starting at s, which under the same urn model is also a closed-form function of (α, c). Containment and the discovery curve are then two projections of one process: independent fits agree on (α, c) when the urn is homogeneous, so any disagreement is itself a measurement. Applied to Common Crawl (2020--2025, domain granularity) and to the German Academic Web (GAW, URL granularity), the two projections disagree on both archives, and a two-component urn with a persistent core fraction κ alongside shell parameters (α_partial, c_partial) reconciles the disagreement. A residual on c_partial remains, signaling that the shell itself is not homogeneous; κ is recorded as the scalar entry point to a rank-resolved generalization, which is left to follow-up work. web archive \and crawl coverage \and discovery curve \and urn model \and two-component model \and URL lifetime

We present Music Arena, an open platform for scalable human preference evaluation of text-to-music (TTM) models. Soliciting human preferences via listening studies is the gold standard for evaluation in TTM, but these studies are expensive to conduct and difficult to compare, as study protocols may differ across systems. Moreover, human preferences might help researchers align their TTM systems or improve automatic evaluation metrics, but an open and renewable source of preferences does not currently exist. We aim to fill these gaps by offering *live* evaluation for TTM. In Music Arena, real-world users input text prompts of their choosing and compare outputs from two TTM systems, and their preferences are used to compile a leaderboard. While Music Arena follows recent evaluation trends in other AI domains, we also design it with key features tailored to music: an LLM-based routing system to navigate the heterogeneous type signatures of TTM systems, and the collection of *detailed* preferences including listening data and natural language feedback. We also propose a rolling data release policy with user privacy guarantees, providing a renewable source of preference data and increasing platform transparency. Through its standardized evaluation protocol, transparent data access policies, and music-specific features, Music Arena not only addresses key challenges in the TTM ecosystem but also demonstrates how live evaluation can be thoughtfully adapted to unique characteristics of specific AI domains. Music Arena is available at: https://music-arena.org

This paper reports on the second "Throughput" phase of the Tracking Machine Learning (TrackML) challenge on the Codalab platform. As in the first "Accuracy" phase, the participants had to solve a difficult experimental problem linked to tracking accurately the trajectory of particles as e.g. created at the Large Hadron Collider (LHC): given O(10^5) points, the participants had to connect them into O(10^4) individual groups that represent the particle trajectories which are approximated helical. While in the first phase only the accuracy mattered, the goal of this second phase was a compromise between the accuracy and the speed of inference. Both were measured on the Codalab platform where the participants had to upload their software. The best three participants had solutions with good accuracy and speed an order of magnitude faster than the state of the art when the challenge was designed. Although the core algorithms were less diverse than in the first phase, a diversity of techniques have been used and are described in this paper. The performance of the algorithms are analysed in depth and lessons derived.

Decision models support routing, guardrails, and output verification in LLM systems. However, item accuracy alone cannot establish whether models make correct decisions across controlled variations or assign reliable confidence to their predictions. We introduce DecisionPivot, a diagnostic benchmark comprising 48 tasks and 40,033 items across seven decision roles. Its core track contains 7,800 items from 13 generated task families, organized into contrast groups with computed answers verified by independently implemented solvers. The complementary reference track contains 32,233 items from 35 public test sets, preserving their original evaluation metrics. We evaluate 40 decision models and 14 general models using documented, model-specific protocols, assessing item and contrast-group accuracy, probability ordering, calibration, and empirical selective coverage. Our results reveal substantial differences obscured by aggregate scores: two 27B models separated by only 2.6 points on the reference track achieve contrast-group accuracies of 70.1% and 44.2%. Across 2,400 binary contrast pairs, Perplexity Decider correctly ranks the positive member above the negative member in 92.3% of pairs but answers both members correctly in only 74.1%, exposing a gap between probability ordering and decision correctness. Models with similar item accuracy also differ in the reliability of their high-confidence predictions. On a separate 1,296-item subset, GLM-5.3 achieves 93.5\% contrast-group accuracy at a higher reasoning setting, compared with 81.5% at its lowest effort. By linking errors to pivotal details and difficulty levels, DecisionPivot provides fine-grained diagnostics of decision failures that aggregate evaluation alone cannot reveal.

In LLM serving, reusing the KV cache of prompts across requests is critical for reducing TTFT and serving costs. Cache-affinity scheduling, which co-locates requests with the same prompt prefix to maximize KV cache reuse, often conflicts with load-balancing scheduling that distributes requests evenly across compute instances. Existing schedulers fail to reconcile this trade-off as they operate within a single mapping space, typically applying cache-affinity routing to a subset of requests and load-balanced routing to the rest, without a unified solution to achieve both goals. To address this limitation, we propose DualMap, a dual-mapping scheduling strategy for distributed LLM serving that achieves both cache affinity and load balancing. Its key idea is to map each request to two candidate instances via two independent hash functions based on the request prompt, then intelligently select the better candidate based on current system states. This design increases the likelihood that requests with shared prefixes are co-located, while evenly dispersing distinct prefixes across the cluster via ``the power of two choices''. To make DualMap robust under dynamic and skewed real-world workloads, we incorporate three techniques: 1) SLO-aware request routing, which prioritizes cache affinity but switches to load-aware scheduling when TTFT exceeds the SLO, enhancing load balance without sacrificing cache reuse; 2) hotspot-aware rebalancing, which dynamically migrates requests from overloaded to underloaded instances, mitigating hotspots and rebalancing the system; 3) lightweight dual-hash-ring scaling, which leverages a dual-hash-ring mapping to support fast and low-overhead instance scaling without costly global remapping. Experiments on real-world workloads show that DualMap improves effective request capacity by up to 2.25times under the same TTFT SLO constraints compared with SOTA work.