The study audited seven public educational prediction datasets using four pre‑modeling reliability checks—baseline gap, split instability, null separation, and metadata adequacy under group‑aware holdout. Only three datasets passed all checks; the others failed either group‑aware generalization tests or lacked necessary provenance metadata. The audit revealed that cross‑group fragility, rather than weak iid performance, was the dominant failure mode, and that increasing model complexity did not resolve these structural issues.
By Yan Ma, Lizhuo Zhang
The paper introduces a dataset of 10,015 real scam and spam phone calls collected over 53 days using an active voice‑agent honeypot. Each call is recorded, transcribed, and automatically labeled, yielding 328,869 turn‑level transcripts and 895 hours of audio from 5,665 distinct numbers. The corpus distinguishes between predatory‑but‑legal lead generation and outright scams, with labels validated by human review and technical checks on realism.
By Ethan Traister, Dennis Tsang Ng, Siyu Zhang, Huaiyu Guo, Tommy Duong, Tyler Wu, Yuchen Zhou, Xingyu Shen, Jiaqi Wu, Simiao Ren
WebArxiv is a reproducible benchmark designed to evaluate multimodal web agents on arXiv-related tasks. It consists of 510 static, time‑invariant tasks that require multi‑constraint paper retrieval, fine‑grained content extraction, and cross‑paper comparison, each with a deterministic ground truth. The benchmark highlights challenges for foundation‑model agents, such as over‑reliance on fixed interaction histories, and introduces a lightweight dynamic‑memory mechanism to improve adaptive retrieval and reasoning.
By Zihao Sun, Zijing Shi, Ling Chen
TrackEverything is a 3D point tracker that overcomes the trade‑off between sparse long‑horizon tracking and dense short‑clip tracking by representing videos as persistent 3D scene tracks in world coordinates. It introduces voxel‑based de‑duplication at sliding‑window boundaries, a two‑stage refinement process (endpoint refiner and lightweight trajectory refiner), and a 3D WAFT module that replaces memory‑heavy 4D correlation volumes with efficient feature sampling. The method can track all visible points in videos longer than 1000 frames using only 40 GB of GPU memory, outperforming existing dense trackers on short clips and matching sparse trackers on long sequences.
By Ayush Jain, Sreeharsha Paruchuri, Ishita Gupta, Fan Zhang, Tanner Schmidt, Jakob Engel, Katerina Fragkiadaki, Adam W. Harley
Forecast-Dojo is a replayable environment designed to benchmark and train large language model (LLM) forecasting agents. It integrates resolved prediction‑market questions with dated news, enabling agents to research events and revisit predictions at successive historical dates. The platform includes 1,568 Polymarket events, 18.8 million dated news articles, and supports repeated evaluation, training interactions, and outcome feedback, with evidence that research tools lower Brier scores across 12 tested models, though all models still lag behind historical market forecasts.
By Liqin Ye, Haorui Wang, Fardin Ahmed, Rongzhi Zhang, Yuan He, Ziyuan Lin, Yanbin Yin, Jing Peng, Michael Galarnyk, Sudheer Chava, Chao Zhang
PartHackBench is a new methodology for evaluating partial-credit tool agents that eliminates confounding factors by using a private certifier to ensure that compared trajectories match in both current-state predicate satisfaction and standardized agent attribution. The approach measures score inflation only after certification, and in experiments on 18 held-out tasks it produced matched adversaries for 15 tasks, revealing mean inflation of 0.252 and detecting no strict rollbacks. The study shows that semantic LLM judges are vulnerable to evaluator-targeted attacks, whereas PB-CSTE current-state controls guarantee zero inflation by construction.
By Hongye Yang, Zhihao Xie, Shengjun Xiong
TimeBraid is a family of unified models that combine pretrained language models with pretrained time‑series foundation models using interleaved global residual attention layers. The models inherit instruction following, reasoning, and continuous‑signal perception, fusing both modalities into a shared representation space for understanding and generation. The design focuses on aligning representation spaces, grounding language in temporal structure, balancing understanding with generation, and maintaining stable joint optimization, supported by 2.2 M curated series‑text pairs and 4.9 M instruction‑tuning samples. Across diverse benchmarks, TimeBraid competes with larger general‑purpose and task‑specific models.
By Xinyue Wang, Jiacheng Pang, Kun Zhou, Kexin Zhang, Defu Cao, Fan Feng, Faisal, Songyao Jin, Yan Liu, Biwei Huang
BridgeMem is a new method for temporal knowledge graph forecasting that focuses on pair‑specific transition evidence, adding a residual correction to the log scores of a frozen full‑vocabulary forecaster. It retrieves and encodes prior events between a query actor and candidate, converting them into a likelihood‑ratio correction via a support‑adaptive empirical‑Bayes reader. Across five benchmarks, BridgeMem outperforms nine baselines from 2021–2026, improving filtered MRR and Hits@{1,3,10} metrics by up to 0.0216.
By Zeyan Li, Libing Chen, Shengda Zhuo, Yin Tang, Jianfeng Xu
The paper investigates why latent neural surrogate solvers, which compress physical system dynamics into a lower‑dimensional space, often fail during long‑horizon autoregressive rollouts. It demonstrates that training the latent representation only for reconstruction leads to instability, and proposes a set of training interventions—Koopman operator learning, Hamming noise injection, and multi‑step rollout fine‑tuning—that align the latent space with long‑horizon forecasting. These interventions reduce long‑rollout error by about 40 % and achieve accuracy comparable to full‑resolution models while using far fewer floating‑point operations and GPU memory, enabling stable extrapolation in mesoscale crystal‑plasticity simulations of high‑cycle fatigue.
By Andreas E. Robertson, Ashley T. Lenau, John D. Shimanek, Benjamin A. Jasperson, Vivek Oommen, David L. Damm, Krishna Garikipati, Remi Dingreville
The paper introduces Selective Supervision for Direct-OPD (S$^2$D-OPD), a refinement of Direct On-Policy Distillation that filters out states where the teacher’s policy change is minimal, as measured by the teacher‑reference Jensen‑Shannon divergence. By masking low‑divergence states and keeping only the top 10% of states per response, S$^2$D-OPD improves held‑out accuracy on AIME and HMMT benchmarks across multiple teacher‑student pairs without additional forward passes.
By Yibo Zhao, Zixuan Yang, Yunshi Lan, Xiang Li
TOLA is a diffusion‑based text image super‑resolution method that eliminates iterative image‑text diffusion by using a one‑step latent adaptation framework. It employs a confidence‑weighted text conditioning module to build a reliable semantic condition and a lightweight latent residual correction module to fix structured residual errors, thereby preserving text fidelity. Experiments show TOLA outperforms existing diffusion‑based TSR methods, achieving at least 2.72 dB higher PSNR on the CTR‑TSR‑Test benchmark.
By Yike Xu, Yue Shi, Yong Guo, Jiezhang Cao
The paper presents a new approach for specific emitter identification (SEI) that combines an integrated complex variational mode decomposition algorithm with a temporal convolutional network and a spatial attention mechanism. This method improves feature extraction from complex-valued signals and adaptively emphasizes informative segments, leading to higher identification accuracy. Experiments show the model reaches 96% accuracy using only 10 symbols and no prior knowledge, demonstrating its effectiveness in low‑data scenarios.
By Chenyu Zhu, Zeyang Li, Ziyi Xie, Jie Zhang
The paper presents a unified formalism for proactive robot assistance, organized into three levels, and introduces a framework for unprompted proactive assistance. It demonstrates that offline evaluation overestimates performance and proposes a closed-loop evaluation using an adaptive human model. The authors also present GAP, a method that learns from passive observation to anticipate user goals and act, outperforming prior state-of-the-art methods in closed-loop tests.
By Maithili Patel, Sonia Chernova
The paper proposes using lightweight, calibrated System One decision models—specifically JEV and Laya—to improve autonomous penetration-testing harnesses that rely on large language models (LLMs). It defines four key decision points (finding adjudication, severity recalibration, agent pruning, and confirmation loops) and presents a NeuroSploit case study showing differences in severity distribution, runtime, and grading when using TypeSafe System One. The authors review existing System One specifications, discuss various RL-based training approaches, and introduce Rave, a domain‑adapted model with a proposed training and evaluation framework.
By Joas Antonio dos Santos Barbosa
The paper examines how hallucinations arise in multi-stage video‑understanding agents by aligning existing benchmarks with the stages of temporal grounding, visual observation, and reasoning. It introduces a causal stage‑intervention protocol that isolates each stage while keeping the downstream task constant, revealing that grounding errors dominate downstream hallucinations and that correct region location matters more than precise temporal overlap. The study also shows that current benchmark scores poorly predict causal sensitivity and can fail under distribution shift, advocating for stage‑aware evaluation methods.
By Shuzhi Gong, Fengze Sun, Yuansan Liu
arXiv:2609. 28506v1 Announce Type: new Abstract: TW3Cast is a time-series forecasting system that reaches position 3 of 130 entries on the GIFT-Eval benchmark by mean MASE rank, as of 2026-09-14.
By Nathan Thierry, Andre-Louis Rochet
The paper introduces AdvRole, an adversarial closed‑loop curriculum for training role‑playing agents with large language models. It alternates between an Actor that learns to role‑play and a Rewriter that edits character profiles and dialogue contexts into hard scenarios, using a performance‑gap reward to target the Actor’s weaknesses. Experiments on English, Chinese, and a new multilingual benchmark demonstrate that AdvRole consistently outperforms baseline methods.
By Zheng Zhang, Liu Liu, Qi Chai, Deheng Ye, Peilin Zhao, Mao Zheng, Hao Wang
The paper audits a multilingual affective generation benchmark that uses emoji summaries for Bangla, English, and Hindi sentences. It finds that the benchmark’s conclusions are largely artifacts of the measurement instrument, with no system significantly outperforming another when annotators are treated as random factors. The study shows that annotator identity and output length drive most variance, and proposes a new stable metric called emoji‑affect decodability.
By Fardeen Sadab, Adib Sakhawat
The paper introduces Reinforcement Learning with Verifiable Rewards (RLVR) applied to small search agents, specifically training a Qwen3.5-0.8B model with Group Relative Policy Optimization and an interleaved Wikipedia-search tool on the MuSiQue dataset. Experiments varying reward shapes across three seeds show that RLVR can achieve a 3.8‑fold improvement over an untrained baseline, with the best run reaching a 0.352 average exact match. The study finds that the sparse exact‑match reward, standard in larger models, performs poorly for small models, indicating that reward design must be tailored rather than scaled down from large‑model recipes.
By Gaurisankar Jayadas, Aske Plaat, \'Alvaro Serra-G\'omez, Sandheep P
The paper introduces GRAFT, a Graph-based Faithful sTep-level credit-assignment framework that constructs a trajectory graph from rollout trajectories, recovers node state-values via Bellman iteration, and assigns step-level advantages based on node value differences. It also proposes Graph GAE to further reduce state-value estimation bias. Experiments on multi-turn agentic benchmarks demonstrate consistent improvements over GRPO and other recent agentic RL algorithms.
By Xincheng Yao, Haobo Fu, Weiming Liu, Chongyang Zhang