{
  "schema_version": 3,
  "paper": {
    "title": "ExeWorldBench: Benchmarking Code-Based Executable 3D World Modeling",
    "kind": "benchmark",
    "primary_source_ids": [
      "paper",
      "dataset",
      "repo"
    ],
    "earliest_arxiv_date": "2026-10-10",
    "earliest_arxiv_timestamp_utc": "2026-10-10T20:34:57Z"
  },
  "artifacts": {
    "article": "content/_posts/090-exeworldbench.md",
    "deck": "lib/decks/exeworldbench.html",
    "embed_src": "/lib/decks/exeworldbench.html"
  },
  "sources": [
    {
      "id": "paper",
      "kind": "paper",
      "title": "ExeWorldBench: Benchmarking Code-Based Executable 3D World Modeling",
      "url": "https://arxiv.org/pdf/2610.15166v1",
      "revision": "arXiv:2610.15166v1",
      "used_for": [
        "problem",
        "method",
        "results",
        "limitations"
      ]
    },
    {
      "id": "arxiv-entry",
      "kind": "metadata",
      "title": "ExeWorldBench arXiv v1 submission history and reuse license",
      "url": "https://arxiv.org/abs/2610.15166v1",
      "revision": "v1: 2026-10-10T20:34:57Z",
      "used_for": [
        "title",
        "published-date",
        "license"
      ]
    },
    {
      "id": "repo",
      "kind": "code",
      "title": "ExeWorldBench official repository",
      "url": "https://github.com/Helen1p/ExeWorldBench/tree/06bb54244a7240da61c23f2429efc865c5c2197c",
      "revision": "06bb54244a7240da61c23f2429efc865c5c2197c",
      "used_for": [
        "release-boundary",
        "availability"
      ]
    },
    {
      "id": "dataset",
      "kind": "dataset",
      "title": "ExeWorldBench tasks, generation prompts and judge prompts",
      "url": "https://huggingface.co/datasets/HelenPeng/ExeWorldBench/tree/c6dc85e412af91046b5a58d783324df7baf846f4",
      "revision": "c6dc85e412af91046b5a58d783324df7baf846f4",
      "used_for": [
        "task-counts",
        "visible-inputs",
        "judge",
        "real-cases"
      ]
    },
    {
      "id": "blender",
      "kind": "documentation",
      "title": "Blender RigidBodyObject Python API",
      "url": "https://docs.blender.org/api/current/bpy.types.RigidBodyObject.html",
      "revision": "current documentation inspected 2026-10-15",
      "used_for": [
        "Part 0 background"
      ]
    },
    {
      "id": "blender-ops",
      "kind": "documentation",
      "title": "Blender rigidbody operators",
      "url": "https://docs.blender.org/api/current/bpy.ops.rigidbody.html",
      "revision": "current documentation inspected 2026-10-15",
      "used_for": [
        "Part 0 background"
      ]
    },
    {
      "id": "physcode",
      "kind": "paper",
      "title": "PhysCodeBench: Benchmarking Physics-Aware Symbolic Simulation of 3D Scenes via Self-Corrective Multi-Agent Refinement",
      "url": "https://arxiv.org/abs/2604.23580v2",
      "revision": "arXiv:2604.23580v2",
      "used_for": [
        "related-work: abstract and Sections 3–4"
      ]
    },
    {
      "id": "worldcoder",
      "kind": "paper",
      "title": "WorldCoder-Bench: Benchmarking Physically Grounded 3D World Synthesis",
      "url": "https://arxiv.org/abs/2606.01869",
      "revision": "arXiv release inspected 2026-10-15",
      "used_for": [
        "related-work: abstract and Sections 2–3"
      ]
    },
    {
      "id": "visphy",
      "kind": "paper",
      "title": "VisPhyWorld: Probing Physical Reasoning via Code-Driven Video Reconstruction",
      "url": "https://arxiv.org/abs/2602.13294",
      "revision": "arXiv abstract inspected 2026-10-15",
      "used_for": [
        "related-work: abstract only"
      ]
    },
    {
      "id": "mpmworlds",
      "kind": "paper",
      "title": "MPMWorlds: Material-Point-Method Simulations for Inferring and Extrapolating Physical Dynamics",
      "url": "https://arxiv.org/abs/2606.01538",
      "revision": "arXiv abstract inspected 2026-10-15",
      "used_for": [
        "related-work: abstract only"
      ]
    },
    {
      "id": "3dcode",
      "kind": "paper",
      "title": "3DCodeBench: Benchmarking Agentic Procedural 3D Modeling Via Code",
      "url": "https://arxiv.org/abs/2606.01057",
      "revision": "arXiv abstract inspected 2026-10-15",
      "used_for": [
        "related-work: abstract only"
      ]
    },
    {
      "id": "pace",
      "kind": "paper",
      "title": "PACE-Bench: Benchmarking Physics Adaptation via Code Evolution in Dynamic Environments",
      "url": "https://arxiv.org/abs/2608.14441",
      "revision": "arXiv abstract inspected 2026-10-15",
      "used_for": [
        "related-work: abstract only"
      ]
    },
    {
      "id": "editorial-reference",
      "kind": "editorial",
      "title": "GroundingPI：把精确定位训练成物理智能的感知基础 · Zhixiong Zhang",
      "url": "https://rookiexiong7.github.io/blog/posts/groundingpi/",
      "revision": "inspected 2026-10-15",
      "used_for": [
        "layout and teaching choices only; not a scientific evidence source"
      ]
    }
  ],
  "paper_reading": {
    "main_agent_complete_read": true,
    "appendices_included": true,
    "extracted_text_reviewed": true,
    "rendered_pages_reviewed": true,
    "page_count": 21,
    "rendered_page_ranges": [
      {
        "start": 1,
        "end": 21
      }
    ],
    "source_visuals_expected": 18,
    "visual_inventory_complete": true,
    "reading_evidence": "Complete extracted text and all 21 rendered pages inspected, including original figures and tables and task specifications. No numbered Algorithm present."
  },
  "note_sections": [
    {
      "role": "problem",
      "locator": "1. Motivation：程序能运行，机械装置未必能工作"
    },
    {
      "role": "related-work",
      "locator": "2. Related Work：任务更强调机械组合，评分仍有自己的边界"
    },
    {
      "role": "method",
      "locator": "3. Benchmark / Protocol：从任务包到一份评分"
    },
    {
      "role": "experiments",
      "locator": "4. Experiments：修好执行错误，仍接不通机械过程"
    },
    {
      "role": "future-work",
      "locator": "5. Discussion：什么已经说明，什么还值得补实验"
    },
    {
      "role": "summary",
      "locator": "6. Conclusion：值得读的是任务设计，读分数要带着条件"
    }
  ],
  "terminology": [
    {
      "term": "bpy",
      "kind": "operation",
      "source_ids": [
        "blender",
        "blender-ops"
      ],
      "source_locators": [
        {
          "source_id": "blender",
          "locator": "RigidBodyObject reference"
        },
        {
          "source_id": "blender-ops",
          "locator": "rigidbody.object_add"
        }
      ],
      "article_quote": "Python 接口 bpy 可以创建物体、设置碰撞与关节、推进模拟并保存场景。"
    },
    {
      "term": "keyframe",
      "kind": "operation",
      "source_ids": [
        "blender",
        "paper"
      ],
      "source_locators": [
        {
          "source_id": "blender",
          "locator": "RigidBodyObject.kinematic and animation context"
        },
        {
          "source_id": "paper",
          "locator": "§3.1 and public task prohibitions"
        }
      ],
      "article_quote": "给物体设置关键帧（keyframe），是直接指定“第几帧到哪个位置”。"
    },
    {
      "term": "harness",
      "kind": "actor",
      "source_ids": [
        "paper",
        "dataset"
      ],
      "source_locators": [
        {
          "source_id": "paper",
          "locator": "§4.2 and harness-based rows Tables 1–4"
        },
        {
          "source_id": "dataset",
          "locator": "t3/task2/053/053.txt opening tool/outputs specification"
        }
      ],
      "article_quote": "把模型接到文件系统、shell、Blender 等工具并组织多轮执行的外层软件，在这里称为 harness"
    },
    {
      "term": "VLM judge",
      "kind": "model-call",
      "source_ids": [
        "paper",
        "dataset"
      ],
      "source_locators": [
        {
          "source_id": "paper",
          "locator": "§4.1, p.5"
        },
        {
          "source_id": "dataset",
          "locator": "eval/t1/prompt1.txt visual inputs and JSON output"
        }
      ],
      "article_quote": "论文用 Opus 5.5 作为 <strong>VLM judge</strong>：一次独立的多模态模型调用读取评分提示和图像，输出结构化分数。"
    },
    {
      "term": "Executable Causal Mechanism",
      "kind": "metric",
      "source_ids": [
        "paper",
        "dataset"
      ],
      "source_locators": [
        {
          "source_id": "paper",
          "locator": "§4.1, p.6"
        },
        {
          "source_id": "dataset",
          "locator": "eval/t2/prompt1.txt and eval/t3/prompt1_1.txt causal segments"
        }
      ],
      "article_quote": "Executable Causal Mechanism（<code>ECM</code>，0–10）看组件动作是否完成，以及前一级是否通过机械作用触发后一级。"
    },
    {
      "term": "Goal Satisfaction",
      "kind": "metric",
      "source_ids": [
        "paper",
        "dataset"
      ],
      "source_locators": [
        {
          "source_id": "paper",
          "locator": "§4.1, Equations (1)–(2)"
        },
        {
          "source_id": "dataset",
          "locator": "eval/t3/prompt2_1.txt Initial/Behavior/Final conditions"
        }
      ],
      "article_quote": "Goal Satisfaction（<code>G</code>）用于 T3-2。judge 分别对初始状态 I、中间行为 B、最终状态 F 的每个条件给 0 或 1。"
    },
    {
      "term": "object_A",
      "kind": "state",
      "source_ids": [
        "dataset",
        "paper"
      ],
      "source_locators": [
        {
          "source_id": "dataset",
          "locator": "t3/task2/053/053.txt Required Objects and R05"
        },
        {
          "source_id": "paper",
          "locator": "task 053, pp.20–21"
        }
      ],
      "article_quote": "object_A 控制 B 的进入；A 完全离开共享区时，应通过物理作用改变它的位置或姿态，从而放行 B。"
    }
  ],
  "coverage": [
    {
      "id": "central-claim",
      "claim_type": "author-claim",
      "source_ids": [
        "paper"
      ],
      "source_locators": [
        {
          "source_id": "paper",
          "locator": "pp.1–2, Abstract and Introduction; §4.2"
        }
      ],
      "evidence": [
        {
          "artifact": "article",
          "quote": "作者要测的是：模型理解了一个装置之后，能否把它真正搭出来并运行完整过程。"
        },
        {
          "artifact": "article",
          "quote": "执行率与机械正确性明显脱节。"
        }
      ]
    },
    {
      "id": "task-provenance",
      "claim_type": "inspected-implementation",
      "source_ids": [
        "paper",
        "dataset"
      ],
      "source_locators": [
        {
          "source_id": "paper",
          "locator": "§3.1–3.2, pp.3–5; Appendix A.1, Table 5"
        },
        {
          "source_id": "dataset",
          "locator": "repository file inventory: t1/task (50), t2/task (100), t3/task1 (63), t3/task2 (60); README.md"
        }
      ],
      "evidence": [
        {
          "artifact": "article",
          "quote": "T1、T2 还提供加入运动参考帧的变体，但它们是同一道题的另一种输入设置，不能再次计入总题数。"
        },
        {
          "artifact": "article",
          "quote": "T1 从公开机械动画中采样。"
        },
        {
          "artifact": "article",
          "quote": "T2 由作者先制作有预定义运动的原子组件代码库。"
        },
        {
          "artifact": "article",
          "quote": "当前公开版本没有给出足以量化污染程度的审计。"
        }
      ]
    },
    {
      "id": "annotation-and-check-construction",
      "claim_type": "author-claim",
      "source_ids": [
        "paper",
        "dataset"
      ],
      "source_locators": [
        {
          "source_id": "paper",
          "locator": "§3.2.1–3.2.3, pp.4–5; Figure 1"
        },
        {
          "source_id": "dataset",
          "locator": "eval/t1/prompt1.txt; eval/t2/prompt1.txt; eval/t3/prompt1_1.txt and prompt2_1.txt"
        }
      ],
      "evidence": [
        {
          "artifact": "article",
          "quote": "作者采用模型辅助构建、过滤与人工复核，而不是把全部任务和检查都称为人工标注。"
        },
        {
          "artifact": "article",
          "quote": "最后五名人工标注者排除仍然难以推断或有歧义的题目。"
        },
        {
          "artifact": "article",
          "quote": "论文也未报告这五名标注者的逐题分歧或一致性系数。"
        },
        {
          "artifact": "article",
          "quote": "检查 judge 与人工评分的一致性、已知错误实现的检出率。"
        }
      ]
    },
    {
      "id": "public-vs-hidden-inputs",
      "claim_type": "inspected-implementation",
      "source_ids": [
        "paper",
        "dataset"
      ],
      "source_locators": [
        {
          "source_id": "paper",
          "locator": "§3.1–3.2, §4.2"
        },
        {
          "source_id": "dataset",
          "locator": "README.md; t1/task/000/000.txt; t1/task_image/000/000.txt; t2/task/04_01/04_01.txt; api_prompt/t1/prompt.txt; api_prompt/t2/prompt.txt; t3/task1/12_10/12_10.txt; t3/task2/053/053.txt"
        }
      ],
      "evidence": [
        {
          "artifact": "article",
          "quote": "三条构建线索不在公开 T1 主设置的模型输入中"
        },
        {
          "artifact": "article",
          "quote": "T1/T2 主设置不提供完整参考动画；T1 构建线索和 T2 组件实现也不提供给被测 Agent。"
        },
        {
          "artifact": "article",
          "quote": "公开生成提示没有将最终评分反馈交给 Agent 的步骤"
        },
        {
          "artifact": "article",
          "quote": "没有公开操作系统层的文件隔离配置。"
        }
      ]
    },
    {
      "id": "evaluation-formulas",
      "claim_type": "reproduced-calculation",
      "source_ids": [
        "paper",
        "dataset"
      ],
      "source_locators": [
        {
          "source_id": "paper",
          "locator": "§4.1, p.6, Equations (1)–(2); Tables 1/6 and 4/9"
        },
        {
          "source_id": "dataset",
          "locator": "eval/t2/prompt1.txt: causal segments and cutoff; eval/t3/prompt1_1.txt: N segments; eval/t3/prompt2_1.txt: binary I/B/F and overall integer ECM"
        }
      ],
      "evidence": [
        {
          "artifact": "article",
          "quote": "G 是先按题算条件满足比例，再对题取平均。"
        },
        {
          "artifact": "article",
          "quote": "G_i = (P_i^I + P_i^B + P_i^F) / (N_i^I + N_i^B + N_i^F)"
        },
        {
          "artifact": "article",
          "quote": "主表让执行失败的题记零，分母保留全部题目"
        },
        {
          "artifact": "article",
          "quote": "0.94 × 5.40 = 5.08"
        },
        {
          "artifact": "article",
          "quote": "某处断链后，无法由连续机械传递到达的后续段全部记 0"
        }
      ]
    },
    {
      "id": "checker-code-and-cases",
      "claim_type": "inspected-implementation",
      "source_ids": [
        "paper",
        "dataset",
        "repo"
      ],
      "source_locators": [
        {
          "source_id": "paper",
          "locator": "Appendix A.3, Figure 9 p.19; Appendix A.4, task specifications pp.20–21"
        },
        {
          "source_id": "dataset",
          "locator": "t3/task2/053/053.txt: Initial, R01–R13, Final; eval/t3/prompt2_1.txt; full tree metadata, 2117 siblings without executable runner files"
        },
        {
          "source_id": "repo",
          "locator": "06bb542: README.md and tracked file inventory (web assets only)"
        }
      ],
      "evidence": [
        {
          "artifact": "article",
          "quote": "它属于前述模型辅助生成、校验、人工复核的数据路径，但没有公开逐题署名的标注记录。"
        },
        {
          "artifact": "article",
          "quote": "E_start_B < E_enter_A"
        },
        {
          "artifact": "article",
          "quote": "R06 又单独要求 A 完全离开"
        },
        {
          "artifact": "article",
          "quote": "公开评分提示是视觉判断，未提供读取这些几何事件的程序实现。"
        },
        {
          "artifact": "article",
          "quote": "Rules 如何转换成评分提示的 B 条目，逐项映射没有公开"
        },
        {
          "artifact": "article",
          "quote": "可以按公开提示尝试生成和判分，但还不能完整复现作者的执行与汇总流程。"
        },
        {
          "artifact": "article",
          "quote": "以下是依据规则推导的失败例，不是作者公布的模型日志"
        }
      ]
    },
    {
      "id": "model-results",
      "claim_type": "author-claim",
      "source_ids": [
        "paper"
      ],
      "source_locators": [
        {
          "source_id": "paper",
          "locator": "§4.2–4.7, Tables 1–4, Figures 2–4; Appendix Tables 6–9 and Figures 5–9"
        }
      ],
      "evidence": [
        {
          "artifact": "article",
          "quote": "Codex 达到 G=90.14%、ECM=5.82；Claude Code 为 G=75.19%、ECM=5.17。"
        },
        {
          "artifact": "article",
          "quote": "Kimi 修复后 79% 的代码能执行，但 ECM 仍为 0"
        },
        {
          "artifact": "article",
          "quote": "Astra API 的 ECM=4.08，高于这道题 Codex 的 3.33"
        },
        {
          "artifact": "article",
          "quote": "Astra API 从 5.08 降至 3.40"
        },
        {
          "artifact": "article",
          "quote": "两者不是严格匹配预算的单次生成比较。"
        }
      ]
    },
    {
      "id": "related-work-comparison",
      "claim_type": "interpretation",
      "source_ids": [
        "paper",
        "physcode",
        "worldcoder",
        "visphy",
        "mpmworlds",
        "3dcode",
        "pace"
      ],
      "source_locators": [
        {
          "source_id": "paper",
          "locator": "§2, p.3"
        },
        {
          "source_id": "physcode",
          "locator": "v2 Abstract; §3–4 expert assertion construction, engine state and conservation-law checks"
        },
        {
          "source_id": "worldcoder",
          "locator": "Abstract; §2–3 StateProbe, hidden runtime contracts, mutation hardening"
        },
        {
          "source_id": "visphy",
          "locator": "Abstract, 209 scenes and 108 templates"
        },
        {
          "source_id": "mpmworlds",
          "locator": "Abstract, 2D MPM with deformable objects and fluids"
        },
        {
          "source_id": "3dcode",
          "locator": "Abstract, 12 VLMs and 3DCodeArena"
        },
        {
          "source_id": "pace",
          "locator": "Abstract, 144 source-to-target adaptation pairs and limited attempts"
        }
      ],
      "evidence": [
        {
          "artifact": "article",
          "quote": "需要分别比较任务内容和评价证据。"
        },
        {
          "artifact": "article",
          "quote": "v2 有 1,200 个专家验证样例"
        },
        {
          "artifact": "article",
          "quote": "StateProbe 执行动作并读取内部状态"
        },
        {
          "artifact": "article",
          "quote": "其余四篇的范围与规模来自各自原始摘要，不跨论文比较分数。"
        }
      ]
    },
    {
      "id": "limitations",
      "claim_type": "unknown",
      "source_ids": [
        "paper",
        "dataset",
        "repo"
      ],
      "source_locators": [
        {
          "source_id": "paper",
          "locator": "§4.1–4.7, pp.5–10: budgets/repeats not reported; Appendix A.1–A.4, pp.13–21"
        },
        {
          "source_id": "dataset",
          "locator": "eval/t3/prompt2_1.txt: no actual intervention required, overall ECM schema; t3/task2/053/053.txt: no final-hold duration, B mapping or movement threshold"
        },
        {
          "source_id": "repo",
          "locator": "tracked file inventory at 06bb542; no runner, repair diffs or component implementation"
        }
      ],
      "evidence": [
        {
          "artifact": "article",
          "quote": "评分提示中的“如果移除 A，B 是否仍会动”也是辅助推理问题"
        },
        {
          "artifact": "article",
          "quote": "它不是评价器真的移除了零件再运行一遍。"
        },
        {
          "artifact": "article",
          "quote": "图中没有各级样本量与误差条"
        },
        {
          "artifact": "article",
          "quote": "它们不是同题逐步去提示的受控实验。"
        },
        {
          "artifact": "article",
          "quote": "不能分离模型能力、执行反馈、更多尝试与外部修复各自的贡献。"
        }
      ]
    }
  ],
  "source_visuals": [
    {
      "id": "figure-1",
      "kind": "figure",
      "label": "Figure 1",
      "source_id": "paper",
      "source_locator": "p.4, figure 1",
      "status": "reused",
      "rationale": "T1/T2/T3 construction paths; original makes who generated, inferred, validated and reviewed tasks visible.",
      "placements": [
        {
          "artifact": "article",
          "locator": "Figure 1 · 数据构建。"
        },
        {
          "artifact": "deck",
          "locator": "T1 怎样保证一张静态图有足够运动线索"
        }
      ],
      "permission_status": "verified-allowed",
      "permission_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.4, CC BY 4.0",
      "asset_paths": [
        "/lib/papers/exeworldbench/figure-1.png",
        "/lib/papers/exeworldbench/figure-1-t1-source.png",
        "/lib/papers/exeworldbench/figure-1-t1-validation.png",
        "/lib/papers/exeworldbench/figure-1-t2-library.png",
        "/lib/papers/exeworldbench/figure-1-t2-composition.png"
      ]
    },
    {
      "id": "figure-2",
      "kind": "figure",
      "label": "Figure 2",
      "source_id": "paper",
      "source_locator": "p.9, figure 2",
      "status": "reused",
      "rationale": "Mechanical-stage grouping of ECM and nonmonotonic long-horizon behavior, including uncertainty limits.",
      "placements": [
        {
          "artifact": "article",
          "locator": "Figure 2 · 横轴 Step"
        },
        {
          "artifact": "deck",
          "locator": "机械级数增加，ECM 总体下降"
        }
      ],
      "permission_status": "verified-allowed",
      "permission_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.9, CC BY 4.0",
      "asset_paths": [
        "/lib/papers/exeworldbench/figure-2.png"
      ]
    },
    {
      "id": "figure-3",
      "kind": "figure",
      "label": "Figure 3",
      "source_id": "paper",
      "source_locator": "p.9, figure 3",
      "status": "reused",
      "rationale": "T1 motion-reference intervention including the Astra negative result; original values preserved.",
      "placements": [
        {
          "artifact": "article",
          "locator": "Figure 3 · T1 加入 motion reference。"
        }
      ],
      "permission_status": "verified-allowed",
      "permission_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.9, CC BY 4.0",
      "asset_paths": [
        "/lib/papers/exeworldbench/figure-3.png"
      ]
    },
    {
      "id": "figure-4",
      "kind": "figure",
      "label": "Figure 4",
      "source_id": "paper",
      "source_locator": "p.10, figure 4",
      "status": "reused",
      "rationale": "T2 motion-reference results, limited gains and negative cases.",
      "placements": [
        {
          "artifact": "article",
          "locator": "Figure 4 · T2 加入运动参考。"
        },
        {
          "artifact": "deck",
          "locator": "增加运动帧通常帮助有限"
        }
      ],
      "permission_status": "verified-allowed",
      "permission_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.10, CC BY 4.0",
      "asset_paths": [
        "/lib/papers/exeworldbench/figure-4.png"
      ]
    },
    {
      "id": "figure-5",
      "kind": "figure",
      "label": "Figure 5",
      "source_id": "paper",
      "source_locator": "p.15, figure 5",
      "status": "reused",
      "rationale": "Category-specific I/B/F and overall goal radar panels with different denominators.",
      "placements": [
        {
          "artifact": "article",
          "locator": "Figure 5 · I / B / F 类别表现。"
        }
      ],
      "permission_status": "verified-allowed",
      "permission_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.15, CC BY 4.0",
      "asset_paths": [
        "/lib/papers/exeworldbench/figure-5.png"
      ]
    },
    {
      "id": "figure-6",
      "kind": "figure",
      "label": "Figure 6",
      "source_id": "paper",
      "source_locator": "p.16, figure 6",
      "status": "reused",
      "rationale": "Original T1 reference and outputs; four times per model, not four alternative designs.",
      "placements": [
        {
          "artifact": "article",
          "locator": "Figure 6 · T1 的机构还原案例。"
        }
      ],
      "permission_status": "verified-allowed",
      "permission_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.16, CC BY 4.0",
      "asset_paths": [
        "/lib/papers/exeworldbench/figure-6.png"
      ]
    },
    {
      "id": "figure-7",
      "kind": "figure",
      "label": "Figure 7",
      "source_id": "paper",
      "source_locator": "p.17, figure 7",
      "status": "reused",
      "rationale": "T2 model output case showing Astra API above Codex in a particular task.",
      "placements": [
        {
          "artifact": "article",
          "locator": "Figure 7 · T2 代表例。"
        }
      ],
      "permission_status": "verified-allowed",
      "permission_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.17, CC BY 4.0",
      "asset_paths": [
        "/lib/papers/exeworldbench/figure-7.png"
      ]
    },
    {
      "id": "figure-8",
      "kind": "figure",
      "label": "Figure 8",
      "source_id": "paper",
      "source_locator": "p.18, figure 8",
      "status": "reused",
      "rationale": "Released 12_10 task tied to original 12-component and two-ball comparison.",
      "placements": [
        {
          "artifact": "article",
          "locator": "Figure 8 · T3-1 的真实任务"
        }
      ],
      "permission_status": "verified-allowed",
      "permission_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.18, CC BY 4.0",
      "asset_paths": [
        "/lib/papers/exeworldbench/figure-8.png"
      ]
    },
    {
      "id": "figure-9",
      "kind": "figure",
      "label": "Figure 9",
      "source_id": "paper",
      "source_locator": "p.19, figure 9",
      "status": "reused",
      "rationale": "Released 053 crossing task tied to V/ECM/G values; full figure plus Codex detail for initial orientation.",
      "placements": [
        {
          "artifact": "article",
          "locator": "Figure 9 ·"
        },
        {
          "artifact": "deck",
          "locator": "G=100% 仍可能只有 ECM=5"
        }
      ],
      "permission_status": "verified-allowed",
      "permission_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.19, CC BY 4.0",
      "asset_paths": [
        "/lib/papers/exeworldbench/figure-9.png",
        "/lib/papers/exeworldbench/figure-9-detail-codex.png"
      ]
    },
    {
      "id": "table-1",
      "kind": "table",
      "label": "Table 1",
      "source_id": "paper",
      "source_locator": "p.7, table 1",
      "status": "reused",
      "rationale": "All-task T1 results with executability and structural/motion metrics before and after repair.",
      "placements": [
        {
          "artifact": "article",
          "locator": "Table 1 · T1 主表"
        }
      ],
      "permission_status": "verified-allowed",
      "permission_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.7, CC BY 4.0",
      "asset_paths": [
        "/lib/papers/exeworldbench/table-1.png"
      ]
    },
    {
      "id": "table-2",
      "kind": "table",
      "label": "Table 2",
      "source_id": "paper",
      "source_locator": "p.7, table 2",
      "status": "reused",
      "rationale": "All-task T2 results preserve every model and column; key executable-but-zero-ECM result.",
      "placements": [
        {
          "artifact": "article",
          "locator": "Table 2 · T2 主表"
        },
        {
          "artifact": "deck",
          "locator": "能运行的程序，仍可能接不通整条链"
        }
      ],
      "permission_status": "verified-allowed",
      "permission_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.7, CC BY 4.0",
      "asset_paths": [
        "/lib/papers/exeworldbench/table-2.png"
      ]
    },
    {
      "id": "table-3",
      "kind": "table",
      "label": "Table 3",
      "source_id": "paper",
      "source_locator": "p.8, table 3",
      "status": "reused",
      "rationale": "All-task T3-1 results, separate from T2 due to different task distributions.",
      "placements": [
        {
          "artifact": "article",
          "locator": "Table 3 · T3-1 主表"
        }
      ],
      "permission_status": "verified-allowed",
      "permission_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.8, CC BY 4.0",
      "asset_paths": [
        "/lib/papers/exeworldbench/table-3.png"
      ]
    },
    {
      "id": "table-4",
      "kind": "table",
      "label": "Table 4",
      "source_id": "paper",
      "source_locator": "p.8, table 4",
      "status": "reused",
      "rationale": "All-task T3-2 results retain E/V/ECM/G distinctions.",
      "placements": [
        {
          "artifact": "article",
          "locator": "Table 4 · T3-2 主表"
        },
        {
          "artifact": "deck",
          "locator": "Codex 的 90.14% 是条件满足率"
        }
      ],
      "permission_status": "verified-allowed",
      "permission_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.8, CC BY 4.0",
      "asset_paths": [
        "/lib/papers/exeworldbench/table-4.png"
      ]
    },
    {
      "id": "table-5",
      "kind": "table",
      "label": "Table 5",
      "source_id": "paper",
      "source_locator": "p.13, table 5",
      "status": "reused",
      "rationale": "T2 component frequency table, 850 occurrences rather than 850 tasks.",
      "placements": [
        {
          "artifact": "article",
          "locator": "Table 5 · 组件出现次数"
        }
      ],
      "permission_status": "verified-allowed",
      "permission_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.13, CC BY 4.0",
      "asset_paths": [
        "/lib/papers/exeworldbench/table-5.png"
      ]
    },
    {
      "id": "table-6",
      "kind": "table",
      "label": "Table 6",
      "source_id": "paper",
      "source_locator": "p.13, table 6",
      "status": "reused",
      "rationale": "T1 executable-only quality denominator and recomputation of penalized M.",
      "placements": [
        {
          "artifact": "article",
          "locator": "Table 6 · T1，conditioned quality。"
        }
      ],
      "permission_status": "verified-allowed",
      "permission_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.13, CC BY 4.0",
      "asset_paths": [
        "/lib/papers/exeworldbench/table-6.png"
      ]
    },
    {
      "id": "table-7",
      "kind": "table",
      "label": "Table 7",
      "source_id": "paper",
      "source_locator": "p.14, table 7",
      "status": "reused",
      "rationale": "T2 executable-only quality shows mechanics still weak after failures removed.",
      "placements": [
        {
          "artifact": "article",
          "locator": "Table 7 · T2，conditioned quality。"
        }
      ],
      "permission_status": "verified-allowed",
      "permission_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.14, CC BY 4.0",
      "asset_paths": [
        "/lib/papers/exeworldbench/table-7.png"
      ]
    },
    {
      "id": "table-8",
      "kind": "table",
      "label": "Table 8",
      "source_id": "paper",
      "source_locator": "p.14, table 8",
      "status": "reused",
      "rationale": "T3-1 executable-only quality distinguishes before/after repair populations.",
      "placements": [
        {
          "artifact": "article",
          "locator": "Table 8 · T3-1，conditioned quality。"
        }
      ],
      "permission_status": "verified-allowed",
      "permission_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.14, CC BY 4.0",
      "asset_paths": [
        "/lib/papers/exeworldbench/table-8.png"
      ]
    },
    {
      "id": "table-9",
      "kind": "table",
      "label": "Table 9",
      "source_id": "paper",
      "source_locator": "p.15, table 9",
      "status": "reused",
      "rationale": "T3-2 executable-only goal scores; reweight Kimi 49/60 to all-task score.",
      "placements": [
        {
          "artifact": "article",
          "locator": "Table 9 · T3-2，conditioned quality。"
        }
      ],
      "permission_status": "verified-allowed",
      "permission_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.15, CC BY 4.0",
      "asset_paths": [
        "/lib/papers/exeworldbench/table-9.png"
      ]
    }
  ],
  "figures": [
    {
      "path": "/lib/papers/exeworldbench/figure-1.png",
      "origin": "paper-original",
      "source_visual_id": "figure-1",
      "source_id": "paper",
      "source_locator": "p.4, figure 1",
      "license_status": "verified-allowed",
      "license_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.4, CC BY 4.0",
      "used_in": [
        "article"
      ]
    },
    {
      "path": "/lib/papers/exeworldbench/figure-1-t1-source.png",
      "origin": "paper-original",
      "source_visual_id": "figure-1",
      "source_id": "paper",
      "source_locator": "p.4, Figure 1 T1 stages 1–3; original-pixel bounds (0,0,529,531), explicitly labeled crop",
      "license_status": "verified-allowed",
      "license_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.4, CC BY 4.0; detail crop only, original retained in article",
      "used_in": [
        "deck"
      ]
    },
    {
      "path": "/lib/papers/exeworldbench/figure-1-t2-library.png",
      "origin": "paper-original",
      "source_visual_id": "figure-1",
      "source_id": "paper",
      "source_locator": "p.4, Figure 1 T2 component library; original-pixel bounds (529,0,1148,399), explicitly labeled crop",
      "license_status": "verified-allowed",
      "license_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.4, CC BY 4.0; detail crop only, original retained in article",
      "used_in": [
        "deck"
      ]
    },
    {
      "path": "/lib/papers/exeworldbench/figure-1-t1-validation.png",
      "origin": "paper-original",
      "source_visual_id": "figure-1",
      "source_id": "paper",
      "source_locator": "p.4, Figure 1 T1 stages 4–5; original-pixel bounds (0,531,529,924), explicitly labeled crop",
      "license_status": "verified-allowed",
      "license_basis": "arXiv:2610.15166v1 links to CC BY 4.0, verified 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.4, CC BY 4.0; stage crop only, original retained in article",
      "used_in": ["deck"]
    },
    {
      "path": "/lib/papers/exeworldbench/figure-1-t2-composition.png",
      "origin": "paper-original",
      "source_visual_id": "figure-1",
      "source_id": "paper",
      "source_locator": "p.4, Figure 1 T2 stages 2–5; original-pixel bounds (529,399,1148,924), explicitly labeled crop",
      "license_status": "verified-allowed",
      "license_basis": "arXiv:2610.15166v1 links to CC BY 4.0, verified 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.4, CC BY 4.0; stage crop only, original retained in article",
      "used_in": ["deck"]
    },
    {
      "path": "/lib/papers/exeworldbench/figure-2.png",
      "origin": "paper-original",
      "source_visual_id": "figure-2",
      "source_id": "paper",
      "source_locator": "p.9, figure 2",
      "license_status": "verified-allowed",
      "license_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.9, CC BY 4.0",
      "used_in": [
        "article",
        "deck"
      ]
    },
    {
      "path": "/lib/papers/exeworldbench/figure-3.png",
      "origin": "paper-original",
      "source_visual_id": "figure-3",
      "source_id": "paper",
      "source_locator": "p.9, figure 3",
      "license_status": "verified-allowed",
      "license_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.9, CC BY 4.0",
      "used_in": [
        "article"
      ]
    },
    {
      "path": "/lib/papers/exeworldbench/figure-4.png",
      "origin": "paper-original",
      "source_visual_id": "figure-4",
      "source_id": "paper",
      "source_locator": "p.10, figure 4",
      "license_status": "verified-allowed",
      "license_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.10, CC BY 4.0",
      "used_in": [
        "article",
        "deck"
      ]
    },
    {
      "path": "/lib/papers/exeworldbench/figure-5.png",
      "origin": "paper-original",
      "source_visual_id": "figure-5",
      "source_id": "paper",
      "source_locator": "p.15, figure 5",
      "license_status": "verified-allowed",
      "license_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.15, CC BY 4.0",
      "used_in": [
        "article"
      ]
    },
    {
      "path": "/lib/papers/exeworldbench/figure-6.png",
      "origin": "paper-original",
      "source_visual_id": "figure-6",
      "source_id": "paper",
      "source_locator": "p.16, figure 6",
      "license_status": "verified-allowed",
      "license_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.16, CC BY 4.0",
      "used_in": [
        "article"
      ]
    },
    {
      "path": "/lib/papers/exeworldbench/figure-7.png",
      "origin": "paper-original",
      "source_visual_id": "figure-7",
      "source_id": "paper",
      "source_locator": "p.17, figure 7",
      "license_status": "verified-allowed",
      "license_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.17, CC BY 4.0",
      "used_in": [
        "article"
      ]
    },
    {
      "path": "/lib/papers/exeworldbench/figure-8.png",
      "origin": "paper-original",
      "source_visual_id": "figure-8",
      "source_id": "paper",
      "source_locator": "p.18, figure 8",
      "license_status": "verified-allowed",
      "license_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.18, CC BY 4.0",
      "used_in": [
        "article"
      ]
    },
    {
      "path": "/lib/papers/exeworldbench/figure-9.png",
      "origin": "paper-original",
      "source_visual_id": "figure-9",
      "source_id": "paper",
      "source_locator": "p.19, figure 9",
      "license_status": "verified-allowed",
      "license_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.19, CC BY 4.0",
      "used_in": [
        "article",
        "deck"
      ]
    },
    {
      "path": "/lib/papers/exeworldbench/figure-9-detail-codex.png",
      "origin": "paper-original",
      "source_visual_id": "figure-9",
      "source_id": "paper",
      "source_locator": "p.19, figure 9 (explicitly labeled detail crop)",
      "license_status": "verified-allowed",
      "license_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.19, CC BY 4.0; detail crop only, original retained in article",
      "used_in": [
        "article",
        "deck"
      ]
    },
    {
      "path": "/lib/papers/exeworldbench/table-1.png",
      "origin": "paper-original",
      "source_visual_id": "table-1",
      "source_id": "paper",
      "source_locator": "p.7, table 1",
      "license_status": "verified-allowed",
      "license_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.7, CC BY 4.0",
      "used_in": [
        "article"
      ]
    },
    {
      "path": "/lib/papers/exeworldbench/table-2.png",
      "origin": "paper-original",
      "source_visual_id": "table-2",
      "source_id": "paper",
      "source_locator": "p.7, table 2",
      "license_status": "verified-allowed",
      "license_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.7, CC BY 4.0",
      "used_in": [
        "article",
        "deck"
      ]
    },
    {
      "path": "/lib/papers/exeworldbench/table-3.png",
      "origin": "paper-original",
      "source_visual_id": "table-3",
      "source_id": "paper",
      "source_locator": "p.8, table 3",
      "license_status": "verified-allowed",
      "license_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.8, CC BY 4.0",
      "used_in": [
        "article"
      ]
    },
    {
      "path": "/lib/papers/exeworldbench/table-4.png",
      "origin": "paper-original",
      "source_visual_id": "table-4",
      "source_id": "paper",
      "source_locator": "p.8, table 4",
      "license_status": "verified-allowed",
      "license_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.8, CC BY 4.0",
      "used_in": [
        "article",
        "deck"
      ]
    },
    {
      "path": "/lib/papers/exeworldbench/table-5.png",
      "origin": "paper-original",
      "source_visual_id": "table-5",
      "source_id": "paper",
      "source_locator": "p.13, table 5",
      "license_status": "verified-allowed",
      "license_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.13, CC BY 4.0",
      "used_in": [
        "article"
      ]
    },
    {
      "path": "/lib/papers/exeworldbench/table-6.png",
      "origin": "paper-original",
      "source_visual_id": "table-6",
      "source_id": "paper",
      "source_locator": "p.13, table 6",
      "license_status": "verified-allowed",
      "license_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.13, CC BY 4.0",
      "used_in": [
        "article"
      ]
    },
    {
      "path": "/lib/papers/exeworldbench/table-7.png",
      "origin": "paper-original",
      "source_visual_id": "table-7",
      "source_id": "paper",
      "source_locator": "p.14, table 7",
      "license_status": "verified-allowed",
      "license_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.14, CC BY 4.0",
      "used_in": [
        "article"
      ]
    },
    {
      "path": "/lib/papers/exeworldbench/table-8.png",
      "origin": "paper-original",
      "source_visual_id": "table-8",
      "source_id": "paper",
      "source_locator": "p.14, table 8",
      "license_status": "verified-allowed",
      "license_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.14, CC BY 4.0",
      "used_in": [
        "article"
      ]
    },
    {
      "path": "/lib/papers/exeworldbench/table-9.png",
      "origin": "paper-original",
      "source_visual_id": "table-9",
      "source_id": "paper",
      "source_locator": "p.15, table 9",
      "license_status": "verified-allowed",
      "license_basis": "arXiv:2610.15166v1 Rights to this article links to CC BY 4.0; verified from arXiv abs and license deed on 2026-10-15. https://arxiv.org/abs/2610.15166v1; https://creativecommons.org/licenses/by/4.0/",
      "credit": "Peng et al., ExeWorldBench v1, p.15, CC BY 4.0",
      "used_in": [
        "article"
      ]
    }
  ],
  "qa": {
    "reading_layout": {
      "deck_first": true,
      "toc_h2_only": true,
      "surface": "neutral-light",
      "body_font": "serif",
      "body_font_px": 16,
      "line_height": 1.85,
      "figures_centered": true,
      "article_visuals_interleaved": true,
      "internal_scroll_for_tables_and_code": true,
      "page_horizontal_overflow": false
    },
    "viewports": [
      {
        "width": 1600,
        "height": 1000,
        "passed": true
      },
      {
        "width": 1512,
        "height": 982,
        "passed": true
      },
      {
        "width": 390,
        "height": 844,
        "passed": true
      }
    ],
    "tested_surfaces": {
      "article_body": true,
      "embedded_deck": true,
      "standalone_deck": true
    },
    "language_review": {
      "technical_terms_contextualized": true,
      "actors_disambiguated": true,
      "read_aloud_complete": true,
      "facts_rechecked_after_rewrite": true
    },
    "interactions": {
      "buttons": true,
      "keyboard": true,
      "hash_navigation": true,
      "fullscreen": true,
      "escape": true,
      "standalone_return_to_article": true,
      "immersive_mode_exit": true,
      "usable_state_after_exit": true
    },
    "inspected_slides": [
      "opening",
      "densest",
      "largest-table-or-code",
      "final"
    ],
    "browser_mode": "isolated task-owned headless Chromium; no user browser attachment",
    "validated_at_utc": "2026-10-15T08:08:02Z",
    "source_sha256": {
      "article": "e7f06ca380c445eafff2e2b40558630af68bfdc78ee659844c2bcdb8470302e7",
      "deck": "7df87e728e2651f736e485cb9195662563cc05bff493df1945e37f53fd1644b8",
      "editorial_css": "739bfdb45c63e3320d5a05f6687bf0c7f37f4a654ae9f1fa10075a6666a37690"
    },
    "evidence": {
      "evidence_store": "Retained in the local task QA workspace; screenshots and raw browser logs are not deployed.",
      "screenshots": "output/playwright/exeworldbench-{article,embedded,standalone}-{1600,1512,390}-NN.png",
      "actual_size_checks": "Embedded slides 4, 5, 6, 7 and 14 inspected individually at 898px; dense originals, article prose and comparison table inspected at inline size. Full article scroll captures and slide contact sheets reviewed at all specified widths.",
      "mobile_footer_checks": "All 18 embedded slides scrolled to their source footers, captured and checked for reachable controls and zero horizontal page overflow.",
      "interaction_script": "tmp/exeworldbench/interaction-qa.js",
      "final_interaction_result": "All 14 interaction groups passed; errors=[]; failedRequests=[]. Includes exact Copy text, light/dark refresh, six TOC anchors, image fit/clear/zoom/drag, focus restoration, stable return, and nested DOM-fullscreen/Escape exits.",
      "syntax_and_arithmetic": "HTML inline JavaScript parses; Python teaching examples syntax-checked; score-example.py rerun: 4.75, 62.50%, 66.67%, 5.08, 18.13%."
    },
    "external_service_boundary": {
      "local_comments": "Twikoo form and empty comment list rendered in final local checks; no comment was submitted.",
      "earlier_errors": "Earlier captures encountered Twikoo ERR_CONNECTION_CLOSED and an opaque rejection 0. Raw logs retained in the QA workspace; later final interaction checks were clean.",
      "raw_logs": [
        ".playwright-cli/console-2026-10-15T06-53-00-079Z.log",
        ".playwright-cli/console-2026-10-15T07-43-36-666Z.log"
      ],
      "production_comments": "Independently verified on the production article after release c32cca7 on 2026-10-15: form and empty list rendered; four Twikoo API responses returned 200. No comment submitted. Production verification script retained in the local QA workspace.",
      "production_verification": "All eight production interaction groups passed on release c32cca7: English title and dates, original image loading and centering, assets, comments, image viewer, embedded navigation/focus exit, exact code copy, mobile reading, and standalone return. All 23 original/cropped image resources returned 200. Article/deck page errors and failed-request arrays were empty; raw console retained a standalone-deck favicon.ico 404, addressed by adding the site's existing SVG icon link."
    },
    "limitations": [
      "Blender background snippets syntax-checked only, not executed on Blender 5.2.",
      "No official runner, repair implementation, component library or per-item judge logs released at inspected revisions.",
      "Native OS-window fullscreen is not covered by headless DOM-fullscreen testing."
    ]
  }
}