{ "title": "Awesome Jev research papers", "url": "https://awesomejev.github.io/papers/", "updated": "2026-10-17", "scope": "Curated direct preprints; findings remain author-reported. No experimental reproduction implied.", "papers": [ { "category": "Papers", "title": "JevAgentHarness: Benchmarking, Deploying and Training Decision Models That Share Control with LLMs", "by": "Zhu et al. · 2026-10-15 · v1", "date": "2026-10-15", "url": "https://arxiv.org/abs/2610.21412v1", "en": "Studies fitted confidence thresholds for sharing agent control with decision models.", "zh": "研究通过拟合置信度阈值，让决策模型参与 agent 控制。" }, { "category": "Papers", "title": "System1-Arena: Which Decisions Can Be Left to System 1?", "by": "Chen et al. · 2026-10-15 · v1", "date": "2026-10-15", "url": "https://arxiv.org/abs/2610.21244v1", "en": "A benchmark spanning matched requests and closed-loop tasks. Experiments have not been reproduced here.", "zh": "涵盖匹配请求与闭环任务的基准；本项目尚未复现其实验。" }, { "category": "Papers", "title": "Jev for Scientific Decisions: Evaluating Semantic Choices and Their Consequences", "by": "Deng, Fan, Zhang & Xie · 2026-10-15 · v3", "date": "2026-10-15", "first_submitted": "2026-09-21", "url": "https://arxiv.org/abs/2609.24965v3", "en": "In ten prepared scientific cases, the authors separate semantic selections, computed quantities and final labels; comparison-model errors can be hidden by unchanged final labels.", "zh": "作者在十个预设科学案例中分别评估语义选择、计算得到的数值与最终标签；对比模型的错误可能被不变的最终标签掩盖。" }, { "category": "Papers", "title": "Selective Reasoning for Typed Decision Models: When to Start, Read, and Stop", "by": "Zheng & Chen · 2026-10-14 · v1", "date": "2026-10-14", "url": "https://arxiv.org/abs/2610.19644v1", "en": "Gates reasoning start, candidate-answer reading and generation stopping; evidence is limited to a frozen Qwen backbone and three task families.", "zh": "控制推理启动、候选答案读取与生成停止的时机；证据限于冻结的 Qwen 骨干及三类任务。" }, { "category": "Papers", "title": "Typed Decision Models on Tabular Data", "by": "Zhang, Ye & Zha · 2026-10-14 · v1", "date": "2026-10-14", "url": "https://arxiv.org/abs/2610.19536v1", "en": "Tests classification and verification under content-preserving table rewrites across three task families; confidence can flag instability without proving correctness.", "zh": "在三类表格任务中检验保持内容不变的改写对分类与验证的影响；置信度可提示不稳定，但不证明正确性。" }, { "category": "Papers", "title": "More Choices, Fewer Decisions: Ordinal-Scale Bias in JEV-like Direct-Decision Models", "by": "Gao et al. · 2026-10-14 · v2", "date": "2026-10-14", "first_submitted": "2026-09-30", "url": "https://arxiv.org/abs/2609.38827v2", "en": "Audits ordinal label underuse across selected decision models; finer scales increase compression, while accuracy, final-label utilization and averaged candidate probabilities remain distinct measures.", "zh": "审查所选决策模型对有序标签的使用不足；细化量表会加重压缩，准确率、最终标签利用率和平均候选概率仍是不同测量。" }, { "category": "Papers", "title": "DecisionRobust: Benchmarking Robustness for JEV Style Decision Models Across Text and Vision", "by": "Lan et al. · 2026-10-13 · v1", "date": "2026-10-13", "url": "https://arxiv.org/abs/2610.19118v1", "en": "Text and vision attacks with label-validity checks and held-out evaluation models.", "zh": "结合标签有效性检查与留出评测模型，考察文本和视觉攻击。" }, { "category": "Papers", "title": "Jeva: System One Models from First Principles", "by": "Li & Wang · 2026-10-13 · v1", "date": "2026-10-13", "url": "https://arxiv.org/abs/2610.18773v1", "en": "A decision-model architecture exploring null handling, option permutations, scaling, and caching. Its design is Jeva’s, not Jev’s disclosed internals.", "zh": "探索空答案、选项排列、规模扩展与缓存的决策架构。这是 Jeva 的设计，并非 Jev 已披露的内部实现。" }, { "category": "Papers", "title": "VideoJev: System One Model for Video Understanding and Interaction", "by": "Zhu et al. · 2026-10-13 · v1", "date": "2026-10-13", "url": "https://arxiv.org/abs/2610.18615v1", "en": "Structured video decisions with temporal spans and candidate scores; cached and joint inference reuse visual computation.", "zh": "以时间区间和候选分数表达视频决策，通过缓存与联合推理复用视觉计算。" }, { "category": "Papers", "title": "Reading Right, Acting Wrong: Diagnosing Decision Failures of System One Models in Agent Loops", "by": "Li, Krishnan & Padman · 2026-10-13 · v1", "date": "2026-10-13", "url": "https://arxiv.org/abs/2610.18579v1", "en": "Pairs factual probes with action choices to diagnose decision failures.", "zh": "将事实探针与行动选择配对，诊断决策失败。" }, { "category": "Papers", "title": "Judges Follow the Reference: The Derivation Gap in Verifying Tool-Call Arguments", "by": "Ma et al. · 2026-10-13 · v1", "date": "2026-10-13", "url": "https://arxiv.org/abs/2610.18497v1", "en": "Tool-argument judges including Jev reject some correct derived values without visible references; model-generated references are evaluated on three benchmarks.", "zh": "包括 Jev 在内的工具参数评判模型，在缺少可见参考时会拒绝部分正确推导值；模型生成的参考在三个基准上接受评测。" }, { "category": "Papers", "title": "Same Attack, Different Judge", "by": "Li et al. · 2026-10-13 · v1", "date": "2026-10-13", "url": "https://arxiv.org/abs/2610.17820v1", "en": "A controlled adversarial comparison of Jev, LLM judges, and purpose-built safety classifiers.", "zh": "对 Jev、LLM 评判模型与专用安全分类器进行受控对抗比较。" }, { "category": "Papers", "title": "LLM-as-Jev: LLMs Are Already Jev-Style Decision Models -- When and How to Fine-Tune Them", "by": "Li & Wagle · 2026-10-13 · v3", "date": "2026-10-13", "first_submitted": "2026-10-01", "url": "https://arxiv.org/abs/2610.02076v3", "en": "Compares frozen Qwen decision readouts with optional KL-anchored fine-tuning and budgeted reasoning; reported gains depend on backbone size and routing workload, not disclosure of hosted Jev internals.", "zh": "比较冻结 Qwen 的决策读出、带 KL 锚定的可选微调与限预算推理；所报收益依赖骨干规模和路由任务，并不揭示托管 Jev 内部实现。" }, { "category": "Papers", "title": "Decision Models vs. LLMs as Rubric Judges", "by": "Rao & Callison-Burch · 2026-10-13 · v3", "date": "2026-10-13", "url": "https://arxiv.org/abs/2609.29769v3", "en": "Compares rubric judges and examines correlated errors that can limit cascade improvements.", "zh": "比较评分准则评判模型，考察相关错误如何限制级联系统的收益。" }, { "category": "Papers", "title": "Releasing the Inner Jev: A Simple Recipe for Open Decision Models", "by": "Geuter, Cheng & Alvarez-Melis · 2026-10-12 · v1", "date": "2026-10-12", "url": "https://arxiv.org/abs/2610.17205v1", "en": "Answer-letter readout and supervised LoRA baselines; evaluation includes model-generated labels.", "zh": "答案字母概率读出与监督式 LoRA 基线；评测包含模型生成的标签。" }, { "category": "Papers", "title": "GraphDecisionBench: Benchmarking System One Models on Graph Tasks", "by": "Yang, Zhang & Zhao · 2026-10-12 · v3", "date": "2026-10-12", "first_submitted": "2026-10-05", "url": "https://arxiv.org/abs/2610.06354v3", "en": "Compares graph decisions across fourteen model-interface configurations; local adjacency, graph-text evidence use and sequential solution quality are separate capabilities.", "zh": "比较十四种模型—接口配置的图决策；局部邻接识别、图文证据利用与序列构造解的质量是不同能力。" }, { "category": "Papers", "title": "General Decision Models: Benchmarking and Insights Beyond Jev", "by": "Duan et al. · 2026-10-12 · v2", "date": "2026-10-12", "url": "https://arxiv.org/abs/2610.03935v2", "en": "Bilingual decision evaluation and reliability when local decisions compose into larger systems.", "zh": "双语决策评估，以及局部决策组合成更大系统时的可靠性。" }, { "category": "Papers", "title": "Suggested Is Not Superseded: A Pre-Registered Evaluation of a System-One Model as a Memory-Retirement Gate", "by": "Chen & Le · 2026-10-11 · v1", "date": "2026-10-11", "url": "https://arxiv.org/abs/2610.15710v1", "en": "Jev memory-retirement evaluation separates confirmed updates from suggestions; mixed LLM/human labels, shared errors and selection on a small development set constrain cascade claims.", "zh": "Jev 旧记忆停用评测区分已确认更新与变化暗示；LLM／人工混合标注、共享错误及小型开发集上的选择限制了级联收益主张。" }, { "category": "Papers", "title": "SanSi: A Looped Typed Decision Model for System 1.5 Thinking", "by": "Gan, Lee & Kang · 2026-10-10 · v2", "date": "2026-10-10", "url": "https://arxiv.org/abs/2610.07730v2", "en": "Looped hidden-state computation for typed decisions; additional loops trade compute for accuracy.", "zh": "通过隐状态循环计算输出类型化决策；额外循环以计算成本换取准确率。" }, { "category": "Papers", "title": "JevCascade: Readout Stability and Inference-Time Compute Allocation for Prefill-Only Decision Models", "by": "Li & Chen · 2026-10-10 · v2", "date": "2026-10-10", "url": "https://arxiv.org/abs/2610.07716v2", "en": "Candidate-menu interventions and confidence cascades; ranking stability does not establish probability calibration.", "zh": "研究候选集合干预与置信度级联；排序稳定性并不等同概率校准。" }, { "category": "Papers", "title": "Jev in Medicine: A Benchmark Evaluation", "by": "Madrid-García & Merino-Barbancho · 2026-10-10 · v4", "date": "2026-10-10", "first_submitted": "2026-09-27", "url": "https://arxiv.org/abs/2609.34024v4", "en": "Preliminary medical-benchmark evaluation of Jev 1.13 reports task-dependent accuracy, calibration and weak unanswerable-question detection; it does not establish clinical deployment validity.", "zh": "Jev 1.13 的初步医学基准评测报告了随任务变化的准确率、校准表现及较弱的不可回答问题识别能力；这不能证明临床部署有效性。" }, { "category": "Papers", "title": "Same Scores, Different Decisions: Evaluating JEV and Language Models for Legal Document Understanding", "by": "Zhang et al. · 2026-10-10 · v2", "date": "2026-10-10", "first_submitted": "2026-09-23", "url": "https://arxiv.org/abs/2609.27678v2", "en": "ContractNLI request variations can change which answers are wrong while aggregate accuracy stays unchanged; stable answers may preserve errors. This benchmark does not validate real legal deployment.", "zh": "ContractNLI 请求变化可能改变具体错误答案，而总体准确率保持不变；稳定答案也可能保留错误。该基准不能验证真实法律场景的部署有效性。" }, { "category": "Papers", "title": "Does System One Transfer? Calibrated Typed-Decision Models, Label Budgets, and Translation for Indonesian Regional Languages", "by": "Fadli et al. · 2026-10-09 · v1", "date": "2026-10-09", "first_submitted": "2026-10-09", "url": "https://arxiv.org/abs/2610.14112v1", "en": "Tests Jev and multilingual Laya on Indonesian regional-language tasks; stable expected calibration error can coexist with worse error ranking, while human translation, label budgets and cascade choices change comparisons.", "zh": "在印尼地方语言任务中测试 Jev 与多语言 Laya；预期校准误差稳定仍可伴随错误排序变差，人类翻译、标签预算与级联选择会改变比较条件。" }, { "category": "Papers", "title": "JEV-as-a-Judge: Accept When Confident, Escalate When Unsure", "by": "Li et al. · 2026-10-09 · v5", "date": "2026-10-09", "url": "https://arxiv.org/abs/2609.26550v5", "en": "Studies confidence-based escalation from Jev to a reasoning judge, including limits under distribution shift.", "zh": "研究基于置信度从 Jev 升级到推理评判模型的方法，以及分布变化下的局限。" }, { "category": "Papers", "title": "One Word Opens the Gate: The Option-Channel Attack on Typed Decision Models as Agent Guardrails", "by": "Azizi, Baghaei Potraghloo & Pedram · 2026-10-08 · v1", "date": "2026-10-08", "url": "https://arxiv.org/abs/2610.12292v1", "en": "Tests option-channel attacks on seven open-weight models; reversed decisions can remain confident. The reported experiments do not test hosted Jev.", "zh": "测试七个开放权重模型的选项通道攻击；决策翻转后仍可能高置信。报告中的实验未测试托管 Jev。" }, { "category": "Papers", "title": "Specialized Decision Models vs. General-Purpose LLMs: Benchmarking Jev Across Knowledge, Reasoning, and Multilingual Tasks", "by": "Li et al. · 2026-10-08 · v1", "date": "2026-10-08", "url": "https://arxiv.org/abs/2610.11978v1", "en": "Knowledge, reasoning and multilingual multiple-choice evaluation of Jev; reported rankings and mathematical weaknesses apply to the selected models and benchmarks.", "zh": "在知识、推理和多语言选择题中评测 Jev；报告的排名及数学弱点仅适用于所选模型与基准。" }, { "category": "Papers", "title": "MetaEncoder: Exploring the Limit of Bi-Encoders for Multimodal System One Decision Making with Natural Language Interface", "by": "Cheng et al. · 2026-10-08 · v1", "date": "2026-10-08", "url": "https://arxiv.org/abs/2610.11316v1", "en": "Bi-encoder approach to multimodal decisions and retrieval through natural-language interfaces.", "zh": "通过自然语言接口实现多模态决策与检索的双编码器方法。" }, { "category": "Papers", "title": "JevOut: Natural Context Can Flip Decision Models", "by": "Xu et al. · 2026-10-08 · v2", "date": "2026-10-08", "url": "https://arxiv.org/abs/2609.30243v2", "en": "Optimizes context additions around unchanged questions and options in four decision systems; results concern initially correct decisions, and human validity checks cover a sampled subset of successful attacks.", "zh": "在四种决策系统中围绕不变的题目和选项优化上下文补充；结果针对原本答对的决策，人工有效性检查仅覆盖成功攻击的抽样子集。" }, { "category": "Papers", "title": "Type-Safe Is Not Error-Free: Typed Decision Models Follow the Option Name, Not the Definition Bound to It", "by": "Sun, Xu, Shi & Yang · 2026-10-08 · v3", "date": "2026-10-08", "url": "https://arxiv.org/abs/2609.26758v3", "en": "Tests conflicts between option names and definitions despite structurally valid outputs.", "zh": "检验输出结构合法时，选项名称与定义之间的冲突。" }, { "category": "Papers", "title": "Same-Number Citation Swaps: Stress-Testing Jev as a Financial Evidence Judge", "by": "Xu et al. · 2026-10-06 · v1", "date": "2026-10-06", "url": "https://arxiv.org/abs/2610.08675v1", "en": "Same-number citation swaps probe financial-evidence verification; rejecting wrong roles trades off with accepting equivalent support in a constructed test.", "zh": "以同数值引用替换检验金融证据验证；在构造测试中，识别财务项目错配与接受等价证据之间存在权衡。" }, { "category": "Papers", "title": "Token-Efficient Multi-Agent Collaboration via System One-Guided Computational Division of Labor", "by": "Zhou et al. · 2026-10-06 · v1", "date": "2026-10-06", "first_submitted": "2026-10-06", "url": "https://arxiv.org/abs/2610.08155v1", "en": "Separates bounded multi-agent coordination from LLM reasoning through a controller–reader loop; reported GPT-4o-token and end-to-end latency reductions concern complete framework comparisons on seven benchmarks.", "zh": "通过控制器—读取器循环分离多智能体的有界协调与 LLM 推理；报告的 GPT-4o token 用量和端到端时延下降来自七个基准上的完整框架比较。" }, { "category": "Papers", "title": "JEV versus LLMs: Accuracy, Cost and Calibration on Seven Political Science Replications", "by": "Denney & DiGiuseppe · 2026-10-06 · v2", "date": "2026-10-06", "url": "https://arxiv.org/abs/2610.06625v2", "en": "Studies accuracy, cost, and calibration in seven political-science replication tasks.", "zh": "在七项政治学复现任务中考察准确率、成本与校准。" }, { "category": "Papers", "title": "Benchmarking System One decision models against trained classifiers and language models for automated decision gates", "by": "Rafe & Das · 2026-10-06 · v2", "date": "2026-10-06", "first_submitted": "2026-09-29", "url": "https://arxiv.org/abs/2610.00346v2", "en": "A common-request harness compares typed decision models, generative readouts and trained classifiers; accuracy, selective coverage and out-of-scope acceptance depend on label access and task.", "zh": "用共同请求框架比较类型化决策模型、生成模型读出与训练分类器；准确率、选择性覆盖率和域外接纳取决于是否有任务标签及具体任务。" }, { "category": "Papers", "title": "SoK: Semantic Decision Engines in Network Control Loops", "by": "Li et al. · 2026-10-05 · v1", "date": "2026-10-05", "first_submitted": "2026-10-05", "url": "https://arxiv.org/abs/2610.06425v1", "en": "Surveys semantic engines in network control loops; separates valid decisions, action feasibility, service checks and deadline attainment, with bounded event-model tests rather than production validation.", "zh": "综述网络控制环中的语义引擎，区分决策格式有效、动作可行、服务核查与期限达成；验证来自有界事件模型测试，而非生产部署。" }, { "category": "Papers", "title": "Calibrated to Whom? Persona and Language Effects on Cultural Values in JEV", "by": "Asseri & Asseri · 2026-10-03 · v2", "date": "2026-10-03", "url": "https://arxiv.org/abs/2609.36399v2", "en": "English/Arabic persona-conditioned Jev survey audit; the effects concern simulated respondents in one model, not measurements of real populations.", "zh": "对 Jev 进行英语和阿拉伯语人物设定条件下的问卷审查；观察的是单一模型的模拟受访者，并非对真实人群的测量。" }, { "category": "Papers", "title": "System One Models for Agent Security Decisions", "by": "Liu · 2026-10-03 · v3", "date": "2026-10-03", "url": "https://arxiv.org/abs/2609.33401v3", "en": "Examines reliability, group-specific confident errors, and selective automation in security decisions.", "zh": "研究安全决策中的可靠性、不同群体的高置信错误与选择性自动化。" }, { "category": "Papers", "title": "Batched Speech Decisions Without Decoding: Single-Token Supervision Lets a Frozen LLM Hear Beyond the Transcript", "by": "Jin et al. · 2026-10-02 · v1", "date": "2026-10-02", "first_submitted": "2026-10-02", "url": "https://arxiv.org/abs/2610.02638v1", "en": "DuplexJev connects ASR hidden states to a frozen LLM for batched answer-token readout; connector-specific speech tasks and eight-GPU timing do not establish single-device or live-agent latency.", "zh": "DuplexJev 将 ASR 隐藏状态接入冻结 LLM，通过答案 token 读出批量决策；连接器相关的语音任务结果与八 GPU 计时，不能证明单设备或实时 agent 延迟。" }, { "category": "Papers", "title": "Labels Override Definitions in Jev-Style Typed Decision Models", "by": "Azizi, Baghaei Potraghloo & Pedram · 2026-10-01 · v1", "date": "2026-10-01", "first_submitted": "2026-10-01", "url": "https://arxiv.org/abs/2610.02586v1", "en": "Studies option-label bias in open-weight decision models; prompt-rendering changes with fixed weights implicate the label/definition interface, without testing hosted Jev.", "zh": "研究开放权重决策模型的选项标签偏置；固定权重后改变提示渲染，指向标签／定义接口的影响，未测试托管 Jev。" }, { "category": "Papers", "title": "HakemBench: A Turkish Benchmark of Typed Decisions", "by": "Teke · 2026-10-01 · v1", "date": "2026-10-01", "first_submitted": "2026-10-01", "url": "https://arxiv.org/abs/2610.02293v1", "en": "Turkish typed-decision evaluation reports decision quality, Brier-based probability quality and risk–coverage alongside input probes; most labels are AI-derived, and the lab’s model used feedback from earlier test results.", "zh": "土耳其语类型化决策评测同时报告决策质量、基于 Brier 的概率质量、风险—覆盖及输入探查；多数标签由 AI 产生，实验室自身模型还使用过先前测试结果的反馈。" }, { "category": "Papers", "title": "AudioJev: Direct Audio Decisions with Order-Calibrated Probabilities", "by": "Lv et al. · 2026-10-01 · v1", "date": "2026-10-01", "first_submitted": "2026-10-01", "url": "https://arxiv.org/abs/2610.01293v1", "en": "Learns audio-conditioned candidate probabilities with aligned option-permutation training; order stability and empirical correctness calibration remain different properties.", "zh": "通过按候选身份对齐的选项重排训练学习音频条件概率；顺序稳定性与概率符合实际正确率仍是不同性质。" }, { "category": "Papers", "title": "Beyond Answer Confidence: A Controlled Audit of Self-Knowledge in a Black-Box Decision Model", "by": "Shankaranarayana, Runje & Jannink · 2026-10-01 · v1", "date": "2026-10-01", "url": "https://arxiv.org/abs/2610.01006v1", "en": "Controls information and surface cues to distinguish Jev answer confidence from self-knowledge; black-box behavior does not establish access to an internal knowledge state.", "zh": "通过控制信息与表面线索，区分 Jev 的答案置信度与是否知道答案的自评；黑盒表现并不证明能读取内部知识状态。" }, { "category": "Papers", "title": "AnyJev Technical Report", "by": "Zhang et al. · 2026-09-30 · v1", "date": "2026-09-30", "url": "https://arxiv.org/abs/2610.00831v1", "en": "Training-free decision readout with label-prior and option-order correction.", "zh": "无需训练的决策读出，修正标签先验与选项顺序偏差。" }, { "category": "Papers", "title": "JevSpawn: Adaptive Agentic Inference through Compositional Action Spaces", "by": "Su & Huang · 2026-09-30 · v1", "date": "2026-09-30", "url": "https://arxiv.org/abs/2610.00437v1", "en": "Composes finite action fields and explores alternatives using interaction feedback.", "zh": "组合有限行动字段，利用交互反馈探索候选路径。" }, { "category": "Papers", "title": "OmniMed-Jev: Calibrating LVLM Confidence for Trustworthy Medical Multimodal Decisions via System One", "by": "Tang & Chen · 2026-09-30 · v1", "date": "2026-09-30", "first_submitted": "2026-09-30", "url": "https://arxiv.org/abs/2610.00381v1", "en": "Medical candidate-conditioned decisions are compared with a matched generative baseline; calibration gains do not isolate interface effects from training differences or establish clinical readiness.", "zh": "将医学候选条件决策与匹配的生成式基线比较；校准收益未分离接口与训练差异的影响，也不能证明临床部署成熟度。" }, { "category": "Papers", "title": "Bongard: Training Machine Intuition", "by": "Ding, Jin & Ji · 2026-09-30 · v1", "date": "2026-09-30", "url": "https://arxiv.org/abs/2609.39111v1", "en": "Shared evidence encoding and candidate-probability readout, trained with representation and action-outcome feedback.", "zh": "共享证据编码与候选概率读出，结合表征学习和行动结果反馈训练。" }, { "category": "Papers", "title": "Evaluating and Benchmarking the System One Model Jev", "by": "Deußer et al. · 2026-09-29 · v1", "date": "2026-09-29", "url": "https://arxiv.org/abs/2609.37647v1", "en": "An early benchmark study of Jev. Read the task selection and evaluation setup alongside its reported results.", "zh": "Jev 的早期基准研究。阅读报告结果时，同时检查任务选择与评测设置。" }, { "category": "Papers", "title": "Chinese-Jev: Bringing System One Models to Chinese-Language Tasks", "by": "Wang et al. · 2026-09-29 · v1", "date": "2026-09-29", "url": "https://arxiv.org/abs/2609.36965v1", "en": "A Chinese typed-decision model and proposed benchmark; domain-specific training calls for independent validation.", "zh": "面向中文的类型化决策模型与新基准；领域专用训练的效果仍需独立验证。" }, { "category": "Papers", "title": "Mnemon: Raw Records, Fast Judgments, Slow Thoughts", "by": "Wang · 2026-09-28 · v1", "date": "2026-09-28", "url": "https://arxiv.org/abs/2609.36059v1", "en": "Keeps raw memory records, separating Jev relevance judgments from LLM search planning and answer generation.", "zh": "保留原始记忆记录，将 Jev 相关性判断与 LLM 检索规划、回答生成分开。" }, { "category": "Papers", "title": "Sys1Cal-v1: A Dataset for Probability Calibration", "by": "Porcedda · 2026-09-28 · v1", "date": "2026-09-28", "url": "https://arxiv.org/abs/2609.35342v1", "en": "A controlled calibration study using constructed probabilities. Interpretations of latent uncertainty remain hypotheses.", "zh": "使用构造概率进行受控校准测试；对潜在不确定性的解释仍属研究假说。" }, { "category": "Papers", "title": "JEV as a Judge for Agent Trace Security: An Empirical Comparison with Generative LLM Judges", "by": "Wang & Gao · 2026-09-28 · v1", "date": "2026-09-28", "url": "https://arxiv.org/abs/2609.34862v1", "en": "Compares Jev and generative judges on retrospective agent-trace risk classification; dataset rankings vary, and successful-call latency and valid-judgment cost must be read with coverage.", "zh": "比较 Jev 与生成式评判模型对历史 agent 轨迹的风险分类；排名因数据集而异，成功调用耗时和有效判断成本需结合覆盖率解读。" }, { "category": "Papers", "title": "Beyond Calibration: Do Typed Decisions Obey the Probability Axioms?", "by": "Li et al. · 2026-09-27 · v1", "date": "2026-09-27", "url": "https://arxiv.org/abs/2609.33209v1", "en": "Tests coherence between logically related probabilities, separately from accuracy and calibration.", "zh": "检验逻辑相关命题的概率一致性，与准确率和校准分别考察。" }, { "category": "Papers", "title": "Typed Decision Models: An Early Evidence Audit and Evaluation Checklist", "by": "Tang & Zheng · 2026-09-26 · v1", "date": "2026-09-26", "url": "https://arxiv.org/abs/2609.32160v1", "en": "An evidence audit and a checklist for evaluating typed decision models.", "zh": "围绕类型化决策模型的证据审查与评估清单。" }, { "category": "Papers", "title": "JevAdvBench", "by": "Hu et al. · 2026-09-25 · v1", "date": "2026-09-25", "url": "https://arxiv.org/abs/2609.31142v1", "en": "A black-box adversarial benchmark. A changed decision is not necessarily an incorrect decision.", "zh": "黑盒对抗评测基准。决策发生变化，并不必然代表决策错误。" }, { "category": "Papers", "title": "LAVOIR: Teaching a Single-Pass Decision Encoder When and What to Ask with Amortized Value of Information", "by": "Yilmaz, Tasdemir & Gozay · 2026-09-25 · v1", "date": "2026-09-25", "url": "https://arxiv.org/abs/2609.30706v1", "en": "Estimates whether asking for missing information would improve a typed decision.", "zh": "估计追问缺失信息是否有助于改进类型化决策。" }, { "category": "Papers", "title": "Jev-Mobile: Jev as an Executor for Mobile GUI Agents", "by": "Zhang · 2026-09-24 · v1", "date": "2026-09-24", "first_submitted": "2026-09-24", "url": "https://arxiv.org/abs/2609.30186v1", "en": "Uses VLM local-goal planning with Jev action selection over Android accessibility-tree candidates; reported time and API-cost savings are conditional on successful trajectories and the tested controller configurations.", "zh": "用 VLM 规划局部目标，再由 Jev 从 Android 无障碍树候选中选动作；报告的时延和 API 成本节省仅针对成功轨迹及所测控制器配置。" }, { "category": "Papers", "title": "Just Ask Jev: Reinforcement Learning for Calibrated Decisions as a Zero-Shot Detector of AI Alignment Failures", "by": "Guo et al. · 2026-09-24 · v1", "date": "2026-09-24", "first_submitted": "2026-09-24", "url": "https://arxiv.org/abs/2609.29429v1", "en": "Benchmarks Jev alignment-failure detection while separating question wording from input context; benchmark-scorer labels dominate, and context gains often depend on fields encoding the label.", "zh": "评测 Jev 对 AI 对齐失效的检测，分别改变提问与输入上下文；参考标签主要来自基准评分器，上下文收益常依赖编码标签的字段。" }, { "category": "Papers", "title": "From Text Decisions to Pixels", "by": "Zhou et al. · 2026-09-24 · v1", "date": "2026-09-24", "url": "https://arxiv.org/abs/2609.29283v1", "en": "A Jev-style visual-choice study separating adaptation gains, candidate-readout validity, and calibration limits.", "zh": "Jev 风格视觉选择模型研究，区分适配收益、候选读出有效性与校准局限。" }, { "category": "Papers", "title": "Decision Hijacking: Prompt Injection Attacks on Jev's Typed Probabilistic Decisions", "by": "Wu & Lim · 2026-09-23 · v1", "date": "2026-09-23", "url": "https://arxiv.org/abs/2609.28613v1", "en": "Audits Jev injection on reconstructed InjecAgent cases, separating probability shifts from attacker-target selections; optimization results are checked on fresh calls, within this benchmark setting.", "zh": "在重构的 InjecAgent 案例中审查 Jev 提示注入，区分概率偏移与攻击目标选择；优化结果通过新调用验证，结论限于该基准设置。" }, { "category": "Papers", "title": "Can Jev Judge Radiology Reports? Evaluating a System One Model for Clinical Factuality", "by": "Huang et al. · 2026-09-23 · v1", "date": "2026-09-23", "first_submitted": "2026-09-23", "url": "https://arxiv.org/abs/2609.27607v1", "en": "Jev checks support between generated and reference radiology reports in both directions; expert agreement depends on report size and error definitions and does not establish prospective clinical effectiveness.", "zh": "Jev 双向检查生成与参考放射学报告之间的陈述支持关系；与专家的一致性受报告长度和错误定义影响，并不证明前瞻性临床效果。" }, { "category": "Papers", "title": "KITE: Scaling Jev Population Experiments with Sparse Flagship Calibration", "by": "Li · 2026-09-23 · v1", "date": "2026-09-23", "first_submitted": "2026-09-23", "url": "https://arxiv.org/abs/2609.27535v1", "en": "Jev state tables and sparse flagship-model anchors support population simulations with shared human–model discrepancy estimates; larger simulations do not add human evidence or establish prospective intervention effects.", "zh": "Jev 状态表与少量强模型锚点支持纳入共享人—模型差异估计的群体模拟；扩大模拟规模不会增加人类证据，也不能确立前瞻性干预效果。" }, { "category": "Papers", "title": "JEV-Star: Fast, Low-Cost StarCraft II Control with Language-Model Planning", "by": "Ma, Zhao, Zeng & Zhao · 2026-09-23 · v1", "date": "2026-09-23", "first_submitted": "2026-09-23", "url": "https://arxiv.org/abs/2609.27331v1", "en": "Combines Jev action selection with persistent language-model planning for StarCraft II; limited full-game evidence and simultaneous interface changes prevent isolating the planning contribution.", "zh": "在星际争霸 II 中结合 Jev 动作选择与持续的语言模型规划；完整对局证据有限，同时发生的接口变化使规划贡献无法单独归因。" }, { "category": "Papers", "title": "Evaluating Decision Models for Text Annotation in Computational Social Science", "by": "Ibrahim & Zaki · 2026-09-23 · v2", "date": "2026-09-23", "url": "https://arxiv.org/abs/2609.24574v2", "en": "Evaluates zero-shot social-science annotation and confidence-routed LLM cascades; lower costs coexist with task-dependent errors and overconfidence, so confidence cannot substitute for label validation.", "zh": "评估零样本社会科学标注与按置信度分流的 LLM 级联；低成本仍伴随任务相关误差和过度自信，置信度不能替代标签验证。" }, { "category": "Papers", "title": "REFLEX with Jev for Efficient Selective Control in LLM Agents", "by": "Wu & Lim · 2026-09-22 · v1", "date": "2026-09-22", "url": "https://arxiv.org/abs/2609.26532v1", "en": "Confidence-gated Jev control with LLM fallback; gains depend on action alternatives and baseline routing.", "zh": "以置信度控制 Jev 与 LLM 回退；收益取决于候选动作与基线路由。" }, { "category": "Papers", "title": "Visual Jev: Accurate and Efficient Decisions from Shared Visual Context", "by": "Yu & Yao · 2026-09-22 · v1", "date": "2026-09-22", "first_submitted": "2026-09-22", "url": "https://arxiv.org/abs/2609.25845v1", "en": "An independent visual decision architecture shares image context across batched questions using an LM-head readout; warm amortized speedups trade higher peak memory, while accuracy gains concentrate on trained task families.", "zh": "独立的视觉决策架构以语言模型头读出，让批处理问题共享图像上下文；预热后的摊销加速伴随更高峰值内存，准确率收益集中于训练覆盖的任务家族。" }, { "category": "Papers", "title": "Jev-Mem: System-One-Controlled Agentic Memory", "by": "Jiang et al. · 2026-09-21 · v1", "date": "2026-09-21", "url": "https://arxiv.org/abs/2609.23986v1", "en": "Studies Jev as a controller for agent memory. Reported outcomes concern the complete memory workflow.", "zh": "研究以 Jev 控制 agent 记忆；报告结果衡量完整记忆工作流。" }, { "category": "Papers", "title": "this-that-model-1.0", "by": "Cheng et al. · 2026-09-20 · v1", "date": "2026-09-20", "url": "https://arxiv.org/abs/2609.23886v1", "en": "An open typed-decision model using hidden-state readout. The abstract reports a small external comparison.", "zh": "采用隐藏状态读出的开放类型化决策模型。摘要报告了一项小规模外部比较。" } ] }