Add 107 trajectories (phase 2, remaining runs; batch S03)

phase2/ispatialgym/alignment_axial/trajectories/ispatialgym-alignment-axial-i00-privileged--ebkrz3s.json

@@ -0,0 +1,205 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-alignment-axial-i00-privileged--ebkrz3s",
+ "taskKey": "ispatialgym/alignment_axial",
+ "dag_sha256": "b530f6c690202b00faaa215c65e0558d1cf88fb1c92dad0ed74a8f9008f88b8c",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 255,
+   "sim_steps": 242,
+   "rows": 255,
+   "max_rows": 1612,
+   "replay_success": 1,
+   "instance_ok": 1,
+   "steps": 242,
+   "success_step": 242,
+   "symbolic_success": 1,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 1,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.0,
+   "peak_processes": 0.0,
+   "peak_disk_gb": 0.0,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-alignment-axial-i00-privileged--ebkrz3s/replay.mp4",
+  "duration_s": 8.2,
+  "fps": 10.0,
+  "frames": 82,
+  "sha256": "69f43711d903f87fb0b3dde836e7c87022af321bfd23c23bff1146df6df43fd1"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "game_box_aligned": {
+    "method": "visual",
+    "intervals": [
+     [
+      8.1,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s，绿色 Clue 游戏盒平放在白色餐桌上，但其长边相对桌前边斜置，尚未摆正。"
+     },
+     {
+      "t": 6.6,
+      "log_step": null,
+      "text_zh": "观察：视频 6.0–6.6 s 镜头重新转向餐桌；视频 6.6 s 游戏盒仍平放在原桌面区域，长边已近似平行桌前边。视频 6.6–8.1 s 逐帧可见盒子保持此朝向，无夹爪持有盒子；对比依据是盒边与桌边的相对方向，不是画面转动。"
+     },
+     {
+      "t": 8.1,
+      "log_step": null,
+      "text_zh": "推断：inputs/run.json 的评分器字段 success=1、replay_success=1、symbolic_success=1、deterministic=1，支持游戏盒朝向、桌面关系、非持有及连续 45 控制步联合静置均通过。结合视频 8.1 s 末帧的摆正状态，将包含静置窗口的整步完成定位于视频约 8.1 s；无状态日志，无法从画面精确复核速度阈值或首次验收帧，时间精度约 0.5 s。"
+     },
+     {
+      "t": null,
+      "log_step": 44,
+      "text_zh": "观察：log step 43 的脚本先重置；log step 44 用修正后的动作映射完成转动、释放、收臂和保持，并覆盖保存 255 行轨迹；log step 45 检查其形状为 (255,23)。log step 50 的 agent 自述放弃后续侧面接触实验、保留此前版本，故这些实验不计入本次评分轨迹。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：视频约 1.0–6.0 s 镜头偏离游戏盒，未拍到机器人作用端与盒子的接触，且无状态接触日志。log step 44、56 描述右夹爪在盒顶接触并转动，但画面不能确认具体接触方式，因此实际接触保留未知。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "game_box_orientation_error",
+   "unit": "rad",
+   "threshold": 0.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_object_robot_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_object_robot_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_robot_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_joint_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "评分轨迹将绿色 Clue 游戏盒从桌面斜置变为沿桌边横放，游戏盒摆正节点最终完成，评分器确认成功及确定性回放。视频 6.6–8.1 s 支持盒子平放、未被持有并维持摆正，包含静置验收的完成时刻约记为视频 8.1 s。操作期间镜头偏离目标，实际接触方式未知；无状态日志，朝向误差、速度和漂移数值未报告，联合验收依���分器成功字段判断，未把 agent 自测数值填入原生指标。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "游戏盒摆正节点 final=true、视频约 8.1 s 完成与评分器成功字段一致；视频 0 s 斜置、6.6–8.1 s 沿桌边摆正，接触未知合理（镜头偏离），未发现需改动的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:19:36Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-alignment-axial-i00-privileged--ebkrz3s/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-alignment-axial-i00-privileged--ebkrz3s/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "abb54ae7bfa37c827c6147e7f6b6283f4200c477385e7737209a49c239f4391f",
+   "video_sha256": "69f43711d903f87fb0b3dde836e7c87022af321bfd23c23bff1146df6df43fd1",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/alignment_axial/trajectories/ispatialgym-alignment-axial-i00-privileged--gpax5rn.json

@@ -0,0 +1,205 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-alignment-axial-i00-privileged--gpax5rn",
+ "taskKey": "ispatialgym/alignment_axial",
+ "dag_sha256": "b530f6c690202b00faaa215c65e0558d1cf88fb1c92dad0ed74a8f9008f88b8c",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 1075,
+   "sim_steps": 640,
+   "rows": 1075,
+   "max_rows": 1612,
+   "replay_success": 1,
+   "instance_ok": 1,
+   "steps": 640,
+   "success_step": 640,
+   "symbolic_success": 1,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 16.952,
+   "peak_memory_gb": 8.766,
+   "peak_processes": 22,
+   "peak_disk_gb": 1.97,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-alignment-axial-i00-privileged--gpax5rn/replay.mp4",
+  "duration_s": 21.5,
+  "fps": 10.0,
+  "frames": 215,
+  "sha256": "7e6f5f70d8ba6165b98b750168e9f087fac483c23e3c8905e9c281abb0b5a865"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "game_box_aligned": {
+    "method": "visual",
+    "intervals": [
+     [
+      21.3,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–2.0 s，绿色 Clue 游戏盒平放于白色餐桌上，盒边相对桌边明显斜置，初态未摆正。"
+     },
+     {
+      "t": 14.5,
+      "log_step": null,
+      "text_zh": "观察：视频 10.5–17.5 s，机器人右夹爪在游戏盒右侧盒缘附近沿桌面推进，盒子相对桌边逐渐转正，始终留在桌面上；夹爪未包夹盒子。视频 17.5–20.0 s，右夹爪向上撤离，盒子没有随夹爪抬起。"
+     },
+     {
+      "t": 21.3,
+      "log_step": null,
+      "text_zh": "观察：视频 20.0–21.4 s，右夹爪与盒子分离，游戏盒保持平放，封面文字朝向及盒边相对桌边的横放关系与参考成功录像（inputs/oracle/replay.mp4，视频 6.5–7.5 s）一致；逐帧查看视频最后一秒，未见盒子再转动。inputs/run.json 的评分器字段 success=1、replay_success=1、symbolic_success=1、deterministic=1、success_step=640。"
+     },
+     {
+      "t": 21.3,
+      "log_step": 78,
+      "text_zh": "推断：本次评分在第 640 控制步首次通过联合验收；任务说明给出的控制率为 30 Hz，因此对应视频约 640/30=21.33 s，将本节点首次成立时间记为视频约 21.3 s。时间映射已用 log step 78 的抬手命令核对：log step 80 输出抬手前累计 525 控制步，抬手持续 70 控制步，预测视频 17.5–19.8 s，与实际撤手画面一致。该节点包含连续 45 控制步静置要求，故不在盒子刚转正时提前计为完成；完整朝向误差和速度阈值依评分器成功结果推断通过，录像本身不能数值复核。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "推/扫",
+      "note_zh": "观察：视频 10.5–17.5 s，机器人右夹爪的并拢手指抵住绿色 Clue 游戏盒右侧盒缘并沿桌面推进，使盒子转向；视频 17.5–20.0 s 夹爪向上撤离，盒子留在桌面，未见包夹或提起。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "game_box_orientation_error",
+   "unit": "rad",
+   "threshold": 0.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_object_robot_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_object_robot_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_robot_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_joint_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "观察：机器人用右夹爪抵住盒缘推动游戏盒转正后撤手，视频 20.0–21.4 s 盒子保持平放、摆正，评分器判定成功，最终完成 1/1 节点，无失败断点。推断：��静置窗口的正式完成时间约为视频 21.3 s；无状态日志，完整旋转误差、速度和漂移阈值只能依据评分器验收结果，原生指标未报告数值；log step 120 的终态测量来自 agent 自行运行至 1075 控制步的回放，超出本次在 640 控制步结束的评分录像，不作为本次终态数值。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "节点 game_box_aligned 的成立、时间（视频约 21.3 s，由 640 控制步/30 Hz 映射并在撤手画面处核对）和 contact_actual（手指/夹爪·推/扫）均与视频帧一致，成功运行无断点，未发现需改动内容。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:26:14Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-alignment-axial-i00-privileged--gpax5rn/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-alignment-axial-i00-privileged--gpax5rn/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "c91661f6e96ce4aafea1191a391574a792b9447b2188d368fa1663e69ece575b",
+   "video_sha256": "7e6f5f70d8ba6165b98b750168e9f087fac483c23e3c8905e9c281abb0b5a865",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/alignment_axial/trajectories/ispatialgym-alignment-axial-i00-privileged--h99ba4r.json

@@ -0,0 +1,205 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-alignment-axial-i00-privileged--h99ba4r",
+ "taskKey": "ispatialgym/alignment_axial",
+ "dag_sha256": "b530f6c690202b00faaa215c65e0558d1cf88fb1c92dad0ed74a8f9008f88b8c",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 420,
+   "sim_steps": 374,
+   "rows": 420,
+   "max_rows": 1612,
+   "replay_success": 1,
+   "instance_ok": 1,
+   "steps": 374,
+   "success_step": 374,
+   "symbolic_success": 1,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 16.901,
+   "peak_memory_gb": 9.373,
+   "peak_processes": 22,
+   "peak_disk_gb": 1.376,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-alignment-axial-i00-privileged--h99ba4r/replay.mp4",
+  "duration_s": 12.6,
+  "fps": 10.0,
+  "frames": 126,
+  "sha256": "80fe0b0591944f92752e72fe40244c0b6bea3cd2ba08e36579fca215ccb6ed21"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "game_box_aligned": {
+    "method": "visual",
+    "intervals": [
+     [
+      12.5,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 绿色 Clue 游戏盒平放在白色餐桌上，但盒边相对桌边斜置，初始未摆正。"
+     },
+     {
+      "t": 8.8,
+      "log_step": 28,
+      "text_zh": "观察：视频 7.2–8.8 s，右夹爪沿桌面抵推盒子靠机器人一侧的角部，盒子相对桌边转动而未离开桌面；视频 8.8 s 盒子长边已近似平行于桌子短边，封面方向与成功参考视频 7.0 s 的摆正结果一致。此时机器人手臂仍在运动，尚不据此认定完整静置验收成立。"
+     },
+     {
+      "t": 11.0,
+      "log_step": 29,
+      "text_zh": "观察：视频 9.2–11.1 s 右夹爪后撤并抬起，盒子留在原处；视频 11.2–12.5 s（含最后一秒逐帧）所见盒边与封面保持不动，夹爪未夹持盒子，末帧局部盒面被前臂遮挡。log step 29 的脚本在撤臂后保持 90 控制步并保存共 420 行，log step 31 对该交付文件启动新进程回放，log step 34 返回相同终态哈希。"
+     },
+     {
+      "t": 12.5,
+      "log_step": 29,
+      "text_zh": "推断：评分器 reward.success、replay_success、symbolic_success、deterministic 均为 1，success_step=374，支持本目标含静置验收最终成立。任务规定 30 Hz；log step 29 保存 420 行，末尾 90 行保持，因此撤臂结束在累计控制步 330，即视频约 11.0 s，已与视频 11.0–11.1 s 的撤臂结束核对。由评分器成功控制步 374/30≈12.47，取已检查的末帧视频 12.5 s 为节点达成锚点；并非将 agent 工作时间换算成视频时间。无状态日志，精确速度与漂移阈值依评分器成功判定，不能从图像独立量测。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "推/扫",
+      "note_zh": "观察：视频 7.2–8.8 s 右夹爪在绿色 Clue 游戏盒靠机器人一侧的角部接触并沿桌面抵推，使盒子原地转向；视频 8.2–9.2 s 可见两指张开，未合拢包夹盒子，随后撤开。接触点有部分遮挡；推断：结合该段盒边相对桌边的转动与 log step 28 保持夹爪张开并向前推的交付动作脚本，实际方式为推/扫。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "game_box_orientation_error",
+   "unit": "rad",
+   "threshold": 0.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_object_robot_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_object_robot_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_robot_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_joint_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人以张开的右夹爪抵推游戏盒角部，盒子约在视频 8.8 s 由斜置转为沿桌边摆正，随后右臂撤开并保持静止；完整静置验收约在视频 12.5 s 达成，最终完成 1/1 节点。评分器记录成功、回放成功且确定性一致，本次无失败断点。没有状态日志，接触点部分遮挡，完整旋转误差及速度、漂移指标无法独立量测；七项原生指标均未由评分器报告数值，保留为空。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "成功运行，唯一节点 game_box_aligned 判为达成、final=true，与评分器成功及视频所见（盒子被抵推后摆正并静置）一致；时钟换算（控制步 374/30≈12.5 s）有核对说明，接触标注为手指/夹爪·推/扫合理，无失败断点。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:35:24Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-alignment-axial-i00-privileged--h99ba4r/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-alignment-axial-i00-privileged--h99ba4r/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "0549e03647c98a40eaab95825be9883a8dc303e5da7301e63d047b750b3e2a33",
+   "video_sha256": "80fe0b0591944f92752e72fe40244c0b6bea3cd2ba08e36579fca215ccb6ed21",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/alignment_axial/trajectories/ispatialgym-alignment-axial-i00-privileged--pvjmajp.json

@@ -0,0 +1,221 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-alignment-axial-i00-privileged--pvjmajp",
+ "taskKey": "ispatialgym/alignment_axial",
+ "dag_sha256": "b530f6c690202b00faaa215c65e0558d1cf88fb1c92dad0ed74a8f9008f88b8c",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 80,
+   "sim_steps": 80,
+   "rows": 80,
+   "max_rows": 1612,
+   "replay_success": 0,
+   "instance_ok": 1,
+   "steps": 80,
+   "success_step": 0,
+   "symbolic_success": 0,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 29.965,
+   "peak_memory_gb": 18.284,
+   "peak_processes": 35,
+   "peak_disk_gb": 2.392,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-alignment-axial-i00-privileged--pvjmajp/replay.mp4",
+  "duration_s": 2.8,
+  "fps": 10.0,
+  "frames": 28,
+  "sha256": "ca11a55a284b503ed0c4d85a09d195b546769d29f4f00098e77ee62f5738d53a"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "game_box_aligned": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：评分视频 0.0 s 的绿色 Clue 游戏盒平放在白色餐桌上，盒子长边相对桌子近边斜置；与成功参考视频 7.9 s 相同视角下盒子横向摆正的外观不同。"
+     },
+     {
+      "t": 2.7,
+      "log_step": 124,
+      "text_zh": "观察：逐帧查看评分视频 0.0–2.7 s（含视频 1.8–2.7 s 的全部末段帧），盒子始终留在原处并保持同一斜向，只有轻微视角变化，没有夹爪接近、夹持或推动盒子。log step 124 的 agent 脚本录制并保存 80 行 hold_action，未包含抓取或转动段。"
+     },
+     {
+      "t": null,
+      "log_step": 150,
+      "text_zh": "观察：log step 147 检查交付文件仍为 80×23、最后 50 行相同；log step 150 的另一次会话测试输出 after close inhand None 和 GRASP FAILED - kept safe file，未覆盖保底文件。评分器字段 steps=80、symbolic_success=0、replay_success=0；该测试中的抓取动作不是本次评分视频中的动作。"
+     }
+    ],
+    "contact_actual": []
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "game_box_orientation_error",
+   "unit": "rad",
+   "threshold": 0.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_object_robot_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_object_robot_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_robot_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_joint_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "game_box_aligned",
+   "t": 0.0,
+   "log_step": 124,
+   "where": "首步前",
+   "evidence_zh": "观察：评分视频 0.0–2.7 s 中游戏盒从未改变初始斜向，也未见机器人末端进入盒子所在区域；log step 124 的脚本仅保存 80 行保持动作，log step 147 确认仍交付该文件。推断：评分尝试没有启动游戏盒摆正操作，故断在该节点的首步前，而不是会话测试中的抓取失败时刻。"
+  },
+  "phenomena": [
+   {
+    "cat": "没尝试",
+    "item": "没尝试",
+    "t": 0.0,
+    "log_step": 124,
+    "evidence_zh": "观察：评分视频 0.0–2.7 s 全部帧中，游戏盒持续原位斜放，无夹爪操作；log step 124 的交付脚本只有 80 行 hold_action，log step 150 明确保留该文件。推断：最终交付的评分尝试未开始摆正盒子，会话中的抓取尝试没有进入交付轨迹。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "t": null,
+    "log_step": 150,
+    "source_zh": "日志",
+    "evidence_zh": "观察：会话测试 log step 127、128 均输出 after close: inhand None 及 GRASP FAILED - not saved；log step 139 的新���话脚本把成功夹持作为继续转动和保存的条件，log step 150 输出 after close inhand None、GRASP FAILED - kept safe file。agent 在 log step 151 自述抓取均未成功，因此始终保留保底文件。推断：agent 未能实现可用的抓取转动控制，导致最终交付只有保持动作；这些会话结果仅用于解释根因，不作为评分尝试的接触现象或节点进度。",
+    "note_zh": "推断：会话记录可支持控制层面的失败，但不足以区分接触几何、末端轨迹或辅助抓取触发问题，故不指定控制子类。"
+   }
+  ],
+  "gap_zh": "观察：inputs/ 中有评分视频、完整运行记录及成功参考视频，但没有状态日志；评分器未报告 DAG 所列朝向误差、速度或窗口漂移数值，故全部原生指标 value 留空。推断：游戏盒未被操作且仍斜置，足以判节点未达；会话抓取测试未包含在评分视频内，其失败的具体控制机制无法由现有画面复核。",
+  "unmet": {
+   "game_box_aligned": false
+  }
+ },
+ "summary_zh": "评分视频 0.0–2.7 s 中游戏盒始终在桌面原位斜放，游戏盒摆正节点从未达成，最终为 0/1；断点为首步前，现象为没尝试。log step 124、147、150 表明，agent 在会话抓取控制未成功后交付并保留了 80 行保持动作，而没有交付抓取转动段。没有状态日志或原生误差读数，控制失败的具体子类无法确定。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "标注与证据一致：评分视频 0–2.7 s 盒子始终原位斜放，节点未达；断点首步前/没尝试，根因控制来自会话抓取失败日志（符合 FAIL-22），log step id 引用正确。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:47:17Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-alignment-axial-i00-privileged--pvjmajp/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-alignment-axial-i00-privileged--pvjmajp/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "33e167f7f44f3df326f87ac0c7b599401d588f0b58bfe065a509caef47d26016",
+   "video_sha256": "ca11a55a284b503ed0c4d85a09d195b546769d29f4f00098e77ee62f5738d53a",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/alignment_axial/trajectories/ispatialgym-alignment-axial-i00-privileged--xstibnf.json

@@ -0,0 +1,205 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-alignment-axial-i00-privileged--xstibnf",
+ "taskKey": "ispatialgym/alignment_axial",
+ "dag_sha256": "b530f6c690202b00faaa215c65e0558d1cf88fb1c92dad0ed74a8f9008f88b8c",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 200,
+   "sim_steps": 184,
+   "rows": 200,
+   "max_rows": 1612,
+   "replay_success": 1,
+   "instance_ok": 1,
+   "steps": 184,
+   "success_step": 184,
+   "symbolic_success": 1,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 16.941,
+   "peak_memory_gb": 10.291,
+   "peak_processes": 22,
+   "peak_disk_gb": 1.211,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-alignment-axial-i00-privileged--xstibnf/replay.mp4",
+  "duration_s": 6.3,
+  "fps": 10.0,
+  "frames": 63,
+  "sha256": "580c7f40cdbbf9c5d35d2599f9aa47a4baad35c1e1e13423e4e335d77da27203"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "game_box_aligned": {
+    "method": "visual",
+    "intervals": [
+     [
+      6.2,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–0.2 s，绿色 Clue 游戏盒平放在白色餐桌上，盒子长边相对桌前沿斜置；初态尚未摆正。"
+     },
+     {
+      "t": 4.8,
+      "log_step": null,
+      "text_zh": "观察：视频 3.7–4.8 s 相机逐渐转回餐桌，游戏盒重新完整进入画面；视频 4.8–6.2 s 盒子平放桌面，长边近似平行桌前沿，封面文字朝向与 oracle 视频 7.0–8.0 s 的成功终态一致。视频 5.3–6.2 s 逐帧检查未见盒子再转斜。"
+     },
+     {
+      "t": 6.2,
+      "log_step": null,
+      "text_zh": "推断：inputs/run.json 的评分器字段 success=1、replay_success=1、symbolic_success=1、deterministic=1，success_step=184，结合视频 6.2 s 的摆正终态，将本步含连续 45 控制步静置的完整验收判为达成。完整验收时刻近似锚定视频 6.2 s（约半秒精度）；未把盒子刚摆正或相机刚转回的时刻当作静置验收完成，也未将 agent 工作时间换算为视频时间。"
+     },
+     {
+      "t": 2.0,
+      "log_step": 424,
+      "text_zh": "观察：视频 1.0–3.6 s 只露出盒子局部或盒子完全出画，未见机器人夹爪与盒子的接触。log step 424 的交付生成脚本命令左夹爪下降、推角、上撤和回位，log step 428 记录保存 200 行动作；这些日志说明交付意图，不能确认画外实际接触的作用端及方式。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：视频操作段相机偏离游戏盒，夹爪与盒子的接触区域未入画；没有状态日志，无法核实实际由哪一作用端接触盒子及接触方式。左夹爪推角仅见于 log step 424 的脚本，不据此确认实际推/扫或抓握。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "game_box_orientation_error",
+   "unit": "rad",
+   "threshold": 0.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_object_robot_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_object_robot_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_robot_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_joint_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "评分尝试最终完成“游戏盒摆正”节点（1/1）：视频初态盒子斜置，视频 4.8–6.2 s 可见其横放桌面并保持稳定，评分器确认确定性回放成功。完整静置验收近似锚定视频 6.2 s；操作期间接触区域出画，实际接触标为未知，缺少状态日志和评分器原生指标数值，无法独立复核朝向、速度及��移阈值。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "游戏盒摆正节点终态达成与视频及评分器字段一致（视频初态盒子斜置，4.8–6.2 s 盒子横放稳定）；contact_actual 因接触出画标未知合理；log step 424/428 存在，时间未换算，未发现需要改动的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:32:41Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-alignment-axial-i00-privileged--xstibnf/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-alignment-axial-i00-privileged--xstibnf/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "f5008bc2fceddd6cb82a8ffc1d13f5a34cbd6c9ed0bf563979c9366d17995fa7",
+   "video_sha256": "580c7f40cdbbf9c5d35d2599f9aa47a4baad35c1e1e13423e4e335d77da27203",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/alignment_axial/trajectories/ispatialgym-alignment-axial-i00-privileged--yjhcca9.json

@@ -0,0 +1,215 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-alignment-axial-i00-privileged--yjhcca9",
+ "taskKey": "ispatialgym/alignment_axial",
+ "dag_sha256": "b530f6c690202b00faaa215c65e0558d1cf88fb1c92dad0ed74a8f9008f88b8c",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 397,
+   "sim_steps": 310,
+   "rows": 397,
+   "max_rows": 1612,
+   "replay_success": 1,
+   "instance_ok": 1,
+   "steps": 310,
+   "success_step": 310,
+   "symbolic_success": 1,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 16.87,
+   "peak_memory_gb": 8.512,
+   "peak_processes": 24,
+   "peak_disk_gb": 2.693,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-alignment-axial-i00-privileged--yjhcca9/replay.mp4",
+  "duration_s": 10.5,
+  "fps": 10.0,
+  "frames": 105,
+  "sha256": "0ee2f97b69ceedfbeac9bc0da875d356851ea555a227af510820113c5707c82e"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "game_box_aligned": {
+    "method": "visual",
+    "intervals": [
+     [
+      10.4,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 的绿色 Clue 游戏盒平放在白色餐桌上，盒边相对桌边斜置；视频 0.0–1.3 s 头部视角持续转向右侧，不能把盒子在画面中的角度变化当成盒子实际旋转。"
+     },
+     {
+      "t": 1.4,
+      "log_step": null,
+      "text_zh": "观察：逐帧检查视频 0.0–10.4 s；游戏盒在视频约 1.4 s 完全离开画面，此后至视频 10.4 s 只有桌角、门和室外景物可见，夹爪与游戏盒的接触、盒子转动及最终朝向均未直接入镜。"
+     },
+     {
+      "t": null,
+      "log_step": 154,
+      "text_zh": "观察：log step 154 中 agent 的调试脚本重跑转动方案、追加保持并保存 337 行；log step 162 将同一交付文件的保持尾段延长至 397 行；log step 180 的独立回放输出 instance_exact=true、steps=397。上述是会话与交付记录，不把其自测盒子偏角作为本次视频的逐帧状态。"
+     },
+     {
+      "t": 10.4,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器字段 success=1、replay_success=1、symbolic_success=1、deterministic=1，success_step=310；视频 9.4–10.4 s 逐帧均未显示游戏盒，因而没有直接显示目标未满足的画面。"
+     },
+     {
+      "t": 10.4,
+      "log_step": null,
+      "text_zh": "推断：依评分器已通过的联合验收，将唯一目标节点终态判为 true，并将含静置窗口的首次达成保守锚定在视频末帧 10.4 s（约半秒精度，实际阈值跨越时刻不可见）；不从 agent 时间或会话自测推算视频时刻。"
+     },
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：参照录像 inputs/oracle/replay.mp4 在其视频 6.5–7.9 s 可见游戏盒沿桌边横放，与本次视频 0.0 s 的斜置不同；本次录像末段没有回到该视角。该参照仅说明达成外观，不作为本次发生接触或达标时刻的证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本次视频 0.0–10.4 s 没有显示夹爪接触游戏盒，且无状态日志。agent 在 log step 185 自述左侧闭合夹爪抵住盒侧面并绕轴推动，但实际作用端与接触方式无法由画面核实，故不将其自述或 DAG 的典型接触方案直接标为实际接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "game_box_orientation_error",
+   "unit": "rad",
+   "threshold": 0.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_object_robot_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_object_robot_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_robot_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_joint_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "评分器确认交付轨迹成功且可确定复现，唯一的“游戏盒摆正”节点终态完成，含静置验收的达成时间保守锚定在视频末段 10.4 s。视频初态可见游戏盒斜置，随后头部视角转离目标；视频约 1.4–10.4 s 游戏盒与操作接触均不在画面内，因此实际接触记为未知，不用会话自测替代本次画面。没有状态日志，无法直接复核旋转误差、各项速度及漂移数值，评分器也未报告这些原生指标的数值。",
+ "conf": 6,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "评分器 success=1，视频仅 0–1.4 s 可见游戏盒斜置，其后相机转离桌面；单节点 true、接触未知、无失败层均与证据一致，时间锚点保守且已标推断。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:36:20Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-alignment-axial-i00-privileged--yjhcca9/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-alignment-axial-i00-privileged--yjhcca9/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "fb67dc4891343816523460068cdebc334d981f550ef36d14375232b926dde159",
+   "video_sha256": "0ee2f97b69ceedfbeac9bc0da875d356851ea555a227af510820113c5707c82e",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/alignment_axial/trajectories/ispatialgym-alignment-axial-i00-standard--pj9uzw7.json

@@ -0,0 +1,277 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-alignment-axial-i00-standard--pj9uzw7",
+ "taskKey": "ispatialgym/alignment_axial",
+ "dag_sha256": "b530f6c690202b00faaa215c65e0558d1cf88fb1c92dad0ed74a8f9008f88b8c",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "live_matches_replay": 1,
+   "finalized": 1,
+   "missing_log": 0,
+   "n_actions": 1612,
+   "sim_steps": 1612,
+   "max_steps": 1612,
+   "steps": 822,
+   "elapsed_s": 1290.2,
+   "symbolic_success": 0,
+   "success_step": 0,
+   "video_rendered": 1,
+   "gpu_lost": 0,
+   "replay_error": 0,
+   "instance_ok": 1,
+   "sim_startups": 1,
+   "resumes": 0,
+   "resume_state_mismatch": 0,
+   "resume_max_tcp_dev_m": 0.0,
+   "resume_max_joint_dev": 0.0,
+   "gave_up": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 0.469,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.949,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-alignment-axial-i00-standard--pj9uzw7/replay.mp4",
+  "duration_s": 53.9,
+  "fps": 10.0,
+  "frames": 539,
+  "sha256": "b8312d65883d95ec28ae1ec7f8c781b19398cb400da81e38d2a7afd391e80b6f"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "game_box_aligned": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–1.5 s，绿色 Clue 游戏盒平放在白桌上，盒边相对桌边斜置；成功参考录像的视频 6.5–8.0 s 则显示盒子沿桌边横放。参考录像只用于辨认目标外观，不作为本次进度证据。"
+     },
+     {
+      "t": 6.5,
+      "log_step": 76,
+      "text_zh": "观察：视频 5.0–6.5 s 左夹爪下降、闭合后抬离，盒子仍留在桌面；视频 8.5–18.5 s 多次重复下探和抬爪也未见盒子随爪离桌。agent 在 log step 76 自述第一次闭爪未持住盒子。"
+     },
+     {
+      "t": 30.5,
+      "log_step": 101,
+      "text_zh": "观察：视频 21.5–23.5 s 左夹爪从侧面贴近盒边；视频 27.0–30.5 s 左夹爪抵住盒面作横向动作后撤开，盒子仍平放且相对桌边斜置。log step 101 中 agent 将动作描述为侧向推动。"
+     },
+     {
+      "t": 53.8,
+      "log_step": 143,
+      "text_zh": "观察：视频 34.5–38.5 s、42.5–46.0 s、48.5–51.0 s 左夹爪继续下压、推动并撤回；视频 52.9–53.8 s 的逐帧末段显示盒子未被持有、仍斜置，未出现目标横放外观。log step 143 的在线服务返回 control_steps=1612、done=True，原因为控制步数耗尽；评分字段 live_success=0、replay_success=0、live_matches_replay=1。"
+     },
+     {
+      "t": 53.8,
+      "log_step": null,
+      "text_zh": "推断：结合整段视频中各次撤爪后的盒边方向及末段静止画面，本节点从未达到目标朝向并完成静置验收；此判断依据可见偏转，不仅依据零分。没有状态日志，不能给出精确旋转误差或静置速度。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "压/擦",
+      "note_zh": "观察：视频 27.0–30.0 s、35.0–38.0 s、43.0–45.5 s 及 49.0–50.5 s，机器人左夹爪指端从上方抵住绿色 Clue 盒面并擦动；夹爪撤离后盒子仍留桌面，未见两指夹持盒子随爪移动，故不记抓握。"
+     },
+     {
+      "who": "手指/夹爪",
+      "how": "推/扫",
+      "note_zh": "观察：视频 21.5–23.5 s 左夹爪从盒子侧边抵推，后续顶面接触伴随横向推动；作用对象为绿色 Clue 游戏盒。与下压接触先后及伴随发生，均为同一左夹爪的实际操作。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "game_box_orientation_error",
+   "unit": "rad",
+   "threshold": 0.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_object_robot_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_object_robot_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_robot_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_joint_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "game_box_aligned",
+   "t": 6.5,
+   "log_step": 76,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 5.0–6.5 s 机器人左夹爪首次下探闭合并抬起，盒子仍在原桌面，未建立持有；后续多次试夹、下压和推动均未把本节点完成。视频 52.9–53.8 s 盒子仍相对桌边斜置，故首次失败的游戏盒摆正步骤始终未修复。"
+  },
+  "unmet": {
+   "game_box_aligned": false
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没建立接触",
+    "t": 6.5,
+    "log_step": 76,
+    "evidence_zh": "观察：视频 5.0–6.5 s 左夹爪尝试闭合后抬起，盒子始终留在桌面；视频 8.5–18.5 s 的重复试夹以及视频 21.5–24.0 s 的侧向试夹也未见盒子被夹持。这里缺少的是计划中的抓握接触，并非声称夹爪从未碰到盒面；log step 76 中 agent 也自述首次闭爪失败。"
+   },
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 30.5,
+    "log_step": null,
+    "evidence_zh": "观察：视频 30.5–34.0 s 左夹爪撤离后，盒边仍相对桌边斜置；视频 38.5–42.0 s、46.0–48.0 s 和末段视频 52.9–53.8 s 均保留明显偏角。后续推动未将盒子转为成功参考所示的沿桌边横放。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "认知",
+    "item": "推理/规划",
+    "source_zh": "日志 / 视频",
+    "evidence_zh": "推断：task/task_dag.json:29、39、45、67 说明目标是把游戏盒绕竖直轴转到目标朝向，盒子经竖直转轴连到桌面，并非可任意抓起搬移的自由物体。inputs/run.json 的 log step 53 中 agent 自述准备抓取，log step 76 自述重新下探、闭爪、抬起后再放置；视频 5.0、5.5、6.0、6.5 s 及 8.5、12.0、18.5 s 显示左夹爪多次下探与抬起而盒子留在桌面。log step 101 中 agent 转而计划侧推改变桌面位置，log step 118 按目标位置作向前微调，log step 129 又按其估计的约 6 cm 横向位置差决定反向推动；这些位置与位移均为 agent 的估计，不是已核实的物体测量。log step 145 的动作记录同样列出沿 x/y 的平移推移。视频 27.0、30.5、38.0、45.0、50.0 s 可见夹爪抵推盒子，视频 53.0、53.4、53.8 s 盒子仍斜置。抓起搬运及围绕位置偏移作直线推移的策略没有针对受转轴约束的盒子建立目标朝向修正，支持认知·推理/规划；试抓未持有不能独立证明可行方案上的控制执行失败。"
+   }
+  ],
+  "gap_zh": "没有提供状态日志；现有头部视角可判断盒子未摆正，但早段盒子位于画面左侧且部分被夹爪遮挡，不能逐次确认指端接触对、夹持力或具体控制误差。评分器未报告 DAG 所列七项原生指标的数值，均填 null；不能从画面核算 0.15 弧度阈值及全部 45 控制步静置条件。日志若干工具输出本身被截断；agent 关于末端停在盒面之上等自述未被独立测量核实，现有试抓失败证据不能区分错误搬运策略与额外的控制执行问题，因此根因仅保留推理/规划。"
+ },
+ "summary_zh": "机器人用左夹爪多次试夹、下压和推动绿色 Clue 游戏盒，但没有将盒子摆正，节点始终为 0/1。agent 先按抓起搬运、后按位置偏移制定直线推移方案，没有针对盒子的竖直转轴约束建立朝向修正策略；盒子到视频 53.8 s 仍斜置，在线回合因控制步数耗尽结束。缺少状态日志、精确朝向和接触力指标，无法复核数值阈值，也不足以独立支持控制根因。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 2,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "fix",
+    "conf": 6,
+    "summary_zh": "节点未达成、断点（首次试抓）和视频末帧斜置均与画面一致；但根因「控制」缺少独立证据：前期失败的是“抓起并搬到某位置”这一目标/策略本身（盒子以竖直转轴约束在桌面、任务只要求转向），应归规划，而不是接触执行失败。",
+    "n_issues": 1,
+    "n_major": 1
+   },
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 2,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "节点未达成、断点（首次试抓，约视频 6.5 s）、接触标签与末帧斜置均与画面一致；根因已按上轮意见改为推理/规划，无需再改。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:49:47Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-alignment-axial-i00-standard--pj9uzw7/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-alignment-axial-i00-standard--pj9uzw7/04_review_r1"
+   },
+   {
+    "stage": "reconcile",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-alignment-axial-i00-standard--pj9uzw7/05_reconcile_r1"
+   },
+   {
+    "stage": "traj_review",
+    "round": 2,
+    "ok": true,
+    "space": "work/runs/ispatialgym-alignment-axial-i00-standard--pj9uzw7/04_review_r2"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "3eced630c50682c04709506cebff742e0174db55ca704ebb54daa4098cb966ff",
+   "video_sha256": "b8312d65883d95ec28ae1ec7f8c781b19398cb400da81e38d2a7afd391e80b6f",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/alignment_axial/trajectories/ispatialgym-alignment-axial-i00-standard--ut4xsxt.json

@@ -0,0 +1,224 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-alignment-axial-i00-standard--ut4xsxt",
+ "taskKey": "ispatialgym/alignment_axial",
+ "dag_sha256": "b530f6c690202b00faaa215c65e0558d1cf88fb1c92dad0ed74a8f9008f88b8c",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "live_success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "live_matches_replay": 1,
+   "finalized": 1,
+   "missing_log": 0,
+   "n_actions": 608,
+   "sim_steps": 608,
+   "max_steps": 1612,
+   "steps": 203,
+   "elapsed_s": 420.4,
+   "symbolic_success": 1,
+   "success_step": 608,
+   "video_rendered": 1,
+   "gpu_lost": 0,
+   "replay_error": 0,
+   "instance_ok": 1,
+   "sim_startups": 1,
+   "resumes": 0,
+   "resume_state_mismatch": 0,
+   "resume_max_tcp_dev_m": 0.0,
+   "resume_max_joint_dev": 0.0,
+   "gave_up": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 0.329,
+   "peak_processes": 11,
+   "peak_disk_gb": 0.921,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-alignment-axial-i00-standard--ut4xsxt/replay.mp4",
+  "duration_s": 20.4,
+  "fps": 10.0,
+  "frames": 204,
+  "sha256": "9c83d17dbac19c13b3956f20d02299ac1d36a86bd1c456e96bc9bf6cb7716970"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "game_box_aligned": {
+    "method": "visual",
+    "intervals": [
+     [
+      20.3,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 的绿色 Clue 游戏盒平放在白桌上，但盒边相对桌边斜置；对照 oracle 视频 7.0–8.0 s 的摆正朝向，初态尚未完成本节点。"
+     },
+     {
+      "t": 13.6,
+      "log_step": 55,
+      "text_zh": "观察：本次视频 9.1–9.5 s 游戏盒先向一侧转动，视频 11.3–13.6 s 又持续转向；此时门窗和桌边基本不动，确为盒子转动而非相机转动。视频 14.0–16.4 s 盒子继续改变投影位置，视频 16.5–20.3 s 保持摆正后的朝向和桌面位置。"
+     },
+     {
+      "t": 14.0,
+      "log_step": 55,
+      "text_zh": "推断：log step 28 的服务规格为 30 Hz；本次单 episode 无重置，log step 50 的 control_steps=336 与 log step 55 的 control_steps=420 对应视频 11.2–14.0 s，已与视频 11.3–13.6 s 的盒子转动及随后停转核对。因此采用本次 episode 的控制步时钟定位验收，未使用 agent 工作时间。"
+     },
+     {
+      "t": 20.3,
+      "log_step": 63,
+      "text_zh": "观察：log step 63 的仿真服务返回 done=true、reason=success、control_steps=608；评分器 reward 同时给出 live_success=1、replay_success=1、symbolic_success=1、live_matches_replay=1、success_step=608。视频 19.3–20.3 s 逐帧检查，盒子保持桌面位置和朝向，没有末段再次转歪。"
+     },
+     {
+      "t": 20.3,
+      "log_step": 63,
+      "text_zh": "推断：完整节点包含机器人与物体连续 45 控制步静置，不能仅以盒子停转时刻计为完成；服务在 control step 608 首次通过联合验收，对应视频 608/30≈20.27 s，按已检查的视频帧取约 20.3 s 达成并保持至结束。没有状态日志，具体姿态误差和速度阈值依服务成功判定，无法逐项数值复核。"
+     },
+     {
+      "t": 9.5,
+      "log_step": 48,
+      "text_zh": "观察：视频 8.5–16.4 s 密集帧可见盒子移动、转动，但盒子左侧被画面边缘截去，机器人夹爪和接触点始终不在提供的视频中；log step 47 的脚本命令左夹爪闭合，log step 48 返回左夹爪开口约 0.045 m。该读数不能单独证明相向夹持，因此实际接触方式保留未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：视频 9.1–16.4 s 游戏盒确实被操作，但机器人作用端及盒子接触区域在画面外；log step 44、47、49、54、59 记录左夹爪下探、闭合、上移、转腕和张开指令，无法从所提供画面区分实际夹持、压擦或侧推，不沿用 DAG 的典型接触方案。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "game_box_orientation_error",
+   "unit": "rad",
+   "threshold": 0.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_object_robot_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_object_robot_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_robot_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_joint_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "本次完成游戏盒摆正，视频中可见盒子相对固定桌面转向并在末段保持稳定；服务于 control step 608 报告成功，映射为视频约 20.3 s，通过包含静置窗口的完整验收。评分器确认在线与回放均成功且一致，无失败断点。所提供录像未显示夹爪与盒子的接触点，实际接触标为未知；无状态日志，且评分器未报告各项连续指标数值，不能独立复核朝向误差及静置阈值。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "单节点成功运行，control step 608→视频约 20.3 s 的映射、节点达成与末段保持均与帧和评分器字段一致；接触因夹爪在画面外标为未知，合理。未发现需修改之处。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:22:04Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-alignment-axial-i00-standard--ut4xsxt/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-alignment-axial-i00-standard--ut4xsxt/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "04a5d8c466c80f2bfecd09eb19d72c33349496e76e2a9d2aeb2d8c682fe3365c",
+   "video_sha256": "9c83d17dbac19c13b3956f20d02299ac1d36a86bd1c456e96bc9bf6cb7716970",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/camera_relocalization_obstructed/trajectories/ispatialgym-camera-relocalization-obstructed-i00-privileged--46825eb.json

@@ -0,0 +1,237 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-camera-relocalization-obstructed-i00-privileged--46825eb",
+ "taskKey": "ispatialgym/camera_relocalization_obstructed",
+ "dag_sha256": "4e13f8e6fcf7dbba26ce5d2c29955034bada2f217b1eda1751e2281902a318a2",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 817,
+   "sim_steps": 817,
+   "rows": 817,
+   "max_rows": 1324,
+   "replay_success": 0,
+   "instance_ok": 1,
+   "steps": 817,
+   "success_step": 0,
+   "symbolic_success": 0,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 1,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.0,
+   "peak_processes": 0.0,
+   "peak_disk_gb": 0.0,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-camera-relocalization-obstructed-i00-privileged--46825eb/replay.mp4",
+  "duration_s": 27.4,
+  "fps": 10.0,
+  "frames": 274,
+  "sha256": "72e8a809c1cc44ba7848fe12d6fcdc662418602722fe4fb9070aeb247c44abf2"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "camera_at_goal_pose": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 头部相机仍远望整组沙发，厨房岛台只在右后方；视频 0.5–14.0 s 机器人移动并调节相机视角，未呈现稳定的目标构图。"
+     },
+     {
+      "t": 21.8,
+      "log_step": null,
+      "text_zh": "观察：视频 14.0–18.0 s 及视频 20.6–23.0 s 的密集帧显示机器人到达沙发外侧并作微调；视频 21.8–27.3 s 画面基本不动，但沙发靠背、岛台台面与吊灯在画面中的位置仍有偏差。对照成功参考视频 10.0–12.2 s，本次视角较低、岛台台面露出较少，吊灯与窗带投影整体较低；参考仅用于辨认目标视角。"
+     },
+     {
+      "t": 27.3,
+      "log_step": null,
+      "text_zh": "推断：本次末帧视频 27.3 s 仍保持上述偏离目标的视角；结合评分器 replay_success=0、symbolic_success=0、success_step=0，唯一目标节点未通过连续45控制步验收。此判断不把视觉停稳等同于原生速度和漂移阈值已满足，也不由零分单独推断未尝试。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "底盘",
+      "how": "支撑",
+      "note_zh": "推断：视频 0.5–14.0 s 中机器人头部相机持续沿地板平移，绕过沙发接近厨房，视频 20.6–21.8 s 再作小幅平移；结合 log step 86 的底盘驱动脚本与R1Pro轮式结构，车轮接触地板支撑并运送机身和相机。轮地接触点位于头部视野外；未见夹爪抓取任务物体。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "head_camera_position_error",
+   "unit": "m",
+   "threshold": 0.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "head_camera_orientation_error",
+   "unit": "rad",
+   "threshold": 0.45,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_robot_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_robot_joint_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "camera_at_goal_pose",
+   "t": 21.8,
+   "log_step": null,
+   "where": "操作中",
+   "evidence_zh": "观察：机器人在视频 21.8 s 左右停止最后的小幅视角修正，此后至视频 27.3 s 一直保持有偏差的厨房视角；岛台台面、沙发靠背和吊灯的投影未对齐成功参考视频 10.0–12.2 s 所示目标构图。推断：相机重定位步骤未完成，后续静止保持未修复位姿偏差，故断点为该导航步骤的操作中。"
+  },
+  "unmet": {
+   "camera_at_goal_pose": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 21.8,
+    "log_step": null,
+    "evidence_zh": "观察：视频 21.8–27.3 s 本次稳定画面的沙发靠背与岛台台面露出比例、吊灯和窗带的垂直位置，与成功参考视频 10.0–12.2 s 不同；本次视角偏低、岛台台面露出较少。推断：机器人停在偏离目标的相机位姿，静止等待没有消除该偏差。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "认知",
+    "item": "感知",
+    "source_zh": "日志",
+    "evidence_zh": "推断：agent 从图像恢复的目标相机位置存在估计偏差。log step 59 的PnP解和 log step 60 的自述采用 odom (3.27,-4.57,1.12)米；用 log step 43 输出的世界到odom矩阵转换DAG目标，得到约(3.425,-4.628,1.359)米，两者相距约0.291米，超过0.15米容差。log step 86 按该估计设定底盘目标及低位躯干姿态；log step 89 的生成结果和 log step 111 的会话自回放均接近所设位置，却高度偏低，与视频 21.8–27.3 s 的构图偏差相符。这些数值是会话诊断和目标换算，用于根因分析，不充作评分回放的原生误差指标。"
+   },
+   {
+    "cat": "认知",
+    "item": "误以为完成",
+    "source_zh": "agent 自述",
+    "evidence_zh": "观察：log step 112 仅计算回放末图与目标的平均像素差24.52；log step 114 agent 自述“The replayed final view matches the goal image”并清理仿真进程，log step 116 宣布完成、交付已验证。log step 102、111 的回放摘要只说明817行完成和实例匹配，未报告任务成功；评分器最终 replay_success=0。推断：agent 把构图近似和回放执行完成当作任务已完成，停止了后续修正；run.json 记录会话用时3116 s、预算3600 s，停止时约余484 s，并非预算耗尽。"
+   }
+  ],
+  "gap_zh": "无状态日志；评分器未报告DAG列出的8项原生误差、速度和漂移指标，均填null，不能从录像量出精确阈值或分离符号关系与静止窗口各自的失败贡献。log step 111 的位姿来自会话自回放而非评分器状态记录，仅用于根因佐证；节点值由本次录像的可见偏差及评分结果共同判断。底盘轮地接触点不在头部相机画面内，支撑接触依据轮式结构、可见移动与驱动脚本推断。"
+ },
+ "summary_zh": "机器人移动并调节躯干，绕过沙发到达厨房观察位置后保持静止，但相机目标位姿节点始终未通过，最终为0/1。视频 21.8–27.3 s 的稳定构图仍偏低；日志支持目标位姿估计偏差，以及agent误以为画面匹配而停止修正。无状态日志，精确原生位姿误差及静止验收各项指标未报告。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "节点 camera_at_goal_pose 未达成、断点（操作中）、位姿不准、感知与误以为完成均有帧与日志支持；视频 27.3 s 末帧与成功参考的构图相比偏低、岛台露出较少。未发现需要改动的内容。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:23:21Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-camera-relocalization-obstructed-i00-privileged--46825eb/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-camera-relocalization-obstructed-i00-privileged--46825eb/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "0138cac455190b80a8f0bb49a9b50453c1c92c72b9649042a6410dc4a1cf4523",
+   "video_sha256": "72e8a809c1cc44ba7848fe12d6fcdc662418602722fe4fb9070aeb247c44abf2",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/camera_relocalization_obstructed/trajectories/ispatialgym-camera-relocalization-obstructed-i00-privileged--9sngdbr.json

@@ -0,0 +1,212 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-camera-relocalization-obstructed-i00-privileged--9sngdbr",
+ "taskKey": "ispatialgym/camera_relocalization_obstructed",
+ "dag_sha256": "4e13f8e6fcf7dbba26ce5d2c29955034bada2f217b1eda1751e2281902a318a2",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 420,
+   "sim_steps": 364,
+   "rows": 420,
+   "max_rows": 1324,
+   "replay_success": 1,
+   "instance_ok": 1,
+   "steps": 364,
+   "success_step": 364,
+   "symbolic_success": 1,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 16.542,
+   "peak_memory_gb": 10.137,
+   "peak_processes": 24,
+   "peak_disk_gb": 0.628,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-camera-relocalization-obstructed-i00-privileged--9sngdbr/replay.mp4",
+  "duration_s": 12.3,
+  "fps": 10.0,
+  "frames": 123,
+  "sha256": "1d663b12e5ee95b475b8ba178285ae28c69459eb1089f7f19d5a992099c86c61"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "camera_at_goal_pose": {
+    "method": "visual",
+    "intervals": [
+     [
+      12.2,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频0.0 s的头部相机仍远望整组深棕色沙发，白色茶几居中，厨房岛台在右后方，尚不是DAG要求的近距离厨房目标构图。"
+     },
+     {
+      "t": 6.0,
+      "log_step": 86,
+      "text_zh": "观察：视频0.0–10.5 s中，机器人头部相机视角连续平移和转向，沙发逐渐靠近并移到左前景，厨房岛台及白色圆凳逐渐变大；log step 86的交付轨迹生成脚本从初态清空动作记录后，控制底盘平移、转向并同步调整躯干，随后执行停稳动作。"
+     },
+     {
+      "t": 12.2,
+      "log_step": null,
+      "text_zh": "观察：已逐帧检查视频9.0–12.2 s；末段约视频10.8–12.2 s构图基本不变，深棕色沙发占左前景，木色岛台和一排白色圆凳在右前方，后方可见黑色高柜、红色挡板与两盏吊灯，视频12.2 s末帧保持这一视角。"
+     },
+     {
+      "t": 12.2,
+      "log_step": 89,
+      "text_zh": "推断：结合上述末段稳定构图及inputs/run.json评分器字段success=1、replay_success=1、symbolic_success=1、deterministic=1，判定唯一目标节点最终达成，包含原生位姿容差与连续45控制步静止验收；将完成锚点保守置于视频12.2 s，精确首达时刻不能由画面独立测定。log step 89已保存420行动作，log step 91–97是保存后的会话诊断，不作为交付轨迹的节点时间或原生指标。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "底盘",
+      "how": "支撑",
+      "note_zh": "推断：视频0.0–10.5 s的连续室内视点位移与log step 86记录的底盘平移、转向控制一致，R1Pro轮式底盘由木地板支撑并移动机身，把头部相机送到目标视角；第一人称画面不直接显示车轮接触点。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "head_camera_position_error",
+   "unit": "m",
+   "threshold": 0.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "head_camera_orientation_error",
+   "unit": "rad",
+   "threshold": 0.45,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_robot_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_robot_joint_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人移动底盘并调整躯干，使头部相机从远望整组沙发的位置到达厨房目标视角，末段构图稳定；结合评分器回放成功，唯一节点最终完成（1/1），完成锚点保守记为视频12.2 s。log step 89已保存轨迹，log step 98的提供方过滤提示发生在保存及会话验证之后，评分器仍确认该交付成功。无状态日志和独立参考视频，精确首达时刻、位姿误差、速度及窗口漂移不能从画面独立测定，评分字段未报告的八项原生指标均保留null。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "唯一目标节点判为达成（1/1）与视频和评分器成功字段一致：视频0.0 s远望沙发，10.5–12.2 s构图稳定在沙发左前景、岛台与白色圆凳右前方、后方吊灯与红色挡板；完成锚点12.2 s保守且已标为推断，contact_actual底盘支撑合理，failure为null正确。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:25:31Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-camera-relocalization-obstructed-i00-privileged--9sngdbr/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-camera-relocalization-obstructed-i00-privileged--9sngdbr/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "4c4b631dfa39d8b340abf04c463abaff9779f9da408ca495460e1fc73d9b5f42",
+   "video_sha256": "1d663b12e5ee95b475b8ba178285ae28c69459eb1089f7f19d5a992099c86c61",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/camera_relocalization_obstructed/trajectories/ispatialgym-camera-relocalization-obstructed-i00-privileged--kcgghan.json

@@ -0,0 +1,212 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-camera-relocalization-obstructed-i00-privileged--kcgghan",
+ "taskKey": "ispatialgym/camera_relocalization_obstructed",
+ "dag_sha256": "4e13f8e6fcf7dbba26ce5d2c29955034bada2f217b1eda1751e2281902a318a2",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 494,
+   "sim_steps": 459,
+   "rows": 494,
+   "max_rows": 1324,
+   "replay_success": 1,
+   "instance_ok": 1,
+   "steps": 459,
+   "success_step": 459,
+   "symbolic_success": 1,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 16.707,
+   "peak_memory_gb": 9.848,
+   "peak_processes": 23,
+   "peak_disk_gb": 1.868,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-camera-relocalization-obstructed-i00-privileged--kcgghan/replay.mp4",
+  "duration_s": 15.4,
+  "fps": 10.0,
+  "frames": 154,
+  "sha256": "bc26538449aee54cfe822ccee298cc0a165b59ff1eb92cb06dfc17799d4d4fde"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "camera_at_goal_pose": {
+    "method": "visual",
+    "intervals": [
+     [
+      15.3,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 的头部相机远望整组深棕色沙发和白色茶几，厨房岛台仅在右后方，尚非目标观察视角。"
+     },
+     {
+      "t": 10.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–13.9 s 的连续抽帧显示机器人头部视点向沙发端部及厨房方向移动；视频 10.0 s 前景沙发仍在明显移出画面，视图尚在变化。"
+     },
+     {
+      "t": 13.9,
+      "log_step": null,
+      "text_zh": "观察：逐帧检查视频 12.5–15.3 s，末段约视频 13.9 s 起构图保持稳定：深棕色沙发占左前景，木色岛台、一排白色圆凳位于右前方，黑色高柜、红色挡板和两盏吊灯清晰可见；与参照成功录像的视频 12.2 s 构图一致。"
+     },
+     {
+      "t": 15.3,
+      "log_step": null,
+      "text_zh": "推断：本次评分器 reward 的 success=1、replay_success=1、symbolic_success=1，success_step=459；任务文本规定30 Hz控制，459/30=视频约15.3 s，与所查视频末帧及末段停稳相符。节点包含连续45控制步静止验收，因此将完成锚点定在视频约15.3 s，而非刚接近目标或刚停下时；没有状态日志，无法独立复测各项容差及静止阈值。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "底盘",
+      "how": "支撑",
+      "note_zh": "推断：视频 0.0–13.9 s 的头部视点移动，结合 log step 47 的底盘速度控制脚本及 log step 48 最终交付轨迹的drive输出，表明R1Pro通过轮式底盘在室内地板上行驶、支撑机身，将头部相机送到目标附近；轮地接触点不在头部画面内。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "head_camera_position_error",
+   "unit": "m",
+   "threshold": 0.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "head_camera_orientation_error",
+   "unit": "rad",
+   "threshold": 0.45,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_robot_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_robot_joint_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人移动底盘并调节头部视角，最终形成与成功参照一致的厨房构图；约视频 15.3 s 通过连续静止验收，唯一节点完成，最终为1/1，评分器记录成功。视频末段约13.9–15.3 s 的构图稳定，成功字段支持位姿、符号关系及静止要求均获验收��没有状态日志，精确位姿误差、速度和窗口漂移无法独立复测，评分器也未报告这八项原生指标的数值。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "单节点 camera_at_goal_pose 在视频约 15.3 s 达成：视频 0 s 为远望沙发，13.9 s 与 15.3 s 的构图一致且为厨房目标视角；success_step 459/30≈15.3 s 与视频长度相符。底盘接触为推断且已标注，无失败，无问题。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:31:13Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-camera-relocalization-obstructed-i00-privileged--kcgghan/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-camera-relocalization-obstructed-i00-privileged--kcgghan/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "b4e0434dc5dde6f0afeb8a769dfc4665d486e8bcb9439770aacc5cd1493406ad",
+   "video_sha256": "bc26538449aee54cfe822ccee298cc0a165b59ff1eb92cb06dfc17799d4d4fde",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/camera_relocalization_obstructed/trajectories/ispatialgym-camera-relocalization-obstructed-i00-privileged--lvz35dw.json

@@ -0,0 +1,233 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-camera-relocalization-obstructed-i00-privileged--lvz35dw",
+ "taskKey": "ispatialgym/camera_relocalization_obstructed",
+ "dag_sha256": "4e13f8e6fcf7dbba26ce5d2c29955034bada2f217b1eda1751e2281902a318a2",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 428,
+   "sim_steps": 428,
+   "rows": 428,
+   "max_rows": 1324,
+   "replay_success": 0,
+   "instance_ok": 1,
+   "steps": 428,
+   "success_step": 0,
+   "symbolic_success": 0,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 1,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.0,
+   "peak_processes": 0.0,
+   "peak_disk_gb": 0.0,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-camera-relocalization-obstructed-i00-privileged--lvz35dw/replay.mp4",
+  "duration_s": 14.4,
+  "fps": 10.0,
+  "frames": 144,
+  "sha256": "854dbb75969b0dd64b2bb2893548679a614f60e7b83e529624059d5dd6d2eeb6"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "camera_at_goal_pose": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：本次视频 0.0 s 从客厅远望整组深棕色沙发和白色茶几，厨房岛台较远，尚未呈现目标近景；视频 0.0–10.1 s 头部画面持续平移和转向，随后还有末段俯仰调整。"
+     },
+     {
+      "t": 11.0,
+      "log_step": null,
+      "text_zh": "观察：本次视频 10.7–11.6 s 及视频 13.4–14.3 s 的逐帧检查显示画面已基本固定，但沙发弯曲端部仍完整位于左下方，前方可见较大面积木地板；成功参考视频 10.0–12.2 s 的沙发贴近相机并伸出左下边界，岛台和圆凳更大、所在构图不同。两段录像使用相同头部视角，本次终态仍偏离参考目标视点。"
+     },
+     {
+      "t": 14.3,
+      "log_step": 231,
+      "text_zh": "推断：结合本次视频全程的视角变化与上述终态偏差，以及 inputs/run.json 中评分器 replay_success=0、success_step=0、deterministic=1，未发现满足目标位姿并连续静止45控制步的区间，终值为假。log step 209 保存428行交付，log step 231 确认该文件末125行底盘命令为零；后续会话候选不作为本次回放的进度。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "底盘",
+      "how": "支撑",
+      "note_zh": "推断：视频 0.0–10.1 s 头部视角沿地板平移并转向，结合 log step 205 的轮式底盘速度控制及 log step 206 的303步行驶记录，R1Pro通过车轮接触室内地板支撑机身并移动相机；车轮不在头部画面内。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "head_camera_position_error",
+   "unit": "m",
+   "threshold": 0.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "head_camera_orientation_error",
+   "unit": "rad",
+   "threshold": 0.45,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_robot_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_robot_joint_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "camera_at_goal_pose",
+   "t": 11.0,
+   "log_step": 205,
+   "where": "操作中",
+   "evidence_zh": "观察：机器人在本次视频 0.0–10.1 s 移动相机，视频约10.3–10.9 s 调整俯仰后停在偏离目标的视点；视频 11.0–14.3 s 未再修正。与成功参考视频 12.2 s 相比，本次沙发端部离相机更远，岛台尺度和构图仍不符；失败发生在相机定位步骤，尚未达到全部目标关系后的终态验收。"
+  },
+  "unmet": {
+   "camera_at_goal_pose": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 11.0,
+    "log_step": null,
+    "evidence_zh": "观察：本次视频 11.0 s 及视频 14.3 s 的沙发弯曲端部完整落在画面内，岛台与圆凳较小，沙发前方留有明显木地板；成功参考视频 12.2 s 沙发在左前景被画面边界截断，岛台更靠近且占幅更大。机器人停稳后的相机位置和观察方向所形成的构图仍有偏差。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "认知",
+    "item": "感知",
+    "t": 11.0,
+    "log_step": 205,
+    "source_zh": "日志 / 视频",
+    "evidence_zh": "推断：agent 在 log step 197 从图像地标解算出目标里程计位置(2.62,-4.76)米，并在 log step 205 将其写入交付控制脚本；log step 206、209 的会话反馈均为最终相机(2.630,-4.738,1.386)米，距这个自定目标的水平误差仅0.024米。以 log step 11 的初始世界位置(16.400,2.500)米及 log step 103 的初始航向-129.909度换算，该会话末态水平位置约(11.078,3.522)米，距DAG给定真正目标(10.652,2.842)米约0.80米；这是会话数据的坐标推算，非评分器实测指标。本次视频 11.0–14.3 s 与成功参考视频 12.2 s 的稳定构图偏差与此相符，支持目标视点估计错误导致机器人驶向错误位置。"
+   }
+  ],
+  "gap_zh": "没有状态日志，评分器未报告相机位姿误差、原生符号关系逐项真值或45控制步静止窗口的分项读数，故八项原生指标均留空值。头部录像不能直接显示车轮接触；底盘支撑由连续平移画面及交付控制脚本推断。log step 209 会话自测最大关节速度0.07148弧度/秒高于0.02阈值，但不能据此确认评分回放的静止分项是否也失败；未另加控制根因。"
+ },
+ "summary_zh": "机器人移动底盘并调整头部视角后停止，但视频 11.0–14.3 s 的沙发和岛台构图仍偏离成功参考，唯一节点“相机到目标位姿”始终未达成，最终0/1。断点为相机定位操作中的位姿不准；日志表明控制器接近了agent由地标估算的目标点，结合坐标换算及录像偏差，根因支持认知·感知。无状态日志及评分器分项指标，无法量化评分回放的真实位姿误差或确定静止条件是否另有失败。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "唯一节点未达成、终值0、断点与位姿不准现象与视频一致；末帧构图（沙发端部完整、岛台偏小）确实偏离成功参考。根因感知由会话坐标换算推断，已标为推断，可接受。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:35:06Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-camera-relocalization-obstructed-i00-privileged--lvz35dw/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-camera-relocalization-obstructed-i00-privileged--lvz35dw/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "e08365b34e2ff165baf004db8ecc0f56c161383991c65961d9b027f7d3554e07",
+   "video_sha256": "854dbb75969b0dd64b2bb2893548679a614f60e7b83e529624059d5dd6d2eeb6",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/camera_relocalization_obstructed/trajectories/ispatialgym-camera-relocalization-obstructed-i00-privileged--omns6pf.json

@@ -0,0 +1,233 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-camera-relocalization-obstructed-i00-privileged--omns6pf",
+ "taskKey": "ispatialgym/camera_relocalization_obstructed",
+ "dag_sha256": "4e13f8e6fcf7dbba26ce5d2c29955034bada2f217b1eda1751e2281902a318a2",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 253,
+   "sim_steps": 253,
+   "rows": 253,
+   "max_rows": 1324,
+   "replay_success": 0,
+   "instance_ok": 1,
+   "steps": 253,
+   "success_step": 0,
+   "symbolic_success": 0,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 1,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.0,
+   "peak_processes": 0.0,
+   "peak_disk_gb": 0.0,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-camera-relocalization-obstructed-i00-privileged--omns6pf/replay.mp4",
+  "duration_s": 8.6,
+  "fps": 10.0,
+  "frames": 86,
+  "sha256": "f3ad0d54d6643248f46dc21a99053745a0d8d96ddfd844ca7aa7a78ee280abbe"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "camera_at_goal_pose": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：本次视频0.0 s显示整组深棕色沙发和白色茶几，厨房岛台在远处；初始视角不符合节点要求。已以约0.5秒间隔检查完整视频，并逐帧检查视频4.8–8.5 s。"
+     },
+     {
+      "t": 6.0,
+      "log_step": null,
+      "text_zh": "观察：本次视频0.0–5.9 s头部视角逐渐靠近沙发并转向，视频约6.0–8.5 s构图基本停止变化；沙发横段仍横跨画面中央，白色茶几占右下前景，岛台仅在远处露出小块木色端面。成功参考视频10.5–12.2 s则显示沙发退到左前景，岛台和一排白凳清楚地占右前方，两盏吊灯均清晰可见；参考仅用于对照到达目标后的可见构图。"
+     },
+     {
+      "t": 8.5,
+      "log_step": 72,
+      "text_zh": "推断：本次视频8.5 s仍远望厨房，目标构图的尺度和遮挡关系从未出现，故相机到目标位姿的状态从未成立，而非因看不到精确误差而标未知。评分器reward报告replay_success=0、symbolic_success=0、deterministic=1；log step 71保存253行，log step 72核实交付数组为(253,23)且末50行相同，与视频末段停下的过程一致。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "底盘",
+      "how": "支撑",
+      "note_zh": "推断：本次视频0.0–约6.0 s头部视角在地面场景中平移和转向；log step 44的驾驶脚本通过底盘前三个动作通道驱动，log step 68修正坐标转换，log step 71保存该驾驶结果。因此实际导航交互为R1Pro底盘车轮接触地板并支撑机身移动；头部画面不直接显示轮地接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "head_camera_position_error",
+   "unit": "m",
+   "threshold": 0.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "head_camera_orientation_error",
+   "unit": "rad",
+   "threshold": 0.45,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_robot_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_robot_joint_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "camera_at_goal_pose",
+   "t": 6.0,
+   "log_step": 71,
+   "where": "操作中",
+   "evidence_zh": "观察：本次视频约6.0 s机器人结束主要平移和转向，随后至视频8.5 s保持沙发横挡中景、茶几在右下、岛台远小于目标的视角，导航结果未修正。log step 71记录驾驶控制器进入hold并保存253行，log step 72核实交付文件；断���锚定在本次视频停在错误观察位置之时。"
+  },
+  "unmet": {
+   "camera_at_goal_pose": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 6.0,
+    "log_step": null,
+    "evidence_zh": "观察：本次视频6.0–8.5 s机器人头部相机停在沙发内侧附近，白色茶几仍占右下前景，岛台和凳子远小于目标构图且被沙发遮挡；成功参考视频10.5–12.2 s的岛台及整排白凳在右前方清楚展开。停下后的相机观察位置与朝向组合明显没有达到目标视角。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "认知",
+    "item": "感知",
+    "t": 6.0,
+    "log_step": 60,
+    "source_zh": "日志",
+    "evidence_zh": "推断：agent的目标视角定位估计偏离了实际任务目标。log step 53的程序用72×72灰度图平均绝对差搜索底盘目标，log step 60选定估计底盘世界位置(14.15,-0.225,0.0053)米、偏航-3.03弧度；log step 68修正世界与里程计坐标转换后，交付生成器在log step 71报告距该估计底盘目标0.024米、偏航差约0.029弧度并进入hold。然而本次视频6.0–8.5 s的厨房构图仍显著不同于成功参考视频10.5–12.2 s，支持错误的目标位姿估计引导了停止位置。log step 73也明确将目标称为自身图像匹配估计并承认位置容差可能未满足；未把已修正的早期坐标系错误或运行预算当作本次失败原因。"
+   }
+  ],
+  "gap_zh": "没有状态日志或本次评分回放的逐步头部相机世界位姿，无法量化原生位置/朝向误差、符号关系各项真值以及45控制步静止窗口内的速度和漂移；底盘轮地接触依据视角运动、驾驶代码和机体结构推断。log step 71的trans=0.3286、rot=0.1304是agent脚本对自身图像匹配参考外参的诊断量，不是评分器相对于真实目标的原生误差，故native_metrics全部保留null；定位根因有程序、日志和构图差异支持，但无法进一步确定图像匹配偏差的具体来源。"
+ },
+ "summary_zh": "机器人执行了底盘导航，在本次视频约6.0 s停下并保持到视频8.5 s，但厨房岛台仍在远处且被沙发遮挡，唯一节点“相机到目标位姿”全程未达到，最终0/1。断点是导航停在错误观察位姿，现象为位姿不准；结合图像匹配目标及驾驶收敛日志，推断根因是感知中的目标位姿估计偏差。缺少评分回放的相机外参与静止窗口数据，不能给出原生数值误差。",
+ "conf": 6,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "节点未达成、断点、位姿不准现象和感知根因均有视频与日志支持；与成功参考对比，本次末段仍远望厨房，岛台远小于目标构图。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:28:11Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-camera-relocalization-obstructed-i00-privileged--omns6pf/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-camera-relocalization-obstructed-i00-privileged--omns6pf/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "5d5a55295119df9aa7dffe3ad0eb80eeeaf1900266f7d9f2e1dd6fbebf51a534",
+   "video_sha256": "f3ad0d54d6643248f46dc21a99053745a0d8d96ddfd844ca7aa7a78ee280abbe",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/camera_relocalization_obstructed/trajectories/ispatialgym-camera-relocalization-obstructed-i00-privileged--ps5njyo.json

@@ -0,0 +1,237 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-camera-relocalization-obstructed-i00-privileged--ps5njyo",
+ "taskKey": "ispatialgym/camera_relocalization_obstructed",
+ "dag_sha256": "4e13f8e6fcf7dbba26ce5d2c29955034bada2f217b1eda1751e2281902a318a2",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 582,
+   "sim_steps": 582,
+   "rows": 582,
+   "max_rows": 1324,
+   "replay_success": 0,
+   "instance_ok": 1,
+   "steps": 582,
+   "success_step": 0,
+   "symbolic_success": 0,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 16.629,
+   "peak_memory_gb": 11.313,
+   "peak_processes": 22,
+   "peak_disk_gb": 0.765,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-camera-relocalization-obstructed-i00-privileged--ps5njyo/replay.mp4",
+  "duration_s": 19.5,
+  "fps": 10.0,
+  "frames": 195,
+  "sha256": "d1582099545dd61515e78252d87a2e1449868886bce7f2ece2701e487559f461"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "camera_at_goal_pose": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 为远望整组沙发和厨房的初始视角；视频约 1.0–14.0 s，相机持续向厨房移动，篮子从右前方移至画面下缘后离开视野，尚未形成目标静止视角。"
+     },
+     {
+      "t": 16.8,
+      "log_step": null,
+      "text_zh": "观察：视频 14.0–19.4 s 已逐渐停下；视频 16.8、18.4、19.4 s 的沙发右端仍伸入画面中央，岛台、圆凳和吊灯整体位置偏高，天花板可见面积较少。对照成功参考录像视频 10.0–12.2 s，参考视角的沙发更多留在左侧、岛台及圆凳的透视与遮挡不同；此处是本次录像的观察，参考录像仅用于辨认达标构图。"
+     },
+     {
+      "t": 19.4,
+      "log_step": 449,
+      "text_zh": "推断：全程抽查及视频 14.0–19.4 s 每约 0.2 s 检查均未见目标视角完成后保持；评分器 replay_success=0、success_step=0、symbolic_success=0 与可见偏差一致，因此节点从未达到，末态为假。log step 445、449 保存的是重置后运行的582行轨迹，和评分器 n_actions=582 一致；后续调试的额外静止步不计入该交付。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "底盘",
+      "how": "支撑",
+      "note_zh": "推断：R1Pro轮式底盘通过车轮与室内地板的支承接触移动相机；视频约 1.0–16.8 s 的连续场景视差与 log step 445 的底盘平移、转向动作一致。头部视角不直接展示车轮接触面；未见抓取或搬动篮子。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "head_camera_position_error",
+   "unit": "m",
+   "threshold": 0.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "head_camera_orientation_error",
+   "unit": "rad",
+   "threshold": 0.45,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_robot_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_robot_joint_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "camera_at_goal_pose",
+   "t": 16.8,
+   "log_step": 445,
+   "where": "操作中",
+   "evidence_zh": "观察：视频约 16.8 s 机器人结束接近并停在一个构图仍有偏差的视角，直至视频 19.4 s 未修正；沙发伸入中央、厨房在画面中的高度和透视均不同于成功参考录像视频 10.0–12.2 s。log step 445 的交付脚本以底盘目标(11.35, 2.48)米、偏航−132度结束接近并追加80个保持动作，log step 449 返回582行；评分器回放未成功。"
+  },
+  "unmet": {
+   "camera_at_goal_pose": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 16.8,
+    "log_step": null,
+    "evidence_zh": "观察：视频 16.8–19.4 s 相机停下时，前景沙发的右端占据中央较大区域，厨房岛台和两盏吊灯偏向画面上部；成功参考录像视频 10.0–12.2 s 的沙发右缘更靠左、厨房构图更低且透视不同。机器人已导航接近厨房，但停止后的观察位姿仍未对齐目标。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "认知",
+    "item": "感知",
+    "source_zh": "日志",
+    "evidence_zh": "推断：agent 的目标相机定位估计有误。log step 418 的PnP图像定位输出目标位置(11.175454, 2.305586, 1.410091)米，与任务DAG criterion_zh 给出的原生目标(10.652477, 2.841637, 1.363528)米相距约0.750米，超过0.15米容差；log step 427 的局部图像搜索仍选择底盘(11.35, 2.48)米、偏航−132度，log step 445 据此生成交付，log step 449 的底盘终点(11.363310, 2.486448)米已很接近这一自定目标。结合视频 16.8–19.4 s 可见构图偏差，支持目标位姿估计错误，而非仅凭失败评分归因。上述数值来自会话定位与控制结果，不作为评分回放的原生误差指标。"
+   },
+   {
+    "cat": "认知",
+    "item": "误以为完成",
+    "source_zh": "agent 自述 / 日志",
+    "evidence_zh": "推断：错误的完成判断使 agent 停止修正并保留此交付。log step 464 自述最终相机“within 3.47 cm of ground truth”且静止验收达标；但其比较依据是自己估计的目标，视频 16.8–19.4 s 仍有明显构图偏差。log step 449 对已保存582行轨迹报告的最后45步速度最大值为0.001465米/秒、0.005348弧度/秒和0.020177弧度/秒，均略超任务阈值；log step 451、454 是保存后额外静止调试，log step 456、459 确认交付仍为582行。log step 463 显示尚余269.9秒预算，agent 随后仍以达标结论结束。"
+   }
+  ],
+  "gap_zh": "无状态日志；视频为头部相机视角，无法独立量出评分回放的完整相机位姿、轮地接触、关节速度和45控制步窗口漂移。评分器未报告DAG列出的八项原生连续指标，均填null；会话中的相机查询和速度检查只用于根因证据，不替代回放指标。视频末段按约0.2秒密度检查至视频 19.4 s，足以确认构图仍有偏差，但不能单独确定所有静止或符号检查的失败分项。"
+ },
+ "summary_zh": "机器人移动底盘绕过前景篮子并接近厨房，视频约 16.8 s 后停下，但沙发、岛台和吊灯的构图仍偏离目标，唯一节点“相机到目标位姿”未达到（0/1）。日志显示 agent 将估计错误的相机目标转为底盘终点，并误判位姿和静止验收已完成后结束；582行交付的确定性回放未成功。缺少状态日志和评分侧连续指标，无法逐项量化精确位姿及静止窗口误差。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "标签与视频、日志一致：视频 16.8 s 的构图与参考录像 11 s 明显不同（沙发、岛台位置与朝向偏差），节点未达到，断点、位姿不准、感知和误以为完成均有日志依据。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:41:00Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-camera-relocalization-obstructed-i00-privileged--ps5njyo/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-camera-relocalization-obstructed-i00-privileged--ps5njyo/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "90c49991d24378e27dfca6738bd6d320d7cda13bbfae01ad2891a7a5a9668d9c",
+   "video_sha256": "d1582099545dd61515e78252d87a2e1449868886bce7f2ece2701e487559f461",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/camera_relocalization_obstructed/trajectories/ispatialgym-camera-relocalization-obstructed-i00-privileged--wbkttui.json

@@ -0,0 +1,264 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-camera-relocalization-obstructed-i00-privileged--wbkttui",
+ "taskKey": "ispatialgym/camera_relocalization_obstructed",
+ "dag_sha256": "4e13f8e6fcf7dbba26ce5d2c29955034bada2f217b1eda1751e2281902a318a2",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 199,
+   "sim_steps": 199,
+   "rows": 199,
+   "max_rows": 1324,
+   "replay_success": 0,
+   "instance_ok": 1,
+   "steps": 199,
+   "success_step": 0,
+   "symbolic_success": 0,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 14.913,
+   "peak_memory_gb": 9.895,
+   "peak_processes": 20,
+   "peak_disk_gb": 3.05,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-camera-relocalization-obstructed-i00-privileged--wbkttui/replay.mp4",
+  "duration_s": 6.8,
+  "fps": 10.0,
+  "frames": 68,
+  "sha256": "ee0b1f6605caa78348dd7e1848161a74e76e2d92d2973aac7ff769c3b083fa89"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "camera_at_goal_pose": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 的头部相机画面远望完整L形沙发、中央白色矮桌和远处厨房，初态不具备目标图中沙发占左前景、岛台及圆凳位于右前方的构图。"
+     },
+     {
+      "t": 3.6,
+      "log_step": null,
+      "text_zh": "观察：逐帧检查视频 0.0–6.7 s；视频约 1.6–3.6 s 相机平移，随后约视频 3.7–4.8 s 转向，始终仍远望整组沙发，未出现目标近景视角。"
+     },
+     {
+      "t": 6.7,
+      "log_step": null,
+      "text_zh": "观察：视频 4.8–6.7 s 画面基本停住，但沙发仍横跨中景，前方大面积为空地，篮子位于画面中右部，厨房仍在远处。成功参考录像的参考视频 10.0–12.2 s 中，沙发已占左前景，木色岛台及白色圆凳清晰占右前方；该参考仅用于辨认目标视角。"
+     },
+     {
+      "t": null,
+      "log_step": 229,
+      "text_zh": "观察：log step 188 截取原轨迹前109行，加30行原地转向和60行静止；log step 229 将转向段改为-0.4，log step 230 确认最终文件为199行且末45行底盘命令为零。评分器报告 replay_success=0、symbolic_success=0、deterministic=1；本节点未达成还由上述明显不同的录像构图独立支持，非仅凭零分判断。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "底盘",
+      "how": "支撑",
+      "note_zh": "推断：视频约 1.6–4.8 s 的相机平移与转向，结合 log step 188、229 中最终交付的底盘平移及原地转向命令，表明R1Pro通过车轮在室内地板上支撑并移动机身来调整头部相机；头部视角未直接拍到车轮接地点。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "head_camera_position_error",
+   "unit": "m",
+   "threshold": 0.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "head_camera_orientation_error",
+   "unit": "rad",
+   "threshold": 0.45,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_robot_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_robot_joint_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "camera_at_goal_pose",
+   "t": 4.8,
+   "log_step": 229,
+   "where": "操作中",
+   "evidence_zh": "观察：视频约 4.8 s 完成末段转向并开始停留，此时整组沙发仍在中景，厨房岛台仍很远；视频 4.8–6.7 s 没有进一步驶近目标视角。log step 229 保存的最终版本仅修改原地转向，未增��接近厨房的平移；本次相机重定位步骤始终未完成。"
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 4.8,
+    "log_step": null,
+    "evidence_zh": "观察：视频 4.8–6.7 s 停住后的画面仍以整组沙发、白色矮桌和大片空地为主，岛台在远处；与成功参考录像的参考视频 10.0–12.2 s 中沙发占左前景、岛台和圆凳占右前方的近景明显不同，相机停在偏离目标的位置与朝向。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "认知",
+    "item": "感知",
+    "evidence_zh": "推断：agent 的图像特征匹配把目标相机位姿估得过于接近当前视角。log step 171 的拟合给出剩余平移约[−0.007,0.171,−0.035]米，log step 172 据此自述只需再横移约17厘米；log step 188 最终保留短程平移并接原地转向，log step 229 只改转向方向。log step 209 中 agent 已自述先前拟合被重复地板纹理主导、回放末帧仍明显不同于目标；本次视频 4.8–6.7 s 仍远望沙发及厨房，也表明这种近距离目标位姿估计未正确描述所需视角。未把该拟合的数值当作真实相机误差。",
+    "source_zh": "日志"
+   },
+   {
+    "cat": "交付",
+    "item": "交付操作出错",
+    "evidence_zh": "观察：log step 229 将最终轨迹的 a[109:139,2] 改为 -0.4 并保存；log step 230 仅确认文件为199行、数值有限、末45行底盘命令为零及转向段为 -0.4，未重新回放，log step 231 也明确自述最终编辑未回放、结果未验证。log step 209 已自述此前回放末帧仍明显不同于目标。评分器报告 deterministic=1、replay_success=0、symbolic_success=0；评分视频 4.8–6.7 s 仍远望沙发和厨房，未呈现目标近景。",
+    "source_zh": "日志"
+   }
+  ],
+  "gap_zh": "无状态日志，评分器未报告相机位置/朝向误差、速度或窗口漂移数值，无法量化偏差及核验45控制步内的精确静止阈值。头部视角未显示车轮接地点；底盘支撑依据可见移动及最终命令推断。log step 231 明确说明最终转向修改未由agent再回放，会话中较早回放和探索的位姿读数不作为本次评分轨迹的状态事实。",
+  "unmet": {
+   "camera_at_goal_pose": false
+  }
+ },
+ "summary_zh": "机器人在评分录像中执行短程平移与转向后停住，但全程未呈现目标厨房近景，相机到目标位姿节点最终为0/1。断点为视频约 4.8 s 停在错误视角，现象为位姿不准；日志支持图像特征匹配导致的目标位姿误估，并确认agent保存最终转向修改后未经回放即交付，补记交付操作出错。无状态日志及原生误差指标，精确位姿和静止窗口无法量化。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 2,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "fix",
+    "conf": 7,
+    "summary_zh": "节点未达成、断点、位姿不准现象均有视频支持（评分录像全程远望沙发，参考录像10–12 s为岛台近景）。缺少一个有日志证据的根因：agent 在 log step 229 未经回放就改了转向符号并交付，log step 231 自己承认未验证。",
+    "n_issues": 1,
+    "n_major": 1
+   },
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 2,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "节点未达成、断点、位姿不准、感知与交付操作出错根因均有视频与日志支持；评分录像 6.5 s 仍远望沙发，参考录像 11 s 为岛台近景。未发现需修改的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:47:46Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-camera-relocalization-obstructed-i00-privileged--wbkttui/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-camera-relocalization-obstructed-i00-privileged--wbkttui/04_review_r1"
+   },
+   {
+    "stage": "reconcile",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-camera-relocalization-obstructed-i00-privileged--wbkttui/05_reconcile_r1"
+   },
+   {
+    "stage": "traj_review",
+    "round": 2,
+    "ok": true,
+    "space": "work/runs/ispatialgym-camera-relocalization-obstructed-i00-privileged--wbkttui/04_review_r2"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "0d9e20471e1d776de9e45aeff0a65e550714dd3f5bb96f24409918b665a903bd",
+   "video_sha256": "ee0b1f6605caa78348dd7e1848161a74e76e2d92d2973aac7ff769c3b083fa89",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/camera_relocalization_obstructed/trajectories/ispatialgym-camera-relocalization-obstructed-i00-standard--6qvacql.json

@@ -0,0 +1,226 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-camera-relocalization-obstructed-i00-standard--6qvacql",
+ "taskKey": "ispatialgym/camera_relocalization_obstructed",
+ "dag_sha256": "4e13f8e6fcf7dbba26ce5d2c29955034bada2f217b1eda1751e2281902a318a2",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "live_success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "live_matches_replay": 1,
+   "finalized": 1,
+   "missing_log": 0,
+   "n_actions": 836,
+   "sim_steps": 836,
+   "max_steps": 1324,
+   "steps": 836,
+   "elapsed_s": 567.7,
+   "symbolic_success": 1,
+   "success_step": 836,
+   "video_rendered": 1,
+   "gpu_lost": 0,
+   "replay_error": 0,
+   "instance_ok": 1,
+   "sim_startups": 1,
+   "resumes": 0,
+   "resume_state_mismatch": 0,
+   "resume_max_tcp_dev_m": 0.0,
+   "resume_max_joint_dev": 0.0,
+   "gave_up": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 0.329,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.903,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-camera-relocalization-obstructed-i00-standard--6qvacql/replay.mp4",
+  "duration_s": 28.0,
+  "fps": 10.0,
+  "frames": 280,
+  "sha256": "f151acf6456e779568af557e7f5d81ddef5f789177c20e809f8e61097e499af1"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "camera_at_goal_pose": {
+    "method": "visual",
+    "intervals": [
+     [
+      27.9,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 机器人头部相机仍远望整组沙发，厨房岛台在远处；与成功参考视频 12.2 s 的近距离厨房构图明显不同，初态未达标。"
+     },
+     {
+      "t": 13.0,
+      "log_step": 55,
+      "text_zh": "观察：视频 8.5–12.0 s 头部相机画面明显侧倾，视频 12.5–13.5 s 逐渐恢复水平，随后沿沙发与玻璃墙之间前进；log step 55 的服务底盘位姿读数显示机身已恢复直立，机器人继续完成同一个导航目标。"
+     },
+     {
+      "t": 27.9,
+      "log_step": 66,
+      "text_zh": "观察：视频 24.9–26.4 s 机器人继续小幅平移和转向，视频 26.6–27.9 s 构图保持稳定；视频 27.9 s 沙发占左前景，木色岛台及一排白色圆凳在右前方，后方黑色高柜、红色厨房挡板及两盏吊灯均清楚可见，与成功参考视频 12.2 s 的构图一致。log step 66 的模拟器服务返回 done=true、reason=success、control_steps=836；评分字段 live_success、replay_success、symbolic_success、deterministic、live_matches_replay 均为1。"
+     },
+     {
+      "t": 27.9,
+      "log_step": 66,
+      "text_zh": "推断：log step 36 报告控制频率30 Hz，log step 66 的 control step 836 对应836/30=27.867秒，取已检查的最近视频帧27.9 s 作为节点首次完成锚点，精度约0.1 s；这一控制步映射已用视频 24.9–26.4 s 的最后微调及随后停止核对。节点含45控制步验收，故不在刚到达相似构图时提前计为完成；最终按本次原生成功结果记为达成。"
+     },
+     {
+      "t": 27.9,
+      "log_step": 66,
+      "text_zh": "观察：本次没有状态日志，评分字段未报告精确相机位姿误差或静止窗口的逐步指标；log step 66 同时打印的终态关节速度部分超过DAG所列0.02弧度/秒，服务却明确返回成功。推断：无法用该终态读数重建原生静止窗口，按本次实际评分规则认定完成，不自行声称已逐项验证全部阈值。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "底盘",
+      "how": "支撑",
+      "note_zh": "推断：视频 0.0–8.5 s、12.5–22.0 s 及24.9–26.4 s 的头部视角平移、log step 39、55、57、60、66 的底盘位姿变化共同支持R1Pro车轮在室内地板上滚动并支撑机身，将头部相机送往厨房目标视角；未见夹爪抓取或搬运场景物体。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "head_camera_position_error",
+   "unit": "m",
+   "threshold": 0.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "head_camera_orientation_error",
+   "unit": "rad",
+   "threshold": 0.45,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_robot_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_robot_joint_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人沿沙发与玻璃墙之间移动，中途头部画面侧倾后恢复，最终完成唯一的“相机到目标位姿”节点（1/1）。视频 27.9 s 的稳定厨房构图与成功参考视频 12.2 s 一致，log step 66 的服务成功回执及本次live/replay成功字段共同支持终态达成，无失败断点。缺少状态日志和原生逐步验收指标，且log step 66 的部分终态关节速度读数高于DAG阈值，无法独立重建静止窗口；依本次实际成功判定标注，未报告的原生指标保留null。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "单节点 camera_at_goal_pose 在视频 27.9 s 达成，与 log step 66 的 control_steps=836、30 Hz（836/30=27.87 s）、done=true/success 及成功参考视频 12.2 s 的构图一致；初态（0 s）为远望整组沙发，未达标；8–12 s 画面侧倾、13.5 s 后恢复水平，与证据一致。contact_actual 为底盘·支撑，合理；无失败断点，无需修改。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:28:37Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-camera-relocalization-obstructed-i00-standard--6qvacql/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-camera-relocalization-obstructed-i00-standard--6qvacql/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "c7817bbb3cdeb1667e3f59061569ae49dc7fabe494cc5eb68d96991f4d5e4e83",
+   "video_sha256": "f151acf6456e779568af557e7f5d81ddef5f789177c20e809f8e61097e499af1",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/configuration_matching_articulation_open_small_scale/trajectories/ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--2kpcmbe.json

@@ -0,0 +1,188 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--2kpcmbe",
+ "taskKey": "ispatialgym/configuration_matching_articulation_open_small_scale",
+ "dag_sha256": "d02aeb1ff9d6bc57baa2a14ae97e39013d8e5fefeea4780a10ac0ca80c4f15e7",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 228,
+   "sim_steps": 228,
+   "rows": 228,
+   "max_rows": 2382,
+   "replay_success": 0,
+   "instance_ok": 1,
+   "steps": 228,
+   "success_step": 0,
+   "symbolic_success": 0,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 13.603,
+   "peak_memory_gb": 9.723,
+   "peak_processes": 21,
+   "peak_disk_gb": 4.07,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--2kpcmbe/replay.mp4",
+  "duration_s": 7.7,
+  "fps": 10.0,
+  "frames": 77,
+  "sha256": "3c251cc4cb68119549d75dc15a523a451dd2c44e83cab1a6230b79edc229b35d"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "blender_lid_opened": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–0.8 s 搅拌机直立在白桌上，圆盖平扣杯口；视频约 0.9–5.1 s 目标及夹爪接触区离开头部相机视野。"
+     },
+     {
+      "t": 5.3,
+      "log_step": 89,
+      "text_zh": "观察：视频 5.3–7.6 s（包括最后一秒逐帧）搅拌机重新完整可见，圆盖仍水平盖住杯口，没有掀起；成功示例视频 26.0 s 的盖呈侧立姿态。log step 89 回显 gen4 保存 228 行、终态 j_lid 约 0；log step 94 核实交付仍为 (228,23)，评分器 n_actions=228、symbolic_success=0。"
+     },
+     {
+      "t": 1.7,
+      "log_step": 87,
+      "text_zh": "推断：log step 87 的 gen4 三次尝试末尾（生成轨迹累计 52、99、146 控制步）均测得 j_lid 约 0，log step 89 保存的正是此运行；结合视频开始与结束均闭盖，未见也无反馈支持任何达标区间。后续 gen5 的反向推盖修改不能计入交付进度。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：视频约 0.9–5.1 s 搅拌机及右夹爪操作区在唯一头部视图之外；开始及结尾也未见夹爪包夹杯盖。log step 73、82–83 的脚本确实安排右臂接近盖缘并尝试推盖，但不能据指令认定实际抓握、按压或勾住，故实际作用端与接触方式记未知。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "blender_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "blender_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "blender_lid_joint_error",
+   "unit": "rad",
+   "threshold": 0.4305347800398469,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "blender_lid_opened",
+   "t": 1.7,
+   "log_step": 87,
+   "where": "操作中",
+   "evidence_zh": "推断：断在打开搅拌机盖，首次未修复尝试以视频约 1.7 s 为导航锚。log step 87 记录所交付 gen4 的首次尝试至累计 52 控制步仍 j_lid=0，后两次也为 0；log step 89 保存该 228 行运行。按任务 30 Hz，52/30≈1.73 s；脚本在累计 146 步结束尝试后用 12 步回初姿，158/30≈5.27 s，与视频 5.3 s 重新看到闭盖搅拌机的时刻核对一致。操作区此前离开视野，具体接触失效瞬间无法定位；视频 5.3–7.6 s 确认盖未打开。"
+  },
+  "unmet": {
+   "blender_lid_opened": false
+  },
+  "phenomena": [],
+  "root_causes": [
+   {
+    "cat": "认知",
+    "item": "推理/规划",
+    "source_zh": "日志",
+    "evidence_zh": "推断：生成交付轨迹的脚本将铰链开盖路径的方向算反：log step 73 的公式沿正向轴旋转，log step 82–83 保留该方向用于 gen4，log step 87 三次打印初始切向 tan≈[0,0,-1]，即将闭盖的自由边向杯内下方移动，而所需运动是向上掀盖；同一运行三次末尾 j_lid 均约 0。agent 直到 log step 88 才自述发现方向相反并同时翻转切向及旋转方向；log step 89、94 显示交付仍是此前的 228 行文��。此归因依据事先构造的错误开盖路径，不将 agent 关于碰到玻璃的自述当成已观察到的碰撞。"
+   }
+  ],
+  "gap_zh": "观察：输入只有头部相机视频，没有状态日志；视频约 0.9–5.1 s 操作区在视野外，无法区分未碰到盖、夹取失败或接触后受阻，故现象层留空、实际接触记未知。推断：根因的方向错误由交付生成脚本及其切向输出支持，但不能据此断言实际接触力、碰撞位置或跟踪误差。评分器未报告 DAG 的三个误差指标，native_metrics 均留 null；不能用会话自测数值冒充评分器指标。"
+ },
+ "summary_zh": "评分轨迹尝试打开搅拌机盖，但视频 5.3–7.6 s 显示盖仍平扣杯口，最终完成 0/1 节点。交付生成脚本把开盖路径初始方向设为向下，推断为开盖方向规划错误；后续反向修改没有进入最终 228 行交付。操作接触区在视频约 0.9–5.1 s 离开视野且无状态日志，故实际接触未知，具体接触失败现象留空。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "视频 0.5/1.7/3.0/5.3/7.6 s 帧与标签一致：操作区约 0.9–5.1 s 离开头部视野，5.3 s 后搅拌机圆盖仍平扣，节点未达成；断点、接触未知、现象留空并附 gap_zh 均合理，log 的 gen4 保存 228 行、j_lid≈0 与引用相符。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:40:05Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--2kpcmbe/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--2kpcmbe/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "b810a67c26989467b37937e6dc1d9294a0c74528737af48c21f8f961db2ab068",
+   "video_sha256": "3c251cc4cb68119549d75dc15a523a451dd2c44e83cab1a6230b79edc229b35d",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/configuration_matching_articulation_open_small_scale/trajectories/ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--5spzola.json

@@ -0,0 +1,198 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--5spzola",
+ "taskKey": "ispatialgym/configuration_matching_articulation_open_small_scale",
+ "dag_sha256": "d02aeb1ff9d6bc57baa2a14ae97e39013d8e5fefeea4780a10ac0ca80c4f15e7",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 371,
+   "sim_steps": 371,
+   "rows": 371,
+   "max_rows": 2382,
+   "replay_success": 0,
+   "instance_ok": 1,
+   "steps": 371,
+   "success_step": 0,
+   "symbolic_success": 1,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 1,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.0,
+   "peak_processes": 0.0,
+   "peak_disk_gb": 0.0,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--5spzola/replay.mp4",
+  "duration_s": 12.5,
+  "fps": 10.0,
+  "frames": 125,
+  "sha256": "6e8519a62cc1f5a1910a6274f5620355541ecc5d545627dc6f364ecc02173d18"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "blender_lid_opened": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 搅拌机直立在白色桌面上，圆盖平扣在透明杯口上，开盖节点初始不成立。"
+     },
+     {
+      "t": 8.5,
+      "log_step": null,
+      "text_zh": "观察：视频 8.5–9.9 s 搅拌机重新完整进入视野，圆盖仍近水平覆在杯口上；视频 11.5–12.4 s 逐帧检查的终态相同。与参考成功视频 13.0 s、17.0 s 的杯口露出、盖子立在侧方不同，本次未达到目标开度。"
+     },
+     {
+      "t": 12.4,
+      "log_step": 55,
+      "text_zh": "推断：log step 55 的 m8 测试记录六段扫动后的盖关节仅约 0、0、0、0、0.243、0.196 rad，log step 56 明确将该测试的全部 371 行保存为交付轨迹；评分器报告同为 371 行且 deterministic=1、replay_success=0。结合视频中的接近、离开视野、撤回及近闭盖终态，判断该节点全程未达成；中段画面外的关节数值仅为所交付脚本的会话诊断支持，未当作回放状态日志。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：视频 1.0–1.8 s 可见右侧夹爪向搅拌机方向接近，随后盖缘和夹爪接触区落到画面右下方之外；视频 2.0–8.4 s 的主要操作未显示接触点。log step 55 的脚本意图用右夹爪指垫压推盖缘，但画面不能区分实际碰到盖缘、杯身还是未夹住，故不据脚本意图标抓握或压推。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "blender_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "blender_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "blender_lid_joint_error",
+   "unit": "rad",
+   "threshold": 0.4305347800398469,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "blender_lid_opened",
+   "t": 8.5,
+   "log_step": 55,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 1.0–1.8 s 机器人右夹爪接近盖子后移出视野，视频 8.5 s 撤回过程中再次看见的圆盖仍覆在杯口，直至视频 12.4 s 均未修复。log step 55–56 确认交付的是盖缘推压后沿铰链弧线扫动的 m8 轨迹。以视频 8.5 s 首次清楚重新看见失败结果作导航锚点，具体失效接触时刻在画面外。"
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 8.5,
+    "log_step": null,
+    "evidence_zh": "观察：视频 8.5–9.9 s 和视频 11.5–12.4 s，机器人撤离后圆盖仍大致水平、覆盖透明杯口，未像参考成功视频 13.0 s、17.0 s 那样在杯口侧方竖起；盖子最终开度明显不足。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "抓取和接触设计",
+    "alt_feasible": "未知",
+    "log_step": 55,
+    "source_zh": "日志",
+    "evidence_zh": "推断：log step 55 的已交付 m8 脚本选用右夹爪单侧指垫水平压向盖缘，再闭合夹爪并扫动末端位置，未设置另一只手稳定机身。该测试回显夹���闭合开口为 0，盖关节最后只有 0.1956 rad，同时机身位置从初始约 (8.5362,3.1081,0.8884) m 变到 (8.3996,3.2739,0.8924) m；log step 56 保存了该测试轨迹。结合视频 8.5–12.4 s 的近闭盖结果，所选接触几何未有效把末端扫动转化为盖子绕铰链的开启，支持接触设计不适配的归因；具体碰撞面和是否曾夹住仍未得到回放接触数据验证。"
+   }
+  ],
+  "gap_zh": "主要开盖操作位于视频右下边界之外，无法直接确认接触对象、接触方式及具体失效时刻，也不能逐帧核验中段最大开度。inputs/ 无状态日志，评分器未报告三项原生位姿误差或45步稳定窗口诊断，故 native_metrics 均留 null；节点未达成结合可见终态与明确交付的 m8 诊断推断，未把其他会话测试当作进度。agent 在 log step 55、58 声称盖顶无碰撞网格，这一解释未经独立验证，不据此归咎环境。",
+  "unmet": {
+   "blender_lid_opened": false
+  }
+ },
+ "summary_zh": "本次交付的 371 行轨迹尝试用右夹爪推压盖缘并沿铰链弧线扫动，唯一开盖节点未完成，终态为 0/1。视频 8.5–12.4 s 可见圆盖仍近水平覆在杯口，已交付脚本的测试反馈支持接触设计未能有效带动开盖。主要接触区在画面外，无状态日志，实际接触方式、精确失效时刻与稳定窗口细节无法核实。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "唯一节点未达成、断点、现象和根因均有视频终态（盖近水平覆在杯口）与已交付 m8 轨迹的日志数值支持；contact_actual 标未知符合画面看不到接触的事实，未发现需要改内容的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:41:48Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--5spzola/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--5spzola/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "9f008df96c013b1b0d04d22ce6c4bd157f1526b3408ed92b96b593366ba0401f",
+   "video_sha256": "6e8519a62cc1f5a1910a6274f5620355541ecc5d545627dc6f364ecc02173d18",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/configuration_matching_articulation_open_small_scale/trajectories/ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--7tqdhtt.json

@@ -0,0 +1,218 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--7tqdhtt",
+ "taskKey": "ispatialgym/configuration_matching_articulation_open_small_scale",
+ "dag_sha256": "d02aeb1ff9d6bc57baa2a14ae97e39013d8e5fefeea4780a10ac0ca80c4f15e7",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 225,
+   "sim_steps": 225,
+   "rows": 225,
+   "max_rows": 2382,
+   "replay_success": 0,
+   "instance_ok": 1,
+   "steps": 225,
+   "success_step": 0,
+   "symbolic_success": 0,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 16.977,
+   "peak_memory_gb": 10.515,
+   "peak_processes": 25,
+   "peak_disk_gb": 0.655,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--7tqdhtt/replay.mp4",
+  "duration_s": 7.6,
+  "fps": 10.0,
+  "frames": 76,
+  "sha256": "de9297ff67bee31b76faa4b901af60fe4836bacfc379e4a33be9faeb88eaf0c1"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "blender_lid_opened": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 搅拌机圆盖平扣透明杯口，初态未打开。"
+     },
+     {
+      "t": 3.0,
+      "log_step": 539,
+      "text_zh": "观察：视频 1.8–2.4 s 左夹爪尚不可见，杯体与平扣的盖面仍有部分露在画面左下边缘；视频约 2.8–3.2 s 左夹爪部件从画面左缘出现，视频 3.6–4.5 s 可见左夹爪在盖上方移动，视频 5.0–5.5 s 撤离，可见盖面未被带起。log step 539 是实际保存轨迹的脚本：左夹爪闭合，执行接近、弧线扫动、撤离及60步保持。"
+     },
+     {
+      "t": 7.5,
+      "log_step": null,
+      "text_zh": "观察：逐帧检查视频 6.5–7.5 s，左臂已撤离，圆盖仍横盖杯口，未出现成功参考视频 18.0 s 所见的杯口露出、盖翻到侧方的状态。评分器 replay_success=0、symbolic_success=0；未完成的判断来自可见闭盖状态。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：视频约 2.8–3.2 s 左夹爪部件从画面左缘出现，视频 3.6–4.5 s 可见其在搅拌机盖上方移动，视频 5.0–5.5 s 撤离；盖缘与手指接近处多在画面左边界之外，没有可辨认的夹住或勾住盖缘画面。无法确定是否短暂擦碰及具体接触方式，不能将log step 539的推盖意图当作实际推/扫接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "blender_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "blender_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "blender_lid_joint_error",
+   "unit": "rad",
+   "threshold": 0.4305347800398469,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "blender_lid_opened",
+   "t": 3.0,
+   "log_step": 539,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 1.8–2.4 s 左夹爪尚不可见，部分杯体与平扣的盖面仍在画面左下边缘；视频约 2.8–3.2 s 左夹爪部件从画面左缘出现，视频 3.6–4.5 s 可见其在盖上方移动，视频 5.0–5.5 s 撤离，而可见盖面未被带起，至视频 7.5 s 仍未开盖。log step 539保存的225步轨迹正是该左臂扫动和撤离动作；断点锚在这次未能带起盖的操作，而非会话中的其他测试。"
+  },
+  "unmet": {
+   "blender_lid_opened": false
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没建立接触",
+    "t": 3.0,
+    "log_step": null,
+    "evidence_zh": "推断：视频约 2.8–3.2 s 左夹爪部件从画面左缘出现，视频 3.6–4.5 s 可见其在盖上方移动而盖未随动，视频 5.0–5.5 s 夹爪撤离后盖仍平扣杯口；结合这些画面，判断未建立能夹住或勾带盖缘并使其绕铰链转动的有效接触。视频 1.8–2.4 s 夹爪不可见，不能据此观察其动作；这里不排除画外曾有短暂擦碰，也没有已持盖后滑脱的可见过程。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "source_zh": "日志 / 视频",
+    "evidence_zh": "推断：log step 539中agent脚本以固定初始夹爪朝向求各弧线航点逆解、插值关节指令，并持续闭合左夹爪，执行过程未根据接触或盖关节状态修正动作；视频约 2.8–3.2 s 左夹爪部件从画面左缘出现，视频 3.6–4.5 s 可见左末端在盖上方移动，视频 5.0–5.5 s 撤离而盖未被带起。这支持开盖操作的控制失败；缺少本次接触与末端跟踪数据，不能进一步确定是接触几何还是执行跟踪问题。"
+   }
+  ],
+  "gap_zh": "观察：inputs/没有状态日志；视频的盖缘与手指接近处多被左画面边界截断，因此实际接触填未知，现象中的有效接触未建立是结合盖始终不随动与末端撤离的推断，不能排除短暂擦碰。评分器未报告DAG列出的三项原生误差值，均填null；无法精确核验机身漂移、盖角度或45步稳定窗口，但可见闭盖已足以判定节点不成立。会话测试的关节读数未作为本次节点时间或数值。"
+ },
+ "summary_zh": "评分轨迹中机器人左臂接近搅拌机、向上扫动并撤离，圆盖从初态到视频 7.5 s 均平扣杯口，打开搅拌机盖节点未完成，最终0/1。断点在开盖操作，推断左末端未建立带动盖缘的有效接触，根因归控制；接触区域贴近画面左边界且无状态日志，实际接触方式和更细的控制原因无法确定。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 2,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "fix",
+    "conf": 7,
+    "summary_zh": "节点未达成（盖始终平扣杯口，评分器 success=0）、断点、现象与根因类别均有依据，无需改动；只有证据文字中夹爪出现的时间说早了。",
+    "n_issues": 1,
+    "n_major": 0
+   },
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 2,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "节点未达成（盖始终平扣杯口，评分器 success=0）、断点、接触未知、现象与根因均有依据；上一轮指出的夹爪出现时间已改正，无新问题。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:51:07Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--7tqdhtt/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--7tqdhtt/04_review_r1"
+   },
+   {
+    "stage": "reconcile",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--7tqdhtt/05_reconcile_r1"
+   },
+   {
+    "stage": "traj_review",
+    "round": 2,
+    "ok": true,
+    "space": "work/runs/ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--7tqdhtt/04_review_r2"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "4fabf7afc2b49c20ccc5b08d2296b6cd4aaab12e3a1a8553b04da8bc0a9b1fdc",
+   "video_sha256": "de9297ff67bee31b76faa4b901af60fe4836bacfc379e4a33be9faeb88eaf0c1",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/configuration_matching_articulation_open_small_scale/trajectories/ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--8jff6cp.json

@@ -0,0 +1,204 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--8jff6cp",
+ "taskKey": "ispatialgym/configuration_matching_articulation_open_small_scale",
+ "dag_sha256": "d02aeb1ff9d6bc57baa2a14ae97e39013d8e5fefeea4780a10ac0ca80c4f15e7",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 200,
+   "sim_steps": 200,
+   "rows": 200,
+   "max_rows": 2382,
+   "replay_success": 0,
+   "instance_ok": 1,
+   "steps": 200,
+   "success_step": 0,
+   "symbolic_success": 1,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 16.886,
+   "peak_memory_gb": 9.912,
+   "peak_processes": 21,
+   "peak_disk_gb": 3.126,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--8jff6cp/replay.mp4",
+  "duration_s": 6.8,
+  "fps": 10.0,
+  "frames": 68,
+  "sha256": "91bba3eeb2787b4aefc07c1338d5550ebccfb4d0262d69d0412e4e4730b3b651"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "blender_lid_opened": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–1.6 s，搅拌机直立在白桌上，圆盖平扣杯口，开盖节点初始不成立。"
+     },
+     {
+      "t": 4.6,
+      "log_step": null,
+      "text_zh": "观察：逐帧检查视频 2.8–5.4 s，机器人左夹爪抵近并拨动盖缘，圆盖仅小幅翘起；可见盖面仍接近杯口，未出现成功参考视频 14.0–18.0 s 中杯口露出的开盖状态。"
+     },
+     {
+      "t": 6.7,
+      "log_step": null,
+      "text_zh": "观察：视频 5.7–6.7 s 的全部末段帧中，机器人左夹爪与杯盖基本停住，盖仍处于小开度，搅拌机留在桌上；评分器 reward.replay_success=0、success_step=0，虽有 symbolic_success=1，仍未通过节点所要求的完整开度、位姿与稳定联合验收。"
+     },
+     {
+      "t": null,
+      "log_step": 293,
+      "text_zh": "观察：log step 293 中 agent 的脚本将动作行截取为前200行并保存，log step 310、323 的两次新进程回放均报告200行且终态哈希同为8172ba39f53eee1e。log step 325 中 agent 自述盖约开0.24 rad且不匹配目标图；该角度是会话自述，不作为评分视频的精确关节读数。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "推/扫",
+      "note_zh": "观察：视频 2.8–5.1 s，机器人左夹爪的指端抵住盖缘并推拨，使盖小幅翘起，未见两指夹牢盖后带动其绕铰链持续翻开。推断：结合log step 281、285的会话接触诊断（仅left_gripper_finger_link1接触盖，辅助抓握未建立），实际作用为单侧指端推拨；未将闭爪指令本身计为抓握。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "blender_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "blender_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "blender_lid_joint_error",
+   "unit": "rad",
+   "threshold": 0.4305347800398469,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "blender_lid_opened",
+   "t": 4.6,
+   "log_step": null,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 4.3–5.1 s，机器人左夹爪在盖缘处调整，盖仅保持小幅翘起；随后视频 5.7–6.7 s 动作停住，未将盖继续掀到目标开度。唯一开盖步骤已实际尝试，但此后没有修复。"
+  },
+  "unmet": {
+   "blender_lid_opened": false
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "接触力异常",
+    "t": 4.6,
+    "log_step": null,
+    "evidence_zh": "观察：视频 3.0–5.1 s，机器人左夹爪抵着盖缘操作，盖只发生小幅转动，始终没有像成功参考视频 14.0–18.0 s 那样让杯口露出；视频 5.7–6.7 s 仍停在该小开度。推断：指端接触未产生足够的开盖作用，表现为盖未被推开到位；画面不支持已抓牢后滑落。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "抓取和接触设计",
+    "alt_feasible": "未知",
+    "t": 4.6,
+    "log_step": 281,
+    "evidence_zh": "推断：左夹爪采用单侧指端抵推盖缘的接触布局，没有形成能随盖绕铰链运动的稳定夹持或勾持。依据是视频 3.0–5.1 s 中指端贴近盖缘而盖仅小幅翘起，以及交付前同段动作的会话诊断：log step 281 仅报告left_gripper_finger_link1接触lid、grasp左右均为None；log step 285 微调后仍仅单指接触，log step 287 抬高末端后接触集合为空且盖未继续打开。log step 293 保存这段动作的前200行；这些会话诊断用于解释接触设计，不转作评分视频的精确关节或接触状态。",
+    "source_zh": "日志 / 视频"
+   }
+  ],
+  "gap_zh": "观察：inputs/中没有状态日志；评分器未报告DAG所列的机身位置误差、朝向误差和盖关节误差，也没有逐步稳定窗口或接触力数据。视频为单一头部视角，操作时盖的一部分及指端位于画面左沿，无法量化接触力或精确角度；小开度未达目标仍可从可见盖面和成功参考的杯口状态判断。log step 290的交互终态哈希71c50db4ae368c79与log step 310、323的新进程回放终态哈希8172ba39f53eee1e不同，因此不把交互自测的精确角度、接触对或机身误差当作评分回放事实。"
+ },
+ "summary_zh": "观察：评分轨迹中机器人左夹爪接触并推拨搅拌机盖，但盖仅小幅翘起，唯一开盖节点从未完成，最终为0/1。推断：单侧指端接触未形成持续带动盖绕铰链开启的有效接触布局，断点为打开搅拌机盖的操作中。观察：无状态日志和评分回放的精确误差或接触力记录，部分盖缘受画面左沿遮挡，根因细节结合会话诊断推断。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "逐帧核对：评分视频中左夹爪只把盖推起小幅，盖未达目标开度（会话读数约0.24 rad，远低于目标1.435±0.43 rad，且新进程回放终态哈希一致），节点未达成、断点、接触力异常现象与抓取和接触设计根因均有证据支持，未发现需要修改的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:47:47Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--8jff6cp/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--8jff6cp/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "97ab97a9b184aa28d7b1b9f1d6929ef949d948a8fb163bdfe3925ac66422078a",
+   "video_sha256": "91bba3eeb2787b4aefc07c1338d5550ebccfb4d0262d69d0412e4e4730b3b651",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/configuration_matching_articulation_open_small_scale/trajectories/ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--b9alyoh.json

@@ -0,0 +1,207 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--b9alyoh",
+ "taskKey": "ispatialgym/configuration_matching_articulation_open_small_scale",
+ "dag_sha256": "d02aeb1ff9d6bc57baa2a14ae97e39013d8e5fefeea4780a10ac0ca80c4f15e7",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 1334,
+   "sim_steps": 1334,
+   "rows": 1334,
+   "max_rows": 2382,
+   "replay_success": 0,
+   "instance_ok": 1,
+   "steps": 1334,
+   "success_step": 0,
+   "symbolic_success": 1,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 17.039,
+   "peak_memory_gb": 9.085,
+   "peak_processes": 26,
+   "peak_disk_gb": 1.377,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--b9alyoh/replay.mp4",
+  "duration_s": 44.6,
+  "fps": 10.0,
+  "frames": 446,
+  "sha256": "b8f0e0c2d153b903cef5ab61b3721d4eb55ea6a09c4b4f896db322bc142a9e3e"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "blender_lid_opened": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 搅拌机直立在白色桌面，圆盖平扣杯口，尚未打开。"
+     },
+     {
+      "t": 5.2,
+      "log_step": null,
+      "text_zh": "观察：视频 5.2 s 左夹爪二指位于圆盖自由缘两侧；视频 12.0–21.0 s 夹爪与圆盖一起转动，盖逐渐离开杯口，机身仍直立在桌面。"
+     },
+     {
+      "t": 23.5,
+      "log_step": null,
+      "text_zh": "观察：视频 19.0–23.5 s 的密集帧显示左夹爪继续带动盖向上掀，杯口逐渐露出，随后盖与搅拌机移出画面左边缘。成功参考视频约 18.0–20.0 s 也显示夹爪掀盖、杯口露出；此参考只用于识别动作外观。"
+     },
+     {
+      "t": 44.5,
+      "log_step": null,
+      "text_zh": "观察：视频 43.6–44.5 s 仍看不到搅拌机及盖；评分器 reward 报告 steps=1334、success_step=0、replay_success=0、symbolic_success=1、deterministic=1，且没有仿真或评分器错误。"
+     },
+     {
+      "t": null,
+      "log_step": 71,
+      "text_zh": "推断：本节点包含目标位姿与连续45控制步原生验收；结合视频中的实际开盖动作及评分器未在任何控制步触发成功，节点从未完整达成。log step 64 已交付1334行文件，log step 66 的自测脚本明确重放该文件，log step 71 打印盖角1.91144 rad；该自测值相对DAG目标1.43512 rad偏大约0.47633 rad，超过0.43053 rad容差，支持终态开度过大的解释，但不是正式评分回放的逐帧状态。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "观察：视频 5.2 s 左夹爪二指夹住圆盖自由缘，视频 12.0–23.5 s 夹爪随圆盖转动，将盖绕侧铰链掀起。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "blender_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "blender_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "blender_lid_joint_error",
+   "unit": "rad",
+   "threshold": 0.4305347800398469,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "blender_lid_opened",
+   "t": null,
+   "log_step": 67,
+   "where": "操作中",
+   "evidence_zh": "推断：视频 5.2–23.5 s 机器人左夹爪已实际抓盖并持续掀开；该开盖步骤最终未通过原生验收。log step 64 交付1334行轨迹，log step 67 的生成脚本指定开盖终值THETA=1.9 rad，log step 71 对同一文件的自测终值约1.91144 rad，高于DAG允许上界约1.86565 rad；评分器replay_success=0。盖在操作后段已移出画面，不能可靠定位越过容差的具体视频时刻，故时间留空。"
+  },
+  "unmet": {
+   "blender_lid_opened": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": null,
+    "log_step": 71,
+    "evidence_zh": "推断：视频 19.0–23.5 s 机器人夹爪带盖连续掀起，随后目标移出左边缘。结合log step 64交付的文件、log step 66明确读取���文件的重放脚本、log step 71打印终态盖角1.91144 rad，以及DAG目标1.43512 rad和0.43053 rad容差，开盖终值过大是有依据的失败现象；正式评分器symbolic_success=1而replay_success=0与此相容。精确终态角未从视频直接量得。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "认知",
+    "item": "感知",
+    "log_step": 72,
+    "evidence_zh": "推断：agent 自述“凭目标图目测选了1.9 rad，没有匹配目标相机视角”（log step 72）；log step 67的生成脚本确实设置THETA=1.9，log step 71同文件自测到约1.91144 rad。与DAG目标关节1.43512 rad比较，agent对目标图开度的估计偏大，使所选终值落在容差之外。",
+    "source_zh": "agent 自述"
+   }
+  ],
+  "gap_zh": "没有模拟器状态日志或逐步原生验收诊断；视频约 24.0 s 后搅拌机及盖处于画外，视频 43.6–44.5 s 的密集末帧也无法补足。开度过大的判断由已交付文件的自测及目标容差共同支持，自测数值不当作正式回放的原生指标；无法确定正式终态精确误差、越界视频时刻，或是否另有稳定检查同时失败。评分器未报告DAG三项原生误差，native_metrics.value均留空。"
+ },
+ "summary_zh": "机器人左夹爪实际夹住盖缘并掀开搅拌机盖，但“打开搅拌机盖”节点包含目标开度与连续45控制步原生验收，整段均未完整达成，终态为0/1。已交付文件的自测盖角约1.911 rad，超出目标允许上界约1.866 rad，结合agent承认目测选择1.9 rad，推断目标图开度估计偏大导致位姿不准。视频后段盖在画外且没有状态日志，精确失败时刻与其他稳定条件无法核实。",
+ "conf": 6,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "标签与视频、log一致：左夹爪确实抓盖掀开（视频 5.2–23.5 s），评分器 replay_success=0；已交付文件自测盖角 1.911 rad 超出 DAG 上界 1.866 rad（log step 71），位姿不准与感知（目测 1.9 rad，log step 67/72）作为推断标注合理，时间留空与 gap_zh 诚实。未发现需改动的内容。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:30:08Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--b9alyoh/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--b9alyoh/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "1d696d41f28e01419d466ad4a8fd3bd2dcd4ef2d852b657424e47737dc83b9f4",
+   "video_sha256": "b8f0e0c2d153b903cef5ab61b3721d4eb55ea6a09c4b4f896db322bc142a9e3e",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/configuration_matching_articulation_open_small_scale/trajectories/ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--mhvsi3q.json

@@ -0,0 +1,174 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--mhvsi3q",
+ "taskKey": "ispatialgym/configuration_matching_articulation_open_small_scale",
+ "dag_sha256": "d02aeb1ff9d6bc57baa2a14ae97e39013d8e5fefeea4780a10ac0ca80c4f15e7",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 60,
+   "sim_steps": 60,
+   "rows": 60,
+   "max_rows": 2382,
+   "replay_success": 0,
+   "instance_ok": 1,
+   "steps": 60,
+   "success_step": 0,
+   "symbolic_success": 0,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 14.497,
+   "peak_memory_gb": 10.214,
+   "peak_processes": 25,
+   "peak_disk_gb": 0.638,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--mhvsi3q/replay.mp4",
+  "duration_s": 2.1,
+  "fps": 10.0,
+  "frames": 21,
+  "sha256": "f564bc0f148cdfec7f70bf0df70ee6e182bfd65f7611c58f2632cf028c168f77"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "blender_lid_opened": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：逐帧检查本次视频 0.0–2.0 s（全部21帧），白桌上的搅拌机圆盖始终水平扣在透明杯口，未见夹爪接近、掀盖或机身搬动；视频 1.0–2.0 s 的末段也没有变化。"
+     },
+     {
+      "t": 2.0,
+      "log_step": null,
+      "text_zh": "观察：本次视频 2.0 s 末帧杯盖仍闭合；对照成功参考视频 24.0 s，参考圆盖已近竖直掀起并露出杯口。本次缺少可见的开盖结果，节点不成立。"
+     },
+     {
+      "t": null,
+      "log_step": 27,
+      "text_zh": "推断：log step 27 的检查脚本将 hold_action 重复60次写入交付文件，评分器报告 rows=60、n_actions=60、deterministic=1，且本次视频全程保持初态，三者支持评分录像对应这段保底保持轨迹；log step 69–74 的移动、合爪及 ATTACHED 输出属于会话测试，不能作为评分轨迹的接触或进度。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "blender_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "blender_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "blender_lid_joint_error",
+   "unit": "rad",
+   "threshold": 0.4305347800398469,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 77,
+   "where": "首步前",
+   "evidence_zh": "观察：本次视频 0.0–2.0 s 未出现开盖操作；log step 27 的脚本保存60步保持初态的保底轨迹，log step 69–74 仍在调试抓取，log step 77 最后输出提供方过滤提示“This request was blocked by Gemini's filters”。日志没有后续保存并验证开盖交付的记录，无法定位已执行操作中的物理失败时刻。"
+  },
+  "unmet": {
+   "blender_lid_opened": false
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：输入没有状态日志，评分器也未报告三项原生误差值；本次视频 0.0–2.0 s 只有保底保持轨迹。log step 74 的会话测试输出 ATTACHED，log step 75 记录预算尚余1775秒，随后 log step 77 出现提供方过滤提示，未见最终开盖交付保存及回放验证。推断：这是任务结果确定前的外部过滤中断，按 CASES.md FAIL-5 处理；保留目录 success=0 和可见的0/1进度，但不归入机器人能力失败，不标没尝试或从会话抓取测试推导本次现象、根因；断点保留 unknown／首步前。"
+ },
+ "summary_zh": "评分视频全程保持初态，搅拌机盖未打开，唯一节点未达成，终态进度为0/1。log step 27 保存的是60步保持初态的保底轨迹；log step 69–74 的抓取调试没有进入本次录像。log step 77 显示提供方过滤中断，尚无最终方案保存并验证的证据，故按外部中断保留空现象、空根因；无状态日志及原生误差读数。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "评分视频全程保持初态，盖未开；log step 27 只保存了60步保持轨迹，step 77 为提供方过滤中断，之后无开盖交付。按 FAIL-5 保留 unknown/首步前、空现象空根因合理，未发现需要修改的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:19:14Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--mhvsi3q/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--mhvsi3q/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "f8dec3f738ba160b0bf1daf47b26df80f81c128008617066e237886d9004b46e",
+   "video_sha256": "f564bc0f148cdfec7f70bf0df70ee6e182bfd65f7611c58f2632cf028c168f77",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/configuration_matching_articulation_open_small_scale/trajectories/ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--szjzbpw.json

@@ -0,0 +1,191 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--szjzbpw",
+ "taskKey": "ispatialgym/configuration_matching_articulation_open_small_scale",
+ "dag_sha256": "d02aeb1ff9d6bc57baa2a14ae97e39013d8e5fefeea4780a10ac0ca80c4f15e7",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 121,
+   "sim_steps": 121,
+   "rows": 121,
+   "max_rows": 2382,
+   "replay_success": 0,
+   "instance_ok": 1,
+   "steps": 121,
+   "success_step": 0,
+   "symbolic_success": 0,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 1,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.0,
+   "peak_processes": 0.0,
+   "peak_disk_gb": 0.0,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--szjzbpw/replay.mp4",
+  "duration_s": 4.2,
+  "fps": 10.0,
+  "frames": 42,
+  "sha256": "7c1c504e9b9117db0cb81c545abe55b10c5ccf3eb2eabfde7c4b27f763968a47"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "blender_lid_opened": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：评分视频 0.0–4.1 s 的全部42帧中，搅拌机留在白色桌面，圆盖始终平扣杯口；机器人仅靠近桌边和调整姿态，画面右下方出现的夹爪始终与搅拌机分离，没有抓盖、勾盖或掀盖动作。最后一秒的视频 3.2–4.1 s 仍是闭盖。"
+     },
+     {
+      "t": 4.1,
+      "log_step": 236,
+      "text_zh": "观察：log step 236 的 agent 脚本先 reset，再执行底盘前进、停底盘、关节回中性位和55步保持，保存121行轨迹；该步自测盖关节约为0 rad。log step 240、248、250确认最终文件仍为121行，log step 251明确自述交付文件不达开盖目标；评分器字段 n_actions=121、deterministic=1、replay_success=0，与视频中的基线动作一致。此处日志仅用于识别交付内容，未把会话测试计为评分进度。"
+     },
+     {
+      "t": 4.1,
+      "log_step": null,
+      "text_zh": "推断：参考成功视频 8–18 s 中，圆盖转离杯口、杯口露出；本次评分视频 0.0–4.1 s 从未出现这种开盖形态，故节点从未成立，而非仅凭零分判未完成。参考视频只用于对照外观，不作为本次动作证据。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "blender_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "blender_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "blender_lid_joint_error",
+   "unit": "rad",
+   "threshold": 0.4305347800398469,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "blender_lid_opened",
+   "t": 0.0,
+   "log_step": 236,
+   "where": "首步前",
+   "evidence_zh": "观察：评分视频 0.0–4.1 s 只有靠近桌边和姿态调整，杯盖一直闭合，没有开始操作杯盖；log step 236 保存的就是不含开盖动作的121步基线，log step 248、250核对最终仍交付该文件。故以视频 0.0 s 为首步未启动的导航锚点；会话中其他开盖测试不在评分轨迹内。"
+  },
+  "unmet": {
+   "blender_lid_opened": false
+  },
+  "phenomena": [
+   {
+    "cat": "没尝试",
+    "item": "没尝试",
+    "t": 0.0,
+    "log_step": 236,
+    "evidence_zh": "观察：评分视频 0.0–4.1 s 中，机器人未把夹爪伸到杯盖上进行抓握、勾拉或推动，盖始终扣住杯口；log step 236 的最终保存脚本只包含前进、回中性位和静止保持。虽然机器人有准备性运动，交付的基线没有启动开盖操作。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "t": null,
+    "log_step": 219,
+    "source_zh": "日志",
+    "evidence_zh": "推断：会话自测表明 agent 未得到有效的开盖控制：log step 199 的闭爪及后续运动中，自测 j_lid 约维持0.413 rad；log step 219 的按压拖动自测末端逆解位置残差约0.00096 m，但 j_lid 仍约0；log step 229 的深压自测也仅约0.047 rad。随后 agent 在 log step 236 reset并保存无开盖动作的基线，log step 251自述这些操作未能打开盖���交付文件不达目标。因此以会话证据追溯为控制能力未解决；不把上述测试当作本次视频中的接触现象，也不把停止或时间预算当作根因。"
+   }
+  ],
+  "gap_zh": "没有仿真状态日志，评分器未报告三项原生误差数值；闭盖及未启动操作可由视频直接判断，但精确机身误差、细微稳定量不可复核。根因仅由会话自测与最终保存记录推断，缺少接触力和可靠的抓持状态证据，无法细分接触设计、轨迹或反馈控制；agent 在 log step 251 关于资产无法抓取、所有合法接触均不可行的说法未经验证，不能据此归因任务或环境。"
+ },
+ "summary_zh": "评分轨迹仅靠近桌边、调整姿态后保持静止，视频 0.0–4.1 s 杯盖始终闭合，唯一节点“打开搅拌机盖”未达到，最终为0/1。断点在首步前，现象为没尝试：log step 236 保存的是未含开盖操作的121步基线；会话中的开盖控制测试未解决目标，支持宽泛的控制根因。无状态日志和原生误差数值，控制子类及精确稳定情况无法核实，不能把会话测试中的接触或物体倾倒记入评分轨迹。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "视频 0.0–4.1 s 杯盖始终闭合，节点未达成、final=false 正确；首步前/没尝试，以及依据会话证据的控制根因，均符合 CASES.md FAIL-22 的处理方式，未发现需改动的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:30:28Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--szjzbpw/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-configuration-matching-articulation-open-small-scale-i00-privileged--szjzbpw/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "05b24c49a5de832d10584ed3f3a2b2d2122f323a414f01ff5343fa62203b3656",
+   "video_sha256": "7c1c504e9b9117db0cb81c545abe55b10c5ccf3eb2eabfde7c4b27f763968a47",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/configuration_matching_block_design/trajectories/ispatialgym-configuration-matching-block-design-i00-privileged--hm2cjxz.json

@@ -0,0 +1,277 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-configuration-matching-block-design-i00-privileged--hm2cjxz",
+ "taskKey": "ispatialgym/configuration_matching_block_design",
+ "dag_sha256": "b80b8b64018446fb9084afd442651aea5f42a11bee539a14524ad1124ad51a8f",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 239,
+   "sim_steps": 239,
+   "rows": 239,
+   "max_rows": 8298,
+   "replay_success": 0,
+   "instance_ok": 1,
+   "steps": 239,
+   "success_step": 0,
+   "symbolic_success": 0,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 1,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.0,
+   "peak_processes": 0.0,
+   "peak_disk_gb": 0.0,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-configuration-matching-block-design-i00-privileged--hm2cjxz/replay.mp4",
+  "duration_s": 8.1,
+  "fps": 10.0,
+  "frames": 81,
+  "sha256": "49f1d91d9a2d412b43ad012da0b95f3a83ab4b45604cba515261048c2d861600"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 5,
+  "base": 1,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "tray_on_table": {
+    "method": "visual",
+    "intervals": [
+     [
+      0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s，空的四格木托盘开口向上，位于白桌靠玻璃一侧，四块积木散放在其右侧桌面。"
+     },
+     {
+      "t": 0.8,
+      "log_step": 66,
+      "text_zh": "推断：视频 0.0–0.8 s 托盘与白桌的相对位置不变，随后因机器人头部相机随底盘接近而移出主要视野；视频 6.0–8.0 s 左下角仍见木色盘沿。全视频未见物体操作，log step 66 的交付生成日志只有底盘移动成功、四次搬放均报错，支持托盘持续留在原桌面区域。"
+     }
+    ]
+   },
+   "blocks_in_layout_1": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s，木托盘四格均为空，四块积木分散在托盘外的白桌上，未形成任何入盘目标格位。"
+     },
+     {
+      "t": 8.0,
+      "log_step": null,
+      "text_zh": "观察：逐帧检查视频 0.0–8.0 s，积木随相机运动改变画面位置，但保持桌面上的散放关系，未见夹爪接近、拿起或入盘；视频 7.0–8.0 s 三块积木完整可见，近处第四块仍在画面下缘，未达到至少1块入盘的条件。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "底盘",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.0–约 5.0 s 机器人头部视角随底盘向积木接近并作位置修正；log step 49 的生成代码及 log step 66 的底盘结果相符。轮式底盘承载机器人移动，为首块积木操作作准备；没有执行夹爪与积木的接触。"
+     }
+    ]
+   },
+   "blocks_in_layout_2": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s，木托盘四格均为空，四块积木分散在托盘外的白桌上，未形成任何入盘目标格位。"
+     },
+     {
+      "t": 8.0,
+      "log_step": null,
+      "text_zh": "观察：逐帧检查视频 0.0–8.0 s，积木随相机运动改变画面位置，但保持桌面上的散放关系，未见夹爪接近、拿起或入盘；视频 7.0–8.0 s 三块积木完整可见，近处第四块仍在画面下缘，未达到至少2块入盘的条件。"
+     }
+    ]
+   },
+   "blocks_in_layout_3": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s，木托盘四格均为空，四块积木分散在托盘外的白桌上，未形成任何入盘目标格位。"
+     },
+     {
+      "t": 8.0,
+      "log_step": null,
+      "text_zh": "观察：逐帧检查视频 0.0–8.0 s，积木随相机运动改变画面位置，但保持桌面上的散放关系，未见夹爪接近、拿起或入盘；视频 7.0–8.0 s 三块积木完整可见，近处第四块仍在画面下缘，未达到至少3块入盘的条件。"
+     }
+    ]
+   },
+   "blocks_in_layout_4": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s，木托盘四格均为空，四块积木分散在托盘外的白桌上，未形成任何入盘目标格位。"
+     },
+     {
+      "t": 8.0,
+      "log_step": null,
+      "text_zh": "观察：逐帧检查视频 0.0–8.0 s，积木随相机运动改变画面位置，但保持桌面上的散放关系，未见夹爪接近、拿起或入盘；视频 7.0–8.0 s 三块积木完整可见，近处第四块仍在画面下缘，未达到至少4块入盘的条件。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "group_layout_position_error",
+   "unit": "m",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "group_layout_orientation_error",
+   "unit": "rad",
+   "threshold": 0.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "group_pose_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "group_pose_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "blocks_in_layout_1",
+   "t": 0.0,
+   "log_step": 66,
+   "where": "首步前",
+   "evidence_zh": "观察：视频 0.0–8.0 s 只有机器人视角随底盘接近及停住，四块积木始终散放，未出现首块积木的实际抓取或搬放。log step 66 核对的交付文件为 239×23 个有限动作数值，生成结果中首块及其余三块的 place 调用均为 IndexError；log step 67 中 agent 也自述只保存了底盘接近与保持动作。视频 0.0 s 为未启动首个操作步骤的导航锚点，并非可见的异常发生时刻。"
+  },
+  "unmet": {
+   "tray_on_table": true,
+   "blocks_in_layout_1": false,
+   "blocks_in_layout_2": false,
+   "blocks_in_layout_3": false,
+   "blocks_in_layout_4": false
+  },
+  "phenomena": [
+   {
+    "cat": "没尝试",
+    "item": "没尝试",
+    "t": 0.0,
+    "log_step": 66,
+    "evidence_zh": "观察：视频 0.0–8.0 s 未见机器人机械臂或夹爪对积木执行接近抓取、搬运或摆放；四块积木保持原有散放关系，末秒也没有物体操作。交付视频中的首块入盘步骤未启动。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "log_step": 66,
+    "evidence_zh": "观察：log step 49 的 agent 生成脚本在每次 place 异常后捕获异常并继续，log step 64 使该脚本在末尾保持并保存动作；log step 66 核对这一交付文件为 239 行，同时四个 place 均记录 IndexError（index 1 is out of bounds for axis 0 with size 1），底盘结果则正常。视频 0.0–8.0 s 的接近后无积木操作与该日志相符；agent 自述异常在首个航点前中断搬放（log step 67）。",
+    "source_zh": "日志",
+    "note_zh": "候选名称：控制脚本异常。按 DECISIONS.md D38 使用控制·控制；异常来自 agent 自写的搬放辅助代码，不能据此归为仿真或评分器故障。"
+   }
+  ],
+  "gap_zh": "无状态日志或成功参考视频。托盘自视频约 0.9 s 后大部分位于视野外，其持续留桌结合初态、盘沿及没有物体操作的证据推断；末段近处第四块也只有局部可见。评分字段未报告四项原生误差值，因此 native_metrics 均为 null。生成日志只给出捕获的索引异常，没有最终调用栈，无法独立定位具体出错代码行，也无法把异常精确对应到视频时刻。"
+ },
+ "summary_zh": "评分轨迹只完成底盘接近与保持；托盘初始留桌状态保留，四块积木均未入盘，最终节点为 1/5。断点为第1块积木入盘的首步前，交付生成日志中的四次搬放调用均因 agent 自写控制代码的索引异常中断，与视频无积木操作相符。托盘后段大部分出画，其持续留桌为上下文推断；没有状态日志和原生误差读数。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "标签与证据一致：托盘开局成立并保留，四块积木均未入盘，断点首步前+没尝试；交付日志 step 66 的四次 IndexError 支持控制脚本异常根因，视频时间均在 8.1 s 内且未转换 agent 时间。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:22:40Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-configuration-matching-block-design-i00-privileged--hm2cjxz/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-configuration-matching-block-design-i00-privileged--hm2cjxz/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "c880b2d0cf123cd452aa56f46888741bbcd20e1162a7d47eaad33b8043dc2af6",
+   "video_sha256": "49f1d91d9a2d412b43ad012da0b95f3a83ab4b45604cba515261048c2d861600",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/configuration_matching_block_design/trajectories/ispatialgym-configuration-matching-block-design-i00-privileged--jsc5bka.json

@@ -0,0 +1,312 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-configuration-matching-block-design-i00-privileged--jsc5bka",
+ "taskKey": "ispatialgym/configuration_matching_block_design",
+ "dag_sha256": "b80b8b64018446fb9084afd442651aea5f42a11bee539a14524ad1124ad51a8f",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 1720,
+   "sim_steps": 1720,
+   "rows": 1720,
+   "max_rows": 8298,
+   "replay_success": 0,
+   "instance_ok": 1,
+   "steps": 1720,
+   "success_step": 0,
+   "symbolic_success": 0,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 23.92,
+   "peak_memory_gb": 16.722,
+   "peak_processes": 34,
+   "peak_disk_gb": 1.373,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-configuration-matching-block-design-i00-privileged--jsc5bka/replay.mp4",
+  "duration_s": 57.5,
+  "fps": 10.0,
+  "frames": 575,
+  "sha256": "7cb13d2525d79d8db68385060987adf1566a1e86fdd1792bdd364a54885b8079"
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 5,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "tray_on_table": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": 0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s、0.5 s 可见木托盘开口向上，位于白桌靠玻璃一侧，初始条件成立；视频 1.0 s 托盘已接近画面下边缘，随后离开视野。"
+     },
+     {
+      "t": 57.4,
+      "log_step": null,
+      "text_zh": "观察：视频 2.0–57.4 s 的检查帧中，托盘所在近端桌面区域在头部相机画外；视频 56.5 s、57.0 s、57.4 s 也未重新显示托盘，无法核实其终态或完整成立区间。intervals 为 null 表示后续不可判定，并非把离开视野标成条件失去。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：视频中未见夹爪接触托盘；操作期间托盘与夹爪均在画外，不能确认是否发生直接或经积木传递的接触。"
+     }
+    ]
+   },
+   "blocks_in_layout_1": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": 0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–1.0 s 可见四块彩色积木分散在白桌上，木托盘为空，尚无积木占据盘中目标格位。"
+     },
+     {
+      "t": 57.4,
+      "log_step": null,
+      "text_zh": "观察：视频 2.0–57.4 s 的检查帧只显示桌面远端及厨房、客厅背景，抓取位置、夹爪与盘中目标格位均在画外；无法确认何时有至少1块积木同时满足格位及完整朝向要求。"
+     },
+     {
+      "t": null,
+      "log_step": 44,
+      "text_zh": "观察：log step 44 的录制脚本安排左夹爪依次搬运四块积木，log step 49 将这份1720行动作文件作为交付；这些指令只能表明尝试范围，视频没有显示积木离开原位并进入格位，不能据此记为完成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：log step 44 的交付脚本使用左二指夹爪尝试搬运积木；视频中具体接触过程在画外，无法确认实际夹持方式，也无法把某次搬运认定为第1个有效格位。"
+     }
+    ]
+   },
+   "blocks_in_layout_2": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": 0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–1.0 s 可见四块彩色积木分散在白桌上，木托盘为空，尚无积木占据盘中目标格位。"
+     },
+     {
+      "t": 57.4,
+      "log_step": null,
+      "text_zh": "观察：视频 2.0–57.4 s 的检查帧只显示桌面远端及厨房、客厅背景，抓取位置、夹爪与盘中目标格位均在画外；无法确认何时有至少2块积木同时满足格位及完整朝向要求。"
+     },
+     {
+      "t": null,
+      "log_step": 44,
+      "text_zh": "观察：log step 44 的录制脚本安排左夹爪依次搬运四块积木，log step 49 将这份1720行动作文件作为交付；这些指令只能表明尝试范围，视频没有显示积木离开原位并进入格位，不能据此记为完成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：log step 44 的交付脚本使用左二指夹爪尝试搬运积木；视频中具体接���过程在画外，无法确认实际夹持方式，也无法把某次搬运认定为第2个有效格位。"
+     }
+    ]
+   },
+   "blocks_in_layout_3": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": 0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–1.0 s 可见四块彩色积木分散在白桌上，木托盘为空，尚无积木占据盘中目标格位。"
+     },
+     {
+      "t": 57.4,
+      "log_step": null,
+      "text_zh": "观察：视频 2.0–57.4 s 的检查帧只显示桌面远端及厨房、客厅背景，抓取位置、夹爪与盘中目标格位均在画外；无法确认何时有至少3块积木同时满足格位及完整朝向要求。"
+     },
+     {
+      "t": null,
+      "log_step": 44,
+      "text_zh": "观察：log step 44 的录制脚本安排左夹爪依次搬运四块积木，log step 49 将这份1720行动作文件作为交付；这些指令只能表明尝试范围，视频没有显示积木离开原位并进入格位，不能据此记为完成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：log step 44 的交付脚本使用左二指夹爪尝试搬运积木；视频中具体接触过程在画外，无法确认实际夹持方式，也无法把某次搬运认定为第3个有效格位。"
+     }
+    ]
+   },
+   "blocks_in_layout_4": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": 0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–1.0 s 可见四块彩色积木分散在白桌上，木托盘为空，尚无积木占据盘中目标格位。"
+     },
+     {
+      "t": 57.4,
+      "log_step": null,
+      "text_zh": "观察：视频 2.0–57.4 s 的检查帧只显示桌面远端及厨房、客厅背景，抓取位置、夹爪与盘中目标格位均在画外；无法确认何时有至少4块积木同时满足格位及完整朝向要求。"
+     },
+     {
+      "t": null,
+      "log_step": 44,
+      "text_zh": "观察：log step 44 的录制脚本安排左夹爪依次搬运四块积木，log step 49 将这份1720行动作文件作为交付；这些指令只能表明尝试范围，视频没有显示积木离开原位并进入格位，不能据此记为完成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：log step 44 的交付脚本使用左二指夹爪尝试搬运积木；视频中具体接触过程在画外，无法确认实际夹持方式，也无法把某次搬运认定为第4个有效格位。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "group_layout_position_error",
+   "unit": "m",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "group_layout_orientation_error",
+   "unit": "rad",
+   "threshold": 0.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "group_pose_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "group_pose_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": null,
+   "where": "未知",
+   "evidence_zh": "观察：视频 0.0 s 托盘为空且四块积木散放，随后头部视角转离操作区域；检查至视频 57.4 s 未见抓取、释放或格位结果。评分器 reward 报告 success=0、symbolic_success=0、1720行动作，但没有逐节点结果；无法从画面定位第一个未修复的失败步骤或发生时刻。"
+  },
+  "phenomena": [],
+  "root_causes": [
+   {
+    "cat": "交付",
+    "item": "交付操作出错",
+    "log_step": 49,
+    "t": null,
+    "source_zh": "日志 / agent 自述 / 评分器字段",
+    "evidence_zh": "推断：log step 44 明确在已有调试Session中调用 s.reset() 后录制并保存1720行动作，log step 49 把该文件复制为最终交付；log step 62–65 在第二个新进程直接回放同一文件，打印的托盘中心变为 [5.9064,4.0351,0.8252]，且积木3中心高度为0.8509米，与录制结果 log step 48 的0.8329551米不同。log step 66–67 又在reset后复测，输出的托盘位置再次不同，说明交付对初态变化不稳健；log step 72 确认最终动作文件仍为原1720行，agent 在 log step 75 自述新进程回放不匹配目标但仍保留原文件。评分器字段 deterministic=1、replay_success=0支持按实际失败交付标注；此处归因于录制及保留已知未通过复测的文件，不将本地复测中的具体物理结果当作视频节点事实。"
+   }
+  ],
+  "gap_zh": "观察：inputs/有完整头部视频和run.json，但无状态日志、腕部录像或逐格位评分结果；已按视频每0.5 s检查全程，并查看视频 0.0、0.5、1.0、1.5、56.5、57.4 s 的大图。操作区从视频约 2.0 s 起始终在画外，故无法核实五个节点的终值、实际接触及第一个未修复的物理失败，现象层留空，断点及位置为未知。托盘初始成立和积木初始未入盘已记录于节点证据；现有区间字段无法表达随后整段未知，故不以消失于视野的时刻伪造条件丢失。log step 48、65、67、70属于不同录制或复测，log step 75的“三块入盘、第四块在盘沿、朝向错误”是agent的总结，均不替代本次视频证据；总失败也不能推出某一个计数节点为false。四项原生误差未在reward中报告，值为null。",
+  "unmet": {
+   "tray_on_table": null,
+   "blocks_in_layout_1": null,
+   "blocks_in_layout_2": null,
+   "blocks_in_layout_3": null,
+   "blocks_in_layout_4": null
+  }
+ },
+ "summary_zh": "观察：评分器判定1720行交付失败；视频开局托盘在白桌上、四块积木散放，随后操作区离开头部视野，五个节点的终态均无法核实。推断：日志支持交付操作出错——agent在reset后的调试Session录制轨迹，新进程复测已出现不同结果，最终仍保留原文件。观察：视频直至 57.4 s 均未重新显示操作结果，无法定位具体物理断点或填写失败现象，不能把agent自述的三块入盘计作已完成。",
+ "conf": 4,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "视频确认：0–1 s 托盘与四块积木可见，约 2 s 起头部相机转离桌面，直到 57.4 s 均无操作区画面，故五个节点 unknown、断点未知、现象层留空并有 gap_zh 是诚实的；交付·交付操作出错有 log step 44/49/63/72 支持（reset 后录制、未在新进程验证即交付），标为推断恰当。未发现需要改动的内容。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:35:07Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-configuration-matching-block-design-i00-privileged--jsc5bka/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-configuration-matching-block-design-i00-privileged--jsc5bka/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "aafa46ea4a73a903c2469e55c25a0cc1194004332d12ac0418cfbf5549a5a626",
+   "video_sha256": "7cb13d2525d79d8db68385060987adf1566a1e86fdd1792bdd364a54885b8079",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/configuration_matching_block_design/trajectories/ispatialgym-configuration-matching-block-design-i00-privileged--uy8s8yq.json

@@ -0,0 +1,313 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-configuration-matching-block-design-i00-privileged--uy8s8yq",
+ "taskKey": "ispatialgym/configuration_matching_block_design",
+ "dag_sha256": "b80b8b64018446fb9084afd442651aea5f42a11bee539a14524ad1124ad51a8f",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 1261,
+   "sim_steps": 1261,
+   "rows": 1261,
+   "max_rows": 8298,
+   "replay_success": 0,
+   "instance_ok": 1,
+   "steps": 1261,
+   "success_step": 0,
+   "symbolic_success": 0,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 13.714,
+   "peak_memory_gb": 9.842,
+   "peak_processes": 22,
+   "peak_disk_gb": 1.409,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-configuration-matching-block-design-i00-privileged--uy8s8yq/replay.mp4",
+  "duration_s": 42.2,
+  "fps": 10.0,
+  "frames": 422,
+  "sha256": "dcdb49e64071572f322306bc38050437ce95f00a2a47cfbd0491d54f0d1562e4"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 5,
+  "base": 1,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "tray_on_table": {
+    "method": "visual",
+    "intervals": [
+     [
+      0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 木托盘开口向上位于白桌靠玻璃一侧，盘中为空。"
+     },
+     {
+      "t": 31.2,
+      "log_step": null,
+      "text_zh": "观察：视频 11.3–12.2 s 机器人在托盘中留下一块积木，托盘仍平放原桌面；视频 31.2 s 左下边缘再次可见托盘仍在桌上。"
+     },
+     {
+      "t": 42.1,
+      "log_step": null,
+      "text_zh": "推断：视频 12.4–42.1 s 机器人左夹爪只在其余三块桌面积木附近反复下降、抬起，没有搬动托盘的动作；结合视频 31.2 s 的再次显露，托盘留在桌上判为持续成立。"
+     }
+    ]
+   },
+   "blocks_in_layout_1": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 四块彩色积木均散放在白桌，木托盘为空。"
+     },
+     {
+      "t": 11.7,
+      "log_step": 108,
+      "text_zh": "观察：视频 9.2–10.8 s 机器人左夹爪夹起原先红顶的积木，旧位置变空，积木随夹爪移到托盘上方；视频 11.3–11.9 s 松开后该块留在盘中，但明显斜立，红顶及蓝黄斜分面相对托盘格线倾斜，并未平正占据图示格位。"
+     },
+     {
+      "t": 31.2,
+      "log_step": 112,
+      "text_zh": "观察：视频 12.0–12.2 s 盘中积木保持倾斜；视频 31.2 s 画面左下角仍可见该块边缘。log step 112 是 agent 脚本在会话中计算的单块误差（位置约0.008米、朝向约0.462弧度），仅与可见倾斜相互印证，不作为原生组布局评分。"
+     },
+     {
+      "t": 42.1,
+      "log_step": null,
+      "text_zh": "推断：视频 12.4–42.1 s 机器人未返回托盘纠正该块，另外三块也未被搬入；依据可见的错误朝向及后续动作，第1块完整格位与朝向判据始终未达成，不能只因进盘或顶面为红色记为完成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "视频 9.2–11.3 s 机器人左二指夹爪夹住红顶积木并搬运到托盘上方后张开释放；此前同一块的空夹重试未搬起积木。"
+     }
+    ]
+   },
+   "blocks_in_layout_2": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 13.4,
+      "log_step": null,
+      "text_zh": "观察：视频 13.0–13.8 s 机器人左夹爪对桌面后方蓝顶积木下降合拢后抬起，积木仍留在桌面；视频 14.8–16.4 s 又对前方蓝黄斜分面的积木做同类动作，没有搬入第二块。"
+     },
+     {
+      "t": 42.1,
+      "log_step": null,
+      "text_zh": "推断：视频 0.0–42.1 s 仅一块曾被搬到托盘，且其朝向不符；其余三块留在桌面，所以至少2块分别占据合格格位的条件从未成立。按可互换积木计数，不把本节点固定绑定某一积木身份。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "视频 13.0–13.8 s 机器人左二指夹爪下降包围桌面后方蓝顶积木并合拢尝试取块，随后抬离，该块仍在桌面；后��又反复尝试。作为进盘一块后补齐下一块的实际取块尝试记录，不表示布局计数已增加。"
+     }
+    ]
+   },
+   "blocks_in_layout_3": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 24.4,
+      "log_step": null,
+      "text_zh": "观察：视频 16.8–17.8 s 机器人左夹爪对桌面黄顶积木下降、合拢后离开；视频 22.4–24.8 s 又在前方及后方积木间重试，原来三块仍散放桌面，没有第三块入盘。"
+     },
+     {
+      "t": 42.1,
+      "log_step": null,
+      "text_zh": "推断：视频 0.0–42.1 s 仅一块曾被搬到托盘，且其朝向不符；其余三块留在桌面，所以至少3块分别占据合格格位的条件从未成立。按可互换积木计数，不把本节点固定绑定某一积木身份。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "视频 14.8–16.4 s 机器人左二指夹爪下降包围桌面前方蓝黄斜分面朝上的积木并合拢，抬起后积木仍在桌面；视频 22.4–23.6 s、25.4–26.8 s 及38.8–40.4 s 重复尝试。按后续待放积木的操作顺序记录接触，不固定可互换积木的合格槽位身份。"
+     }
+    ]
+   },
+   "blocks_in_layout_4": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 39.6,
+      "log_step": null,
+      "text_zh": "观察：视频 31.4–40.4 s 机器人左夹爪继续逐块下降、合拢和抬起；视频 41.2–42.1 s 三块积木仍分散留在白桌，托盘内从未形成四块二乘二布局。"
+     },
+     {
+      "t": 42.1,
+      "log_step": null,
+      "text_zh": "推断：视频 0.0–42.1 s 仅一块曾被搬到托盘，且其朝向不符；其余三块留在桌面，所以至少4块分别占据合格格位的条件从未成立。按可互换积木计数，不把本节点固定绑定某一积木身份。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "视频 16.8–17.8 s 机器人左二指夹爪下降包围桌面黄顶积木并合拢，随后未将其搬到托盘；视频 27.6–28.8 s 及33.8–35.0 s 又做同类尝试。记录补齐四块布局的取块动作，不表示第4块格位达成，也不固定实体身份。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "group_layout_position_error",
+   "unit": "m",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "group_layout_orientation_error",
+   "unit": "rad",
+   "threshold": 0.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "group_pose_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "group_pose_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "blocks_in_layout_1",
+   "t": 11.3,
+   "log_step": 108,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 11.3–11.9 s 机器人左夹爪张开并撤离托盘，所放积木留在格内但明显倾斜；视频 12.0–12.2 s 仍未摆正。视频 12.4–42.1 s 机器人转而反复取其余三块，既未修正该块，也未搬入其他合格积木；这是首个一直未修复的布局步骤。"
+  },
+  "unmet": {
+   "tray_on_table": true,
+   "blocks_in_layout_1": false,
+   "blocks_in_layout_2": false,
+   "blocks_in_layout_3": false,
+   "blocks_in_layout_4": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 11.3,
+    "log_step": 108,
+    "evidence_zh": "观察：视频 11.3–11.9 s 机器人释放的积木以明显倾斜的姿态停在托盘格内，红顶和蓝黄斜分面相对托盘格线歪斜，未获得图示的平正朝向；视频 12.0–12.2 s 仍保持此姿态。没有观察到一个先已平正达标、再倒下的阶段。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "t": 11.3,
+    "log_step": 108,
+    "evidence_zh": "推断：log step 108 的 agent 控制脚本把所持积木移到槽位上方，以设定高度0.862米张开夹爪让其下落，随后直接撤手并转向其他积木，没有落座后的朝向纠正；视频 10.6–11.9 s 与这段先悬在格位上方、再释放且留下倾斜积木的动作相符。该放置控制未把积木稳定送到目标朝向；log step 108、112 的脚本也读到朝向误差约0.462弧度。具体是释放姿态偏差还是下落中的接触造成倾斜，证据不足以细分。",
+    "source_zh": "日志"
+   }
+  ],
+  "gap_zh": "没有状态日志和成功示范；视频为单一头部视角，托盘在后半段多次出画，最终维持状态依据视频 31.2 s 的再显露及后续未返回托盘的动作推断。缺少原生槽位指派、逐步姿态误差与接触��，无法精确确定释放瞬间的姿态或是否碰到隔板，也不能验证静止窗口各速度阈值；因此根因只到控制层。评分器未报告DAG列出的四项原生误差，均填null，未用agent脚本的单块误差替代。log step 117–118 确认交付1261步；log step 121 的45度夹爪重试在log step 123–125仍未形成新保存结果，故不纳入视频动作判断。"
+ },
+ "summary_zh": "托盘从起始一直留在白桌上，最终满足1/5个节点；机器人在视频约11.3 s 将一块积木放入托盘，但姿态明显倾斜，完整的第1块布局步骤未达成，其余三块经反复抓取仍留在桌面。断点为第1块积木入盘，放置控制未获得目标朝向，之后没有修复；无状态日志且后段托盘部分出画，具体接触机制与精确误差不能由视频确定。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "节点值、断点（第1块入盘，视频约11.3 s 释放后倾斜留在盘中）、位姿不准现象和停在“控制”的根因均与视频及 log step 108/112/113 相符；未发现需改内容的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:35:46Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-configuration-matching-block-design-i00-privileged--uy8s8yq/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-configuration-matching-block-design-i00-privileged--uy8s8yq/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "92534896c672ae6b244b7577a2c2088ac5b4cb9fb7f436b4be5a256bb53c2213",
+   "video_sha256": "dcdb49e64071572f322306bc38050437ce95f00a2a47cfbd0491d54f0d1562e4",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/configuration_matching_block_design/trajectories/ispatialgym-configuration-matching-block-design-i00-privileged--y556sur.json

@@ -0,0 +1,302 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-configuration-matching-block-design-i00-privileged--y556sur",
+ "taskKey": "ispatialgym/configuration_matching_block_design",
+ "dag_sha256": "b80b8b64018446fb9084afd442651aea5f42a11bee539a14524ad1124ad51a8f",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 830,
+   "sim_steps": 830,
+   "rows": 830,
+   "max_rows": 8298,
+   "replay_success": 0,
+   "instance_ok": 1,
+   "steps": 830,
+   "success_step": 0,
+   "symbolic_success": 0,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 16.876,
+   "peak_memory_gb": 10.051,
+   "peak_processes": 22,
+   "peak_disk_gb": 3.995,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-configuration-matching-block-design-i00-privileged--y556sur/replay.mp4",
+  "duration_s": 27.8,
+  "fps": 10.0,
+  "frames": 278,
+  "sha256": "c5af3378b8497c44f61728732e71afff53d0e3db55efee83430aa66b011c6ae2"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 5,
+  "base": 1,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "tray_on_table": {
+    "method": "visual",
+    "intervals": [
+     [
+      0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 木托盘开口向上位于白桌靠玻璃门一侧；视频 15.8–18.6 s、22.3–27.7 s 托盘仍在同一桌面区域，没有翻倒或离桌。"
+     },
+     {
+      "t": 27.7,
+      "log_step": null,
+      "text_zh": "观察：视频 27.7 s 托盘仍承载盘内积木；本节点只计托盘保留关系，完整布局及45控制步共同静止属于联合成功验收。"
+     }
+    ]
+   },
+   "blocks_in_layout_1": {
+    "method": "visual",
+    "intervals": [
+     [
+      11.0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 15.8,
+      "log_step": null,
+      "text_zh": "观察：视频 15.8–16.6 s 托盘后排有两块积木；后左一块红顶，面向前方可见蓝黄侧面，与后左目标格位关系相符；后右一块却以蓝黄斜分面朝上。视频 18.6 s、24.3–27.7 s 后左红顶积木继续留在原格。"
+     },
+     {
+      "t": 11.0,
+      "log_step": 153,
+      "text_zh": "推断：后左有效积木来自视频 7.5–9.0 s 左夹爪夹起的红顶积木，原散放点随后空出；落盘处当时在画面左缘之外。log step 153 的最终交付脚本每次搬运185控制步，第二次开始松爪为第325控制步，按30赫兹约为视频 10.83 s，故取视频约11.0 s（误差约0.5 s）。log step 158 确认交付830行，与视频27.8 s相符；第三次松爪第510–530控制步对应视频17.0–17.67 s，已用视频17.0–18.0 s可见的放置核对。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "机器人左二指夹爪搬运积木；视频7.5–9.0 s可见夹爪合拢包夹红顶积木，积木随夹爪离开原位，随后进入托盘后左格。首轮搬运的夹取大部在画面外。"
+     }
+    ]
+   },
+   "blocks_in_layout_2": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 15.8,
+      "log_step": null,
+      "text_zh": "观察：视频 15.8–16.6 s 两块已放入后排的积木中，后右积木顶面呈蓝黄斜分，目标要求此格蓝顶，故两块入盘不等于两个有效格位。视频 18.6 s 新增红顶积木位于前左一侧，前右格仍空；视频 27.7 s 此布局未修复，另一蓝顶积木在盘外。"
+     },
+     {
+      "t": 0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 四块积木均散放桌面，托盘为空；视频 0.0–27.7 s 的顺序搬运没有形成两个位置和朝向同时正确的不同槽位。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "机器人左夹爪依次夹持并向托盘搬运第二、第三块积木，尝试增加有效积木数；视频13.8–15.8 s第三次搬运时积木被两指夹住，视频17.0–18.2 s松爪。"
+     }
+    ]
+   },
+   "blocks_in_layout_3": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 18.6,
+      "log_step": null,
+      "text_zh": "观察：视频 18.6 s 机器人左夹爪已退出，盘中三块积木是后左红顶、后右蓝黄斜分顶、前左红顶；后右朝向和前左顶色均���符合图示，前右格为空，未形成三个有效格位；视频 27.7 s 未修复。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "第三轮入盘尝试中，机器人左二指夹爪夹起桌面蓝黄斜分顶积木并搬入托盘；松爪后它红顶朝上且落在前左一侧，有效槽位数未增加。"
+     }
+    ]
+   },
+   "blocks_in_layout_4": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 23.5,
+      "log_step": null,
+      "text_zh": "观察：视频 20.0–22.9 s 左夹爪夹住最后一块并搬向托盘；视频 23.3–24.5 s 松爪后蓝顶积木留在托盘左外侧白桌上，前右槽仍空。视频 26.3–27.7 s 此结果稳定，四块有效布局从未形成。"
+     },
+     {
+      "t": 27.7,
+      "log_step": null,
+      "text_zh": "观察：评分器字段 replay_success=0、symbolic_success=0、success_step=0；这些字段只佐证联合目标未达，节点判假依据为视频中的盘外积木及错误朝向。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "视频20.0–22.9 s机器人左二指夹爪夹持最后一块积木搬向托盘，视频23.3–24.5 s打开并撤离；积木留在盘外，没有占据第四个有效槽位。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "group_layout_position_error",
+   "unit": "m",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "group_layout_orientation_error",
+   "unit": "rad",
+   "threshold": 0.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "group_pose_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "group_pose_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "blocks_in_layout_2",
+   "t": 11.0,
+   "log_step": 153,
+   "where": "操作中",
+   "evidence_zh": "推断：视频约11.0 s第二次搬运结束后，首次得到一个有效后左格，但未得到第二个有效格；视频15.8–16.6 s可见后右积木仍是蓝黄斜分顶而非蓝顶。该两块计数目标此后一直未修复：视频18.6 s第三次放置错在前左，视频24.3–27.7 s第四块又留在盘外。按可互换积木的有效计数，断点为第二块有效积木入盘；视频约11.0 s为交付脚本控制步推断的导航锚点，具体映射见节点证据。"
+  },
+  "unmet": {
+   "tray_on_table": true,
+   "blocks_in_layout_1": true,
+   "blocks_in_layout_2": false,
+   "blocks_in_layout_3": false,
+   "blocks_in_layout_4": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 11.0,
+    "log_step": null,
+    "evidence_zh": "观察：视频15.8–16.6 s后右积木顶面是蓝黄斜分，目标后右格要求蓝顶；视频18.6 s新增红顶积木位于应为蓝顶的前左区域，而前右空着；视频23.3–24.5 s最后一块在托盘外侧松开，视频27.7 s仍在盘外。机器人完成了抓取和搬运，但释放的积木位置或朝向不合目标，始终没有第二个有效槽位。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "source_zh": "日志 / 视频",
+    "evidence_zh": "推断：log step 153 的最终生成器把目标积木槽坐标直接作为left_tcp目标，所有积木都使用同一q_down夹爪姿态，没有以夹持后的积木相对夹爪变换补偿物体目标位姿，也没有根据落盘后实际位姿修正。与视频15.8–18.6 s错朝向、前排错位和视频23.3–24.5 s盘外释放相符，支持物体位姿控制不足；现有证据不能再区分接触设计、跟踪与逆解各自贡献。"
+   },
+   {
+    "cat": "认知",
+    "item": "误以为完成",
+    "source_zh": "日志 / 视频",
+    "evidence_zh": "观察：agent在log step 157查看其最终场景图，log step 161仍自述最终图“confirms all four blocks are seated in their respective slots inside the tray”，随后结束；log step 160记录尚余369.9墙钟秒。视频24.3–27.7 s却清楚显示一块在盘外、后右蓝黄斜分顶且前右空缺。推断：错误完成判断使agent未再纠正尚未满足的有效格位，维持了上述失败布局。"
+   }
+  ],
+  "gap_zh": "无状态日志、无成功示范，也无评分器逐槽指派、布局误差和静止窗口数值；四项原生误差均未报告，保留null。前两次落盘大部在相机左缘外，首次有效格位的约视频11.0 s由最终交付脚本控制步推断，约有0.5 s误差，后续视频15.8–16.6 s确认格位结果；视觉可判明显错误和后左关系，但不能精确复算0.03米/0.15弧度阈值。未把agent其他测试脚本的结果当作本次评分轨迹。"
+ },
+ "summary_zh": "机器人依��搬运四块积木，托盘始终留在白桌，后左一块达到图示格位，最终可确认2/5个节点。第二块有效积木入盘未完成：后右朝向错误，后续前排放置错位且最后一块在盘外；生成器缺乏物体位姿补偿，agent仍误认布局完成并停止。无状态日志及逐槽误差，首次有效放置时间由交付控制步推断，精确容差无法复算。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "节点值、断点（第二块有效积木）、现象与根因与视频一致；托盘全程在桌上，后左红顶有效，后右蓝黄斜分朝向错误，最后一块在盘外。未发现需改动的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:21:38Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-configuration-matching-block-design-i00-privileged--y556sur/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-configuration-matching-block-design-i00-privileged--y556sur/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "532a5afbc82b2dee5c766d839fa2b86162b1e9ab60a75f1b03fcc1268c0b54d0",
+   "video_sha256": "c5af3378b8497c44f61728732e71afff53d0e3db55efee83430aa66b011c6ae2",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/configuration_matching_block_design/trajectories/ispatialgym-configuration-matching-block-design-i00-privileged--zxck9lm.json

@@ -0,0 +1,228 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-configuration-matching-block-design-i00-privileged--zxck9lm",
+ "taskKey": "ispatialgym/configuration_matching_block_design",
+ "dag_sha256": "b80b8b64018446fb9084afd442651aea5f42a11bee539a14524ad1124ad51a8f",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 50,
+   "sim_steps": 50,
+   "rows": 50,
+   "max_rows": 8298,
+   "replay_success": 0,
+   "instance_ok": 1,
+   "steps": 50,
+   "success_step": 0,
+   "symbolic_success": 0,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 17.072,
+   "peak_memory_gb": 9.006,
+   "peak_processes": 21,
+   "peak_disk_gb": 0.637,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-configuration-matching-block-design-i00-privileged--zxck9lm/replay.mp4",
+  "duration_s": 1.8,
+  "fps": 10.0,
+  "frames": 18,
+  "sha256": "9c25454b92d5a254b616b5f4990f92e7365ec78b70cd6c32e85f85fd4daf93fb"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 5,
+  "base": 1,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "tray_on_table": {
+    "method": "visual",
+    "intervals": [
+     [
+      0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–1.7 s 全部18帧中，木色浅托盘开口向上留在白桌靠玻璃一侧，未翻倒、移离或落地；视频 1.7 s 末帧仍成立。"
+     }
+    ]
+   },
+   "blocks_in_layout_1": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 1.7,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–1.7 s 全部18帧中，四块彩色积木一直分散在白桌上、均在空托盘外；没有机器人夹爪进入工作区域，也没有积木离开原位或进入托盘，因而不满足至少1块占据图示目标格位。"
+     }
+    ]
+   },
+   "blocks_in_layout_2": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 1.7,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–1.7 s 全部18帧中，四块彩色积木一直分散在白桌上、均在空托盘外；没有机器人夹爪进入工作区域，也没有积木离开原位或进入托盘，因而不满足至少2块占据图示目标格位。"
+     }
+    ]
+   },
+   "blocks_in_layout_3": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 1.7,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–1.7 s 全部18帧中，四块彩色积木一直分散在白桌上、均在空托盘外；没有机器人夹爪进入工作区域，也没有积木离开原位或进入托盘，因而不满足至少3块占据图示目标格位。"
+     }
+    ]
+   },
+   "blocks_in_layout_4": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 1.7,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–1.7 s 全部18帧中，四块彩色积木一直分散在白桌上、均在空托盘外；没有机器人夹爪进入工作区域，也没有积木离开原位或进入托盘，因而不满足至少4块占据图示目标格位。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "group_layout_position_error",
+   "unit": "m",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "group_layout_orientation_error",
+   "unit": "rad",
+   "threshold": 0.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "group_pose_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "group_pose_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 151,
+   "where": "首步前",
+   "evidence_zh": "观察：log step 19 中 agent 的脚本将初始关节姿态重复50行保存为交付轨迹，评分器记录 rows=50、n_actions=50；视频 0.0–1.7 s 中四块积木始终在空托盘外。log step 147–148 仍在运行新的抓取测试，log step 151 随后报告 Gemini 过滤拦截；日志没有后续保存完整操作轨迹或验证交付完成的记录。推断：评分回放为早期保持姿态的占位轨迹，实际操作交付尚未完成就被外部过滤中���，不能定位一次已交付物理操作的失败断点。"
+  },
+  "unmet": {
+   "tray_on_table": true,
+   "blocks_in_layout_1": false,
+   "blocks_in_layout_2": false,
+   "blocks_in_layout_3": false,
+   "blocks_in_layout_4": false
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：log step 151 明确报告模型提供方过滤拦截，log step 150 的运行用时监测仍有549.9秒预算；log step 19 仅保存50行初始姿态，后续日志没有保存并验证完整交付的记录。推断：按外部中断规则和D37，此次不是已完成并验证交付后的中断，故不将占位回放未操作积木归为“没尝试”，也不从会话测试移植物理失败现象或根因。log step 138、144 的四积木测试及误差读数属于会话试验，与视频 0.0–1.7 s 的空托盘场景不一致，不能计入本次节点进度。输入无状态日志；评分器未报告四项原生位姿误差，故这些数值留空。"
+ },
+ "summary_zh": "观察：视频 0.0–1.7 s 中托盘始终留在白桌上，四块积木均散放盘外，最终仅托盘节点成立，进度1/5。推断：评分回放对应 log step 19 保存的50步初始姿态占位轨迹，后续会话抓取及摆放测试未成为本次交付。观察：log step 151 出现模型过滤中断，且无完整交付验证记录；按外部因素处理，断点未知，现象与根因留空。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "节点进度1/5（仅托盘在桌上）与视频一致；log step 19 确实保存了50行初始姿态，之后无完整交付保存，step 151 为模型过滤拦截，按外部因素处理（断点未知、现象根因为空并有gap_zh）合理。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:38:48Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-configuration-matching-block-design-i00-privileged--zxck9lm/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-configuration-matching-block-design-i00-privileged--zxck9lm/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "771d2197e227c6215f83a1e6e7cbfdcd7eda27ee398b3c7209d2412fd14fa616",
+   "video_sha256": "9c25454b92d5a254b616b5f4990f92e7365ec78b70cd6c32e85f85fd4daf93fb",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/configuration_matching_block_design/trajectories/ispatialgym-configuration-matching-block-design-i00-standard--tgch5hj.json

@@ -0,0 +1,341 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-configuration-matching-block-design-i00-standard--tgch5hj",
+ "taskKey": "ispatialgym/configuration_matching_block_design",
+ "dag_sha256": "b80b8b64018446fb9084afd442651aea5f42a11bee539a14524ad1124ad51a8f",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "live_matches_replay": 1,
+   "finalized": 1,
+   "missing_log": 0,
+   "n_actions": 2175,
+   "sim_steps": 2175,
+   "max_steps": 8298,
+   "steps": 1950,
+   "elapsed_s": 2809.5,
+   "symbolic_success": 0,
+   "success_step": 0,
+   "video_rendered": 1,
+   "gpu_lost": 0,
+   "replay_error": 0,
+   "instance_ok": 1,
+   "sim_startups": 1,
+   "resumes": 0,
+   "resume_state_mismatch": 0,
+   "resume_max_tcp_dev_m": 0.0,
+   "resume_max_joint_dev": 0.0,
+   "gave_up": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 0.974,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.886,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-configuration-matching-block-design-i00-standard--tgch5hj/replay.mp4",
+  "duration_s": 72.6,
+  "fps": 10.0,
+  "frames": 726,
+  "sha256": "0326ec8c5618931e6c02454bf39193ddc8c5747665255de71500ee10f2f9697a"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 5,
+  "base": 1,
+  "peak": 1,
+  "final": 0,
+  "nodes": {
+   "tray_on_table": {
+    "method": "visual",
+    "intervals": [
+     [
+      0,
+      66.3
+     ]
+    ],
+    "final": false,
+    "evidence": [
+     {
+      "t": 24.0,
+      "log_step": null,
+      "text_zh": "观察：视频24.0–25.0 s木托盘开口向上留在白桌靠玻璃的一侧，盘内为空；视频35.0–36.0 s及43.0–64.0 s仍在桌上。"
+     },
+     {
+      "t": 24.0,
+      "log_step": 54,
+      "text_zh": "推断：DAG记载托盘起态已在目标区域，结合log step 54仅开始操作红顶积木及视频24.0 s的托盘状态，将托盘在桌上的区间起点记为视频0 s；视频前段相机朝向室内，托盘不在画面中。"
+     },
+     {
+      "t": 66.3,
+      "log_step": null,
+      "text_zh": "观察：视频65.7–66.1 s左夹爪贴近托盘边缘，视频66.2 s撤臂时托盘被带向桌边，视频66.3 s已从桌面消失；视频68.7–72.5 s桌面全景仍无托盘。"
+     },
+     {
+      "t": 66.3,
+      "log_step": 213,
+      "text_zh": "推断：视频66.2–66.3 s托盘离开桌面且未恢复，结合agent在log step 213自述“托盘已掉下桌面”，判定托盘在桌上条件失去；落地位置不在相机视野内。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "推/扫",
+      "note_zh": "观察：视频65.7–66.3 s左夹爪在推动红顶积木及撤臂过程中碰动托盘边缘，将托盘带离桌面；这是意外碰触，未见夹持托盘或放回操作。"
+     }
+    ]
+   },
+   "blocks_in_layout_1": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 24.0,
+      "log_step": null,
+      "text_zh": "观察：视频23.5–25.0 s四块积木分散在白桌上，木托盘为空；视频35.0–36.0 s红顶积木已靠近托盘外沿，其余三块仍散放。"
+     },
+     {
+      "t": 50.4,
+      "log_step": null,
+      "text_zh": "观察：视频48.6–50.0 s左夹爪靠近并推动盘外红顶积木，视频50.2–50.4 s撤臂后红顶积木仍在盘外，托盘格位为空；视频55.0–58.5 s再次推动后亦未入格。"
+     },
+     {
+      "t": 66.3,
+      "log_step": null,
+      "text_zh": "观察：视频64.8–66.1 s红顶积木始终在托盘外侧，左夹爪向盘缘推进；视频66.3 s托盘消失后四块积木仍散放桌上，视频72.5 s终帧同样未形成入盘布局。"
+     },
+     {
+      "t": 35.0,
+      "log_step": 157,
+      "text_zh": "推断：log step 54、69、89、142、146记录的是针对同一红顶积木的接近、闭爪与试抬，log step 157称此前未抓住并改为推动；结合视频24.0 s、35.0 s和后续各次推动后的盘外位置，无证据支持曾达到一个有效目标格位。早期接触被头相机朝向遮去，不能由闭爪指令认定抓握成功。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "推/扫",
+      "note_zh": "观察：视频49.2–50.2 s、56.5–58.2 s及64.8–66.2 s左二指夹爪推动或拨动红顶积木及盘缘；未见两指夹住积木并将其提离桌面，不标抓握。"
+     }
+    ]
+   },
+   "blocks_in_layout_2": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 24.0,
+      "log_step": null,
+      "text_zh": "观察：视频24.0–25.0 s托盘为空，四块积木分散在盘外；视频35.0–36.0 s及43.0–66.1 s机器人只在红顶积木附近操作，其余三块仍在桌面原区域，未出现至少2块占据不同目标格位的布局。"
+     },
+     {
+      "t": 72.5,
+      "log_step": null,
+      "text_zh": "观察：视频66.3 s托盘离开桌面，视频68.7–72.5 s全景中四块积木均散放桌上，无至少2块入盘的结果。"
+     }
+    ]
+   },
+   "blocks_in_layout_3": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 24.0,
+      "log_step": null,
+      "text_zh": "观察：视频24.0–25.0 s托盘为空，四块积木分散在盘外；视频35.0–36.0 s及43.0–66.1 s机器人只在红顶积木附近操作，其余三块仍在桌面原区域，未出现至少3块占据不同目标格位的布局。"
+     },
+     {
+      "t": 72.5,
+      "log_step": null,
+      "text_zh": "观察：视频66.3 s托盘离开桌面，视频68.7–72.5 s全景中四块积木均散放桌上，无至少3块入盘的结果。"
+     }
+    ]
+   },
+   "blocks_in_layout_4": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 24.0,
+      "log_step": null,
+      "text_zh": "观察：视频24.0–25.0 s托盘为空，四块积木分散在盘外；视频35.0–36.0 s及43.0–66.1 s机器人只在红顶积木附近操作，其余三块仍在桌面原区域，未出现至少4块占据不同目标格位的布局。"
+     },
+     {
+      "t": 72.5,
+      "log_step": null,
+      "text_zh": "观察：视频66.3 s托盘离开桌面，视频68.7–72.5 s全景中四块积木均散放桌上，无至少4块入盘的结果。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "group_layout_position_error",
+   "unit": "m",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "group_layout_orientation_error",
+   "unit": "rad",
+   "threshold": 0.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "group_pose_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "group_pose_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "blocks_in_layout_1",
+   "t": 50.2,
+   "log_step": 157,
+   "where": "操作中",
+   "evidence_zh": "观察：视频49.2–50.4 s机器人左夹爪推动红顶积木后撤离，积木仍留在托盘外，第一块入格始终未完成；视频64.8–66.3 s继续处理同一积木时又碰动托盘并使其离开桌面。log step 157明确从失败的抓取改为将该积木推入格位，log step 213承认托盘落下后停止；该步骤之后未获修复。"
+  },
+  "unmet": {
+   "tray_on_table": false,
+   "blocks_in_layout_1": false,
+   "blocks_in_layout_2": false,
+   "blocks_in_layout_3": false,
+   "blocks_in_layout_4": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 50.4,
+    "log_step": null,
+    "evidence_zh": "观察：视频49.2–50.4 s左夹爪对红顶积木作入盘方向的推动，但撤臂后积木仍在盘外且与目标格位分离；视频58.5 s及66.3 s仍未入盘，没有形成第一个有效格位。"
+   },
+   {
+    "cat": "运动",
+    "item": "被挡/碰撞",
+    "t": 66.2,
+    "log_step": null,
+    "evidence_zh": "观察：视频65.7–66.1 s左夹爪进入托盘边缘区域，视频66.2 s撤臂时托盘随夹爪向桌边移动，视频66.3 s托盘离开桌面；被碰坏的是原先成立的托盘在桌上状态。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "抓取和接触设计",
+    "alt_feasible": "未知",
+    "source_zh": "视频 / 日志",
+    "evidence_zh": "推断：视频49.2–50.4 s及64.8–66.3 s显示左夹爪采用低位推动，积木未越过盘缘，夹爪却碰动托盘并在撤离时把托盘带离桌面；这支持本次积木—夹爪—盘缘的接触几何不适配。log step 157明确选择从抓取改为推动，log step 201以闭爪命令和同一高度的两端关节插值完成推动，log step 203直接返回初始臂姿；没有已验证成功的替代抓法，故alt_feasible为未知。"
+   }
+  ],
+  "gap_zh": "视频前段及视频25.5–34.0 s、36.5–42.0 s头相机朝向室内，无法直接确认早期每次闭爪的接触细节；没有状态日志、接触力读数或成功示范，不能量化抓取偏差及原生布局误差。托盘落地位置在画外，但视频66.2–66.3 s的离桌过程和视频72.5 s桌面全景可判定其未恢复。四项原生误差评分器均未报告，native_metrics保留null。"
+ },
+ "summary_zh": "托盘在桌上的初始条件成立，但没有积木达到盘内有效格位，进度峰值为1/5，终态为0/5。机器人反复尝试处理同一红顶积木，低位推动未将其送入格位，并在视频66.2–66.3 s撤臂时碰动托盘使其离开桌面，断点为第1块积木入盘；接触几何不适配有视频和指令记录支持。早期抓取被头相机朝向遮去，且无状态日志或原生误差数值，不能确定更细的抓取失败原因。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 2,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "fix",
+    "conf": 7,
+    "summary_zh": "断点（第1块入盘未达成）、托盘在视频66.2–66.3 s离桌、根因（抓取和接触设计）与现有帧相符；但托盘节点及blocks_in_layout_1/2/3/4中“视频24.0 s、35.0 s托盘在桌上、盘内为空”的观察在这些帧里并不成立（托盘不在画面内，约44 s后才出现），应改为推断；底盘“支撑”接触取自撤离段，与该节点的操作无关。",
+    "n_issues": 2,
+    "n_major": 0
+   },
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 2,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "断点（第1块入盘未完成）、托盘在视频66.2–66.3 s离桌、根因（抓取和接触设计）与帧相符。第一轮审查所称24 s/35 s无托盘不成立：视频24.0 s帧左下角可见木托盘且盘内为空，因此不再提出该点。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:48:05Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-configuration-matching-block-design-i00-standard--tgch5hj/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-configuration-matching-block-design-i00-standard--tgch5hj/04_review_r1"
+   },
+   {
+    "stage": "reconcile",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-configuration-matching-block-design-i00-standard--tgch5hj/05_reconcile_r1"
+   },
+   {
+    "stage": "traj_review",
+    "round": 2,
+    "ok": true,
+    "space": "work/runs/ispatialgym-configuration-matching-block-design-i00-standard--tgch5hj/04_review_r2"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "7cdc3aa0aa7416d3f04dd6d8291f21abb1a38ed47b34716fb02972ed3608e4a8",
+   "video_sha256": "0326ec8c5618931e6c02454bf39193ddc8c5747665255de71500ee10f2f9697a",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/inter_object_relation_satisfaction_nextto_small_scale/trajectories/ispatialgym-inter-object-relation-satisfaction-nextto-small-scale-i00-privileged--5u8sytn.json

@@ -0,0 +1,248 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-inter-object-relation-satisfaction-nextto-small-scale-i00-privileged--5u8sytn",
+ "taskKey": "ispatialgym/inter_object_relation_satisfaction_nextto_small_scale",
+ "dag_sha256": "a1913428fea25766fa07bf571634053c1cc8aa302d6a8d8b88da8834e534f9f8",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 70,
+   "sim_steps": 70,
+   "rows": 70,
+   "max_rows": 2076,
+   "replay_success": 0,
+   "instance_ok": 1,
+   "steps": 70,
+   "success_step": 0,
+   "symbolic_success": 0,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 1,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.0,
+   "peak_processes": 0.0,
+   "peak_disk_gb": 0.0,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-inter-object-relation-satisfaction-nextto-small-scale-i00-privileged--5u8sytn/replay.mp4",
+  "duration_s": 2.5,
+  "fps": 10.0,
+  "frames": 25,
+  "sha256": "62689c0800e905eb329bebd8bb01f133c6a9d4d32f1bab6c66a82cb82ed60a01"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 1,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "pencil_at_goal": {
+    "method": "visual",
+    "intervals": [
+     [
+      0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：评分视频 0.0–2.4 s 的全部25帧中，黄色彩铅始终平放在白色台面同一位置，朝向未见变化；无夹爪进入其附近。结合DAG确认彩铅初态即合格，判为从视频 0.0 s 保持至结束。"
+     }
+    ]
+   },
+   "marker_nextto_pencil": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.4,
+      "log_step": 157,
+      "text_zh": "观察：评分视频 0.0–2.4 s 的全部25帧中，红帽黑身记号笔一直留在黄色彩铅右侧原处，两笔间隙未缩小，无抓取、推移或搬运；成功参考视频 12.0–17.5 s 显示的紧邻并排外观在本次评分视频中始终未出现。log step 154、157 保存并复核的是70行相同的初态保持动作，评分器 n_actions=70、symbolic_success=0，与视频一致。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "pencil_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "pencil_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "marker_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "marker_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_joint_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "marker_nextto_pencil",
+   "t": 0.0,
+   "log_step": 154,
+   "where": "首步前",
+   "evidence_zh": "观察：评分视频 0.0–2.4 s 中机器人没有开始靠近、夹取或推移记号笔，两笔始终分离于初始位置。log step 154 将交付文件改为70行初态保持动作，log step 157 重生成并确认所有行相同；会话中的抓取、推移调试未交付，因此断在记号笔邻彩铅这一步启动之前。"
+  },
+  "unmet": {
+   "pencil_at_goal": true,
+   "marker_nextto_pencil": false
+  },
+  "phenomena": [
+   {
+    "cat": "没尝试",
+    "item": "没尝试",
+    "t": 0.0,
+    "log_step": 154,
+    "evidence_zh": "观察：评分视频 0.0–2.4 s 全程未见夹爪接近两笔或两笔移动，末帧仍保持初始间隔；log step 154、157 明确提交并复核70���相同的静止动作，未包含会话中的摆放尝试。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "log_step": 146,
+    "source_zh": "日志",
+    "evidence_zh": "推断：会话内的操作控制未能形成可交付的靠近两笔动作，促使agent退回静止基线。log step 130 的脚本输出显示夹爪闭合、抬升后彩铅高度仍约0.810 m；log step 134、138、140 的多轮推移输出中两笔横向间距仍约0.090 m；log step 146 记录逆解返回成功，但执行后左末端距所设目标的x、z误差仍为0.191 m、0.234 m。agent 在 log step 156、161 明确称这些会话操作失败使其选择保留初态，随后 log step 154、157 的保存与复核记录确认最终仅交付静止动作。上述是会话证据，只用于解释最终选择，不作为评分视频中的接触或进度。",
+    "note_zh": "缺少会话操作的视频和接触记录，无法区分具体接触布局、轨迹执行或反馈问题，根因停在控制；agent所称桌面阻挡前臂未独立核实。"
+   }
+  ],
+  "gap_zh": "无状态日志，评分器未报告DAG列出的10项连续指标，不能从视频精确量出位置、姿态误差或联合静置速度，native_metrics 均为null。评分视频完整且两笔可见，足以判断保持初态与未启动摆放；控制根因来自会话脚本输出及agent交付说明，会话操作本身没有随附视频，具体阻挡接触及控制子类型未核实。"
+ },
+ "summary_zh": "评分视频 0.0–2.4 s 全程保持初态：彩铅留在原位成立，记号笔邻彩铅始终未达成，最终为1/2节点。断点为记号笔邻彩铅／首步前，现象为没尝试；会话日志显示多次抓取、推移控制未奏效，agent遂在 log step 154、157 改交并复核静止基线，这些调试动作不计入评分轨迹。无状态日志和连续误差指标，具体控制失效机制无法核实。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "评分视频仅2.4 s，两笔始终原位，彩铅成立、记号笔邻彩铅未达成，断点首步前/没尝试，根因控制（来自会话证据）均与视频、log和CASES FAIL-22一致；未发现需要修改的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:39:13Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-nextto-small-scale-i00-privileged--5u8sytn/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-nextto-small-scale-i00-privileged--5u8sytn/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "9a5fd2160abc49e0b3e244f25a9c43bd29fa3ab61421fd834d37614eff005bd5",
+   "video_sha256": "62689c0800e905eb329bebd8bb01f133c6a9d4d32f1bab6c66a82cb82ed60a01",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/inter_object_relation_satisfaction_nextto_small_scale/trajectories/ispatialgym-inter-object-relation-satisfaction-nextto-small-scale-i00-privileged--cdnazq4.json

@@ -0,0 +1,268 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-inter-object-relation-satisfaction-nextto-small-scale-i00-privileged--cdnazq4",
+ "taskKey": "ispatialgym/inter_object_relation_satisfaction_nextto_small_scale",
+ "dag_sha256": "a1913428fea25766fa07bf571634053c1cc8aa302d6a8d8b88da8834e534f9f8",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 768,
+   "sim_steps": 768,
+   "rows": 768,
+   "max_rows": 2076,
+   "replay_success": 0,
+   "instance_ok": 1,
+   "steps": 768,
+   "success_step": 0,
+   "symbolic_success": 1,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 16.888,
+   "peak_memory_gb": 8.841,
+   "peak_processes": 22,
+   "peak_disk_gb": 1.409,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-inter-object-relation-satisfaction-nextto-small-scale-i00-privileged--cdnazq4/replay.mp4",
+  "duration_s": 25.7,
+  "fps": 10.0,
+  "frames": 257,
+  "sha256": "6e3f69a560c353c1c80261216518b106c5850bc51a438c18779a2899b5d89df8"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 1,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "pencil_at_goal": {
+    "method": "visual",
+    "intervals": [
+     [
+      0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–3.2 s，黄色彩铅平放白色台面，长轴朝向与成功参照的初态相符。"
+     },
+     {
+      "t": 25.6,
+      "log_step": null,
+      "text_zh": "推断：视频 19.7–25.6 s，彩铅仍平放台面、长轴方向基本保持，只向记号笔靠近了小段距离；按DAG的宽位置容差与绕长轴等价朝向判为保留。视频 3.5–19.5 s 两物离开画面，未观察到移出容差后再恢复的过程，故不凭松手或相机转动切断区间；该区间为前后帧支持的估计，不能量出完整三维位姿误差。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：视频 3.5–19.5 s 彩铅和机器人作用端均不在头部相机视野内；前后帧确认彩铅被移近记号笔。log step 98–109 的脚本及叙述指向右夹爪侧推，但无该次重放的接触对或腕部视频，不能直接确认指面和接触方式。"
+     }
+    ]
+   },
+   "marker_nextto_pencil": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–3.2 s，红帽黑身记号笔与黄色彩铅明显分离，尚未呈目标中的紧邻排列。"
+     },
+     {
+      "t": 20,
+      "log_step": null,
+      "text_zh": "观察：视频 19.7–25.6 s，两物已靠近，但记号笔红帽向画面右侧、黑身下端向左侧倾斜，彩铅仍近乎竖直；与视频 0.0 s 记号笔反方向的倾斜、以及成功参照视频 16.0–17.5 s 两笔近乎平行的外观不同。"
+     },
+     {
+      "t": 25.6,
+      "log_step": null,
+      "text_zh": "推断：视频 24.6–25.6 s 逐帧检查中斜置结果一直保留，未见后续摆正；结合评分器 reward.symbolic_success=1、replay_success=0、deterministic=1，判为已实现近邻符号关系，但未实现本节点要求的目标位姿，final=false。无逐物误差字段，不能把符号关系成立等同于此复合节点达标；视野外是否短暂达标无法确认，不增记达成区间。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：视频 0.0 s 与视频 19.7–25.6 s 的物体间距及记号笔朝向发生变化，说明机器人确实工作于此关系；操作段不在相机视野内，不能判定右指直接碰记号笔还是由彩铅间接带动，也不能确认抓握。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "pencil_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "pencil_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "marker_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "marker_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_joint_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "marker_nextto_pencil",
+   "t": 19.7,
+   "log_step": null,
+   "where": "操作中",
+   "evidence_zh": "推断：机器人为建立两笔近邻关系而工作，视频 19.7 s 两物重新进入画面时已靠近，但记号笔转斜，直到视频 25.6 s 结束也未摆正。将此节点作为未修复断点；视频 19.7 s 是首次重见结果的导航锚点，实际转歪发生在视频约 3.5–19.5 s 的视野外，不能精确定位。"
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 19.7,
+    "log_step": null,
+    "evidence_zh": "观察：视频 19.7–25.6 s 记号笔红帽朝右、笔身下端朝左，相对黄色彩铅呈可见夹角；视频 0.0 s 的记号笔倾向相反，成功参照视频 16.0–17.5 s 则两笔近乎平行。两物距离缩短后留下了错误朝向，末尾没有修正。"
+   }
+  ],
+  "root_causes": [],
+  "gap_zh": "操作接触段在视频约 3.5–19.5 s 完全离开头部视野，故两个实际接触标签填未知；无状态日志或逐物位姿、关节速度、45步联合静置检查值。记号笔朝向失败据前后帧和参照推断，无法精确测量0.3 rad容差，亦不能排除额外的静置失败。log step 105、107 算出的彩铅旋转1.112 rad来自会话检查，未应用DAG的绕长轴等价朝向，不能据此宣称彩铅节点失败。log step 130、133 表明最终768行含未重放验证的追加尾段，但没有证据证明追加导致此次失败；侧推如何造成记号笔偏转及其具体控制根因不可确定，因此根因留空。",
+  "unmet": {
+   "pencil_at_goal": true,
+   "marker_nextto_pencil": false
+  }
+ },
+ "summary_zh": "机器人实际把黄色彩铅移向记号笔，最终近邻符号关系成立；彩铅仍平放并大致保持目标方向，按可视判据保留该节点。记号笔在两物重新入镜后呈斜置且未修正，记号笔邻彩铅节点未达到完整位姿要求，评分器重放失败。操作接触过程不在视野内，无逐物误差和静置记录，不能确定具体控制根因。",
+ "conf": 6,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "标签与视频一致：视频 0.0 s 与 25.6 s 帧对比，彩铅基本不动、记号笔由左倾变为右倾斜置；3.5–19.5 s 操作段确在视野外（视频 3.0、10.0 s 帧为转头），未知接触、断点锚点、位姿不准、根因留空并带 gap_zh 都有依据。无需修改。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:29:39Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-nextto-small-scale-i00-privileged--cdnazq4/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-nextto-small-scale-i00-privileged--cdnazq4/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "538f1fc879bdab1f252f000fb465c8a7be9539142fd94da7c1a6294a2b6176a1",
+   "video_sha256": "6e3f69a560c353c1c80261216518b106c5850bc51a438c18779a2899b5d89df8",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/inter_object_relation_satisfaction_nextto_small_scale/trajectories/ispatialgym-inter-object-relation-satisfaction-nextto-small-scale-i00-privileged--efq9r73.json

@@ -0,0 +1,275 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-inter-object-relation-satisfaction-nextto-small-scale-i00-privileged--efq9r73",
+ "taskKey": "ispatialgym/inter_object_relation_satisfaction_nextto_small_scale",
+ "dag_sha256": "a1913428fea25766fa07bf571634053c1cc8aa302d6a8d8b88da8834e534f9f8",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 285,
+   "sim_steps": 285,
+   "rows": 285,
+   "max_rows": 2076,
+   "replay_success": 0,
+   "instance_ok": 1,
+   "steps": 285,
+   "success_step": 0,
+   "symbolic_success": 1,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 16.871,
+   "peak_memory_gb": 10.261,
+   "peak_processes": 21,
+   "peak_disk_gb": 0.667,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-inter-object-relation-satisfaction-nextto-small-scale-i00-privileged--efq9r73/replay.mp4",
+  "duration_s": 9.6,
+  "fps": 10.0,
+  "frames": 96,
+  "sha256": "9b67f5012f5d357113a3a4f89d1d58b06fe512629e0c6822b9be7621c260bef6"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 1,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "pencil_at_goal": {
+    "method": "visual",
+    "intervals": [
+     [
+      0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 黄色彩铅已平放在白色台面上；视频 0.0–6.9 s 机器人右夹爪接近并推移的是红帽记号笔，彩铅未被夹起。"
+     },
+     {
+      "t": 9.5,
+      "log_step": null,
+      "text_zh": "观察：视频 7.2–9.5 s 彩铅仍平放在原来所在的台面区域，未见明显平移或转歪；结合初态即合格，按可见位姿判据记为持续保留。相机随机器人躯干运动，图像中的位置变化不当作彩铅世界位移。"
+     }
+    ]
+   },
+   "marker_nextto_pencil": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 4.5,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 两笔明显分离；视频 4.5–6.9 s 机器人以张开的右夹爪从记号笔侧面推移，没有夹住笔身并搬起。"
+     },
+     {
+      "t": 7.1,
+      "log_step": null,
+      "text_zh": "观察：视频 7.0–7.3 s 机器人右夹爪上撤时，记号笔迅速横移并转动，越到彩铅的另一侧；视频 7.3–9.5 s 两笔虽已靠近，但记号笔黑身与黄色彩铅呈张开的斜向排列。参考成功视频 16.0–17.5 s 的两笔贴近、近乎平行外观，本次未见对应的摆正结果。"
+     },
+     {
+      "t": 9.5,
+      "log_step": 367,
+      "text_zh": "推断：按视频中撤手后的朝向偏差，记号笔的完整目标摆放未确认达到，判本节点未完成；不是仅据零分否定相邻关系。评分器 reward 的 symbolic_success=1、replay_success=0、success=0，说明符号检查的正值不能替代本节点的关系与位姿合取验收；log step 367 仅检查动作数组末尾相同，未提供目标姿态误差。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "推/扫",
+      "note_zh": "观察：视频约 4.5–7.1 s，机器人张开的右二指夹爪从侧面抵推红帽黑身记号笔，随后上撤；未见相向夹紧形成抓握。"
+     },
+     {
+      "who": "底盘",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.0–1.0 s 机器人底盘前移，带机体靠近桌上的记号笔，为右臂推移作准备；log step 368 的交付轨迹分段记录也包含起始30步前进。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "pencil_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "pencil_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "marker_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "marker_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_joint_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "marker_nextto_pencil",
+   "t": 7.1,
+   "log_step": 359,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 7.0–7.3 s，机器人在推移记号笔后撤起右夹爪，记号笔随即横移转斜；视频 7.3–9.5 s 未再修正其朝向。log step 359 生成并保存最后一次推移、撤手和保持轨迹，log step 367–371 确认并重新生成285行动作文件；本断点按交付视频中的这次摆放偏差定位。"
+  },
+  "unmet": {
+   "pencil_at_goal": true,
+   "marker_nextto_pencil": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 7.1,
+    "log_step": null,
+    "evidence_zh": "观察：视频 7.1–7.3 s 机器人撤手时记号笔发生转动，视频 7.3–9.5 s 红帽黑身记号笔与彩铅形成斜向张开的排列，未呈参考成功视频 16.0–17.5 s 的近乎平行摆放；两笔靠近后未再被摆正。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "认知",
+    "item": "误以为完成",
+    "t": null,
+    "log_step": 377,
+    "source_zh": "日志 / agent 自述",
+    "evidence_zh": "推断：agent 在 log step 367 以末45行动作完全相同检查静置，在 log step 370 将交付说明写为两笔平行且精确接触，log step 377 又自述目标已在仿真中执行并验证，随后结束；log step 372 尚余839.9 s预算。该完成信念使其停止修正，而交付视频 7.3–9.5 s 仍有笔身斜向偏差，最终评分 success=0。动作命令相同不等于物体完整位姿与实际速度已通过验收。"
+   }
+  ],
+  "gap_zh": "观察：inputs/ 只有运行记录、视频及成功参考，没有本次状态日志；评分记录未报告逐物体位置/完整四元数姿态误差或45步联合静置窗口各项实测值。视频能看出推移后转斜，但无法精确区分记号笔绕长轴转动、平面朝向误差及微小残余速度各自对判负的贡献，也不能确认撤手瞬间是否曾短暂满足全部位姿容差；节点时间线按可见摆放判据保守标注，未把会话中其他重置试验的位姿数值当作交付事实。"
+ },
+ "summary_zh": "黄色彩铅从视频 0.0 s 到末帧保持平放，机器人前移后用右夹爪推近记号笔；视频约 7.1 s 撤手时记号笔转斜，最终两笔靠近但未形成参考成功轨迹中的平行摆放，按可见判据完成1/2节点。agent 在 log step 377 自述已验证完成并停止，评分器确定性重放仍失败；缺少完整姿态与联合静置实测数据，不能进一步锁定具体越限分量。",
+ "conf": 6,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "节点值、断点与原因与视频一致：视频 7.0–7.3 s 撤手时记号笔转斜，末帧两笔呈明显 V 形斜向排列，而参考成功视频末帧近乎平行；彩铅保持，1/2 合理。symbolic_success=1 与 replay_success=0 并存，已在推断中如实说明，无需改动。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:33:48Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-nextto-small-scale-i00-privileged--efq9r73/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-nextto-small-scale-i00-privileged--efq9r73/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "37aaee3c74c292b3a2f5e71c8e10705d250a8394356c54240d4b9ba16c0eac1a",
+   "video_sha256": "9b67f5012f5d357113a3a4f89d1d58b06fe512629e0c6822b9be7621c260bef6",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/inter_object_relation_satisfaction_nextto_small_scale/trajectories/ispatialgym-inter-object-relation-satisfaction-nextto-small-scale-i00-privileged--f8fun8m.json

@@ -0,0 +1,295 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-inter-object-relation-satisfaction-nextto-small-scale-i00-privileged--f8fun8m",
+ "taskKey": "ispatialgym/inter_object_relation_satisfaction_nextto_small_scale",
+ "dag_sha256": "a1913428fea25766fa07bf571634053c1cc8aa302d6a8d8b88da8834e534f9f8",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 244,
+   "sim_steps": 244,
+   "rows": 244,
+   "max_rows": 2076,
+   "replay_success": 0,
+   "instance_ok": 1,
+   "steps": 244,
+   "success_step": 0,
+   "symbolic_success": 0,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 20.309,
+   "peak_memory_gb": 25.16,
+   "peak_processes": 77,
+   "peak_disk_gb": 1.391,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-inter-object-relation-satisfaction-nextto-small-scale-i00-privileged--f8fun8m/replay.mp4",
+  "duration_s": 8.3,
+  "fps": 10.0,
+  "frames": 83,
+  "sha256": "a8e8e4000fb264b1d82fca45a05714cac16eb367beb55726500150b5776514be"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 1,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "pencil_at_goal": {
+    "method": "visual",
+    "intervals": [
+     [
+      0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 黄色彩铅平放在白色台面上；逐帧检查视频 0.0–8.2 s，彩铅始终在记号笔左侧，未见机器人把它夹起、推走或明显转歪。"
+     },
+     {
+      "t": 8.2,
+      "log_step": null,
+      "text_zh": "推断：视频 2.3 s、4.8 s 和 8.2 s 中彩铅相对桌面的位姿外观与初态一致，支持初态已合格的彩铅目标状态持续保持；视角随机器人躯干改变，不把画面中的整体投影移动当作彩铅位移。"
+     }
+    ]
+   },
+   "marker_nextto_pencil": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 1.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 及 1.0–2.3 s，红帽黑身记号笔与黄色彩铅明显分离；机器人右夹爪从右侧接近，随后张开手指横向扫向记号笔。"
+     },
+     {
+      "t": 4.8,
+      "log_step": null,
+      "text_zh": "观察：逐帧检查视频 2.4–4.6 s，右夹爪遮住部分记号笔；视频 4.8–5.4 s 重新露出的黑色笔杆仍与彩铅隔开，没有显示搬到彩铅紧邻处的结果。视频 5.5–8.2 s，包括最后一秒的每帧，笔帽及露出笔杆仍处在彩铅右侧的分离位置。"
+     },
+     {
+      "t": 8.2,
+      "log_step": null,
+      "text_zh": "推断：本次视频 8.2 s 两笔间隔仍明显，区别于参考成功视频 16.0–17.5 s 中记号笔紧挨彩铅的外观，故记号笔相邻目标未达到；评分器 reward.symbolic_success=0、replay_success=0 与此相符，并非仅凭零分判未达。"
+     },
+     {
+      "t": null,
+      "log_step": 134,
+      "text_zh": "观察：log step 134 最终生成文件为 244 行，校验打印 match executed: False；log step 141 所述推移距离及偏航取自会话测试，不能替代本次评分视频中的物体移动证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：视频 1.0–4.6 s 机器人右侧张开的二指夹爪接近并扫过记号笔所在区域，未见夹紧或抬起；关键笔身被手指遮挡，单一头部视角不能确认实际接触点及施力方式。已检查接近前和扫过后的连续帧，未把会话中自述的单指推移直接当作本次实际接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "pencil_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "pencil_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "marker_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "marker_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_joint_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "marker_nextto_pencil",
+   "t": 4.8,
+   "log_step": null,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 2.4–4.6 s 机器人右夹爪向两笔方向扫动，视频约 4.8 s 改变姿态后露出的记号笔仍与彩铅分离；视频 5.5–8.2 s 机器人停留，未再把记号笔移到紧邻位置。推断：失败发生在记号笔摆放步骤，后续静候没有修复，断点以扫动结束附近的视频 4.8 s 作导航锚点。"
+  },
+  "unmet": {
+   "pencil_at_goal": true,
+   "marker_nextto_pencil": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 4.8,
+    "log_step": null,
+    "evidence_zh": "观察：视频 4.8–5.4 s 可见记号笔露出夹爪下方的笔杆，至视频 8.2 s 仍与黄色彩铅保持明显横向间隔，没有形成参考成功视频 16.0–17.5 s 可见的紧邻并排结果。推断：此次摆放的位置未到目标；视频不支持把会话报告的偏航角当作此次失败现象。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "交付",
+    "item": "交付操作出错",
+    "source_zh": "日志",
+    "evidence_zh": "推断：最终交付重新拼装了不等同于会话执行的动作。log step 123 保存 204 行实验动作及 sols10；log step 124–125 的代码继续覆盖同一 sols10 文件。log step 128 生成脚本却把旧 final_rows10 的末行用作 HOLD，与已覆盖的 IK 解组合；log step 134 再生成 244 行后明确打印 match executed: False，仍保留该文件交付。log step 139–140 的新进程输出只确认执行 244 行及初态匹配，没有任务成功字段；评分视频 4.8–8.2 s 也未出现会话测试自述的到位推移。这支持动作保存与重建过程出错，而非评分器重放故障。"
+   }
+  ],
+  "gap_zh": "无状态日志，单一头部视角在视频 2.4–4.6 s 遮住关键笔身，不能确认右手是否真正碰到记号笔、具体接触方式或微小位移，故该节点实际接触记为未知，不臆标未接触或抓握。评分器未报告 DAG 所列逐物位姿误差及静置速度、漂移指标，全部原生指标值保留 null；不能从视频量出毫米级静止条件。会话测试中的物体位姿不能移用于最终重新生成的交付文件。"
+ },
+ "summary_zh": "评分轨迹中彩铅持续保持初始目标位姿，最终为 1/2 节点；机器人右夹爪接近并扫过记号笔区域，但扫动后两笔仍明显分离，记号笔邻彩铅步骤未完成。日志显示最终动作由不同实验的数据重新组合，生成后仍打印 match executed: False，支持交付操作出错；会话自述的推移和偏航不作为本次视频事实。没有状态日志，夹爪遮挡使实际接触方式及精确静置指标无法确认。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 2,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "fix",
+    "conf": 6,
+    "summary_zh": "视频末帧（8.2 s）中记号笔笔帽与露出的笔杆已紧靠彩铅右侧、夹爪压在其上，外观接近参考成功视频 16.5–17 s，并非标注所称的“明显分离”；评分器仅给整体 symbolic_success=0，无法据此判 marker 节点未达或判“位置未到目标”，现有 final=false 与位姿不准现象证据不成立。",
+    "n_issues": 2,
+    "n_major": 2
+   },
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 2,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "独立复查末帧：8.2 s 彩铅与红色笔帽仍相隔约一支笔长的一半，与 0.0 s 初态间隔相近，明显大于参考视频 17 s 的紧邻外观；标注 final=false、位姿不准、交付根因与第 1 轮回应一致，未发现需改动之处。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:50:54Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-nextto-small-scale-i00-privileged--f8fun8m/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-nextto-small-scale-i00-privileged--f8fun8m/04_review_r1"
+   },
+   {
+    "stage": "reconcile",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-nextto-small-scale-i00-privileged--f8fun8m/05_reconcile_r1"
+   },
+   {
+    "stage": "traj_review",
+    "round": 2,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-nextto-small-scale-i00-privileged--f8fun8m/04_review_r2"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "c8f80a112dc1ea21fc2b03e99ec9820714b7e3a6d1e852f052a3858ebd0d7098",
+   "video_sha256": "a8e8e4000fb264b1d82fca45a05714cac16eb367beb55726500150b5776514be",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/inter_object_relation_satisfaction_nextto_small_scale/trajectories/ispatialgym-inter-object-relation-satisfaction-nextto-small-scale-i00-privileged--izyumdi.json

@@ -0,0 +1,253 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-inter-object-relation-satisfaction-nextto-small-scale-i00-privileged--izyumdi",
+ "taskKey": "ispatialgym/inter_object_relation_satisfaction_nextto_small_scale",
+ "dag_sha256": "a1913428fea25766fa07bf571634053c1cc8aa302d6a8d8b88da8834e534f9f8",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 629,
+   "sim_steps": 523,
+   "rows": 629,
+   "max_rows": 2076,
+   "replay_success": 1,
+   "instance_ok": 1,
+   "steps": 523,
+   "success_step": 523,
+   "symbolic_success": 1,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 16.947,
+   "peak_memory_gb": 9.799,
+   "peak_processes": 22,
+   "peak_disk_gb": 2.502,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-inter-object-relation-satisfaction-nextto-small-scale-i00-privileged--izyumdi/replay.mp4",
+  "duration_s": 17.6,
+  "fps": 10.0,
+  "frames": 176,
+  "sha256": "bac070d5b8015504d48fff35f9c305b8db3d4651a0fb6657041557c20314bfbf"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 1,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "pencil_at_goal": {
+    "method": "visual",
+    "intervals": [
+     [
+      0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 黄色细彩铅已平放在白色台面，红帽黑身记号笔位于其右侧且明显分离；视频全程至 17.5 s，彩铅未见明显平移或长轴转歪。"
+     },
+     {
+      "t": 17.5,
+      "log_step": null,
+      "text_zh": "推断：视频 12.2–17.5 s 彩铅仍留在原区域，末段两笔未见明显运动；结合评分器 reward.success=1、replay_success=1、symbolic_success=1，判定彩铅目标位姿最终合格。精确位置、等价朝向误差及静置速度不能仅凭画面量出。"
+     }
+    ]
+   },
+   "marker_nextto_pencil": {
+    "method": "visual",
+    "intervals": [
+     [
+      12.2,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 9.6,
+      "log_step": null,
+      "text_zh": "观察：视频 7.5–9.4 s 机器人右夹爪下降到记号笔红帽附近，两指从两侧闭合；视频 9.6 s 起黑色笔身随夹爪移动，原先与彩铅分离的笔身位置空出，随后右夹爪带笔向彩铅靠近。"
+     },
+     {
+      "t": 12.2,
+      "log_step": null,
+      "text_zh": "推断：对照视频 11.6–12.7 s 的逐帧近景，右夹爪携带的记号笔与黄色彩铅间距逐渐缩小，约在视频 12.2 s（时间不确定约±0.3 s）形成近距离并排、长轴近乎平行的外观；至视频 17.5 s 未见关系丢失。本节点 freeze_held=false，不把夹爪尚在笔旁视作未达成。"
+     },
+     {
+      "t": 17.5,
+      "log_step": null,
+      "text_zh": "观察：视频 16.5–17.5 s 每隔 0.1 s 的末段画面中，两笔相对位置及右夹爪外观保持稳定；夹爪遮挡红帽及部分笔身。评分器 reward.success=1、deterministic=1、replay_success=1、success_step=523、symbolic_success=1，记录本次评分重放通过；未用 agent 会话中早期试验的转笔或失放现象标注本次轨迹。"
+     },
+     {
+      "t": 1.5,
+      "log_step": 46,
+      "text_zh": "推断：视频 0.5–2.0 s 台面和两笔在视野中一起接近，与 log step 44 的生成脚本先 drive(0.2,0.0,30)、log step 46 由该脚本生成最终版本及 log step 50 安装该版本的记录相符，判定机器人底盘在抓笔前向桌前进。此处只据视频定位时间，未将 agent 工作时间换算为视频时间。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "观察：视频 9.4–12.7 s，机器人右侧二指夹爪夹住红帽附近的记号笔，笔身随夹爪抬移并靠近彩铅；左夹爪未参与此搬运。"
+     },
+     {
+      "who": "底盘",
+      "how": "支撑",
+      "note_zh": "推断：视频 0.5–2.0 s，机器人轮式底盘支撑机体向桌前进以接近记号笔，随后停下，由右夹爪完成搬运；前进同时得到交付生成脚本（log step 44、46、50）的支持。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "pencil_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "pencil_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "marker_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "marker_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_joint_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "观察：黄色彩铅从视频 0.0 s 起保持原位，机器人右夹爪夹住记号笔并移向彩铅，约在视频 12.2 s 形成相邻且近乎平行的摆放，两个节点最终均达成。观察：视频末段至 17.5 s 未见明显运动，评分器记录 success=1、replay_success=1，联合终态验收通过，无失败断点。推断：首次相邻时间有约±0.3 s 的视觉不确定性；没有状态日志，夹爪遮挡部分笔身，无法独立量出精确位姿与静置指标，DAG 的全部原生指标均未由评分器报告数值，且所给 oracle 为本次运行，不能作为独立参考。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "逐帧核对：彩铅全程原位，记号笔在视频约 9.4 s 被右夹爪夹住，11.6–12.7 s 靠近彩铅，12.2 s 并排，至 17.5 s 保持；节点值、时间、接触、无断点均成立。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:19:26Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-nextto-small-scale-i00-privileged--izyumdi/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-nextto-small-scale-i00-privileged--izyumdi/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "87e2fbcc5407ca31a0d145a548f358897794965f49ce01d87fee851c0bd60d5e",
+   "video_sha256": "bac070d5b8015504d48fff35f9c305b8db3d4651a0fb6657041557c20314bfbf",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/inter_object_relation_satisfaction_nextto_small_scale/trajectories/ispatialgym-inter-object-relation-satisfaction-nextto-small-scale-i00-privileged--ztv8py3.json

@@ -0,0 +1,275 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-inter-object-relation-satisfaction-nextto-small-scale-i00-privileged--ztv8py3",
+ "taskKey": "ispatialgym/inter_object_relation_satisfaction_nextto_small_scale",
+ "dag_sha256": "a1913428fea25766fa07bf571634053c1cc8aa302d6a8d8b88da8834e534f9f8",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 702,
+   "sim_steps": 676,
+   "rows": 702,
+   "max_rows": 2076,
+   "replay_success": 1,
+   "instance_ok": 1,
+   "steps": 676,
+   "success_step": 676,
+   "symbolic_success": 1,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 1,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.0,
+   "peak_processes": 0.0,
+   "peak_disk_gb": 0.0,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-inter-object-relation-satisfaction-nextto-small-scale-i00-privileged--ztv8py3/replay.mp4",
+  "duration_s": 22.7,
+  "fps": 10.0,
+  "frames": 227,
+  "sha256": "b353a420f03ad042ad1c56b6fa07c8245046dd30d70158df101661e282f1ce73"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 1,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "pencil_at_goal": {
+    "method": "visual",
+    "intervals": [
+     [
+      0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 黄色细彩铅平放于白色台面，红帽记号笔与其分离；与参考成功视频 0.0 s 的初态一致。"
+     },
+     {
+      "t": 6.3,
+      "log_step": null,
+      "text_zh": "观察：视频 3.0–5.4 s 机器人右夹爪下降并在彩铅端部合拢，彩铅仅有轻微台面内偏移；视频 5.7–6.6 s 夹爪抬起而彩铅仍留在台面，视频 7.2–10.8 s 空手移动期间彩铅也未随手升起。"
+     },
+     {
+      "t": 22.6,
+      "log_step": null,
+      "text_zh": "推断：视频 0.0–22.6 s 彩铅始终留在原工作区域且无明显转歪，视频 21.6–22.6 s 仍与记号笔并排平放；结合评分器 success=1、replay_success=1、symbolic_success=1，判定该位姿目标从初态持续保持至末帧。视频不能直接量出毫米级位移与姿态误差。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "推/扫",
+      "note_zh": "推断：视频 3.6–5.4 s 右夹爪手指合拢时轻微拨动黄色彩铅端部；视频 5.7–6.6 s 彩铅未被带起，不将合拢动作记为已夹持搬运。"
+     },
+     {
+      "who": "底盘",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.0–1.2 s 机器人向台面前进以接近两支笔，轮式底盘支撑机体完成本次操作的工作位置调整。"
+     }
+    ]
+   },
+   "marker_nextto_pencil": {
+    "method": "visual",
+    "intervals": [
+     [
+      18.3,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 红帽黑身记号笔与黄色彩铅明显分离；视频 10.8–12.3 s 记号笔仍在原台面位置，机器人右夹爪转向它。"
+     },
+     {
+      "t": 15.3,
+      "log_step": null,
+      "text_zh": "观察：视频 13.2–14.7 s 右夹爪在记号笔处闭合；视频 15.0–16.8 s 记号笔露出的黑色笔端随右手离开原处、升高并搬移，彩铅仍在台面，支持实际夹持搬运。"
+     },
+     {
+      "t": 18.3,
+      "log_step": null,
+      "text_zh": "推断：视频 17.1–18.3 s 机器人右手携记号笔下降至彩铅旁，露出的黑色笔端已靠近彩铅；视频 19.8–20.4 s 夹爪撤开后可见两笔在台面近距离并排、长轴大致平行，视频 21.6–22.6 s 保持该关系。因前臂遮住接触区，首次到达时间估为视频 18.3 s，误差约 0.5 s；此节点 freeze_held=false，不以撤手时刻代替到达时刻。"
+     },
+     {
+      "t": 22.6,
+      "log_step": null,
+      "text_zh": "推断：视频 22.0–22.6 s 两笔并排且未见继续移动，与参考成功视频 16.0–17.5 s 的目标关系一致；本次评分器 success=1、deterministic=1、replay_success=1、symbolic_success=1，支持最终关系、位姿及联合静置通过。"
+     },
+     {
+      "t": null,
+      "log_step": 74,
+      "text_zh": "观察：log step 74 的最终脚本运行输出 SAVED 702；评分器报告 n_actions=702、rows=702，表明交付的是最后保存的轨迹。此日志仅用于核对交付，不把较早测试的结果或 agent 时间映射为视频时刻。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "观察：视频 13.2–19.8 s 机器人右侧二指夹爪夹住红帽黑身记号笔，笔端随右手抬升、搬至黄色彩铅旁并下放，随后夹爪张开撤离。"
+     },
+     {
+      "who": "底盘",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.0–1.2 s 机器人轮式底盘前进以接近两支笔，支撑机体完成记号笔摆放前的工作位置调整。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "pencil_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "pencil_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "marker_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "marker_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_max_joint_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "本次重放成功，彩铅目标位姿从视频 0.0 s 持续保持，右夹爪轻微拨动彩铅但未将其提起，随后夹持记号笔搬至彩铅旁，约视频 18.3 s 起两节点均满足并保持至视频 22.6 s。评分器确认确定性重放及符号目标成功，无失败断点；前臂遮挡使首次相邻时刻约有 0.5 s 不确定性，且无状态日志，精确位姿误差和45控制步联合静置依评分结果确认，评分器未报告各原生指标数值，均填 null。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "成功重放，两节点终态与首次到达时间（约18.3 s，因手臂遮挡）与帧一致；无失败断点，抓握/搬运、底盘标注合理。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:22:55Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-nextto-small-scale-i00-privileged--ztv8py3/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-nextto-small-scale-i00-privileged--ztv8py3/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "4b43d75dc87d3d1d66d413227dbac9fa955e95c073a66150c3e56ea8646f2392",
+   "video_sha256": "b353a420f03ad042ad1c56b6fa07c8245046dd30d70158df101661e282f1ce73",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/inter_object_relation_satisfaction_ontop_small_scale/trajectories/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--48zm7wr.json

@@ -0,0 +1,281 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--48zm7wr",
+ "taskKey": "ispatialgym/inter_object_relation_satisfaction_ontop_small_scale",
+ "dag_sha256": "149991ba89630c9372ba8c97b9c9950caaf63efb5952e71b5bf7eb73fa2a6fa7",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 700,
+   "sim_steps": 700,
+   "rows": 700,
+   "max_rows": 1900,
+   "replay_success": 0,
+   "instance_ok": 1,
+   "steps": 700,
+   "success_step": 0,
+   "symbolic_success": 1,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 16.979,
+   "peak_memory_gb": 10.513,
+   "peak_processes": 21,
+   "peak_disk_gb": 1.781,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--48zm7wr/replay.mp4",
+  "duration_s": 23.5,
+  "fps": 10.0,
+  "frames": 235,
+  "sha256": "7a967721258932a53e4b65499b408d05e6d2815c5908b9f9b99e35830befc28b"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 1,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "platter_at_target": {
+    "method": "visual",
+    "intervals": [
+     [
+      0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–1.4 s 红盘平放在台面，包装在盘旁；视频 12.6–23.4 s 红盘重新完整进入画面，盘相对台边、窗框的位置和长轴方向与开局一致，未见被搬走或明显转动。"
+     },
+     {
+      "t": 23.4,
+      "log_step": 220,
+      "text_zh": "推断：DAG记载红盘初态已在位置及姿态容差内；视频首尾固定参照一致，log step 220 的交付生成脚本只搬运包装、不安排移动红盘，故红盘目标位姿从开局保持到终态。中段视野外的微小扰动无法定量复核。"
+     }
+    ]
+   },
+   "package_on_platter": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–1.4 s 包装在红盘左侧台面上；视频 12.5–13.5 s 相机转回时，包装已离开原处、横放在红盘上，视频 13.5–23.4 s 始终由盘面承托，末秒没有滑落。首尾包装的白色中央商标、绿色边带及两端图案朝向基本相同。"
+     },
+     {
+      "t": 23.4,
+      "log_step": 220,
+      "text_zh": "推断：DAG记载包装初态全姿态误差约0.439弧度，大于0.3弧度容差；log step 220 的实际交付生成脚本在抓取、抬升、横移、放下时沿用同一q_rot，没有目标姿态修正，视频首尾包装纹理也保留原朝向。因此虽完成物理上盘，仍未完成含全姿态要求的本节点；未将视野外的计划放下时刻记成达成。"
+     },
+     {
+      "t": 23.4,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分字段为symbolic_success=1、replay_success=0、success_step=0、deterministic=1；符号目标成立没有带来整体通过。该字段未给出包装姿态误差或静止窗口的分项值，不能据此单独判定具体超差量。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：视频中搬运及夹爪闭合、释放区域都在头部相机视野外，只有首尾物体转移可见，无法直接判定实际接触端和接触方式。log step 220 计划用右夹爪抓取，但不据此把计划抓握记为已观察接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "platter_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "platter_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "package_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "package_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_window",
+   "unit": "control_steps",
+   "threshold": 45,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "max_body_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_body_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_robot_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_body_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_body_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_joint_coordinate_drift",
+   "unit": "joint_coordinate",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "package_on_platter",
+   "t": 12.6,
+   "log_step": 220,
+   "where": "操作中",
+   "evidence_zh": "推断：视频 12.6 s 包装随相机回转重新可见时已在盘上，但包装纹理朝向基本保留视频 0.0 s 的初始朝向；结合DAG记载的初始朝向超差和log step 220 不调整物体姿态的搬运方案，断点为包装上盘步骤的姿态验收未达。之后直到视频 23.4 s 没有再次操作修正；评分器replay_success=0。视频 12.6 s 仅为首次清楚看到结果的导航锚点，真正放下发生于视野外。"
+  },
+  "unmet": {
+   "platter_at_target": true,
+   "package_on_platter": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 12.6,
+    "log_step": 220,
+    "evidence_zh": "推断：视频 0.0 s 与视频 13.5–23.4 s 包装的商标和端部图案朝向基本相同，表现为移到盘上而未完成所需的姿态调整；DAG初始全姿态误差约0.439弧度，超出0.3弧度要求，log step 220 的搬运也始终保持同一夹爪目标朝向。包装终态在盘上，并非掉落或未搬动；具体终态角度未由评分器报告。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "认知",
+    "item": "推理/规划",
+    "evidence_zh": "推断：log step 220 的最终生成脚本将所有抓取、抬升、移盘、放下航点都设为同一q_rot，目标位置发生变化而目标朝向不变；log step 221 生成并保存该700行交付。DAG明确初态包装朝向尚不达标，视频首尾纹理也表明主要只发生平移，支持方案遗漏包装目标姿态调整，而非逆解未执行其已设定的目标。",
+    "source_zh": "日志 / 视频"
+   },
+   {
+    "cat": "认知",
+    "item": "误以为完成",
+    "evidence_zh": "观察：log step 267 的新进程重放输出只报告700步完成、终态哈希及初态匹配，没有任务成功字段；agent在log step 272 却自述“两次重放成功”、物体朝向误差小于0.01弧度并结束。log step 270 显示仍余864.9秒工作预算，而实际评分字段replay_success=0、success_step=0；这些记录支持把重放执行完成误当作任务验收通过，影响了停止决定。",
+    "source_zh": "日志 / agent 自述 / 评分器字段"
+   }
+  ],
+  "gap_zh": "本次无状态日志，inputs/oracle/oracle.json 确认没有oracle视频，inputs/也不含可供直接比较的目标图；视频约 1.8–12.3 s 操作区域在画面外，因此实际抓握接触和中途瞬时姿态无法确定。包装朝向未达依赖首尾纹理、DAG的初始超差与交付脚本的联合推断，缺少本次终态四元数及原生误差分项；不能给出精确角度或完全排除联合静止验收另有问题。log step 224 的速度读数来自会话内生成测试，并非新进程评分重放，未作为本次原生指标；symbolic_success=1也不能单独证明全姿态达标。"
+ },
+ "summary_zh": "红盘保持初始目标位姿；包装确实从台面转移到盘上，但首尾纹理及交付脚本支持其沿用初始朝向，含全姿态容差的上盘节点未达，终态为1/2。agent遗漏目标姿态修正，又将重放正常完成和哈希一致误当作验收成功后停止，评分仍为失败。搬运接触在视野外，且缺少原生位姿与静止窗口分项，朝向失败为联合证据推断。",
+ "conf": 6,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "标签与视频、log一致：包装在视频 12.5 s 起已在红盘上，朝向与开局相近，交付脚本全程沿用同一抓取朝向 q_rot，断点和两个根因有证据支持；姿态超差为联合推断，已如实标注。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:34:01Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--48zm7wr/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--48zm7wr/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "6bdd450022eb7cc184ee1c5422a923f8d91d14aa25415bf20bd060f2c5d3e228",
+   "video_sha256": "7a967721258932a53e4b65499b408d05e6d2815c5908b9f9b99e35830befc28b",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/inter_object_relation_satisfaction_ontop_small_scale/trajectories/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--aeg2ccq.json

@@ -0,0 +1,298 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--aeg2ccq",
+ "taskKey": "ispatialgym/inter_object_relation_satisfaction_ontop_small_scale",
+ "dag_sha256": "149991ba89630c9372ba8c97b9c9950caaf63efb5952e71b5bf7eb73fa2a6fa7",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 410,
+   "sim_steps": 410,
+   "rows": 410,
+   "max_rows": 1900,
+   "replay_success": 0,
+   "instance_ok": 1,
+   "steps": 410,
+   "success_step": 0,
+   "symbolic_success": 0,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 16.813,
+   "peak_memory_gb": 10.025,
+   "peak_processes": 21,
+   "peak_disk_gb": 3.097,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--aeg2ccq/replay.mp4",
+  "duration_s": 13.8,
+  "fps": 10.0,
+  "frames": 138,
+  "sha256": "cebc8b5abdf10fca233cf13525c5b44622cf186f9a3fb34a8bc2872bc832f847"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 1,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "platter_at_target": {
+    "method": "visual",
+    "intervals": [
+     [
+      0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–3.0 s 红盘平放在白色台面，包装在盘左侧；DAG 记载冻结初态红盘已满足目标位姿。"
+     },
+     {
+      "t": 8.1,
+      "log_step": null,
+      "text_zh": "推断：视频 3.5–8.7 s 红盘相对台面边沿未见滑动或转动，画面中的位置改变随头部视角变化；视频 9.0–13.7 s 盘逐渐只剩左下角局部，但未见红盘被碰动，按保持初态目标位姿标为持续成立。"
+     }
+    ]
+   },
+   "package_on_platter": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–3.0 s 包装仍在红盘旁的台面上，上盘条件未成立。"
+     },
+     {
+      "t": 10.8,
+      "log_step": 134,
+      "text_zh": "观察：视频 9.6–11.1 s 左下角出现包装局部并向红盘区域下降，视频 11.1–13.7 s 包装局部与盘面投影重叠且最终稳定；夹爪及包装全貌在画外。log step 134 将原485行测试轨迹截为前335行并重复末行75次，最终交付410行闭爪版本；此前释放后的测试终态不当作本次完成证据。"
+     },
+     {
+      "t": 11.1,
+      "log_step": 100,
+      "text_zh": "推断：log step 100 的脚本以包装初始姿态确定固定夹爪朝向，各搬运路点只改变位置；agent 在 log step 124 明确自述将保持包装原朝向。DAG 记载包装初始全姿态误差约0.439弧度，超过0.3弧度容差；结合视频末段没有可见再调整及评分器 symbolic_success=0、replay_success=0，判断包装虽移入盘区，完整上盘位姿条件未达到。最终姿态超差为结合计划的推断，并非从局部末帧量得；未把零总分单独用作未搬运的证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：视频 3.5–13.7 s 抓取区域和夹爪在画外，只有末段包装局部可见。log step 134 的交付动作保留左夹爪闭合指令，但画面不能确认实际包夹形态，且没有接触状态日志，故实际作用端及接触方式记未知。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "platter_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "platter_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "package_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "package_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_window",
+   "unit": "control_steps",
+   "threshold": 45,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "max_body_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_body_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_robot_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_body_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_body_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_joint_coordinate_drift",
+   "unit": "joint_coordinate",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "package_on_platter",
+   "t": 11.1,
+   "log_step": 134,
+   "where": "操作中",
+   "evidence_zh": "推断：视频 9.6–11.1 s 机器人将包装局部送入红盘区域，随后至视频 13.7 s 停住；log step 100、124 的搬运方案保持包装初始朝向，log step 134 的最终交付保留这一放置过程并直接闭爪等待。包装初始姿态超出DAG容差，未见后续纠正，评分器 symbolic_success=0、success=0；断点定位包装上盘步骤，视频 11.1 s 仅为放置结束附近的导航锚点。"
+  },
+  "unmet": {
+   "platter_at_target": true,
+   "package_on_platter": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 11.1,
+    "log_step": 100,
+    "evidence_zh": "推断：视频 10.2–13.7 s 包装已进入红盘的可见局部，但完整姿态在画外。log step 100 生成固定朝向的平移搬运，log step 124 自述保持原朝向；DAG 给出的该初态朝向误差约0.439弧度大于0.3弧度验收阈值，因此将包装朝向未对齐保留为位姿不准的推断，包装并非未曾移动。末态精确误差没有评分器分项数值可核实；log step 134 删除释放段并闭爪交付是否影响验收，也未被整体失败字段排除。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "认知",
+    "item": "推理/规划",
+    "log_step": 124,
+    "evidence_zh": "推断：agent 在 log step 124 将包装原朝向判断为接近目标并决定全程保持；log step 100 的路点只规划从包装初始位置平移到红盘中心，未规划目标朝向修正。DAG 记载原朝向误差已超过容差，因此保留目标朝向规划遗漏这一可能归因。另据 log step 134 的截取代码及输出，agent 删除原485行版本的释放和撤离段，以前335行加75行末动作重复交付410行版本，末动作左夹爪指令为闭合；log step 135 自述此举是为保持目标图中的夹爪接触。该选择是否因未释放而导致失败，无法由评分器 symbolic_success=0 判定，不能将朝向未修正视为唯一原因；log step 127 自述的释放版测试成功也不作为本次评分成功证据。",
+    "source_zh": "日志"
+   }
+  ],
+  "gap_zh": "头部视角在视频约 3.5 s 后将抓取区移出画面，视频 9.6–13.7 s 只看见包装和红盘局部，实际夹爪接触、完整支撑关系和终态全姿态无法直接测量；没有逐帧状态日志，oracle也没有视频。朝向不达标及规划归因依据DAG初态误差、脚本和可见搬运作推断，不能精确确认超差数值及首次出现时刻。log step 134 删除释放段并保存410行闭爪版本，log step 135 明确自述保留辅助抓握；这一选择可能影响盘上关系或联合验收，但实际夹持情况及其失败作用未获直接确认。DAG package_on_platter.sources 记载oracle的一种成功方式为释放并停住，note_zh 同时说明原生关系实现及是否另有放手谓词缺失，故不能据oracle新增松手要求。评分器仅给 symbolic_success=0 等整体字段，未给各物体误差、静止窗口或单项关系，无法区分朝向超差与未释放等可能解释；因此所有原生数值指标留空，也不将最终闭爪本身断言为失败原因。"
+ },
+ "summary_zh": "红盘初态已达标，视频中未见其被扰动，最终计1/2个节点；包装从盘旁被移入红盘区域，但完整上盘位姿条件判断未达到。断点在包装上盘：脚本保留原本超出朝向容差的初始姿态，保留位姿不准及目标朝向规划遗漏为推断；log step 134/135 删除释放段、闭爪交付是否另致失败仍未排除。末段仅有左下角局部可见，评分器无分项结果，无法区分朝向超差与未释放等解释，因此置信度由6降至5。",
+ "conf": 5,
+ "review": {
+  "status": "agreed",
+  "rounds": 2,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "fix",
+    "conf": 6,
+    "summary_zh": "节点值、时间、断点(package_on_platter)与视频一致：红盘初态保持，9.0–13.7 s 左下角确实只见红盘局部与包装一角重叠。问题在归因：位姿不准/规划遗漏依据DAG初态误差0.439 rad+保持朝向推断，但没有排除另一同样可能的解释——agent 在 log step 127→134 刻意把已释放版(485行)改成不释放、夹爪闭合的410行版本(oracle成功方式为释放并停住)，评分器只给 symbolic_success=0，无法区分朝向超差与未释放。",
+    "n_issues": 1,
+    "n_major": 0
+   },
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 2,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "节点值（红盘保持、包装上盘未达成）、断点 package_on_platter 与视频一致；视频 3 s 后头部转向、抓取区出画，末段仅见红盘局部，标签已把朝向超差与未释放写成非唯一推断并降置信度，上一轮意见已妥善处理，无需再改。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:50:38Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--aeg2ccq/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--aeg2ccq/04_review_r1"
+   },
+   {
+    "stage": "reconcile",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--aeg2ccq/05_reconcile_r1"
+   },
+   {
+    "stage": "traj_review",
+    "round": 2,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--aeg2ccq/04_review_r2"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "c21628f0013a94763ad3e67515aba91015ff4b8947db201b8f7f509c8df86274",
+   "video_sha256": "cebc8b5abdf10fca233cf13525c5b44622cf186f9a3fb34a8bc2872bc832f847",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/inter_object_relation_satisfaction_ontop_small_scale/trajectories/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--axhrimm.json

@@ -0,0 +1,264 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--axhrimm",
+ "taskKey": "ispatialgym/inter_object_relation_satisfaction_ontop_small_scale",
+ "dag_sha256": "149991ba89630c9372ba8c97b9c9950caaf63efb5952e71b5bf7eb73fa2a6fa7",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 357,
+   "sim_steps": 357,
+   "rows": 357,
+   "max_rows": 1900,
+   "replay_success": 0,
+   "instance_ok": 1,
+   "steps": 357,
+   "success_step": 0,
+   "symbolic_success": 0,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 1,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.0,
+   "peak_processes": 0.0,
+   "peak_disk_gb": 0.0,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--axhrimm/replay.mp4",
+  "duration_s": 12.0,
+  "fps": 10.0,
+  "frames": 120,
+  "sha256": "d776e41ef44b3d350d468cbebc2cdbf30607ceae3e78211dd3cd8009e4b30025"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "platter_at_target": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–1.0 s，红色椭圆盘平放于白色台面，包装在盘旁；结合DAG记载的冻结初态，红盘开局位姿达标。"
+     },
+     {
+      "t": 1.2,
+      "log_step": null,
+      "text_zh": "观察：视频约 1.1 s 红盘开始离开画面，视频 1.2–11.9 s 未再显示红盘；末段只有包装边角进入画面右下方。没有状态日志及红盘位姿评分字段，无法确定离画后是否保持位置和朝向，完整区间及终态留未知，不将离画当作失去达标状态。"
+     }
+    ]
+   },
+   "package_on_platter": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–1.1 s，香肠包装放在红盘左侧的白色台面上，与盘面分离；此段没有上盘。"
+     },
+     {
+      "t": 11.9,
+      "log_step": null,
+      "text_zh": "观察：已逐帧查看视频 0.0–11.9 s；视频 1.2 s 后抓取区离画，视频约 8.8–11.9 s 仅包装边角在右下方可见，邻接区域为白色台面，未见红盘承托。评分器 reward.symbolic_success=0、replay_success=0、n_actions=357。"
+     },
+     {
+      "t": 11.9,
+      "log_step": null,
+      "text_zh": "推断：结合初态盘与包装分离、末段露出的包装边角周围仍是白色台面，以及评分器原生关系检查 symbolic_success=0，包装上盘终态判为未达；没有可核实的达标区间。空区间不表示已看清离画期间每次抓取，遮挡限制见 gap_zh。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：视频 1.2–11.9 s 的主要操作区不在画面中；log step 120 的最终脚本命令右夹爪下降、闭合及提起，但视频未显示右夹爪与包装的接触，无法确认实际是夹持、按压还是未碰到，不能按典型方案记抓握。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "platter_position_error",
+   "value": null,
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better"
+  },
+  {
+   "name": "platter_orientation_error",
+   "value": null,
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better"
+  },
+  {
+   "name": "package_position_error",
+   "value": null,
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better"
+  },
+  {
+   "name": "package_orientation_error",
+   "value": null,
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better"
+  },
+  {
+   "name": "settle_window",
+   "value": null,
+   "unit": "control_steps",
+   "threshold": 45,
+   "direction": "higher_better"
+  },
+  {
+   "name": "max_body_linear_speed",
+   "value": null,
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better"
+  },
+  {
+   "name": "max_body_angular_speed",
+   "value": null,
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better"
+  },
+  {
+   "name": "max_robot_joint_speed",
+   "value": null,
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better"
+  },
+  {
+   "name": "max_window_body_position_drift",
+   "value": null,
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better"
+  },
+  {
+   "name": "max_window_body_orientation_drift",
+   "value": null,
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better"
+  },
+  {
+   "name": "max_window_joint_coordinate_drift",
+   "value": null,
+   "unit": "joint_coordinate",
+   "threshold": 0.05,
+   "direction": "lower_better"
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "package_on_platter",
+   "t": null,
+   "log_step": 121,
+   "where": "操作中",
+   "evidence_zh": "推断：log step 120 启动最后一次右夹爪抓取及移向红盘的脚本，log step 121 的该次输出记录包装提起检查失败、终态仍在原处并保存357行；log step 122 核实交付为357行，评分器也重放357步且 symbolic_success=0，故失败步骤为包装上盘。视频 1.2 s 后操作区离画，不能定位抓取失败的可见时刻，t留空；此前重置测试中的掉落不作为本次断点。"
+  },
+  "unmet": {
+   "platter_at_target": null,
+   "package_on_platter": false
+  },
+  "phenomena": [],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "t": null,
+    "log_step": 121,
+    "evidence_zh": "推断：agent脚本在 log step 121 报告闭爪后包装位移仅0.0002米，执行提起动作后包装高度仍为0.839米、平面位置仍为[0.694,0.097]，随后记录no retry并保存357行；log step 116、119 的独立测试也记录提起后包装仍在台面高度。结合 log step 122 的交付行数及评分器357步失败结果，证据支持抓取执行环节的控制问题；无法进一步判定指垫几何、闭合量或跟踪的具体责任。",
+    "source_zh": "日志"
+   }
+  ],
+  "gap_zh": "观察：输入有完整12.0秒视频和运行日志，但没有状态日志；oracle.json注明成功参考无视频。视频约 1.2 s 后红盘及主要夹爪操作区域离画，视频 11.0–11.9 s 逐帧也只见包装一小角，不能核实抓握、推压或短暂上盘，因此接触填未知，失败现象留空，且不能给出可靠断点视频时刻；红盘仅初态可确认达标，完整区间及终态留未知。log step 121 的坐标是agent脚本诊断，作为根因的会话证据，不能替代不可见的逐帧接触；log step 123 关于夹爪闭合轴的猜测未经验证，未据此细分根因。评分器未给出DAG所列位姿误差和静止窗口指标，native_metrics均留空。"
+ },
+ "summary_zh": "红盘在视频开局已处于目标位姿，包装在盘旁；评分器原生关系检查未通过，包装上盘终态未达，红盘终态因离画无法确认。最后交付的357步轨迹对应一次抓取失败的尝试，会话诊断支持控制问题，但视频约 1.2 s 后未展示主要操作区，实际接触、具体失败现象及红盘后续位姿无法核实。",
+ "conf": 5,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "标注保守且与证据一致：视频约1.2 s后主要操作区离画，红盘终态与实际接触留未知、包装上盘终态为否（评分器 symbolic_success=0）、断点定为 package_on_platter 均有日志和评分字段支持，无需修改。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:45:58Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--axhrimm/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--axhrimm/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "446009885b6ef9c2d87e4f69aa5921a6e7846be6e4cb4d0215076c0bfdb20dab",
+   "video_sha256": "d776e41ef44b3d350d468cbebc2cdbf30607ceae3e78211dd3cd8009e4b30025",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/inter_object_relation_satisfaction_ontop_small_scale/trajectories/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--nms8bjx.json

@@ -0,0 +1,293 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--nms8bjx",
+ "taskKey": "ispatialgym/inter_object_relation_satisfaction_ontop_small_scale",
+ "dag_sha256": "149991ba89630c9372ba8c97b9c9950caaf63efb5952e71b5bf7eb73fa2a6fa7",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 1895,
+   "sim_steps": 1895,
+   "rows": 1895,
+   "max_rows": 1900,
+   "replay_success": 0,
+   "instance_ok": 1,
+   "steps": 1895,
+   "success_step": 0,
+   "symbolic_success": 1,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 17.085,
+   "peak_memory_gb": 9.95,
+   "peak_processes": 23,
+   "peak_disk_gb": 3.121,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--nms8bjx/replay.mp4",
+  "duration_s": 63.3,
+  "fps": 10.0,
+  "frames": 633,
+  "sha256": "ec64fbbe255794c70b559e879a423ec481f44f98e92936f205aee85f534027b5"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 1,
+  "peak": 2,
+  "final": 1,
+  "nodes": {
+   "platter_at_target": {
+    "method": "visual",
+    "intervals": [
+     [
+      0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 红盘已平放在台面目标区域，包装在盘左侧；视频全程粗查及视频 36.0–43.5 s、55.5–57.4 s、60.5–63.2 s 密查中，红盘未被移走或明显转向。"
+     },
+     {
+      "t": 63.2,
+      "log_step": 143,
+      "text_zh": "推断：红盘位姿从初态保持到视频 63.2 s。DAG记载初态位姿已达标；log step 143 所示最后交付片段的诊断盘中心为(7.09798,3.16221,0.82074)米，距DAG目标约0.002米，朝向变化也远小于0.3弧度，与视频末帧盘仍在原处一致；微小振动不等同于本节点位姿失效。"
+     }
+    ]
+   },
+   "package_on_platter": {
+    "method": "visual",
+    "intervals": [
+     [
+      38.0,
+      55.9
+     ]
+    ],
+    "final": false,
+    "evidence": [
+     {
+      "t": 12.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–10.0 s 包装在红盘左侧的台面上；视频 12.0 s 包装已离开原位置，随右夹爪抬起；视频 15.0–18.6 s 右手将它移到盘上方并下降。此时到达盘的投影区域不等于已有盘面支撑。"
+     },
+     {
+      "t": 38.0,
+      "log_step": 138,
+      "text_zh": "推断：视频约 38.0 s 微量下降后，包装在盘中央接触盘面且保持一定倾角，暂记完整上盘节点成立，起点约有0.5秒不确定性。log step 132、136 的交付动作先持物倾斜再逐级降低；log step 138 的同段诊断在高度0.868米时OnTop为false、降到0.866米时为true。此处OnTop是仿真器的包装与盘接触且盘位于下方的关系检查，不是成功总判据。"
+     },
+     {
+      "t": 41.3,
+      "log_step": 138,
+      "text_zh": "推断：交付动作共1895个30赫兹控制步，对应视频约63.2秒；log step 136 的0.866米高度段结束于控制步1240，即视频约41.3 s，视频中可核对为持物停在盘面的阶段；控制步330附近开始上提对应视频约11.0 s，也与视频 10.0 s 在旧位置、12.0 s 已抬起一致。视频 38.0–55.8 s 的持物倾角及位置结合log step 132 的倾斜目标与log step 129–130 的目标图拟合，支持该段近似达标；没有原生逐帧姿态误差可精确复核边界。"
+     },
+     {
+      "t": 55.9,
+      "log_step": 142,
+      "text_zh": "观察：视频 55.5–55.8 s 包装仍由夹爪围住；视频约 55.9–56.3 s 右夹爪完全张开，包装改变倾角、落平一些，随后留在盘上。视频 60.5–63.2 s 包装未再被抓起恢复倾角。"
+     },
+     {
+      "t": 63.2,
+      "log_step": 143,
+      "text_zh": "推断：最终盘上关系存在，但包装完整目标姿态未保持。log step 143 为log step 142 保存、log step 144 随即复制交付的终态诊断：包装四元数约(-0.002526,-0.001656,0.959100,0.283052)，相对log step 52 所示初态四元数仅转动0.08596弧度；结合DAG记载的初态目标朝向误差0.43906弧度，旋转距离三角不等式给出终态目标朝向误差至少0.35310弧度，超过0.3弧度。该推断以交付片段诊断与视频所见释放后变平一致为依据；评分字段symbolic_success=1仅确认关系，不能抵消朝向要求。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "观察：视频约 8.0–11.0 s 右二指夹爪夹住包装两侧；视频 12.0 s 包装随右手离开原台面，随后被搬到盘上。机器人用同一右夹爪持物下降、微调高度及开度，视频约 55.9 s 完全张开后包装留在盘面，夹爪仍停在包装旁。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "platter_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "platter_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "package_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "package_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_window",
+   "unit": "control_steps",
+   "threshold": 45,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "max_body_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_body_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_robot_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_body_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_body_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_joint_coordinate_drift",
+   "unit": "joint_coordinate",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "package_on_platter",
+   "t": 55.9,
+   "log_step": 142,
+   "where": "操作中",
+   "evidence_zh": "观察：视频约 55.9 s 右夹爪在盘上张开，包装随之改变倾角；视频 56.3–63.2 s 没有恢复倾角的动作。推断：包装虽留在盘上，释放后未保持完整目标姿态，失败落在香肠包装上盘步骤；log step 143 的交付终态诊断与DAG初态误差合用，给出包装终态朝向误差下界约0.353弧度，高于0.3弧度。"
+  },
+  "unmet": {
+   "platter_at_target": true,
+   "package_on_platter": false
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "放后不稳",
+    "t": 55.9,
+    "log_step": 142,
+    "evidence_zh": "观察：视频 55.5–56.3 s 机器人按计划在红盘上张开右夹爪，包装留在盘面却改变倾角，视频 63.2 s 仍是释放后的姿态。推断：这是到盘后的释放使包装姿态变化，非在盘外松手；log step 143 的保存终态诊断还记录包装角速度0.08443弧度/秒、红盘角速度0.03603弧度/秒，均高于任务静止阈值0.005弧度/秒，支持释放后没有形成所需稳定终态。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "抓取和接触设计",
+    "alt_feasible": "未知",
+    "t": 55.9,
+    "log_step": 142,
+    "source_zh": "日志 / 视频",
+    "evidence_zh": "推断：机器人采用先夹持倾斜包装、逐级压低到盘面、再完全松开的接触方案，但最终盘面支承没有保留所需倾角。log step 132 明确设置包装倾角(-0.18,-0.08)弧度，log step 136、139 调低高度并调整夹爪开度，log step 142 最后完全打开夹爪；视频 55.9–56.3 s 看到随张开产生的姿态变化，log step 143 终态包装已近乎平放且残留角运动。证据支持释放与支承几何不适配，无法据此断言所有夹持或放置方案均不可行。"
+   }
+  ],
+  "gap_zh": "无逐帧状态日志，且成功oracle没有视频；目标图原文件及原生逐帧目标误差未随本阶段输入提供。包装在视频约38.0–55.9 s 完整达标的区间，尤其倾角是否始终在0.3弧度内，依据画面、交付动作和目标图拟合近似推断；终态朝向超差有旋转距离下界支持，但交付片段的诊断数值并非评分器直接输出的原生指标。评分器只报symbolic_success=1、replay_success=0，未逐项报告位姿和静止窗口失败项；不能从视频精确测出微小速度，也不把会话中其他重置试验当成评分轨迹。"
+ },
+ "summary_zh": "红盘从视频 0.0 s 起保持目标位姿；右夹爪将包装搬到盘上，视频约 38.0–55.9 s 暂时达到放置及倾斜姿态，最终保留1/2个节点。视频约 55.9 s 完全松开后包装倾角改变，交付终态诊断支持朝向超差且仍有残余角运动，评分器虽报symbolic_success=1，整体success仍为0；接触与释放方案未保持所需终态。缺少逐帧状态、目标图原文件及原生���差，临时达标区间属于近似推断。",
+ "conf": 6,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "节点值、时间、断点（package_on_platter，操作中）与现象/根因均有帧和log支持，未发现需改动的错误。红盘全程在原位，包装约38 s起在盘上、55.9 s松爪后变平，终态朝向与静止窗口未达标与success=0一致。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:47:06Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--nms8bjx/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--nms8bjx/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "e083d63bb1109392ab2c6d9871dcf2862ca875d3d57f943b2cfacc2019040980",
+   "video_sha256": "ec64fbbe255794c70b559e879a423ec481f44f98e92936f205aee85f534027b5",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/inter_object_relation_satisfaction_ontop_small_scale/trajectories/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--tl5va93.json

@@ -0,0 +1,265 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--tl5va93",
+ "taskKey": "ispatialgym/inter_object_relation_satisfaction_ontop_small_scale",
+ "dag_sha256": "149991ba89630c9372ba8c97b9c9950caaf63efb5952e71b5bf7eb73fa2a6fa7",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 575,
+   "sim_steps": 575,
+   "rows": 575,
+   "max_rows": 1900,
+   "replay_success": 0,
+   "instance_ok": 1,
+   "steps": 575,
+   "success_step": 0,
+   "symbolic_success": 1,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 16.997,
+   "peak_memory_gb": 10.586,
+   "peak_processes": 29,
+   "peak_disk_gb": 0.64,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--tl5va93/replay.mp4",
+  "duration_s": 19.3,
+  "fps": 10.0,
+  "frames": 193,
+  "sha256": "cd41512b34c056bcd493099f5debc8feb29a86bc7285578822828bc61d870f00"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "platter_at_target": {
+    "method": "visual",
+    "intervals": [
+     [
+      0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 红盘平放于白色台面，包装在其左侧；DAG注明红盘初态已在目标位姿容差内。"
+     },
+     {
+      "t": 19.2,
+      "log_step": null,
+      "text_zh": "推断：对照视频 0.0–19.2 s 的台面参照，红盘始终平放且没有明显平移或旋转；视频 11.5 s 包装落盘及视频 18.2–19.2 s 逐帧检查均未见红盘被推走，故保留其初始达标状态。相机视角随机器人躯干动作改变，不据红盘的屏幕坐标变化判为移位。"
+     }
+    ]
+   },
+   "package_on_platter": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": 6.5,
+      "log_step": 159,
+      "text_zh": "观察：视频 4.5–6.5 s 左侧画面中的二指夹爪合拢夹住包装，随后包装离开原来的台面位置，随夹爪上升并向红盘移动；log step 159 的最终生成动作使用 left_tcp 和左夹爪，与此动作顺序相符。"
+     },
+     {
+      "t": 14,
+      "log_step": null,
+      "text_zh": "观察：视频约 11.5 s 包装已下降到红盘面，视频 12.5–14.0 s 夹爪张开并上撤，包装留在盘面；视频 18.2–19.2 s 逐帧可见包装仍横放在盘上，未见滑落或宏观晃动。盘上放置这一几何结果已发生，时间约有 0.5 s 不确定性。"
+     },
+     {
+      "t": 19.2,
+      "log_step": 79,
+      "text_zh": "推断：本节点还要求包装完整姿态误差≤0.3弧度；现有输入没有目标图或完整目标四元数，没有重放状态日志或逐物体误差字段。log step 79 的154度偏航是agent按图像轮廓拟合的自定目标，log step 167 的约0.005弧度也是agent相对该估计的自述，不能验证评分器目标。虽已观察到上盘，仍无法确定整个节点何时达标或终值，故完整节点保留未知。"
+     },
+     {
+      "t": null,
+      "log_step": 166,
+      "text_zh": "观察：log step 164 保存575行动作，log step 166 确认交付数组为(575,23)且数值有限；评分器记录同为575步、deterministic=1、symbolic_success=1、replay_success=0、success=0，且infra_error=sim_error=grader_error=0。symbolic_success只报告符号层结果，未给出本节点的位姿验收结果。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "观察：视频约 5.0–12.5 s 机器人左手二指夹爪夹住香肠包装两侧，提离台面后搬到红盘上，再张开并向上撤离；由 log step 159 的 left_tcp 动作确认作用手。未见另一只手参与此步骤。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "platter_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "platter_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "package_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "package_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_window",
+   "unit": "control_steps",
+   "threshold": 45,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "max_body_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_body_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_robot_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_body_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_body_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_joint_coordinate_drift",
+   "unit": "joint_coordinate",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "package_on_platter",
+   "t": 14,
+   "log_step": null,
+   "where": "终态验收",
+   "evidence_zh": "观察：视频约 11.5–14.0 s 机器人已将包装放到红盘面并撤开夹爪，视频 14.0–19.2 s 包装留在盘上、红盘保持原位；评分器仍记录 replay_success=0、success_step=0。推断：可定位到包装放置后的最终位姿及联合静止验收，具体失败瞬间和未通过的合取项无法由当前记录确定；视频 14.0 s 仅作放置结束后的导航锚点。"
+  },
+  "unmet": {
+   "platter_at_target": true,
+   "package_on_platter": null
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：inputs/只有本次重放视频、运行记录及oracle元数据，没有状态日志，oracle也没有视频；该工作目录没有可查看的目标图或完整目标四元数。评分器只给总体失败和symbolic_success=1，未报告两物体的位姿误差、速度或45步联合静止验收明细；log step 163–164 的 settled_by_drift 读数来自agent会话测试，不是评分重放的验收数据。推断：能确认包装已抓起、搬运、落盘和松爪，不能确认包装是否在完整姿态容差内，也不能在目标姿态偏差、细微速度超阈值或其他验收项之间判别；因此包装完整节点未知，现象与根因留空，不把会话测试中的抖动或agent的完成自述当作重放失败原因。"
+ },
+ "summary_zh": "红盘从初态起保持目标位姿；机器人左夹爪抓起香肠包装，于视频约 11.5–14.0 s 放到红盘上并松爪，包装在视频 19.2 s 仍留在盘面。评分器记录 symbolic_success=1 但最终 success=0，断点定位于包装放置后的终态验收；包装完整姿态及联合静止验收缺少重放明细，完整包装节点保留未知，现象和根因不作无证据归因。",
+ "conf": 6,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "标注与视频一致：左夹爪抓起包装、约 11.5 s 放到红盘上并松爪，红盘未被移动；包装完整位姿及联合静止验收缺少明细，保持 unknown 并留空现象/根因有依据。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:28:46Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--tl5va93/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--tl5va93/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "b968566e3ce5e8709cd3033db8920b54ac34e8bcae4259239c18b662de5e88e7",
+   "video_sha256": "cd41512b34c056bcd493099f5debc8feb29a86bc7285578822828bc61d870f00",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/inter_object_relation_satisfaction_ontop_small_scale/trajectories/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--ubd7spm.json

@@ -0,0 +1,281 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--ubd7spm",
+ "taskKey": "ispatialgym/inter_object_relation_satisfaction_ontop_small_scale",
+ "dag_sha256": "149991ba89630c9372ba8c97b9c9950caaf63efb5952e71b5bf7eb73fa2a6fa7",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 1800,
+   "sim_steps": 1800,
+   "rows": 1800,
+   "max_rows": 1900,
+   "replay_success": 0,
+   "instance_ok": 1,
+   "steps": 1800,
+   "success_step": 0,
+   "symbolic_success": 1,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 1,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 33.295,
+   "peak_memory_gb": 16.573,
+   "peak_processes": 32,
+   "peak_disk_gb": 1.379,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "over_gpu_memory_gb": 1
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--ubd7spm/replay.mp4",
+  "duration_s": 60.1,
+  "fps": 10.0,
+  "frames": 601,
+  "sha256": "249e60ceca59b40d15719a72b394c11e52fb604c96c5cef6ff1c90faae3b08fc"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 1,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "platter_at_target": {
+    "method": "visual",
+    "intervals": [
+     [
+      0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 红色椭圆盘平放于白色台面，包装在盘左侧；DAG 的红盘初态基准已在目标位姿容差内。"
+     },
+     {
+      "t": 60.0,
+      "log_step": null,
+      "text_zh": "推断：视频 0.0–60.0 s 的全程粗采样、视频 10.5–16.2 s 的放置密集帧及视频 59.0–60.0 s 的逐帧检查中，红盘相对台面的位置和长轴方向均保持，机器人没有抓取或推移红盘；据初态已达标的基准，将红盘位姿节点保持为真。包装和机械臂的阴影变化不作为盘体倾倒或移动。"
+     }
+    ]
+   },
+   "package_on_platter": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 包装在红盘左侧台面上，尚未上盘。"
+     },
+     {
+      "t": 7.4,
+      "log_step": null,
+      "text_zh": "观察：视频 4.6–6.2 s 右夹爪从两侧夹住包装，视频 6.6–7.4 s 包装随夹爪升离台面，原来的台面位置空出；随后右臂将包装搬向红盘，包装的长轴方向及朝上的标签面基本保持。"
+     },
+     {
+      "t": 12.0,
+      "log_step": null,
+      "text_zh": "观察：视频 10.5–12.3 s 右夹爪将包装降至红盘上，视频 14.7–15.6 s 夹爪撤离时包装留在盘面；视频 16.0–60.0 s 的粗采样及视频 59.0–60.0 s 的逐帧检查均显示包装仍横放于盘上，没有滚出或再次抓取。这里只确认盘面承托关系。"
+     },
+     {
+      "t": 60.0,
+      "log_step": 103,
+      "text_zh": "推断：本节点还要求包装全姿态误差≤0.3弧度；DAG criterion_zh 及其初态来源说明包装初始误差为0.4390613弧度，且无姿态对称豁免。视频中的抓取、平移、放下及末帧未显示包装重定向；log step 64 的脚本沿用同一末端四元数，log step 90 只改落点，agent 在 log step 103 也明确自述保持原偏航及水平姿态。由初态误差与视频中基本保留的姿态推断，包装虽已上盘，仍未满足完整姿态验收，因此不记达标区间；不是由总分0反推未上盘。会话 log step 32、91 的里程计姿态读数近乎相同，仅作为这一推断的辅助证据，并非评分回放逐帧状态。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "观察：机器人右手二指夹爪在视频约5.2 s 夹住香肠包装，视频约6.6–12.0 s 提起并搬运到红盘，随后张开并在视频约14.7–15.6 s 撤开；左手未参与此操作，未见底盘行驶。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "platter_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "platter_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "package_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "package_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_window",
+   "unit": "control_steps",
+   "threshold": 45,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "max_body_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_body_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_robot_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_body_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_body_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_joint_coordinate_drift",
+   "unit": "joint_coordinate",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "package_on_platter",
+   "t": 12.0,
+   "log_step": 64,
+   "where": "操作中",
+   "evidence_zh": "推断：视频约12.0 s 机器人把包装以基本保留初始朝向的姿态放到红盘上，随后松手撤开，直到视频60.0 s 未再纠正姿态；DAG 给出的包装初态朝向误差0.4390613弧度大于0.3弧度容差，故上盘步骤的完整位姿验收未完成。log step 64 的放置脚本固定末端朝向，log step 103 的最终报告明确称保持原偏航。评分字段 symbolic_success=1、replay_success=0，与物体关系成立但整体未验收成功相容，不能仅凭 symbolic_success 判定全部位姿达标。"
+  },
+  "unmet": {
+   "platter_at_target": true,
+   "package_on_platter": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 12.0,
+    "log_step": 103,
+    "evidence_zh": "推断：视频约12.0 s 包装被放上红盘，但视频0.0 s、7.4 s、15.6 s 和60.0 s 对比显示包装朝向基本保留，未见目标姿态所需的重定向；结合DAG初态全姿态误差0.4390613弧度及0.3弧度容差，判为包装朝向未达标。视频末尾包装仍由盘面承托，不是掉落或滑出。agent 在 log step 103 对轨迹的描述同样称保持偏航不变。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "认知",
+    "item": "感知",
+    "evidence_zh": "推断：agent 在 log step 103 将目标图要求解释为“keeping its yaw”，并称根据目标图拟合相机、估计平移目标；log step 64 的脚本只搬移包装且保持固定末端朝向，log step 90 只调整放置位置。视频约6.6–15.6 s 的实际搬放与这一估计一致；但DAG的包装初态全姿态误差已经超过目标容差，保持原姿态无法满足目标，支持目标图中包装朝向估计错误，而非已规划旋转后未跟踪到位。",
+    "source_zh": "日志"
+   }
+  ],
+  "gap_zh": "观察：inputs/ 没有 states.jsonl，也没有成功参照视频；评分字段未报告各对象位姿误差或45控制步静止窗口的逐步结果，全部原生指标值因此保留为null。推断：包装姿态不达标由DAG初态误差、视频中未重定向的搬放过程及脚本共同判断，无法给出评分回放的精确终态角度。log step 93、98–100 是生成最终交付文件的新进程试验，确实记录包装和盘的静止判定失败；log step 101 复制其1800行文件交付，但这些会话采样不替代评分回放的速度日志。视频末尾无明显宏观晃动，故不把agent在log step 103自述的微小抖动另标为放后不稳，也不据此断言任务或环境异常。"
+ },
+ "summary_zh": "红盘从视频0.0 s 起保持目标位姿；右夹爪成功抓起包装，在视频约12.0 s 放到盘上并撤开，但包装基本保留初始姿态，结合DAG的初始朝向误差推断其未满足0.3弧度姿态容差，最终节点为1/2。断点为香肠包装上盘，现象为位姿不准，日志支持agent把目标图误估为保持原朝向即可；评分器报告symbolic_success=1但replay_success=0。没有评分回放的位姿和速度日志，精确姿态误差及静止窗口失败情况不能直接核验，微小抖动仅见会话诊断。",
+ "conf": 6,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "红盘全程保持、包装在视频约12 s 放上盘并撤开，节点结果1/2、断点package_on_platter、位姿不准的推断（初态朝向误差0.439>0.3 rad，agent会话里包装四元数始终0.441/0.897即偏航未变）与帧相符；因无评分回放的位姿/速度日志，朝向与静止窗口哪个先失败无法由准入证据区分，文件已在gap_zh中如实说明，未发现必须修改的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:46:55Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--ubd7spm/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--ubd7spm/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "cf90ec09963a79e5008357ada3ed571fb01e35cb4a7ff449d003034bdd8c7d0c",
+   "video_sha256": "249e60ceca59b40d15719a72b394c11e52fb604c96c5cef6ff1c90faae3b08fc",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/inter_object_relation_satisfaction_ontop_small_scale/trajectories/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--ygkjwdq.json

@@ -0,0 +1,303 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--ygkjwdq",
+ "taskKey": "ispatialgym/inter_object_relation_satisfaction_ontop_small_scale",
+ "dag_sha256": "149991ba89630c9372ba8c97b9c9950caaf63efb5952e71b5bf7eb73fa2a6fa7",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 331,
+   "sim_steps": 331,
+   "rows": 331,
+   "max_rows": 1900,
+   "replay_success": 0,
+   "instance_ok": 1,
+   "steps": 331,
+   "success_step": 0,
+   "symbolic_success": 1,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 1,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.0,
+   "peak_processes": 0.0,
+   "peak_disk_gb": 0.0,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--ygkjwdq/replay.mp4",
+  "duration_s": 11.2,
+  "fps": 10.0,
+  "frames": 112,
+  "sha256": "f22eb03f54eb035174ec8322625a2c1a5f29da8088d614ea69c56c16320c3240"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": null,
+  "peak": null,
+  "final": 1,
+  "nodes": {
+   "platter_at_target": {
+    "method": "visual",
+    "intervals": [
+     [
+      0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0、0.5、1.0 s，红色椭圆盘平放在白色台面上，包装在盘旁；DAG注明此冻结初态的盘位姿已达标。"
+     },
+     {
+      "t": 11.1,
+      "log_step": 204,
+      "text_zh": "推断：视频 5.5–11.1 s 的右下角持续可见红盘边缘，未见盘被搬走或翻转；log step 194 的脚本明确读取最终交付的331行文件重放，log step 204 得到盘中心约(7.0971,3.1618,0.8207)米，与初始位置相同。结合初态达标及交付脚本只操作包装，判盘目标位姿保持；镜头移动和裁边使这一连续保持为上下文推断，非逐帧原生位姿测量。"
+     }
+    ]
+   },
+   "package_on_platter": {
+    "method": "visual",
+    "intervals": null,
+    "final": false,
+    "evidence": [
+     {
+      "t": 0,
+      "log_step": 12,
+      "text_zh": "观察：视频 0.0–1.5 s，包装仍在红盘旁的台面上，未在盘上；log step 12 记录其初始四元数约(0.000612,-0.000023,0.946104,0.323864)，对应偏航约142.21°、倾斜约0.07°。"
+     },
+     {
+      "t": 8.5,
+      "log_step": null,
+      "text_zh": "观察：视频 6.0、7.4–8.6、9.5 s 及末秒视频 10.2–11.1 s，画面右下角逐渐露出的包装位于红盘上；包装和夹爪的大部分始终在画面外，未直接看到完整拿起、闭夹或松开过程。评分器 reward.symbolic_success=1，但 replay_success=0、success_step=0，符号关系成立不等于本节点包含的位姿条件全部成立。"
+     },
+     {
+      "t": 11.1,
+      "log_step": 204,
+      "text_zh": "推断：log step 194–204 明确重放最终交付的331行文件，结果为包装中心约(7.0880,3.1498,0.8599)米、倾斜0.31°、偏航141.01°，盘上终态与视频 8.5–11.1 s 的可见局部一致；评分器也确认331行、deterministic=1且符号目标成立。DAG记初始包装朝向误差0.4390613弧度且无对称豁免，交付文件复放终态与初态的转角上界约为(1.20+0.07+0.31)°=0.0276弧度，故终态距目标的全姿态误差至少约0.411弧度，仍大于0.3弧度，final=false。该估计取自交付文件的会话复放，非评分器原生误差字段；镜外操作是否曾短暂满足完整条件及其时间无法确定，intervals保留null。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：视频 1.5 s 后取放区域大多移出视野，视频 6.0–11.1 s 仅见包装和红盘的局部，夹爪不在画面内；虽交付脚本命令右夹爪闭合、搬运并张开，现有视频及无状态日志的输入不能确认实际对向夹持或按压接触，故接触方式未知。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "platter_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "platter_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "package_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "package_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_window",
+   "unit": "control_steps",
+   "threshold": 45,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "max_body_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_body_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_robot_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_body_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_body_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_joint_coordinate_drift",
+   "unit": "joint_coordinate",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "package_on_platter",
+   "t": 8.0,
+   "log_step": 204,
+   "where": "操作中",
+   "evidence_zh": "推断：视频 7.4–8.6 s 已可见包装落在红盘上的局部，之后直到视频 11.1 s 未见再次调整；log step 190 的交付脚本保持初始包装朝向，log step 204 的交付文件复放结果也基本保持该朝向。结合DAG的初态朝向误差，包装虽到盘上仍未满足完整目标位姿，且未修复。视频 8.0 s 仅作放置末段的导航锚点，不表示看到了松手瞬间。"
+  },
+  "unmet": {
+   "platter_at_target": true,
+   "package_on_platter": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 8.0,
+    "log_step": 204,
+    "evidence_zh": "推断：视频 8.5–11.1 s 的可见局部显示包装在盘上，评分器 symbolic_success=1；但log step 204 对交付文件复放测得包装仍近似初始朝向（偏航141.01°、倾斜0.31°，初始约142.21°、0.07°）。由DAG的初始目标朝向误差0.4390613弧度减去约0.0276弧度的朝向变化上界，终态目标朝向误差仍至少约0.411弧度，超过0.3弧度。此处为结合交付记录的姿态推断，裁边画面本身不能量出误差。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "认知",
+    "item": "感知",
+    "source_zh": "日志",
+    "evidence_zh": "观察：log step 211将目标图的位姿解释为“platter centre, pack's original yaw”；log step 190的交付脚本用初始四元数计算yaw_pack，并以同一末端朝向生成抓取、搬运及放置目标。task/task_dag.json:78、100–102记录包装无姿态对称豁免、初始目标朝向误差0.4390613弧度且容差为0.3弧度；log step 194–204对最终交付的331行文件复放测得包装偏航141.01°、倾斜0.31°，仍接近初始偏航142.21°、倾斜0.07°。推断：agent从目标图估计包装目标朝向时错误地沿用初始朝向，交付动作也保持了这一估计；按初末朝向变化上界约0.0276弧度，终态目标朝向误差仍至少约0.411弧度，说明这一错误估计导致上盘后朝向仍超差。此为日志与DAG初态事实结合的推断，非由裁边视频直接量得。"
+   },
+   {
+    "cat": "认知",
+    "item": "误以为完成",
+    "source_zh": "日志",
+    "evidence_zh": "观察：log step 194 的交付文件复放检查仅打印包装位置、倾斜、偏航和离盘中心距离，没有将全姿态与目标比较或读取原生45步联合静止验收；log step 210 只确认末45行命令相同，log step 211 随即自述“The task is complete”，并称目标采用包装原来的偏航角。推断：agent将居中平放及恒定命令当作完整成功而结束，未再修正超容差的目标朝向；run.wall_s=3053、time_limit_s=3600，结束时仍有约547秒墙钟预算。"
+   }
+  ],
+  "gap_zh": "观察：inputs/有完整视频和run.json，但无states.jsonl；oracle.json明确无参考视频。头部相机随躯干转动，主要取放及夹爪接触在画面外，末段仅见盘和包装局部，因此实际抓握/按压方式未知，包装完整位姿的历史成立时段无法确定。推断：红盘持续保持及包装终态朝向不合格均借助上下文和交付文件的会话复放记录；没有评分器逐物体误差、速度、漂移或静止窗口字段，不能精确验证其数值，也不能排除镜外短暂达标，故包装intervals=null，全部native_metrics.value=null；不将早期会话试验中的倾翻或抓取失败算作本次现象。"
+ },
+ "summary_zh": "推断：红盘保持初始达标位姿；包装被移到盘上，评分器symbolic_success=1，但最终仍近似初始朝向，结合DAG初始朝向误差推得包装全姿态超出容差，末态完成1/2节点。观察：agent在log step 211把居中平放和末尾恒定动作视为完成后结束，���评分器replay_success=0。推断：断点为香肠包装上盘的位姿验收，根因为将目标朝向错误估为初始朝向（认知·感知），以及误以为完成；操作和夹爪大多在镜外，无状态日志或原生误差字段，包装历史达标区间及实际接触未知。",
+ "conf": 6,
+ "review": {
+  "status": "agreed",
+  "rounds": 2,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "fix",
+    "conf": 6,
+    "summary_zh": "节点值（红盘保持、包装上盘final=false）、断点package_on_platter、现象位姿不准与视频/log一致；但根因漏掉了真正导致朝向错的原因：agent从目标图估计目标位姿时把包装目标偏航定为“原偏航”，与DAG的0.439弧度初始朝向误差矛盾，属对目标朝向的错误估计。",
+    "n_issues": 1,
+    "n_major": 1
+   },
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 2,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "节点值、断点、现象与根因（感知、误以为完成）与视频、log和评分器字段一致；第1轮已采纳的修改无新问题。裁边画面无法直接量朝向，标签已标为推断。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:51:12Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--ygkjwdq/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--ygkjwdq/04_review_r1"
+   },
+   {
+    "stage": "reconcile",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--ygkjwdq/05_reconcile_r1"
+   },
+   {
+    "stage": "traj_review",
+    "round": 2,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-privileged--ygkjwdq/04_review_r2"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "74feb2a01948c3ae08a08a5956f491032f95edc58a94de6a699bf011d2577198",
+   "video_sha256": "f22eb03f54eb035174ec8322625a2c1a5f29da8088d614ea69c56c16320c3240",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/inter_object_relation_satisfaction_ontop_small_scale/trajectories/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-standard--tuiw6bm.json

@@ -0,0 +1,299 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-standard--tuiw6bm",
+ "taskKey": "ispatialgym/inter_object_relation_satisfaction_ontop_small_scale",
+ "dag_sha256": "149991ba89630c9372ba8c97b9c9950caaf63efb5952e71b5bf7eb73fa2a6fa7",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "live_matches_replay": 1,
+   "finalized": 1,
+   "missing_log": 0,
+   "n_actions": 1900,
+   "sim_steps": 1900,
+   "max_steps": 1900,
+   "steps": 1855,
+   "elapsed_s": 2603.4,
+   "symbolic_success": 0,
+   "success_step": 0,
+   "video_rendered": 1,
+   "gpu_lost": 0,
+   "replay_error": 0,
+   "instance_ok": 1,
+   "sim_startups": 1,
+   "resumes": 0,
+   "resume_state_mismatch": 0,
+   "resume_max_tcp_dev_m": 0.0,
+   "resume_max_joint_dev": 0.0,
+   "gave_up": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 0.651,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.956,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-standard--tuiw6bm/replay.mp4",
+  "duration_s": 63.5,
+  "fps": 10.0,
+  "frames": 635,
+  "sha256": "4f084733e881239d51c3c5f30cffe217c356c0b13a5d431a5e28a5cd570222b4"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 1,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "platter_at_target": {
+    "method": "visual",
+    "intervals": [
+     [
+      0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 红盘平放在白色台面上；DAG记载该冻结初态的红盘位姿已达标。视频 0.0–21.0 s 的台面操作中，红盘相对台边的位置和长轴方向未见明显改变。"
+     },
+     {
+      "t": 48,
+      "log_step": null,
+      "text_zh": "观察：视频 46.0–49.0 s 红盘重新完整出现在台面上，仍平放在原工作区，盘内为空；视频 48.0 s 包装在盘前方的地板上。"
+     },
+     {
+      "t": 63.4,
+      "log_step": 186,
+      "text_zh": "推断：结合视频 0.0–21.0 s、46.0–49.0 s 红盘未见被移走，及 log step 161–186 机器人针对地板包装执行抓取和向盘上搬运的记录，红盘目标位姿按保持处理。视频 49.5–63.4 s 未显示红盘，终段精确容差没有逐物体检查器读数，末值为上下文推断。"
+     }
+    ]
+   },
+   "package_on_platter": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 包装在红盘左侧的台面上，未在盘内。视频 2.5–8.5 s 左夹爪靠近、闭合和抬起后，包装仍留在台面，未随夹爪完成拿起。"
+     },
+     {
+      "t": 18.4,
+      "log_step": 77,
+      "text_zh": "观察：视频 14.0–18.1 s 包装仍由盘旁台面承托；视频 18.2 s 左夹爪抬起时包装没有随爪上升，而在视频 18.3–18.6 s 越过台边下落。红盘始终为空。"
+     },
+     {
+      "t": 48,
+      "log_step": 157,
+      "text_zh": "观察：视频 48.0 s 包装清楚可见于机器人与台面之间的木地板，红盘仍为空；此前的地板恢复操作没有把包装送上盘。agent 在 log step 157 也自述此前侧向抓取在包装上方闭合。"
+     },
+     {
+      "t": 63.4,
+      "log_step": 189,
+      "text_zh": "推断：视频 48.0 s 包装仍在地板，视频 49.5–63.4 s 的头部视野没有提供其成功上盘的证据；评分器终态 symbolic_success=0、live_success=0、replay_success=0，支持上盘关系最终未成立。log step 177 的“已抬起”仅为 agent 自述，无法从所提供视频核实，不据此计入上盘区间。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "推/扫",
+      "note_zh": "观察：视频 2.5–8.5 s、14.0–18.6 s 左夹爪接近并闭合时拨动盘旁包装，包装沿台面移动并最终掉出台边；未见有效对向夹持后稳定随爪抬起，故不把闭爪命令记成实际抓握。后段地板拾取的接触方式被头部视角遮挡。"
+     },
+     {
+      "who": "底盘",
+      "how": "支撑",
+      "note_zh": "观察：视频 20.0–21.0 s、46.0–49.0 s 机器人在包装上盘步骤的地板恢复过程中移动底盘、后退及重新靠近台面；log step 87、149、156 分别记录 back、back_more、forward 的执行。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "platter_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "platter_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "package_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "package_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "settle_window",
+   "unit": "control_steps",
+   "threshold": 45,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "max_body_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_body_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_robot_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_body_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_body_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_joint_coordinate_drift",
+   "unit": "joint_coordinate",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "package_on_platter",
+   "t": 6.5,
+   "log_step": 50,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 6.5–8.5 s 第一次闭爪后的抬臂未带起包装，包装仍在盘旁；视频 18.2–18.6 s 再次抬爪后包装滑出台边，视频 48.0 s 仍在地板，始终未观察到上盘完成。评分器终态 symbolic_success=0；log step 189 的机器人服务报告控制步耗尽结束，未修复该步骤。"
+  },
+  "unmet": {
+   "platter_at_target": true,
+   "package_on_platter": false
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没建立接触",
+    "t": 6.5,
+    "log_step": 50,
+    "evidence_zh": "观察：视频 2.5–8.5 s 左夹爪虽闭合并抬起，包装仍在台面；视频 14.0–18.1 s 夹爪再次围近包装，视频 18.2 s 抬爪时包装仍贴台面并随后滑落，未显示形成能带起包装的夹持接触。这里指所需的有效夹持未建立，已有的拨动接触不等于抓稳。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "source_zh": "视频",
+    "evidence_zh": "推断：视频 2.5–8.5 s 和14.0–18.6 s 机器人左夹爪已到包装附近并执行闭合、抬升，但实际接触只拨动包装，未完成稳定拿起，支持抓取执行层面的控制问题。log step 48 的夹爪开口为0，亦不能代替包装随爪抬起的视觉证据；缺少接触力和清晰指间视图，不能进一步区分接触设计、跟踪或反馈子类。"
+   }
+  ],
+  "gap_zh": "观察：inputs/ 无逐帧状态日志，成功参考 inputs/oracle/oracle.json 的 video_kind=none。所提供回放只有头部视角，视频 22.0–27.0 s 被机器人近处部件遮挡，视频 28.0–45.0 s、49.5–63.4 s 多朝向室内而看不到操作区；已逐帧查看视频 62.4–63.4 s，仍看不到夹爪、包装和红盘。后段地板拾取是否曾形成夹持、最后放置的具体落点与红盘终态精确位姿无法直接核实；红盘保持与包装最终未达标结合可见上下文和评分器字段推断。各原生位姿误差与静止指标均未报告，未编造数值。"
+ },
+ "summary_zh": "红盘开局已达目标位姿，视频可见期间未见移位，按持续保持计入；香肠包装未观察到上盘，最终进度为1/2。左夹爪多次闭合未建立有效夹持，视频 18.2–18.6 s 包装滑出台边，视频 48.0 s 仍在地板；失败定位为包装上盘步骤的抓取控制问题。后半段操作区长期不在头部视野内，没有状态日志及逐物体终态误差，红盘末值及包装末值有上下文和评分器推断。",
+ "conf": 6,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "抽查视频 0/5/8/18/18.5/48 s：包装开局在盘旁台面，左夹爪闭合抬起未带起，18.5 s 包装已落到地板，48 s 在地板上、红盘仍在台面且为空，与标注一致；断点、现象和根因有证据支撑且标注了不确定性，未发现需要改动的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:39:46Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-standard--tuiw6bm/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-ontop-small-scale-i00-standard--tuiw6bm/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "7fe2e34ea103938fc221a4606746f6d71d65f49690a4b6cd24cb2f94dd1c3342",
+   "video_sha256": "4f084733e881239d51c3c5f30cffe217c356c0b13a5d431a5e28a5cd570222b4",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/inter_object_relation_satisfaction_touching_small_scale/trajectories/ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-privileged--dnei5mk.json

@@ -0,0 +1,243 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-privileged--dnei5mk",
+ "taskKey": "ispatialgym/inter_object_relation_satisfaction_touching_small_scale",
+ "dag_sha256": "28d0c1ba95380927c7f64c8a65a012a975a7dd662ef095211d362d931e923cef",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 90,
+   "sim_steps": 90,
+   "rows": 90,
+   "max_rows": 3142,
+   "replay_success": 0,
+   "instance_ok": 1,
+   "steps": 90,
+   "success_step": 0,
+   "symbolic_success": 0,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 12.532,
+   "peak_memory_gb": 9.746,
+   "peak_processes": 20,
+   "peak_disk_gb": 3.092,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-privileged--dnei5mk/replay.mp4",
+  "duration_s": 3.1,
+  "fps": 10.0,
+  "frames": 31,
+  "sha256": "28b400722b24748c3e96d98720f8ea89895308f959755c41b5aa119539bc2feb"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 1,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "tissue_box_at_goal": {
+    "method": "visual",
+    "intervals": [
+     [
+      0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s，蓝色纸巾盒放在白色桌面左侧；逐帧检查视频 0.0–3.0 s，蓝盒没有可见平移、转动或离开桌面，末帧视频 3.0 s 仍在原处。"
+     },
+     {
+      "t": 3.0,
+      "log_step": null,
+      "text_zh": "推断：DAG 指明蓝盒初态已在目标容差内，且本次视频 0.0–3.0 s 始终未见其位姿改变，故纸巾盒就位从视频 0.0 s 持续至结束；无状态日志，不能独立重算毫米级位姿误差。"
+     }
+    ]
+   },
+   "napkin_box_touching_tissues": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s，橙白色卫生巾盒在蓝盒右侧，两盒侧面间有清楚可见的桌面缝隙。视频 0.0–3.0 s 的全部31帧中，两盒保持这一分离排列，没有夹爪进入盒间或推动卫生巾盒。"
+     },
+     {
+      "t": 3.0,
+      "log_step": 86,
+      "text_zh": "观察：视频 3.0 s 末帧仍有明显盒间缝隙；log step 86 检查交付数组得到 shape=(90,23)、constant=True，元数据为 hold stationary for 90 control steps，与本次视频的未搬动盒子相符。成功参考视频约 7.0–8.5 s 则可见橙白盒已紧靠蓝盒侧面；参考只用于辨认相贴外观。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "max_target_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_target_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_robot_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_joint_drift",
+   "unit": "joint_unit",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "napkin_box_touching_tissues",
+   "t": 0.0,
+   "log_step": 86,
+   "where": "首步前",
+   "evidence_zh": "观察：本次视频 0.0–3.0 s 的两盒始终分开，机器人没有启动使橙白盒移向蓝盒的动作；log step 86 和 log step 97 确认最终交付90行静止轨迹。相贴步骤根本未开始，蓝盒已有的就位状态则未被破坏。"
+  },
+  "phenomena": [
+   {
+    "cat": "没尝试",
+    "item": "没尝试",
+    "t": 0.0,
+    "log_step": 86,
+    "evidence_zh": "观察：视频 0.0–3.0 s 全部31帧中，卫生巾盒没有向蓝盒��动，盒间缝隙一直可见，未见夹爪接近或接触两盒；log step 86 的交付动作数组90行完全相同，计划字段明确为保持静止。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "认知",
+    "item": "感知",
+    "log_step": 62,
+    "evidence_zh": "推断：agent 在 log step 62 自述两盒已有相同左右顺序、中心仅约15 cm，因而决定保持静止，并将目标图与初态的主要差别归为相机角度和机器人姿态；log step 86 显示这一判断落实为90行静止交付。本次视频 0.0–3.0 s 可见盒间缝隙，成功参考视频约 7.0–8.5 s 可见两盒侧面紧靠，说明 agent 对目标与当前盒间关系的视觉比较遗漏了需消除的间隙，导致没有启动相贴动作。",
+    "source_zh": "agent 自述"
+   }
+  ],
+  "gap_zh": "无逐帧状态日志，评分器未报告DAG所列8项连续指标，故这些数值均为null；不能从录像精确核验完整四元数误差及45控制步窗口的微小速度、漂移阈值。相贴未达由可见缝隙判定，感知根因由log step 62的比较判断与交付行为推断。",
+  "unmet": {
+   "tissue_box_at_goal": true,
+   "napkin_box_touching_tissues": false
+  }
+ },
+ "summary_zh": "观察：本次视频 0.0–3.0 s 保留了初态已就位的蓝色纸巾盒，但橙白盒始终与其有明显缝隙，最终仅1/2节点成立，相贴步骤在首步前未启动。推断：agent 在 log step 62 将场景差异主要归于相机姿态，随后交付90行静止动作，支持视觉关系判断错误的感知根因。无状态日志及连续指标读数，精细位姿与静止窗口无法独立量化。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "标注与证据一致：视频0–3 s两盒始终有缝隙且无机器人动作，log step 86/97 确认交付为90行静止轨迹；蓝盒就位从0起成立，final=1/2，断点首步前/没尝试合理，感知根因已标为推断。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:25:21Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-privileged--dnei5mk/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-privileged--dnei5mk/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "476ef90cf18445cb4a9e23acd89fe52b1cce16eeae6be5b8f18a7a59792414ea",
+   "video_sha256": "28b400722b24748c3e96d98720f8ea89895308f959755c41b5aa119539bc2feb",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/inter_object_relation_satisfaction_touching_small_scale/trajectories/ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-privileged--iztuvbl.json

@@ -0,0 +1,234 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-privileged--iztuvbl",
+ "taskKey": "ispatialgym/inter_object_relation_satisfaction_touching_small_scale",
+ "dag_sha256": "28d0c1ba95380927c7f64c8a65a012a975a7dd662ef095211d362d931e923cef",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 290,
+   "sim_steps": 273,
+   "rows": 290,
+   "max_rows": 3142,
+   "replay_success": 1,
+   "instance_ok": 1,
+   "steps": 273,
+   "success_step": 273,
+   "symbolic_success": 1,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 16.979,
+   "peak_memory_gb": 10.646,
+   "peak_processes": 26,
+   "peak_disk_gb": 1.206,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-privileged--iztuvbl/replay.mp4",
+  "duration_s": 9.2,
+  "fps": 10.0,
+  "frames": 92,
+  "sha256": "f70dbbd9ab511acc8575c5cabd97c79829bf87d4b67072657c6557d6b76cf4b9"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 1,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "tissue_box_at_goal": {
+    "method": "visual",
+    "intervals": [
+     [
+      0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s，蓝色纸巾盒已放在白色桌面左侧；视频 0.0–9.1 s 的抽帧及视频 3.8–5.7 s、8.1–9.1 s 的逐帧检查均未见它被夹爪直接触碰、抬离桌面或明显挪动、转歪。"
+     },
+     {
+      "t": 9.1,
+      "log_step": null,
+      "text_zh": "推断：蓝盒从初态保持原位至视频 9.1 s；结合 DAG 所述初态已达标及评分器 reward 中 success=1、replay_success=1、symbolic_success=1、deterministic=1，判定其目标位姿全程保持。视频不能独立量出完整三维位置和朝向误差。"
+     }
+    ]
+   },
+   "napkin_box_touching_tissues": {
+    "method": "visual",
+    "intervals": [
+     [
+      5.0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s，橙白色卫生巾盒在蓝色纸巾盒右侧，两盒之间有明显间隔。"
+     },
+     {
+      "t": 4.1,
+      "log_step": 299,
+      "text_zh": "观察：视频 3.8–4.0 s 右夹爪接近橙白盒，视频约 4.1–5.2 s 夹爪指面从盒子右侧推动它，盒子沿桌面向蓝盒滑动并转正，没有被夹起；log step 299 的交付生成脚本记录了右臂推盒、抬手、回撤、保持并保存轨迹的顺序，与视频一致。"
+     },
+     {
+      "t": 5.0,
+      "log_step": null,
+      "text_zh": "观察：视频 4.8–5.2 s，两盒间距收拢至侧面相贴的外观；以视频约 5.0 s 为达成锚点，存在约 0.2 s 的视觉不确定性。视频 5.3–6.5 s 机器人抬起右夹爪并退开，橙白盒留在桌面蓝盒右侧。"
+     },
+     {
+      "t": 9.1,
+      "log_step": null,
+      "text_zh": "推断：视频 8.1–9.1 s 逐帧显示两盒维持相贴外观且无明显运动；评分器 reward 中 success=1、replay_success=1、symbolic_success=1，支持最终目标关系、两盒位姿及共同静止验收通过。缺少状态日志，不能从像素独立核验毫米级间距、完整朝向及45控制步内的速度和漂移。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "推/扫",
+      "note_zh": "观察：视频约 4.1–5.2 s，机器人右夹爪的指面接触橙白色卫生巾盒右侧，将它沿桌面推向蓝色纸巾盒；没有双指包夹盒体或抬起搬运，随后右夹爪抬起退开。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "max_target_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_target_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_robot_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_joint_drift",
+   "unit": "joint_unit",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "蓝色纸巾盒从视频 0.0 s 起保持桌面目标位姿；机器人用右夹爪推橙白盒，约在视频 5.0 s 达成两盒相贴，退手后保持至视频 9.1 s，最终完成2/2节点。评分器记录 success=1、replay_success=1、symbolic_success=1、deterministic=1，本次无失败断点。没有状态日志，精确位姿误差和共同45控制步静止条件由成功评分支持而非逐项实测；DAG所列八项原生连续指标在评分器字段中均未报告，故值为null。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "成功运行，两节点标注与视频一致：蓝盒全程未动，右夹爪约 4.1–5.2 s 推动橙白盒至相贴，contact_actual 为 手指/夹爪·推/扫，无失败断点，final=2/2 有评分器字段支持。未发现需改内容的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:32:28Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-privileged--iztuvbl/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-privileged--iztuvbl/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "09302ff0e74d668c58e33dd7a96aaa8e6f4b8bbd54aadceda867cd31b62fedd2",
+   "video_sha256": "f70dbbd9ab511acc8575c5cabd97c79829bf87d4b67072657c6557d6b76cf4b9",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/inter_object_relation_satisfaction_touching_small_scale/trajectories/ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-privileged--jcdhyag.json

@@ -0,0 +1,247 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-privileged--jcdhyag",
+ "taskKey": "ispatialgym/inter_object_relation_satisfaction_touching_small_scale",
+ "dag_sha256": "28d0c1ba95380927c7f64c8a65a012a975a7dd662ef095211d362d931e923cef",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 440,
+   "sim_steps": 396,
+   "rows": 440,
+   "max_rows": 3142,
+   "replay_success": 1,
+   "instance_ok": 1,
+   "steps": 396,
+   "success_step": 396,
+   "symbolic_success": 1,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 1,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 32.623,
+   "peak_memory_gb": 18.227,
+   "peak_processes": 31,
+   "peak_disk_gb": 1.942,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "over_gpu_memory_gb": 1
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-privileged--jcdhyag/replay.mp4",
+  "duration_s": 13.3,
+  "fps": 10.0,
+  "frames": 133,
+  "sha256": "6167dbda10895b8b33cb01d3eca3d4a7cdb5752837c934aeb1fa91ad56da7a01"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 1,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "tissue_box_at_goal": {
+    "method": "visual",
+    "intervals": [
+     [
+      0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 蓝色纸巾盒已在桌面左侧，橙白色卫生巾盒与它有间隙；视频全程蓝盒未离开桌面。视频约 6.5–7.5 s 橙白盒被右夹爪推来后，蓝盒有小幅平移、转动，未见明显偏离目标区域或倾倒。"
+     },
+     {
+      "t": 13.2,
+      "log_step": null,
+      "text_zh": "观察：视频 12.0–13.2 s 蓝盒仍直立于桌面、位于橙白盒左侧，终态布局与成功参照的相贴布局一致；本次评分器 reward.success=1、replay_success=1、symbolic_success=1、deterministic=1。"
+     },
+     {
+      "t": 13.2,
+      "log_step": null,
+      "text_zh": "推断：结合初态已就位、全程仅小幅受推及评分器成功验收，蓝盒目标位姿从视频 0.0 s 持续成立；无状态日志，不能逐帧直接测量三维位置及完整朝向误差。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "推/扫",
+      "note_zh": "观察：视频约 6.5–7.5 s 右侧张开的夹爪推橙白色卫生巾盒，橙白盒与蓝色纸巾盒相贴后带着蓝盒小幅移动；作用链为右夹爪→橙白盒→蓝盒，未见夹爪直接夹持蓝盒。"
+     }
+    ]
+   },
+   "napkin_box_touching_tissues": {
+    "method": "visual",
+    "intervals": [
+     [
+      6.5,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 两盒之间有明显间隙，橙白色卫生巾盒相对蓝盒转斜；视频 4.0–6.0 s 机器人右侧张开的夹爪在橙白盒右侧向左推，橙白盒在桌面平移并转正，未被夹起。"
+     },
+     {
+      "t": 6.5,
+      "log_step": null,
+      "text_zh": "观察：视频 6.1–6.7 s 两盒侧面的可见间隙逐渐消失，视频约 6.5 s 橙白盒已侧贴蓝盒，随后右夹爪继续小幅推移两盒；首次相贴时间按图像估计，约有 0.3 s 不确定性。"
+     },
+     {
+      "t": 9.6,
+      "log_step": null,
+      "text_zh": "观察：视频 8.0–9.6 s 右夹爪向右撤开并抬起，两盒均留在桌面原处；视频 10.0–13.2 s 两盒保持蓝盒在左、橙白盒在右的相贴关系，未见重新分离或倒下。"
+     },
+     {
+      "t": 13.2,
+      "log_step": null,
+      "text_zh": "推断：视频约 6.5 s 起可见相贴关系持续到末帧；评分器 reward.success=1、success_step=396、symbolic_success=1 进一步确认终态两盒关系、目标位姿及共同45控制步静止窗口通过验收。无状态日志，精确接触间隙和静止阈值由评分结果支持，未从agent会话中的自测数字替代视频判断。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "推/扫",
+      "note_zh": "观察：视频约 4.3–7.5 s 机器人右侧夹爪保持张开，以指部侧面推动橙白色卫生巾盒沿桌面向蓝盒滑动并调整朝向；视频 8.0–9.6 s 夹爪撤开，未见两指合拢夹住盒体或将其提起。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "max_target_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_target_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_robot_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_joint_drift",
+   "unit": "joint_unit",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "本次评分回放成功：蓝色纸巾盒从视频 0.0 s 起保持就位，机器人用张开的右夹爪将橙白色卫生巾盒推到蓝盒旁，约视频 6.5 s 起两盒相贴，最终两个节点均成立。机器人撤手并回位后，两盒至视频 13.2 s 仍保持相贴，评分器确认目标关系、位姿及共同静止验收成功，且确定性检查通过。没有状态日志，首次相贴时间约有 0.3 s 误差，毫米级间隙、完整三维位姿与速度窗口无法逐帧独立测量；评分器未报告八项原生指标的具体值。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "两节点（蓝盒就位自0 s、橙白盒相贴约6.5 s）、终值、接触（右夹爪推/扫，间接推蓝盒）与评分器成功一致，视频帧复核未见错误；无失败层，合理。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:48:13Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-privileged--jcdhyag/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-privileged--jcdhyag/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "40fcd703b70f5c9aee363f961347df506454423c9ecdba4d9ca15eb7a5ca2182",
+   "video_sha256": "6167dbda10895b8b33cb01d3eca3d4a7cdb5752837c934aeb1fa91ad56da7a01",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/inter_object_relation_satisfaction_touching_small_scale/trajectories/ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-privileged--jdxxgbt.json

@@ -0,0 +1,246 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-privileged--jdxxgbt",
+ "taskKey": "ispatialgym/inter_object_relation_satisfaction_touching_small_scale",
+ "dag_sha256": "28d0c1ba95380927c7f64c8a65a012a975a7dd662ef095211d362d931e923cef",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 385,
+   "sim_steps": 351,
+   "rows": 385,
+   "max_rows": 3142,
+   "replay_success": 1,
+   "instance_ok": 1,
+   "steps": 351,
+   "success_step": 351,
+   "symbolic_success": 1,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 1,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.0,
+   "peak_processes": 0.0,
+   "peak_disk_gb": 0.0,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-privileged--jdxxgbt/replay.mp4",
+  "duration_s": 11.8,
+  "fps": 10.0,
+  "frames": 118,
+  "sha256": "56cd3185695522467733338248dc9151748e652328cf81bb8a4e99177f600401"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 1,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "tissue_box_at_goal": {
+    "method": "visual",
+    "intervals": [
+     [
+      0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 蓝色纸巾盒位于白色桌面左侧，与右侧橙白色卫生巾盒分开；结合DAG所列已达标初态，本节点从开局计入。"
+     },
+     {
+      "t": 6.8,
+      "log_step": null,
+      "text_zh": "观察：视频 5.8–8.3 s 的逐帧局部图显示机器人夹爪从蓝盒左侧推它向右滑动；蓝盒始终由桌面支撑，未被夹起，也未见明显转歪。"
+     },
+     {
+      "t": 11.7,
+      "log_step": null,
+      "text_zh": "推断：视频 10.7–11.7 s 蓝盒与橙白盒相贴并保持不动；run.json 的评分器字段 success=1、replay_success=1、symbolic_success=1、deterministic=1 支持终态位姿通过原生验收。结合蓝盒仅在桌面短距离平移且无明显转动，将其目标位姿判为全程保持；视频无法直接量出精确三维误差。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "推/扫",
+      "note_zh": "观察：视频 5.8–8.3 s 机器人单夹爪从蓝色纸巾盒左侧推它向橙白盒滑动，没有两指包夹蓝盒；log step 65 的生成脚本将此作用端标为左夹爪。随后夹爪向左并抬高退开。"
+     }
+    ]
+   },
+   "napkin_box_touching_tissues": {
+    "method": "visual",
+    "intervals": [
+     [
+      7.3,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 两盒均在桌面上，蓝盒在左、橙白盒在右，中间有清楚可见的缝隙，尚未相贴。"
+     },
+     {
+      "t": 7.3,
+      "log_step": null,
+      "text_zh": "观察：视频 5.8–7.7 s 逐帧局部图中，机器人单夹爪推动蓝盒向右，逐渐消除两盒之间的间隙；视频约 7.3 s 两盒侧面相贴，蓝左橙白右，均未离开桌面。首次相贴时间为视觉估计，误差约0.3 s。"
+     },
+     {
+      "t": 11.7,
+      "log_step": null,
+      "text_zh": "观察：视频 8.3–10.0 s 机器人夹爪撤离两盒；视频 10.7–11.7 s 逐帧检查，两盒仍在同一桌面位置相贴，未见滑开或倾倒。"
+     },
+     {
+      "t": 11.7,
+      "log_step": null,
+      "text_zh": "推断：run.json 的评分器字段 success=1、replay_success=1、symbolic_success=1、deterministic=1，结合视频末尾相贴且静止的画面，确认本节点终态通过目标关系、位姿及共同静止验收；不把log step 69 的会话内自测坐标当作本次重放的状态日志。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "推/扫",
+      "note_zh": "观察：视频 5.8–8.3 s 单夹爪直接推蓝色纸巾盒左侧，使蓝盒在桌面向右滑到橙白色卫生巾盒旁相贴；夹爪未直接抓握或搬运橙白盒。log step 65 的生成脚本标明使用左夹爪；这是推动伙伴蓝盒达成两盒相贴的实际方案。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "max_target_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_target_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_robot_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_joint_drift",
+   "unit": "joint_unit",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "蓝色纸巾盒的目标位姿从视频 0.0 s 保持，机器人左夹爪推动蓝盒靠近橙白盒，约视频 7.3 s 达成两盒相贴，最终两个节点均成立（2/2）。视频 10.7–11.7 s 两盒保持相贴，评分器报告 success=1、replay_success=1、deterministic=1，本次无失败断点。缺少状态日志和逐项原生指标数值，无法从画面独立核验微小朝向误差与45控制步静止阈值；终态精确验收依据评分器成功结果，相贴时刻为视觉估计。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "成功运行，两节点与相贴时刻（约视频7.3 s）、推/扫接触与无失败断点均与帧一致，评分器success=1支持终态。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:33:44Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-privileged--jdxxgbt/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-privileged--jdxxgbt/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "b5291e5ae74dbc6418cb089670e8f39c62cb66fecc0bf27fb1212c9fd8a2f0cf",
+   "video_sha256": "56cd3185695522467733338248dc9151748e652328cf81bb8a4e99177f600401",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/inter_object_relation_satisfaction_touching_small_scale/trajectories/ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-privileged--jieb4lv.json

@@ -0,0 +1,247 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-privileged--jieb4lv",
+ "taskKey": "ispatialgym/inter_object_relation_satisfaction_touching_small_scale",
+ "dag_sha256": "28d0c1ba95380927c7f64c8a65a012a975a7dd662ef095211d362d931e923cef",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 70,
+   "sim_steps": 70,
+   "rows": 70,
+   "max_rows": 3142,
+   "replay_success": 0,
+   "instance_ok": 1,
+   "steps": 70,
+   "success_step": 0,
+   "symbolic_success": 0,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 1,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.0,
+   "peak_processes": 0.0,
+   "peak_disk_gb": 0.0,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-privileged--jieb4lv/replay.mp4",
+  "duration_s": 2.5,
+  "fps": 10.0,
+  "frames": 25,
+  "sha256": "062c553e22be8a850547eed1c6e08a6e3ee399b97a68fbcd332d7e685f7ff34d"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 1,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "tissue_box_at_goal": {
+    "method": "visual",
+    "intervals": [
+     [
+      0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0,
+      "log_step": null,
+      "text_zh": "观察：视频0.0 s中蓝色纸巾盒平放在白色桌面左侧；逐帧检查视频0.0–2.4 s，盒体未离开原处、未明显转动或离开桌面。"
+     },
+     {
+      "t": 2.4,
+      "log_step": 134,
+      "text_zh": "推断：蓝盒初态已满足DAG所列目标位姿，视频2.4 s仍保持该初态，因此纸巾盒就位从视频0.0 s持续成立；log step 134确认agent用70行保持初态的文件覆盖了会话推盒试验。精确三维误差未由评分器报告。"
+     }
+    ]
+   },
+   "napkin_box_touching_tissues": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0,
+      "log_step": null,
+      "text_zh": "观察：视频0.0 s中橙白色卫生巾盒在蓝盒右侧，两盒侧面之间有可见桌面间隙。"
+     },
+     {
+      "t": 2.4,
+      "log_step": null,
+      "text_zh": "观察：逐帧检查视频0.0–2.4 s（包括视频1.5–2.4 s最后一秒），两盒位置和间隙持续不变，没有夹爪进入两盒周围实施操作；视频2.4 s仍未相贴。成功参考录像在参考视频9.1 s显示两盒侧面紧邻，本段录像缺少该可见结果。"
+     },
+     {
+      "t": null,
+      "log_step": 134,
+      "text_zh": "观察：log step 134中agent运行保存静止回退轨迹的脚本，输出为70×23且末行等于首行；log step 136说明已覆盖183行推盒试验。评分器字段n_actions=70、deterministic=1、replay_success=0，与视频中的保持初态一致。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "max_target_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_target_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_robot_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_joint_drift",
+   "unit": "joint_unit",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "napkin_box_touching_tissues",
+   "t": 0,
+   "log_step": 134,
+   "where": "首步前",
+   "evidence_zh": "观察：视频0.0–2.4 s中两盒始终分离，机器人未开始移盒；log step 134把会话推盒轨迹覆盖为70行保持初态，log step 135–136确认最终交付该静止方案。两盒相贴这一步在评分轨迹中未启动，且未在后续修复。"
+  },
+  "unmet": {
+   "tissue_box_at_goal": true,
+   "napkin_box_touching_tissues": false
+  },
+  "phenomena": [
+   {
+    "cat": "没尝试",
+    "item": "没尝试",
+    "t": 0,
+    "log_step": 134,
+    "evidence_zh": "观察：视频0.0–2.4 s逐帧可见橙白盒和蓝盒均停在原处，侧面间隙始终存在，机器人没有向盒子伸手或推近；log step 134保存的是保持初态文件。会话中另一次推盒试验没有出现在该评分轨迹中。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "source_zh": "日志",
+    "evidence_zh": "推断：agent未能把已识别的推盒目标实现为有效控制，故最终评分轨迹退回静止初态。log step 108明确知道两盒需推近；会话生成脚本在log step 118、120–121修正关节插值的维度和索引错误，log step 124–125又暴露间隙坐标计算错误；最后一轮会话试验的log step 133输出显示间隙从0.0459 m变为0.0473 m，蓝盒世界Y坐标从约3.085 m变为3.2378 m。log step 134明确因该试验把两盒横向带偏而覆盖成静止回退文件。上述是会话控制证据，不能计为评分轨迹的接触或失败现象；尚不足以区分接触设计与轨迹跟踪子类。"
+   }
+  ],
+  "gap_zh": "没有状态日志；视频能判定两盒始终有间隙及蓝盒保持初态，但无法精确核验完整三维位姿误差、每步速度和共同45控制步静止窗口。评分器未报告DAG列出的八项原生指标，故其value均为null。控制根因来自会话脚本和试验输出，无法进一步确定最后一次试验偏移的具体接触机制。"
+ },
+ "summary_zh": "评分轨迹保持初态，蓝色纸巾盒就位从视频0.0 s持续至末帧视频2.4 s，最终1/2节点达成；橙白盒与蓝盒始终有明显间隙，两盒相贴这一步未启动。agent在会话推盒控制未奏效后改交70行静止回退轨迹（log step 133–136），断点为两盒相贴／首步前，现象为没尝试。无状态日志和原生误差、速度读数，细粒度验收及控制子类无法核实。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "标注与视频、log一致：评分轨迹为70行静止初态，蓝盒就位从0 s成立，两盒始终有间隙；断点首步前/没尝试，根因控制来自会话试验证据，符合FAIL-22。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:19:14Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-privileged--jieb4lv/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-privileged--jieb4lv/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "57417ad86454190ed9f2b45a02cddb2a146302f1447dd06feeef455f3b607d7d",
+   "video_sha256": "062c553e22be8a850547eed1c6e08a6e3ee399b97a68fbcd332d7e685f7ff34d",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/inter_object_relation_satisfaction_touching_small_scale/trajectories/ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-privileged--sgdfrsl.json

@@ -0,0 +1,266 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-privileged--sgdfrsl",
+ "taskKey": "ispatialgym/inter_object_relation_satisfaction_touching_small_scale",
+ "dag_sha256": "28d0c1ba95380927c7f64c8a65a012a975a7dd662ef095211d362d931e923cef",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "deterministic": 1,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "n_actions": 2330,
+   "sim_steps": 776,
+   "rows": 2330,
+   "max_rows": 3142,
+   "replay_success": 1,
+   "instance_ok": 1,
+   "steps": 776,
+   "success_step": 776,
+   "symbolic_success": 1,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 17.021,
+   "peak_memory_gb": 9.731,
+   "peak_processes": 21,
+   "peak_disk_gb": 1.99,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-privileged--sgdfrsl/replay.mp4",
+  "duration_s": 26.0,
+  "fps": 10.0,
+  "frames": 260,
+  "sha256": "1cc5a07547affb09543aef0359b7d4d1183e6329ab3ea2515192f94b77fc5c14"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 1,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "tissue_box_at_goal": {
+    "method": "visual",
+    "intervals": [
+     [
+      0.0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频0.0 s，蓝色纸巾盒平放在白色桌面左侧，橙白色卫生巾盒在其右方且留有明显缝隙；蓝盒初态位置和朝向与成功参考录像的蓝盒相符。"
+     },
+     {
+      "t": 23.0,
+      "log_step": null,
+      "text_zh": "观察：视频0.0–23.0 s，机器人右夹爪只抓取、抬起和移动右侧橙白盒；蓝盒始终留在桌面原区域，未被提起或明显转歪，橙白盒移到它旁边贴近后也没有把它推离目标区域。"
+     },
+     {
+      "t": 25.9,
+      "log_step": null,
+      "text_zh": "推断：视频25.0–25.9 s逐帧可见蓝盒仍平放、两盒位置稳定，结合inputs/run.json评分器的success=1、replay_success=1、symbolic_success=1，蓝盒最终位姿达标；全过程没有可见的超容差位移，区间记为从视频0.0 s持续至结束。"
+     }
+    ]
+   },
+   "napkin_box_touching_tissues": {
+    "method": "visual",
+    "intervals": [
+     [
+      19.5,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频0.0 s，两盒之间有可见桌面缝隙，橙白盒相对蓝盒带偏角，尚未相贴。"
+     },
+     {
+      "t": 16.0,
+      "log_step": 71,
+      "text_zh": "观察：视频12.0–14.0 s，机器人右二指夹爪在橙白盒两侧闭合；视频15.0–17.0 s盒体随夹爪抬高，原桌面位置空出，随后在视频17.5–20.0 s向蓝盒移动并转正。log step 71的会话工具输出报告右手建立抓握，与录像中的实际抬起相符；接触为单右夹爪抓握。"
+     },
+     {
+      "t": 19.5,
+      "log_step": null,
+      "text_zh": "观察：视频17.0 s橙白盒与蓝盒仍有可见缝隙且带偏角，视频18.0 s两盒靠近，视频19.0–20.0 s橙白盒已转正并到达蓝盒右侧，侧面贴近、原来的明显缝隙消失；视频20.0–23.5 s两盒保持并排贴近，没有可见的再次分离，右夹爪留在橙白盒顶部并遮挡部分盒体。"
+     },
+     {
+      "t": 19.5,
+      "log_step": null,
+      "text_zh": "推断：结合视频19.0–20.0 s的并排相贴外观、视频20.0–25.9 s两盒没有明显再次分离，以及inputs/run.json评分器报告success=1、replay_success=1、symbolic_success=1，首次达标近似记在视频19.5 s，视觉定时误差约1 s；精确接触距离、桌面支撑与完整朝向误差无法从单头部画面直接测量。"
+     },
+     {
+      "t": 25.9,
+      "log_step": null,
+      "text_zh": "观察：视频25.0–25.9 s逐帧中，两盒保持并排相贴，右夹爪仍靠近橙白盒，尚未退离；评分器字段success_step=776、steps=776、rows=2330，说明这段评分录像在交付文件全部动作执行完之前已成功结束。log step 117、135记录的2330步会话自测终态发生在另一段完整重放中，不用其后续退手、底盘后退或停车状态延长本次录像的判断。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "观察：视频约12.0–14.0 s机器人右平行二指夹爪包夹橙白色卫生巾���，视频15.0–20.0 s抓持盒体抬起、转正并移向蓝色纸巾盒，视频约19.0–20.0 s橙白盒到达蓝盒右侧并贴近，之后至视频25.9 s两盒保持并排，右夹爪仍留在橙白盒顶部；左夹爪没有同时操作目标，评分录像中未出现底盘行驶。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "max_target_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_target_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_robot_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_joint_drift",
+   "unit": "joint_unit",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "蓝色纸巾盒从视频0.0 s起保持就位，机器人用右夹爪抓起橙白色卫生巾盒、转正并移到蓝盒右侧，约在视频19.5 s形成并排相贴状态，最终达成2/2节点。评分器报告成功并在776个控制步结束，本次视频25.9 s末帧中右夹爪仍靠近盒体；log step 117、135所述2330步完整会话自测的退手和停车终态不属于这段已提前结束的评分录像。无状态日志，部分接触区域被夹爪遮挡，相贴起点有约1 s视觉不确定性，精确位姿及共同静止验收由评分成功支持；评分器未报告DAG八项原生指标的数值，均保留null。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 2,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "fix",
+    "conf": 7,
+    "summary_zh": "终态与2/2成功无误，但napkin_box_touching_tissues的起点23.0 s与证据不符：视频约19–20 s起橙白盒已贴着蓝盒，20.0–23.0 s帧几乎不变，没有所谓“21.0–23.2 s下降”。应把区间起点改到约19.5 s并更正证据文字。",
+    "n_issues": 1,
+    "n_major": 1
+   },
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 2,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "2/2 终态与评分器成功一致；napkin_box_touching_tissues 起点约19.5 s 与帧相符，接触标注（右夹爪抓握）合理，未发现需改动的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:50:15Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-privileged--sgdfrsl/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-privileged--sgdfrsl/04_review_r1"
+   },
+   {
+    "stage": "reconcile",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-privileged--sgdfrsl/05_reconcile_r1"
+   },
+   {
+    "stage": "traj_review",
+    "round": 2,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-privileged--sgdfrsl/04_review_r2"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "093e73a72ab937619693b863a5a88569c82e8cf547b66d545169a738017a6753",
+   "video_sha256": "1cc5a07547affb09543aef0359b7d4d1183e6329ab3ea2515192f94b77fc5c14",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/ispatialgym/inter_object_relation_satisfaction_touching_small_scale/trajectories/ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-standard--tdhwd8z.json

@@ -0,0 +1,292 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-standard--tdhwd8z",
+ "taskKey": "ispatialgym/inter_object_relation_satisfaction_touching_small_scale",
+ "dag_sha256": "28d0c1ba95380927c7f64c8a65a012a975a7dd662ef095211d362d931e923cef",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "live_matches_replay": 1,
+   "finalized": 1,
+   "missing_log": 0,
+   "n_actions": 2660,
+   "sim_steps": 2660,
+   "max_steps": 3142,
+   "steps": 2660,
+   "elapsed_s": 2910.6,
+   "symbolic_success": 0,
+   "success_step": 0,
+   "video_rendered": 1,
+   "gpu_lost": 0,
+   "replay_error": 0,
+   "instance_ok": 1,
+   "sim_startups": 1,
+   "resumes": 0,
+   "resume_state_mismatch": 0,
+   "resume_max_tcp_dev_m": 0.0,
+   "resume_max_joint_dev": 0.0,
+   "gave_up": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 0.449,
+   "peak_processes": 9,
+   "peak_disk_gb": 1.001,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-standard--tdhwd8z/replay.mp4",
+  "duration_s": 88.8,
+  "fps": 10.0,
+  "frames": 888,
+  "sha256": "582535c6f7fcf6ee14481bd1820356f444541442740db2df7e25bc2a1be7f0d1"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 1,
+  "peak": 1,
+  "final": 0,
+  "nodes": {
+   "tissue_box_at_goal": {
+    "method": "visual",
+    "intervals": [
+     [
+      0,
+      43.0
+     ]
+    ],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 蓝色纸巾盒在白桌面左侧，位置和朝向与成功参考视频 0.0 s 的同一初态相符；本次视频 6.0 s、20.0 s、38.7 s 至39.7 s 仍见它留在桌面原处。"
+     },
+     {
+      "t": 46.4,
+      "log_step": null,
+      "text_zh": "观察：视频 39.7 s 后相机转离两盒；视频 46.4 s 至50.0 s 再见蓝盒时，它已在机器人前方的地板上，橙白盒仍在桌边。视频 40.0–46.2 s 未显示蓝盒跌落过程，不能确定碰它的机器人部位。"
+     },
+     {
+      "t": 43.0,
+      "log_step": null,
+      "text_zh": "推断：蓝盒从初始就位状态变为落地未就位，失效发生在视频 39.7–46.4 s 的不可见时段；区间终点粗记为视频 43.0 s，取前后观测界限中间的近似锚点，并非观测到该时刻落地。"
+     },
+     {
+      "t": 88.7,
+      "log_step": 247,
+      "text_zh": "观察：视频 50.0–81.0 s 的机械臂恢复动作未将蓝盒带回桌面；视频 84.0–88.7 s，含末秒逐帧画面，蓝盒一直在地板上，终态不满足桌面目标。agent 最后也自述蓝盒仍在地板（log step 247），此处终值由画面确定。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：机器人右夹爪在视频 50.0–63.0 s、68.0–81.0 s 参与蓝盒地板拾取恢复尝试，但可见时夹爪仍在盒上方，未见夹住或带起蓝盒；视频 39.7–46.2 s 蓝盒离位过程在画外，缺少接触对记录，无法确定实际触盒部位和方式。"
+     }
+    ]
+   },
+   "napkin_box_touching_tissues": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 蓝盒和橙白色卫生巾盒之间有明显桌面间隙；成功参考视频 8.0 s、9.1 s 则显示两盒长侧面紧挨，供可见判据比较。"
+     },
+     {
+      "t": 11.5,
+      "log_step": 78,
+      "text_zh": "观察：视频 10.0–11.5 s，机器人右二指夹爪夹住橙白盒并将其抬离桌面，原位置不再留有该盒；视频 11.5–13.4 s 它随右夹爪移动，尚未形成桌面相贴关系。log step 78 的服务读数同时报告 control_steps=345、右夹爪开口约0.071 m，仅作为持物动作的辅助证据。"
+     },
+     {
+      "t": 20.0,
+      "log_step": 81,
+      "text_zh": "观察：视频约13.5–13.8 s 右夹爪张开，橙白盒离开夹爪；视频 17.5 s、20.0 s 至21.0 s 手臂移开后，两盒仍有明显间隔，橙白盒朝向也未与蓝盒并齐。log step 81 记录这次放置及随后静置后服务仍为 done=false。"
+     },
+     {
+      "t": 38.7,
+      "log_step": null,
+      "text_zh": "观察：视频 22.0–36.0 s 头部视野多次转离两盒；视频 37.5–39.7 s 两盒重新出现时仍未侧面贴齐，橙白盒位于蓝盒右后方。其��视频 46.4 s 起蓝盒已在地板，视频 88.7 s 橙白盒仍在桌边，始终未见成功参考视频中两盒在桌面长侧贴齐的状态。"
+     },
+     {
+      "t": 88.7,
+      "log_step": null,
+      "text_zh": "推断：结合前后连续可见的间隙、橙白盒离桌搬动、后续未完成的重抓和蓝盒落地后的分离终态，没有可计入的相贴区间；画外重抓时段不以 agent 的持握或放置自述计为达成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "观察：视频约10.0–13.5 s，机器人右二指平行夹爪从两侧夹住橙白色卫生巾盒，将其抬起、横移后张爪释放；未见双臂同时配合。后续重抓未见再次成功搬至蓝盒旁，不把空夹读数当作抓住。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "max_target_position_error",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_target_orientation_error",
+   "unit": "rad",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_settle_angular_speed",
+   "unit": "rad/s",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_robot_joint_speed",
+   "unit": "rad/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_position_drift",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_orientation_drift",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_window_joint_drift",
+   "unit": "joint_unit",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "napkin_box_touching_tissues",
+   "t": 13.5,
+   "log_step": 81,
+   "where": "操作中",
+   "evidence_zh": "观察：视频约13.5–13.8 s 机器人右夹爪释放首次搬动的橙白盒，视频 17.5–21.0 s 退手后两盒仍分离且未并齐；后续视频 37.5–39.7 s 仍未修复，视频 46.4 s 以后蓝盒落地，视频 88.7 s 仍未相贴。推断：最先未被完成修复的步骤是两盒相贴，断点锚定首次错误落位，而非后来的蓝盒落地或会话截止。"
+  },
+  "unmet": {
+   "tissue_box_at_goal": false,
+   "napkin_box_touching_tissues": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 13.5,
+    "log_step": 81,
+    "evidence_zh": "观察：视频 11.5–13.4 s 机器人持橙白盒横移，视频约13.5–13.8 s 张爪放下；视频 17.5 s、20.0 s 与21.0 s 显示橙白盒落点仍与蓝盒有间隙，朝向未并齐，未形成成功参考视频 8.0–9.1 s 所见的桌面侧面相贴。推断：释放时未将盒子送到所需相贴位姿；不是已正确放到目标后才失稳。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "反馈控制",
+    "t": 13.5,
+    "log_step": 167,
+    "source_zh": "日志",
+    "evidence_zh": "观察：log step 123 的 agent 交接记录说明首次下放目标为 TCP [0.61,0.074,0.83] m，但释放时实测高度约0.918 m；log step 167 读取本次实时 episode 保存的 release_meta，明确在 control step 415 得到右 TCP [0.617,0.079,0.918] m、夹爪开口0.100 m，log step 83 的静置后服务读数也保持约0.918 m。视频约13.5–13.8 s 确认夹爪张开，视频 20.0 s 确认落位有偏差。推断：末端尚未下放到指令高度，脚本已进入释放和静置，没有在释放前使实际下放位姿收敛；这支持反馈控制问题。此归因不将 IK 返回的小数值误差当作实际执行到位。"
+   }
+  ],
+  "gap_zh": "无状态日志，不能精确核验两盒的三维误差、原生 Touching 的微小间隙容差或共同45控制步静止窗口，评分器也未报告 DAG 所列八项原生指标，故其值均为 null。视频 22.0–36.0 s 多次转离两盒；蓝盒在视频 39.7 s 尚在桌面，视频 46.4 s 已落地，中间离位过程在画外，故其就位区间终点视频 43.0 s 是约±3.5 s 的估计，实际碰撞部位与接触方式未知。日志 control_steps 属于单次实时 episode，2660/30≈88.67 s 与视频长度一致，control step 415/30≈13.83 s 的释放在视频约13.5–13.8 s 已核对；此映射仅辅助核实服务读数，未把 agent 工作时间当作视频时间。"
+ },
+ "summary_zh": "本次开局纸巾盒已就位，进度为1/2；机器人右夹爪夹起橙白盒，但视频约13.5 s 首次落位后仍未与蓝盒贴齐，后续修正未完成。日志显示释放时末端仍高于其下放目标，支持反馈控制未收敛；蓝盒随后落地且恢复失败，终态0/2，与评分��未成功一致。蓝盒离位发生在视频39.7–46.4 s的画外时段，区间终点只能估计，且无状态日志核验微小位姿及静止指标。",
+ "conf": 6,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "标签与视频一致：视频 0.0 s 蓝盒就位，约13.5–13.8 s 右夹爪释放橙白盒后两盒仍有间隙，46.4 s 起蓝盒落地；断点、位姿不准与反馈控制（log step 167 的 release_meta z≈0.918 m 对比下放目标0.83 m）均有证据支持，蓝盒离位在画外，已按推断标注。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T02:38:44Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-standard--tdhwd8z/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ispatialgym-inter-object-relation-satisfaction-touching-small-scale-i00-standard--tdhwd8z/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "29815e50fb03a6abd7f647d6aaa29ca0fbe301c8572056448867916e77a1fa8f",
+   "video_sha256": "582535c6f7fcf6ee14481bd1820356f444541442740db2df7e25bc2a1be7f0d1",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}