{ "schema_version": 1, "read_only_audit": true, "workspace": ".", "scope": "backfill-inventory.json中54个new Google/DeepMind官方源身份；不覆盖全部521、不是全网查新", "summary": { "original_new_count": 54, "disposition_counts": { "genuine_substantive_review_gap_not_accepted_implementation": 48, "out_of_scope_labor_policy": 1, "conceptual_position_no_direct_implementation_target": 1, "conceptual_report_no_direct_implementation_target": 1, "out_of_scope_normative_ethics": 1, "already_triaged_p2_not_accepted_delivery": 1, "conceptual_white_paper_no_direct_implementation_target": 1 }, "existing_implementation_found_count": 0, "existing_substantive_terminal_found_count": 0, "duplicate_identities_within_54": 0, "accepted_implementation_commitments_established_this_audit": 0 }, "method": [ "读AGENTS.md及docs/architecture.md；禁止DeepXiv/Superpowers，未训练/下载/改仓库", "完整读取ledger和manifest并做canonical ID与题名模糊映射；54 ID均无命中", "docs/src/tests做54个精确题名与23个独特别名的全量文本搜索；排除历史审计材料后仅有发现测试fixture与不相关Flames图注", "两条范围外和三条白皮书/观点判断基于已有官方缓存摘要的主题/文体初筛；不冒称全文审查或既有仓库终态" ], "full_file_checks": [ { "path": "docs/paper-discovery-ledger.json", "line_count": 123324, "full_file_raw_identity_search_matches": [] }, { "path": "docs/research-manifest.json", "line_count": 39593, "full_file_raw_identity_search_matches": [] } ], "important_semantics": [ "new表示没有current catalog或substantive terminal，不代表新发表、已接受或必须实现", "54全部是Google/DeepMind官方源信号，并非54篇都已确认一作Google隶属", "PoTRE已有P2范围初筛但没有实施承诺；已有P2不等于高成本/拒绝终态", "官网URL重复已由54的canonical ID集合去重，没有新增可减掉的重复身份", "2范围外和3非直接实施对象是本轮审计判断，未回写ledger；54现存标签仍然原样。" ], "papers": [ { "id": "2405.02699", "title": "Platform Competition in the Autobidding World", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/platform-competition-in-the-autobidding-world/" ], "original_publication_dates_in_prior_audit": [ "2024-05-04" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "分析自动出价广告平台竞争，相关于广告机制；原稿已于 2024 年公开，2026 会议年份不算首发。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 21, "json_pointer": "/papers/0" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 1010, "json_pointer": "/official_pool/63" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 10114, "json_pointer": "/records/216" } ] }, { "id": "2507.23590", "title": "Identifying Hearing Difficulty Moments in Conversational Audio", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/identifying-hearing-difficulty-moments-in-conversational-audio/" ], "original_publication_dates_in_prior_audit": [ "2025-07-31" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "音频语言模型检测听力困难片段，属于多模态应用候选，需公开数据与真实模型验证。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 30, "json_pointer": "/papers/1" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 397, "json_pointer": "/official_pool/24" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 4572, "json_pointer": "/records/101" } ] }, { "id": "2509.22963", "title": "Reinforcement Learning with Discrete Diffusion Policies for Combinatorial Action-Spaces", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/reinforcement-learning-with-discrete-diffusion-policies-for-combinatorial-action-spaces/" ], "original_publication_dates_in_prior_audit": [ "2025-09-26" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "用强化学习优化离散扩散策略，具有后训练方法相关性；首发为 2025 年，不是 2026 年本窗口新文。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 39, "json_pointer": "/papers/2" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 662, "json_pointer": "/official_pool/41" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 7066, "json_pointer": "/records/156" } ] }, { "id": "2510.03217", "title": "Abstain and Validate: A Dual-LLM Policy for Reducing Noise in Agentic Program Repair", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/abstain-and-validate-a-dual-llm-policy-for-reducing-noise-in-agentic-program-repair/" ], "original_publication_dates_in_prior_audit": [ "2025-10-03" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "弃权和补丁验证双LLM策略，适合代码Agent历史候选；过滤成功率需报告保留率。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 48, "json_pointer": "/papers/3" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 775, "json_pointer": "/official_pool/48" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 7988, "json_pointer": "/records/174" } ] }, { "id": "2510.17670", "title": "On-the-Fly OVD Adaptation with FLAME: Few-shot Localization via Active Marginal-Samples Exploration", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/on-the-fly-ovd-adaptation-with-flame-few-shot-localization-via-active-marginal-samples-exploration/" ], "original_publication_dates_in_prior_audit": [ "2025-10-20" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "开放词汇检测适配的 FLAME；保留为历史多模态候选。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 57, "json_pointer": "/papers/4" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 40, "json_pointer": "/official_pool/1" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 142, "json_pointer": "/records/4" } ] }, { "id": "2511.02767", "title": "Dynamic Reflections: Probing Video Representations with Text Alignment", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://deepmind.google/research/publications/193694/" ], "original_publication_dates_in_prior_audit": [ "2026-04-23" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "官方目录日期2026-04-23早于10月8日；本轮做时间范围核验，不把目录旧论文计为新发表或声称原稿全文重审。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 66, "json_pointer": "/papers/5" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 1908, "json_pointer": "/official_pool/121" }, { "path": "docs/research-audits/oct15-review-closure/google-deepmind-source-review.json", "line": 623, "json_pointer": "/records/19" } ] }, { "id": "2511.03782", "title": "Expert evaluation of LLM world models: A high-Tc superconductivity case study", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/expert-evaluation-of-llm-world-models-a-high-tc-superconductivity-case-study/" ], "original_publication_dates_in_prior_audit": [ "2025-11-05" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "专家科学问答及多模态文献RAG评测可作Agent/多模态公共评测历史候选。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 75, "json_pointer": "/papers/6" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 254, "json_pointer": "/official_pool/15" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 2772, "json_pointer": "/records/58" } ] }, { "id": "2511.10865", "title": "Towards A Human-in-the-Loop Framework for Reliable Patch Evaluation using an LLM-as-a-Judge", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/towards-a-human-in-the-loop-framework-for-reliable-patch-evaluation-using-an-llm-as-a-judge/" ], "original_publication_dates_in_prior_audit": [ "2025-11-14" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "人类修订逐 bug 评分准则后用 LLM 判断补丁有效性，属于 Agent 评测；arXiv 原稿为 2025 年。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 84, "json_pointer": "/papers/7" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 758, "json_pointer": "/official_pool/47" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 7924, "json_pointer": "/records/173" } ] }, { "id": "2511.11552", "title": "DocLens: A Tool-Augmented Multi-Agent Framework for Long Visual Document Understanding", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/doclens-a-tool-augmented-multi-agent-framework-for-long-visual-document-understanding/" ], "original_publication_dates_in_prior_audit": [ "2025-11-14" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "DocLens 用视觉文档解析、证据镜头与裁决改进长文档 QA，属于多模态 Agent；2025 年首发。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 93, "json_pointer": "/papers/8" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 882, "json_pointer": "/official_pool/55" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 8709, "json_pointer": "/records/188" } ] }, { "id": "2601.00372", "title": "LLM-Powered Analysis of IoT User Reviews: Tracking and Ranking Security and Privacy Concerns", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/llm-powered-analysis-of-iot-user-reviews-tracking-and-ranking-security-and-privacy-concerns/" ], "original_publication_dates_in_prior_audit": [ "2026-01-01" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "微调 LLM 识别并聚类物联网评论中的安全隐私问题，相关于内容理解；1 月首发，不依据离线评论结果推定线上 A/B。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 102, "json_pointer": "/papers/9" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 583, "json_pointer": "/official_pool/36" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 6289, "json_pointer": "/records/139" } ] }, { "id": "2601.19066", "title": "Dynamic Cogeneration of Bug Reproduction Test in Agentic Program Repair", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/dynamic-cogeneration-of-bug-reproduction-test-in-agentic-program-repair/" ], "original_publication_dates_in_prior_audit": [ "2026-01-27" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "动态协同生成 bug 复现和修复测试；120 个内部 Google bug 不能充当本地公共评测。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 111, "json_pointer": "/papers/10" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 792, "json_pointer": "/official_pool/49" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 8063, "json_pointer": "/records/175" } ] }, { "id": "2601.21149", "title": "Mobility-Embedded POIs: Learning What a Place Is and How It’s Used from Human Movement", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/mobility-embedded-pois-learning-what-a-place-is-and-how-its-used-from-human-movement/" ], "original_publication_dates_in_prior_audit": [ "2026-01-29" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "POI文本表示与移动轨迹对比学习，适合基础模型/LLM应用历史候选；不能凭地图任务推断线上A/B。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 120, "json_pointer": "/papers/11" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 1130, "json_pointer": "/official_pool/71" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 11219, "json_pointer": "/records/239" } ] }, { "id": "2602.02164", "title": "Co-RedTeam: Orchestrated Security Discovery and Exploitation with LLM Agents", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/co-redteam-orchestrated-security-discovery-and-exploitation-with-llm-agents/" ], "original_publication_dates_in_prior_audit": [ "2026-02-02" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "CoRedTeam 执行反馈驱动的网络安全协作 Agent；历史候选。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 139, "json_pointer": "/papers/13" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 724, "json_pointer": "/official_pool/45" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 7682, "json_pointer": "/records/169" } ] }, { "id": "2602.02690", "title": "Outrunning LLM Cutoffs: A Live Kernel Crash Resolution Benchmark for All,", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/outrunning-llm-cutoffs-a-live-kernel-crash-resolution-benchmark-for-all/" ], "original_publication_dates_in_prior_audit": [ "2026-02-02" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "LivekBench 以实时 kernel crash resolution 降低截止时间污染，属于历史 Agent benchmark。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 157, "json_pointer": "/papers/15" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 1228, "json_pointer": "/official_pool/77" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 12024, "json_pointer": "/records/255" } ] }, { "id": "2602.06276", "title": "Learning from Attribution Sets", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/learning-from-attribution-sets/" ], "original_publication_dates_in_prior_audit": [ "2026-02-06" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "利用归因集合进行统计估计，相关于广告转化建模；已确认 2 月原稿，工业线上证据仍需另审。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 166, "json_pointer": "/papers/16" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 980, "json_pointer": "/official_pool/61" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 9655, "json_pointer": "/records/207" } ] }, { "id": "2602.07266", "title": "ADCanvas: Accessible and Conversational Audio Description Authoring for Blind and Low Vision Creators", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/adcanvas-accessible-and-conversational-audio-description-authoring-for-blind-and-low-vision-creators/" ], "original_publication_dates_in_prior_audit": [ "2026-02-06" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "对话式 VQA、脚本生成与无障碍音频描述 Agent；12 人用户研究不能替代能力 benchmark。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 184, "json_pointer": "/papers/18" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 57, "json_pointer": "/official_pool/2" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 893, "json_pointer": "/records/21" } ] }, { "id": "2602.07918", "title": "Causal Armor: Efficient Indirect Prompt Injection Guardrails via Causal Attribution", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/causal-armor-efficient-indirect-prompt-injection-guardrails-via-causal-attribution/" ], "original_publication_dates_in_prior_audit": [ "2026-02-08" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "因果消融归因触发选择性IPI净化和CoT遮蔽，适合Agent安全历史候选。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 193, "json_pointer": "/papers/19" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 707, "json_pointer": "/official_pool/44" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 7608, "json_pointer": "/records/168" } ] }, { "id": "2602.12481", "title": "Online Advertising with Spatial Interactions", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/online-advertising-with-spatial-interactions/" ], "original_publication_dates_in_prior_audit": [ "2026-02-12" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "广告空间外部性分配算法理论；保留历史相关项，未查到线上证据前不接受工业复现。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 202, "json_pointer": "/papers/20" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 1027, "json_pointer": "/official_pool/64" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 10179, "json_pointer": "/records/217" } ] }, { "id": "2602.16938", "title": "ConvApparel: A Benchmark Dataset and Validation Framework for User Simulators in Conversational Recommenders", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/convapparel-a-benchmark-dataset-and-validation-framework-for-user-simulators-in-conversational-recommenders-2/" ], "original_publication_dates_in_prior_audit": [ "2026-02-18" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "ConvApparel 对话式用户模拟与评测；历史候选，需核对公开数据和可执行模拟器。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 220, "json_pointer": "/papers/22" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 365, "json_pointer": "/official_pool/22" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 4440, "json_pointer": "/records/99" } ] }, { "id": "2602.17554", "title": "A Theoretical Framework for Modular Learning of Robust Generative Models", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/a-theoretical-framework-for-modular-learning-of-robust-generative-models/" ], "original_publication_dates_in_prior_audit": [ "2026-02-19" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "以门控专家和鲁棒目标组织模块化生成模型，属于基础模型训练；2 月原稿而非本轮新文。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 229, "json_pointer": "/papers/23" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 950, "json_pointer": "/official_pool/59" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 9224, "json_pointer": "/records/198" } ] }, { "id": "2603.01293", "title": "Theoretical Perspectives on Data Quality and Synergistic Effects in Pre- and Post-Training Reasoning Models", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/theoretical-perspectives-on-data-quality-and-synergistic-effects-in-pre-and-post-training-reasoning-models/" ], "original_publication_dates_in_prior_audit": [ "2026-03-01" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "用理论模型解释预训练与后训练的数据质量协同，相关于基础模型/后训练；3 月首发。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 248, "json_pointer": "/papers/25" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 1211, "json_pointer": "/official_pool/76" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 11903, "json_pointer": "/records/253" } ] }, { "id": "2603.06981", "title": "Diffusion Controller: Framework, Algorithms and Parameterization", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/diffusion-controller-framework-algorithms-and-parameterization/" ], "original_publication_dates_in_prior_audit": [ "2026-03-07" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "DiffCon 将扩散采样表述为控制问题，并提供 SFT/RLFT 控制器；属于多模态与后训练，3 月首发。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 266, "json_pointer": "/papers/27" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 615, "json_pointer": "/official_pool/38" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 6864, "json_pointer": "/records/153" } ] }, { "id": "2603.10465", "title": "MoXaRt: Audio-Visual Object-Guided Sound Interaction for XR", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/moxart-audio-visual-object-guided-sound-interaction-for-xr-2/" ], "original_publication_dates_in_prior_audit": [ "2026-03-11" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "视听对象引导声音交互与分离，属于多模态历史候选。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 275, "json_pointer": "/papers/28" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 474, "json_pointer": "/official_pool/29" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 3400, "json_pointer": "/records/109" } ], "dedup_note_zh": "继承Google记录71与109使用两个官网URL指同一论文；当前54中只有一个ID，已去重。" }, { "id": "2603.14609", "title": "GroundSet: A Cadastral-Grounded Dataset for Spatial Understanding with Vector Data", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/groundset-a-cadastral-grounded-dataset-for-spatial-understanding-with-vector-data/" ], "original_publication_dates_in_prior_audit": [ "2026-03-15" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "GroundSet公开空间理解数据、模型与指令微调，适合多模态历史候选。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 284, "json_pointer": "/papers/29" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 1356, "json_pointer": "/official_pool/85" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 13367, "json_pointer": "/records/285" } ] }, { "id": "2603.14769", "title": "POLCA: Stochastic Generative Optimization with LLM", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/polca-stochastic-generative-optimization-with-llm-2/" ], "original_publication_dates_in_prior_audit": [ "2026-03-16" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "POLCA随机反馈优先队列+多样性与历史摘要优化，适合RSI/Agent历史候选。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 293, "json_pointer": "/papers/30" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 1476, "json_pointer": "/official_pool/93" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 15601, "json_pointer": "/records/340" } ] }, { "id": "2603.24591", "title": "Vibe Coding XR: Accelerating AI + XR Prototyping with XR Blocks and Gemini", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/vibe-coding-xr-accelerating-ai-xr-prototyping-with-xr-blocks-and-gemini/" ], "original_publication_dates_in_prior_audit": [ "2026-03-25" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "自然语言到WebXR程序的开源工作流与VCXR60试验，适合代码Agent历史候选。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 311, "json_pointer": "/papers/32" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 207, "json_pointer": "/official_pool/12" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 2418, "json_pointer": "/records/51" } ] }, { "id": "2603.28896", "title": "Fisheye3R: Adapting Unified 3D Feed-Forward Foundation Models to Fisheye Lenses", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/fisheye3r-adapting-unified-3d-feed-forward-foundation-models-to-fisheye-lenses/" ], "original_publication_dates_in_prior_audit": [ "2026-03-30" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "Fisheye3R 以适配 token 扩展三维基础模型到鱼眼成像，属于多模态；3 月原稿已经定位。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 329, "json_pointer": "/papers/34" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 1162, "json_pointer": "/official_pool/73" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 11369, "json_pointer": "/records/241" } ] }, { "id": "2603.29470", "title": "AI-Simulated Expert Panels for Socio-Technical Scenarios and Decision Guidance", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/ai-simulated-expert-panels-for-socio-technical-scenarios-and-decision-guidance/" ], "original_publication_dates_in_prior_audit": [ "2026-03-31" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "LLM专家面板与交叉影响一致性情景生成，保留Agent历史候选，不能当真实专家真值。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 338, "json_pointer": "/papers/35" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 1506, "json_pointer": "/official_pool/95" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 15773, "json_pointer": "/records/343" } ] }, { "id": "2603.29791", "title": "Reasoning-Driven Synthetic Data Generation and Evaluation", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/reasoning-driven-synthetic-data-generation-and-evaluation/" ], "original_publication_dates_in_prior_audit": [ "2026-03-31" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "Simula 用推理式 Agent 生成和评价无种子合成数据，属于基础模型数据与 Agent；已找到 3 月 arXiv 原稿。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 347, "json_pointer": "/papers/36" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 190, "json_pointer": "/official_pool/11" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 2354, "json_pointer": "/records/50" } ] }, { "id": "2604.05364", "title": "TFRBench: A Reasoning Benchmark for Evaluating Forecasting Systems", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/tfrbench-a-reasoning-benchmark-for-evaluating-forecasting-systems/" ], "original_publication_dates_in_prior_audit": [ "2026-04-07" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "TFRBench多Agent生成可验证预测推理的公开基准，适合Agent/evolve评测历史候选。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 356, "json_pointer": "/papers/37" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 1179, "json_pointer": "/official_pool/74" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 11584, "json_pointer": "/records/246" } ] }, { "id": "2604.09803", "title": "MAGE: Modality-Agnostic Music Generation and Editing", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/mage-modality-agnostic-music-generation-and-editing/" ], "original_publication_dates_in_prior_audit": [ "2026-04-10" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "流Transformer的多模态音乐生成/编辑及动态模态mask，适合多模态历史候选。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 365, "json_pointer": "/papers/38" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 1281, "json_pointer": "/official_pool/80" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 12496, "json_pointer": "/records/265" } ] }, { "id": "2604.20329", "title": "Image Generators are Generalist Vision Learners", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://deepmind.google/research/publications/240658/" ], "original_publication_dates_in_prior_audit": [ "2026-04-22" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "官方目录日期2026-04-22早于10月8日；本轮做时间范围核验，不把目录旧论文计为新发表或声称原稿全文重审。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 384, "json_pointer": "/papers/40" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 1925, "json_pointer": "/official_pool/122" }, { "path": "docs/research-audits/oct15-review-closure/google-deepmind-source-review.json", "line": 656, "json_pointer": "/records/20" } ] }, { "id": "2604.23099", "title": "ProEval: Proactive Failure Discovery and Efficient Performance Estimation for Generative AI Evaluation", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://deepmind.google/research/publications/238239/" ], "original_publication_dates_in_prior_audit": [ "2026-04-25" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "官方目录日期2026-04-25早于10月8日；本轮做时间范围核验，不把目录旧论文计为新发表或声称原稿全文重审。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 393, "json_pointer": "/papers/41" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 1891, "json_pointer": "/official_pool/120" }, { "path": "docs/research-audits/oct15-review-closure/google-deepmind-source-review.json", "line": 590, "json_pointer": "/records/18" } ] }, { "id": "2604.27723", "title": "Optimized Deferral for Imbalanced Settings", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/optimized-deferral-for-imbalanced-settings/" ], "original_publication_dates_in_prior_audit": [ "2026-04-30" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "MILD 面向不均衡任务优化 LLM 延迟/转交策略，保留为历史推理路由候选。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 402, "json_pointer": "/papers/42" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 916, "json_pointer": "/official_pool/57" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 8864, "json_pointer": "/records/191" } ] }, { "id": "2605.03767", "title": "Did US Worker Retraining Reduce Participant Automation Exposure?", "original_status": "new", "audit_disposition": "out_of_scope_labor_policy", "judgement_zh": "范围外（社会政策实证研究）；不是本项目技术复现待办", "basis_zh": "官方摘要研究2300万WIOA参与记录的职业再培训与自动化暴露，不是推荐/基础模型/后训练/Agent/多模态算法。此为主题初筛，不是阅读全文或实验复现。", "source_urls": [ "https://deepmind.google/research/publications/239849/" ], "original_publication_dates_in_prior_audit": [ "2026-05-06" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "官方目录日期2026-05-06早于10月8日；本轮做时间范围核验，不把目录旧论文计为新发表或声称原稿全文重审。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 430, "json_pointer": "/papers/45" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 1874, "json_pointer": "/official_pool/119" }, { "path": "docs/research-audits/oct15-review-closure/google-deepmind-source-review.json", "line": 557, "json_pointer": "/records/17" }, { "path": ".cache/oct12-scan/official/8f7cf5eb35ea837ef078f09d64a0062c7abe678c1e64ebd15a7e742e2b638671.html", "line": 1, "source_url": "https://deepmind.google/research/publications/239849/", "location": "official Abstract and ScholarlyArticle JSON-LD; cached primary source", "note": "仅本轮主题/文体初筛；不是全文审查终态" } ] }, { "id": "2605.26340", "title": "Scientist-One: Verifiable Autonomous Research via Chain-of-Evidence", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/scientist-one-verifiable-autonomous-research-via-chain-of-evidence/" ], "original_publication_dates_in_prior_audit": [ "2026-05-25" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "Scientist One 用证据链连接论文结果与代码；官方标题/作者别名匹配到该预印本。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 448, "json_pointer": "/papers/47" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 839, "json_pointer": "/official_pool/52" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 8281, "json_pointer": "/records/179" } ] }, { "id": "2605.29729", "title": "Realistic honeypot evaluations for scheming propensity", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://deepmind.google/research/publications/253391/" ], "original_publication_dates_in_prior_audit": [ "2026-05-28" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "官方目录日期2026-05-28早于10月8日；本轮做时间范围核验，不把目录旧论文计为新发表或声称原稿全文重审。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 457, "json_pointer": "/papers/48" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 1840, "json_pointer": "/official_pool/117" }, { "path": "docs/research-audits/oct15-review-closure/google-deepmind-source-review.json", "line": 491, "json_pointer": "/records/15" } ] }, { "id": "2605.30174", "title": "LiveSVG: Zero-Shot SVG Animation via Video Generation", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/livesvg-zero-shot-svg-animation-via-video-generation/" ], "original_publication_dates_in_prior_audit": [ "2026-05-28" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "用冻结视频生成模型产生目标视频，再以可微渲染拟合 SVG 动画，属于多模态生成；已确认 5 月首发。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 475, "json_pointer": "/papers/50" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 553, "json_pointer": "/official_pool/34" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 6124, "json_pointer": "/records/136" } ] }, { "id": "2605.30322", "title": "Gram: Assessing sabotage propensities via automated alignment auditing", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://deepmind.google/research/publications/252981/" ], "original_publication_dates_in_prior_audit": [ "2026-05-28" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "官方目录日期2026-05-28早于10月8日；本轮做时间范围核验，不把目录旧论文计为新发表或声称原稿全文重审。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 484, "json_pointer": "/papers/51" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 1857, "json_pointer": "/official_pool/118" }, { "path": "docs/research-audits/oct15-review-closure/google-deepmind-source-review.json", "line": 524, "json_pointer": "/records/16" } ] }, { "id": "2606.00131", "title": "AI-PROPELLER: Warehouse-Scale Inter-Procedural Code Layout Optimization with AlphaEvolve", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/ai-propeller-warehouse-scale-inter-procedural-code-layout-optimization-with-alphaevolve/" ], "original_publication_dates_in_prior_audit": [ "2026-05-28" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "AlphaEvolve演化编译布局策略并用硬件计数器奖励，适合RSI/Agent历史高优候选。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 493, "json_pointer": "/papers/52" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 536, "json_pointer": "/official_pool/33" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 6007, "json_pointer": "/records/134" } ] }, { "id": "2606.00369", "title": "Quantifying the Salience of Geo-Cultural Values for Pluralistic Safety Alignment", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://deepmind.google/research/publications/225819/" ], "original_publication_dates_in_prior_audit": [ "2026-07-10" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "官方目录日期2026-07-10早于10月8日；本轮做时间范围核验，不把目录旧论文计为新发表或声称原稿全文重审。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 502, "json_pointer": "/papers/53" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 1697, "json_pointer": "/official_pool/108" }, { "path": "docs/research-audits/oct15-review-closure/google-deepmind-source-review.json", "line": 204, "json_pointer": "/records/6" } ] }, { "id": "2606.03237", "title": "Solipsistic superintelligence is unlikely to be cooperative", "original_status": "new", "audit_disposition": "conceptual_position_no_direct_implementation_target", "judgement_zh": "相关多Agent合作观点；不是一个明确算法实现欠账", "basis_zh": "官方摘要呼吁将相互依存作为设计原则并构建动态评测平台，属于研究范式/观点。没有在本轮确认一个论文级可执行复现目标，不能等同于接受待实现。", "source_urls": [ "https://deepmind.google/research/publications/231466/" ], "original_publication_dates_in_prior_audit": [ "2026-06-04" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "官方目录日期2026-06-04早于10月8日；本轮做时间范围核验，不把目录旧论文计为新发表或声称原稿全文重审。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 511, "json_pointer": "/papers/54" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 1823, "json_pointer": "/official_pool/116" }, { "path": "docs/research-audits/oct15-review-closure/google-deepmind-source-review.json", "line": 458, "json_pointer": "/records/14" }, { "path": ".cache/oct12-scan/official/a82ad022f39b97837c035ff1a7ebd8e6a95abc20c8a9c32884f2ec13ed7fd4bd.html", "line": 1, "source_url": "https://deepmind.google/research/publications/231466/", "location": "official Abstract and ScholarlyArticle JSON-LD; cached primary source", "note": "仅本轮主题/文体初筛；不是全文审查终态" } ] }, { "id": "2606.12683", "title": "From AGI to ASI", "original_status": "new", "audit_disposition": "conceptual_report_no_direct_implementation_target", "judgement_zh": "相关AGI/ASI路线报告；不是一个可直接交付的实现", "basis_zh": "官方摘要研究AGI到ASI的四条可能路径与瓶颈，包含递归改进，但属于前瞻报告。可列研究资料，不应仅因无代码被算成实现TODO；未做全文审查。", "source_urls": [ "https://deepmind.google/research/publications/239142/" ], "original_publication_dates_in_prior_audit": [ "2026-06-12" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "官方目录日期2026-06-12早于10月8日；本轮做时间范围核验，不把目录旧论文计为新发表或声称原稿全文重审。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 520, "json_pointer": "/papers/55" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 1806, "json_pointer": "/official_pool/115" }, { "path": "docs/research-audits/oct15-review-closure/google-deepmind-source-review.json", "line": 425, "json_pointer": "/records/13" }, { "path": ".cache/oct12-scan/official/4583ab7e6fd90b008be907196b912a74e50ec0f1eabba34eea47f2843d10f576.html", "line": 1, "source_url": "https://deepmind.google/research/publications/239142/", "location": "official Abstract and ScholarlyArticle JSON-LD; cached primary source", "note": "仅本轮主题/文体初筛；不是全文审查终态" } ] }, { "id": "2607.02515", "title": "PointDiT: Pixel-Space Diffusion for Monocular Geometry Estimation", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/pointdit-pixel-space-diffusion-for-monocular-geometry-estimation/" ], "original_publication_dates_in_prior_audit": [ "2026-07-02" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "PointDiT 以像素空间扩散预测单目三维几何，属于多模态；7 月首发及官方代码已定位。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 529, "json_pointer": "/papers/56" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 1339, "json_pointer": "/official_pool/84" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 13304, "json_pointer": "/records/284" } ] }, { "id": "2607.03906", "title": "The Case for Globally Beneficial Technology", "original_status": "new", "audit_disposition": "out_of_scope_normative_ethics", "judgement_zh": "范围外（技术收益分配伦理论证）；不是技术复现待办", "basis_zh": "官方摘要论述技术创新收益归属与五类道德理由，不提供本项目技术算法的实现对象。此为主题初筛，不是全文审查DONE。", "source_urls": [ "https://deepmind.google/research/publications/260960/" ], "original_publication_dates_in_prior_audit": [ "2026-07-06" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "官方目录日期2026-07-06早于10月8日；本轮做时间范围核验，不把目录旧论文计为新发表或声称原稿全文重审。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 538, "json_pointer": "/papers/57" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 1714, "json_pointer": "/official_pool/109" }, { "path": "docs/research-audits/oct15-review-closure/google-deepmind-source-review.json", "line": 237, "json_pointer": "/records/7" }, { "path": ".cache/oct12-scan/official/6eeaaac11d2fd504daeb465524db27bdca4ce756e1acb94c7e32f45612cd6b45.html", "line": 1, "source_url": "https://deepmind.google/research/publications/260960/", "location": "official Abstract and ScholarlyArticle JSON-LD; cached primary source", "note": "仅本轮主题/文体初筛；不是全文审查终态" } ] }, { "id": "2607.13614", "title": "VIP-MINGLE: A Corpus for Videoconference and In-Person Multimodal Interaction in Group Language Engagement", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/vip-mingle-a-corpus-for-videoconference-and-in-person-multimodal-interaction-in-group-language-engagement/" ], "original_publication_dates_in_prior_audit": [ "2026-07-15" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "跨场景音视频群体交互语料，适合多模态数据候选。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 547, "json_pointer": "/papers/58" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 1442, "json_pointer": "/official_pool/91" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 15350, "json_pointer": "/records/334" } ] }, { "id": "2607.20268", "title": "PoTRE: Test-Time Reasoning inspired by Cognitive Heterogeneity", "original_status": "new", "audit_disposition": "already_triaged_p2_not_accepted_delivery", "judgement_zh": "已有P2/query-collision初筛；并非已接受待交付", "basis_zh": "独立历史候选表将PoTRE标为p2-deferred-review、p2-or-query-collision，full_text_review_required=false，implementation_batch=null。未发现实质全文终态或实现；若未来提升优先级需另审，不是新增实施承诺。", "source_urls": [ "https://research.google/pubs/potre-test-time-reasoning-inspired-by-cognitive-heterogeneity/" ], "original_publication_dates_in_prior_audit": [ "2026-07-22" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "PoTRE 结合异质认知角色进行测试时推理，属于 Agent/推理；已确认 7 月首发。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 556, "json_pointer": "/papers/59" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 809, "json_pointer": "/official_pool/50" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 8159, "json_pointer": "/records/177" }, { "path": "docs/paper-audits/2026-historical-candidates.json", "line": 40535, "note": "下方full_text_review_required=false、plan_status=p2-or-query-collision、implementation_batch=null" } ] }, { "id": "2607.25537", "title": "Visual prompt engineering for video models", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://deepmind.google/research/publications/264392/" ], "original_publication_dates_in_prior_audit": [ "2026-07-28" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "官方目录日期2026-07-28早于10月8日；本轮做时间范围核验，不把目录旧论文计为新发表或声称原稿全文重审。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 565, "json_pointer": "/papers/60" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 1665, "json_pointer": "/official_pool/106" }, { "path": "docs/research-audits/oct15-review-closure/google-deepmind-source-review.json", "line": 140, "json_pointer": "/records/4" } ] }, { "id": "2608.05578", "title": "Detecting Safety Training Modification in Language Models via Activation Analysis", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/ams-detecting-unsafe-and-tampered-language-models-via-activation-analysis/" ], "original_publication_dates_in_prior_audit": [ "2026-08-06" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "激活空间检测安全训练修改，适合基础模型安全评测；行为微调盲区需保留。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 583, "json_pointer": "/papers/62" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 333, "json_pointer": "/official_pool/20" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 3808, "json_pointer": "/records/82" } ] }, { "id": "2608.11367", "title": "Gaze Target Estimation Anywhere with Concepts", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/gaze-target-estimation-anywhere-with-concepts/" ], "original_publication_dates_in_prior_audit": [ "2026-08-11" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "GazeAnywhere/官方 AnyGaze 概念条件凝视目标估计；已知 arXiv v1 早于窗口，最早会议版本日期另待核定。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 592, "json_pointer": "/papers/63" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 286, "json_pointer": "/official_pool/17" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 3457, "json_pointer": "/records/74" } ] }, { "id": "2608.25924", "title": "Visual General Intelligence: A White Paper", "original_status": "new", "audit_disposition": "conceptual_white_paper_no_direct_implementation_target", "judgement_zh": "相关白皮书；可保留资料阅读，不应计入实施欠账", "basis_zh": "官方摘要明确是视觉通用智能白皮书，讨论原则、模态、benchmark和学习范式，不提出一个单独可执行复现目标。全文尚未审，不据此声称已完成方法审查。", "source_urls": [ "https://deepmind.google/research/publications/270149/" ], "original_publication_dates_in_prior_audit": [ "2026-08-26" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "官方目录日期2026-08-26早于10月8日；本轮做时间范围核验，不把目录旧论文计为新发表或声称原稿全文重审。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 601, "json_pointer": "/papers/64" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 1633, "json_pointer": "/official_pool/104" }, { "path": "docs/research-audits/oct15-review-closure/google-deepmind-source-review.json", "line": 76, "json_pointer": "/records/2" }, { "path": ".cache/oct12-scan/official/8fdda604a586b76e5214f1e0b8eb62e644e0105c7f29b65d8ebaa4e32b6f4570.html", "line": 1, "source_url": "https://deepmind.google/research/publications/270149/", "location": "official Abstract and ScholarlyArticle JSON-LD; cached primary source", "note": "仅本轮主题/文体初筛；不是全文审查终态" } ] }, { "id": "2609.01588", "title": "Designing Proactive Thought Partners for Writing", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://deepmind.google/research/publications/265605/" ], "original_publication_dates_in_prior_audit": [ "2026-09-01" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "官方目录日期2026-09-01早于10月8日；本轮做时间范围核验，不把目录旧论文计为新发表或声称原稿全文重审。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 610, "json_pointer": "/papers/65" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 1616, "json_pointer": "/official_pool/103" }, { "path": "docs/research-audits/oct15-review-closure/google-deepmind-source-review.json", "line": 43, "json_pointer": "/records/1" }, { "path": "tests/test_discovery_sources.py", "line": 216, "note": "发现解析测试只用其标题/日期构造HTML fixture；不是论文方法实现" } ] }, { "id": "2609.01834", "title": "Architecting Conversational Data Systems for Stateless LLM APIs", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/architecting-conversational-data-systems-for-stateless-llm-apis/" ], "original_publication_dates_in_prior_audit": [ "2026-09-01" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "将会话状态持久化与无状态 LLM API 解耦，属于 Agent 记忆架构；arXiv 首发在 9 月 1 日。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 619, "json_pointer": "/papers/66" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 457, "json_pointer": "/official_pool/28" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 4918, "json_pointer": "/records/108" } ] }, { "id": "2609.12464", "title": "Beyond Vector Similarity: Hierarchical Context-Aware Graph RAG vs Standard RAG in Enterprise Code Migration", "original_status": "new", "audit_disposition": "genuine_substantive_review_gap_not_accepted_implementation", "judgement_zh": "真正待实质审查；尚未接受实现", "basis_zh": "Google审计仅做身份/日期/摘要范围归档；原记录fulltext_review_completed=false，没有已接受实施、拒绝或高成本终态。不能把缺实现自动翻译为实现TODO。", "source_urls": [ "https://research.google/pubs/beyond-vector-similarity-hierarchical-context-aware-graph-rag-vs-standard-rag-in-enterprise-code-migration/", "https://research.google/pubs/beyond-vector-similarity-hierarchical-context-aware-graph-rag-vs-standard-rag-in-enterprise-code-migration-2/" ], "original_publication_dates_in_prior_audit": [ "2026-09-11" ], "first_author_google_affiliation_independently_verified_this_audit": false, "existing_implementation_found": false, "existing_metrics_found": false, "existing_substantive_terminal_decision_found": false, "ledger_identity_present": false, "manifest_identity_present": false, "inherited_reason_zh": "AST图关系与上下文缓存辅助代码迁移，适合代码Agent历史候选；词面指标不能替代编译正确性。", "inherited_fulltext_review_completed": false, "evidence": [ { "path": "docs/research-audits/harness2-priority-repair/backfill-inventory.json", "line": 628, "json_pointer": "/papers/67" }, { "path": "docs/research-audits/harness2-priority-repair/google-audit.json", "line": 504, "json_pointer": "/official_pool/31" }, { "path": "docs/research-audits/oct15-review-closure/google-source-review.json", "line": 5322, "json_pointer": "/records/116" } ], "dedup_note_zh": "Google官方两个URL是同一论文（official_pool两行）；54清单已按canonical arXiv ID合并，不能再次从54扣除。" } ], "cached_author_block_insurance": { "population": 535, "cached_header_covered_count": 45, "cached_header_missing_count": 490, "covered": [ { "id": "2610.09027", "batch": "20261012-four-track-review-and-frozen16", "evidence": [ { "path": ".cache/oct12-scan/fulltext/2610.09027.json", "line": 71, "field": "affiliation", "header": "David Dobre Leo Schwinn Gauthier Gidel Spandana Gella Perouz Taslakian Pierre-André Noël Affiliation: [ Affiliation: [ Affiliation: [ Affiliation: [ Affiliation: [" } ] }, { "id": "2610.09127", "batch": "20261012-four-track-review-and-frozen16", "evidence": [ { "path": ".cache/oct12-scan/fulltext/2610.09127.json", "line": 63, "field": "affiliation", "header": "Gennadiy Savrasov Affiliation: Lomonosov Moscow State University; Maksim Elistratov Affiliation: Lomonosov Moscow State University; Nikita Gavrilov Affiliation: Lomonosov Moscow State University; Albert Garifullin Affiliation: Lomonosov Moscow State University; Oleg Pavlov Affiliation: Lomonosov Moscow State University; Soslan Kabisov Affiliation: Lomonosov Moscow State University; Vladimir Frolov Affiliation: Lomonosov Moscow State University; Anton Konushin Affiliation: Lomonosov Moscow State University; Andrey Kuznetsov Affiliation: Innopolis University; Affiliation: FusionBrain Lab Dmitrii Zhemchuzhnikov † † thanks: Corresponding author: zhemchuzhnikovds@my.msu.ru . Affiliation: Lomonosov Moscow State University;" } ] }, { "id": "2610.09159", "batch": "20261012-four-track-review-and-frozen16", "evidence": [ { "path": ".cache/oct12-scan/fulltext/2610.09159.json", "line": 71, "field": "affiliation", "header": "Param Biyani MATS parambiyani8@gmail.com Krishnamurthy (Dj) Dvijotham Google DeepMind" } ] }, { "id": "2610.09252", "batch": "20261012-four-track-review-and-frozen16", "evidence": [ { "path": ".cache/oct12-scan/fulltext/2610.09252.json", "line": 48, "field": "affiliation", "header": "Seyoung Jeong Jong Pil Yun Sang Jun Lee" } ] }, { "id": "2610.09430", "batch": "20261012-four-track-review-and-frozen16", "evidence": [ { "path": ".cache/oct12-scan/fulltext/2610.09430.json", "line": 55, "field": "affiliation", "header": "Ling Li Affiliation: Tsinghua University Qiuyu Shen Affiliation: Northwestern Polytechnical University Zheng Jiang Affiliation: Tsinghua University Qinwei Ma Affiliation: Tsinghua University Yuxuan Liu Affiliation: Tsinghua University Zhidong Deng Affiliation: Tsinghua University" } ] }, { "id": "2610.09484", "batch": "20261012-four-track-review-and-frozen16", "evidence": [ { "path": ".cache/oct12-scan/fulltext/2610.09484.json", "line": 64, "field": "affiliation", "header": "Hoang Phan Affiliation: Meta Superintelligence Labs Affiliation: New York University Work done at Meta Dat Huynh Affiliation: Meta Superintelligence Labs Andrey Zhmoginov Qi Zeng Affiliation: Meta Superintelligence Labs Wancen Mu Affiliation: Meta Superintelligence Labs Yue Cao Affiliation: Meta Superintelligence Labs Shengjie Bi Affiliation: Meta Superintelligence Labs Yun He Affiliation: Meta Superintelligence Labs Changdae Oh Affiliation: Meta Superintelligence Labs Affiliation: University of Wisconsin - Madison Work done at Meta Deren Lei Affiliation: Meta Superintelligence Labs" } ] }, { "id": "2610.09624", "batch": "20261012-four-track-review-and-frozen16", "evidence": [ { "path": ".cache/oct12-scan/fulltext/2610.09624.json", "line": 77, "field": "affiliation", "header": "Xijie Gong Affiliation: Mohamed bin Zayed University of Artificial Intelligence Affiliation: University of Electronic Science and Technology of China Tingxu Han Affiliation: Mohamed bin Zayed University of Artificial Intelligence Affiliation: Nanjing University Jiahao Zhang Affiliation: Mohamed bin Zayed University of Artificial Intelligence Wei Song Affiliation: Griffith University Ziqi Ding Affiliation: University of New South Wales Hanqi Yan Affiliation: King’s College London Youcheng Sun Affiliation: Mohamed bin Zayed University of Artificial Intelligence Lijie Hu Affiliation: Mohamed bin Zayed University of Artificial Intelligence" } ] }, { "id": "2610.09761", "batch": "20261012-four-track-review-and-frozen16", "evidence": [ { "path": ".cache/oct12-scan/fulltext/2610.09761.json", "line": 56, "field": "affiliation", "header": "Shengkai Ma Zhenyu Hou Weihua Cao" } ] }, { "id": "2610.09856", "batch": "20261012-four-track-review-and-frozen16", "evidence": [ { "path": ".cache/oct12-scan/fulltext/2610.09856.json", "line": 52, "field": "affiliation", "header": "Wenjie Liao Affiliation: Waseda University jie3040@akane.waseda.jp Liangjie Zhao Affiliation: Adelaide University zhaoliangjie55@gmail.com , Jimmy.Cao@adelaide.edu.au Zehong Cao Affiliation: Adelaide University zhaoliangjie55@gmail.com , Jimmy.Cao@adelaide.edu.au" } ] }, { "id": "2610.09935", "batch": "20261012-four-track-review-and-frozen16", "evidence": [ { "path": ".cache/oct12-scan/fulltext/2610.09935.json", "line": 67, "field": "affiliation", "header": "Zitong Yao Affiliation: Sun Yat-sen University , Guangzhou , China email: yaozt@mail2.sysu.edu.cn , Jiangrong Wu Note: Corresponding author. Affiliation: The Hong Kong University of Science and Technology , Hong Kong , China email: jiangrongwu@cse.ust.hk , Yixi Lin Affiliation: Sun Yat-sen University , Guangzhou , China email: linyx98@mail2.sysu.edu.cn , Anrui Huang Affiliation: Sun Yat-sen University , Guangzhou , China email: huanganr3@mail2.sysu.edu.cn , Luoyun Zhang Affiliation: Sun Yat-sen University , Guangzhou , China email: zhangk367@mail2.sysu.edu.cn and Yuhong Nan Affiliation: Sun Yat-sen University , Guangzhou , China email: nanhy@mail.sysu.edu.cn" } ] }, { "id": "2610.09952", "batch": "20261012-four-track-review-and-frozen16", "evidence": [ { "path": ".cache/oct12-scan/fulltext/2610.09952.json", "line": 62, "field": "affiliation", "header": "Artem Filippov Affiliation: Lomonosov Moscow State University , Moscow , Russia email: artyom.filippov@graphics.cs.msu.ru , Aleksandr Gushchin Affiliation: MSU Institute for Artificial Intelligence , Moscow , Russia email: alexander.gushchin@graphics.cs.msu.ru , Kirill Koltsov Affiliation: Lomonosov Moscow State University , Moscow , Russia email: kirill.e.koltsov@mail.ru , Dmitriy Vatolin Affiliation: MSU Institute for Artificial Intelligence , Moscow , Russia email: dmitriy@graphics.cs.msu.ru and Anastasia Antsiferova Affiliation: MSU Institute for Artificial Intelligence , Moscow , Russia email: aantsiferova@graphics.cs.msu.ru" } ] }, { "id": "2610.10062", "batch": "20261012-four-track-review-and-frozen16", "evidence": [ { "path": ".cache/oct12-scan/fulltext/2610.10062.json", "line": 59, "field": "affiliation", "header": "Obada Kraishan Affiliation: College of Media and Communication Texas Tech University Lubbock, TX, USA omareikr@ttu.edu" } ] }, { "id": "2610.10071", "batch": "20261012-four-track-review-and-frozen16", "evidence": [ { "path": ".cache/oct12-scan/fulltext/2610.10071.json", "line": 64, "field": "affiliation", "header": "Ran Zhou Affiliation: School of Intelligence Science and Technology, Nanjing University Xueming Han Affiliation: Jiutian Research Correspondence: renyuxiang@nju.edu.cn Jiaheng Liu Affiliation: School of Intelligence Science and Technology, Nanjing University Yuyao Zhang Affiliation: Jiutian Research Correspondence: renyuxiang@nju.edu.cn Fanyu Meng Affiliation: Jiutian Research Correspondence: renyuxiang@nju.edu.cn Junlan Feng Affiliation: Jiutian Research Correspondence: renyuxiang@nju.edu.cn Yuxiang Ren Affiliation: School of Intelligence Science and Technology, Nanjing University" } ] }, { "id": "2610.10120", "batch": "20261012-four-track-review-and-frozen16", "evidence": [ { "path": ".cache/oct12-scan/fulltext/2610.10120.json", "line": 60, "field": "affiliation", "header": "Hengbo XiaoYoolee.ai Boyao ZhangUniversity of Science and Technology of China Purui LiuPeking University Yuxuan ZhengPeking University Haoran YinTianjin University Haibo LiuYoolee.ai Fan ZhangYoolee.ai" } ] }, { "id": "2610.10305", "batch": "20261012-four-track-review-and-frozen16", "evidence": [ { "path": ".cache/oct12-scan/fulltext/2610.10305.json", "line": 65, "field": "affiliation", "header": "Eduardo Santos-Escriche † † thanks: Equal contribution Affiliation: Technical University of Munich Affiliation: Munich Center for Machine Learning Valerie Engelmayer 1 1 footnotemark: 1 Affiliation: Technical University of Munich Affiliation: Munich Center for Machine Learning Ya-Wei Eileen Lin Affiliation: Technical University of Munich Affiliation: Munich Center for Machine Learning Stefanie Jegelka Affiliation: Technical University of Munich Affiliation: Munich Center for Machine Learning Affiliation: Massachusetts Institute of Technology" } ] }, { "id": "2610.10786", "batch": "20261012-four-track-review-and-frozen16", "evidence": [ { "path": ".cache/oct12-scan/fulltext/2610.10786.json", "line": 63, "field": "affiliation", "header": "Syamantak Kumar Jiang Guo Hassan Hamad 2 2 footnotemark: 2 Hideo Kobayashi 2 2 footnotemark: 2 Yi Xiang 2 2 footnotemark: 2 Yezhou Yang 2 2 footnotemark: 2 Yanjun Qi 2 2 footnotemark: 2 Daniele Bonadiman 2 2 footnotemark: 2 Jiarong Jiang 2 2 footnotemark: 2 † † thanks: University of Texas at Austin. Correspondence: syamantak@utexas.edu . † † thanks: Amazon. Correspondence: gujiang@amazon.com ." } ] }, { "id": "2610.11008", "batch": "20261012-four-track-review-and-frozen16", "evidence": [ { "path": ".cache/oct12-scan/fulltext/2610.11008.json", "line": 74, "field": "affiliation", "header": "Yujiao Chen Affiliation: Massachusetts Institute of Technology Email: yujiaoch@mit.edu Affiliation: Primitive Labs Email: yujiao@primitivelabs.ai" } ] }, { "id": "2610.11123", "batch": "20261012-four-track-review-and-frozen16", "evidence": [ { "path": ".cache/oct12-scan/fulltext/2610.11123.json", "line": 71, "field": "affiliation", "header": "Xiaolong Li Xiaohan Xu Jinyang Li Xinnuo Xu Ge Qu Nan Huo Jack Williams Reynold Cheng The University of Hong Kong  Microsoft Research The BIRD Team {xia01ong,xxh24,jl0725}@connect.hku.hk https://github.com/bird-bench/GenUI-Agent" } ] }, { "id": "2610.11136", "batch": "20261012-four-track-review-and-frozen16", "evidence": [ { "path": ".cache/oct12-scan/fulltext/2610.11136.json", "line": 67, "field": "affiliation", "header": "Yuchen Miao Affiliation: Sydney Smart Technology College, Northeastern University, China Zijun Wang Affiliation: Sydney Smart Technology College, Northeastern University, China Chang Han Affiliation: Sydney Smart Technology College, Northeastern University, China Yurui Shi Affiliation: Taiyuan University of Technology Mingtai Zhang Affiliation: School of Resources, Environment and Materials, Guangxi University Siyang Xu Affiliation: School of Computer and Communication EngineeringNortheastern University at Qinhuangdao, China" } ] }, { "id": "2610.11226", "batch": "20261012-four-track-review-and-frozen16", "evidence": [ { "path": ".cache/oct12-scan/fulltext/2610.11226.json", "line": 62, "field": "affiliation", "header": "Yanlong Zhao † † thanks: Equal Contribution: Yanlong Zhao , Xiaoyuan Cheng and Huihang Liu . Affiliation: University of Science and Technology of China Xiaoyuan Cheng 1 1 footnotemark: 1 Affiliation: University College London Huihang Liu 1 1 footnotemark: 1 Affiliation: Shanghai University of Finance and Economics Baihua He Affiliation: University of Science and Technology of China Xinyu Zhang Affiliation: University of Science and Technology of China Affiliation: AMSS, Chinese Academy of Sciences Harrison Bo Hua Zhu Affiliation: University of Copenhagen Affiliation: Imperial College London Affiliation: Technical University of Denmark Wenlong Chen Affiliation: Imperial College London Li Zeng Affiliation: Shenzhen Research Institute of Big Data Zhuo Sun † † thanks: Correspondence Author. Correspondence to Zhuo Sun . Affiliation: Shanghai University of Finance and Economics Affiliation: Imperial College London" } ] }, { "id": "2610.11384", "batch": "20261012-four-track-review-and-frozen16", "evidence": [ { "path": ".cache/oct12-scan/fulltext/2610.11384.json", "line": 56, "field": "affiliation", "header": "Hangxi Guo Affiliation: The Chinese University of Hong Kong, Shenzhen Fengyuan Liu Affiliation: The Chinese University of Hong Kong, Shenzhen Yue Wang Affiliation: Shanghai AI Laboratory Affiliation: University of Science and Technology of China Yuhua Qi Affiliation: The Chinese University of Hong Kong, Shenzhen Haoyi Xiong Affiliation: Independent Fei Sun Affiliation: Institute of Computing Technology, CAS Mengnan Du Email: keepstudying931@gmail.com, mengnandu@cuhk.edu.cn * Equal contribution, † Corresponding author Affiliation: The Chinese University of Hong Kong, Shenzhen" } ] }, { "id": "2610.11573", "batch": "20261012-four-track-review-and-frozen16", "evidence": [ { "path": ".cache/oct12-scan/fulltext/2610.11573.json", "line": 66, "field": "affiliation", "header": "Yi Wen Affiliation: City University of Hong Kong Email: yiwen23-c@my.cityu.edu.hk Derong Xu Affiliation: City University of Hong Kong Email: wangyichao5@huawei.com Pengyue Jia Affiliation: City University of Hong Kong Email: xianzhao@cityu.edu.hk Yichao Wang Affiliation: Huawei Noah’s Ark Lab Yingyi Zhang Affiliation: City University of Hong Kong Maolin Wang Affiliation: City University of Hong Kong Junyi Li Affiliation: City University of Hong Kong Wenlin Zhang Affiliation: City University of Hong Kong Xiaopeng Li Affiliation: City University of Hong Kong Yong Liu Affiliation: Huawei Noah’s Ark Lab Xiangyu Zhao Affiliation: City University of Hong Kong" } ] }, { "id": "2610.11578", "batch": "20261012-four-track-review-and-frozen16", "evidence": [ { "path": ".cache/oct12-scan/fulltext/2610.11578.json", "line": 66, "field": "affiliation", "header": "Xin Yin Note: Equal contribution. Affiliation: Zhejiang University , China , Yiang Zhang Affiliation: Zhejiang University , China , Zhiyuan Peng Affiliation: Shanghai Jiao Tong University , China , Chao Ni Affiliation: Zhejiang University , China , Zhe Cui Affiliation: Hithink Research , China and Xiaohua Xin Affiliation: National Industrial Information Security Development Research Center , China" } ] }, { "id": "2610.11862", "batch": "20261012-four-track-review-and-frozen16", "evidence": [ { "path": ".cache/oct12-scan/fulltext/2610.11862.json", "line": 58, "field": "affiliation", "header": "Nihal Jain Shuangjie Yao Begum Cicekdag Zhuo Zhang Suman Jana Affiliation: Columbia University Affiliation: {nihal.jain, sy3437, bc2975, zz3474, sj2754}@columbia.edu" } ] }, { "id": "2610.12248", "batch": "20261012-four-track-review-and-frozen16", "evidence": [ { "path": ".cache/oct12-scan/fulltext/2610.12248.json", "line": 68, "field": "affiliation", "header": "Heeseung Kim Affiliation: Department of AI, University of Seoul Email: gmltmd789@uos.ac.kr" } ] }, { "id": "2610.12611", "batch": "20261012-four-track-review-and-frozen16", "evidence": [ { "path": ".cache/oct12-scan/fulltext/2610.12611.json", "line": 56, "field": "affiliation", "header": "Tian Jin 1,∗ Wenhao Wang 2,∗ Jing Kang 1 Haoting Shi 1 Rui Ye 1 Yijun Lu 1 Siheng Chen 1,† 1 Shanghai Jiaotong University 2 Zhejiang University * Equal contribution. † Corresponding author. Correspondence: sihengc@sjtu.edu.cn" } ] }, { "id": "2610.12655", "batch": "20261012-four-track-review-and-frozen16", "evidence": [ { "path": ".cache/oct12-scan/fulltext/2610.12655.json", "line": 60, "field": "affiliation", "header": "Jiatan Huang 1 Zheyuan Zhang 2 Ziming Li 1 Tianyi Ma 2 Mingchen Li 3 Yanfang Ye 2 Chuxu Zhang 1, † \\dagger Affiliation: 1 University of Connecticut 2 University of Notre Dame 3 University of Massachusetts, Amherst {jiatan.huang,chuxu.zhang}@uconn.edu † \\dagger Corresponding author" } ] }, { "id": "2610.12794", "batch": "20261012-four-track-review-and-frozen16", "evidence": [ { "path": ".cache/oct12-scan/fulltext/2610.12794.json", "line": 60, "field": "affiliation", "header": "Seung Won Wilson Yoo Affiliation: Snowflake Chang Zeng † † thanks: Work done during internship at Snowflake. Affiliation: UMass Amherst Aditi Somayajula Affiliation: Stanford Sahithi Ankireddy Affiliation: Stanford Rohan Kosalge Affiliation: UC Berkeley Avishree Khare Affiliation: UPenn John Mitchell Affiliation: Snowflake Affiliation: Stanford Aditya G. Parameswaran Affiliation: Snowflake Affiliation: UC Berkeley Rajhans Samdani Affiliation: Snowflake Anupam Datta Affiliation: Snowflake" } ] }, { "id": "2610.12957", "batch": "20261012-four-track-review-and-frozen16", "evidence": [ { "path": ".cache/oct12-scan/fulltext/2610.12957.json", "line": 67, "field": "affiliation", "header": "Gangyi Li LuYao Wang Jingfu Sun Jingyi Yuan Hang Liu Hao Zong Yujia Hu Tingting Gao Jianyong Wang Xu Yang Affiliation: JD.COM" } ] }, { "id": "2610.13144", "batch": "20261012-four-track-review-and-frozen16", "evidence": [ { "path": ".cache/oct12-scan/fulltext/2610.13144.json", "line": 54, "field": "affiliation", "header": "Songmiao Wang Yuchen Wang Chenglin Yu Hongxia Yang Ming Li † † thanks: Corresponding author: ming.li@polyu.edu.hk Affiliation: The Hong Kong Polytechnic University" } ] }, { "id": "2610.13166", "batch": "20261012-four-track-review-and-frozen16", "evidence": [ { "path": ".cache/oct12-scan/fulltext/2610.13166.json", "line": 62, "field": "affiliation", "header": "Shun Zou 1,2∗ Ziyu Ma 2∗ Yi Zou 1,2∗ Yong Wang 2†‡ Lin Chen 1 Zehui Chen 1 Guanhua Chen 3 Xiangxiang Chu 2 Feng Zhao 1‡ 1 University of Science and Technology of China 2 AMAP, Alibaba Group 3 Southern University of Science and Technology Porject Page Data & Model" } ] }, { "id": "2610.18157", "batch": "oct15-missed", "evidence": [ { "path": ".cache/oct15-review-closure/paper-slices/2610.18157.json", "line": 6, "field": "front", "header": "Fan Out or Relay? Ownership-Aware Routing for LLM Agents Querying Siloed Tables Zhexi Feng Affiliation: Department of Electrical and Computer Engineering Affiliation: University of California San Diego Affiliation: La Jolla, CA 92093, USA Email: zhf023@ucsd.edu " } ] }, { "id": "2610.18684", "batch": "oct15-missed", "evidence": [ { "path": ".cache/oct15-review-closure/paper-slices/2610.18684.json", "line": 6, "field": "front", "header": "Knowing the User Is Not Knowing What We Established: Grounding Awareness in Personal Agents Jiajun Chen † † thanks: Equal contribution. † † thanks: Work done while at NYU Shanghai. Affiliation: Carnegie Mellon University Xinyue Chen 1 1 footnotemark: 1 Affiliation: Carnegie Mellon University Yicheng Jiang 2 2 footnotemark: 2 Affiliation: New York University Hua Shen Affiliation: New York University Affiliation: NYU Shanghai " } ] }, { "id": "2610.18780", "batch": "oct15-missed", "evidence": [ { "path": ".cache/oct15-review-closure/paper-slices/2610.18780.json", "line": 6, "field": "front", "header": "BLADE: Auditing Information Use for Failure Attribution in LLM Multi-Agent Systems Xiaoguang Guo Yaohao Long Chuxu Zhang † University of Connecticut {xiaoguang.guo, chuxu.zhang}@uconn.edu robinlong5678@gmail.com † Corresponding author " } ] }, { "id": "2610.19131", "batch": "oct15-missed", "evidence": [ { "path": ".cache/oct15-review-closure/paper-slices/2610.19131.json", "line": 6, "field": "front", "header": "Plant Memory: Technical Language Processing and a Verified Maintenance Ontology for Agent-Accessible Failure Knowledge from SAP PM Work Orders Akash Raj Affiliation: M.Tech, School of Artificial Intelligence and Data Science Affiliation: Indian Institute of Technology Jodhpur Email: m25ai1116@iitj.ac.in Affiliation: ORCID: 0009-0005-5666-3967 October 2026 " } ] }, { "id": "2610.19139", "batch": "oct15-missed", "evidence": [ { "path": ".cache/oct15-review-closure/paper-slices/2610.19139.json", "line": 6, "field": "front", "header": "Online Tokenization with Optimal Delay in Linear Time Mina Dalirrooyfard Affiliation: Morgan Stanley Email: minad@mit.edu Christopher Ye † † thanks: Supported by NSF HDR TRIPODS Phase II grant 2217058 (EnCORE Institute). This work was completed as part of an internship at Morgan Stanley Machine Learning Research. Affiliation: University of California San Diego Email: czye@ucsd.edu " } ] }, { "id": "2610.19316", "batch": "oct15-missed", "evidence": [ { "path": ".cache/oct15-review-closure/paper-slices/2610.19316.json", "line": 6, "field": "front", "header": "Truthful and Expected-Payment-Optimal Model Routing in Graph-Based Agentic Workflows Jiayuan Liu 1,2 Mingyu Guo 3 Jiarui Gan 4 Vincent Conitzer 1,2 1 Carnegie Mellon University 2 Foundations of Cooperative AI Lab (FOCAL) 3 Adelaide University 4 University of Oxford jiayuan4@cs.cmu.edu mingyu.guo@adelaide.edu.au jiarui.gan@cs.ox.ac.uk conitzer@cs.cmu.edu " } ] }, { "id": "2610.19345", "batch": "oct15-missed", "evidence": [ { "path": ".cache/oct15-review-closure/paper-slices/2610.19345.json", "line": 6, "field": "front", "header": "When Is Smaller Greener? Right-Sizing Language Models for Agentic AI Journal: JORS Sajid Siraj Email: s.siraj@leeds.ac.uk Address: Leeds University Business School, University of Leeds, Leeds, United Kingdom Address: Ca’Foscari University of Venice, Venice, Italy Maria Barbati Email: maria.barbati@unive.it Corresponding author: Corresponding author Address: Ca’Foscari University of Venice, Venice, Italy " } ] }, { "id": "2610.19474", "batch": "oct15-missed", "evidence": [ { "path": ".cache/oct15-review-closure/paper-slices/2610.19474.json", "line": 6, "field": "front", "header": "Beyond Retrieval: Evidence Access and Event Aggregation in Agent Memory Zerui Yang † † thanks: Corresponding author Quan Xue Chaorui Zhang Xueyan Niu † † thanks: Corresponding author Xuanyu Zheng Affiliation: Xiyuan Wang , Haolin Ruan , Guangcheng Song , Weixi Zhang , Wei Han , Bo Bai Affiliation: Huawei Technologies Ltd. Affiliation: Correspondence: richard.yangpolyu@gmail.com , zhang.chaorui@huawei.com " } ] }, { "id": "2610.19659", "batch": "oct15-missed", "evidence": [ { "path": ".cache/oct15-review-closure/paper-slices/2610.19659.json", "line": 6, "field": "front", "header": "Measuring and Reducing Unusable Replies of LLM Agents Yilin Jiang Affiliation: East China Normal University, Shanghai, China Affiliation: The Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China Email: ftan@mail.ecnu.edu.cn Fei Tan † † thanks: Corresponding author. Affiliation: East China Normal University, Shanghai, China Xuanyu Yin Affiliation: East China Normal University, Shanghai, China Guanchu Wang Affiliation: Shanghai Artificial Intelligence Laboratory, Shanghai, China Xia Hu Affiliation: Shanghai Artificial Intelligence Laboratory, Shanghai, China Yutao Yue Affiliation: The Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China Affiliation: Institute of Deep Perception Technology, JITRI, Wuxi, China " } ] }, { "id": "2610.19801", "batch": "oct15-missed", "evidence": [ { "path": ".cache/oct15-review-closure/paper-slices/2610.19801.json", "line": 6, "field": "front", "header": "Delta Tokenizer: Reference-Identical Tokenization of Session Deltas and Segment-Parallelism for Long Prompts on Multicore CPUs Zhenyu Zhang Affiliation: Arizona State University Email: zzhan641@asu.edu Zhichao Cao Affiliation: Arizona State University Email: Zhichao.Cao@asu.edu " } ] }, { "id": "2610.19818", "batch": "oct15-missed", "evidence": [ { "path": ".cache/oct15-review-closure/paper-slices/2610.19818.json", "line": 6, "field": "front", "header": "One Monitor Does Not Fit All: Formalism-Agnostic Runtime Verification for Agentic AI Angelo Ferrando Affiliation: University of Modena and Reggio Emilia, Modena, Italy Email: angelo.ferrando@unimore.it " } ] }, { "id": "2610.20067", "batch": "oct15-missed", "evidence": [ { "path": ".cache/oct15-review-closure/paper-slices/2610.20067.json", "line": 6, "field": "front", "header": "Which Knowledge Should LLMs Trust? Uncovering a Generalizable Steering Direction Across Knowledge Conflicts Yibin Lou Affiliation: State Key Laboratory of Multimedia Information Processing, Peking University Affiliation: School of Computer Science, Peking University Affiliation: Southern University of Science and Technology Email: 12310513@mail.sustech.edu.cn Yi Hu Affiliation: Institute for Artificial Intelligence, Peking University Jianhui Chen Affiliation: School of Computer Science, Peking University Chenpeng Wang Affiliation: YiXin-AI Lab, YIXIN, Beijing, China Muhan Zhang Affiliation: Institute for Artificial Intelligence, Peking University Liangming Pan † † thanks: Corresponding author: liangmingpan@pku.edu.cn Affiliation: State Key Laboratory of Multimedia Information Processing, Peking University Affiliation: School of Computer Science, Peking University " } ] }, { "id": "2610.20303", "batch": "oct15-missed", "evidence": [ { "path": ".cache/oct15-review-closure/paper-slices/2610.20303.json", "line": 6, "field": "front", "header": "SpectrumQuant: Spectrum Alignment for LLM Quantization Dwaipayan Haldar † † thanks: These authors contributed equally to this work. Affiliation: Indian Institute of Science Affiliation: Bengaluru-560012, Karnataka, India Email: dwaipayanh@iisc.ac.in Indranil Patra Note: Indian Institute of Science Affiliation: Bengaluru-560012, Karnataka, India Email: indranilp@iisc.ac.in Azhar Yousuf Note: Indian Institute of Science Affiliation: Bengaluru-560012, Karnataka, India Email: azharyousuf@iisc.ac.in Chandra Sekhar Seelamantula Affiliation: Indian Institute of Science Affiliation: Bengaluru-560012, Karnataka, India Email: css@iisc.ac.in " } ] }, { "id": "2610.20413", "batch": "oct15-missed", "evidence": [ { "path": ".cache/oct15-review-closure/paper-slices/2610.20413.json", "line": 6, "field": "front", "header": "When Does Momentum Help in Reinforcement Learning with Verifiable Rewards? Kaifei Wang † † thanks: Peking University. Email: {wkf5094,shanzikang}@stu.pku.edu.cn , wanglw@pku.edu.cn . Zikang Shan 1 1 footnotemark: 1 Liwei Wang 1 1 footnotemark: 1 Yinyu Ye † † thanks: Shanghai Jiao Tong University, SIMIS, and Stanford University. Email: yinyu-ye@stanford.edu . Han Zhong † † thanks: Shanghai Jiao Tong University. Email: han.zhong@sjtu.edu.cn . " } ] } ], "missing_ids": [ "2510.15831", "2602.16490", "2610.00010", "2610.00012", "2610.00025", "2610.00040", "2610.00097", "2610.00182", "2610.00204", "2610.00251", "2610.00313", "2610.00354", "2610.00372", "2610.00385", "2610.00399", "2610.00400", "2610.00431", "2610.00432", "2610.00435", "2610.00487", "2610.00497", "2610.00499", "2610.00541", "2610.00559", "2610.00568", "2610.00573", "2610.00574", "2610.00604", "2610.00613", "2610.00648", "2610.00651", "2610.00661", "2610.00663", "2610.00666", "2610.00675", "2610.00710", "2610.00717", "2610.00767", "2610.00779", "2610.00864", "2610.00878", "2610.00885", "2610.00899", "2610.00952", "2610.00977", "2610.00979", "2610.00981", "2610.00983", "2610.01017", "2610.01023", "2610.01037", "2610.01054", "2610.01076", "2610.01127", "2610.01128", "2610.01172", "2610.01192", "2610.01238", "2610.01243", "2610.01278", "2610.01286", "2610.01316", "2610.01492", "2610.01496", "2610.01499", "2610.01506", "2610.01508", "2610.01513", "2610.01537", "2610.01560", "2610.01564", "2610.01569", "2610.01595", "2610.01605", "2610.01625", "2610.01637", "2610.01640", "2610.01649", "2610.01670", "2610.01687", "2610.01710", "2610.01741", "2610.01769", "2610.01815", "2610.01892", "2610.01917", "2610.01921", "2610.01955", "2610.01962", "2610.02005", "2610.02019", "2610.02091", "2610.02140", "2610.02150", "2610.02161", "2610.02190", "2610.02191", "2610.02196", "2610.02202", "2610.02204", "2610.02206", "2610.02298", "2610.02304", "2610.02323", "2610.02324", "2610.02330", "2610.02344", "2610.02349", "2610.02351", "2610.02359", "2610.02361", "2610.02388", "2610.02396", "2610.02404", "2610.02425", "2610.02444", "2610.02460", "2610.02462", "2610.02472", "2610.02478", "2610.02480", "2610.02523", "2610.02525", "2610.02542", "2610.02569", "2610.02588", "2610.02593", "2610.02597", "2610.02614", "2610.02616", "2610.02617", "2610.02638", "2610.02657", "2610.02666", "2610.02670", "2610.02683", "2610.02687", "2610.02695", "2610.02697", "2610.02700", "2610.02710", "2610.02713", "2610.02718", "2610.02740", "2610.02781", "2610.02814", "2610.02828", "2610.02832", "2610.02835", "2610.02840", "2610.02841", "2610.02856", "2610.02876", "2610.02880", "2610.02885", "2610.02887", "2610.02925", "2610.02928", "2610.02951", "2610.02952", "2610.02970", "2610.02986", "2610.03002", "2610.03014", "2610.03020", "2610.03034", "2610.03036", "2610.03039", "2610.03063", "2610.03084", "2610.03089", "2610.03099", "2610.03102", "2610.03123", "2610.03190", "2610.03198", "2610.03199", "2610.03213", "2610.03226", "2610.03315", "2610.03389", "2610.03448", "2610.03516", "2610.03524", "2610.03574", "2610.03585", "2610.03591", "2610.03631", "2610.03632", "2610.03695", "2610.03741", "2610.03780", "2610.03796", "2610.03820", "2610.03894", "2610.03938", "2610.03984", "2610.03994", "2610.04002", "2610.04008", "2610.04009", "2610.04047", "2610.04118", "2610.04132", "2610.04134", "2610.04139", "2610.04140", "2610.04152", "2610.04178", "2610.04196", "2610.04204", "2610.04225", "2610.04261", "2610.04292", "2610.04299", "2610.04303", "2610.04318", "2610.04328", "2610.04344", "2610.04371", "2610.04375", "2610.04379", "2610.04381", "2610.04409", "2610.04429", "2610.04432", "2610.04488", "2610.04506", "2610.04512", "2610.04517", "2610.04518", "2610.04555", "2610.04575", "2610.04580", "2610.04596", "2610.04604", "2610.04607", "2610.04611", "2610.04616", "2610.04641", "2610.04646", "2610.04651", "2610.04658", "2610.04687", "2610.04699", "2610.04721", "2610.04753", "2610.04779", "2610.04792", "2610.04794", "2610.04805", "2610.04824", "2610.04845", "2610.04851", "2610.04868", "2610.04882", "2610.04889", "2610.04899", "2610.04915", "2610.04916", "2610.04918", "2610.04921", "2610.04927", "2610.04940", "2610.04950", "2610.04961", "2610.04991", "2610.05023", "2610.05024", "2610.05025", "2610.05026", "2610.05030", "2610.05039", "2610.05047", "2610.05066", "2610.05097", "2610.05099", "2610.05107", "2610.05124", "2610.05126", "2610.05132", "2610.05139", "2610.05162", "2610.05176", "2610.05185", "2610.05190", "2610.05247", "2610.05273", "2610.05274", "2610.05282", "2610.05284", "2610.05295", "2610.05300", "2610.05303", "2610.05308", "2610.05318", "2610.05334", "2610.05342", "2610.05367", "2610.05373", "2610.05383", "2610.05398", "2610.05402", "2610.05418", "2610.05425", "2610.05473", "2610.05481", "2610.05484", "2610.05532", "2610.05538", "2610.05540", "2610.05563", "2610.05572", "2610.05615", "2610.05622", "2610.05630", "2610.05637", "2610.05664", "2610.05685", "2610.05697", "2610.05709", "2610.05715", "2610.05744", "2610.05774", "2610.05778", "2610.05833", "2610.05842", "2610.05861", "2610.05872", "2610.05879", "2610.05885", "2610.05896", "2610.05897", "2610.05935", "2610.05940", "2610.05978", "2610.05981", "2610.05993", "2610.05994", "2610.06001", "2610.06021", "2610.06026", "2610.06064", "2610.06100", "2610.06147", "2610.06161", "2610.06191", "2610.06192", "2610.06193", "2610.06226", "2610.06269", "2610.06286", "2610.06324", "2610.06404", "2610.06406", "2610.06413", "2610.06439", "2610.06446", "2610.06454", "2610.06479", "2610.06598", "2610.06611", "2610.06616", "2610.06617", "2610.06636", "2610.06637", "2610.06647", "2610.06648", "2610.06650", "2610.06666", "2610.06671", "2610.06686", "2610.06729", "2610.06748", "2610.06750", "2610.06751", "2610.06782", "2610.06804", "2610.06817", "2610.06829", "2610.06830", "2610.06843", "2610.06844", "2610.06892", "2610.06932", "2610.06947", "2610.06950", "2610.06963", "2610.06964", "2610.06966", "2610.06993", "2610.07043", "2610.07118", "2610.07121", "2610.07132", "2610.07177", "2610.07191", "2610.07212", "2610.07247", "2610.07258", "2610.07261", "2610.07289", "2610.07311", "2610.07324", "2610.07332", "2610.07335", "2610.07348", "2610.07349", "2610.07376", "2610.07385", "2610.07389", "2610.07457", "2610.07522", "2610.07532", "2610.07533", "2610.07557", "2610.07572", "2610.07585", "2610.07587", "2610.07639", "2610.07643", "2610.07645", "2610.07654", "2610.07657", "2610.07689", "2610.07706", "2610.07729", "2610.07731", "2610.07739", "2610.07758", "2610.07767", "2610.07774", "2610.07809", "2610.07810", "2610.07816", "2610.07819", "2610.07832", "2610.07848", "2610.07851", "2610.07860", "2610.07862", "2610.07868", "2610.07887", "2610.07898", "2610.07903", "2610.07925", "2610.07935", "2610.07940", "2610.07972", "2610.07987", "2610.07990", "2610.08048", "2610.08076", "2610.08101", "2610.08106", "2610.08108", "2610.08133", "2610.08155", "2610.08159", "2610.08162", "2610.08164", "2610.08183", "2610.08341", "2610.08364", "2610.08401", "2610.08403", "2610.08414", "2610.08452", "2610.08463", "2610.08514", "2610.08560", "2610.08586", "2610.08621", "2610.08622", "2610.08639", "2610.08647", "2610.08649", "2610.08659", "2610.08662", "2610.08670", "2610.08674", "2610.08678", "2610.08691", "2610.08713", "2610.08718", "2610.08773", "2610.08775", "2610.08778", "2610.08779", "2610.08791" ], "google_meta_positive": [ { "id": "2610.09159", "title": "SpecGuard: Proving a Task Is Broken Before the Agent Cheats", "evidence": { "path": ".cache/oct12-scan/fulltext/2610.09159.json", "line": 71, "field": "affiliation", "header": "Param Biyani MATS parambiyani8@gmail.com Krishnamurthy (Dj) Dvijotham Google DeepMind" } }, { "id": "2610.09484", "title": "MIMESIS: Learning User Simulators as Training Environments for Interactive Agents", "evidence": { "path": ".cache/oct12-scan/fulltext/2610.09484.json", "line": 64, "field": "affiliation", "header": "Hoang Phan Affiliation: Meta Superintelligence Labs Affiliation: New York University Work done at Meta Dat Huynh Affiliation: Meta Superintelligence Labs Andrey Zhmoginov Qi Zeng Affiliation: Meta Superintelligence Labs Wancen Mu Affiliation: Meta Superintelligence Labs Yue Cao Affiliation: Meta Superintelligence Labs Shengjie Bi Affiliation: Meta Superintelligence Labs Yun He Affiliation: Meta Superintelligence Labs Changdae Oh Affiliation: Meta Superintelligence Labs Affiliation: University of Wisconsin - Madison Work done at Meta Deren Lei Affiliation: Meta Superintelligence Labs" } } ], "primary_web_supplemented_ids": [ "2610.07289", "2610.06616", "2610.00497", "2602.16490", "2510.15831" ], "total_raw_author_block_coverage_including_web": 50, "remaining_no_raw_author_block": 485, "warning": "oct07 490独立原文缓存缺失；不是535篇机构穷尽核验。使用模型不作为机构证据。" }, "priority_pending_shortlist": [ { "id": "2610.07289", "title": "Catching Developers in the Flow: Low-Latency Agentic Program Repair at Google Scale", "ledger_evidence": { "path": "docs/paper-discovery-ledger.json", "line": 19305 }, "review_evidence": { "path": "docs/research-audits/oct07-review-closure/paper-register.json", "line": 8694 }, "repo_status": "implementation-pending", "manifest_found": false, "actual_implementation_or_metrics_found": false, "source_batch": "oct07-fixed-window-review-closure", "affiliation": "一作及全部作者均Google（New York）；官方HTML作者块。", "upstream_code": "Data Availability明确无法公开reproduction package或数据；未发现作者源码。", "low_cost_object": "可另定公开APR任务上的忠实有界核心：pre/post abstention过滤、真实LLM ReAct generate-and-validate、实际补丁应用/测试执行与时延记录。不能声称内部Google生产复现；尚未证明该独立小实验全部成本。", "full_experiment_cost": "完整生产验证依赖私有源码、内部CI/开发工具、内部模型和数百万changes；公共复现不可得，应关闭完整原环境复现而不是无限实施欠账。", "recommended_disposition": "保留已有accepted但明确分出公共核心范围再评估；完整Google生产实验不复现（私有资源），不是待GPU队列。", "web_evidence": [ { "url": "https://arxiv.org/html/2610.07289v1", "locations": [ "作者块 line53", "§3 filters/ReAct", "Data Availability lines454-456" ] } ] }, { "id": "2610.06616", "title": "Video Encoders Built on Image Representations", "ledger_evidence": { "path": "docs/paper-discovery-ledger.json", "line": 21553 }, "review_evidence": { "path": "docs/research-audits/oct07-review-closure/paper-register.json", "line": 11663 }, "repo_status": "implementation-pending", "manifest_found": false, "actual_implementation_or_metrics_found": false, "source_batch": "oct07-fixed-window-review-closure", "affiliation": "一作Stanford；DeepMind共同作者仅advisory capacity，不能宣传Google一作论文。", "upstream_code": "所读官方HTML未发现论文作者源码仓库；未据此宣称绝对未发布。", "low_cost_object": "原文Appendix C.1明确training-free selection-only独立变体：对真实冻结图像encoder特征保留source position，执行question relevance/DPP diversity/cross-frame correspondence的固定token预算选择；不带learned refiner。可先做CPU算法恒等/预算/源位置诊断；真实7B/8B checkpoint小样本需另核资源与GPU门禁，不把随机特征当能力复现。", "full_experiment_cost": "Appendix C.3：65K Video-ChatGPT子集、2epochs、8×A100×12h=96GPUh，仅refiner；另三backbone/13benchmark。完整实验成本过高，不复现（建议审查DONE）；低成本selection-only不能被96GPUh统一否定。", "recommended_disposition": "关闭完整refiner/全表实验；将selection-only作为明确保真边界的可负担核心候选，未实现。", "web_evidence": [ { "url": "https://arxiv.org/html/2610.06616v1", "locations": [ "作者块 lines61-62", "Appendix C.1 lines333-355", "Appendix C.3 lines369-372" ] } ] }, { "id": "2610.00497", "title": "Gumbel Straight Flow: Distilling Autoregressive Models into One-step Flow Maps", "ledger_evidence": { "path": "docs/paper-discovery-ledger.json", "line": 40409 }, "review_evidence": { "path": "docs/research-audits/oct07-review-closure/paper-register.json", "line": 37202 }, "repo_status": "implementation-pending", "manifest_found": false, "actual_implementation_or_metrics_found": false, "source_batch": "oct07-fixed-window-review-closure", "affiliation": "原文作者脚注：一作工作作为Google DeepMind Amsterdam Student Researcher完成。", "upstream_code": "所读官方HTML无作者实现仓库链接；未作全网穷尽未发布断言。", "low_cost_object": "可另定小词表/短序列真实AR teacher+flow-map student的L1：posterior Gumbel sampling、单调Gaussianization、diagonal CE与off-diagonal semigroup self-consistency、实际梯度和1-step sampling。必须两个真实训练模型且保留核心损失，不可仅直线插值fixture；该缩小实验仍未估算/实现。", "full_experiment_cost": "Appendix C原文batch512，teacher/GSF-FM/GSF最高1M steps；主学生167.8M，另多数据/条件任务及teacher训练。完整公平campaign估计远超短时单A100/A30，成本过高不复现（建议审查DONE）。", "recommended_disposition": "完整实验高成本关闭；廉价公式/训练机制仅为待定有界候选，不是现成GPU交付承诺。", "web_evidence": [ { "url": "https://arxiv.org/html/2610.00497v1", "locations": [ "作者脚注 line65", "Appendix C Algorithms1-2 lines546-593", "Training details lines594-598", "Table9 line617" ] } ] }, { "id": "2602.16490", "title": "From Growing to Looping: A Unified View of Iterative Computation in LLMs", "ledger_evidence": { "path": "docs/paper-discovery-ledger.json", "line": 41715 }, "review_evidence": { "path": "docs/research-audits/oct07-review-closure/paper-register.json", "line": 38978 }, "repo_status": "implementation-pending", "manifest_found": false, "actual_implementation_or_metrics_found": false, "source_batch": "oct07-fixed-window-review-closure", "affiliation": "TUM/Helmholtz AI/Google Paradigms of Intelligence合作；非可确认为Google单独主导。", "upstream_code": "既有原审及本轮所读官方HTML未找到作者实现仓库。", "low_cost_object": "可另定小Transformer真实训练机制：复制中间层及AdamW状态实现depth-growing，再在真实前向上复用该block做inference looping；记录参数/训练/推理预算，并只在validation选择loop配置。仅实现复制/loop不等于验证论文推理收益，完整缩小训练方案还需廉价成本核定。", "full_experiment_cost": "正文§4.1：SmolLM360M/1.7B分别约200B/400B tokens；22benchmark、growing/looping配方与后续训练对照。完整训练明显超当前单卡/本机预算，成本过高不复现（建议审查DONE）。", "recommended_disposition": "关闭原规模训练；有界真实小模型机制需单独验收，不把200B预训练留为实施欠账。", "web_evidence": [ { "url": "https://arxiv.org/html/2602.16490v1", "locations": [ "作者/机构 lines52-53", "§4.1 line130-131", "§3.1-3.2/Table1", "Appendix A" ] } ] }, { "id": "2510.15831", "title": "VISTA: A Test-Time Self-Improving Video Generation Agent", "ledger_evidence": { "path": "docs/paper-discovery-ledger.json", "line": 41744 }, "review_evidence": { "path": "docs/research-audits/oct07-review-closure/paper-register.json", "line": 39011 }, "repo_status": "implementation-pending", "manifest_found": false, "actual_implementation_or_metrics_found": false, "source_batch": "oct07-fixed-window-review-closure", "affiliation": "官方project：一作Google Student Researcher/NUS，其他Google作者。", "upstream_code": "官方project只见论文、demo和prompts，未见作者源码入口；所读页面不是全网穷尽。", "low_cost_object": "能廉价验证的只是tournament/champion/三维批评/DTPA提示修订编排合同；没有真实视频生成和多模态judge时只能L0/诊断，不是核心复现。定义性路径仍需每轮实际重生成视频，因此本轮不认为已存在可直接接单的低成本能力实现。", "full_experiment_cost": "正文§4：100 MovieGen prompts+161内部prompts，Gemini2.5Flash+Veo3；§4.3每轮约0.7M tokens+28视频，视频生成token不包含。多轮、多baseline完整campaign高成本，且内部数据不可得；建议完整实验成本过高不复现（审查DONE）。", "recommended_disposition": "完整VISTA复现高成本关闭；保留资料，不将固定视频/分数fixture当实现或evolve能力。", "web_evidence": [ { "url": "https://arxiv.org/html/2510.15831v1", "locations": [ "§4 Experiments lines250-257", "§4.3 Cost Analysis lines324-327" ] }, { "url": "https://g-vista.github.io/", "locations": [ "作者机构 lines2-5", "项目公开范围" ] } ] }, { "id": "2610.09159", "title": "SpecGuard: Proving a Task Is Broken Before the Agent Cheats", "ledger_evidence": { "path": "docs/paper-discovery-ledger.json", "line": 43318 }, "review_evidence": { "path": "docs/research-audits/oct12-review-closure/agent-reviews.json", "line": 10859 }, "repo_status": "implementation-pending", "manifest_found": false, "actual_implementation_or_metrics_found": false, "source_batch": "20261012-four-track-review-and-frozen16", "affiliation": "一作Param Biyani为MATS；合作者Krishnamurthy Dvijotham为Google DeepMind，非Google一作。", "upstream_code": "作者真实公共Apache2.0仓库https://github.com/prmbiy/specguard，含src/conflict_certifier、configs、data、scripts、tests、RESULTS。", "low_cost_object": "优先可落地的忠实小范围：真实模型从D+C但不读T生成spec.lean；独立TestAgent生成tests.lean；真实Certifier连接并交Lean kernel，保留conflict/no-conflict/inconclusive和语义忠实性审计。少量公开/自建任务可不训练模型；不得手写gold规范绕过autoformalization或把Lean通过等同原意正确。", "full_experiment_cost": "§5同模型三Agent，SpecAgent120bash turns，各Agent10次Lean反馈，task timeout1h；349SWE任务+模型/重复/基线+187VERINA+95LiveCodeBench等。论文报告$0.26/task，仅349任务单臂估$90.74、三重复约$272.22（由论文均费推算，不是本地支出），完整campaign多小时应成本关闭；不能据此否定真实少量任务核心。", "recommended_disposition": "保留小样本真实Lean核心作为最优先可负担候选，仍未实现；完整多模型全基准不承诺复现。", "web_evidence": [ { "url": "https://arxiv.org/html/2610.09159v1", "locations": [ "作者块 line84", "方法§4 lines164-181", "实验limits lines195-200", "费用 line217" ] }, { "url": "https://github.com/prmbiy/specguard", "locations": [ "root dirs lines187-203", "README Pipeline lines218-225" ] } ] } ], "priority_pending_shortlist_boundary": "6项为现存证据中明确Google关联且当前未交付的短名单，不宣称535机构全覆盖；5项结构化oct07接受+1项oct12关键词推断。全文成本建议为本轮只读审计建议，仓库未回写终态。" }