Add 122 trajectories (phase 2, remaining runs; batch S43)

phase2/embodied-olympics-fun/box_jump/trajectories/embodied-olympics-fun-box-jump-i00-privileged--ebyqjqt.json

@@ -0,0 +1,203 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-box-jump-i00-privileged--ebyqjqt",
+ "taskKey": "embodied-olympics-fun/box_jump",
+ "dag_sha256": "94c6447a4b03c43587761e86e851339c4316a0cf5855b294978de845b6362c39",
+ "outcome": {
+  "success": 0,
+  "reward": 0.3333,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "n_actions": 840,
+   "sim_steps": 474,
+   "horizon": 2000,
+   "episode_over": 1,
+   "progress": 0.3333,
+   "stages_reached": 1,
+   "stages_total": 3,
+   "fell": 0,
+   "replay_s": 4.59,
+   "bad_trajectory": 0,
+   "metric_air_s": 0.546,
+   "metric_pelvis_rise_m": 0.302,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 117,
+   "agent_resets": 2320,
+   "agent_replays": 2,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 5.547,
+   "peak_processes": 29,
+   "peak_disk_gb": 0.05,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-box-jump-i00-privileged--ebyqjqt/replay.mp4",
+  "duration_s": 4.76,
+  "fps": 25.0,
+  "frames": 119,
+  "sha256": "283b9131edba407b5d503cb319de57154db07c93e58fb984cb27932570430987"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "box_jump_stood": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00 s 机器人双脚站在棕色箱前地面；视频 1.20–3.20 s 向箱前靠近并停住，视频 3.52–4.16 s 屈膝、摆臂准备起跳，尚未上箱。"
+     },
+     {
+      "t": 4.44,
+      "log_step": null,
+      "text_zh": "观察：逐帧查看视频 3.52–4.72 s；机器人约于视频 4.20 s 双脚离地，视频 4.28–4.60 s 双脚仍在箱顶以下、箱前一侧，随后下降；视频 4.72 s 双脚落向箱前地面，未出现双脚落在箱顶或箱顶站直保持。"
+     },
+     {
+      "t": 4.72,
+      "log_step": 110,
+      "text_zh": "观察：log step 109 重新生成最终 840 行交付轨迹；log step 110 对该文件的两次新进程重放均报告 failed=\"the jump landed on the floor\"、stages=[\"take-off\"]、success=false，最终状态哈希相同；inputs/run.json 的评分器字段 deterministic=1、stages_reached=1、stages_total=3，与本次视频的起跳后落回地面一致。原生起跳阶段通过不等于整个“跳上箱并站直”节点通过。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "机器人左右脚在箱前地面承重、行走靠近、屈膝蹬地起跳；末段双脚返回箱前地面（视频 0.00–4.72 s，log step 110 的最终重放亦报告落地面）。未形成箱顶双脚支撑。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "air_s",
+   "unit": "s",
+   "threshold": 0.1,
+   "direction": "higher_better",
+   "value": 0.546
+  },
+  {
+   "name": "pelvis_rise_m",
+   "unit": "m",
+   "threshold": 0.1,
+   "direction": "higher_better",
+   "value": 0.302
+  },
+  {
+   "name": "pelvis_above_box_m",
+   "unit": "m",
+   "threshold": 1.0,
+   "direction": "higher_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "box_jump_stood",
+   "t": 4.44,
+   "log_step": 110,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 4.28–4.60 s 机器人已腾空但双脚仍低于箱顶，随后落回箱前地面，视频 4.72 s 仍未上箱；log step 110 对最终交付文件的重放明确报告“the jump landed on the floor”，只确认起跳阶段。失败发生在跳箱动作的腾跃、落点衔接中，未到箱顶站立验收。"
+  },
+  "unmet": {
+   "box_jump_stood": false
+  },
+  "phenomena": [
+   {
+    "cat": "全身",
+    "item": "步态",
+    "t": 4.44,
+    "log_step": 110,
+    "evidence_zh": "观察：视频 4.20–4.72 s 机器人双脚离地后，脚端没有越上箱顶，而是在箱前下降并落回地面；机器人躯干未翻倒。log step 110 的最终文件重放仅达到 take-off，随后因落在地面结束，表现为跳跃未上箱。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "log_step": 110,
+    "evidence_zh": "推断：log step 97 记录 agent 使用参数化低层关节控制脚本安排蹬伸、空中收腿和箱顶落地；log step 105 的参数试验中，基线落在地面、加快收腿的变体落在箱侧。log step 109 将最终参数生成交付文件，log step 110 ��次重放均只完成起跳并落地面；视频 4.28–4.60 s 的脚端始终低于箱顶与之吻合。因此可归于该跳跃控制序列未能实现所需的腾跃及收腿动作，证据不足以进一步指定控制子类。",
+    "source_zh": "日志 / 视频"
+   }
+  ],
+  "gap_zh": "没有状态日志，无法核实逐控制步的关节跟踪、力矩、接触对或速度，无法将控制失败细分到轨迹、反馈或特定参数。评分器未报告 pelvis_above_box_m；单侧画面不足以判断毫米级脚箱接触及法线，未额外认定箱侧碰撞。agent 在 log step 113 自述的力矩/行程上限解释未经独立验证。"
+ },
+ "summary_zh": "机器人靠近箱子、屈膝摆臂并完成起跳，评分器记录腾空 0.546 s、骨盆上升 0.302 m，但视频 4.28–4.72 s 显示双脚未越上箱顶而落回箱前地面，唯一语义节点“跳上箱并站直”从未达成，最终为 0/1（原生阶段为 1/3）。失败现象为全身·步态，最终交付脚本的跳跃控制未实现上箱；缺少状态日志，具体控制机制和细微接触无法细分。",
+ "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、评分器字段一致：约视频 4.2 s 起跳腾空，双脚始终低于箱顶并落回箱前地面，节点 box_jump_stood 未达成（final=false，0/1，原生 1/3）。断点、步态现象、控制根因（停在“控制”未细分）均有证据支持，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": "2026-10-31T00:56:01Z",
+  "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/embodied-olympics-fun-box-jump-i00-privileged--ebyqjqt/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-box-jump-i00-privileged--ebyqjqt/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "917cacd9b157fbe8294c74153ea093f01c885119a01034881edec4154efd9e25",
+   "video_sha256": "283b9131edba407b5d503cb319de57154db07c93e58fb984cb27932570430987",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/box_jump/trajectories/embodied-olympics-fun-box-jump-i00-privileged--kfxjcam.json

@@ -0,0 +1,189 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-box-jump-i00-privileged--kfxjcam",
+ "taskKey": "embodied-olympics-fun/box_jump",
+ "dag_sha256": "94c6447a4b03c43587761e86e851339c4316a0cf5855b294978de845b6362c39",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "n_actions": 410,
+   "sim_steps": 410,
+   "horizon": 2000,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 3,
+   "stages_total": 3,
+   "fell": 0,
+   "replay_s": 7.8,
+   "bad_trajectory": 0,
+   "metric_air_s": 0.345,
+   "metric_pelvis_above_box_m": 1.022,
+   "metric_pelvis_rise_m": 0.292,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 47,
+   "agent_resets": 264,
+   "agent_replays": 2,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 0.741,
+   "peak_processes": 10,
+   "peak_disk_gb": 0.252,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-box-jump-i00-privileged--kfxjcam/replay.mp4",
+  "duration_s": 4.12,
+  "fps": 25.0,
+  "frames": 103,
+  "sha256": "be45f8d1567a992230ebde01f08ae633566081d0025189ff52ccb83fbf91b885"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "box_jump_stood": {
+    "method": "visual",
+    "intervals": [
+     [
+      4.08,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.6,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–0.60 s，机器人在棕色箱前双脚支撑地面，屈膝、前倾后伸腿蹬起；箱子仍在机器人前方。"
+     },
+     {
+      "t": 0.8,
+      "log_step": null,
+      "text_zh": "观察：视频 0.72–0.84 s，机器人双脚离开地面并收腿向箱顶运动，可见脚与地面之间的间隙；视频 0.96–1.08 s，双脚落在箱顶，机器人屈膝承重，双手伸在身体前方，未见手或躯干支撑箱子。"
+     },
+     {
+      "t": 2.12,
+      "log_step": null,
+      "text_zh": "观察：视频 1.12–2.12 s，机器人双脚留在箱顶，逐渐伸膝抬高骨盆、竖起躯干；视频 2.24–4.08 s 保持近直立姿态，最后一秒各帧仍见脚在箱顶，没有跌落。"
+     },
+     {
+      "t": 4.08,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器字段 success=1、replay_success=1、deterministic=1，stages_reached=3、stages_total=3、fell=0；报告腾空 0.345 s、骨盆上升 0.292 m、骨盆高于箱顶 1.022 m。"
+     },
+     {
+      "t": 4.08,
+      "log_step": null,
+      "text_zh": "推断：视频 2.12 s 附近进入站直保持，至视频末帧 4.08 s 仍在箱顶直立；结合评分器三阶段通过和 success=1，将含连续站立窗口的整步完成锚定在视频约 4.08 s，精度约半秒，而非首次落箱时。没有逐控制步状态日志，精确站立计数、速度与接触法线由评分结果佐证。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–0.60 s 机器人左右脚在地面承重并蹬起；视频 0.72–0.84 s 腾空收腿；视频约 1.00–4.08 s 左右脚落在棕色箱顶承重，机器人伸腿站直。双臂摆动，未见手臂接触箱子。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "air_s",
+   "unit": "s",
+   "threshold": 0.1,
+   "direction": "higher_better",
+   "value": 0.345
+  },
+  {
+   "name": "pelvis_rise_m",
+   "unit": "m",
+   "threshold": 0.1,
+   "direction": "higher_better",
+   "value": 0.292
+  },
+  {
+   "name": "pelvis_above_box_m",
+   "unit": "m",
+   "threshold": 1.0,
+   "direction": "higher_better",
+   "value": 1.022
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人从箱前地面起跳、收腿越向箱顶，视频约 1.00 s 双脚落箱，随后由蹲伏逐渐站直并保持至视频 4.08 s，完成 1/1 个节点。评分器确认起跳、落地、站立三阶段全部通过，两次重放一致，无失败断点。没有状态日志，毫米级接触位置、法线、精确速度和连续站立计数不能由画面独立核验；整步完成时刻结合视频末帧与成功判定近似标注。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "标注与视频和评分器字段一致：起跳腾空、约1.0 s落箱、站直保持至视频末帧；节点完成、contact_actual（腿/躯干 支撑）、无失败断点均成立。",
+    "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": "2026-10-31T00:53:18Z",
+  "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/embodied-olympics-fun-box-jump-i00-privileged--kfxjcam/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-box-jump-i00-privileged--kfxjcam/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "af3fdb4affa572bcbfc076f8ef421ae10268de46ea8a2be3244d62a93d15019f",
+   "video_sha256": "be45f8d1567a992230ebde01f08ae633566081d0025189ff52ccb83fbf91b885",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/box_jump/trajectories/embodied-olympics-fun-box-jump-i00-privileged--n3uwo5y.json

@@ -0,0 +1,194 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-box-jump-i00-privileged--n3uwo5y",
+ "taskKey": "embodied-olympics-fun/box_jump",
+ "dag_sha256": "94c6447a4b03c43587761e86e851339c4316a0cf5855b294978de845b6362c39",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "n_actions": 400,
+   "sim_steps": 400,
+   "horizon": 2000,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 3,
+   "stages_total": 3,
+   "fell": 0,
+   "replay_s": 4.52,
+   "bad_trajectory": 0,
+   "metric_air_s": 0.467,
+   "metric_pelvis_above_box_m": 1.021,
+   "metric_pelvis_rise_m": 0.435,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 8,
+   "agent_resets": 0,
+   "agent_replays": 8,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 4.167,
+   "peak_processes": 25,
+   "peak_disk_gb": 0.242,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-box-jump-i00-privileged--n3uwo5y/replay.mp4",
+  "duration_s": 4.04,
+  "fps": 25.0,
+  "frames": 101,
+  "sha256": "4158683cc3c16f39731585765ed013079cedfd76d36a710d53d562a2fb37beea"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "box_jump_stood": {
+    "method": "visual",
+    "intervals": [
+     [
+      4.0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.6,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–0.60 s，机器人在箱前地面由站立下蹲，双脚留在地面，手臂摆动，尚未跳上箱。"
+     },
+     {
+      "t": 0.88,
+      "log_step": null,
+      "text_zh": "观察：视频 0.64–0.76 s 机器人伸腿蹬地，视频 0.80–1.08 s 双脚离地并向箱顶移动，腿部在空中收起；身体未见先行搭靠箱子。"
+     },
+     {
+      "t": 1.2,
+      "log_step": null,
+      "text_zh": "观察：视频 1.12–1.24 s，机器人双脚落到棕色箱的顶面，屈膝缓冲；视频 1.40–1.80 s 仍在箱顶屈膝，随后逐渐伸腿抬高骨盆。"
+     },
+     {
+      "t": 4.0,
+      "log_step": null,
+      "text_zh": "观察：视频约 2.00 s 后机器人已在箱顶接近直立，视频 3.00–4.00 s 逐帧检查可见双脚仍在顶面、躯干保持直立，未见跌落或手臂触箱；最后一帧为视频 4.00 s。"
+     },
+     {
+      "t": 4.0,
+      "log_step": 92,
+      "text_zh": "观察：评分器 reward 报告 success=1、replay_success=1、deterministic=1、stages_reached=3/3、fell=0；log step 92 保存最终 400 行轨迹并重放成功，指标与本次 reward 一致，log step 93 再次重放成功且终态哈希一致。"
+     },
+     {
+      "t": 4.0,
+      "log_step": 92,
+      "text_zh": "推断：结合评分器完成全部阶段的结果与视频 2.00–4.00 s 的箱顶站立，完整的连续站立验收窗口在视频约 4.00 s 完成，因此此时才计节点达成。无状态日志，无法从侧视图独立量出接触法线、毫米级接触高度及速度；这些精确验收条件依据评分器成功结果确认。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "机器人左右脚先支撑地面并蹬地起跳，腾空后以双脚落到箱顶承重，随后伸腿在箱顶站立保持；接触端为双脚，对象依次为地面和箱顶。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "air_s",
+   "unit": "s",
+   "threshold": 0.1,
+   "direction": "higher_better",
+   "value": 0.467
+  },
+  {
+   "name": "pelvis_rise_m",
+   "unit": "m",
+   "threshold": 0.1,
+   "direction": "higher_better",
+   "value": 0.435
+  },
+  {
+   "name": "pelvis_above_box_m",
+   "unit": "m",
+   "threshold": 1.0,
+   "direction": "higher_better",
+   "value": 1.021
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人完成了唯一节点“跳上箱并站直”：视频约 0.80 s 腾空、约 1.20 s 双脚落上箱顶，随后伸腿站立，至视频约 4.00 s 完成保持窗口，最终进度 1/1。评分器确认全部三阶段通过且重放确定性一致，故无失败断点。没有状态日志，细微接触几何和精确速度无法独立从画面量测，依据评分器成功结果确认。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "节点 box_jump_stood 在视频约 4.00 s 达成与画面和评分器结果一致；双脚起跳、箱顶落地、站直均可见，contact_actual 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": "2026-10-31T00:55:15Z",
+  "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/embodied-olympics-fun-box-jump-i00-privileged--n3uwo5y/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-box-jump-i00-privileged--n3uwo5y/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "3d8f905439cbd4a06ad0461bb0b1aa4b5282a88895be4b1648b674096df8b02c",
+   "video_sha256": "4158683cc3c16f39731585765ed013079cedfd76d36a710d53d562a2fb37beea",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/box_jump/trajectories/embodied-olympics-fun-box-jump-i00-privileged--nnavfyr.json

@@ -0,0 +1,231 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-box-jump-i00-privileged--nnavfyr",
+ "taskKey": "embodied-olympics-fun/box_jump",
+ "dag_sha256": "94c6447a4b03c43587761e86e851339c4316a0cf5855b294978de845b6362c39",
+ "outcome": {
+  "success": 0,
+  "reward": 0.3333,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "n_actions": 680,
+   "sim_steps": 680,
+   "horizon": 2000,
+   "episode_over": 1,
+   "progress": 0.3333,
+   "stages_reached": 1,
+   "stages_total": 3,
+   "fell": 0,
+   "replay_s": 9.46,
+   "bad_trajectory": 0,
+   "metric_air_s": 0.53,
+   "metric_pelvis_rise_m": 0.132,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 209,
+   "agent_resets": 70,
+   "agent_replays": 2,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 1.052,
+   "peak_processes": 11,
+   "peak_disk_gb": 0.842,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-box-jump-i00-privileged--nnavfyr/replay.mp4",
+  "duration_s": 6.84,
+  "fps": 25.0,
+  "frames": 171,
+  "sha256": "cfdb5eafbc96830a395415a6f5a930cf30e7bc1475c93cdf4299fc37ffff4da8"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "box_jump_stood": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00 s 机器人站在箱前地面；视频 0.00–5.20 s 的抽帧显示机器人向箱前靠近并停稳，尚未在箱顶支撑。"
+     },
+     {
+      "t": 6.4,
+      "log_step": null,
+      "text_zh": "观察：视频 5.80–6.80 s 已逐帧检查；机器人从下蹲伸腿起跳，视频约 6.36–6.72 s 腾空时一腿向箱顶前伸、另一腿向后屈曲，骨盆仍在箱前，未形成双脚箱顶支撑。"
+     },
+     {
+      "t": 6.8,
+      "log_step": 583,
+      "text_zh": "观察：视频 6.48–6.68 s 前伸脚仍悬在箱顶上方，另一腿向后屈曲，骨盆留在箱前；视频 6.72–6.80 s 前伸脚下降至箱顶前缘一带，另一腿落向箱前地面，末帧呈低位跪落姿态，未形成双脚箱顶承重或站直保持。交付文件的裁判重放结果（log step 583）仅记录 take-off 阶段，失败原因为 the jump landed on the floor，触地部位为 left_shin；评分器字段 stages_reached=1、stages_total=3。"
+     },
+     {
+      "t": 6.8,
+      "log_step": null,
+      "text_zh": "推断：整段视频未出现双脚落箱并伸腿站稳的条件，故合并节点“跳上箱并站直”从未成立；原生起跳阶段的 1/3 进度不等于此语义节点完成。成功参考视频在其视频约 1.00 s 显示双脚箱顶蹲伏、视频 2.00–4.08 s 显示箱顶站直，仅用于比较可见完成姿态。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 5.80–6.32 s 机器人左右脚以地面支撑并蹬起；视频 6.72–6.80 s 前伸脚下降至箱顶前缘一带，另一腿在箱前落下，全程未形成双脚箱顶承重。交付重放 log step 583 确认左小腿（left_shin）触地。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "air_s",
+   "value": 0.53,
+   "unit": "s",
+   "threshold": 0.1,
+   "direction": "higher_better"
+  },
+  {
+   "name": "pelvis_rise_m",
+   "value": 0.132,
+   "unit": "m",
+   "threshold": 0.1,
+   "direction": "higher_better"
+  },
+  {
+   "name": "pelvis_above_box_m",
+   "value": null,
+   "unit": "m",
+   "threshold": 1.0,
+   "direction": "higher_better"
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "box_jump_stood",
+   "t": 6.8,
+   "log_step": 583,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 6.72–6.80 s 机器人在跳箱下降过程中没有以双脚承接在箱顶，身体在箱前下落至小腿触地；交付重放的裁判输出（log step 583）确认 left_shin 触地并判负，只有 take-off 阶段成立，之后没有修复。"
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 6.8,
+    "log_step": 583,
+    "evidence_zh": "观察：视频 6.48–6.68 s 机器人一脚前伸至箱顶上方，另一腿向后屈曲，骨盆仍在箱前；视频 6.72–6.80 s 前伸脚下降至箱顶前缘一带，另一腿落向箱前地面，未形成双脚落箱的目标位姿。交付重放 log step 583 记录 left_shin 触地、the jump landed on the floor，只有 take-off 阶段成立。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "t": 6.6,
+    "log_step": 583,
+    "source_zh": "视频 / 日志",
+    "evidence_zh": "推断：视频 6.36–6.80 s 中机器人的前伸腿与后屈腿未能在下降前协调到箱顶，骨盆持续下沉直至小腿落地，表明该跳跃控制未实现双脚落箱的身体运动；交付重放 log step 583 的有效起跳指标 air_s=0.53、pelvis_rise_m=0.132 与随后触地判负支持失败在起跳后的落地控制，而非没有起跳。现有证据不足以确定更细的控制子类。"
+   }
+  ],
+  "gap_zh": "观察：本作业没有逐帧状态日志；视频为单一侧视，不能精确核实脚箱接触法线、接触力或控制跟踪误差，pelvis_above_box_m 未被评分器报告。日志中大量会话测试输出为空，最终生成器修改内容也未完整展示；因此无法进一步区分轨迹、反馈控制或参数问题。log step 585–594 是交付验证后的其他测试，未作为本评分轨迹的动作证据。",
+  "unmet": {
+   "box_jump_stood": false
+  }
+ },
+ "summary_zh": "观察：机器人靠近箱子、下蹲并完成裁判认可的起跳，但下降时双脚落点未能同时到达箱顶，视频约 6.80 s 左小腿触地，唯一节点“跳上箱并站直”未完成（0/1）。推断：断点为跳箱操作中的落脚位姿不准，根因可支持到控制层；缺少逐帧状态和完整控制记录，无法确定更细的控制机制。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 2,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "fix",
+    "conf": 6,
+    "summary_zh": "节点、断点、时钟均正确（0/1，断点 box_jump_stood，视频 6.8 s 左小腿触地）。问题在现象：机器人从未获得支撑，跳得不够远/高，脚只够到箱沿后翻落箱前，现象“平衡”（失去支撑或翻倒）定义不贴切，应为位姿不准（落脚点没到箱顶）。",
+    "n_issues": 2,
+    "n_major": 1
+   },
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 2,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "单节点未达成、断点 box_jump_stood（视频 6.8 s，左小腿触地，log step 583 的裁判事件 fail t=6.798）、时钟与现象（运动·位姿不准）、根因停在「控制」均与视频和日志相符；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": "2026-10-31T01:09:46Z",
+  "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/embodied-olympics-fun-box-jump-i00-privileged--nnavfyr/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-box-jump-i00-privileged--nnavfyr/04_review_r1"
+   },
+   {
+    "stage": "reconcile",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-box-jump-i00-privileged--nnavfyr/05_reconcile_r1"
+   },
+   {
+    "stage": "traj_review",
+    "round": 2,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-box-jump-i00-privileged--nnavfyr/04_review_r2"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "4934d4eb9874da3a6b8499d76f698947b90bcd4949e6bb58d3d15c82d3597bb4",
+   "video_sha256": "cfdb5eafbc96830a395415a6f5a930cf30e7bc1475c93cdf4299fc37ffff4da8",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/box_jump/trajectories/embodied-olympics-fun-box-jump-i00-privileged--ofxbtle.json

@@ -0,0 +1,195 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-box-jump-i00-privileged--ofxbtle",
+ "taskKey": "embodied-olympics-fun/box_jump",
+ "dag_sha256": "94c6447a4b03c43587761e86e851339c4316a0cf5855b294978de845b6362c39",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "n_actions": 50,
+   "sim_steps": 50,
+   "horizon": 2000,
+   "episode_over": 0,
+   "progress": 0.0,
+   "stages_reached": 0,
+   "stages_total": 3,
+   "fell": 0,
+   "replay_s": 2.53,
+   "bad_trajectory": 0,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 2918,
+   "agent_resets": 79,
+   "agent_replays": 1,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 1,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 33.989,
+   "peak_processes": 17,
+   "peak_disk_gb": 0.275,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "over_memory_gb": 1,
+   "session_log_tampered": 1,
+   "session_log_from_monitor": 1
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-box-jump-i00-privileged--ofxbtle/replay.mp4",
+  "duration_s": 0.52,
+  "fps": 25.0,
+  "frames": 13,
+  "sha256": "21ecfeadb8f7b8adaac4c43f2348c408ac1f52c68be79ba4aaaf82fd2ad0a0a6"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "box_jump_stood": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": 15,
+      "text_zh": "观察：评分视频 0.00–0.48 s 的全部 13 帧中，机器人始终双脚站在箱前地面，未见下蹲蹬地、腾空、触箱或登上箱顶；视频 0.48 s 的末帧仍为原处站姿。log step 15 中 agent 的脚本执行 50 次初始 hold 动作后保存，log step 16 的生成脚本也只有这 50 次 hold。该节点的跳跃工作未启动，故不记录任务接触；原地站立的背景支撑不算跳箱尝试。"
+     },
+     {
+      "t": 0.48,
+      "log_step": 18,
+      "text_zh": "观察：log step 18 回放已保存文件，返回 executed_rows=50、success=false、stages=[]、metrics={}、episode_over=false；最终评分器同样记录 n_actions=50、sim_steps=50、stages_reached=0、fell=0、deterministic=1。结合视频 0.00–0.48 s 的箱前站姿，跳上箱并站直从未达到；并非已跳上但未保持。"
+     },
+     {
+      "t": null,
+      "log_step": 354,
+      "text_zh": "观察：log step 354 检查交付文件，actions shape 仍为 (50,141)。成功参考视频约 2.00–4.00 s 可见机器人双脚支撑在箱顶并伸直站立；该参考只用于核对可见完成姿态，不作为本次进度证据。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "air_s",
+   "unit": "s",
+   "threshold": 0.1,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "pelvis_rise_m",
+   "unit": "m",
+   "threshold": 0.1,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "pelvis_above_box_m",
+   "unit": "m",
+   "threshold": 1.0,
+   "direction": "higher_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "box_jump_stood",
+   "t": 0.0,
+   "log_step": 15,
+   "where": "首步前",
+   "evidence_zh": "观察：评分视频 0.00–0.48 s 全部帧仅显示机器人在箱前保持原地站姿，没有开始起跳。log step 15 保存 50 步 hold 基线，log step 18 回放确认无阶段完成；最终评分器记录的 50 个动作与该基线相符。因此失败停在“跳上箱并站直”首步前；视频 0.00 s 仅作未启动步骤的导航锚点。"
+  },
+  "unmet": {
+   "box_jump_stood": false
+  },
+  "phenomena": [
+   {
+    "cat": "没尝试",
+    "item": "没尝试",
+    "t": 0.0,
+    "log_step": 15,
+    "evidence_zh": "观察：评分视频 0.00–0.48 s 中机器人只在箱前原地站立，双脚没有开始蹬起或移动到箱上。log step 15 的已保存动作是 50 步 hold，log step 18 对该文件的回放 stages=[]；后续会话中的起跳测试未出现在这段评分视频里。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "log_step": 349,
+    "source_zh": "日志",
+    "evidence_zh": "推断：agent 未能在会话调试中生成合规落箱的跳跃控制，最终保留了静站基线。证据是 log step 344 的候选仅达 take-off 后落回地面，log step 349 的优化日志多次返回 stages=[take-off]、did not land on the feet、success=False；log step 360 的三组落地测试均未合规落箱，直到 log step 615 的测试仍以 the jump landed on the floor 结束。log step 354 确认交付文件仍只��初始 50 步，最终视频也只显示静站。这些测试仅支持根因判断，其落地失败不作为评分轨迹的现象或进度；现有证据不足以细分控制子类。"
+   }
+  ],
+  "gap_zh": "没有逐步状态日志，无法区分会话测试中落箱失败究竟来自接触设计、轨迹求解还是反馈跟踪；根因只保留控制大类。视频完整但仅 0.52 s，后续会话调试不属于评分片段；原生 air_s、pelvis_rise_m、pelvis_above_box_m 均未由最终评分器报告，保留 null。"
+ },
+ "summary_zh": "评分轨迹仅为箱前 50 步静站基线，机器人未启动跳跃，唯一节点“跳上箱并站直”始终未达到，最终 0/1。断点为首步前，现象为没尝试；会话多次跳跃测试无法合规落箱，支持控制方面的根因推断，测试中的运动不计入本次轨迹。缺少逐步状态日志，不能进一步区分控制子类，三个原生指标也未报告。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "评分轨迹只有 50 步 hold 基线，视频 0.00/0.24/0.48 s 机器人始终站在箱前；节点未达、断点首步前、没尝试、控制根因（来自会话证据，gap_zh 已说明）均有证据支持，与 FAIL-22/PhMAE2F 处理一致。",
+    "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": "2026-10-31T00:53:42Z",
+  "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/embodied-olympics-fun-box-jump-i00-privileged--ofxbtle/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-box-jump-i00-privileged--ofxbtle/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "bdc4c9811873b864e853f4c7bca2437e81e94b882a94d717719114c9ff9320dc",
+   "video_sha256": "21ecfeadb8f7b8adaac4c43f2348c408ac1f52c68be79ba4aaaf82fd2ad0a0a6",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/box_jump/trajectories/embodied-olympics-fun-box-jump-i00-privileged--ylstfpm.json

@@ -0,0 +1,202 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-box-jump-i00-privileged--ylstfpm",
+ "taskKey": "embodied-olympics-fun/box_jump",
+ "dag_sha256": "94c6447a4b03c43587761e86e851339c4316a0cf5855b294978de845b6362c39",
+ "outcome": {
+  "success": 0,
+  "reward": 0.3333,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "n_actions": 301,
+   "sim_steps": 301,
+   "horizon": 2000,
+   "episode_over": 1,
+   "progress": 0.3333,
+   "stages_reached": 1,
+   "stages_total": 3,
+   "fell": 0,
+   "replay_s": 3.7,
+   "bad_trajectory": 0,
+   "metric_air_s": 0.355,
+   "metric_pelvis_rise_m": 0.458,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 10,
+   "agent_resets": 0,
+   "agent_replays": 5,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 3.998,
+   "peak_processes": 27,
+   "peak_disk_gb": 0.984,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-box-jump-i00-privileged--ylstfpm/replay.mp4",
+  "duration_s": 3.04,
+  "fps": 25.0,
+  "frames": 76,
+  "sha256": "e37c8576cae912de3f5dcc205852564fce4ab226b117ff5d37a2af7c04d53173"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "box_jump_stood": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.24,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–2.16 s，机器人在箱前由站立转为屈膝蹬地；视频 2.24–2.44 s 双脚离地并向箱子上方移动，已实际尝试跳箱。"
+     },
+     {
+      "t": 2.52,
+      "log_step": null,
+      "text_zh": "观察：视频 2.48–2.76 s，机器人一脚短暂踩住箱顶，另一腿向后抬起，躯干越来越前倾；视频 2.84–3.00 s 双脚离开箱顶、身体继续向箱子前方翻转，整段未出现箱顶双脚伸直站稳保持。"
+     },
+     {
+      "t": null,
+      "log_step": 326,
+      "text_zh": "观察：log step 326 对最终交付的 301 行轨迹重放，裁判返回 did not land with both feet on the box，stages 只有 take-off；其 air_s=0.355、pelvis_rise_m=0.458 与本次 reward 相同。reward 还记录 deterministic=1、stages_reached=1/3、success=0，与视频所见未完成落地和站立一致。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–2.20 s，机器人双脚交替或共同支撑地面并蹬地起跳；视频 2.48–2.76 s 至少一脚在箱顶短暂承重，另一腿在后方摆动，随后脚离开箱顶。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "air_s",
+   "value": 0.355,
+   "unit": "s",
+   "threshold": 0.1,
+   "direction": "higher_better"
+  },
+  {
+   "name": "pelvis_rise_m",
+   "value": 0.458,
+   "unit": "m",
+   "threshold": 0.1,
+   "direction": "higher_better"
+  },
+  {
+   "name": "pelvis_above_box_m",
+   "value": null,
+   "unit": "m",
+   "threshold": 1.0,
+   "direction": "higher_better"
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "box_jump_stood",
+   "t": 2.6,
+   "log_step": 326,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 2.48–2.76 s，机器人在一脚短暂踩上箱顶后继续前倾、另一腿留在后方，随后视频 2.84–3.00 s 身体翻转离开箱面，未恢复双脚支撑；log step 326 的最终交付文件重放由裁判以 did not land with both feet on the box 结束，跳上箱并站直未完成。"
+  },
+  "phenomena": [
+   {
+    "cat": "全身",
+    "item": "平衡",
+    "t": 2.6,
+    "log_step": null,
+    "evidence_zh": "观察：视频 2.52–2.72 s，机器人在箱顶仅有短暂单脚支撑，躯干前倾接近水平且另一腿向后甩起；视频 2.84–3.00 s 支撑脚也离开箱顶，机器人向前翻转，未能在箱顶稳定下来。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "evidence_zh": "推断：log step 281、307 的控制脚本修改分别加入按脚位置触发伸腿和髋偏航反馈，log step 315 保存最终参数并生成 301 行轨迹，log step 322、326 验证其可重复重放但仍无法双脚落地；结合视频 2.48–3.00 s 一脚着箱后持续前倾并失去支撑，说明生成的跳跃与落地控制未能稳定身体姿态和双脚支撑。证据不足以进一步确定是具体轨迹、反馈律还是增益设置的问题。",
+    "source_zh": "日志、视频"
+   }
+  ],
+  "gap_zh": "没有状态日志，无��逐控制步核验左右脚接触对、接触法线和骨盆速度；视频与最终文件的裁判重放足以确认节点未完成，但不能据此确定具体控制参数或姿态失稳机制。",
+  "unmet": {
+   "box_jump_stood": false
+  }
+ },
+ "summary_zh": "机器人在箱前屈膝蹬地并腾空，一脚短暂落到箱顶，随后前倾翻转并离开箱面，未完成“跳上箱并站直”，节点最终为 0/1。最终交付轨迹的重放仅通过起跳阶段，双脚落地失败；现象为失衡，根因归于控制，未进一步细分。缺少状态日志，精确接触条件和具体控制失稳机制无法核验。",
+ "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 box_jump_stood 未达成、断点、平衡现象与控制根因均与视频和最终重放的裁判结果一致，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": "2026-10-31T00:56:13Z",
+  "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/embodied-olympics-fun-box-jump-i00-privileged--ylstfpm/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-box-jump-i00-privileged--ylstfpm/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "ec9b28416e6aeb2cc8a1500c971ae3e1d4c4f9ccae21e2419fd3a4df87a89fd3",
+   "video_sha256": "e37c8576cae912de3f5dcc205852564fce4ab226b117ff5d37a2af7c04d53173",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/box_jump/trajectories/embodied-olympics-fun-box-jump-i00-standard--amqkdmm.json

@@ -0,0 +1,225 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-box-jump-i00-standard--amqkdmm",
+ "taskKey": "embodied-olympics-fun/box_jump",
+ "dag_sha256": "94c6447a4b03c43587761e86e851339c4316a0cf5855b294978de845b6362c39",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 680,
+   "sim_steps": 4163,
+   "calls": 79,
+   "rows": 2509,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 5191.1,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 3,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 5.356,
+   "peak_processes": 19,
+   "peak_disk_gb": 0.384,
+   "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/embodied-olympics-fun-box-jump-i00-standard--amqkdmm/replay.mp4",
+  "duration_s": 6.84,
+  "fps": 25.0,
+  "frames": 171,
+  "sha256": "1be9542858b373475ab3e5da13b4d059c2b532d297b9f769126f92dd8973a0b7"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "box_jump_stood": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：本次视频 0.00–5.80 s，机器人始终在棕色箱前地面上，期间屈膝、调整脚位后重新站立，没有上箱；视频 6.00–6.52 s 才连续下蹲并蹬地，随后双脚离开地面向箱子运动。"
+     },
+     {
+      "t": 6.8,
+      "log_step": 111,
+      "text_zh": "观察：逐帧检查本次视频 5.80–6.80 s，视频 6.60–6.76 s 机器人腾空后屈膝收腿，视频 6.76–6.80 s 膝部附近的腿部贴到箱体近侧上沿，双脚仍低于箱顶并在箱体外侧，躯干前倾、膝盖弯曲；全片没有双脚在箱顶站直保持的阶段。log step 111 的服务状态明确记录最后回合 episode=6、control_steps=680、done=true、reason=failed，评分字段 graded_by_live=1、live_success=0。"
+     },
+     {
+      "t": 6.8,
+      "log_step": null,
+      "text_zh": "推断：本次跳跃没有建立后续站直保持所需的箱顶双脚支撑，故唯一节点从未满足。成功参考视频 inputs/oracle/replay.mp4 的视频约 1.00 s 可见双脚在箱顶蹲伏，参考视频约 2.00–4.00 s 可见伸腿直立；本次相同侧视中的这些可见达标姿态始终未出现。参考仅用于识别达标外观。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：本次视频 0.00–6.52 s，机器人左右脚在地面承重、调整脚位及蹬地，此处“支撑”指地面阶段；随后腾空，视频 6.76–6.80 s 可见膝部附近的腿部贴到棕色箱近侧上沿，双脚仍低于箱顶并位于箱体外侧，未建立箱顶脚部支撑。画面中双臂摆动，没有可见的手臂触箱。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "air_s",
+   "unit": "s",
+   "threshold": 0.1,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "pelvis_rise_m",
+   "unit": "m",
+   "threshold": 0.1,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "pelvis_above_box_m",
+   "unit": "m",
+   "threshold": 1.0,
+   "direction": "higher_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "box_jump_stood",
+   "t": 6.8,
+   "log_step": 111,
+   "where": "操作中",
+   "evidence_zh": "观察：本次视频 6.52–6.80 s，机器人蹬地腾空后向棕色箱运动，视频 6.76–6.80 s 膝部附近的腿部碰到箱体近侧上沿，双脚仍在箱顶高度以下、箱体外侧，没有形成箱顶双脚支撑；视频在该次接触时结束。log step 111 的最后评分回合 E6 随后由机器人服务返回 done=true、reason=failed，log step 112 的 hold_lean 为空，未继续进入脚本的站直保持控制。"
+  },
+  "unmet": {
+   "box_jump_stood": false
+  },
+  "phenomena": [
+   {
+    "cat": "全身",
+    "item": "步态",
+    "t": 6.8,
+    "log_step": 111,
+    "evidence_zh": "观察：本次视频 6.60–6.80 s，机器人腾空跳向箱子，视频 6.76–6.80 s 膝部附近的腿部碰到箱体近侧上沿，双脚仍低于箱顶且在箱体外侧，没有落上顶面；视频 6.80 s 仍屈膝前倾，未出现翻倒。推断：该次跳跃偏短，未完成��上箱所需的落脚位置。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "evidence_zh": "推断：接近阶段的实际移动方向与脚本预期不符，使后续开环跳跃从偏远的位置启动。log step 79 的脚本用 walk_rows(a.vx, ...) 接近箱子，并把测得距离减目标距离后换算为同方向第二段行走；log step 111 的最后回合 E6 测距却由 d_face=0.585 m 增至 0.660 m，目标为 D=0.30 m，log step 112 确认第二段时长 t2_cmd=0.45。与此一致，本次视频 1.00–5.50 s 机器人脚位向远离箱体的方向移动，视频 6.80 s 腿部撞到箱沿而脚仍低于箱顶、未落上顶面；agent 在 log step 112 自述“+vx moves the robot backward”。这些证据支持接近与跳跃执行的控制失配，尚不能细分具体控制子类。",
+    "source_zh": "日志 / 视频 / agent 自述"
+   }
+  ],
+  "gap_zh": "没有状态日志或逐接触裁判明细，无法核实箱沿接触的具体腿别和连杆、法线、局部高度，以及触发判负的精确条款；评分字段 first_reason_code=3 在现有记录中没有释义。接近方向失配有视频和日志支持，但其由接口语义误配、控制实现还是跟踪误差造成，不能进一步确定；根因保留为“控制”。评分器未报告 air_s、pelvis_rise_m、pelvis_above_box_m，均填 null；日志中的离线优化数值不属于本次评分回合。"
+ },
+ "summary_zh": "最后评分回合为 E6（log step 111）；机器人完成了地面准备和一次向箱子的腾空跳跃，但在视频 6.76–6.80 s 腿部碰到箱体近侧上沿，双脚仍低于箱顶、未落上顶面，未建立箱顶站直保持，节点最终为 0/1。接近动作实际使箱前距离增大，随后开环跳跃偏短，现象为“步态”，根因据视频与日志判为“控制”。缺少状态日志及接触裁判明细，无法进一步核实接触法线和具体判负条款。",
+ "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0/1、断点(box_jump_stood)、步态+控制的大方向与视频、log(E6 d_face 0.585→0.66、t2_cmd 0.45、control_steps 680)一致。但视频 6.76–6.80 s 看到的接触是大腿/膝部撞上箱体近侧上沿，脚仍在箱体顶面以下、箱子左侧，并非“脚落在箱体近侧上沿”，contact_actual 与断点证据对接触部位的描述有误。",
+    "n_issues": 2,
+    "n_major": 1
+   },
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 2,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "节点 0/1、断点 box_jump_stood、步态+控制与视频和 log 一致；上轮接触部位问题已更正，未发现新错误。",
+    "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": "2026-10-31T01:09:01Z",
+  "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/embodied-olympics-fun-box-jump-i00-standard--amqkdmm/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-box-jump-i00-standard--amqkdmm/04_review_r1"
+   },
+   {
+    "stage": "reconcile",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-box-jump-i00-standard--amqkdmm/05_reconcile_r1"
+   },
+   {
+    "stage": "traj_review",
+    "round": 2,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-box-jump-i00-standard--amqkdmm/04_review_r2"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "2f5e6125042606f7b8e04d0654cac31259b6f542641ae658b4d43b98569cec6e",
+   "video_sha256": "1be9542858b373475ab3e5da13b4d059c2b532d297b9f769126f92dd8973a0b7",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/box_jump/trajectories/embodied-olympics-fun-box-jump-i00-standard--fhhrrru.json

@@ -0,0 +1,202 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-box-jump-i00-standard--fhhrrru",
+ "taskKey": "embodied-olympics-fun/box_jump",
+ "dag_sha256": "94c6447a4b03c43587761e86e851339c4316a0cf5855b294978de845b6362c39",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 124,
+   "sim_steps": 124,
+   "calls": 5,
+   "rows": 105,
+   "resets": 0,
+   "reset_budget": 50,
+   "sim_startups": 1,
+   "elapsed_s": 3556.8,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 0.971,
+   "peak_processes": 8,
+   "peak_disk_gb": 0.292,
+   "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/embodied-olympics-fun-box-jump-i00-standard--fhhrrru/replay.mp4",
+  "duration_s": 1.28,
+  "fps": 25.0,
+  "frames": 32,
+  "sha256": "5580a7592b5e8166533a2f33c826fc430e4c87f7fb84baafefdf39bd269358b0"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "box_jump_stood": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.68,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–0.80 s，机器人在棕色箱前以双脚支撑地面并屈膝下蹲，随后伸腿蹬起；此时尚未在箱顶站立。"
+     },
+     {
+      "t": 1.24,
+      "log_step": 109,
+      "text_zh": "观察：视频 0.96–1.24 s，机器人腾空收腿后双脚落向箱子前缘，末帧仍深屈膝、骨盆位于箱前，未出现箱顶伸直站立及连续保持。log step 109 的实时机器人服务（执行动作并返回裁判状态的接口）报告 episode=1、done=True、reason=failed、outcome=\"landed on the side of the box\"，stages_reached=1/3。"
+     },
+     {
+      "t": 1.24,
+      "log_step": 109,
+      "text_zh": "推断：log step 109 的本回合 control_steps=124，按 log step 4 的 control_dt=0.01 s 得仿真 1.24 s；与视频 1.24 s 的最终落脚画面核对一致，视频封装总长为 1.28 s。起跳阶段通过不等于整项跳箱站直完成；本次节点始终未满足。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–0.88 s，机器人左右脚在地面承重并蹬起；视频 1.08–1.24 s，脚到达箱子前缘。log step 109 的实时裁判确认发生箱侧落脚接触，未建立合规双脚箱顶支撑；手臂摆动，未见手接触箱子。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "air_s",
+   "unit": "s",
+   "threshold": 0.1,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "pelvis_rise_m",
+   "unit": "m",
+   "threshold": 0.1,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "pelvis_above_box_m",
+   "unit": "m",
+   "threshold": 1.0,
+   "direction": "higher_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "box_jump_stood",
+   "t": 1.24,
+   "log_step": 109,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 1.08–1.24 s，机器人收腿后在箱子靠近起跳点的前缘落脚，未能进入合规箱顶站立；log step 109 的实时裁判以“landed on the side of the box”终止同一回合，之后没有修复或继续站直。"
+  },
+  "unmet": {
+   "box_jump_stood": false
+  },
+  "phenomena": [
+   {
+    "cat": "全身",
+    "item": "步态",
+    "t": 1.24,
+    "log_step": 109,
+    "evidence_zh": "观察：视频 0.96–1.24 s，机器人已腾空向箱子跳跃，但落脚仅到箱前缘，脚后部仍在顶面外，未形成双脚在顶面的合规落地；log step 109 的实时裁判确认箱侧落脚。视频结束前没有身体倒地，跳跃落点不足以完成上箱。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "log_step": 108,
+    "source_zh": "日志",
+    "evidence_zh": "推断：log step 108 的控制脚本预先按固定时间生成深蹲、蹬伸、收腿和落地的 lowcmd 关节目标及力矩，一次提交整段动作，没有根据实时飞行或脚箱位置修改后续目标。log step 112 对同次实时回合反馈与本地模型的比较显示，仿真 1.00 s 实测髋角约 -0.459 rad，而本地模型为 -0.940 rad；两者动态响应已有明显差异。结合视频 1.08–1.24 s 的前缘落脚�� log step 109 的判负，支持该低层跳跃控制未实现所需落脚轨迹；不能进一步确定是哪个增益、力矩或阶段切换参数导致。"
+   }
+  ],
+  "gap_zh": "没有逐步状态日志或脚箱接触点、法线记录，单侧视频不能精确重建是哪只脚先触箱侧及毫米级接触位置；箱侧接触由本回合实时裁判反馈确认。日志中的本地模型测试不属于评分回合，其成功不能计入节点进度；现有证据不足以把控制根因细分到特定参数或反馈机制。三个原生指标均未在 run.json 的评分器字段中报告，故保留 null。"
+ },
+ "summary_zh": "机器人完成了下蹲、双脚蹬地和腾空收腿，但在视频 1.24 s 落到箱子前缘，实时裁判报告箱侧落脚并终止，唯一节点“跳上箱并站直”未达成（0/1）。固定时间生成的低层控制轨迹未实现合规箱顶落脚；本地模型成功测试不计入本次进度。缺少逐步接触与状态日志，无法细分具体接触几何和控制参数原因。",
+ "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：节点 box_jump_stood 未达成，视频 1.24 s 落在箱前缘，log step 109 实时裁判 failed / landed on the side of the box，control_steps=124 对应 1.24 s；断点、步态现象和控制根因有证据支持，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": "2026-10-31T00:53:49Z",
+  "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/embodied-olympics-fun-box-jump-i00-standard--fhhrrru/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-box-jump-i00-standard--fhhrrru/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "dd726f55beeb510f59288d59756e75b97d2f294d5c62b77d84d22b58b817203b",
+   "video_sha256": "5580a7592b5e8166533a2f33c826fc430e4c87f7fb84baafefdf39bd269358b0",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/box_jump/trajectories/embodied-olympics-fun-box-jump-i00-standard--fktgc5j.json

@@ -0,0 +1,196 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-box-jump-i00-standard--fktgc5j",
+ "taskKey": "embodied-olympics-fun/box_jump",
+ "dag_sha256": "94c6447a4b03c43587761e86e851339c4316a0cf5855b294978de845b6362c39",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 401,
+   "sim_steps": 11657,
+   "calls": 330,
+   "rows": 10088,
+   "resets": 50,
+   "reset_budget": 50,
+   "sim_startups": 1,
+   "elapsed_s": 3151.7,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 0.366,
+   "peak_processes": 11,
+   "peak_disk_gb": 0.421,
+   "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/embodied-olympics-fun-box-jump-i00-standard--fktgc5j/replay.mp4",
+  "duration_s": 4.04,
+  "fps": 25.0,
+  "frames": 101,
+  "sha256": "bd20f8235ac02bea7fde605ca3df6bfc6dd696ee00025f8c30766e57b25229ad"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "box_jump_stood": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.8,
+      "log_step": 208,
+      "text_zh": "观察：视频 0.00–2.60 s，机器人在箱前站立并小步接近；视频 2.80–3.36 s 屈膝下蹲，视频 3.40–3.76 s 伸腿后腾空，但双脚未上箱。log step 208 启动最后一次回合 p；此前重置回合不计入本轨迹。"
+     },
+     {
+      "t": 3.92,
+      "log_step": 209,
+      "text_zh": "观察：视频 3.72–4.00 s，机器人双脚向身后收起，身体向箱子反方向偏移，随后前倾下降，视频 4.00 s 膝部落到箱前地垫；全段没有双脚箱顶支撑或在箱顶站直。log step 209 的机器人服务状态为 episode=51、done=true、reason=failed，与评分字段 live_success=0 一致。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–3.64 s，机器人左右脚在箱前地垫承重、行走及蹬起；腾空后视频 4.00 s 膝部落在地垫上，未形成箱顶脚部支撑。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "air_s",
+   "value": null,
+   "unit": "s",
+   "threshold": 0.1,
+   "direction": "higher_better"
+  },
+  {
+   "name": "pelvis_rise_m",
+   "value": null,
+   "unit": "m",
+   "threshold": 0.1,
+   "direction": "higher_better"
+  },
+  {
+   "name": "pelvis_above_box_m",
+   "value": null,
+   "unit": "m",
+   "threshold": 1.0,
+   "direction": "higher_better"
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "box_jump_stood",
+   "t": 3.9,
+   "log_step": 208,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 3.72–4.00 s，机器人在跳箱过程中双脚收向身后，身体未到箱顶便前倾下降，以膝部落回箱前地垫；跳上箱并站直始终未完成。log step 209 的机器人服务随后报告本回合 done=true、reason=failed。"
+  },
+  "unmet": {
+   "box_jump_stood": false
+  },
+  "phenomena": [
+   {
+    "cat": "全身",
+    "item": "平衡",
+    "t": 3.96,
+    "log_step": null,
+    "evidence_zh": "观察：视频 3.84–4.00 s，机器人下降时躯干持续前倾，双脚仍在膝部后方，没有恢复脚底承重姿态，最终膝部着垫呈跪倒姿态，箱子仍在其前方。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "evidence_zh": "推断：log step 205 为最后回合设置向箱顶前伸双脚的落地姿态，log step 208 执行该回合；视频 3.60–4.00 s 的实际运动却向远离箱子的方向偏移，屈腿后未及时恢复脚底支撑，导致跪落地垫，支持起跳与落地的全身运动控制未实现目标。现有证据不能进一步确定是关节目标、力矩还是跟踪问题。",
+    "source_zh": "视频 / 日志"
+   }
+  ],
+  "gap_zh": "没有状态日志；评分器只报告失败，未给最后回合的具体违规项和 air_s、pelvis_rise_m、pelvis_above_box_m 数值，不能核实精确起跳判据、接触法线或控制子类。agent 在 log step 211 自述可能擦到箱侧，但视频 3.72–4.00 s 显示机器人落在箱前，未见触箱，因此不采用该候选解释。"
+ },
+ "summary_zh": "评分回合中机器人接近箱子、下蹲并腾空，但没有跳到箱顶，唯一节点“跳上箱并站直”始终���达成，最终为 0/1。视频 3.72–4.00 s 显示机器人偏离箱子、前倾并跪落地垫，断点为跳箱操作中，支持全身运动控制失败。缺少状态日志与具体裁判违规字段，无法确定控制子类；agent 在 log step 211 提出的擦箱侧解释没有画面支持。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "视频 2.8–4.0 s 帧与标注一致：机器人起跳后向箱子反方向腾空，未上箱，以膝部落在地垫；唯一节点未达成、断点和单一现象/根因有画面支持，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": "2026-10-31T00:53:52Z",
+  "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/embodied-olympics-fun-box-jump-i00-standard--fktgc5j/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-box-jump-i00-standard--fktgc5j/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "c4fb2e01e9c91295ef38a3f382179448db2a5fe671ae7459ee0875e6462bec53",
+   "video_sha256": "bd20f8235ac02bea7fde605ca3df6bfc6dd696ee00025f8c30766e57b25229ad",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/box_jump/trajectories/embodied-olympics-fun-box-jump-i00-standard--kigfv4k.json

@@ -0,0 +1,198 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-box-jump-i00-standard--kigfv4k",
+ "taskKey": "embodied-olympics-fun/box_jump",
+ "dag_sha256": "94c6447a4b03c43587761e86e851339c4316a0cf5855b294978de845b6362c39",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 243,
+   "sim_steps": 16810,
+   "calls": 548,
+   "rows": 3692,
+   "resets": 46,
+   "reset_budget": 50,
+   "sim_startups": 1,
+   "elapsed_s": 7025.7,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 0.573,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.873,
+   "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/embodied-olympics-fun-box-jump-i00-standard--kigfv4k/replay.mp4",
+  "duration_s": 2.44,
+  "fps": 25.0,
+  "frames": 61,
+  "sha256": "3146b2b5a282725a3ebc3da0afe4f58a10727e5c063a1e09f00c03f8228e3433"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "box_jump_stood": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 1.8,
+      "log_step": 285,
+      "text_zh": "观察：视频 0.00–1.80 s，机器人一直在棕色箱前地面，随后屈膝下蹲；视频 2.00–2.40 s，机器人蹬伸、短暂腾空并向前倾倒，双脚始终未落到箱顶，更未在箱顶站直保持。"
+     },
+     {
+      "t": 2.4,
+      "log_step": 285,
+      "text_zh": "观察：log step 285 的机器人服务返回最后第 47 回合 done=true、reason=failed、outcome=\"the jump landed on the floor\"，stages_reached=1/3，仿真结束时间为 2.43 s；视频 2.40 s 末帧中机器人前扑，脚已回到箱前地面，与该回合结果一致。评分字段 graded_by_live=1、n_actions=243，与该回合 control_steps=243 一致；前面重置回合的试验不计入本轨迹。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–2.00 s，机器人双脚在箱前地面承重、屈膝并蹬伸；视频 2.40 s，脚回落箱前地面。中间短暂腾空，未见双脚与箱顶建立支撑。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "air_s",
+   "unit": "s",
+   "threshold": 0.1,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "pelvis_rise_m",
+   "unit": "m",
+   "threshold": 0.1,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "pelvis_above_box_m",
+   "unit": "m",
+   "threshold": 1.0,
+   "direction": "higher_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "box_jump_stood",
+   "t": 2.2,
+   "log_step": 285,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 2.12–2.40 s，机器人腾空后躯干持续向前旋转，双脚拖在身后且未越上箱顶，末帧呈前扑姿态、脚落回箱前地面；log step 285 的机器人服务记录该最后回合仅达到起跳阶段，随后以“the jump landed on the floor”判负，跳上箱并站直始终未完成。"
+  },
+  "phenomena": [
+   {
+    "cat": "全身",
+    "item": "平衡",
+    "t": 2.2,
+    "log_step": 285,
+    "evidence_zh": "观察：视频 2.12–2.40 s，机器人由接近直立的腾空姿态不断前倾，头肩向箱子方向下落、双脚留在后方，末帧躯干大幅倾斜并前扑，未恢复脚在身体下方的稳定支撑；log step 285 的机器人服务确认跳后触地并终止回合。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "t": 2.2,
+    "log_step": 285,
+    "source_zh": "日志",
+    "evidence_zh": "推断：log step 285 中，agent 的控制脚本对双侧髋、膝、踝发送固定时长的关节目标插值来下蹲和蹬伸，随后切换到零速度 walk 控制器；视频 2.12–2.40 s 显示该序列未抑制躯干向前旋转，也未把双脚带到箱顶落点，因而本次跳跃的全身运动控制未成功。现有观测不足以进一步区分轨迹设计或反馈跟踪失稳。"
+   }
+  ],
+  "gap_zh": "没有逐控制步状态日志；视频可判定未上箱和前扑失衡，但不能精确测量起跳增高、腾空时长、足底接触法线或箱顶相对骨盆高度。评分器未报告 DAG 的三个原生指标数值，因此均填 null；控制根因的具体子类缺少实测目标跟踪与接触反馈证据。",
+  "unmet": {
+   "box_jump_stood": false
+  }
+ },
+ "summary_zh": "评分轨迹为 log step 285 的最后第 47 回合：机器人在箱前下蹲、蹬伸并短暂腾空，但随后前扑失衡、落回箱前地面，“跳上箱并站直”从未达到，最终为 0/1。机器人服务确认仅达到起跳阶段后因落地失败；日志中的控制序列与画面支持控制层面的归因，但缺少状态日志，不能确定具体控制子类或原生指标数值。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "标签与视频、log step 285 一致：最后第47回合起跳后前扑落回箱前地面，未上箱；节点 false、断点、平衡现象和控制根因（推断、未细分子类）均有证据支持。",
+    "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": "2026-10-31T00:53:57Z",
+  "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/embodied-olympics-fun-box-jump-i00-standard--kigfv4k/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-box-jump-i00-standard--kigfv4k/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "916fcb4f4c4b0caf855a2f1860d5ba6b256da95ed793894b9d91b9b86e1ac35d",
+   "video_sha256": "3146b2b5a282725a3ebc3da0afe4f58a10727e5c063a1e09f00c03f8228e3433",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/box_jump/trajectories/embodied-olympics-fun-box-jump-i00-standard--y5mse2x.json

@@ -0,0 +1,218 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-box-jump-i00-standard--y5mse2x",
+ "taskKey": "embodied-olympics-fun/box_jump",
+ "dag_sha256": "94c6447a4b03c43587761e86e851339c4316a0cf5855b294978de845b6362c39",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 277,
+   "sim_steps": 9303,
+   "calls": 342,
+   "rows": 9303,
+   "resets": 48,
+   "reset_budget": 50,
+   "sim_startups": 1,
+   "elapsed_s": 1450.7,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 0.403,
+   "peak_processes": 10,
+   "peak_disk_gb": 1.212,
+   "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/embodied-olympics-fun-box-jump-i00-standard--y5mse2x/replay.mp4",
+  "duration_s": 2.8,
+  "fps": 25.0,
+  "frames": 70,
+  "sha256": "3691c5835d8236a902a361a139b530091ff071eefcd18e699012b7a8a10ace06"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "box_jump_stood": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–1.80 s，机器人位于棕色箱前，双脚在地面，尚未到箱顶。"
+     },
+     {
+      "t": 2.6,
+      "log_step": 163,
+      "text_zh": "观察：视频 2.00–2.36 s，机器人屈膝后伸腿起跳；视频 2.40–2.76 s，机器人向箱子方向持续前倾，末帧身体倒向箱沿，双脚仍在箱前且低于顶面，全程没有双脚落在箱顶或箱顶站直保持。log step 163 的最后一次实时回合返回裁判结果“did not land on the feet”，stages_reached=1/3、control_steps=277；该结果与视频的失败落地一致。原生起跳阶段不单列为已完成语义节点。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–2.36 s，机器人双脚在箱前地面承重并蹬地；视频 2.40–2.76 s，机器人起跳后双脚离地，躯干前倾至接近水平并移到箱沿上方，脚未形成箱顶支撑；侧视图无法确认身体是否已触箱。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "air_s",
+   "unit": "s",
+   "threshold": 0.1,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "pelvis_rise_m",
+   "unit": "m",
+   "threshold": 0.1,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "pelvis_above_box_m",
+   "unit": "m",
+   "threshold": 1.0,
+   "direction": "higher_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "box_jump_stood",
+   "t": 2.6,
+   "log_step": 163,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 2.40–2.76 s，机器人起跳后逐渐前翻，身体倒向箱沿，未能以双脚落在箱顶；log step 163 的最后一次实时回合以“did not land on the feet”、1/3 原生阶段结束。随后只有脚本整理及最终报告（log step 165–166），没有修复这次评分回合。"
+  },
+  "unmet": {
+   "box_jump_stood": false
+  },
+  "phenomena": [
+   {
+    "cat": "全身",
+    "item": "平衡",
+    "t": 2.6,
+    "log_step": 163,
+    "evidence_zh": "观察：视频 2.40–2.76 s，机器人躯干从前倾发展为接近水平，双腿留在后下方，身体倒向箱沿；机器人没有建立双脚箱顶支撑，裁判在 log step 163 判为“did not land on the feet”。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "evidence_zh": "推断：log step 144 的脚本按固定关节关键帧执行下蹲、蹬伸、空中姿态、缓冲和站立，只检查裁判是否结束；log step 148、161 修改这些关键帧参数，log step 163 用该脚本执行最后一次实时回合。视频 2.40–2.76 s 中机器人未把双脚提至箱顶，却持续前翻，说明该跳跃控制未实现所需腾空姿态及落地支撑；证据不足以进一步区分轨迹设计与反馈跟踪问题。",
+    "source_zh": "日志 / 视频"
+   }
+  ],
+  "gap_zh": "无状态日志，无法确定末段首先触箱的具体身体链接、接触力和精确速度，也无法把控制问题细分。log step 166 中 agent 自述是膝/胫先触箱，但侧视画面仅能确认身体倒向箱沿、双脚未上箱顶，故不采用��具体链接判断。评分器未报告 air_s、pelvis_rise_m、pelvis_above_box_m 数值，三项原生指标留空。"
+ },
+ "summary_zh": "评分回合完成了裁判认可的起跳（log step 163：原生阶段 1/3），但“跳上箱并站直”节点始终未达成，最终为 0/1。视频 2.40–2.76 s 显示机器人起跳后前翻、身体倒向箱沿，未能双脚落到箱顶，断点为该跳箱步骤的操作中，根因为控制。缺少状态及接触力日志，不能精确确定首次触箱链接或进一步细分控制原因；成功参照视频可见箱顶双脚落地后伸腿站稳，与本回合末态明显不同。",
+ "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 box_jump_stood、平衡 + 控制的结论与视频一致（视频 2.8 s ≈ 277 控制步，对应评分回合）；仅有一处接触观察写得过满。",
+    "n_issues": 1,
+    "n_major": 0
+   },
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 2,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "节点未达成、断点 box_jump_stood、平衡 + 控制与视频及 log step 163（did not land on the feet，1/3，277 步）一致；上一轮的接触措辞已修正，未发现新错误。",
+    "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": "2026-10-31T01:08:12Z",
+  "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/embodied-olympics-fun-box-jump-i00-standard--y5mse2x/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-box-jump-i00-standard--y5mse2x/04_review_r1"
+   },
+   {
+    "stage": "reconcile",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-box-jump-i00-standard--y5mse2x/05_reconcile_r1"
+   },
+   {
+    "stage": "traj_review",
+    "round": 2,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-box-jump-i00-standard--y5mse2x/04_review_r2"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "0a05b09a59f4622be1a6f1f11b2db76ca566cef4226c988d6463cb96cd951439",
+   "video_sha256": "3691c5835d8236a902a361a139b530091ff071eefcd18e699012b7a8a10ace06",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/box_jump/trajectories/embodied-olympics-fun-box-jump-i00-standard--yldfzxs.json

@@ -0,0 +1,210 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-box-jump-i00-standard--yldfzxs",
+ "taskKey": "embodied-olympics-fun/box_jump",
+ "dag_sha256": "94c6447a4b03c43587761e86e851339c4316a0cf5855b294978de845b6362c39",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 257,
+   "sim_steps": 2003,
+   "calls": 349,
+   "rows": 2003,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 5644.2,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.665,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.939,
+   "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/embodied-olympics-fun-box-jump-i00-standard--yldfzxs/replay.mp4",
+  "duration_s": 2.6,
+  "fps": 25.0,
+  "frames": 65,
+  "sha256": "cbbde695e5c0188e16e7d32be780f8ecdb9e8fa22e8e95b09bc46a8d4d1a80fa"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "box_jump_stood": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": 221,
+      "text_zh": "观察：视频 0.00 s 机器人站在箱前地面，视频 0.40–2.00 s 下蹲并前倾；log step 221 的服务返回 episode=6、control_steps=0、resets_left=0，log step 252 最终仍为 episode=6。因此只标最后一次实时回合，不将前五次测试或独立模型演练计入节点。"
+     },
+     {
+      "t": 2.24,
+      "log_step": 230,
+      "text_zh": "观察：视频 2.12–2.24 s 机器人蹬伸双腿、离开地面，视频 2.28–2.40 s 腾空收腿向箱顶移动；视频 2.44–2.56 s 脚到达箱缘附近，躯干仍前倾、膝关节弯曲，没有出现箱顶站直保持。"
+     },
+     {
+      "t": 2.56,
+      "log_step": 252,
+      "text_zh": "观察：log step 252 的机器人实时服务终局字段为 outcome=\"landed on the side of the box\"、stages_reached=1/3、control_steps=257、sim_time_s=2.57；log step 253 再次读取同一终局。推断：该回合只通过起跳检查，侧面接触使落地验收失败，跳上箱并站直节点从未成立。控制周期为 log step 7 的 0.01 s，257×0.01=2.57 s，与视频 2.56 s 末帧的箱缘接触位置核对一致；此处仅据控制步对齐两种证据，未使用 agent 工作时间。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–2.12 s 机器人左右脚在地面承重并蹬伸起跳；中途腾空，未完成合规箱顶双脚站稳。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频 2.44–2.56 s 机器人脚到达箱缘，log step 252 的服务裁判确认落地时接触箱侧面。候选名称：碰撞接触；这是脚与箱侧面的非预期接触，不能确认为承重支撑，也不是主动推扫、压擦或击打。单一侧视不能可靠区分是哪一只脚触及哪个侧面。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "air_s",
+   "unit": "s",
+   "threshold": 0.1,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "pelvis_rise_m",
+   "unit": "m",
+   "threshold": 0.1,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "pelvis_above_box_m",
+   "unit": "m",
+   "threshold": 1.0,
+   "direction": "higher_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "box_jump_stood",
+   "t": 2.56,
+   "log_step": 252,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 2.44–2.56 s 机器人正从腾空收腿转为落箱，脚落向箱缘附近，尚未站直；log step 252 的第六回合实时服务在 control_steps=257 时判定 landed on the side of the box，只通过 1/3 阶段并立即结束。推断：失败发生在跳箱步骤的落地过程中，之后没有修复机会。"
+  },
+  "unmet": {
+   "box_jump_stood": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "被挡/碰撞",
+    "t": 2.56,
+    "log_step": 252,
+    "evidence_zh": "观察：视频 2.40–2.56 s 机器人下降时脚来到箱顶边缘，未形成双脚都在顶面内的稳定支撑；log step 252 的服务裁判明确报告 landed on the side of the box，��触箱侧面终止了本次跳跃。视频截至末帧没有显示机器人倒地。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "认知",
+    "item": "感知",
+    "t": 2.52,
+    "log_step": 252,
+    "source_zh": "日志 / 视频 / agent 自述",
+    "evidence_zh": "观察：视频 2.48、2.52、2.56 s 中机器人相对箱子连续向前运动，没有约半米的后退。log step 252 中 agent 脚本打印的 sensor pose 前后位置分量却从 0.4919 m 突变到 -0.0100 m，随后接近 0 m，同时 IKerr 从 0.00035 升到 0.858–1.055；这些是脚本估计值而非真实状态。log step 194、247、251 的代码显示，脚本每次将 estimate3 输出的位置直接送入固定世界足端目标的逆解，再立即发送控制指令；log step 253 中 agent 自述近面离开视野后位置估计跳变并产生错误末段指令。推断：明显错误的位置估计进入落地控制，造成末段足端修正失准，促成侧面接触；近面误识别的具体图像机制未能独立核实。"
+   }
+  ],
+  "gap_zh": "没有仿真状态日志；视频只有场地侧视，无法独立分辨触箱的是左脚还是右脚、具体侧面及接触法线，侧面接触类型由最后实时回合的服务裁判终局确认。位置估计突变及其进入逆解有日志和代码证据，但没有真实三维轨迹或对照试验，不能量化它与此前双脚外展各自对失败的贡献。评分记录未报告 air_s、pelvis_rise_m、pelvis_above_box_m 的数值，原生指标保留 null。"
+ },
+ "summary_zh": "最后一次评分回合中，机器人从地面下蹲、蹬伸并腾空向箱顶移动，但落地触及箱侧面，裁判只确认起跳阶段，跳上箱并站直节点始终未达成，最终为 0/1。落地前脚本的位置估计突变约 0.50 m，并直接用于足端逆解，支持感知错误促成末段落地失准的归因。缺少真实状态日志和其他视角，无法精确确认触箱脚及接触法线，也无法量化该估计错误与已有足端外展的各自贡献。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "节点未达成、断点在跳箱落地、被挡/碰撞与感知根因均有视频末段（脚落箱缘）、log step 252 服务终局及脚本位置估计突变支持；未发现需修改的内容。",
+    "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": "2026-10-31T01:01:28Z",
+  "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/embodied-olympics-fun-box-jump-i00-standard--yldfzxs/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-box-jump-i00-standard--yldfzxs/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "6435f94af0749759081b340d7a98a92ae6317c4c52ec60e1dc3704054dcf34ca",
+   "video_sha256": "cbbde695e5c0188e16e7d32be780f8ecdb9e8fa22e8e95b09bc46a8d4d1a80fa",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch_and_shoot/trajectories/embodied-olympics-fun-catch-and-shoot-i00-privileged--c5wpdf4.json

@@ -0,0 +1,191 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-and-shoot-i00-privileged--c5wpdf4",
+ "taskKey": "embodied-olympics-fun/catch_and_shoot",
+ "dag_sha256": "f6a6d84d578d7c1084b9432c694b8d43d8fa3421ae8d0142dbb9a316794fb326",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "n_actions": 337,
+   "sim_steps": 337,
+   "horizon": 1500,
+   "episode_over": 1,
+   "progress": 0.0,
+   "stages_reached": 0,
+   "stages_total": 3,
+   "fell": 0,
+   "replay_s": 5.16,
+   "bad_trajectory": 0,
+   "metric_ball_height_m": 0.114,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 146,
+   "agent_resets": 155,
+   "agent_replays": 0,
+   "agent_cost_self_reported": 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/embodied-olympics-fun-catch-and-shoot-i00-privileged--c5wpdf4/replay.mp4",
+  "duration_s": 3.4,
+  "fps": 25.0,
+  "frames": 85,
+  "sha256": "977769ec7a26cace0c1515c26b8f07b8d94540aefbab68ecd0d33c7207329111"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "ball_caught": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.4,
+      "log_step": null,
+      "text_zh": "观察：视频 2.28–2.40 s，传球机送出的橙色球到达机器人抬起的迎球手附近；视频 2.44–2.52 s 球已与双手分开并向画面右上方飞去，未出现两手共同承接并持续控制的阶段。"
+     },
+     {
+      "t": 3.36,
+      "log_step": null,
+      "text_zh": "观察：视频 2.56–3.36 s，篮球持续远离机器人，最后落到禁区地板，机器人保持站立、双手空；inputs/run.json 的评分器字段 stages_reached=0、progress=0、deterministic=1，与未完成接球一致。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "其他",
+      "note_zh": "推断：视频 2.36–2.44 s，篮球经过机器人抬起的左手附近后改变飞行方向并弹离；左手发生瞬时碰球，右手虽在身前准备承接，但未见与球共同接触。候选：击打（接球时的短暂碰触反弹）；球未被单手包住，也未被双掌夹住，不能标抓握或持续托举。具体指、掌接触部位受远景遮挡。"
+     }
+    ]
+   },
+   "shot_through_ring": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.52,
+      "log_step": null,
+      "text_zh": "观察：视频 2.44–3.36 s，球从接球动作中弹离，飞行高度始终低于篮圈，随后落地；机器人未先接住球再进行投篮，球也未从上方穿过篮圈。"
+     },
+     {
+      "t": null,
+      "log_step": 191,
+      "text_zh": "观察：log step 162 编写接球、闭合和小幅跟随的 gen.py，log step 163 修复向量长度并生成轨迹；log step 191 明确重新执行 gen.py 恢复最终交付文件。log step 226 自述交付为 337 行、接球未完成；评分器记录 n_actions=337、stages_reached=0。后续测试中的近接球及躯干触球不能算作此交付轨迹的投篮。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [],
+ "failure": {
+  "breakpoint": {
+   "ref": "ball_caught",
+   "t": 2.4,
+   "log_step": null,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 2.36–2.48 s，机器人正在迎接来球，球经过迎球手后立即弹向画面右上方，未被两手共同控制；视频 2.52–3.36 s 球一直远离双手，最后落地，接球失败未修复。"
+  },
+  "unmet": {
+   "ball_caught": false,
+   "shot_through_ring": false
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没建立接触",
+    "t": 2.4,
+    "log_step": null,
+    "evidence_zh": "观察：视频 2.36–2.48 s 只有迎球手附近的瞬时拦碰，球随即飞离，另一只手未把球合围接住；视频 2.52–3.36 s 双手始终空着。推断：接球要求的双手持续有效接触从未建立，不能把一次碰球算作持球后丢失。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "抓取和接触设计",
+    "alt_feasible": "已验证有",
+    "source_zh": "视频 / 日志",
+    "evidence_zh": "观察：log step 162 的交付脚本预设两手迎球姿态，在固定控制步闭合手指并作小幅跟随，log step 191 重新执行此脚本生成最终轨迹；视频 2.32–2.52 s 双手分开，来球只经过迎球侧��附近便被弹出，没有进入双手共同承接的区域。推断：交付动作的接触布局与闭合、跟随组合未形成能够捕获来球的双手接触，属于接触设计失败。成功参考 inputs/oracle/replay.mp4 的参考视频 2.44–3.00 s 可见球被双手持续承接，说明存在可行替代；参考视频仅用于验证替代可行，不作为本次接触证据。"
+   }
+  ],
+  "gap_zh": "本次没有状态日志；已逐帧检查视频 2.08–3.36 s，但远景中手球相交瞬间受手臂与身体遮挡，不能精确确定各指、掌接触对及毫秒接触时长。未完成双手控制由接触前后的连续画面及评分器阶段数共同支持。log step 184 及后续测试采用不同姿态，最终自述混入了这些测试中的掉球、躯干犯规；这些细节不用于本次现象或节点判定。"
+ },
+ "summary_zh": "本次交付在双手接球阶段失败，两个节点均未达成（0/2）：视频约 2.4 s 来球经过迎球手后弹离，至视频 3.36 s 落地，机器人未摔倒。双手接球接触未建立，交付动作的接触布局与闭合、跟随组合未能捕获来球；log step 191 确认最终文件重新由 gen.py 生成，后续实验中的近接球和躯干犯规不属于这段视频。无状态日志，细小接触部位及精确接触时长不能确定。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "两个节点均未达成、断点 ball_caught（操作中）、接触现象与接触设计根因均与视频和日志相符，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": "2026-10-31T01:06:57Z",
+  "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/embodied-olympics-fun-catch-and-shoot-i00-privileged--c5wpdf4/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-privileged--c5wpdf4/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "660da50fc5303661dbfe7dd0cc244e4d2a380b555871a337e64815eff576a703",
+   "video_sha256": "977769ec7a26cace0c1515c26b8f07b8d94540aefbab68ecd0d33c7207329111",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch_and_shoot/trajectories/embodied-olympics-fun-catch-and-shoot-i00-privileged--gzgbxmp.json

@@ -0,0 +1,192 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-and-shoot-i00-privileged--gzgbxmp",
+ "taskKey": "embodied-olympics-fun/catch_and_shoot",
+ "dag_sha256": "f6a6d84d578d7c1084b9432c694b8d43d8fa3421ae8d0142dbb9a316794fb326",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "n_actions": 661,
+   "sim_steps": 661,
+   "horizon": 1500,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 3,
+   "stages_total": 3,
+   "fell": 0,
+   "replay_s": 5.81,
+   "bad_trajectory": 0,
+   "metric_ball_height_m": 2.775,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 194,
+   "agent_resets": 320,
+   "agent_replays": 32,
+   "agent_cost_self_reported": 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/embodied-olympics-fun-catch-and-shoot-i00-privileged--gzgbxmp/replay.mp4",
+  "duration_s": 6.64,
+  "fps": 25.0,
+  "frames": 166,
+  "sha256": "a4b9add3bb5d2c19c65747713771ec3501d3886ecbc5a3bc583811624dfe0f66"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "ball_caught": {
+    "method": "visual",
+    "intervals": [
+     [
+      2.64,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 2.64,
+      "log_step": null,
+      "text_zh": "观察：视频 2.20–2.44 s，橙色篮球从左侧传球机方向飞到机器人胸前；视频 2.48–2.88 s，球停留在机器人两手之间并随两臂一起移动，没有弹走或先落地。视频 2.64 s 作为双手接住事件的近似完成锚点。"
+     },
+     {
+      "t": 2.64,
+      "log_step": 242,
+      "text_zh": "观察：log step 242 的所选交付候选重放结果列出 catch、release、through the ring 三阶段；log step 260 核验最终交付的 661 行动作精确再生成、两次新进程重放成功且末态哈希与该候选相同。inputs/run.json 评分字段 replay_success=1、deterministic=1、stages_reached=3。"
+     },
+     {
+      "t": 2.64,
+      "log_step": null,
+      "text_zh": "推断：结合视频 2.48–2.88 s 的持续双手控球及评分器三个阶段全部通过，双手接触达到接球判据；远景不能独立量出每手接触宽限和连续接触时长。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 2.48–2.88 s，机器人左右手共同在篮球两侧承接并相向夹持球体。候选：夹持；画面未显示任何单手以拇指与其余手指相对合拢包住球，双掌共同夹压不按抓握标注，现有条目无准确对应。"
+     }
+    ]
+   },
+   "shot_through_ring": {
+    "method": "visual",
+    "intervals": [
+     [
+      6.6,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 4.8,
+      "log_step": null,
+      "text_zh": "观察：视频 3.20–4.64 s，机器人将两手间的篮球下放至腹部附近；视频 4.68–4.76 s 双臂一起向前上方摆动，视频 4.80 s 球已与双手分离，随后沿高弧线飞向右侧篮筐。"
+     },
+     {
+      "t": 6.6,
+      "log_step": null,
+      "text_zh": "观察：逐帧查看视频 5.60–6.60 s，球在视频约 6.12 s 从画面上方重新出现，视频 6.32–6.60 s 连续下降至篮板前的篮圈及圈下网区；机器人仍站立。投篮命中完成时间近似记为视频 6.60 s。"
+     },
+     {
+      "t": 6.6,
+      "log_step": 260,
+      "text_zh": "观察：inputs/run.json 的评分结果 success=1、replay_success=1、deterministic=1、stages_reached=3、fell=0；log step 260 对最终交付报告精确再生成和两次成功重放。"
+     },
+     {
+      "t": 6.6,
+      "log_step": null,
+      "text_zh": "推断：视频末段的向下入圈轨迹与评分器成功结果相互支持，投篮命中节点已完成；远景及网篮遮挡不足以独立精确测量球心穿越两个验收平面的高度与水平距离。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 3.20–4.76 s，机器人两手共同夹持、下放并向前上方加速篮球，随后球离手飞行。候选：夹持；两手共同夹压与送球，未见单手相对合拢包住球的抓握，现有接触方式无准确对应；视频 4.80 s 后的飞行由球自身运动完成。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [],
+ "failure": null,
+ "summary_zh": "机器人在视频约 2.6 s 双手接住左侧来球，持续控球后于视频约 4.8 s 以下手动作投出，视频约 6.6 s 球从上方下降入圈，两个事件节点均完成。评分器确认 success=1、replay_success=1、deterministic=1，最终交付的精确再生成和两次成功重放见 log step 260，无失败断点。没有状态日志，远景无法独立测量毫秒级双手接触及精确穿圈几何，相关验收结合可见动作和评分结果判定。",
+ "conf": 9,
+ "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，两个事件节点时间与帧一致，评分器 success=1 支持；接触标注合理，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": "2026-10-31T01:08:28Z",
+  "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/embodied-olympics-fun-catch-and-shoot-i00-privileged--gzgbxmp/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-privileged--gzgbxmp/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "feec6b7d6d4d7a9537bf5aaf31b6e45493b703f8ebc1130633bc25f7c4442da0",
+   "video_sha256": "a4b9add3bb5d2c19c65747713771ec3501d3886ecbc5a3bc583811624dfe0f66",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch_and_shoot/trajectories/embodied-olympics-fun-catch-and-shoot-i00-privileged--k2dnyf8.json

@@ -0,0 +1,189 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-and-shoot-i00-privileged--k2dnyf8",
+ "taskKey": "embodied-olympics-fun/catch_and_shoot",
+ "dag_sha256": "f6a6d84d578d7c1084b9432c694b8d43d8fa3421ae8d0142dbb9a316794fb326",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "n_actions": 525,
+   "sim_steps": 525,
+   "horizon": 1500,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 3,
+   "stages_total": 3,
+   "fell": 0,
+   "replay_s": 6.96,
+   "bad_trajectory": 0,
+   "metric_ball_height_m": 2.773,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 62,
+   "agent_resets": 235,
+   "agent_replays": 1,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.795,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.255,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-catch-and-shoot-i00-privileged--k2dnyf8/replay.mp4",
+  "duration_s": 5.28,
+  "fps": 25.0,
+  "frames": 132,
+  "sha256": "8a4696a65d91eede0c0357ae99bee53262cd6416648f658fcea0bf0d3b79ae9c"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "ball_caught": {
+    "method": "visual",
+    "intervals": [
+     [
+      2.56,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 2.32,
+      "log_step": null,
+      "text_zh": "观察：视频 2.20–2.32 s，橙色篮球从画面左侧传球机方向飞向机器人的胸前，来球在空中，未见先触地；视频 2.36–2.88 s 球被机器人上身部分遮挡。"
+     },
+     {
+      "t": 3.2,
+      "log_step": 145,
+      "text_zh": "观察：视频 3.12–3.44 s，机器人转向篮筐，球重新显露在其胸腹前，随双手移动并保持受控；此后球随双手降低至视频 4.00 s，未弹走。log step 145 的已交付文件两次独立重放均报告 catch、release、through the ring 阶段及相同末态哈希，裁判事件记录接球成立于仿真 2.540 s。"
+     },
+     {
+      "t": 2.56,
+      "log_step": 145,
+      "text_zh": "推断：接球事件取视频约 2.56 s，并在出手后保持锁存。交付重放在 log step 145 报告 525 控制步、仿真 5.25 s，结合 log step 141 脚本的每步 0.01 s，与视频 5.28 s（25 fps）长度相符；裁判释放事件的仿真 4.129 s 已在视频 4.12–4.16 s 球离手处核对，故用同一时钟对应关系辅助定位被遮挡的接球成立时刻，误差约一帧。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "未知",
+      "note_zh": "观察：视频 3.12–3.44 s，机器人左右手在胸腹前共同控制篮球，球随双臂转向并保持在手边。远景与背侧视角无法分辨每手是否以拇指和其余手指形成对向包握，或主要由两手相向夹压球面，故具体接触方式未知。"
+     }
+    ]
+   },
+   "shot_through_ring": {
+    "method": "visual",
+    "intervals": [
+     [
+      5.24,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 4.12,
+      "log_step": null,
+      "text_zh": "观察：视频 3.52–4.00 s，机器人把仍受双手控制的球降至腹前，视频 4.04–4.12 s 双臂向前上方送球；视频 4.16 s 球已明显离开双手，随后沿弧线飞向画面右侧篮圈。"
+     },
+     {
+      "t": 5.24,
+      "log_step": 145,
+      "text_zh": "观察：视频 4.40–5.24 s 已逐帧检查，球越过最高点后下降，视频 5.12–5.24 s 球从篮圈上方向圈口下降并进入圈下区域；末帧机器人仍直立。log step 145 的交付文件重放记录 through the ring 与 success 同在仿真 5.243 s 触发，两次重放 success=true、failed=null；本次 reward 的 replay_success=1、deterministic=1、stages_reached=3。"
+     },
+     {
+      "t": 5.24,
+      "log_step": 145,
+      "text_zh": "推断：结合视频末段的下降入圈轨迹及交付重放的成功事件，投篮命中约在视频 5.24 s 成立；接球前置已满足。远景无法直接测量精确穿圈距离和圈下下降一球直径的阈值，这些验收细节由本次裁判成功结果确认。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "未知",
+      "note_zh": "观察：视频 3.52–4.12 s，机器人双手共同控制篮球下放并向前上方加速，视频 4.16 s 后球离手自由飞行。背侧远景不能可靠区分指掌包握、两手夹持或下方承托各自的接触份额，故具体方式未知。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [],
+ "failure": null,
+ "summary_zh": "机器人在视频约 2.56 s 双手接住左侧来球，转向并下放蓄力后于视频约 4.12–4.16 s 出手，视频约 5.24 s 投篮命中，两节点均完成。视频末段的下降入圈与交付重放及评分器成功结果一致，无失败断点。没有状态日志，背侧远景遮挡接球瞬间且无法区分手指包握和双手夹压；接球时刻结合交付重放事件定位，精确接触时长及穿圈阈值由裁判结果确认。",
+ "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，两节点均达成，时间与帧一致（接球约2.56 s、出手约4.16 s、入圈约5.24 s），接触标注为双臂协作+未知，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": "2026-10-31T01:06:22Z",
+  "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/embodied-olympics-fun-catch-and-shoot-i00-privileged--k2dnyf8/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-privileged--k2dnyf8/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "6763c1cc4b2378c2b2e55bf05e69bdbb507f83bbc4577c1d28aa686742e35902",
+   "video_sha256": "8a4696a65d91eede0c0357ae99bee53262cd6416648f658fcea0bf0d3b79ae9c",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch_and_shoot/trajectories/embodied-olympics-fun-catch-and-shoot-i00-privileged--ka7kvqp.json

@@ -0,0 +1,194 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-and-shoot-i00-privileged--ka7kvqp",
+ "taskKey": "embodied-olympics-fun/catch_and_shoot",
+ "dag_sha256": "f6a6d84d578d7c1084b9432c694b8d43d8fa3421ae8d0142dbb9a316794fb326",
+ "outcome": {
+  "success": 0,
+  "reward": 0.6667,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "n_actions": 301,
+   "sim_steps": 301,
+   "horizon": 1500,
+   "episode_over": 1,
+   "progress": 0.6667,
+   "stages_reached": 2,
+   "stages_total": 3,
+   "fell": 0,
+   "replay_s": 4.01,
+   "bad_trajectory": 0,
+   "metric_ball_height_m": 0.11,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 443,
+   "agent_resets": 5,
+   "agent_replays": 5,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.564,
+   "peak_processes": 11,
+   "peak_disk_gb": 0.855,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-catch-and-shoot-i00-privileged--ka7kvqp/replay.mp4",
+  "duration_s": 3.04,
+  "fps": 25.0,
+  "frames": 76,
+  "sha256": "62c23795ad79d799dbf6348134c4566811d185b4cb3f687de5f5ea3fadd20818"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "ball_caught": {
+    "method": "visual",
+    "intervals": [
+     [
+      2.52,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 2.52,
+      "log_step": 497,
+      "text_zh": "观察：视频 2.00–2.40 s，橙色篮球由机器人左侧传球机飞来，未触地；视频 2.44–2.64 s，球到达机器人胸前双手间并短暂随手移动。log step 497 对最终交付文件的重放记录了 catch 与 release，catch 事件为仿真 2.507 s；评分器 stages_reached=2，与画面中的接球相符。接球是锁存事件，后续掉球不撤销该节点。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "未知",
+      "note_zh": "观察：视频 2.44–2.72 s，机器人左、右手共同承接并随球向下移动；球和两手局部重叠，远景及裁剪仍不足以分辨单手的手指包握与双掌夹持，故接触方式标未知。"
+     }
+    ]
+   },
+   "shot_through_ring": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.8,
+      "log_step": 497,
+      "text_zh": "观察：视频 2.64–2.80 s，机器人双手与球下降；视频 2.84–3.00 s，球与双手分离并落向脚前地板，未飞向画面右侧篮圈，也未穿圈。log step 497 的最终文件重放仅有 catch、release 阶段，并以 the ball touched the floor 结束；log step 498 再次重放得到相同末态哈希。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "未知",
+      "note_zh": "观察：视频 2.52–2.80 s，机器人双手在接球后继续伴随篮球下降，随后球脱手；这是投篮前持球阶段的实际双手接触，未见向篮筐的送球动作。远景不能判明各手是否包握，方式标未知。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [],
+ "failure": {
+  "breakpoint": {
+   "ref": "shot_through_ring",
+   "t": 2.8,
+   "log_step": 497,
+   "where": "操作中",
+   "evidence_zh": "观察：视频约 2.52 s 已接住球，随后双手随球下降，视频约 2.80 s 球脱离双手，至视频 3.00 s 落在机器人脚前，投篮命中未完成；机器人仍站立。log step 497 明确保留接球轨迹，最终文件重放在球触地时结束，未交付会话中另行试验的向上投射动作。"
+  },
+  "unmet": {
+   "ball_caught": true,
+   "shot_through_ring": false
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没保持住",
+    "t": 2.8,
+    "log_step": 497,
+    "evidence_zh": "观察：视频 2.44–2.64 s，机器人两手先共同承接篮球；视频 2.68–2.80 s，球与双手一起下降，视频 2.84–3.00 s 球已在手下方继续下落，未被重新接住。log step 68 所列保存脚本始终保持双手闭合命令并跟随球位置，没有主动张手释放段；最终 log step 497 重新运行该脚本得到相同的球触地结果。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "source_zh": "日志",
+    "evidence_zh": "推断：log step 68、126 展示的保存脚本在双手接触后固定手相对篮球的偏移，每步以当前球位置求解双臂目标，双手闭合命令持续为 1.0；它没有后续向篮筐加速或主动释放段。log step 497 选择保留并重新生成此接球轨迹，log step 498 验证两次重放一致；结合视频 2.64–3.00 s 的随球下降与脱手，交付控制未能维持持球并衔接投射。具体是接触承载几何还是轨迹跟踪失配，现有证据不足以细分。"
+   }
+  ],
+  "gap_zh": "无状态日志；视频为远景，虽能确定双手承接、随后脱手落地，但不能独立测量有效接触的 0.10 秒窗口或各手局部接触形式。接球成立同时由最终交付文件的裁判重放记录支持；控制子类不作进一步推定。"
+ },
+ "summary_zh": "评分轨迹在视频约 2.52 s 完成双手接球，最终达到 1/2 个语义节点；接球事件在掉球后仍保留。投篮前的持球跟随未保持住篮球，视频约 2.80 s 脱手后直接落地，未形成向篮筐的投射；日志确认最终交付的是这一接球轨迹，两次重放一致。无状态日志，远景无法判清手指包握或双掌夹持，也不足以细分控制根因。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "接球节点（视频约 2.5 s）、投篮未达成、断点 shot_through_ring、没保持住（球沿双臂滑落，约 2.8 s 起脱手落地）均与视频和 log step 497/498（stages catch+release，球触地，重放哈希一致）相符；未发现需改动的错误。根因“控制”写得保守，已声明不细分。",
+    "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": "2026-10-31T01:08:45Z",
+  "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/embodied-olympics-fun-catch-and-shoot-i00-privileged--ka7kvqp/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-privileged--ka7kvqp/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "e728e208fb923a5f140149daf939a34de844d3d3e4628a3a7e32208e9abee2c8",
+   "video_sha256": "62c23795ad79d799dbf6348134c4566811d185b4cb3f687de5f5ea3fadd20818",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch_and_shoot/trajectories/embodied-olympics-fun-catch-and-shoot-i00-privileged--mdrhw2j.json

@@ -0,0 +1,202 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-and-shoot-i00-privileged--mdrhw2j",
+ "taskKey": "embodied-olympics-fun/catch_and_shoot",
+ "dag_sha256": "f6a6d84d578d7c1084b9432c694b8d43d8fa3421ae8d0142dbb9a316794fb326",
+ "outcome": {
+  "success": 0,
+  "reward": 0.6667,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "n_actions": 304,
+   "sim_steps": 304,
+   "horizon": 1500,
+   "episode_over": 1,
+   "progress": 0.6667,
+   "stages_reached": 2,
+   "stages_total": 3,
+   "fell": 0,
+   "replay_s": 4.93,
+   "bad_trajectory": 0,
+   "metric_ball_height_m": 0.104,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 106,
+   "agent_resets": 677,
+   "agent_replays": 6,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 1.807,
+   "peak_processes": 16,
+   "peak_disk_gb": 0.981,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-catch-and-shoot-i00-privileged--mdrhw2j/replay.mp4",
+  "duration_s": 3.08,
+  "fps": 25.0,
+  "frames": 77,
+  "sha256": "247afd7c85a0fa69183931e6e210fe47b9d95dbae67b0675fafa0077736f4ba8"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "ball_caught": {
+    "method": "visual",
+    "intervals": [
+     [
+      2.6,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 2.6,
+      "log_step": 223,
+      "text_zh": "观察：视频 2.08–2.44 s 橙球从画面左侧飞向机器人；视频 2.48、2.52、2.56、2.60 s 球在胸前双手之间停留，明显减速，未先落地。log step 223 对最终保存的 304 行轨迹重放，裁判阶段为 catch、release；log step 231 两次重放结果相同，支持双手接球已通过。接球达成时刻取视频约 2.6 s，事件此后锁存。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "推断：视频 2.48–2.64 s 左右手共同从球两侧承接、夹住篮球；log step 223 的交付脚本采用固定双掌姿态，双手闭合命令均为 1。候选：夹持；画面和脚本说明支持两手相向夹球，未见单手拇指与其余手指分别包住球面，故不标抓握，现有条目缺少双掌夹持。"
+     }
+    ]
+   },
+   "shot_through_ring": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.76,
+      "log_step": 223,
+      "text_zh": "观察：视频 2.60–2.72 s 机器人双臂仍维持接球姿态；视频 2.76–3.04 s 球从手下持续下落至脚旁地面，始终未飞向画面右侧篮圈。log step 223 最终交付脚本在全局始终发送同一 stand 行，没有接球后的挥臂或主动张手投篮段；最终文件重放只达 catch、release，失败原因是球触地。"
+     },
+     {
+      "t": 3.04,
+      "log_step": 231,
+      "text_zh": "观察：视频 3.04 s 末帧球在机器人脚边，篮圈未有进球；log step 231 对最终文件的两次重放均 success=False、stages=[catch, release]、failed=the ball touched the floor，且末态哈希相同。会话中 log step 214、218、229 的挥臂测试未被写入此最终固定姿态轨迹，不能算作本次视频中的投篮。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "推断：视频约 2.60–2.72 s 投篮前持球阶段仍沿用双手从两侧夹球的接触，随后球从手间向下脱离；候选：夹持。两手共同夹球而非单手各自形成包握，现有条目缺少双掌夹持；交付轨迹没有实际送球挥臂。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [],
+ "failure": {
+  "breakpoint": {
+   "ref": "shot_through_ring",
+   "t": 2.72,
+   "log_step": 223,
+   "where": "操作中",
+   "evidence_zh": "观察：视频约 2.6 s 接球成立后，机器人仍把双手停在胸前；视频 2.72–2.80 s 球向手下脱离并下落，至视频 3.04 s 落到脚旁地面，没有开始向篮圈的飞行。log step 223 保存并重放的最终轨迹只含固定接球姿态，裁判已记 catch、release 后因球触地判负；断点是投篮命中的持球准备阶段。"
+  },
+  "unmet": {
+   "ball_caught": true,
+   "shot_through_ring": false
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没保持住",
+    "t": 2.72,
+    "log_step": 223,
+    "evidence_zh": "观察：视频 2.48–2.64 s 球曾在双手之间被承接，视频 2.76、2.80、2.84 s 球逐渐降到双手下方，之后继续落地，机器人双臂没有送球挥动。log step 223 的交付代码始终保持双手闭合命令为 1，没有计划松手；这是接住后失去持球接触。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "抓取和接触设计",
+    "alt_feasible": "已验证有",
+    "t": 2.72,
+    "log_step": 223,
+    "source_zh": "日志 / 视频",
+    "evidence_zh": "推断：log step 223 最终脚本从开始到结束重复同一组双臂目标并保持闭合命令为 1，采用双掌夹球；视频 2.48–2.64 s 能短暂承接球，但视频 2.76–3.04 s 球在双臂基本不动时从手间下滑至地面，说明该接触姿态不能持续承载篮球以衔接投篮。可行替代已由同任务成功参考验证：参考视频 2.60–3.40 s 双手持续持球，参考视频 3.60–4.60 s 降低持球位置后于参考视频 4.80 s 向上送球，参考视频 6.48–6.60 s 球落入篮圈下方；这些只证明替代持球与送球方案可行，不作为本次运行的进度。"
+   }
+  ],
+  "gap_zh": "没有状态日志；远景及裁剪放大仍不足以精确辨别每根手指的接触、接触力和裁判的毫秒接触窗口，接球时刻为视频近似值，并以最终交付文件重放的 catch 阶段确认。具体掌面受力不足的数值原因无法确定；agent 在 log step 238 自述这些手臂无法完成投篮，仅是其结论，成功参考表明存在可行方案，不据此归因为任务或环境。"
+ },
+ "summary_zh": "本次交付轨迹在视频约 2.6 s 完成双手接球，最终完成 1/2 个 DAG 节点；裁判原生 2/3 阶段进度包含 release，不等于投篮命中。机器人接球后保持固定双掌姿态，球于视频约 2.72 s 开始从手下脱离，视频 3.04 s 触地，断在投篮前持球阶段，接触设计未能持续承载篮球。无状态日志，细小手指接触及精确受力不可核实，接球验收结合最终文件的确定性重放结果判断。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "接球达成、投篮未达成、断点在投篮持球阶段，现象（没保持住）与根因（抓取和接触设计，参考视频验证可行替代）均有帧和裁判阶段支持；时钟均为视频时间，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": "2026-10-31T01:04:43Z",
+  "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/embodied-olympics-fun-catch-and-shoot-i00-privileged--mdrhw2j/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-privileged--mdrhw2j/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "5f2726d38a0277ee74eb9196f538e7baef248598d86f03ea63421da30beaf051",
+   "video_sha256": "247afd7c85a0fa69183931e6e210fe47b9d95dbae67b0675fafa0077736f4ba8",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch_and_shoot/trajectories/embodied-olympics-fun-catch-and-shoot-i00-standard--3l3defb.json

@@ -0,0 +1,195 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-and-shoot-i00-standard--3l3defb",
+ "taskKey": "embodied-olympics-fun/catch_and_shoot",
+ "dag_sha256": "f6a6d84d578d7c1084b9432c694b8d43d8fa3421ae8d0142dbb9a316794fb326",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 310,
+   "sim_steps": 1834,
+   "calls": 12,
+   "rows": 1681,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 930.2,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 1.034,
+   "peak_processes": 7,
+   "peak_disk_gb": 0.365,
+   "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/embodied-olympics-fun-catch-and-shoot-i00-standard--3l3defb/replay.mp4",
+  "duration_s": 3.12,
+  "fps": 25.0,
+  "frames": 78,
+  "sha256": "07c337c9eeb39b5d1c8b70837513ad0ecefca96a43e38deb8306efc9474389ba"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "ball_caught": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.44,
+      "log_step": null,
+      "text_zh": "观察：视频 2.20–2.40 s 橙色来球从左侧传球机飞向机器人，视频 2.44–2.48 s 球经过胸前双手附近，视频 2.52–2.80 s 球继续向画面右侧远离，未随双手停住；视频 3.08 s 球已到地板处。"
+     },
+     {
+      "t": 3.08,
+      "log_step": 183,
+      "text_zh": "观察：log step 183 明确记录重置至 Episode 6 后执行最终脚本，服务返回 failed、the pass was not caught: the ball touched the floor、stages_reached=0/3；该结果与视频末段的落球一致。成功参照视频 2.44–2.88 s 可见球持续留在双手之间，本次视频没有这一承接控制过程。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：视频 2.40–2.52 s 机器人双手在胸前尝试接球，球短暂与手部投影重叠后继续飞走。远景不能分清是否有瞬间擦碰、是哪只手或哪部分手指触球；没有状态接触记录，故不能确认实际接触方式，也不把闭手动作记为已抓握或已双掌夹持。"
+     }
+    ]
+   },
+   "shot_through_ring": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 3.08,
+      "log_step": 183,
+      "text_zh": "观察：视频 2.52–3.08 s 篮球一直远离双手并落向地面，机器人在视频 3.00–3.08 s 仅空手抬臂，没有将接住的球投出，也没有球飞至右侧篮圈并向下穿圈；log step 183 的最终 Episode 6 服务结果为接球未完成、阶段 0/3。"
+     },
+     {
+      "t": 3.08,
+      "log_step": null,
+      "text_zh": "推断：接球前置事件未成立，后续空手抬臂没有开始持球投篮操作，因此投篮命中节点始终为假，未添加该步的球体接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [],
+ "failure": {
+  "breakpoint": {
+   "ref": "ball_caught",
+   "t": 2.48,
+   "log_step": 183,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 2.44–2.52 s 来球经过机器人胸前双手后仍持续飞走，未被两手共同控制；视频 3.08 s 球落到地板处。log step 183 的最终 Episode 6 服务结果确认来球未接住且触地终止，接球失败未修复。"
+  },
+  "unmet": {
+   "ball_caught": false,
+   "shot_through_ring": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "动态时机",
+    "t": 2.48,
+    "log_step": 183,
+    "evidence_zh": "观察：视频 2.40–2.48 s 机器人双手在来球通过时收拢，但视频 2.52–2.64 s 球已从手前继续飞离，双手没有跟随来球并将其截停；视频 2.80–3.08 s 球向地面下落。推断：对运动来球的接球动作没有在通过窗口内形成有效承接，属于动态时机失配；没有先持住再滑落的可见过程。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "log_step": 178,
+    "evidence_zh": "观察：log step 177 为最终方案求解左掌横向 +0.11 m、右掌 -0.05 m 的非对称等待布局；log step 178 实际提交脚本却写入两臂镜像的 Q_READY，并按预定仿真 2.41��2.44 s 收手、仿真 2.44–2.72 s 保持固定关节目标，整段轨迹一次提交且不根据接球读数修正。视频 2.44–2.64 s 显示球掠过后继续飞走；log step 183 记录该脚本执行后的接球失败。推断：接球控制实现没有落实其预定的非对称拦截布局，且没有在漏接时修正，支持控制层归因；缺少本次末端位姿和接触数据，不能进一步确定几何、时序或跟踪误差各自的贡献。",
+    "source_zh": "日志"
+   }
+  ],
+  "gap_zh": "本次没有状态日志；唯一场地远景在视频 2.40–2.52 s 无法分辨篮球与哪只手是否有瞬间擦碰，也无法测量双手有效接触时长，因此接球的 contact_actual 保留未知。log step 183 的持球最大值为 agent 脚本对返回数组的统计，只作辅助，不等同于逐子步接触记录；接球未完成由视频和最终服务判词共同确认。控制根因子类及具体误差贡献证据不足，保留在控制大类。"
+ },
+ "summary_zh": "评分片段对应最终 Episode 6：机器人尝试双手接来球，但球在视频约 2.44–2.52 s 掠过双手后继续飞走，两个节点均未完成，最终进度为 0/2。断点为双手接住球，现象为动态时机失配；脚本中实际等待姿态与所求非对称方案不一致，并按固定时序执行，支持控制层归因。远景和缺少状态接触记录限制了擦碰方式及更细控制根因的判定。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "两节点均未达成、断点 ball_caught（操作中）与服务判词（stages 0/3，球触地）及视频帧一致；控制类根因是基于 log step 177/178 的推断，已标为推断且未过度细分。",
+    "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": "2026-10-31T01:08:03Z",
+  "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/embodied-olympics-fun-catch-and-shoot-i00-standard--3l3defb/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-standard--3l3defb/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "22ecb355c4af4dfddbc5df23e2a09d671e256a87fffdf0e3f3188ca9917270c9",
+   "video_sha256": "07c337c9eeb39b5d1c8b70837513ad0ecefca96a43e38deb8306efc9474389ba",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch_and_shoot/trajectories/embodied-olympics-fun-catch-and-shoot-i00-standard--57xjzk2.json

@@ -0,0 +1,207 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-and-shoot-i00-standard--57xjzk2",
+ "taskKey": "embodied-olympics-fun/catch_and_shoot",
+ "dag_sha256": "f6a6d84d578d7c1084b9432c694b8d43d8fa3421ae8d0142dbb9a316794fb326",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 330,
+   "sim_steps": 1237,
+   "calls": 26,
+   "rows": 1237,
+   "resets": 3,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 1770.5,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 2.206,
+   "peak_processes": 12,
+   "peak_disk_gb": 2.043,
+   "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/embodied-olympics-fun-catch-and-shoot-i00-standard--57xjzk2/replay.mp4",
+  "duration_s": 3.32,
+  "fps": 25.0,
+  "frames": 83,
+  "sha256": "c6070cbd7f80ae93af87853b36ad3c50e67f667581c0390f5c3338c8f037e7dd"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "ball_caught": {
+    "method": "visual",
+    "intervals": [
+     [
+      2.56,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 2.56,
+      "log_step": 75,
+      "text_zh": "观察：视频 2.28–2.44 s 橙球从画面左侧传球机飞至机器人胸腹前，视频 2.48–2.88 s 球停留在两手承接区域，未先触地；log step 75 的最后一次现场尝试最终由服务返回 stages_reached=2、outcome=\"the ball touched the floor\"，与接球后脱手的画面相符。"
+     },
+     {
+      "t": 2.56,
+      "log_step": 75,
+      "text_zh": "推断：两手共同承接并控制来球，且服务已累计接球与释放两个原生阶段，故双手接球事件成立；以视频 2.56 s 为约略完成锚点，毫秒接触窗口不能由远景独立量出。此节点为锁存事件，随后落球不取消。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "托/舀",
+      "note_zh": "观察：视频 2.44–2.88 s 机器人两手共同承接橙色篮球，左手从下方托接，右手在侧面挡住来球；球停在掌根附近，未见单只手以相对指掌接触包握球。"
+     }
+    ]
+   },
+   "shot_through_ring": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.92,
+      "log_step": 75,
+      "text_zh": "观察：视频 2.56–2.88 s 机器人在接球后继续以双手托挡、收紧持球；视频 2.92–3.28 s 球从手前下滑并下降到脚旁，机器人没有完成向篮筐送球的挥臂，球未飞向篮圈。log step 75 的实际执行代码在接球动作后仅持续提交 q_hold 姿态与双手闭合指令，服务最终报告球触地失败。"
+     },
+     {
+      "t": 3.28,
+      "log_step": 79,
+      "text_zh": "观察：视频 3.28 s 末帧篮球已在机器人脚前地面附近，篮筐区域没有球；log step 79 的服务状态为 episode=4、control_steps=330、done=True、reason=\"failed\"，与 log step 75 最后一次现场尝试及评分视频的结束一致。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 2.56–2.88 s 机器人接球后继续用左手承托、右手向球靠拢压持；候选：托挡夹持。两手共同限制球但未见各手包住球面的抓握，单一托/舀也不能完整表达此阶段的侧向收紧；随后球从掌前滑出，未形成投篮送球动作。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [],
+ "failure": {
+  "breakpoint": {
+   "ref": "shot_through_ring",
+   "t": 2.92,
+   "log_step": 75,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 2.48–2.88 s 机器人已将来球留在两手间，接着在持球准备阶段收紧双手；视频 2.92 s 起球向手前下方滑出，直至视频 3.28 s 落到脚旁，未形成投篮。log step 75 显示本次执行只有接球姿态后接 q_hold 收紧持球，随后服务报告球触地、stages_reached=2；接球事件已经完成，后续投篮步骤因持球失效而中断。"
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没保持住",
+    "t": 2.92,
+    "log_step": 75,
+    "evidence_zh": "观察：视频 2.48–2.88 s 橙球短暂停留在机器人两手���前，视频 2.92–3.28 s 球在双臂没有投篮挥动的情况下从手间下滑、脱离并落到脚旁；log step 75 中双手闭合命令仍为 1、1，并未安排主动松手。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "抓取和接触设计",
+    "alt_feasible": "已验证有",
+    "source_zh": "视频 / 日志 / agent 自述",
+    "evidence_zh": "推断：视频 2.48–2.88 s 的接触主要是下方承托加侧面掌根挡球，随后球沿掌前滚滑出去，说明该托挡布局虽能消减来球速度，却不足以持续约束球。log step 75 的执行代码在接球后持续收紧 q_hold、保持双手闭合，仍发生滑落；agent 在 log step 80 也自述球从两掌掌根间滚出。这支持接触承载布局失效，未据此推定摩擦或力矩参数错误。可行替代由独立成功参考验证：oracle 视频 2.60–4.40 s 球始终随双手保持受控并完成投篮蓄势；该参考仅说明同任务存在可行持球方案，不作为本次进度证据。"
+   }
+  ],
+  "gap_zh": "没有状态日志。远景可确认两手承接、球随后滑落及未投篮，但不能独立测量每手连续接触的 0.10 秒窗口、掌指接触对或接触力；接球完成结合最后一次现场尝试的服务阶段数判定，视频完成锚点约有 0.1 秒误差。具体接触几何根因由画面、执行指令及 agent 自述共同推断；未将 agent 本地模拟器中的试投或摔倒计入本次评分轨迹。",
+  "unmet": {
+   "ball_caught": true,
+   "shot_through_ring": false
+  }
+ },
+ "summary_zh": "本次评分轨迹完成双手接球事件，最终为 1/2 个节点；原生阶段数 2 包含接球和释放，不等于两个语义节点均完成。机器人在后续投篮的持球准备阶段未保持住球，视频约 2.92 s 球开始滑落，最终触地，未形成投篮；接触布局不足以持续约束球是有证据支持的控制根因。无状态日志，精确接触窗口和掌指接触细节依赖远景与现场服务阶段反馈结合判断。",
+ "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 2.28–3.28 s 帧与 log step 75/79/80 一致：球在约 2.4 s 被两手承接，约 2.9 s 起滑落触地；服务 stages_reached=2（接球+释放）支持 ball_caught 成立、shot_through_ring 未成立。断点、没保持住现象与抓取和接触设计根因均有证据支持。",
+    "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": "2026-10-31T01:08:38Z",
+  "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/embodied-olympics-fun-catch-and-shoot-i00-standard--57xjzk2/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-standard--57xjzk2/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "619dc413e33162a18fb64403d7201bbf8b1a1b27360a8d8722cdb1be4094a826",
+   "video_sha256": "c6070cbd7f80ae93af87853b36ad3c50e67f667581c0390f5c3338c8f037e7dd",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch_and_shoot/trajectories/embodied-olympics-fun-catch-and-shoot-i00-standard--9zpudbi.json

@@ -0,0 +1,192 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-and-shoot-i00-standard--9zpudbi",
+ "taskKey": "embodied-olympics-fun/catch_and_shoot",
+ "dag_sha256": "f6a6d84d578d7c1084b9432c694b8d43d8fa3421ae8d0142dbb9a316794fb326",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 242,
+   "sim_steps": 1581,
+   "calls": 19,
+   "rows": 1581,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 3100.6,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 6.983,
+   "peak_processes": 13,
+   "peak_disk_gb": 2.039,
+   "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/embodied-olympics-fun-catch-and-shoot-i00-standard--9zpudbi/replay.mp4",
+  "duration_s": 2.44,
+  "fps": 25.0,
+  "frames": 61,
+  "sha256": "55317138cf57aca80a471fe48017dbddb9978245170e4cb740ec414f33e2293e"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "ball_caught": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 1.6,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–1.88 s，机器人站在原处，将两只手逐渐举至头部附近，等待左侧传球机来球。"
+     },
+     {
+      "t": 2.36,
+      "log_step": 187,
+      "text_zh": "观察：视频 2.00–2.40 s 已逐帧检查；篮球从左侧传球机飞向机器人，在视频 2.32–2.40 s 接近并被头肩附近身体轮廓遮住，没有出现两手共同承接、控制球的状态。log step 187 的最后一次现场尝试 a6 返回 stages_reached=0，终止原因为“the ball touched a part of you other than your hands”，两手 holding 标记均未出现；这是机器人服务的回合结果及传感器摘要，不是 agent 的成功自述。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：视频 1.60–2.40 s 可见机器人双臂举起迎球，但末段球与身体重叠，不能辨明具体接触部位和接触方式；log step 187 的服务结果确认发生非手部触球。没有可确认的双手夹持或抓握，不把预设接球姿势当作已发生的接触。"
+     }
+    ]
+   },
+   "shot_through_ring": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.4,
+      "log_step": 187,
+      "text_zh": "观察：视频 0.0–2.40 s 未见机器人持球后投出，也未见球飞至画面右侧篮圈；回合止于来球接近身体的接球阶段。log step 187 的服务返回接球阶段未完成且回合失败，脚本随即退出，未执行后续投篮段。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [],
+ "failure": {
+  "breakpoint": {
+   "ref": "ball_caught",
+   "t": 2.36,
+   "log_step": 187,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 2.32–2.40 s，来球到达机器人头肩附近，机器人双手未共同控制球；log step 187 明确记录最后一次现场尝试 a6 因非手部触球失败，stages_reached=0。接球失败后没有继续回合，故断点为双手接住球。"
+  },
+  "unmet": {
+   "ball_caught": false,
+   "shot_through_ring": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "动态时机",
+    "t": 2.36,
+    "log_step": 187,
+    "evidence_zh": "观察：视频 2.08–2.40 s，篮球持续向机器人飞来，机器人保持双手举在头部附近的迎球姿势，未随来球形成共同承接；球进入身体轮廓后回合结束。log step 187 的服务判定非手部触球、接球阶段未完成。推断：双手没有跟上运动来球并在其到达时建立合法承接，属于动态接球未对上。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "抓取和接触设计",
+    "alt_feasible": "未知",
+    "t": 2.36,
+    "log_step": 187,
+    "source_zh": "日志 / 视频",
+    "evidence_zh": "推断：最终接球姿势的手臂与来球接触几何不适配。log step 176 写出的脚本先插值到固定迎球姿势，再在来球期间保持同一组臂关节目标；log step 186–187 将它替换为面部附近的固定双掌姿势并提前闭手。视频 1.88–2.40 s 确见双手保持头旁而来球进入身体轮廓，log step 187 的服务确认先发生非手部触球而未完成接球，支持接触布局失败。没有本次相同条件下经验证的替代接触方案，故 alt_feasible 为未知。"
+   }
+  ],
+  "gap_zh": "没有状态日志；远景末段球与机器人身体重叠，无法独立确定具体碰撞的是头、肩还是臂段，也无法重建毫秒级手球接触，故 contact_actual 保留未知。log step 190、194 中 agent 关于击中左肘、传球速度与跨次方差的解释未经本次可用证据独立核实，不据此判定任务或环境问题。成功参考仅用于辨认共同控球和投篮的视觉线索。"
+ },
+ "summary_zh": "评分回合是 log step 187 的最后一次现场尝试 a6；机器人举起双手迎接左侧来球，但未共同控球，两个节点均未完成（0/2）。断点在双手接住球：视频约 2.36 s 来球进入身体轮廓，服务报告非手部触球，固定迎球接触布局未能适配来球，尚未进入投篮。没有状态日志且远景遮挡具体碰撞部位，实际接触方式保留未知。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "评分回合为最后一次现场尝试 a6（log step 187，服务返回 stages_reached=0、非手部触球）；视频 1.96–2.40 s 逐帧可见球从传球机飞向机器人，约 2.28–2.32 s 到达头肩附近，2.40 s 球已不在画面，双手未控球。两个节点 false、断点 ball_caught、contact_actual 未知、时钟与 0/2 均与证据一致，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": "2026-10-31T01:10:06Z",
+  "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/embodied-olympics-fun-catch-and-shoot-i00-standard--9zpudbi/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-standard--9zpudbi/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "ed4b74c06266ddc1205964d52643f84d0a1d1db3606951a1aa2222a5903893cc",
+   "video_sha256": "55317138cf57aca80a471fe48017dbddb9978245170e4cb740ec414f33e2293e",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch_and_shoot/trajectories/embodied-olympics-fun-catch-and-shoot-i00-standard--dstsan3.json

@@ -0,0 +1,199 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-and-shoot-i00-standard--dstsan3",
+ "taskKey": "embodied-olympics-fun/catch_and_shoot",
+ "dag_sha256": "f6a6d84d578d7c1084b9432c694b8d43d8fa3421ae8d0142dbb9a316794fb326",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 239,
+   "sim_steps": 1508,
+   "calls": 34,
+   "rows": 1373,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 1059.8,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.308,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.856,
+   "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/embodied-olympics-fun-catch-and-shoot-i00-standard--dstsan3/replay.mp4",
+  "duration_s": 2.4,
+  "fps": 25.0,
+  "frames": 60,
+  "sha256": "8df4f5169f1e5d09f22af0958fd67bb7ca56a1adb61651b0c44ad7de99a74f16"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "ball_caught": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": 82,
+      "text_zh": "观察：视频 0.00–1.96 s，机器人从空手姿态抬起双臂并在胸前偏高处等待；log step 82 的调用记录含一次 reset 和三次 submit，是最终尝试，与此前五次会话测试区分。"
+     },
+     {
+      "t": 2.32,
+      "log_step": null,
+      "text_zh": "观察：逐帧检查视频 2.00–2.36 s，橙球由左侧传球机飞至机器人头肩附近，视频 2.32 s 仍在双手左侧，视频 2.36 s 被上身遮挡；没有出现两手共同承接并持续控制球的画面。成功参考视频 2.48–3.00 s 可清楚看见球停留在胸前两手之间，本次未出现该线索。"
+     },
+     {
+      "t": 2.36,
+      "log_step": 86,
+      "text_zh": "推断：来球直至视频 2.32 s 尚未进入双手，余下画面没有满足连续 0.10 秒共同接触的可能区间，故接球事件未成立。agent 在 log step 86 自述最终仿真 2.39 s 非手部触球、原生阶段为 0/3；这只是对可见接球失败的旁证，不把其三阶段计数当作本 DAG 的两节点分母。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：视频 2.20–2.36 s，机器人双臂保持接球姿态，但球到达时被头肩和手臂遮住，无法核实哪只手是否实际触球、是否形成对向接触，故不按任务典型方案填夹持或抓握。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–2.36 s，机器人双脚着地支撑本体进行抬臂和迎球，未见摔倒。"
+     }
+    ]
+   },
+   "shot_through_ring": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.36,
+      "log_step": 83,
+      "text_zh": "观察：视频 0.00–2.36 s 中机器人只做接球准备和迎球，没有持球后的送球、释放及球飞向篮圈或自上而下穿圈的动作；接球前置未达。agent 在 log step 83 也自述投篮阶段未启动。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [],
+ "failure": {
+  "breakpoint": {
+   "ref": "ball_caught",
+   "t": 2.32,
+   "log_step": 82,
+   "where": "操作中",
+   "evidence_zh": "观察：最终尝试由 log step 82 重置并执行；视频 2.20–2.36 s，来球到达机器人头肩附近，双手仍停在前方接球姿态，未将球共同承接控制，随后视频结束，接球未修复。agent 在 log step 83、86 自述终局在仿真 2.39 s 因非手部触球失败；该终止原因的原始服务输出未保留。"
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "动态时机",
+    "t": 2.32,
+    "log_step": null,
+    "evidence_zh": "观察：视频 2.20–2.32 s，飞行的球快速从机器人左侧靠近头肩，而两手保持在身体前方，没有随来球建立共同承接；视频 2.36 s 球进入上身遮挡区域，始终未出现接住后的停球。推断：机器人未在来球经过时以双手截住运动目标，属于动态接球配合未跟上；没有证据表明曾持球后再失去。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "source_zh": "视频",
+    "evidence_zh": "推断：视频 0.00–1.96 s 机器人已抬起双臂准备接球，目标与步骤顺序正确；视频 2.20–2.36 s 双手的实际位置与运动未能承接飞来的球，接球控制未奏效。视频不足以区分手掌接触几何、拦截轨迹或闭合时序的具体贡献，故保留在控制大类，不细分参数或感知错误。"
+   }
+  ],
+  "gap_zh": "无状态日志。视频只有外部远景，末帧为视频 2.36 s，球在头肩附近被遮挡，不能确定终止接触的具体身体部位，也不能分辨单手短暂接触及手指闭合方式。log step 82、83 的原始服务输出为空；最终仿真 2.39 s 非手部触球及原生阶段 0/3 仅见 agent 在 log step 83、86 的自述，不作为独立裁判观测。接球控制的具体子类缺少可核实的末端位姿、接触力和时序记录。",
+  "unmet": {
+   "ball_caught": false,
+   "shot_through_ring": false
+  }
+ },
+ "summary_zh": "评分视频对应最后一次重置后的接球尝试：机器人抬起双臂迎球，但在视频 2.20–2.36 s 来球到达头肩附近时未形成持续双手控制，两个语义节点均未完成，投篮未启动。断点为双手接住球，现象是未跟上运动来球的动态时机，根因仅能归入控制。末帧遮挡且最终服务输出为空，不能独立确认 agent 自述的非手部触球位置或更细的控制原因。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "两节点均未达成、断点在 ball_caught（操作中）、动态时机+控制的标注与视频和 log 一致；视频 2.20–2.36 s 球到达头肩、双手未共同承接，投篮未启动。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": "2026-10-31T01:08:47Z",
+  "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/embodied-olympics-fun-catch-and-shoot-i00-standard--dstsan3/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-standard--dstsan3/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "ee4e9894567a914ea18311fc14e5399b0daaebbb54d6a0af433f03acda9f469d",
+   "video_sha256": "8df4f5169f1e5d09f22af0958fd67bb7ca56a1adb61651b0c44ad7de99a74f16",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch_and_shoot/trajectories/embodied-olympics-fun-catch-and-shoot-i00-standard--qfl2y8x.json

@@ -0,0 +1,200 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-and-shoot-i00-standard--qfl2y8x",
+ "taskKey": "embodied-olympics-fun/catch_and_shoot",
+ "dag_sha256": "f6a6d84d578d7c1084b9432c694b8d43d8fa3421ae8d0142dbb9a316794fb326",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 274,
+   "sim_steps": 1579,
+   "calls": 101,
+   "rows": 772,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 2481.8,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 1.701,
+   "peak_processes": 16,
+   "peak_disk_gb": 2.046,
+   "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/embodied-olympics-fun-catch-and-shoot-i00-standard--qfl2y8x/replay.mp4",
+  "duration_s": 2.76,
+  "fps": 25.0,
+  "frames": 69,
+  "sha256": "a419df472ee233024d4668a104da1effcde2d7a5ddb47571c466c9d58c245ee4"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "ball_caught": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.4,
+      "log_step": 124,
+      "text_zh": "观察：视频 0.0–0.4 s，机器人双臂从初始姿态屈曲到腹前准备接球，随后基本保持该姿态；并非首步未启动。"
+     },
+     {
+      "t": 2.44,
+      "log_step": 126,
+      "text_zh": "观察：视频 2.32–2.52 s，橙色来球从机器人左侧飞到双手附近后继续向画面右下方运动；视频 2.56–2.72 s 球远离双手并落向地板，全程未见双手共同承接并控制球。log step 126 中本次 episode 6 的服务返回最终 outcome 为“the pass was not caught: the ball touched the floor”，与画面一致。"
+     },
+     {
+      "t": 2.72,
+      "log_step": 128,
+      "text_zh": "观察：log step 126、128 的服务状态均为 episode 6、control_steps=274、done=true、resets_left=0；reward 中 graded_by_live=1、n_actions=274。评分采用这次最后现场尝试，其他测试回合的前臂触球不属于本段视频。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：视频 2.32–2.52 s，机器人两手在腹前作承接准备，球掠过手部后继续飞走，没有形成双手持球。远景中手指仅占少量像素，且球经过时局部被身体遮挡，不能确定是否有瞬间擦触、触及哪只手及具体接触方式；因此不将脚本预设的右掌拦球、左掌托球当作已发生的接触。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.0–2.72 s，机器人双脚持续在地板上承重，站立完成抬臂与等待来球的尝试，未见摔倒。"
+     }
+    ]
+   },
+   "shot_through_ring": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.72,
+      "log_step": 126,
+      "text_zh": "观察：视频 0.0–2.72 s 未见机器人持球后向篮圈投球，来球直接掠过并下落，篮圈处没有入球。log step 126 显示现场尝试在接球失败后结束，随后脚本提交蓄力动作被服务以 episode finished 拒绝；脚本中写有投篮序列并不表示评分局执行过投篮。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [],
+ "failure": {
+  "breakpoint": {
+   "ref": "ball_caught",
+   "t": 2.44,
+   "log_step": 126,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 2.36–2.52 s，机器人已在接球姿态中等待，但球经过手部区域后仍飞向右下方，没有停在双手间；视频 2.72 s 球已接近地面。log step 126 的本次现场裁判返回“the pass was not caught: the ball touched the floor”，接球未完成且未修复，失败发生在双手接住球这一步。"
+  },
+  "unmet": {
+   "ball_caught": false,
+   "shot_through_ring": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "动态时机",
+    "t": 2.44,
+    "log_step": 126,
+    "evidence_zh": "观察：视频 2.32–2.52 s，运动中的来球经过机器人双手附近，双手未随来球形成拦截和持续承接，球继续飞走并下落；视频 2.56–2.72 s 也没有重新接住。此处是对来球的动态拦接失败，未观察到先持住再丢失的过程。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "log_step": 124,
+    "source_zh": "日志",
+    "evidence_zh": "推断：log step 124 的本次控制脚本先用 35 个 dt=0.01 s 的动作移到固定 CATCH 姿态，再调用 seconds=2.2 的观察推进来球过程，期间没有根据球的运动修正接球动作；之后才闭合手指。视频 0.4–2.52 s 可见双臂基本保持姿态而球经过后继续飞走，log step 126 也记录接球失败。该固定姿态与闭合时序未实现对来球的控制；缺少实测手位、接触力和关节跟踪数据，不能进一步确定是接触布局、轨迹还是跟踪误差。"
+   }
+  ],
+  "gap_zh": "观察：inputs 中没有本次状态日志；视频为单一远景，手指分辨率低，视频 2.36–2.44 s 球经过身体附近时有局部遮挡，无法判定是否存在瞬间单手擦触及其方式。推断：视频连续运动和本次裁判结果足以判定两个节点均未完成，但不足以将控制根因细分或量化手位误差；agent 在 log step 140 自述的运动学残差未获独立测量证实。"
+ },
+ "summary_zh": "最后一次现场评分尝试完成了抬臂接球准备，但视频约 2.4 s 来球掠过双手后继续下落，双手接住球和投篮命中均未达成，最终为 0/2。断点为双手接住球，动态拦接未成功，固定接球姿态及后续闭手未控制住来球，根因保留为控制。远景和缺少状态日志限制了瞬时接触判断与控制子类归因；此前测试回合的前臂触球不计入本次轨迹。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "两节点均未达成、断点为 ball_caught（操作中）、现象为运动·动态时机、根因停在控制，均与视频（球约 2.4 s 经过身体左前后转向右下落地）和 log step 126 的裁判返回一致；视频时间均标明视频时钟，未把 agent 时间换算成视频时间，接触写未知符合远景分辨率，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": "2026-10-31T01:08:14Z",
+  "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/embodied-olympics-fun-catch-and-shoot-i00-standard--qfl2y8x/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-standard--qfl2y8x/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "82d2a3efedf48b8cf98b8c0d7cc9ffa7c5b1dae86e1f79b6d329b1cde65b4139",
+   "video_sha256": "a419df472ee233024d4668a104da1effcde2d7a5ddb47571c466c9d58c245ee4",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch_and_shoot/trajectories/embodied-olympics-fun-catch-and-shoot-i00-standard--sawsntx.json

@@ -0,0 +1,212 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-and-shoot-i00-standard--sawsntx",
+ "taskKey": "embodied-olympics-fun/catch_and_shoot",
+ "dag_sha256": "f6a6d84d578d7c1084b9432c694b8d43d8fa3421ae8d0142dbb9a316794fb326",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 274,
+   "sim_steps": 1610,
+   "calls": 24,
+   "rows": 1610,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 6739.9,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 1.175,
+   "peak_processes": 12,
+   "peak_disk_gb": 2.958,
+   "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/embodied-olympics-fun-catch-and-shoot-i00-standard--sawsntx/replay.mp4",
+  "duration_s": 2.76,
+  "fps": 25.0,
+  "frames": 69,
+  "sha256": "2a303b8460815757619c20428b9d5cad6c73b9b929f3d02af562584a1b0f3652"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "ball_caught": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.48,
+      "log_step": null,
+      "text_zh": "观察：本次视频 2.24–2.48 s，橙球从画面左侧传球机飞至机器人身前；机器人一只手高举在头部附近，另一只手在较低的胸腹前，球未停留在双手之间，视频 2.52–2.72 s 继续向右下方飞离并落至地面附近，没有共同持球的阶段。"
+     },
+     {
+      "t": null,
+      "log_step": 159,
+      "text_zh": "观察：log step 159 的服务返回最后一次 episode 6 已结束，outcome 为“the pass was not caught: the ball touched the floor”，stages_reached=0、stages_total=3；run.json 的 graded_by_live=1、n_actions=274 与该次结束时 control_steps=274 一致。此处服务是执行机器人动作并返回裁判结果的仿真接口；原生三阶段不等于 DAG 的两个语义节点。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：视频 2.24–2.72 s，机器人两臂分别在高位及胸腹前做截球、合拢动作，但球未被双手共同控制。远景中球经过低位手时局部重叠，无法判定是否有瞬时擦碰及其接触方式；不将空手合指标为已抓握篮球。"
+     }
+    ]
+   },
+   "shot_through_ring": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.72,
+      "log_step": 159,
+      "text_zh": "观察：本次视频 0.00–2.72 s 未出现接住球后的投篮动作，末段球越过机器人手部后向地面下降，没有飞至右侧篮圈或从上方穿圈；log step 159 的该次现场裁判结果为接球前触地而失败。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [],
+ "failure": {
+  "breakpoint": {
+   "ref": "ball_caught",
+   "t": 2.48,
+   "log_step": 159,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 2.40–2.52 s，机器人正用两臂截接来球，但高位手未与低位手共同包住球，篮球随即越过手部向右下方继续运动；视频 2.72 s 球已降至地面附近。log step 159 的最后一次 episode 6 由仿真服务报告接球前球触地失败，未在本次尝试中修复。"
+  },
+  "unmet": {
+   "ball_caught": false,
+   "shot_through_ring": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "动态时机",
+    "t": 2.48,
+    "log_step": 159,
+    "evidence_zh": "观察：视频 2.24–2.60 s，篮球快速到达机器人身前并从手部旁漏过；来球经过低位手时另一只手仍在头部附近，双手未同步承接，之后合拢时球已向右下方离开。log step 159 的现场裁判报告未接住来球、球触地；按运动目标漏接的口径记录动态时机，不把同一事件重复标为没保持住。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "抓取和接触设计",
+    "alt_feasible": "已验证有",
+    "evidence_zh": "推断：本次截球的手部接触布局没有在来球经过时形成双手共同承接。视频 2.40–2.52 s 可见两手高低分离、球直接越过；log step 23 的 agent 自建离线模型用盒状拳部几何代替灵巧手，log step 125 的模型代码按手指角度切换开放手与拳部碰撞体，log step 154 以该模型生成截球动作，log step 158–159 将修补后的动作用于最后现场尝试。agent 在 log step 161 自述代理拳部比现场拳部更突出，这个具体尺寸解释待核实，但实际承接位置未拦住球有视频支持。可行替代已由成功参考验证：inputs/oracle/replay.mp4 的视频 2.52–3.00 s 显示参考机器人将球控制在身前两手之间；参考只证明替代可行，不作为本次进度证据。",
+    "source_zh": "视频 / 日志 / agent 自述"
+   }
+  ],
+  "gap_zh": "本次没有状态日志，远景视频不足以确定来球经过低位手时是否有瞬时擦碰、具体触点或毫秒级双手接触窗口，因此接球节点的实际手球接触记为未知；未出现持续共同承接及现场裁判的接球前触地结果足以判定该节点未完成。代理拳部与现场手部的具体尺寸差异仅为 agent 在 log step 161 的自述，尚未独立核实，不将其当作已测得的物理事实。"
+ },
+ "summary_zh": "最后一次现场尝试对应本次视频，机器人站立并伸臂截接来球，但视频约 2.48 s 来球从未同步承接的双手旁漏过，随后下落；双手接住球、投篮命中均未完成（0/2）。断点在接球操作中，现象为动态时机，手部接触布局不适配的控制归因由画面及离线模型、动作生成日志共同支持。缺少状态日志和手部近景，瞬时擦碰及 agent 所述手模型尺寸差异无法独立确认。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 2,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "fix",
+    "conf": 6,
+    "summary_zh": "WHERE 标签（接球未达成、断点 ball_caught、投篮未达成、0/2，时钟均为视频时间，log step id 存在）与视频、log step 159 一致；唯一问题是现象“动态时机”缺乏时机证据，更像手位置没拦到球（位姿不准 / 没建立接触）。",
+    "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（0/2）、断点 ball_caught 操作中、动态时机（符合 FAIL-14）与控制·抓取和接触设计均与视频及 log step 159 一致；时钟为视频时间，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": "2026-10-31T01:14:06Z",
+  "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/embodied-olympics-fun-catch-and-shoot-i00-standard--sawsntx/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-standard--sawsntx/04_review_r1"
+   },
+   {
+    "stage": "reconcile",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-standard--sawsntx/05_reconcile_r1"
+   },
+   {
+    "stage": "traj_review",
+    "round": 2,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-standard--sawsntx/04_review_r2"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "cf60865694ffb0badbe9e49900889b98cdbe37b42327f37912096b76807630b9",
+   "video_sha256": "2a303b8460815757619c20428b9d5cad6c73b9b929f3d02af562584a1b0f3652",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch_and_shoot/trajectories/embodied-olympics-fun-catch-and-shoot-i00-standard--uftjmpy.json

@@ -0,0 +1,176 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-and-shoot-i00-standard--uftjmpy",
+ "taskKey": "embodied-olympics-fun/catch_and_shoot",
+ "dag_sha256": "f6a6d84d578d7c1084b9432c694b8d43d8fa3421ae8d0142dbb9a316794fb326",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "calls": 5,
+   "rows": 0,
+   "resets": 0,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 5998.4,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 4,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 1.324,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.431,
+   "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/embodied-olympics-fun-catch-and-shoot-i00-standard--uftjmpy/replay.mp4",
+  "duration_s": 0.04,
+  "fps": 25.0,
+  "frames": 1,
+  "sha256": "c9acf778685c516e740ca1c335e521ac47da75b99b2c8b2397e4472dce4f3830"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "ball_caught": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": 236,
+      "text_zh": "观察：本次视频 0.00 s 的唯一帧（全片 0.04 s）显示机器人直立、双手未持球，没有双手承接来球的过程；log step 236 的现场服务状态为 control_steps=0、rows=0、done=False。评分器字段 n_actions=0、sim_steps=0，支持受评局始终停在初态，接球未完成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器字段 n_actions=0、rows=0、sim_steps=0；log step 236 的现场服务 control_steps=0，无受评动作轨迹交付。"
+     }
+    ]
+   },
+   "shot_through_ring": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": 240,
+      "text_zh": "观察：本次视频 0.00 s 的唯一帧中机器人未持球，篮圈处没有篮球穿过；评分器字段 n_actions=0、sim_steps=0，且全部调用代码中没有现场 submit 调用。log step 239–240 写入的接投球说明描述 agent 自建 LocalSim 的测试，不是受评现场已执行的动作。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器字段 n_actions=0、rows=0、sim_steps=0；log step 236 的现场服务 control_steps=0，无受评动作轨迹交付。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 241,
+   "where": "首步前",
+   "evidence_zh": "观察：log step 241 的最终消息为 Gemini 过滤拦截提示；此前 log step 236 的现场服务报告 control_steps=0、rows=0、done=False，log step 237 仍读取初始关节状态，log step 239–240 仅写测试说明。本次视频 0.00 s 仅为初态，评分器 n_actions=0、sim_steps=0，没有已执行的物理失败或已固定的现场结果。"
+  },
+  "unmet": {
+   "ball_caught": false,
+   "shot_through_ring": false
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：log step 241 记录模型提供方过滤拦截，发生在现场动作提交之前；本次视频存在但仅有视频 0.00 s 的一个初始帧，无状态日志。推断：这是受评结果固定前的外因中断，按 CASES.md FAIL-5 / DECISIONS.md D37 保留 catalog 的 success=0，但不作为机器人能力失败，不标“没尝试”或根因。log step 239–240 的 LocalSim 接投球自述仅属会话测试，没有受评动作可核验其物理表现。"
+ },
+ "summary_zh": "本次受评视频仅有视频 0.00 s 的初始帧，现场服务与评分器均记录零控制步，双手接球和投篮命中均未完成，最终进度为 0/2。log step 241 记录 Gemini 过滤拦截，现场动作尚未提交，故按外因中断记 unknown／首步前，现象和根因留空，catalog 的 success=0 原样保留。日志中的自建 LocalSim 测试不属于受评轨迹，无法据此判断机器人的接投球能力。",
+ "conf": 9,
+ "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：视频仅 0.04 s 单帧，评分器 n_actions=0、sim_steps=0，log step 241 为 Gemini 过滤拦截，此前无任何现场 submit 或交付；两节点 false、breakpoint unknown/首步前、现象与根因为空并有 gap_zh，符合 FAIL-5。",
+    "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": "2026-10-31T01:06:47Z",
+  "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/embodied-olympics-fun-catch-and-shoot-i00-standard--uftjmpy/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-standard--uftjmpy/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "408843b311abe24d41622d008d3197a8ad727223dcdee41deda94e32938c00c6",
+   "video_sha256": "c9acf778685c516e740ca1c335e521ac47da75b99b2c8b2397e4472dce4f3830",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch_and_shoot/trajectories/embodied-olympics-fun-catch-and-shoot-i00-standard--zfpywsf.json

@@ -0,0 +1,211 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-and-shoot-i00-standard--zfpywsf",
+ "taskKey": "embodied-olympics-fun/catch_and_shoot",
+ "dag_sha256": "f6a6d84d578d7c1084b9432c694b8d43d8fa3421ae8d0142dbb9a316794fb326",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 513,
+   "sim_steps": 2099,
+   "calls": 65,
+   "rows": 1377,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 6899.0,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 2.084,
+   "peak_processes": 11,
+   "peak_disk_gb": 1.029,
+   "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/embodied-olympics-fun-catch-and-shoot-i00-standard--zfpywsf/replay.mp4",
+  "duration_s": 5.16,
+  "fps": 25.0,
+  "frames": 129,
+  "sha256": "6e234aa1d76cec8a9979da172c95ddadbd2334bcf009fad608fe5c11b2725c28"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "ball_caught": {
+    "method": "visual",
+    "intervals": [
+     [
+      2.6,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 2.6,
+      "log_step": 294,
+      "text_zh": "观察：视频 2.24–2.44 s 橙色篮球从左侧传球机飞到机器人胸前，视频 2.48–2.96 s 球留在上下相对的双手之间，随后机器人持球下移准备投篮，未先落地。log step 294 的第 6 局服务裁判返回 stages_reached=2/3，与本视频的接球、释放一致；接球时刻按连续持球画面估计为视频约 2.6 s。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 2.48–2.96 s 机器人两只手从球的上、下方共同夹持篮球。候选：夹持；画面显示双掌相向承接、夹压，不能确认任一单手形成手指与拇指的独立包握，现有条目无双掌夹持，故用其他。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 2.24–2.96 s 机器人双脚在地面支撑站姿，双臂接球。"
+     }
+    ]
+   },
+   "shot_through_ring": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 3.8,
+      "log_step": 287,
+      "text_zh": "观察：视频 3.48–3.60 s 机器人把球降到腰腹前，视频 3.64–3.80 s 双臂向前上方送球，视频 3.80 s 球已与手分离并飞向篮筐；log step 287 记录第 6 局执行最终腕部 lowcmd 修正。"
+     },
+     {
+      "t": 4.72,
+      "log_step": 294,
+      "text_zh": "观察：视频 4.32–4.72 s 篮球飞到篮板和篮圈附近，却从篮圈旁下降，视频 4.80–5.12 s 继续落向篮架附近地面，没有从圈内向下穿过；机器人仍站立。log step 294 的服务裁判报告第 6 局 done=True、outcome=the ball touched the floor、stages_reached=2/3，评分字段 live_success=0。成功参考视频 6.48–6.60 s 可见球由上向下进入圈内，本次未出现该线索。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 3.00–3.60 s 机器人双手从上下方共同夹持、下移篮球，视频 3.64–3.76 s 双臂共同向前上方送球，此后球离手飞行。候选：夹持；未见单手独立包握，词表没有双掌相向夹持，故用其他。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 3.00–5.12 s 机器人双脚支撑身体完成送球及收臂，未见摔倒。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [],
+ "failure": {
+  "breakpoint": {
+   "ref": "shot_through_ring",
+   "t": 3.8,
+   "log_step": 287,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 2.48–2.96 s 机器人已控制来球；视频 3.64–3.80 s 主动将球向前上方投出，但视频 4.64–4.88 s 球从篮圈旁落下，未命中且没有补救。log step 294 的第 6 局裁判以球触地结束并报告完成 2/3 个原生阶段（接球、释放）；因此未完成的语义步骤是投篮命中，视频约 3.8 s 的出手作为断点导航。"
+  },
+  "unmet": {
+   "ball_caught": true,
+   "shot_through_ring": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 3.8,
+    "log_step": 294,
+    "evidence_zh": "观察：视频 3.80 s 机器人已将球投出，视频 4.64–4.88 s 球经过篮圈旁并继续下落，未沿圈内通道下降；log step 294 裁判报告 the ball touched the floor、stages_reached=2/3。投出的球没有命中固定篮筐。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "evidence_zh": "推断：log step 234、253、264、278 记录 agent 在第 6 局逐段提交投篮关节指令，log step 287 再以 lowcmd 执行最终腕部修正；视频 3.64–3.80 s 机器人确实完成向前上方送球，视频 4.64–4.88 s 却从圈旁落下。目标和接球后投篮的顺序明确，失败发生在实际出手形成的球路未对准篮圈，支持控制层归因；尚不能确定是接触几何、轨迹跟踪还是具体参数造成。",
+    "source_zh": "日志 / 视频"
+   }
+  ],
+  "gap_zh": "没有状态日志，远景不能精确测量双手有效接触持续时间、手指对向接触、释放速度和旋转或穿圈几何；接球以持续持球画面结合第 6 局服务裁判阶段结果判定，视频约 2.6 s 为估计时刻。日志中离线模型的球路预测不是评分轨迹状态，不能据此进一步确定控制子类；之前重置的五次尝试不用于本次节点进度或失败现象。"
+ },
+ "summary_zh": "第 6 次现场评分尝试在视频约 2.6 s 双手接住来球，随后降球并于视频约 3.8 s 投出，最终完成 1/2 个语义节点。篮球从篮圈旁落下，裁判因球触地判失败；断点为投篮命中，现象为位姿不准，证据支持控制层归因但不足以细分。无状态日志，接球时刻和具体接触细节依靠远景及本局裁判结果交叉判断。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "接球约 2.6 s、投篮未命中、断点在 shot_through_ring 的标注与视频及裁判 stages_reached=2/3 一致；球从圈旁飞过，现象位姿不准与控制归因有证据支持，未发现需修改的问题。",
+    "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": "2026-10-31T01:04:57Z",
+  "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/embodied-olympics-fun-catch-and-shoot-i00-standard--zfpywsf/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-standard--zfpywsf/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "06d8991d2777e1349d949db9adfeb4d568f690f6dc1833a4e45803cc1bb1bb0a",
+   "video_sha256": "6e234aa1d76cec8a9979da172c95ddadbd2334bcf009fad608fe5c11b2725c28",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch_and_shoot/trajectories/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--fuf6v5a.json

@@ -0,0 +1,206 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--fuf6v5a",
+ "taskKey": "embodied-olympics-fun/catch_and_shoot",
+ "dag_sha256": "f6a6d84d578d7c1084b9432c694b8d43d8fa3421ae8d0142dbb9a316794fb326",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 291,
+   "sim_steps": 1623,
+   "calls": 31,
+   "rows": 9000,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 2317.4,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "realtime": 1,
+   "slowdown": 1.0,
+   "clock_s": 16.378,
+   "max_lag_s": 0.0583,
+   "stalls": 0,
+   "stalled_s": 0.0,
+   "stalled_share": 0.0,
+   "held_steps": 0,
+   "late_steps": 0,
+   "requests_refused": 1,
+   "request_busy_s": 0.433,
+   "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/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--fuf6v5a/replay.mp4",
+  "duration_s": 2.92,
+  "fps": 25.0,
+  "frames": 73,
+  "sha256": "662d497893e05bc17cef0c37a32c617fb36647315e678f99d0f0ff111acad8a2"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "ball_caught": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.44,
+      "log_step": 203,
+      "text_zh": "观察：视频 2.28–2.44 s 篮球从左侧飞至机器人胸前双手附近；视频 2.48–2.64 s 球继续向前移出手部区域，视频 2.68–2.88 s 向地面下落，未呈现双手共同控制。log step 203 对最后一次尝试的传感记录只在仿真 2.46 s 显示 holding_left=1、holding_right=0，其后两者均为 0。"
+     },
+     {
+      "t": 2.88,
+      "log_step": 199,
+      "text_zh": "观察：log step 199 显示重置到 episode=6 后执行最终计划；机器人服务在 control_steps=291、sim_time_s=2.91 报告 stages_reached=0/3，原文为“the pass was not caught: the ball touched the floor”。视频末段球离开双手并接近地面，与该结果一致；未将 agent 工作时间换算为视频时间。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "其他",
+      "note_zh": "推断：视频 2.44–2.52 s 来球在手边改变运动方向，结合 log step 203 最终尝试仅左手短暂 holding_left=1，标左手与篮球短暂碰触；候选名称“碰挡”。没有形成抓握、托住或双掌夹持，现有方式不足以准确表示这种瞬时碰触；右手未建立共同持球。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0–2.88 s 机器人双脚在地板上支撑身体，以站立姿态等待并尝试接球。"
+     }
+    ]
+   },
+   "shot_through_ring": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.88,
+      "log_step": 199,
+      "text_zh": "观察：视频 0–2.88 s 机器人仅准备并尝试接球，没有持球投出动作；篮球从手边向地面落下，未飞至右侧篮圈。log step 199 的最终 episode=6 结果为接球未完成、stages_reached=0/3，未进入释放及穿圈阶段。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [],
+ "failure": {
+  "breakpoint": {
+   "ref": "ball_caught",
+   "t": 2.52,
+   "log_step": 199,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 2.44 s 来球抵达双手附近后，视频 2.52–2.64 s 仍继续向前离开手部，视频 2.68–2.88 s 下落；机器人没有完成双手接球。log step 199 的最终 episode=6 由服务判为接球未完成、球触地，接球步骤此后未修复。"
+  },
+  "unmet": {
+   "ball_caught": false,
+   "shot_through_ring": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "动态时机",
+    "t": 2.52,
+    "log_step": 203,
+    "evidence_zh": "观察：视频 2.44–2.64 s 篮球穿过接球区域后离开，双手未及时合拢并随球形成持续承接。log step 203 的最终尝试传感记录显示，仿真 2.43–2.46 s 两手 closure 均为 0，仿真 2.49 s 才为 0.03，随后逐渐增加，而 holding_right 始终为 0。推断：接合动作没有赶上运动来球的��接窗口；不是接住后再掉球。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "抓取和接触设计",
+    "alt_feasible": "已验证有",
+    "source_zh": "日志 / 视频",
+    "evidence_zh": "推断：log step 192 的最终计划让双手保持预设位置、手指开放至仿真 2.45 s，注释设想来球撞到左手后手臂让位、球自行进入双掌间，再逐步闭手；log step 199 确认实际执行该 plan6。视频 2.44–2.68 s 显示该布置只使球在手边偏转，未形成双手承载，随后球离开；log step 203 的单手瞬时 holding 与此一致，因此归为接球接触方案不适配。成功参考视频 2.48–3.04 s 可见另一控制方案将球持续保持在双手之间，说明存在可行替代；参考仅用于验证可行性，不计入本次进度。"
+   }
+  ],
+  "gap_zh": "本次没有状态日志；外部远景可辨球到手边及离开下落，但不能独立解析每根手指的接触、毫秒级双手接触窗口或实际接触力。左手瞬时接触由最终尝试的传感记录辅助判读；agent 在 log step 221 将失败进一步归因于手臂刚度，缺乏独立力学验证，未据此标参数调试或反馈控制。"
+ },
+ "summary_zh": "评分对象为最后的 episode=6：机器人站立接球，来球在手边短暂偏转后下落，双手接住球与投篮命中均未达成，最终为 0/2。断点在接球操作中，闭合动作未赶上来球，预设的开放双掌承接方案未形成共同控制，服务随后判球触地失败。没有状态日志，接触位置与接触力只能有限判读，不能证实 agent 关于刚度的进一步解释。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "视频 1.9–2.9 s 可见来球到手边后偏转下落，接球未完成，0/2 与服务结果一致；断点、现象、根因与证据一致，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": "2026-10-31T01:13:02Z",
+  "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/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--fuf6v5a/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--fuf6v5a/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "5b21567d31a3edc04f783075e1acc9593b0033f5ecddcda0628a6c315a2a89d5",
+   "video_sha256": "662d497893e05bc17cef0c37a32c617fb36647315e678f99d0f0ff111acad8a2",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch_and_shoot/trajectories/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--gkskpcv.json

@@ -0,0 +1,206 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--gkskpcv",
+ "taskKey": "embodied-olympics-fun/catch_and_shoot",
+ "dag_sha256": "f6a6d84d578d7c1084b9432c694b8d43d8fa3421ae8d0142dbb9a316794fb326",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 264,
+   "sim_steps": 1681,
+   "calls": 33,
+   "rows": 6032,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 6369.8,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "realtime": 1,
+   "slowdown": 1.0,
+   "clock_s": 16.808,
+   "max_lag_s": 0.0071,
+   "stalls": 0,
+   "stalled_s": 0.0,
+   "stalled_share": 0.0,
+   "held_steps": 0,
+   "late_steps": 0,
+   "requests_refused": 0,
+   "request_busy_s": 0.446,
+   "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/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--gkskpcv/replay.mp4",
+  "duration_s": 2.68,
+  "fps": 25.0,
+  "frames": 67,
+  "sha256": "9b58a6e801737a7811cb44b8753f11fc9496369356d3d91326a530b632ee0ee9"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "ball_caught": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.44,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–2.20 s，机器人抬起双臂、张开双手等待左侧传球机来球；逐帧检查视频 2.32–2.64 s，篮球进入手部区域后在双手上方弹动，随后向手臂一侧下落，没有出现持续受控的持球状态。"
+     },
+     {
+      "t": 2.64,
+      "log_step": 202,
+      "text_zh": "推断：结合视频 2.44–2.64 s 未稳定承接的运动、log step 202 服务返回的非手部触球终止结果，以及 log step 203–204 中 agent 自述最终尝试未接住球，判定双手接球未完成；不由零分单独反推。服务记录最终 episode 6 的 control_steps=264，log step 3 的 control_dt=0.01 s，264×0.01=视频 2.64 s，与末帧停止位置核对一致。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "托/舀",
+      "note_zh": "观察：视频 2.40–2.64 s，机器人两只张开的手在胸前从球下方迎接、承托来球，球短暂进入掌部上方后弹动，未见手指合拢包握。此处记录实际承接触碰，不表示已完成双手持球；log step 202 服务另确认发生非手部触球，具体部位无法由该视角分清。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–2.64 s，机器人用双脚支撑身体，站立进行迎球，未见摔倒。"
+     }
+    ]
+   },
+   "shot_through_ring": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.64,
+      "log_step": 202,
+      "text_zh": "观察：视频 0.00–2.64 s 只见来球和接球尝试，机器人没有将球投向篮圈，篮球未接近或穿过篮圈；log step 202 确认最终 episode 6 已因非手部触球终止。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [],
+ "failure": {
+  "breakpoint": {
+   "ref": "ball_caught",
+   "t": 2.44,
+   "log_step": 202,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 2.44–2.64 s，机器人正在用双手迎接来球，球在手部上方弹动而未被稳定控制；log step 202 的最终 episode 6 服务结果为“the ball touched a part of you other than your hands”，该次接球未能修复便结束，投篮尚未开始。"
+  },
+  "unmet": {
+   "ball_caught": false,
+   "shot_through_ring": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "动态时机",
+    "t": 2.44,
+    "log_step": 201,
+    "evidence_zh": "观察：视频 2.44–2.64 s，运动中的篮球到达后弹到张开的双手上方，两手未随球建立稳定承接，球又下降至手臂附近；log step 201 最终动作将闭手安排在仿真 2.75 s，log step 202 服务在控制步 264 即结束。推断：接球动作未跟上来球接触后的运动，闭手尚未来得及执行便已失败。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "log_step": 201,
+    "evidence_zh": "推断：log step 162 的动作生成代码在闭手前保持同一迎球关节目标且 grip=0，log step 165 的驱动一次性提交预生成轨迹，观察调用只保存画面，不根据球的运动修改动作；log step 201 最终配置继续采用该开环方案并延迟闭手。视频 2.44–2.64 s 显示来球进入手部区域后弹动且未被控制，log step 202 服务确认非手部触球终止，因此归为接球控制失败；没有足够的实际接触及跟踪数据确定更细子类。",
+    "source_zh": "日志 / 视频"
+   }
+  ],
+  "gap_zh": "没有状态日志，场地远景及放大图仍不能逐手核实实际接触对、0.10 秒有效双手接触累计或接触力；最终非手部触球由 log step 202 服务结果确认，但无法独立辨清具体身体部位，log step 203–204 所称前臂仅是 agent 自述。接球未完成依据可见的未受控球运动和最终尝试记录综合推断；控制子类不再细分。"
+ },
+ "summary_zh": "最终评分采用 episode 6，视频中机器人站立迎接左侧来球，但未形成稳定双手持球，双手接住球及投篮命中均未完成（0/2）。断点在接球操作中：视频约 2.44 s 来球到达后在双手上方弹动，最终因非手部触球结束，开环迎球及闭手动作未能跟上球的运动。缺少状态日志和逐手接触数据，无法精确核实接触时长、违规部位及更细控制原因。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "最终 episode 6 的标签与视频、log step 202 服务结果一致：双手接球未完成、投篮未开始，断点 ball_caught/操作中，时钟换算（264×0.01=2.64 s）有核对；现象“动态时机”和根因“控制”有证据且不过度细分。未发现需要修改的内容。",
+    "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": "2026-10-31T01:15:20Z",
+  "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/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--gkskpcv/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--gkskpcv/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "8db20af6a7bc137827dc3960112b3197503a2b87b6105049876ff72da3e32236",
+   "video_sha256": "9b58a6e801737a7811cb44b8753f11fc9496369356d3d91326a530b632ee0ee9",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch_and_shoot/trajectories/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--hdzisl6.json

@@ -0,0 +1,241 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--hdzisl6",
+ "taskKey": "embodied-olympics-fun/catch_and_shoot",
+ "dag_sha256": "f6a6d84d578d7c1084b9432c694b8d43d8fa3421ae8d0142dbb9a316794fb326",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 326,
+   "sim_steps": 1762,
+   "calls": 707,
+   "rows": 2983,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 767.8,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "realtime": 1,
+   "slowdown": 1.0,
+   "clock_s": 18.003,
+   "max_lag_s": 0.1595,
+   "stalls": 0,
+   "stalled_s": 0.0,
+   "stalled_share": 0.0,
+   "held_steps": 308,
+   "late_steps": 2,
+   "requests_refused": 0,
+   "request_busy_s": 0.743,
+   "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/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--hdzisl6/replay.mp4",
+  "duration_s": 3.28,
+  "fps": 25.0,
+  "frames": 82,
+  "sha256": "ed0312362b207a79d1a250fb6f26e523113021764b6da5010e1a817ab59b0319"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "ball_caught": {
+    "method": "visual",
+    "intervals": [
+     [
+      2.64,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 2.64,
+      "log_step": 144,
+      "text_zh": "观察：视频 2.32–2.48 s，橙色篮球由左侧飞到机器人胸前；视频 2.52–2.72 s，篮球停在两只手之间并随手小幅移动，没有先触地。log step 144 是重置后的第 6 次现场尝试，服务裁判最终返回 stages_reached=2/3（接球、释放），与画面一致。推断：双手持续接触的接球事件约在视频 2.64 s 达成，远景估时误差约 0.1 s；随后离手不取消此事件。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 2.52–2.76 s，机器人两只手从两侧共同夹持胸前的篮球；远景不能确认单手内手指与拇指的对向包握，按可见的双手相向夹持记录。候选：夹持；现有词表没有双手夹压的专门条目，不把手指闭合指令当作已建立抓握的证据（DECISIONS.md D12）。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0–3.24 s，机器人的双脚在地板上支撑身体，接球和挥臂投篮期间未见摔倒。"
+     }
+    ]
+   },
+   "shot_through_ring": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.88,
+      "log_step": 144,
+      "text_zh": "观察：视频 2.76–2.88 s，机器人双手降到胸腹前，篮球与手分开并落到双手下方；视频 2.92–3.00 s，双臂上挥而球仍在低处，视频 3.04–3.24 s 球继续下落到机器人前方地板附近，始终没有飞到右侧篮圈。log step 144 的本局服务裁判返回 the ball touched the floor、stages_reached=2/3，终止于 control_steps=326，投篮未命中。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 2.68–2.84 s，机器人两手共同夹持篮球并向下预摆，随后分手；视频 2.88–3.00 s 两臂继续向上挥动，球已在手下方。候选：夹持；只能确认双手共同夹持、送球，不能确认单手内的对向包握，现有词表缺少双手夹压条目（DECISIONS.md D12）。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0–3.24 s，机器人的双脚在地板上支撑身体，接球和挥臂投篮期间未见摔倒。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [],
+ "failure": {
+  "breakpoint": {
+   "ref": "shot_through_ring",
+   "t": 2.84,
+   "log_step": 144,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 2.64 s 左右机器人已双手接住球；视频 2.80–2.88 s 球在胸腹前离手向下，视频 2.92–3.00 s 双臂才明显上挥，球没有随手被投向篮圈。log step 144 的第 6 次现场尝试由服务裁判以球触地终止，接球和释放两阶段��完成；故失败步骤是投篮命中。"
+  },
+  "unmet": {
+   "ball_caught": true,
+   "shot_through_ring": false
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没保持住",
+    "t": 2.84,
+    "log_step": 144,
+    "evidence_zh": "观察：视频 2.56–2.72 s，机器人双手夹持胸前的篮球；视频 2.80–2.84 s，球在下沉预摆后落到双手下方，视频 2.92–3.00 s 双臂才明显上挥，球继续向下而未随手上抛。inputs/run.json 的 log step 144（第 6 次现场尝试）中，右手持球读数 hold_r 在仿真 2.56–2.76 s 为 1.0，仿真 2.78 s 降为 0.0；右手接触力读数 hf_r 在仿真 2.74、2.76、2.78 s 依次为 1.00、0.64、0.27，最终服务裁判返回球触地、stages_reached=2/3。推断：球在已持住后、完成上挥送球前脱手，属于没保持住。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "t": 2.84,
+    "log_step": 138,
+    "evidence_zh": "推断：agent 的投篮控制没有使持球加速与释放配合起来。log step 138 的最终脚本在触发后按固定序列执行夹球、下沉、14 个控制步的加速挥臂，在挥臂后段降低 grip 指令并立即撤臂；log step 144 确认第 6 次现场执行了该序列。视频 2.76–2.88 s 球已向下离手，视频 2.92–3.00 s 双臂再向上挥动，表明实际送球未实现脚本意图的向前上方加速出手；现有证据不足以进一步区分接触承载、关节跟踪与释放参数。",
+    "source_zh": "日志 / 视频"
+   }
+  ],
+  "gap_zh": "没有状态日志；场地远景足以判断接住、离手与未命中，但不能独立核实双手各自的毫秒接触窗口、单手内手指与拇指的接触分布或球离手速度。接球时刻取视频近似值，完成判定结合本次现场服务裁判的阶段结果；根因保留在控制大类，不能确定具体子类。"
+ },
+ "summary_zh": "评分对象是 log step 144 的第 6 次现场尝试：机器人约在视频 2.64 s 双手接住来球，完成 1/2 个 DAG 节点，接球事件在离手后仍保留。投篮时球约在视频 2.84 s 于胸腹前向下离手，双臂随后上挥却未将球送向篮筐，最终触地，表现为持球过程中没保持住，证据支持投篮控制配合失败。无状态日志，远景不能确定精确接触分布、释放速度或控制子类。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 2,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "fix",
+    "conf": 6,
+    "summary_zh": "时钟、节点（接球达成、投篮未达成）、断点 shot_through_ring 和 contact_actual 均与视频及服务裁判 2/3 阶段一致；唯一问题是现象：球是在预摆/下沉阶段从双手夹持中掉落（没保持住），而非出手姿态偏差（位姿不准）。",
+    "n_issues": 1,
+    "n_major": 1
+   },
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 2,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "接球达成（约视频 2.64 s）、投篮未达成、断点 shot_through_ring、接触没保持住与控制根因均与视频和服务裁判 2/3 阶段一致；第 1 轮意见已采纳，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": "2026-10-31T01:15:05Z",
+  "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/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--hdzisl6/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--hdzisl6/04_review_r1"
+   },
+   {
+    "stage": "reconcile",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--hdzisl6/05_reconcile_r1"
+   },
+   {
+    "stage": "traj_review",
+    "round": 2,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--hdzisl6/04_review_r2"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "3e0f309d0415de76fa088abbb541d95fd365cdd6ad14889f6f13de1f8fdc30a2",
+   "video_sha256": "ed0312362b207a79d1a250fb6f26e523113021764b6da5010e1a817ab59b0319",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch_and_shoot/trajectories/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--hwbzmay.json

@@ -0,0 +1,200 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--hwbzmay",
+ "taskKey": "embodied-olympics-fun/catch_and_shoot",
+ "dag_sha256": "f6a6d84d578d7c1084b9432c694b8d43d8fa3421ae8d0142dbb9a316794fb326",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 239,
+   "sim_steps": 1692,
+   "calls": 40,
+   "rows": 9000,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 1407.7,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "realtime": 1,
+   "slowdown": 1.0,
+   "clock_s": 16.924,
+   "max_lag_s": 0.0587,
+   "stalls": 0,
+   "stalled_s": 0.0,
+   "stalled_share": 0.0,
+   "held_steps": 0,
+   "late_steps": 0,
+   "requests_refused": 0,
+   "request_busy_s": 0.569,
+   "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/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--hwbzmay/replay.mp4",
+  "duration_s": 2.4,
+  "fps": 25.0,
+  "frames": 60,
+  "sha256": "ebe45e239715f99c3901967c1a846cc02d0568a0c71957121174d54f9da3f7af"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "ball_caught": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.24,
+      "log_step": null,
+      "text_zh": "观察：已逐帧检查视频 0.00–2.36 s；机器人先把两手伸到身前，视频 2.00–2.32 s 橙色球从左侧传球机飞近机器人肩侧，双手仍在其身前，未见球被两手共同截停、持住。"
+     },
+     {
+      "t": 2.36,
+      "log_step": 122,
+      "text_zh": "观察：视频 2.36 s 球到达机器人肩侧并与身体轮廓重叠，未进入双手之间；log step 121 启动 episode 6，log step 122 的服务 FINAL_META 报告 control_steps=239、stages_reached=0/3，并因“the ball touched a part of you other than your hands”判负。该服务结果与视频末段相符，接球事件未成立。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：视频 0.80–2.36 s 机器人两臂在身前准备接球，来球最终到达肩侧；log step 122 的服务结果确认发生非手部触球。远景中球与肩、上臂及躯干轮廓重叠，无法区分实际碰到的部位，也无法确认此前是否有极短的手掌或手指接触，故不将准备动作记成已发生的双手抓握或夹持。"
+     }
+    ]
+   },
+   "shot_through_ring": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.36,
+      "log_step": 122,
+      "text_zh": "观察：视频 0.00–2.36 s 仅有机器人准备接球、传球机来球和球到达机器人肩侧，未见机器人持球后向画面右侧篮筐投出，也没有球穿圈；log step 122 的服务在接球前即以非手部触球结束 episode 6。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [],
+ "failure": {
+  "breakpoint": {
+   "ref": "ball_caught",
+   "t": 2.32,
+   "log_step": 122,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 2.20–2.36 s 来球飞到机器人肩侧，双手在身前却未把球共同接住；log step 122 的服务报告最终 episode 6 接球阶段未成立，并因非手部触球终止。本次失败发生在接球操作中，尚未开始投篮。"
+  },
+  "unmet": {
+   "ball_caught": false,
+   "shot_through_ring": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "动态时机",
+    "t": 2.32,
+    "log_step": 122,
+    "evidence_zh": "观察：视频 2.00–2.36 s 球从传球机飞近，机器人双手未随来球完成共同拦截，球直接到达肩侧；log step 122 的服务确认接球前发生非手部触球。推断：机器人未在来球经过的窗口把双手与球对上，属于运动目标的接球错失。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "evidence_zh": "观察：log step 121 启动数值手位逆解与定时接球控制器，并提交动作计划；视频 0.80–2.36 s 机器人已将双手伸出，但来球到达肩侧而未被两手截停，log step 122 的服务随后以非手部触球判负。推断：接球动作未使双手在来球路径上建立有效控制；证据只支持控制层，不能进一步确定为接触设计、逆解跟踪或某个具体参数不当。",
+    "source_zh": "视频 / 日志"
+   }
+  ],
+  "gap_zh": "没有状态日志；现有远景无法分辨视频 2.36 s 的具体非手接触部位及此前是否存在瞬时手部接触，故 contact_actual 保留未知。最终现场尝试的日志未给逐时手球接触记录，不能把控制问题进一步归入具体子类；agent 在 log step 123–124 所称“躯干”仅为其自述。此前五次尝试不用于本次节点进度或失败现象。"
+ },
+ "summary_zh": "评分对应 log step 121–122 的最后一次现场尝试，双手接住球与投篮命中均未完成，节点最终为 0/2。视频约 2.3 s 来球到达机器人肩侧而未被双手承接，服务以非手部触球判负，断点为接球，现象为动态时机，根因只归到控制层。没有状态日志，远景不足以确定具体碰撞部位、瞬时手部接触及控制子类。",
+ "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 2.4 s 对应最后一次现场 episode（control_steps 239≈2.39 s），球约 2.0 s 出现并在 2.3 s 到达机器人肩侧，双手未接住；服务判负“非手部触球”，stages 0/3。两节点未达成、断点为接球、contact_actual 未知均有证据支持。",
+    "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": "2026-10-31T01:14:59Z",
+  "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/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--hwbzmay/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--hwbzmay/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "3dc6b379f61bf9a08a9d8c4bfc109e424fc8a3a91617476e1c82c84821b29704",
+   "video_sha256": "ebe45e239715f99c3901967c1a846cc02d0568a0c71957121174d54f9da3f7af",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch_and_shoot/trajectories/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--lrxseia.json

@@ -0,0 +1,227 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--lrxseia",
+ "taskKey": "embodied-olympics-fun/catch_and_shoot",
+ "dag_sha256": "f6a6d84d578d7c1084b9432c694b8d43d8fa3421ae8d0142dbb9a316794fb326",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 452,
+   "sim_steps": 1869,
+   "calls": 1223,
+   "rows": 6400,
+   "resets": 4,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 6449.0,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "realtime": 1,
+   "slowdown": 1.0,
+   "clock_s": 23.608,
+   "max_lag_s": 0.5084,
+   "stalls": 8,
+   "stalled_s": 4.036,
+   "stalled_share": 0.171,
+   "held_steps": 0,
+   "late_steps": 821,
+   "requests_refused": 4,
+   "request_busy_s": 1.171,
+   "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/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--lrxseia/replay.mp4",
+  "duration_s": 4.56,
+  "fps": 25.0,
+  "frames": 114,
+  "sha256": "230ecbd789aa4498a80b1df4ed81298419aed37b1304a82247460c4bd7888a68"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "ball_caught": {
+    "method": "visual",
+    "intervals": [
+     [
+      2.5,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 2.5,
+      "log_step": 230,
+      "text_zh": "观察：视频 2.32–2.56 s，左翼来球进入机器人双手之间，横向运动被截停，球随双手下沉后被向上送起，接球前未落地。log step 230 的机器人服务（现场仿真与裁判接口）报告 episode=5、stages_reached=2/3，已通过接球及释放阶段；接球成立时间按画面约记为视频 2.50 s。"
+     },
+     {
+      "t": 4.52,
+      "log_step": 230,
+      "text_zh": "推断：接球是锁存事件，视频 2.68 s 球离手以及视频约 4.52 s 触地不会取消已成立的接球。log step 230 的 calls=1223、rows=6400、control_steps_total=1869 与评分器字段相同，确认第 5 次现场尝试为本次评分局；log step 250 只记录启动后续脚本的请求，没有新现场结果，不能代替此局。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "未知",
+      "note_zh": "观察：视频 2.36–2.56 s，机器人左右手在胸腹前共同承接篮球，球留在双手区域并随手运动。远景中手指与部分球面互相遮挡，不能判定各手有无形成对向包握，故不强选抓握或开放双掌夹持。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 2.32–2.56 s，机器人双脚着地支撑接球动作，躯干保持直立。"
+     }
+    ]
+   },
+   "shot_through_ring": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.68,
+      "log_step": 230,
+      "text_zh": "观察：视频 2.52–2.68 s，机器人双手将接到的篮球由腹前向上送出；视频 2.68 s 球已在手上方分离，随后沿高弧线向画面右侧飞行。"
+     },
+     {
+      "t": 3.92,
+      "log_step": 230,
+      "text_zh": "观察：视频 3.76–4.52 s，篮球从篮圈画面左侧下降并落到篮筐前方地板，没有从篮圈上方进入并穿到圈下。log step 230 的第 5 局服务终态为 failed、outcome=\"the ball touched the floor\"、stages_reached=2/3，与画面一致；成功参照视频 6.40–6.60 s 则可见球朝圈内下降，参照只用于辨认穿圈线索。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "未知",
+      "note_zh": "观察：视频 2.52–2.68 s，机器人左右手共同上送篮球，随后球离手；此远景不能分辨单手包握与双掌夹持的具体接触，怎么碰记未知。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 2.52–4.52 s，机器人双脚支撑投篮及随挥动作，未见摔倒。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [],
+ "failure": {
+  "breakpoint": {
+   "ref": "shot_through_ring",
+   "t": 2.68,
+   "log_step": 230,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 2.52–2.68 s，机器人接球后向上送球并释放；视频 3.76–4.52 s 球在圈外下降直至触地，投篮始终未命中。log step 230 的现场裁判已确认接球和释放两阶段、未确认第三阶段，并以球触地结束本局，因此断点是投篮命中。"
+  },
+  "unmet": {
+   "ball_caught": true,
+   "shot_through_ring": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 2.68,
+    "log_step": 230,
+    "evidence_zh": "观察：视频 2.68 s 出手后，篮球虽向篮筐方向升高，但视频 3.76–4.16 s 下降路径明显位于篮圈画面左侧，随后视频约 4.52 s 落地，未穿圈。log step 230 的服务报告球触地失败，接球及释放已完成。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "evidence_zh": "推断：视频 2.52–4.52 s 显示机器人完成送球与释放，但实际出球运动未将球送过固定篮圈，属于投篮动作控制未达所需方向与速度。log step 230 记录控制器选择 POLICY 20、lambda=0.5、ball None，说明预期的球位测量未用于选择投篮变体；log step 218 的脚本以该默认值选择预制运动。该日志支持此次执行缺少相应球位修正，但不足以分清测量、局部接触模型与运动跟踪各自的贡献，根因仅标到控制。",
+    "source_zh": "视频 / 日志"
+   }
+  ],
+  "gap_zh": "没有状态日志；视频远景能确认双手接球及送球，但手指与球面遮挡，无法区分抓握与开放双掌夹持，也不能独立精确测量接球的 0.10 秒接触窗口，接球验收结合本局服务两阶段记录确认。没有本局实际出球速度、接触力和逐步目标跟踪数据，不能进一步确定控制子类；接球时刻约有 0.1 秒误差。"
+ },
+ "summary_zh": "评分局完成双手接住球，随后上送出手，但篮球在篮圈外下降并触地，最终完成 1/2 个节点，断点为投篮命中。视频与 log step 230 的第 5 次现场服务结果一致；log step 250 声称启动的后续尝试没有新服务结果，不作为评分轨迹。投篮出球控制未命中固定目标，但缺少状态及接触数据，具体控制子类和手部接触方式无法确认。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "接球成立（log step 230 服务 stages_reached=2/3，视频约 2.5 s 双手承接）、投篮未命中（球触地）与标注一致；断点 shot_through_ring、位姿不准、控制根因均有证据支持，时钟均为视频时间，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": "2026-10-31T01:14:42Z",
+  "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/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--lrxseia/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--lrxseia/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "916ef5cdab362c6fae0c76be8bfed30a49172d6fe850ce07d1b06b81e093ba29",
+   "video_sha256": "230ecbd789aa4498a80b1df4ed81298419aed37b1304a82247460c4bd7888a68",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch_and_shoot/trajectories/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--mxqsqwa.json

@@ -0,0 +1,205 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--mxqsqwa",
+ "taskKey": "embodied-olympics-fun/catch_and_shoot",
+ "dag_sha256": "f6a6d84d578d7c1084b9432c694b8d43d8fa3421ae8d0142dbb9a316794fb326",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 241,
+   "sim_steps": 1587,
+   "calls": 21,
+   "rows": 5800,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 2673.1,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "realtime": 1,
+   "slowdown": 1.0,
+   "clock_s": 16.482,
+   "max_lag_s": 0.5956,
+   "stalls": 1,
+   "stalled_s": 0.596,
+   "stalled_share": 0.0361,
+   "held_steps": 0,
+   "late_steps": 0,
+   "requests_refused": 1,
+   "request_busy_s": 0.739,
+   "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/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--mxqsqwa/replay.mp4",
+  "duration_s": 2.44,
+  "fps": 25.0,
+  "frames": 61,
+  "sha256": "f0dd66d65af1d23452d11d15bfe5c01306433240e4bcd23872f884ef4070e528"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "ball_caught": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 1.6,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–1.6 s，机器人双脚站立并将双臂伸到胸前迎球；视频 2.08–2.40 s，橙球由左侧传球机飞近机器人，末帧进入其左侧上肢与躯干附近，未见双手共同承接并控制球。"
+     },
+     {
+      "t": 2.4,
+      "log_step": 262,
+      "text_zh": "观察：log step 262 明确运行最后一次现场评分尝试，服务返回 sim_time_s=2.41（仿真秒）、stages_reached=0、resets_left=0，失败原因是“the ball touched a part of you other than your hands”（球触及非手部位）。视频 2.40 s 的来球到达及随即结束与该结果相符；并非由总分零单独判定未接住。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：视频 2.36–2.40 s 来球进入机器人左侧身体轮廓，log step 262 的现场服务确认球碰到非手部位；远景无法分清是臂、肩或躯干，也不能确认此前是否有短暂手部触球。双臂虽伸出迎球，但没有双手夹持或抓握成立的证据。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.0–2.40 s，机器人双腿及双脚在地板上支撑站立，双臂执行迎球动作。"
+     }
+    ]
+   },
+   "shot_through_ring": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.4,
+      "log_step": 262,
+      "text_zh": "观察：完整视频 0.0–2.40 s 只有机器人准备接球及传球机的来球飞行，未见机器人持球、向篮筐投出或球穿篮；log step 262 的最后现场尝试在接球阶段即报告非手触球并以 stages_reached=0 结束。投篮前置未完成，机器人尚未开始投篮。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [],
+ "failure": {
+  "breakpoint": {
+   "ref": "ball_caught",
+   "t": 2.4,
+   "log_step": 262,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 2.32–2.40 s，机器人双臂仍在胸前等待来球，球到达其左侧身体附近而未被双手控制；log step 262 的最后一次现场尝试由裁判以球碰到非手部位终止，接球事件未成立。该断点属于接球操作，日志中的其他重置尝试及本地测试不计入本段视频进度。"
+  },
+  "unmet": {
+   "ball_caught": false,
+   "shot_through_ring": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "被挡/碰撞",
+    "t": 2.4,
+    "log_step": 262,
+    "evidence_zh": "观察：视频 2.36–2.40 s，来球进入机器人左侧身体轮廓，双手未将球共同承接；log step 262 的现场服务裁判确认球碰到手以外的身体部位，并终止本次尝试。该非手部碰撞发生于接球时，未见先接稳后掉球。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "evidence_zh": "推断：log step 261–262 显示 agent 将本地逆解得到的接球关节姿态写入预编程动作并直接运行最后一次尝试；视频 1.6–2.40 s 中机器人已伸臂迎球，但来球仍进入身体附近，现场服务在 log step 262 确认非手触球且接球未成立。该接球控制未实现合法双手承接；现有证据不足以进一步区分接触几何设计、逆解误差或实际关节跟踪问题。",
+    "source_zh": "日志 / 视频"
+   }
+  ],
+  "gap_zh": "无状态日志；视频为单一远景，末帧不足以辨认具体非手触球部位、短暂手部接触及双手接触的毫秒时长。log step 262 提供最后评分尝试的服务判定，但未提供该尝试完整关节跟踪及接触对记录；log step 263 中 agent 关于左前臂位置和本地模型差异的解释仍待核实，因此不据此细分控制根因。"
+ },
+ "summary_zh": "本次评分视频中机器人站立伸出双臂迎接左侧来球，但没有完成双手接球，投篮尚未开始，最终完成 0/2 节点。视频约 2.40 s 来球到达身体附近，最后一次现场尝试的服务结果确认非手触球并终止，断点为双手接住球，根因保守归为控制。远景与缺少状态日志使具体触球部位和控制子原因无法确定，不能将会话中的本地测试当作本次完成证据。",
+ "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 0–2.44 s（视频时钟）只见机器人伸臂、球在末尾到达身体附近；log step 262 的服务判定为球碰非手部位、stages_reached=0，与标注的 ball_caught 未达、断点、被挡/碰撞现象及保守的控制根因一致，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": "2026-10-31T01:14:16Z",
+  "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/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--mxqsqwa/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--mxqsqwa/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "f55b6f307780525cc0b78bcec55ca4f0fdc5680106b1fe801da4da8b873888e4",
+   "video_sha256": "f0dd66d65af1d23452d11d15bfe5c01306433240e4bcd23872f884ef4070e528",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch_and_shoot/trajectories/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--srraf8q.json

@@ -0,0 +1,186 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--srraf8q",
+ "taskKey": "embodied-olympics-fun/catch_and_shoot",
+ "dag_sha256": "f6a6d84d578d7c1084b9432c694b8d43d8fa3421ae8d0142dbb9a316794fb326",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "calls": 4,
+   "rows": 0,
+   "resets": 0,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 7171.7,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 4,
+   "realtime": 1,
+   "slowdown": 1.0,
+   "clock_s": 0.0,
+   "max_lag_s": 0.0,
+   "stalls": 0,
+   "stalled_s": 0.0,
+   "stalled_share": 0.0,
+   "held_steps": 0,
+   "late_steps": 0,
+   "requests_refused": 0,
+   "request_busy_s": 0.246,
+   "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/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--srraf8q/replay.mp4",
+  "duration_s": 0.04,
+  "fps": 25.0,
+  "frames": 1,
+  "sha256": "6624d12bfebe12839cfbb24ef9895764177edf78482ee6c37ce2f9a4358cc9dc"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "ball_caught": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": 4,
+      "text_zh": "观察：视频 0.00 s（全片唯一帧）中机器人站在初始位置，双手前方没有被共同控制的篮球；log step 4 的现场机器人服务返回 control_steps=0、rows=0、running=false。"
+     },
+     {
+      "t": 0.0,
+      "log_step": 15,
+      "text_zh": "观察：log step 15 的三次现场 get_obs(seconds=0.0) 均返回 running=False；评分器记录 n_actions=0、sim_steps=0、rows=0、resets=0，运行记录 sim_total.submit=0。推断：现场评分局始终未启动，未发生双手接球事件。"
+     }
+    ]
+   },
+   "shot_through_ring": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00 s 的完整初态画面未见机器人持球、出手或篮球穿圈；评分器记录 n_actions=0、sim_steps=0、clock_s=0.0。推断：评分局没有推进，接球前置与投篮命中事件均未发生；此判断并非仅依据 success=0。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [],
+ "failure": {
+  "breakpoint": {
+   "ref": "ball_caught",
+   "t": 0.0,
+   "log_step": 15,
+   "where": "首步前",
+   "evidence_zh": "观察：视频 0.00 s 仅显示机器人初始站姿，整段视频只有这一帧；log step 4、15 的现场服务均显示未运行，评分器记录 n_actions=0、sim_steps=0，运行记录 sim_total.submit=0。推断：现场接球动作从未启动，断点为双手接住球之前；视频 0.00 s 仅作初态导航锚点。"
+  },
+  "unmet": {
+   "ball_caught": false,
+   "shot_through_ring": false
+  },
+  "phenomena": [
+   {
+    "cat": "没尝试",
+    "item": "没尝试",
+    "t": 0.0,
+    "log_step": 15,
+    "evidence_zh": "观察：视频 0.00 s 的唯一帧为机器人初态；log step 15 仅作三次不推进仿真的现场观测，返回 running=False，评分器的动作数与仿真步数均为 0。推断：本次现场评分局没有开始接球，离线测试不构成评分局中的尝试。"
+   }
+  ],
+  "root_causes": [],
+  "gap_zh": "观察：本次视频存在但仅一帧，inputs/ 内无状态日志。log step 46 创建 agent 自建的 LocalSim，后续接触、接球及投篮数据来自该离线场景；log step 251 的最后一条 agent 自述仍为运行左臂接球解测试，log step 258 最后调用离线完整流程测试且没有返回文本。运行记录 wall_s=7200 与 time_limit_s=7200 相等。推断：会话预算结束前没有启动现场控制，但现有证据不能确定为何未能将离线方案用于现场，也不足以区分控制子类或认知根因，故根因留空；不将离线物理失败当作评分局现象，不将预算耗尽当作根因。"
+ },
+ "summary_zh": "现场评分局仅保留视频 0.00 s 的初态帧，评分器动作数和仿真步数均为 0，双手接球与投篮命中均未达到（0/2）。断点为双��接住球的首步前，现象为没尝试；agent 的会话工作停留在自建 MuJoCo 场景中的接球、持球转姿和投篮调试。缺少可确认未启动现场控制之深层原因的证据，根因留空，离线测试结果不计作本次进度或接触。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "评分局从未启动（n_actions=0、sim_steps=0、log step 4/15 running=False），两节点未达成、断点 ball_caught 首步前、现象没尝试均有证据支持；根因留空并附 gap_zh 合理，未把预算耗尽或离线测试当作现象/根因。",
+    "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": "2026-10-31T01:13:19Z",
+  "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/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--srraf8q/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--srraf8q/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "0c6edb475787c68d2597eed91882d7cec0f6b698016a0871514b44ba102c33da",
+   "video_sha256": "6624d12bfebe12839cfbb24ef9895764177edf78482ee6c37ce2f9a4358cc9dc",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch_and_shoot/trajectories/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--sywzh9k.json

@@ -0,0 +1,265 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--sywzh9k",
+ "taskKey": "embodied-olympics-fun/catch_and_shoot",
+ "dag_sha256": "f6a6d84d578d7c1084b9432c694b8d43d8fa3421ae8d0142dbb9a316794fb326",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 315,
+   "sim_steps": 594,
+   "calls": 597,
+   "rows": 1200,
+   "resets": 1,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 1989.9,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "realtime": 1,
+   "slowdown": 1.0,
+   "clock_s": 6.33,
+   "max_lag_s": 0.2589,
+   "stalls": 0,
+   "stalled_s": 0.0,
+   "stalled_share": 0.0,
+   "held_steps": 0,
+   "late_steps": 0,
+   "requests_refused": 0,
+   "request_busy_s": 0.528,
+   "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/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--sywzh9k/replay.mp4",
+  "duration_s": 3.16,
+  "fps": 25.0,
+  "frames": 79,
+  "sha256": "ae65ef5de60b0bf4190c3f6073b921a9ea106ae11fa75d76051e29299de7b2f3"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "ball_caught": {
+    "method": "visual",
+    "intervals": [
+     [
+      2.56,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 2.56,
+      "log_step": 88,
+      "text_zh": "观察：视频 2.32–2.44 s，橙色篮球从左侧传球机飞至机器人胸腹前；视频 2.44–2.56 s 球停在两手共同形成的承接位置，未先触地。log step 88 的现场 episode 2 服务结果为 stages_reached=2、stages_total=3，log step 101 再次确认同一结果，支持接球事件已通过裁判；视频到达时刻取约 2.56 s，精确的双手接触累计时刻无法从远景测量。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "托/舀",
+      "note_zh": "观察：视频 2.44–2.64 s，机器人两只手在胸腹前共同兜托橙色篮球；视频 3.04–3.12 s 腿间可见棕橙色球体轮廓向下移动，不能再称球持续在双手处。远景无法确认每只手是否形成拇指与其余手指的对向抓握，未凭闭合命令标抓握。"
+     }
+    ]
+   },
+   "shot_through_ring": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 3.08,
+      "log_step": 89,
+      "text_zh": "观察：视频约 2.44–2.64 s 球处于双手承托区，视频 3.04–3.12 s 腿间出现并向下移动的棕橙色球体轮廓；机器人未朝右侧篮圈挥臂送球，也未见球飞向篮圈或向下穿圈。手边末段仍有橙色小块，不能仅凭这些像素认定球持续被持住，也无法据此确定精确离手时刻。log step 88 的现场 episode 2 服务结果为球触地、stages_reached=2、stages_total=3；log step 89 的服务终态文件记录 sim_time_s=3.15、outcome=\"the ball touched the floor\"，log step 101 再次确认。推断：球在短暂接住后失去承托而下落，画面末段与服务触地结果相容；未将仿真时刻直接换算为视频时刻。"
+     },
+     {
+      "t": 2.56,
+      "log_step": 88,
+      "text_zh": "观察：log step 81 的 make_plan 在准备结束后保持 qc 手臂目标，手指闭合后命令为 [1,1]；log step 86 仅在参数超过七项时启用 give 轨迹，log step 88 实际提交七参数方案，故接球后没有新增投篮挥臂或主动张手段。推断：投篮动作未启动；服务计入 release 的阶段不证明机器人进行了朝篮圈的投篮。log step 90–99 的离线投掷测试发生于该次现场尝试结束后，不属于本视频。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "托/舀",
+      "note_zh": "观察：视频约 2.56 s 接球完成后，机器人双手仍在胸腹前维持承托姿态，随后视频 3.04–3.12 s 球体轮廓在腿间下降；未见朝篮圈挥臂送球。推断：该实际接触属于接球后为投篮持球的阶段，尚未完成主动出手；不据手指闭合命令断言包握。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [],
+ "failure": {
+  "breakpoint": {
+   "ref": "shot_through_ring",
+   "t": 3.08,
+   "log_step": 88,
+   "where": "操作中",
+   "evidence_zh": "观察：视频约 2.56 s 已完成双手接球，之后机器人维持胸腹前承托姿态，视频 3.04–3.12 s 腿间可见球体轮廓下移；没有朝篮圈的挥臂送球。log step 88、89、101 的 episode 2 服务结果为 2/3 阶段、球触地失败。推断：接球事件已完成并锁存，但接球后为投篮持球的阶段失去球，故断点仍为 shot_through_ring，位置为操作中；视频 3.08 s 是失球已可见的导航锚点，不是精确离手时刻。"
+  },
+  "unmet": {
+   "ball_caught": true,
+   "shot_through_ring": false
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没保持住",
+    "t": 3.08,
+    "log_step": 88,
+    "evidence_zh": "观察：视频约 2.44–2.64 s 球被双手共同承接，视频 3.04–3.12 s 棕橙色球体轮廓已在腿间向下移动；机器人始终维持承接姿态，未见朝篮圈挥臂送球。log step 88 的 episode 2 服务结果为 stages_reached=2、stages_total=3、球触地，log step 101 再次确认。推断：已建立的双手承托未能持续保持，球在主动投篮前掉落，符合没保持住（docs/03_trajectory.md Step 3.2；docs/CASES.md FAIL-8）。精确离手时刻和持续持球时长受遮挡及远景限制。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "抓取和接触设计",
+    "alt_feasible": "已验证有",
+    "evidence_zh": "观察：视频约 2.44–2.64 s 双手在胸腹前形成兜托承接位置，而视频 3.04–3.12 s 球体轮廓在腿间下降。log step 81 的 make_plan 从准备姿态进入 qc 后保持手臂目标，手指闭合后保持 [1,1]；log step 86 的 give 分支需超过七个参数，log step 88 实际七参数调用未启用该分支，也未安排主动张手或投篮挥臂。推断：在持续保持承接目标且无主动释放段的方案下失球，支持所用双掌兜托的接触/承载设计不能稳定衔接后续持球阶段，归抓取和接触设计；不能从远景断言手指完全没有包握，也不能分辨接触几何、手臂回弹与实际握持力各自的作用。观察：参考 inputs/oracle/replay.mp4 的参考视频约 2.4–3.4 s 可见双手持球，参考视频约 3.6–4.4 s 持球下摆，参考视频约 4.8–5.2 s 双手上送并出球；inputs/oracle/oracle.json 的 success=1。推断：参考成功验证存在任务级可行替代，但不证明本次承接口袋可持续持球，也不把参考动作当成本次动作。",
+    "source_zh": "视频 / 日志 / 评分器字段"
+   }
+  ],
+  "gap_zh": "无状态日志；手部远景较小，无法独立测量各手接触对及 0.10 秒累计窗口，也无法确认各指是否形成对向包握。视频末段手边仍有橙色小块，但腿间球体轮廓的下降及 episode 2 服务球触地终态支持失球，删除原先“画面与触地无法解释”的判断；精确离手时刻仍不清楚。log step 89 的 track.py 球高仅为 agent 单目估计，未用来给视频定时。根因是由实际保持姿态的程序及失球结果作出的接触设计推断，缺少接触力、实际关节运动与接触几何数据，不能证明无包握或确定具体失稳机理。实际方案缺少投篮段仍由日志支持，但不把它替代已发生的失球现象。"
+ },
+ "summary_zh": "评分尝试完成双手接球，最终完成 1/2 个 DAG 节点；视频末段约 3.04–3.12 s 可见球体在腿间下降，服务确认 episode 2 球触地失败。断点仍为投篮命中步骤，在接球后持球阶段没保持住；结合实际方案保持承接目标且无主动释放段，根因推断为抓取和接触设计。远景无法确定精确离手时刻、手指包握与失稳机理，置信度维持 6。",
+ "conf": 6,
+ "review": {
+  "status": "agreed",
+  "rounds": 3,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "fix",
+    "conf": 7,
+    "summary_zh": "接球节点（约 2.56 s）与 1/2 进度基本正确；但多处观察性文字称视频 2.6–3.12 s 球「从手边滑落、在腿旁下降」，逐帧检查（2.0–3.16 s）显示球一直停在双手之间直到视频结束，该观察不成立，据此建立的「没保持住」现象与根因需要改写。",
+    "n_issues": 3,
+    "n_major": 2
+   },
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 2,
+    "verdict": "fix",
+    "conf": 7,
+    "summary_zh": "接球节点（约 2.56 s）与 1/2 进度正确；但文件反复断言「视频 2.56–3.12 s 球一直在双手处」不成立：放大裁剪显示球在约 3.04 s 之后已离开双手，3.08、3.12 s 帧手处无橙色、球落在腿脚附近，与服务终态 stages 2/3『ball touched the floor』（sim 3.15 s）一致。因此「没尝试」现象、『差异无法解释』的 gap 及根因证据需据此重写。",
+    "n_issues": 4,
+    "n_major": 3
+   },
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 3,
+    "kind": "confirm",
+    "verdict": "agree",
+    "summary_zh": "四条回应均成立：持续持球与『无法解释』的表述已删除，现象改为没保持住（含 FAIL-8 依据），根因改为抓取和接触设计并标为推断，接球接触说明已改；对审查中过强的『手处无橙色』的部分拒绝有据，我看帧也见手边仍有橙色、腿脚处橙色增多。",
+    "n_issues": 0,
+    "n_checks": 4
+   }
+  ],
+  "open_zh": [],
+  "confirm": {
+   "verdict": "confirmed",
+   "summary_zh": "四条回应均成立：持续持球与『无法解释』的表述已删除，现象改为没保持住（含 FAIL-8 依据），根因改为抓取和接触设计并标为推断，接球接触说明已改；对审查中过强的『手处无橙色』的部分拒绝有据，我看帧也见手边仍有橙色、腿脚处橙色增多。",
+   "checks_ok": "4/4",
+   "n_factual_errors": 0
+  }
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T01:19:26Z",
+  "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/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--sywzh9k/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--sywzh9k/04_review_r1"
+   },
+   {
+    "stage": "reconcile",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--sywzh9k/05_reconcile_r1"
+   },
+   {
+    "stage": "traj_review",
+    "round": 2,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--sywzh9k/04_review_r2"
+   },
+   {
+    "stage": "reconcile",
+    "round": 2,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--sywzh9k/05_reconcile_r2"
+   },
+   {
+    "stage": "confirm",
+    "round": 3,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-standard-realtime--sywzh9k/06_confirm_r3"
+   }
+  ],
+  "confirm_prompt": "prompts/06_confirm.md",
+  "confirm_prompt_sha256": "b9f56c095f7ea5e609e58bb21e076f565766c370333d71c7a3512cffb4d78267",
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "07d5371815f0f38d8fbf6cb3e54bb3af9a3a7df8a0a58138fa69fd6da545b6e4",
+   "video_sha256": "ae65ef5de60b0bf4190c3f6073b921a9ea106ae11fa75d76051e29299de7b2f3",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/clean_and_jerk/trajectories/embodied-olympics-fun-clean-and-jerk-i00-privileged--7thjiwr.json

@@ -0,0 +1,219 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-clean-and-jerk-i00-privileged--7thjiwr",
+ "taskKey": "embodied-olympics-fun/clean_and_jerk",
+ "dag_sha256": "f530c1d45bf82223dabd69c9d30cf24e240db1d588782640954595693a7015ca",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "n_actions": 300,
+   "sim_steps": 300,
+   "horizon": 6000,
+   "episode_over": 0,
+   "progress": 0.0,
+   "stages_reached": 0,
+   "stages_total": 4,
+   "fell": 0,
+   "replay_s": 5.3,
+   "bad_trajectory": 0,
+   "metric_bar_height_m": 0.375,
+   "metric_bar_max_height_m": 0.375,
+   "metric_elbow_flex_deg": 89.0,
+   "metric_knee_flex_deg": 20.6,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 47,
+   "agent_resets": 0,
+   "agent_replays": 3,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.69,
+   "peak_processes": 11,
+   "peak_disk_gb": 0.24,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-clean-and-jerk-i00-privileged--7thjiwr/replay.mp4",
+  "duration_s": 3.04,
+  "fps": 25.0,
+  "frames": 76,
+  "sha256": "4f1de7292b08d481a5be08442930d1017d97d3ef746f2551bd9c016cde9aa814"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "bar_at_shoulders": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：本次视频 0.00–3.00 s 中，机器人始终站在木台上，双手保持在杠杆上方且未伸向杠杆；杠铃始终横放在两只棕色支架上，没有离架或升至胸肩前方。已逐帧检查视频 2.00–3.00 s，末帧仍如此，上肩步骤未启动。"
+     },
+     {
+      "t": 3.0,
+      "log_step": 137,
+      "text_zh": "观察：log step 109 的生成脚本在举重失败时另建回合并保存300步站立轨迹，log step 136 再次输出保存该备用轨迹，log step 137 确认交付文件仍为300行；这与本次视频中的站立行为一致。评分器 stages_reached=0、bar_max_height_m=0.375，支持未离架上肩。"
+     }
+    ]
+   },
+   "bar_overhead": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 3.0,
+      "log_step": null,
+      "text_zh": "观察：本次视频 0.00–3.00 s 内杠铃一直留在低位支架上，机器人双臂未上举；视频 2.00–3.00 s 的逐帧检查也未见杠铃到达头顶上方。杠铃过顶及伸臂保持均未发生，本步骤未尝试。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "bar_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": 0.375
+  },
+  {
+   "name": "bar_max_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": 0.375
+  },
+  {
+   "name": "elbow_flex_deg",
+   "unit": "deg",
+   "threshold": 10.0,
+   "direction": "lower_better",
+   "value": 89.0
+  },
+  {
+   "name": "knee_flex_deg",
+   "unit": "deg",
+   "threshold": 10.0,
+   "direction": "lower_better",
+   "value": 20.6
+  },
+  {
+   "name": "time_s",
+   "unit": "s",
+   "threshold": 60.0,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "bar_at_shoulders",
+   "t": 0.0,
+   "log_step": 136,
+   "where": "首步前",
+   "evidence_zh": "观察：视频 0.00–3.00 s 的机器人仅原地站立，双手未接触杠铃，杠铃始终留在支架上。log step 136 明示举重测试失败后保存站立备用轨迹，log step 137 确认最终文件为300行；故受评轨迹从未开始第一个上肩步骤。"
+  },
+  "unmet": {
+   "bar_at_shoulders": false,
+   "bar_overhead": false
+  },
+  "phenomena": [
+   {
+    "cat": "没尝试",
+    "item": "没尝试",
+    "t": 0.0,
+    "log_step": null,
+    "evidence_zh": "观察：视频 0.00、1.00、2.00、3.00 s 的局部放大图均显示机器人双手与杠杆分离、杠铃搁在两只支架上；全程及末秒逐帧检查未见下蹲取杠、抓杠或抬杠。受评轨迹只维持站立，上肩动作没有启动。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "source_zh": "日志",
+    "evidence_zh": "推断：未能生成稳定的举重控制轨迹，导致交付仅含站立备用动作。log step 132、136 的会话测试输出均显示双手已接触杠、lift-off阶段成立，随后骨盆速度���至超过2 m/s，裁判以非脚部触台终止；两次输出均接着说明未保存举重测试、改存站立备用轨迹。log step 109 可见该失败回退分支的代码，log step 137 确认最终300行文件。上述调试记录用于解释备用交付的成因，其离架和失稳不计入本次视频的节点或现象。"
+   }
+  ],
+  "gap_zh": "没有状态日志，无法独立核验调试阶段的接触细节、关节跟踪和失稳力矩来源；根因保留在“控制”层级，不据agent自述进一步确定前馈或参数问题。本次视频清楚显示杠铃未离架，两个节点的否定判断不受该缺口影响。"
+ },
+ "summary_zh": "受评视频全程为站立备用动作，杠铃始终留在支架上，上肩与过顶均未启动，最终完成0/2个节点。断点为杠铃上肩的首步前，现象为没尝试；日志显示举重控制测试反复失稳后，生成脚本保存了300步站立轨迹。缺少状态日志，具体控制失稳机制无法独立核验，调试中曾离架或后倒不计入本次受评进度。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "受评视频3.04 s全程站立，杠铃始终在支架上，两个节点为false、断点首步前/没尝试、根因控制（标为推断，来自会话测试，log 132/136 的 ABORT 与回退保存均已核对）均有证据支持；未发现需改的错误。",
+    "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": "2026-10-31T00:58:43Z",
+  "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/embodied-olympics-fun-clean-and-jerk-i00-privileged--7thjiwr/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-clean-and-jerk-i00-privileged--7thjiwr/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "e6e724fd2676c7118dd326238c67cae4659af62d0e971e8d99b91184f76b2cbf",
+   "video_sha256": "4f1de7292b08d481a5be08442930d1017d97d3ef746f2551bd9c016cde9aa814",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/clean_and_jerk/trajectories/embodied-olympics-fun-clean-and-jerk-i00-privileged--dhfwnco.json

@@ -0,0 +1,239 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-clean-and-jerk-i00-privileged--dhfwnco",
+ "taskKey": "embodied-olympics-fun/clean_and_jerk",
+ "dag_sha256": "f530c1d45bf82223dabd69c9d30cf24e240db1d588782640954595693a7015ca",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "n_actions": 2737,
+   "sim_steps": 2737,
+   "horizon": 6000,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 4,
+   "stages_total": 4,
+   "fell": 0,
+   "replay_s": 46.44,
+   "bad_trajectory": 0,
+   "metric_bar_height_m": 2.133,
+   "metric_bar_max_height_m": 2.137,
+   "metric_elbow_flex_deg": 3.1,
+   "metric_knee_flex_deg": 0.6,
+   "metric_time_s": 27.37,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 176,
+   "agent_resets": 8,
+   "agent_replays": 2,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 2.68,
+   "peak_processes": 47,
+   "peak_disk_gb": 0.248,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-clean-and-jerk-i00-privileged--dhfwnco/replay.mp4",
+  "duration_s": 27.4,
+  "fps": 25.0,
+  "frames": 685,
+  "sha256": "3dfbf3b4ad659c116ddcd7317539c2fc8869c94d51251ed480a9404c890fc480"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "bar_at_shoulders": {
+    "method": "visual",
+    "intervals": [
+     [
+      17.0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 6.4,
+      "log_step": null,
+      "text_zh": "观察：视频 4.0–5.8 s 机器人弯腰以左右手分握杠杆，视频 6.4 s 杠铃两端已离开棕色支架，随后机器人持杠逐渐伸膝站起。"
+     },
+     {
+      "t": 17.0,
+      "log_step": null,
+      "text_zh": "观察：视频 16.0–16.8 s 机器人继续屈肘抬杠，视频 17.0–18.4 s 杠轴停在肩前，双手持续握杠，躯干直立、双膝近乎伸直，双脚并列支撑在木台上；之后才开始向头顶推举。"
+     },
+     {
+      "t": 17.0,
+      "log_step": 88,
+      "text_zh": "推断：结合视频 16.8–18.4 s 的肩前停留和 log step 88 对所交付轨迹的新进程回放结果（success=True，stages 含 clean、overhead、down signal），上肩事件约在视频 17.0 s 达成，时间误差约 0.5 s；肩距、速度等精确阈值由裁判通过结果佐证。该节点为事件，之后离肩上举不撤销完成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "抓握",
+      "note_zh": "视频 4.0–18.4 s，机器人左右灵巧手分别合拢握住杠杆两侧并共同提离支架、抬至肩前；视频 6.4 s 后杠铃随两手离架上升。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "视频 4.0–18.4 s，机器人双脚接触木色举重台，双腿和躯干在俯身抓杠、起身及肩前持杠时承重。"
+     }
+    ]
+   },
+   "bar_overhead": {
+    "method": "visual",
+    "intervals": [
+     [
+      27.36,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 22.0,
+      "log_step": null,
+      "text_zh": "观察：视频 18.6–22.0 s 机器人从肩前持续双手推举杠铃，杠铃到达头顶上方；视频 22.0–25.2 s 仍有小幅姿态调整，不能将刚过顶直接计作稳定保持完成。"
+     },
+     {
+      "t": 27.36,
+      "log_step": null,
+      "text_zh": "观察：视频 25.4–27.36 s 杠铃保持在机器人头顶上方，双手分别握住杠杆、双臂伸直，双腿近乎伸直且双脚并列留在木台上；已密集检查视频 26.4–27.36 s，包括最后一帧，未见脱手或下放。"
+     },
+     {
+      "t": 27.36,
+      "log_step": 88,
+      "text_zh": "推断：视频末尾的过顶保持姿态与评分器 reward 的 success=1、replay_success=1、stages_reached=4/4 一致；log step 88 的交付文件回放也记录 down signal 和成功。由于节点包含保持至下放信号，取视频最后一帧 27.36 s 为约完成时刻，精度约 0.1 s；不把更早仅举高的画面计作完成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "抓握",
+      "note_zh": "视频 18.6–27.36 s，机器人左右灵巧手持续各握杠杆一侧，双臂共同上举并在头顶承载杠铃，终帧仍持杠。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "视频 18.6–27.36 s，机器人双脚持续��撑于木色举重台，双腿与直立躯干承受上举及过顶保持时的重量。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "bar_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": 2.133
+  },
+  {
+   "name": "bar_max_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": 2.137
+  },
+  {
+   "name": "elbow_flex_deg",
+   "unit": "deg",
+   "threshold": 10.0,
+   "direction": "lower_better",
+   "value": 3.1
+  },
+  {
+   "name": "knee_flex_deg",
+   "unit": "deg",
+   "threshold": 10.0,
+   "direction": "lower_better",
+   "value": 0.6
+  },
+  {
+   "name": "time_s",
+   "unit": "s",
+   "threshold": 60.0,
+   "direction": "lower_better",
+   "value": 27.37
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人双手将杠铃提离支架，约在视频 17.0 s 完成肩前稳定上肩，随后推举至头顶并在视频末尾完成保持，两个目标节点最终均成立。评分器记录成功、四个原生阶段全部完成且回放确定，log step 88、94 的交付回放均成功；没有失败断点。缺少状态日志，视频不能独立量出肩距、速度、接触宽限和精确保持计数，上肩时间约有 0.5 s 不确定性，数值验收由裁判结果佐证。",
+ "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，两个目标节点（上肩约视频 17.0 s、过顶保持完成于视频末帧 27.36 s）与帧和评分器结果（stages 4/4、time_s 27.37）一致，接触标注合理，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": "2026-10-31T01:02:59Z",
+  "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/embodied-olympics-fun-clean-and-jerk-i00-privileged--dhfwnco/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-clean-and-jerk-i00-privileged--dhfwnco/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "bbd15726de429d008a5b2e711dd50f7dd1a65432454332bf4864d7dcff8f689f",
+   "video_sha256": "3dfbf3b4ad659c116ddcd7317539c2fc8869c94d51251ed480a9404c890fc480",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/clean_and_jerk/trajectories/embodied-olympics-fun-clean-and-jerk-i00-privileged--esqglcd.json

@@ -0,0 +1,234 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-clean-and-jerk-i00-privileged--esqglcd",
+ "taskKey": "embodied-olympics-fun/clean_and_jerk",
+ "dag_sha256": "f530c1d45bf82223dabd69c9d30cf24e240db1d588782640954595693a7015ca",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "n_actions": 1797,
+   "sim_steps": 1797,
+   "horizon": 6000,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 4,
+   "stages_total": 4,
+   "fell": 0,
+   "replay_s": 39.49,
+   "bad_trajectory": 0,
+   "metric_bar_height_m": 2.19,
+   "metric_bar_max_height_m": 2.19,
+   "metric_elbow_flex_deg": 0.2,
+   "metric_knee_flex_deg": 3.2,
+   "metric_time_s": 17.97,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 70,
+   "agent_resets": 0,
+   "agent_replays": 1,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.728,
+   "peak_processes": 10,
+   "peak_disk_gb": 0.252,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-clean-and-jerk-i00-privileged--esqglcd/replay.mp4",
+  "duration_s": 18.0,
+  "fps": 25.0,
+  "frames": 450,
+  "sha256": "d2c4bcced8f4c00c88a0e427949d5753238a66e03dcb0e0d77de6890872dfda5"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "bar_at_shoulders": {
+    "method": "visual",
+    "intervals": [
+     [
+      11.12,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 5.6,
+      "log_step": null,
+      "text_zh": "观察：视频 2.8–4.8 s 机器人俯身屈膝，左右手分别合握杠杆两侧；视频 5.6–6.0 s 杠铃已离开两只棕色支架，并随机器人起身上升。"
+     },
+     {
+      "t": 11.12,
+      "log_step": null,
+      "text_zh": "观察：视频 9.6–10.6 s 机器人把杠铃从腰前抬至胸肩前；视频 10.6–12.4 s 双手持续持杠，杠轴在肩前基本静止，躯干直立、双膝近伸直、双脚并齐留在木台上。视频 11.12 s 为上肩完成的近似锚点，精度约半秒。"
+     },
+     {
+      "t": 11.12,
+      "log_step": null,
+      "text_zh": "推断：结合上述肩前停留画面与 inputs/run.json 的 success=1、stages_reached=4、replay_success=1、deterministic=1，判定上肩通过；后续上举不撤销该事件。画面不能独立精测肩距、速度及连续保持计数，具体阈值由成功裁判结果佐证。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "抓握",
+      "note_zh": "观察：视频 2.8–6.0 s 左右灵巧手分别合握杠杆两侧并共同提杠离架，视频 10.6–12.4 s 两手持续持杠停在肩前。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0–12.4 s 机器人双脚支撑于木色举重台，屈膝俯身取杠后伸膝站起，持杠上肩时双脚保持并齐。"
+     }
+    ]
+   },
+   "bar_overhead": {
+    "method": "visual",
+    "intervals": [
+     [
+      17.96,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 16.0,
+      "log_step": null,
+      "text_zh": "观察：视频 12.6 s 后机器人由肩前继续上举；视频 14.6–16.0 s 杠铃逐渐升至头顶上方，双肘逐渐伸直。视频 16.0–17.96 s 两手分握杠杆，双臂近伸直、躯干直立、双膝近伸直，双脚并齐留在木台上，杠铃基本静止；已逐段检查视频最后一秒，末帧仍持杠过顶。"
+     },
+     {
+      "t": 17.96,
+      "log_step": null,
+      "text_zh": "推断：将完成保持的时刻记在视频末帧 17.96 s 附近，而不把首次举过头顶记作完成。inputs/run.json 的 success=1、replay_success=1、stages_reached=4/4、deterministic=1 佐证裁判已接受过顶保持及下放信号；终态指标为杠中心相对台面高度 2.19 m、双肘屈曲 0.2°、双膝屈曲 3.2°。无状态日志，保持计数与细小距离、速度阈值不能由画面精确复核。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "抓握",
+      "note_zh": "观察：视频 12.6–17.96 s 左右灵巧手持续分别握住杠杆两侧，双臂共同上举并在头顶上方保持，未见放手。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 12.6–17.96 s 双脚持续支撑在木台上，双腿与躯干保持直立，承载持杠过顶的身体。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "bar_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": 2.19
+  },
+  {
+   "name": "bar_max_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": 2.19
+  },
+  {
+   "name": "elbow_flex_deg",
+   "unit": "deg",
+   "threshold": 10.0,
+   "direction": "lower_better",
+   "value": 0.2
+  },
+  {
+   "name": "knee_flex_deg",
+   "unit": "deg",
+   "threshold": 10.0,
+   "direction": "lower_better",
+   "value": 3.2
+  },
+  {
+   "name": "time_s",
+   "unit": "s",
+   "threshold": 60.0,
+   "direction": "lower_better",
+   "value": 17.97
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人双手提杠离架，在视频约 11.12 s 完成肩前稳定上肩，随后上举并保持至视频末帧 17.96 s 附近，最终两节点均完成；评分器记录成功、四阶段全达成且回放一致，无失败断点。已对照参考成功视频的上肩与过顶姿态；本次无状态日志，节点时间为画面近似锚点，精确几何、速度及保持计数由成功评分佐证。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "抽查帧与标注一致：视频约10.6–12.4 s 杠铃在肩前稳定，14.6–17.9 s 过头顶保持；评分器成功、4/4阶段，无失败断点，接触标注合理。",
+    "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": "2026-10-31T01:00:17Z",
+  "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/embodied-olympics-fun-clean-and-jerk-i00-privileged--esqglcd/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-clean-and-jerk-i00-privileged--esqglcd/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "4bbb564bf9e09a167822b4680272d1bbe2ffb55de04e28c6204b38f72c06feab",
+   "video_sha256": "d2c4bcced8f4c00c88a0e427949d5753238a66e03dcb0e0d77de6890872dfda5",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/clean_and_jerk/trajectories/embodied-olympics-fun-clean-and-jerk-i00-privileged--hrohaub.json

@@ -0,0 +1,239 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-clean-and-jerk-i00-privileged--hrohaub",
+ "taskKey": "embodied-olympics-fun/clean_and_jerk",
+ "dag_sha256": "f530c1d45bf82223dabd69c9d30cf24e240db1d588782640954595693a7015ca",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "n_actions": 976,
+   "sim_steps": 976,
+   "horizon": 6000,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 4,
+   "stages_total": 4,
+   "fell": 0,
+   "replay_s": 19.14,
+   "bad_trajectory": 0,
+   "metric_bar_height_m": 2.122,
+   "metric_bar_max_height_m": 2.123,
+   "metric_elbow_flex_deg": 0.3,
+   "metric_knee_flex_deg": 3.1,
+   "metric_time_s": 9.76,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 87,
+   "agent_resets": 0,
+   "agent_replays": 18,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 5.594,
+   "peak_processes": 21,
+   "peak_disk_gb": 1.31,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-clean-and-jerk-i00-privileged--hrohaub/replay.mp4",
+  "duration_s": 9.8,
+  "fps": 25.0,
+  "frames": 245,
+  "sha256": "c38eeec8ae2b63cbf9ce24ee44cbba03495623481848077a145f4f488273a767"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "bar_at_shoulders": {
+    "method": "visual",
+    "intervals": [
+     [
+      6.3,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 4.32,
+      "log_step": null,
+      "text_zh": "观察：视频 3.60–4.20 s 机器人蹲下以两手持杠，视频 4.32–4.92 s 杠铃离开两只棕色支架，随双手升至腰前，原支架上已无杠铃。"
+     },
+     {
+      "t": 6.3,
+      "log_step": 223,
+      "text_zh": "观察：视频 5.76–6.36 s 杠铃横停在机器人胸肩前，两手分别持杆，躯干直立、双膝接近伸直，双脚留在木台上且未见前后跨步；随后视频 6.48–7.08 s 机器人继续上举。log step 223 的最终保存候选状态记录 clean 阶段在仿真 6.30 s 通过，肩距为 0.117 m。"
+     },
+     {
+      "t": 6.3,
+      "log_step": 230,
+      "text_zh": "推断：log step 223 的候选终态哈希与 log step 230 最终交付的两次新进程成功回放一致，976 个控制步共仿真 9.76 s，即每步 0.01 s；视频最后一帧为 9.76 s，视频约 4.08 s 的离架及约 6.3 s 的肩前停稳也与该记录吻合。因此以视频约 6.3 s 标记上肩验收，通过后按事件节点保留至片尾；没有使用 agent 会话时间换算。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "抓握",
+      "note_zh": "观察：视频 4.32–6.36 s 左右手分别握住银色杠杆两侧，共同将其离架抬到胸肩前，持杠随手移动。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 3.60–6.36 s 双脚踩在木色举重台上，双腿由蹲姿伸展并支撑持杠的身体。"
+     }
+    ]
+   },
+   "bar_overhead": {
+    "method": "visual",
+    "intervals": [
+     [
+      9.76,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 7.8,
+      "log_step": null,
+      "text_zh": "观察：视频 6.48–7.80 s 机器人将杠铃由肩前举过头顶，逐渐伸直双臂；视频 7.80–9.76 s 杠铃持续高于头部、双手分握杠杆，双腿近乎伸直、双脚并列支撑，躯干与杠铃未见明显晃动。已逐帧检查视频 8.80–9.76 s，片尾仍持杠过顶。"
+     },
+     {
+      "t": 9.76,
+      "log_step": 223,
+      "text_zh": "观察：log step 223 的最终保存候选记录仿真 6.99 s 进入 overhead 阶段、仿真 9.76 s 收到 down signal 并成功；log step 230 的最终交付两次独立回放均成功且终态哈希相同。评分器字段 success=1、replay_success=1、deterministic=1、stages_reached=4/4，末态杠高 2.122 m、肘屈曲 0.3 度、膝屈曲 3.1 度。"
+     },
+     {
+      "t": 9.76,
+      "log_step": 230,
+      "text_zh": "推断：依据已用离架和肩前停稳核对的控制步到视频时钟对应关系，976×0.01 s 对应视频 9.76 s；结合视频约 7.80 s 起的稳定过顶姿态和评分器通过，节点在视频 9.76 s 满足完整保持及下放信号验收。进入 overhead 阶段本身尚未完成保持，因此不从视频约 7.0 s 计为达成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "抓握",
+      "note_zh": "观察：视频 6.48–9.76 s 左右手分别握住杠杆两侧，双臂共同上举并持续持住头顶上方的杠铃。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 6.48–9.76 s 双脚持续踩在木色举重台上，双腿伸直支撑躯干和过顶杠铃。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "bar_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": 2.122
+  },
+  {
+   "name": "bar_max_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": 2.123
+  },
+  {
+   "name": "elbow_flex_deg",
+   "unit": "deg",
+   "threshold": 10.0,
+   "direction": "lower_better",
+   "value": 0.3
+  },
+  {
+   "name": "knee_flex_deg",
+   "unit": "deg",
+   "threshold": 10.0,
+   "direction": "lower_better",
+   "value": 3.1
+  },
+  {
+   "name": "time_s",
+   "unit": "s",
+   "threshold": 60.0,
+   "direction": "lower_better",
+   "value": 9.76
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人双手持杠离架，在视频约 6.3 s 完成肩前停稳，再举至头顶并保持，在视频 9.76 s 完成下放信号验收，两节点最终均达成。评分器记录成功且可确定性回放，log step 230 的最终交付两次新进程回放也均成功；日志中的早期调试失败不属于本次视频。缺少状态日志，细小的手指接触、距离和速度阈值不能仅凭远景逐项精测，精确验收由最终交付记录与评分器结果支持。",
+ "conf": 9,
+ "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，两节点时间与帧一致：约4.3 s离架、6.3 s肩前停稳、7.8 s后稳定过顶至片尾9.76 s；接触标注合理，无失败层。",
+    "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": "2026-10-31T01:02:11Z",
+  "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/embodied-olympics-fun-clean-and-jerk-i00-privileged--hrohaub/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-clean-and-jerk-i00-privileged--hrohaub/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "e16e701c4ebb10305ae5c15955763b62379023ca9ce7144c6bedd3fdf47f5e75",
+   "video_sha256": "c38eeec8ae2b63cbf9ce24ee44cbba03495623481848077a145f4f488273a767",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/clean_and_jerk/trajectories/embodied-olympics-fun-clean-and-jerk-i00-privileged--nb2dojv.json

@@ -0,0 +1,236 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-clean-and-jerk-i00-privileged--nb2dojv",
+ "taskKey": "embodied-olympics-fun/clean_and_jerk",
+ "dag_sha256": "f530c1d45bf82223dabd69c9d30cf24e240db1d588782640954595693a7015ca",
+ "outcome": {
+  "success": 0,
+  "reward": 0.25,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "n_actions": 1904,
+   "sim_steps": 1904,
+   "horizon": 6000,
+   "episode_over": 1,
+   "progress": 0.25,
+   "stages_reached": 1,
+   "stages_total": 4,
+   "fell": 0,
+   "replay_s": 31.87,
+   "bad_trajectory": 0,
+   "metric_bar_height_m": 1.339,
+   "metric_bar_max_height_m": 1.347,
+   "metric_elbow_flex_deg": 144.4,
+   "metric_knee_flex_deg": 0.1,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 154,
+   "agent_resets": 4,
+   "agent_replays": 2,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.874,
+   "peak_processes": 13,
+   "peak_disk_gb": 1.078,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-clean-and-jerk-i00-privileged--nb2dojv/replay.mp4",
+  "duration_s": 19.08,
+  "fps": 25.0,
+  "frames": 477,
+  "sha256": "fa7fb2abc0d5b24223e9cdd361b8068b3cb72118d00f6d1300fce140b211a35b"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "bar_at_shoulders": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 11.0,
+      "log_step": 301,
+      "text_zh": "观察：视频约 7.0–10.0 s，机器人蹲下将双手合在杠杆两处；视频 10.5–12.0 s，杠铃离开两只棕色支架并随机器人站起，原支架顶部露出，说明确已拿起，但此时杠仍在髋前，尚非上肩。"
+     },
+     {
+      "t": 18.8,
+      "log_step": 307,
+      "text_zh": "观察：视频 14.0–18.7 s，机器人屈肘提杠，杠仍低于肩关节；视频 18.8–19.04 s，杠铃明显倾斜，未见肩前双手稳定持杠姿态。交付文件的裁判回放（log step 307）仅记录 stages=[lift-off]，以 let go of the bar 结束；生成同一交付的记录（log step 301）报告末态仅右手持杠，未达到 clean。"
+     },
+     {
+      "t": 19.04,
+      "log_step": null,
+      "text_zh": "推断：成功参考视频 17.0、18.48 s 可清楚看见杠杆停在肩前，而本次已检查的全程视频及末帧均无此姿态；结合本次交付裁判的阶段记录，杠铃上肩从未完成。参考视频只用于识别姿态。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "抓握",
+      "note_zh": "观察：视频约 7.0–18.7 s，左右灵巧手合拢在杠杆的两处并共同提起、承载杠铃；末段双肘弯曲提杠时握持丢失。手指细节分辨率有限；log step 301 的交付裁判终止记录补充说明最终仅右手仍持杠。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0–19.04 s，机器人双脚在木色举重台上承重，双腿由蹲姿逐渐伸直，举杠及末段倾斜期间本体仍站立。"
+     }
+    ]
+   },
+   "bar_overhead": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 19.04,
+      "log_step": 307,
+      "text_zh": "观察：已检查视频 0–19.04 s，杠铃始终没有高于头部，末段双肘仍弯曲、杠杆在胸腹前倾斜，机器人没有开始双臂向头顶伸直的推举。交付裁判回放（log step 307）仅通过 lift-off，没有 clean、过顶或下放信号阶段；成功参考视频 26.0、27.36 s 的头顶伸臂持杠姿态在本次全程中未出现。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "bar_height_m",
+   "value": 1.339,
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better"
+  },
+  {
+   "name": "bar_max_height_m",
+   "value": 1.347,
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better"
+  },
+  {
+   "name": "elbow_flex_deg",
+   "value": 144.4,
+   "unit": "deg",
+   "threshold": 10.0,
+   "direction": "lower_better"
+  },
+  {
+   "name": "knee_flex_deg",
+   "value": 0.1,
+   "unit": "deg",
+   "threshold": 10.0,
+   "direction": "lower_better"
+  },
+  {
+   "name": "time_s",
+   "value": null,
+   "unit": "s",
+   "threshold": 60.0,
+   "direction": "lower_better"
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "bar_at_shoulders",
+   "t": 18.8,
+   "log_step": 301,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 18.6–19.04 s，机器人正在屈肘把杠提向肩前，杠铃随后明显倾斜，未形成上肩姿态，机器人仍站立。log step 301 生成的最终交付为 1904 个控制步，裁判记录 let go of the bar、holding=[right]；log step 307 对该交付的独立回放得到相同末态哈希、指标与失败原因。未完成且未修复的是杠铃上肩。"
+  },
+  "unmet": {
+   "bar_at_shoulders": false,
+   "bar_overhead": false
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没保持住",
+    "t": 18.8,
+    "log_step": 301,
+    "evidence_zh": "观察：视频 10.5–12.0 s 显示机器人双手带杠离架并站起，握持曾经建立；视频 18.76–19.04 s 杠轴由近水平变为明显倾斜。最终交付的裁判记录（log step 301）明确因 let go of the bar 判负，终止时 holding=[right]；log step 307 回放复现该结果。这是上肩过程中丢失左手持杠，未见机器人按计划放杠。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "source_zh": "日志",
+    "evidence_zh": "推断：log step 277 给出的生成脚本在提杠阶段持续命令双手闭合，并用双臂雅可比伺服跟踪手部目标；log step 301 仅更改上肩路径后生成最终交付，仍在提杠过程中丢失左手持杠，log step 307 确认可确定复现。结合视频 18.76–19.04 s 的杠铃倾斜，支持上肩运动中的持杠控制未能维持接触。agent 在 log step 315 将其归因于握持几何勉强、转肘时易滑脱，但该细分归因尚无交付回放的逐指接触力或跟踪误差记录独立确认，故保留为控制大类。"
+   }
+  ],
+  "gap_zh": "无状态日志，视频远景不能精确核验每根手指的接触、握持力及瞬时跟踪误差；左手丢失持杠由最终交付的裁判记录补充确认。现有证据能确定上肩过程的持杠控制失败，但不足以区分抓取接触设计与轨迹跟踪问题。原生 time_s 未在 reward 指标中报告，保留 null。"
+ },
+ "summary_zh": "观察：机器人双手提杠离架并站起，但杠铃尚未稳定上肩，视频约 18.8–19.04 s 杠杆倾斜，交付裁判记录左手脱杠并终止；上肩与过顶两个语义节点均未完成（0/2），原生阶段进度为 1/4。推断：断点在杠铃上肩的操作中，现象为没保持住，根因归于控制；缺少逐指接触力与跟踪误差，无法可靠细分。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "两个节点均未达成与视频及裁判记录一致（lift-off 1/4，杠未上肩，末段 19.04 s 倾斜、右手单持、let go of the bar）；断点 bar_at_shoulders/操作中、现象没保持住、根因停在控制均有服务读数（log step 301/307）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": "2026-10-31T00:59:04Z",
+  "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/embodied-olympics-fun-clean-and-jerk-i00-privileged--nb2dojv/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-clean-and-jerk-i00-privileged--nb2dojv/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "a312b5aad15d828476096bb22523a55415e7d5468cbd3fc4b8846ef68ea82524",
+   "video_sha256": "fa7fb2abc0d5b24223e9cdd361b8068b3cb72118d00f6d1300fce140b211a35b",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/clean_and_jerk/trajectories/embodied-olympics-fun-clean-and-jerk-i00-privileged--nsqtcqx.json

@@ -0,0 +1,230 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-clean-and-jerk-i00-privileged--nsqtcqx",
+ "taskKey": "embodied-olympics-fun/clean_and_jerk",
+ "dag_sha256": "f530c1d45bf82223dabd69c9d30cf24e240db1d588782640954595693a7015ca",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "n_actions": 1420,
+   "sim_steps": 1420,
+   "horizon": 6000,
+   "episode_over": 0,
+   "progress": 0.0,
+   "stages_reached": 0,
+   "stages_total": 4,
+   "fell": 0,
+   "replay_s": 14.81,
+   "bad_trajectory": 0,
+   "metric_bar_height_m": 0.375,
+   "metric_bar_max_height_m": 0.375,
+   "metric_elbow_flex_deg": 4.0,
+   "metric_knee_flex_deg": 106.4,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 249,
+   "agent_resets": 49,
+   "agent_replays": 5,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.725,
+   "peak_processes": 11,
+   "peak_disk_gb": 0.894,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-clean-and-jerk-i00-privileged--nsqtcqx/replay.mp4",
+  "duration_s": 14.24,
+  "fps": 25.0,
+  "frames": 356,
+  "sha256": "15b9c7107ed5a596e629cd4c1854cd1b1155c20d505ee9ea2b6318c8ce366ec6"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "bar_at_shoulders": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 4.5,
+      "log_step": null,
+      "text_zh": "观察：视频约 0–4.5 s，机器人由站立屈膝降低身体，双臂向支架上的银色杠杆下探；杠铃仍横跨两只棕色支架。"
+     },
+     {
+      "t": 14.2,
+      "log_step": 347,
+      "text_zh": "观察：视频 9.0–14.2 s 的密集帧中，机器人双手位于杠杆两侧握持位置，双脚支撑木台，始终深屈膝，杠铃套筒仍在支架顶面，没有离架或到达胸肩前。log step 347 对交付所用1420行前缀的执行反馈为双手 held、attempt=false；log step 348 保存该前缀，log step 360 的交付重放返回 stages=[]、最大杠高0.375米，与画面一致。"
+     },
+     {
+      "t": 14.2,
+      "log_step": null,
+      "text_zh": "推断：本次全段未出现成功参考视频 17.0 s 所示的直立持杠上肩姿态，且末帧仍在架上低位，故上肩事件从未成立。成功参考仅用于辨认目标姿态。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "抓握",
+      "note_zh": "观察：视频末段两只灵巧手分别包住杠杆的两处握持位置，并随小幅水平拉动保持在杆上；log step 347 对交付前缀的反馈也记录左右手均持杠。杠铃仍由支架承托，未抬起。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：机器人下蹲接近和双手握杠时，双脚持续踩在木色举重台上，双腿与躯干维持屈膝支撑。"
+     }
+    ]
+   },
+   "bar_overhead": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 14.2,
+      "log_step": 360,
+      "text_zh": "观察：视频全段及视频 13.2–14.2 s 的末秒密集帧中，杠铃始终在机器人膝部附近的支架高度；双手没有携杠向头顶运动。成功参考视频 26.0 s 和 27.36 s 可清楚看到头顶上方持杠、双臂伸直的姿态，本次没有出现；log step 360 的交付重放也返回最大杠高0.375米、stages=[]。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "bar_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": 0.375
+  },
+  {
+   "name": "bar_max_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": 0.375
+  },
+  {
+   "name": "elbow_flex_deg",
+   "unit": "deg",
+   "threshold": 10.0,
+   "direction": "lower_better",
+   "value": 4.0
+  },
+  {
+   "name": "knee_flex_deg",
+   "unit": "deg",
+   "threshold": 10.0,
+   "direction": "lower_better",
+   "value": 106.4
+  },
+  {
+   "name": "time_s",
+   "unit": "s",
+   "threshold": 60.0,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "bar_at_shoulders",
+   "t": null,
+   "log_step": 348,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 0–14.2 s，机器人实际执行了下蹲、伸手握杠及末段小幅水平拉动，交付结束时仍屈膝持着架上的杠铃，上肩未完成。log step 347–348 明确截取并保存1420行前缀；log step 341 的后续伸腿试举不在该前缀内，log step 359 与最终报告 log step 366 均说明保留未离架的稳定检查点。该节点已经开始操作，但交付中没有可定位的突发失效时刻，因此视频断点时刻留空。"
+  },
+  "phenomena": [],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "t": null,
+    "log_step": 356,
+    "source_zh": "日志",
+    "evidence_zh": "推断：未能形成可在双手持杠条件下离架的控制动作，使交付停在抓持检查点。log step 356 从同一1420行前缀继续向上运动的会话试验中，模拟器反馈 held 从双手变为右手再到空，杠中心高度仍为0.475米；log step 362 再从最终交付文件执行站起，反馈在追加动作第5步已无手持杠、第10步双手接触均失去，杠仍留在原高度。agent 在 log step 359 明确据试举结果保留稳定前缀。上述仅作为会话层的控制能力证据，不作为本次评分视频的滑脱现象；不足以区分抓取几何、逆解与跟踪问题。"
+   }
+  ],
+  "gap_zh": "观察：inputs/ 中没有状态日志；视频能确认杠铃未离架及双手在杆上的姿态，但像素不足以分辨逐指接触和精确接触力。交付在稳定抓持与水平拉动后结束，未包含会话日志中继续向上运动后滑脱的片段，因而现象层留空：不能把会话试验的“没保持住”移入评分轨迹，也不能仅因未上肩而虚构位姿控制失准。根因只依据日志中的重复试举反馈归为控制，具体子类仍不确定。",
+  "unmet": {
+   "bar_at_shoulders": false,
+   "bar_overhead": false
+  }
+ },
+ "summary_zh": "机器人完成下蹲接近和双手握杠，但交付视频截至 14.2 s 杠铃仍在支架上，上肩与过顶均未达到，最终为0/2节点。断点是上肩操作；会话中从该检查点继续试举反复失去抓持，支持控制方面的困难，而最终保存的是试举前的稳定前缀。缺少状态日志及可分辨逐指受力的画面，评分轨迹也未包含后续滑脱，故失败现象留空、根因不细分。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "两节点均未达成与视频、grader 字段（stages=0、杠高0.375 m）一致；断点在上肩、现象留空（交付不含上举动作，符合 FAIL-23）、根因停在“控制”均有日志支撑，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": "2026-10-31T01:00:03Z",
+  "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/embodied-olympics-fun-clean-and-jerk-i00-privileged--nsqtcqx/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-clean-and-jerk-i00-privileged--nsqtcqx/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "fbc574f5137a6c76877f8e34c24245c3a2df33ce48428c2b16e509d96c2ab91f",
+   "video_sha256": "15b9c7107ed5a596e629cd4c1854cd1b1155c20d505ee9ea2b6318c8ce366ec6",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/clean_and_jerk/trajectories/embodied-olympics-fun-clean-and-jerk-i00-standard--8hf7zba.json

@@ -0,0 +1,212 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-clean-and-jerk-i00-standard--8hf7zba",
+ "taskKey": "embodied-olympics-fun/clean_and_jerk",
+ "dag_sha256": "f530c1d45bf82223dabd69c9d30cf24e240db1d588782640954595693a7015ca",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "calls": 1,
+   "rows": 0,
+   "resets": 0,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 2209.7,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 4,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 1.036,
+   "peak_processes": 7,
+   "peak_disk_gb": 0.3,
+   "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/embodied-olympics-fun-clean-and-jerk-i00-standard--8hf7zba/replay.mp4",
+  "duration_s": 0.04,
+  "fps": 25.0,
+  "frames": 1,
+  "sha256": "8e8ef36d37cebc8157c6ec733ba26e6e99595c291f0274cd3f0d3f364cf24532"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "bar_at_shoulders": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": 8,
+      "text_zh": "观察：视频 0.00 s 的唯一帧中，机器人站在木台上，双手位于杠铃上方，杠铃仍横放在两只棕色支架上，未离架或到达胸肩前方；这也是视频末帧。log step 8 的现场服务初始观测为 control_steps=0、done=False，评分器最终记录 n_actions=0、sim_steps=0、rows=0，与初态画面一致，杠铃上肩未达成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 n_actions=0、sim_steps=0、rows=0；log step 89 记录提供方过滤拦截，现场没有交付执行轨迹，本节点的轨迹接触不适用。"
+     }
+    ]
+   },
+   "bar_overhead": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": 8,
+      "text_zh": "观察：视频 0.00 s 的唯一帧（亦为末帧）中，杠铃仍在两只支架上，明显低于机器人头部，双臂没有持杠伸直过顶。评分器最终记录 n_actions=0、sim_steps=0，未出现任何现场上举运动，杠铃过顶未达成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 n_actions=0、sim_steps=0、rows=0；log step 89 记录提供方过滤拦截，现场没有交付执行轨迹，本节点的轨迹接触不适用。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "bar_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "bar_max_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "elbow_flex_deg",
+   "unit": "deg",
+   "threshold": 10.0,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "knee_flex_deg",
+   "unit": "deg",
+   "threshold": 10.0,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "time_s",
+   "unit": "s",
+   "threshold": 60.0,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 89,
+   "where": "首步前",
+   "evidence_zh": "观察：log step 8 只调用现场机器人服务的 get_obs(seconds=0.0)，返回 control_steps=0、done=False；评分器最终记录 n_actions=0、sim_steps=0、rows=0。log step 86 启动的是 agent 自建 LocalSim 的本地测试，log step 87 仍称等待该测试，log step 89 随后记录“blocked by Gemini's filters”；视频 0.00 s 的唯一帧为未抓杠的初态。推断：提供方过滤在任何现场举重动作或有效交付前中断了会话，无法定位某个物理步骤的失败时刻。"
+  },
+  "unmet": {
+   "bar_at_shoulders": false,
+   "bar_overhead": false
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：log step 89 记录提供方过滤拦截，之前 log step 86–87 仍在启动、等待本地 LocalSim 测试；评分器最终 n_actions=0、sim_steps=0、rows=0，视频仅有视频 0.00 s 的一帧初态，inputs/ 中没有状态日志���推断：这是现场动作开始前的外因中断，应留待重测并排除出机器人失败归因；保留目录评分 success=0，但不标“没尝试”或任何机器人根因。本地测试不属于评分回合，也无法提供该回合的接触或控制失败证据。"
+ },
+ "summary_zh": "观察：视频 0.00 s 的唯一帧中，机器人尚未抓杠，杠铃仍置于两只支架上；评分器记录 n_actions=0、sim_steps=0，因此上肩和过顶均未达成，最终为0/2。观察：log step 86–87 仍在启动、等待 agent 自建 LocalSim 测试，log step 89 随后记录提供方过滤拦截。推断：会话在现场执行前因外因中断，保留目录 success=0，断点记 unknown／首步前，现象和根因留空；单帧视频与无状态日志不能支持机器人能力归因。",
+ "conf": 9,
+ "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：评分器 n_actions=0、sim_steps=0，现场未执行任何动作，两节点未达成；log step 89 为提供方过滤拦截（外因），无交付，故断点 unknown、现象/根因留空并有 gap_zh，处理正确。",
+    "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": "2026-10-31T01:05:51Z",
+  "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/embodied-olympics-fun-clean-and-jerk-i00-standard--8hf7zba/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-clean-and-jerk-i00-standard--8hf7zba/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "4084ffcdb750f342162109fc81fde170515e83a73c49bc5a7d9c3414e1eea403",
+   "video_sha256": "8e8ef36d37cebc8157c6ec733ba26e6e99595c291f0274cd3f0d3f364cf24532",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/clean_and_jerk/trajectories/embodied-olympics-fun-clean-and-jerk-i00-standard--husetxq.json

@@ -0,0 +1,265 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-clean-and-jerk-i00-standard--husetxq",
+ "taskKey": "embodied-olympics-fun/clean_and_jerk",
+ "dag_sha256": "f530c1d45bf82223dabd69c9d30cf24e240db1d588782640954595693a7015ca",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 938,
+   "sim_steps": 1648,
+   "calls": 24,
+   "rows": 1648,
+   "resets": 3,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 6580.3,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 3,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 1.333,
+   "peak_processes": 10,
+   "peak_disk_gb": 0.476,
+   "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/embodied-olympics-fun-clean-and-jerk-i00-standard--husetxq/replay.mp4",
+  "duration_s": 9.4,
+  "fps": 25.0,
+  "frames": 235,
+  "sha256": "6fbf5db40461e23d74c0eb23a864991b076fc724b38b06e3c27276721695d4b2"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "bar_at_shoulders": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 3.8,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–3.8 s，机器人由站立转为屈膝俯身，双手向低位杠杆靠近；视频 3.8–5.8 s 双手仍在杠杆上方，杠铃搁在两只棕色支架上，未出现持杠起身或胸肩前停杠。"
+     },
+     {
+      "t": 6.16,
+      "log_step": null,
+      "text_zh": "观察：视频 6.04–6.16 s，银色杠铃从支架前缘落到木台，双手没有随杠下落或承载杠铃；视频 6.2–9.36 s 杠铃留在台面，机器人继续俯身并最终前扑，始终没有上肩。"
+     },
+     {
+      "t": null,
+      "log_step": 129,
+      "text_zh": "观察：log step 128 启动重置后的最后一次现场回合；log step 129 的该回合输出中 descend、gripclose 及 GRASP 的 holding_left/right 均为 0.0，服务最终裁判结果为非脚部触台，stages 0/4。这与视频无持杠上肩的过程一致；此处不把日志中的仿真时刻换算为视频时刻。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.0–8.5 s，机器人双脚接触木色举重台，双腿和躯干在下蹲、俯身伸手过程中承担本体支撑；视频 8.6–9.36 s 支撑失稳并前扑。"
+     },
+     {
+      "who": "双臂协作",
+      "how": "未知",
+      "note_zh": "观察：视频 3.8–8.5 s，机器人两手同时下探接近杠杆，但未见手指包围并夹持杠杆。远景不能排除短暂擦碰，也不能确定杠铃滑离支架前是否被手或腿碰动，故不记为抓握。"
+     }
+    ]
+   },
+   "bar_overhead": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 9.36,
+      "log_step": null,
+      "text_zh": "观察：视频全段 0.0–9.36 s 未出现举杠过顶，杠铃先在支架低位，视频约 6.16 s 后在木台上；末帧机器人向前翻倒，双臂没有持杠伸直。成功参考视频的肩前停杠（参考视频 16.0–18.0 s）和过顶保持（参考视频 24.0–27.2 s）在本次视频均未出现。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "bar_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "bar_max_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "elbow_flex_deg",
+   "unit": "deg",
+   "threshold": 10.0,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "knee_flex_deg",
+   "unit": "deg",
+   "threshold": 10.0,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "time_s",
+   "unit": "s",
+   "threshold": 60.0,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "bar_at_shoulders",
+   "t": 6.1,
+   "log_step": 129,
+   "where": "操作中",
+   "evidence_zh": "观察：机器人正在俯身接近杠铃以完成上肩，但视频 6.04–6.16 s 杠铃滑落到木台，双手未形成承载握持；此后一直没有取回杠铃，视频 8.6–9.36 s 又失衡前扑。log step 129 的最后现场回合显示抓取后 holding_left/right 均为 0.0，并以非脚部触台判负。首次未修复的步骤为杠铃上肩。"
+  },
+  "unmet": {
+   "bar_at_shoulders": false,
+   "bar_overhead": false
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没建立接触",
+    "t": 6.1,
+    "log_step": 129,
+    "evidence_zh": "观察：视频 3.8–6.16 s，机器人双手接近杠杆却未将其握起，杠铃随后落到台面；视频 6.2–9.36 s 未见杠铃被任何一手承载。log step 129 的最后回合抓取输出 holding_left/right 均为 0.0。所需双手握持没有建立，不能将杠铃落台解释为已抓住后滑脱。"
+   },
+   {
+    "cat": "全身",
+    "item": "平衡",
+    "t": 8.6,
+    "log_step": 129,
+    "evidence_zh": "观察：视频 8.56–9.36 s，机器人躯干持续向前下方旋转，双臂向侧后方摆动，身体由双脚支撑的俯身姿态翻成前扑姿态；视频 9.28–9.36 s 头和上身降到台面附近。log step 129 中服务裁判报告非脚部触台并结束最后回合。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "evidence_zh": "推断：log step 119 的 agent 控制脚本以本地模型进行逆运动学和编码器前向运动学修正，log step 128 调整握距和悬停高度后在最后回合执行；但视频 3.8–6.16 s 双手未建立杠铃承载，杠铃落台后也未被取回，视频 8.6–9.36 s 支撑姿态进一步失稳。结合 log step 129 的 holding 均为 0.0，可归于抓取及全身运动控制未实现预定动作；现有证据不足以区分具体接触几何、逆解或反馈控制子类。",
+    "source_zh": "视频",
+    "log_step": 129
+   }
+  ],
+  "gap_zh": "本次无状态日志，只有单一远景，无法确定杠铃滑离支架前的具体触碰部位，也不能逐指核验短暂擦碰或触台的首个机器人连杆。log step 119 的脚本中段在输入记录中被截断；log step 132 关于抓取几何及本地模型误差的说法仅为 agent 自述，未用来细分根因。日志有三次重置，已只用最后回合核对本次视频，未将前两回合的脚本异常或碰支架记成本次失败现象。"
+ },
+ "summary_zh": "本次评分回合完成了俯身接近杠铃，但未建立双手持杠；杠铃在视频约 6.1 s 滑落木台，既未上肩也未过顶，最终完成 0/2 个节点。机器人随后在视频约 8.6–9.36 s 失衡前扑，服务以非脚部触台判负，断点为杠铃上肩，根因仅能定位到控制层。单一远景及缺少状态日志使具体触杠部位和控制子类无法确定。",
+ "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（两个均未达成、0/2）、断点 bar_at_shoulders 与首个现象「没建立接触」与视频一致；问题在于第二个现象「平衡」：视频 8.6–9.36 s 的前扑是评分器因非脚部触台终止回合前的跌倒，属于 S14 规定不单独标注的终止事件，且发生在断点之后。",
+    "n_issues": 1,
+    "n_major": 1
+   },
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 2,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "节点均未达成（0/2）、断点 bar_at_shoulders、现象（没建立接触、平衡）与视频一致；第一轮关于删除平衡的意见已被带证据驳回（03 文档将可见跌倒记为全身·平衡），无新证据，不再重复。",
+    "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": "2026-10-31T01:14:08Z",
+  "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/embodied-olympics-fun-clean-and-jerk-i00-standard--husetxq/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-clean-and-jerk-i00-standard--husetxq/04_review_r1"
+   },
+   {
+    "stage": "reconcile",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-clean-and-jerk-i00-standard--husetxq/05_reconcile_r1"
+   },
+   {
+    "stage": "traj_review",
+    "round": 2,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-clean-and-jerk-i00-standard--husetxq/04_review_r2"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "f27205ed94a2b5f6eaff0d8351ae672f4102ae2432a9426c20958f31b507f6af",
+   "video_sha256": "6fbf5db40461e23d74c0eb23a864991b076fc724b38b06e3c27276721695d4b2",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/clean_and_jerk/trajectories/embodied-olympics-fun-clean-and-jerk-i00-standard--jxozvju.json

@@ -0,0 +1,254 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-clean-and-jerk-i00-standard--jxozvju",
+ "taskKey": "embodied-olympics-fun/clean_and_jerk",
+ "dag_sha256": "f530c1d45bf82223dabd69c9d30cf24e240db1d588782640954595693a7015ca",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 6000,
+   "sim_steps": 12893,
+   "calls": 311,
+   "rows": 11958,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 2028.1,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 2.878,
+   "peak_processes": 23,
+   "peak_disk_gb": 0.47,
+   "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/embodied-olympics-fun-clean-and-jerk-i00-standard--jxozvju/replay.mp4",
+  "duration_s": 60.04,
+  "fps": 25.0,
+  "frames": 1501,
+  "sha256": "0558978f577a17ac23535ece1aed63278433ed055e8db4e7c70c810640642e66"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "bar_at_shoulders": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 18.6,
+      "log_step": 91,
+      "text_zh": "观察：视频 4–18.6 s，机器人屈膝俯身，两手在仍搁于棕色支架上的杠杆两侧反复闭合和调整；log step 91 的本次现场回合读数在仿真 18.7 s 报 holding_left/right=1/1，但这只表明双手接触握持，画面没有杠铃离架或上肩。"
+     },
+     {
+      "t": 20.2,
+      "log_step": 92,
+      "text_zh": "观察：视频 19.8–22.2 s，机器人开始起身，两手逐渐离开杠杆，杠铃仍在两只支架上；log step 92 打印本回合服务本体读数，仿真 20.1 s 的双手 holding=1/1，在仿真 20.2 s 变为0/0。视频 24–30 s 机器人空手抬臂到胸肩附近，杠铃未随动。"
+     },
+     {
+      "t": 60.0,
+      "log_step": 97,
+      "text_zh": "观察：视频 42–60 s 机器人重新俯身抓杠，视频 59.0、59.4、59.8、60.0 s 杠铃仍在支架上；log step 93、96 的本回合服务读数反复为左手 holding=1、右手=0，log step 97 返回 episode=6、control_steps=6000、outcome=time、stages_reached=0/4。上肩从未达成。"
+     },
+     {
+      "t": 39.0,
+      "log_step": 91,
+      "text_zh": "推断：采用最后一次现场回合 episode=6 对应本视频；log step 5 给出 control_dt=0.01 s，log step 91 给出 control_steps=3920、sim_time_s=39.2，log step 97 为6000步，分别对应视频约 39.2 s 和60.0 s。已在视频 39.0 s 空手过顶保持、视频 40–43 s 重新下蹲的转折处核对，因此 log step 92 的仿真 20.2 s 可锚定视频约 20.2 s；没有使用 agent 工作时间换算。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "抓握",
+      "note_zh": "机器人左右灵巧手分别在杠杆两侧闭合抓取；视频 18.6–20.1 s 双手位于杆上，本回合服务 holding 读数同时为1，随后抬手丢失接触，未形成成功离架抬升。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "机器人双脚在木台上承重，双腿与躯干配合俯身抓杠、起身和重新下蹲。"
+     }
+    ]
+   },
+   "bar_overhead": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 34.0,
+      "log_step": 91,
+      "text_zh": "观察：视频 31–39 s 机器人将空手举向头顶并保持，银色杠铃清楚可见仍横在低处两只支架上；log step 91 的 press、hold 段服务读数均为 holding=0/0。动作脚本执行了上举段，但没有将杠铃举过顶。"
+     },
+     {
+      "t": 60.0,
+      "log_step": 97,
+      "text_zh": "观察：视频 59.0–60.0 s，机器人俯身在低位杠铃旁，杠铃既未上肩也未过顶；log step 97 服务裁判返回本回合失败。成功参考视频可清楚看到杠铃随手举高，而本次全程没有该物体位移。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "视频 31–39 s 机器人执行空手上举段时双脚仍在木台上支撑本体；双手没有接触杠铃，不记持杠接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "bar_height_m",
+   "value": null,
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better"
+  },
+  {
+   "name": "bar_max_height_m",
+   "value": null,
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better"
+  },
+  {
+   "name": "elbow_flex_deg",
+   "value": null,
+   "unit": "deg",
+   "threshold": 10.0,
+   "direction": "lower_better"
+  },
+  {
+   "name": "knee_flex_deg",
+   "value": null,
+   "unit": "deg",
+   "threshold": 10.0,
+   "direction": "lower_better"
+  },
+  {
+   "name": "time_s",
+   "value": null,
+   "unit": "s",
+   "threshold": 60.0,
+   "direction": "lower_better"
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "bar_at_shoulders",
+   "t": 20.2,
+   "log_step": 92,
+   "where": "操作中",
+   "evidence_zh": "观察：视频约 20.2 s 机器人在上肩步骤的离架抬升阶段丢失双手接触，视频 20.8–24 s 可见手臂向上而杠铃留在支架上；log step 92 的本回合服务 holding 读数由仿真 20.1 s 的1/1变为仿真 20.2 s 的0/0。视频 42–60 s 重抓仍未抬起杠铃，故该步骤一直未修复。"
+  },
+  "unmet": {
+   "bar_at_shoulders": false,
+   "bar_overhead": false
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没保持住",
+    "t": 20.2,
+    "log_step": 92,
+    "evidence_zh": "观察：视频 18.6–20.0 s 双手合在杠杆上，随后机器人起身时双手脱离，视频 21.2–24 s 两手已在杠杆上方而杠铃仍由支架承托；log step 92 的服务读数在仿真 20.1 s 双手 holding=1/1、仿真 20.2 s 双手=0/0。这里丢失的是已建立的握持接触，并非举起后的杠铃坠落。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "t": 20.2,
+    "log_step": 91,
+    "source_zh": "日志 / 视频",
+    "evidence_zh": "推断：log step 88、91 中 agent 的控制脚本在双手 holding 首次为1后切换到预计算抬升轨迹；log step 92 的实际服务读数显示抬升刚开始就双手失去握持，左手食指和中指已闭至1.7 rad，仍未带起杠铃。脚本没有以持续握持作为后续动作的门控，log step 91 在 holding=0/0 时继续执行 curl+rack、press、hold，视频 24–39 s 对应为空手抬臂。因此抓持和抬升的控制未维持所需接触；现有证据不足以进一步确定接触几何、握力或轨迹跟踪哪一项主导。"
+   }
+  ],
+  "gap_zh": "没有状态日志；场馆远景能分辨杠铃留在支架及双手离开，却不能精确还原手指包络、接触力和滑动路径。log step 92 保留了本回合服务 holding 及部分手指角度读数，可支持接触丢失，但不足以区分具体控制子类；原生指标未在评分器 reward 字段报告，均留 null。此前回合的跌倒及局部仿真结果只作会话背景，不计入本次物理进度。"
+ },
+ "summary_zh": "评分采用第六次现场回合：机器人完成俯身抓杠，但约视频 20.2 s 起身时丢失双手握持，杠铃始终留在支架，上肩与过顶均未达成，最终为0/2。随后脚本空手执行上肩和上举动作，末段重抓也未恢复，服务裁判在视频约 60 s 回合时限处判负。根因证据支持抓持及抬升控制失败；缺少接触力和近距离手指状态，不能确定更细的控制机制。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "两个目标节点均未达成（0/2）、断点在上肩步骤离架抬升处、时钟映射（6000 控制步=视频 60 s，log step 91 的 39.2 s 与视频空手过顶对应）均与视频和日志一致；现象“没保持住”与根因“控制”有证据且已声明不确定性，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": "2026-10-31T00:58:04Z",
+  "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/embodied-olympics-fun-clean-and-jerk-i00-standard--jxozvju/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-clean-and-jerk-i00-standard--jxozvju/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "580c29fbecc04a871fa1d7f7e7db42ccf86bc81bec35068f22e9d695e0387d83",
+   "video_sha256": "0558978f577a17ac23535ece1aed63278433ed055e8db4e7c70c810640642e66",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/clean_and_jerk/trajectories/embodied-olympics-fun-clean-and-jerk-i00-standard--rzyz9ev.json

@@ -0,0 +1,237 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-clean-and-jerk-i00-standard--rzyz9ev",
+ "taskKey": "embodied-olympics-fun/clean_and_jerk",
+ "dag_sha256": "f530c1d45bf82223dabd69c9d30cf24e240db1d588782640954595693a7015ca",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "live_success": 1,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 3679,
+   "sim_steps": 4279,
+   "calls": 39,
+   "rows": 4279,
+   "resets": 1,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 475.7,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 3,
+   "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/embodied-olympics-fun-clean-and-jerk-i00-standard--rzyz9ev/replay.mp4",
+  "duration_s": 36.8,
+  "fps": 25.0,
+  "frames": 920,
+  "sha256": "9c5a696f1f57d6912e9e953a9abedad168de16c8bedee29395dea26a6855ee71"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "bar_at_shoulders": {
+    "method": "visual",
+    "intervals": [
+     [
+      23.5,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 8.4,
+      "log_step": 32,
+      "text_zh": "观察：视频 5.6–6.8 s 机器人俯身将左右手分别合在杠杆两处，视频 8.4–9.6 s 银色杠铃随双手升起、两侧套筒离开棕色支架，支架留在原处；机器人双脚持续踩在木台上。"
+     },
+     {
+      "t": 23.5,
+      "log_step": 45,
+      "text_zh": "观察：视频 22.0–23.0 s 机器人把杠铃从胸前抬至肩前，视频 23.0–24.4 s 杠铃停在肩前，双手持杠，躯干直立、双腿近伸直且双脚并齐；与参考视频 17.0–18.0 s 的上肩姿态相符。log step 45 的机器人服务返回第 2 回合 control_steps=2300 和2450，对应肩前停顿。"
+     },
+     {
+      "t": 23.5,
+      "log_step": 56,
+      "text_zh": "推断：按视频 23.0–24.4 s 的肩前稳定停顿，将上肩事件完成锚点取为视频约 23.5 s，误差约0.5秒；缺少逐控制步阶段计数，精确首次通过时刻不可见。log step 56 的服务裁判报告 stages_reached=4/4、outcome=success，支持该回合已通过上肩及其数值验收；后续举杠离肩不撤销本事件。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "抓握",
+      "note_zh": "视频 6.4–24.4 s，机器人左右灵巧手分别合握杠杆的左右两处，双臂共同将杠铃离架、提起并转至肩前。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "视频 0–24.4 s，机器人双脚支承于木色举重台，双腿和躯干在俯身取杠、起身及上肩停顿时承重。"
+     }
+    ]
+   },
+   "bar_overhead": {
+    "method": "visual",
+    "intervals": [
+     [
+      36.79,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 35.0,
+      "log_step": 56,
+      "text_zh": "观察：视频 25.0–35.0 s 机器人双手持续持杠，从肩前缓慢上举并伸肘；视频 35.0–36.76 s 银色杠铃保持在头顶上方，双臂、双腿近伸直，双脚仍并齐落在木台上，未见脱手或回落，与参考视频 25.0–27.36 s 的终态相符。已逐0.2秒检查视频 35.8–36.6 s，并检查最后一帧视频 36.76 s。"
+     },
+     {
+      "t": 36.76,
+      "log_step": 56,
+      "text_zh": "观察：log step 56 中，机器人服务先返回第 2 回合 control_steps=3550、done=False，随后返回 control_steps=3679、sim_time_s=36.79、done=True、reason=success、stages_reached=4、stages_total=4；inputs/run.json 的评分器字段 live_success=1、graded_by_live=1，表明成功验收属于此现场回合。"
+     },
+     {
+      "t": 36.79,
+      "log_step": 56,
+      "text_zh": "推断：log step 7 的服务规格 control_dt=0.01秒，log step 25–26 重置后开始第 2 回合；以该回合 control_steps×0.01 映射视频时间，已用 log step 45 的2300–2450步对应视频 23.0–24.4 s 肩前停顿核对，3679步对应视频 36.79 s，且与视频总长36.8秒及末帧姿态吻合。节点包含稳定保持至下放信号，故在裁判成功时视频 36.79 s 记为达成；无逐帧速度及接触计数，不能仅从较早的举高画面确认全部验收。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "抓握",
+      "note_zh": "视频 25.0–36.76 s，机器人左右手分别环握同一杠杆的左右两处，双臂持续承载、上举并维持头顶持杠。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "视频 25.0–36.76 s，机器人双脚在木色台面上并齐支撑，双腿与躯干在上举和过顶保持期间承重。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "bar_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "bar_max_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "elbow_flex_deg",
+   "unit": "deg",
+   "threshold": 10.0,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "knee_flex_deg",
+   "unit": "deg",
+   "threshold": 10.0,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "time_s",
+   "unit": "s",
+   "threshold": 60.0,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "本次评分视频对应重置后的第2回合：机器人双手握杠离架，约视频 23.5 s 完成肩前稳定上肩，随后上举并保持，服务裁判在映射后的视频 36.79 s 报告成功及全部4个原生阶段完成，两个DAG节点最终均成立。log step 25–26 之前的第1回合已被重置，不属于本次评分轨迹，成功回合无失败断点。缺少状态日志，上肩首次通过时间约有0.5秒误差，精细距离、速度和接触宽限依据现场裁判成功结论确认；评分器reward未返回DAG列出的原生指标数值，故均留空。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "成功回合，两个节点均达成，时间与接触标注与视频和服务裁判读数一致，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": "2026-10-31T00:59:55Z",
+  "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/embodied-olympics-fun-clean-and-jerk-i00-standard--rzyz9ev/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-clean-and-jerk-i00-standard--rzyz9ev/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "d5a20a0ae00aae0aa7093da0a97e498974a3f054d91814faad2992711d403320",
+   "video_sha256": "9c5a696f1f57d6912e9e953a9abedad168de16c8bedee29395dea26a6855ee71",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/clean_and_jerk/trajectories/embodied-olympics-fun-clean-and-jerk-i00-standard--vk7mymb.json

@@ -0,0 +1,230 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-clean-and-jerk-i00-standard--vk7mymb",
+ "taskKey": "embodied-olympics-fun/clean_and_jerk",
+ "dag_sha256": "f530c1d45bf82223dabd69c9d30cf24e240db1d588782640954595693a7015ca",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 412,
+   "sim_steps": 2402,
+   "calls": 27,
+   "rows": 1202,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 1194.5,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.293,
+   "peak_processes": 6,
+   "peak_disk_gb": 0.289,
+   "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/embodied-olympics-fun-clean-and-jerk-i00-standard--vk7mymb/replay.mp4",
+  "duration_s": 4.16,
+  "fps": 25.0,
+  "frames": 104,
+  "sha256": "7bb3669d84346a3d68afe8625dc4ba89f23937f15282f118f396eebb1d51168e"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "bar_at_shoulders": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.4,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–2.4 s，机器人从站立逐渐屈膝俯身，双手朝低位杠杆伸去；银色杠铃仍横跨两只棕色支架。"
+     },
+     {
+      "t": 3.48,
+      "log_step": 262,
+      "text_zh": "观察：视频 2.4–4.12 s，杠铃始终停在支架高度，没有离架并到达胸肩前方；机器人随后倾倒。log step 262 明确重置到第6回合后提交最终动作，服务返回双手 holding 均为0、原生阶段完成数0/4及非脚部触台判负；这里的0/4是服务阶段数，不是本DAG的节点数。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.0–3.2 s，机器人双脚接触木色举重台，支撑屈膝和俯身伸手动作；视频 3.3–4.12 s 支撑失稳，最终发生非脚部触台（log step 262–263 的服务判负）。"
+     },
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：视频 2.4–3.6 s 双手接近杠杆，手指区域像素不足，不能确认是否发生短暂接触或其方式；log step 262 的终态双手 holding 均为0，不能据此证明全程从未接触，也不记为已建立抓握。"
+     }
+    ]
+   },
+   "bar_overhead": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 4.12,
+      "log_step": 262,
+      "text_zh": "观察：视频 0.0–4.12 s，杠铃一直在两支架之间的低位，机器人没有持杠上举，更没有杠铃过顶、双臂伸直的姿态；终帧机器人已倒下。log step 262 显示最终回合在第一段动作中判负，未继续提交包含上肩与过顶动作的第二段。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "bar_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "bar_max_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "elbow_flex_deg",
+   "unit": "deg",
+   "threshold": 10.0,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "knee_flex_deg",
+   "unit": "deg",
+   "threshold": 10.0,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "time_s",
+   "unit": "s",
+   "threshold": 60.0,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "bar_at_shoulders",
+   "t": 3.4,
+   "log_step": 262,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 3.3–4.12 s，机器人在俯身伸手取杠时逐渐向画面左侧倾倒，躯干降至近水平，杠铃仍留在支架上；未完成上肩，也未恢复。log step 262–263 的第6回合服务结果为非脚部触台判负，仿真 4.12 s、control_steps=412，与视频 4.12 s 的倒伏末帧相符；断点取正在尝试的上肩步骤，视频 3.4 s 为失稳导航锚点。"
+  },
+  "unmet": {
+   "bar_at_shoulders": false,
+   "bar_overhead": false
+  },
+  "phenomena": [
+   {
+    "cat": "全身",
+    "item": "平衡",
+    "t": 3.4,
+    "log_step": 262,
+    "evidence_zh": "观察：视频 3.3–3.8 s，机器人俯身后支撑姿态持续偏斜，视频 3.92–4.12 s 身体向画面左侧翻倒、躯干几乎贴近台面；log step 262–263 的基准服务明确报告非脚部触台并终止最终回合。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "source_zh": "日志 / 视频",
+    "evidence_zh": "推断：log step 160 的控制脚本用局部机器人模型计算重心和关节目标；log step 250 的最终修改预先生成俯身与合手轨迹，使用固定增益及静态逆动力学补偿后整段提交，没有根据执行中姿态观测重新调整平衡。log step 262 执行这段脚本时，视频 3.3–4.12 s 显示机器人持续倾倒而未纠正，故可归为全身动作的控制失败；现有证据不足以进一步确定是逆解、动力学模型、增益还是跟踪反馈中的哪一项主导。"
+   }
+  ],
+  "gap_zh": "没有状态日志；视频已检查至末帧，但手指接触细节和最先触台的具体身体部位不能可靠分辨。log step 270 中 agent 将最终失败解释为力矩失衡导致膝部触台，这只是其自述，缺少本回合重心、接触对和跟踪误差记录，不能验证手部质量或踝部柔顺性是否为具体原因；因此根因只保留“控制”。评分器未报告DAG所列的五项原生指标值，均留null。"
+ },
+ "summary_zh": "最终评分的是 log step 262 重置后执行的第6回合：视频 0.0–4.12 s 只出现屈膝俯身伸手取杠，杠铃始终留在低位支架上，上肩与过顶均未完成（0/2）。视频约 3.4 s 起机器人明显失稳，最终倒伏并由服务判定非脚部触台，断点为“杠铃上肩”，现象为全身平衡失败，根因归于控制。缺少状态日志，无法确认短暂手指接触及具体失稳机制，agent 对力矩失衡和膝部触台的解释未独立证实。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "标签与视频、日志一致：最终评分的是 log step 262 重置后的第6回合，视频 0–4.12 s 只有屈膝俯身伸手，杠铃始终留在支架上，两节点均未达成；约 3.4 s 起失稳，4.0 s 前后倒伏，服务判非脚部触台。断点、平衡现象和控制根因均有证据支持，未发现需要修改的问题。",
+    "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": "2026-10-31T00:59:40Z",
+  "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/embodied-olympics-fun-clean-and-jerk-i00-standard--vk7mymb/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-clean-and-jerk-i00-standard--vk7mymb/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "3e494cb4236ed537e93ea2b3fd1727e5167056b96586b038350d842bf76ca4a9",
+   "video_sha256": "7bb3669d84346a3d68afe8625dc4ba89f23937f15282f118f396eebb1d51168e",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/clean_and_jerk/trajectories/embodied-olympics-fun-clean-and-jerk-i00-standard--wm5mamb.json

@@ -0,0 +1,248 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-clean-and-jerk-i00-standard--wm5mamb",
+ "taskKey": "embodied-olympics-fun/clean_and_jerk",
+ "dag_sha256": "f530c1d45bf82223dabd69c9d30cf24e240db1d588782640954595693a7015ca",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 6000,
+   "sim_steps": 9617,
+   "calls": 918,
+   "rows": 9617,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 1891.3,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 0.308,
+   "peak_processes": 13,
+   "peak_disk_gb": 1.454,
+   "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/embodied-olympics-fun-clean-and-jerk-i00-standard--wm5mamb/replay.mp4",
+  "duration_s": 60.04,
+  "fps": 25.0,
+  "frames": 1501,
+  "sha256": "7f8cded28dc21bf6bcd10a7be2b5f941b4975ef3c0f44d968c74cf09bcaaffdb"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "bar_at_shoulders": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 5.2,
+      "log_step": null,
+      "text_zh": "观察：视频 4.6–5.3 s，机器人屈膝下探，双手仍在杠轴上方，杠轴贴近小腿前侧后向前移出两只棕色支架；视频 5.5–6.5 s 杠铃落在木台上，未随双手抬起。"
+     },
+     {
+      "t": 26,
+      "log_step": 136,
+      "text_zh": "观察：视频 16、18、20、23、26、28 s，机器人由蹲姿空手起身并做屈肘动作，杠铃始终留在脚前木台，未到胸肩前；log step 136 的末次回合输出中，下降搜索、合手、起身和上肩后的 holding_left/right 均为 0。"
+     },
+     {
+      "t": 60,
+      "log_step": 146,
+      "text_zh": "观察：视频 59.2、59.6、60.0 s，杠铃仍在脚前木台，机器人保持空手举臂姿势；log step 146 的现场服务在 control_steps=6000 时返回 done=True、failed、outcome=time、stages_reached=0/4。未见上肩所需的持杠与抬升，故本事件从未达成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "推/扫",
+      "note_zh": "推断：视频 4.6–5.3 s 双手尚高于杠轴，前伸的小腿区域与杠轴相交并将其向前碰离支架；实际是腿部碰推杠铃，未见双手包握。单一斜前视角不能区分左右小腿各自的接触时刻。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0–28 s 机器人双脚在木台承重，双腿屈膝下蹲、再伸膝起身以尝试取杠上肩。"
+     }
+    ]
+   },
+   "bar_overhead": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 32,
+      "log_step": 136,
+      "text_zh": "观察：视频 30–35 s 机器人双臂向头顶上举，但手中没有杠铃，银色杠铃清楚地留在脚前木台；log step 136 的末次回合 overhead hold=(0.0,0.0)。这只是空手上举，杠铃过顶条件从未成立。"
+     },
+     {
+      "t": 60,
+      "log_step": 146,
+      "text_zh": "观察：视频 37.2、45、58.8、59.2、59.6、60.0 s，机器人维持空手举臂姿态，杠铃未离开台面；log step 146 的现场服务最终给出 0/4 原生阶段，与未举起杠铃的画面一致。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 30–60.0 s 双脚持续支撑于木台，机器人空手做过顶举臂并保持姿势，双手没有与仍在台面的杠铃接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "bar_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "bar_max_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "elbow_flex_deg",
+   "unit": "deg",
+   "threshold": 10.0,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "knee_flex_deg",
+   "unit": "deg",
+   "threshold": 10.0,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "time_s",
+   "unit": "s",
+   "threshold": 60.0,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "bar_at_shoulders",
+   "t": 5.2,
+   "log_step": null,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 4.6–5.3 s，机器人正在屈膝接近杠铃，双手尚未握杆，杠铃在小腿前侧受碰向前离架，至视频 5.5–6.5 s 已落在木台；此后机器人空手起身、上举，直到视频 60.0 s 也未把杠铃拿起。首次未修复的失败发生在取杠上肩步骤。"
+  },
+  "unmet": {
+   "bar_at_shoulders": false,
+   "bar_overhead": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "被挡/碰撞",
+    "t": 5.2,
+    "log_step": null,
+    "evidence_zh": "观察：视频 4.6–5.3 s 杠轴与机器人前伸的小腿区域相交，杠铃随下蹲动作向前移离支架并下落，双手仍明显高于杠轴；视频 6.5 s 杠铃已停在脚前木台。机器人碰落了要抓取的目标，随后未恢复取杠。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "抓取和接触设计",
+    "alt_feasible": "已验证有",
+    "source_zh": "视频",
+    "evidence_zh": "推断：视频 4.6–5.3 s 显示，机器人为取杠而屈膝时，小腿先进入杠轴所在空间，双手尚未下降到可包握位置，接近姿势没有为腿部避让杠铃、双手先建立抓握留出几何关系，因此杠铃先被碰离支架。视频 16–28 s 的空手起身和台面上的杠铃进一步确认没有建立承载抓握。可行替代仅由成功参考验证：inputs/oracle/replay.mp4 的参考视频 16–18 s 可见双手持杠于肩前，参考视频 25–27.36 s 可见持杠过顶；该参考不用于认定本次进度。"
+   }
+  ],
+  "gap_zh": "无状态日志，单一斜前方视频不能精确区分左右小腿的接触对及短暂指尖接触，也不能测出所有姿态阈值，但杠铃从未被举起足以判定两节点均未达成。agent 在 log step 150、155 自述由手掌或前臂碰落杠铃；视频 4.6–5.3 s 双手仍高于杠轴，画面支持小腿区域碰推，故不采用其具体部位解释。评分器未报告 DAG 所列的五项原生指标数值，均保留 null。"
+ },
+ "summary_zh": "评分轨迹为第五次重置后的末次现场回合：机器人下蹲取杠时在视频约 5.2 s 将杠铃碰离支架，随后空手起身和上举，杠铃一直留在木台，杠铃上肩与过顶均未完成，最终进度为 0/2。断点在杠铃上肩的操作中，视频支持小腿区域先碰杠、双手未建立承载抓握的接近几何问题；现场服务在 log step 146 返回回合时限结束、0/4 原生阶段。没有状态日志，具体左右接触对及细小阈值不可核验，agent 对碰撞部位的自述未获视频支持。",
+ "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 4.6–5.4 s 杠铃被小腿区域碰离支架并落在台面，之后全程空手，两节点均未达成，断点、现象和根因与证据相符，未发现需改内容的问题。",
+    "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": "2026-10-31T01:03:19Z",
+  "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/embodied-olympics-fun-clean-and-jerk-i00-standard--wm5mamb/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-clean-and-jerk-i00-standard--wm5mamb/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "32bf3950159e1c935e51300cc3f3e3def41ad3288c569465d660b40b570e06ed",
+   "video_sha256": "7f8cded28dc21bf6bcd10a7be2b5f941b4975ef3c0f44d968c74cf09bcaaffdb",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/clean_and_jerk/trajectories/embodied-olympics-fun-clean-and-jerk-i00-standard--ykz3cth.json

@@ -0,0 +1,247 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-clean-and-jerk-i00-standard--ykz3cth",
+ "taskKey": "embodied-olympics-fun/clean_and_jerk",
+ "dag_sha256": "f530c1d45bf82223dabd69c9d30cf24e240db1d588782640954595693a7015ca",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 478,
+   "sim_steps": 1636,
+   "calls": 131,
+   "rows": 1636,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 1596.6,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 3,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 1.575,
+   "peak_processes": 12,
+   "peak_disk_gb": 1.261,
+   "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/embodied-olympics-fun-clean-and-jerk-i00-standard--ykz3cth/replay.mp4",
+  "duration_s": 4.8,
+  "fps": 25.0,
+  "frames": 120,
+  "sha256": "532018c3b77a1be395bfeab39a970596d822c997cc22f1a9e066deaa7771e301"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "bar_at_shoulders": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–3.2 s，机器人在木台上小步移动后站立，双手保持在腰胸前方，银色杠铃仍横放在两只棕色支架上；没有双手下探合握或抬杠动作。"
+     },
+     {
+      "t": 3.6,
+      "log_step": null,
+      "text_zh": "观察：视频 3.4–3.8 s，杠杆在机器人小腿前方移动，两个套筒随后滑离支架并落到台面；视频 4.2–4.76 s，机器人前倾倒下，双手始终未将杠铃持至肩前。杠铃离架是滑落，未完成上肩。"
+     },
+     {
+      "t": null,
+      "log_step": 37,
+      "text_zh": "观察：log step 36 重置后只提交向前行走和停步动作；log step 37 固定读取到的关节目标并循环提交 lowcmd 姿态反馈，未提交下蹲抓杠或闭手动作。这与本次视频的站立、失衡过程一致，先前下蹲测试不属于本次评分回合。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.0–4.2 s，机器人双脚在木台上交替踏步、随后承重站立；视频 4.24–4.76 s 支撑失稳并倒向台面。双手没有合握杠铃。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "推/扫",
+      "note_zh": "推断：视频 3.4–3.76 s，机器人小腿与低位杠杆相交，杠铃随即向前移并从支架滑落，表明下肢意外推碰了杠杆；不是双手抓握或主动举起。视角不足以区分具体哪侧小腿。"
+     }
+    ]
+   },
+   "bar_overhead": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 4.76,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–4.76 s，杠铃一直处于支架或台面低位，机器人双臂没有持杠上举；末帧机器人已侧倒在木台上，杠铃仍在脚前台面。没有过顶伸臂保持的姿态。"
+     },
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：成功参考 inputs/oracle/replay.mp4 的参考视频 16–18 s 显示双手将杠铃持在肩前，参考视频 25–26 s 显示杠铃在头顶、双臂伸直；这些在同一场馆视角可见的标志均未出现在本次视频中。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "bar_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "bar_max_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "elbow_flex_deg",
+   "unit": "deg",
+   "threshold": 10.0,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "knee_flex_deg",
+   "unit": "deg",
+   "threshold": 10.0,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "time_s",
+   "unit": "s",
+   "threshold": 60.0,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "bar_at_shoulders",
+   "t": 3.4,
+   "log_step": 37,
+   "where": "首步前",
+   "evidence_zh": "观察：视频 0.0–3.2 s，机器人只踏步和站立，没有开始双手���杠；视频 3.2–4.12 s 躯干逐渐后仰，随后视频 4.24–4.76 s 前倾失去双脚支撑并倒在台上。log step 36–37 的最后回合动作仅为行走、停步和固定关节目标的平衡控制，尚未进入下蹲抓杠；上肩步骤此后没有完成。"
+  },
+  "phenomena": [
+   {
+    "cat": "没尝试",
+    "item": "没尝试",
+    "t": 0.0,
+    "log_step": 37,
+    "evidence_zh": "观察：视频 0.0–4.76 s，机器人双手始终未向低位杠杆下探合握，没有双手举杠动作；log step 36–37 的最后回合只执行行走与姿态保持，尚未启动上肩所需的抓取和抬升。"
+   },
+   {
+    "cat": "全身",
+    "item": "平衡",
+    "t": 3.4,
+    "log_step": 37,
+    "evidence_zh": "观察：视频 3.2–4.12 s 机器人持续后仰摇摆，视频 4.24–4.76 s 转为向前扑倒，双脚支撑丧失、躯干降至台面。agent 在 log step 44 自述最终因“非脚部触台”结束，与末段倒地画面一致；不另标违反约束。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "反馈控制",
+    "evidence_zh": "推断：log step 33 的控制函数使用骨盆 IMU 姿态和角速度给髋、踝叠加恢复力矩；log step 36 在行走后停步，log step 37 保持当时关节目标并循环运行该反馈控制。视频 2.2–3.0 s 尚能站立，随后视频 3.2–4.76 s 摇摆扩大直至倒地，说明该模式切换后的姿态反馈未能稳定机器人，导致抓杠前回合就结束；不是从未尝试本身反推根因。",
+    "source_zh": "日志"
+   }
+  ],
+  "gap_zh": "本次没有状态日志；最终回合 log step 36–39 的工具文本输出为空，log step 42、44 中的精确结束原因和原生阶段数只能作为 agent 自述，倒地事实由视频核验。视频可判断两个节点均未完成，但不能测量精确关节力矩、增益是否不当或推碰杠杆的具体腿侧；五项原生指标未出现在评分器字段中，均填 null。",
+  "unmet": {
+   "bar_at_shoulders": false,
+   "bar_overhead": false
+  }
+ },
+ "summary_zh": "观察：评分视频只包含最后一次重置后的踏步、站立和倒地回合，杠铃未由双手持起，杠铃上肩与杠铃过顶均未达成，最终为 0/2。推断：机器人在上肩首步开始抓杠前，姿态反馈未能抑制摇摆并于视频 4.24–4.76 s 倒地；先前调试回合的下蹲不计入本次进度。缺少状态日志和最终工具返回文本，无法进一步确认失稳的具体参数与精确接触对。",
+ "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（4.8 s，仅最后一次重置回合）与标注一致：杠铃未上肩/过顶，机器人踏步站立后后仰倒地，节点、断点（首步前）、没尝试+平衡、反馈控制根因均有视频和 log step 36–37、44 支撑，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": "2026-10-31T01:01:16Z",
+  "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/embodied-olympics-fun-clean-and-jerk-i00-standard--ykz3cth/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-clean-and-jerk-i00-standard--ykz3cth/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "3f9bcac36b41b316b3c59b0d4747d4d76ee9a29186407d4a6da6ce26f408626e",
+   "video_sha256": "532018c3b77a1be395bfeab39a970596d822c997cc22f1a9e066deaa7771e301",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/dunk_between_legs/trajectories/embodied-olympics-fun-dunk-between-legs-i00-privileged--2jlffhp.json

@@ -0,0 +1,220 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-dunk-between-legs-i00-privileged--2jlffhp",
+ "taskKey": "embodied-olympics-fun/dunk_between_legs",
+ "dag_sha256": "b3a9022ee9aaa233f14f035ce0b249f6a6cd4ac0723730a08b64571a639c20ad",
+ "outcome": {
+  "success": 0,
+  "reward": 0.2,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "n_actions": 1806,
+   "sim_steps": 1806,
+   "horizon": 4000,
+   "episode_over": 0,
+   "progress": 0.2,
+   "stages_reached": 1,
+   "stages_total": 5,
+   "fell": 0,
+   "replay_s": 18.29,
+   "bad_trajectory": 0,
+   "metric_ball_height_m": 0.119,
+   "metric_basket_touch_s": 0.0,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 4,
+   "agent_resets": 0,
+   "agent_replays": 2,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 1.728,
+   "peak_processes": 19,
+   "peak_disk_gb": 0.243,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-dunk-between-legs-i00-privileged--2jlffhp/replay.mp4",
+  "duration_s": 18.08,
+  "fps": 25.0,
+  "frames": 452,
+  "sha256": "b34d6fcc6cc28c950e82090b7f991cc5574b8fe3425b32040882f5897e4f4379"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 3,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "between_legs_bounce": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.4,
+      "log_step": null,
+      "text_zh": "观察：视频0.0 s机器人双手在胸前夹持篮球，视频0.2–0.4 s双手张开后球落到脚前地面；视频0.6–0.8 s球在机器人前方反弹，没有从两腿之间穿过。"
+     },
+     {
+      "t": 1.2,
+      "log_step": null,
+      "text_zh": "观察：视频1.0–2.8 s机器人手臂向球的方向调整，但球持续在手下方自由反弹，没有再次手触球；视频4.0–18.04 s球留在脚前地面，未见胯下穿越及对侧手触球确认。"
+     },
+     {
+      "t": 18.04,
+      "log_step": 125,
+      "text_zh": "观察：log step 123的最终生成脚本只运行release和walk阶段后追加站立并保存；log step 124重生成1806行交付轨迹，log step 125对该文件两次回放均只报告stages=[dribble]、progress=0.2、success=false，与视频中的释放及自由反弹一致。推断：原生首次合法触地阶段不等于本DAG的完整胯下运球步骤。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "候选接触名：双掌夹持。视频0.0 s左右张开的手共同夹持胸前篮球，随后双手张开释放；结合输入任务对初态双掌侧夹的描述判断，没有单手包住球，不符合抓握，现有表缺少双掌侧夹项。未观察到释放后再次拍球接触。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "视频0.0–18.04 s机器人通过双足站立及小幅踏步支撑躯干，尝试跟随脚前反弹的球；没有跌倒。"
+     }
+    ]
+   },
+   "ball_gathered": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.8,
+      "log_step": null,
+      "text_zh": "观察：视频0.2 s以后球离开双手，在视频0.6–2.8 s自由反弹，随后留在脚前；视频17.0–18.04 s机器人空手站立，球仍在地面，未见双手重新接住球。推断：开局持球不计本步骤，且胯下运球前置未完成，因此双手收球未达到。"
+     }
+    ]
+   },
+   "dunk_landed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 18.04,
+      "log_step": 125,
+      "text_zh": "观察：视频0.0–18.04 s机器人一直在开局位置附近活动，没有携球靠近篮圈、起跳或把球送入篮圈；末帧球在脚前地面。log step 125对交付文件的回放仅报告dribble阶段，未报告gather或dunk；评分器metric_basket_touch_s=0.0。推断：扣篮落地站稳未完成，单独站立不满足该节点。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "basket_touch_s",
+   "value": 0.0,
+   "unit": "s",
+   "threshold": 0.3,
+   "direction": "lower_better"
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "between_legs_bounce",
+   "t": 1.2,
+   "log_step": 124,
+   "where": "操作中",
+   "evidence_zh": "观察：视频0.4 s左右球首次落地后反弹，视频0.8–1.2 s机器人未用手再次触到反弹球；后续球的反弹逐渐变低，到视频4.0 s已留在���前，直至视频18.04 s仍未收回或穿过两腿。log step 124生成并保存的交付轨迹只达到dribble阶段，log step 125回放一致。推断：胯下运球中的再次控球接触未建立，整个步骤始终未修复。"
+  },
+  "unmet": {
+   "between_legs_bounce": false,
+   "ball_gathered": false,
+   "dunk_landed": false
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没建立接触",
+    "t": 1.2,
+    "log_step": 124,
+    "evidence_zh": "观察：视频0.6–2.8 s篮球在机器人前方自由反弹，手部调整后仍与球分离，没有形成释放后的下一次手触球；视频4.0–18.04 s手保持在地面球上方。log step 124交付文件生成时打印的右手接触检测touch持续为0。推断：缺少的是运球下一次拍球所需的接触；开局夹持后的主动释放本身不属于没保持住。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "log_step": 124,
+    "evidence_zh": "观察：log step 119–122修改手掌就绪门控、悬停位置与拍球窗口；log step 123把这些控制代码纳入最终生成脚本，log step 124运行该脚本生成交付文件时，打印st=hover、touch=0，球高度降低后仍保持悬停，最后仅stages=[dribble]。视频1.0–2.8 s手球分离及视频4.0–18.04 s球停地面与之相符。推断：该控制器未建立持续拍球所需的手球接触，也没有在球停地面后恢复控球，因而无法继续胯下运球；现有证据不能进一步确定是逆解、跟踪还是门控参数的主因。",
+    "source_zh": "日志"
+   }
+  ],
+  "gap_zh": "没有状态日志或成功参考视频；场馆远景可判断球与手分离及球停地面，但不足以核验初态每根手指的接触位置和接触力。最终生成日志有部分中段输出被截断，未把它当作逐帧状态路径；手部控制失败的更细子类无法确定。agent在log step 127自述手掌滞后约0.15 s及弱伺服，缺少独立测量，故不据此标具体根因。会话内的扣篮测试未进入交付轨迹，不用于本次节点或现象。"
+ },
+ "summary_zh": "机器人释放篮球并获得原生dribble阶段（进度0.2），但球只在身前反弹，未完成胯下运球，DAG三个节点最终均未达到。断点在胯下运球：释放后手未再次触球，球逐渐停在脚前，控制器未恢复控球；双手收球和扣篮未进入实际交付动作。没有状态日志，细分手部控制根因及初态精细接触几何仍有证据缺口。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "远景视频与标注一致：球释放后在脚前反弹、未穿过两腿，三个节点均未达到，断点与现象有证据支持；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": "2026-10-31T00:53:36Z",
+  "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/embodied-olympics-fun-dunk-between-legs-i00-privileged--2jlffhp/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-dunk-between-legs-i00-privileged--2jlffhp/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "1fe4c8a1c1cb300990aac6f7236c0c2a6c63da86c2184d4ee73a6bb665465a17",
+   "video_sha256": "b34d6fcc6cc28c950e82090b7f991cc5574b8fe3425b32040882f5897e4f4379",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/dunk_between_legs/trajectories/embodied-olympics-fun-dunk-between-legs-i00-privileged--bqmzhap.json

@@ -0,0 +1,195 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-dunk-between-legs-i00-privileged--bqmzhap",
+ "taskKey": "embodied-olympics-fun/dunk_between_legs",
+ "dag_sha256": "b3a9022ee9aaa233f14f035ce0b249f6a6cd4ac0723730a08b64571a639c20ad",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "n_actions": 10,
+   "sim_steps": 10,
+   "horizon": 4000,
+   "episode_over": 0,
+   "progress": 0.0,
+   "stages_reached": 0,
+   "stages_total": 5,
+   "fell": 0,
+   "replay_s": 1.93,
+   "bad_trajectory": 0,
+   "metric_ball_height_m": 1.105,
+   "metric_basket_touch_s": 0.0,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 115,
+   "agent_resets": 12792,
+   "agent_replays": 1,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 2.801,
+   "peak_processes": 28,
+   "peak_disk_gb": 0.239,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-dunk-between-legs-i00-privileged--bqmzhap/replay.mp4",
+  "duration_s": 0.12,
+  "fps": 25.0,
+  "frames": 3,
+  "sha256": "68e58a8e2fe8670b3138563742b6ad0cde523449876c02d4e9c6c1f6b0228c2c"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 3,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "between_legs_bounce": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": 57,
+      "text_zh": "观察：视频 0.00、0.04、0.08 s 的全部三帧中，机器人均站在起始位置、双手在胸前持球，未见球离手、触地反弹或穿过两腿；log step 57 的交付生成脚本仅重复初始保持命令10次。"
+     },
+     {
+      "t": 0.08,
+      "log_step": 77,
+      "text_zh": "观察：log step 77 对交付轨迹的回放结果为 executed_rows=10、stages=[]、episode_over=false，与视频末帧仍在起始位置持球一致；胯下运球未启动。"
+     }
+    ]
+   },
+   "ball_gathered": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.08,
+      "log_step": 77,
+      "text_zh": "观察：视频 0.00、0.04、0.08 s 中篮球始终留在机器人胸前双手之间，没有先反弹再接住的过程；log step 77 的交付回放也未达到任何阶段。推断：这是开局持球，不能计为完成胯下反弹后的双手收球，且其前置节点从未达到。"
+     }
+    ]
+   },
+   "dunk_landed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.08,
+      "log_step": 77,
+      "text_zh": "观察：视频 0.00、0.04、0.08 s 中机器人一直站在距篮圈数米的起始位置，篮球在胸前，未见向篮圈移动、扣球或扣篮后的落地过程；log step 77 的交付回放无已达阶段。推断：扣篮落地站稳节点未达到。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "basket_touch_s",
+   "value": 0.0,
+   "unit": "s",
+   "threshold": 0.3,
+   "direction": "lower_better"
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "between_legs_bounce",
+   "t": 0.0,
+   "log_step": 57,
+   "where": "首步前",
+   "evidence_zh": "观察：视频 0.00、0.04、0.08 s 中机器人只在原地胸前持球；log step 57 的交付脚本只执行10次初始保持命令，log step 77 验证该文件执行10行、无已达阶段且未触发终止。推断：评分尝试在胯下运球启动前结束，断点为首步胯下运球。"
+  },
+  "unmet": {
+   "between_legs_bounce": false,
+   "ball_gathered": false,
+   "dunk_landed": false
+  },
+  "phenomena": [
+   {
+    "cat": "没尝试",
+    "item": "没尝试",
+    "t": 0.0,
+    "log_step": 57,
+    "evidence_zh": "观察：视频 0.00、0.04、0.08 s 的全部三帧未见机器人释放或运篮球，身体保持起始站姿；log step 57 的生成脚本仅保持初态，agent 在 log step 86 也明确自述交付的是10行占位轨迹。"
+   }
+  ],
+  "root_causes": [],
+  "gap_zh": "没有状态日志或成功参考视频，但评分视频的全部三帧与交付脚本、回放结果一致，足以判定三个节点均未达到。根因留空：log step 52 的会话测试曾因球碰左大腿失败，log step 76、86 另记录了跳跃与持球困难，但这些测试未被交付，证据不足以确定首步控制未建成的具体根因；不能仅由主动保留占位文件推断规划或交付操作出错。"
+ },
+ "summary_zh": "评分视频 0.00—0.08 s 的全部三帧显示机器人原地胸前持球；三个语义节点均未达到，开局双手持球不计作胯下反弹后的收球。断点为胯下运球的首步前，现象是没尝试：log step 57 生成并在 log step 77 验证的交付仅含10行初始保持命令。会话测试中的碰腿、持球及跳跃问题不属于评分录像，尚不足以确定这次未启动的具体根因，故根因留空。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "视频仅三帧（0.00–0.08 s），机器人原地持球；三个节点均未达到、断点首步前、现象没尝试与 log step 57/77 的10行占位轨迹（grader stages=[]、progress 0）一致。根因留空并有 gap_zh，符合规则（agent 主动放弃不算根因）。",
+    "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": "2026-10-31T00:56:55Z",
+  "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/embodied-olympics-fun-dunk-between-legs-i00-privileged--bqmzhap/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-dunk-between-legs-i00-privileged--bqmzhap/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "844208122cc37df711bfa7bb7a5d5c7b3933e4ebec45ab85aaea9b29972b119b",
+   "video_sha256": "68e58a8e2fe8670b3138563742b6ad0cde523449876c02d4e9c6c1f6b0228c2c",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/dunk_between_legs/trajectories/embodied-olympics-fun-dunk-between-legs-i00-privileged--nteudmz.json

@@ -0,0 +1,247 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-dunk-between-legs-i00-privileged--nteudmz",
+ "taskKey": "embodied-olympics-fun/dunk_between_legs",
+ "dag_sha256": "b3a9022ee9aaa233f14f035ce0b249f6a6cd4ac0723730a08b64571a639c20ad",
+ "outcome": {
+  "success": 0,
+  "reward": 0.2,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "n_actions": 4000,
+   "sim_steps": 4000,
+   "horizon": 4000,
+   "episode_over": 1,
+   "progress": 0.2,
+   "stages_reached": 1,
+   "stages_total": 5,
+   "fell": 0,
+   "replay_s": 34.64,
+   "bad_trajectory": 0,
+   "metric_ball_height_m": 0.119,
+   "metric_basket_touch_s": 0.0,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 174,
+   "agent_resets": 110,
+   "agent_replays": 7,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 1.612,
+   "peak_processes": 12,
+   "peak_disk_gb": 0.989,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-dunk-between-legs-i00-privileged--nteudmz/replay.mp4",
+  "duration_s": 40.04,
+  "fps": 25.0,
+  "frames": 1001,
+  "sha256": "9232d5270beed36be2094018cba1e2d8a5ed89e255ecb01e18add21a9445d620"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 3,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "between_legs_bounce": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.6,
+      "log_step": 71,
+      "text_zh": "观察：视频 0.00–0.48 s 机器人双手在胸前两侧夹持篮球；视频约 0.60–0.68 s 球从双手下方离开，机器人双脚仍在原处支撑。log step 71 的交付控制先给 HOLD，随后持续给固定 RELEASE；视频中未见向胯下通道推球或拍球的动作，只有松开开局夹持。"
+     },
+     {
+      "t": 1.0,
+      "log_step": 177,
+      "text_zh": "观察：视频 0.68–1.44 s 球在机器人身前下降，约视频 0.96–1.00 s 触地后仍在身前上弹，没有从两腿间穿至身后；视频 1.5–4.9 s 后续弹跳逐渐降低，视频 5–40.00 s 球留在脚前地面，未再被手接住。log step 177 对最终交付文件的回放只报告 stages=[dribble]，即原生开始运球阶段，没有 between the legs。"
+     }
+    ]
+   },
+   "ball_gathered": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 1.32,
+      "log_step": null,
+      "text_zh": "观察：视频 1.20–1.44 s 首次反弹的球在双手下方达到最高处后再次下落，没有进入双手；视频 1.5–4.9 s 球继续自由反弹，视频 5–40.00 s 机器人空手停在原位，球在脚前。开局胸前双手持球发生在胯下运球之前，不满足本步的合法收球判据。"
+     }
+    ]
+   },
+   "dunk_landed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 40.0,
+      "log_step": 177,
+      "text_zh": "观察：视频 0–40.00 s 机器人始终留在起始位置，篮球释放后在脚前反弹并落地，未被带至篮圈，也未从篮圈上方扣入；视频 39.00–40.00 s 末段及最后一帧仍为机器人空手站立、球在脚前。log step 177 的最终交付回放结果 success=false，stages=[dribble]。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "basket_touch_s",
+   "value": 0.0,
+   "unit": "s",
+   "threshold": 0.3,
+   "direction": "lower_better"
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "between_legs_bounce",
+   "t": 1.0,
+   "log_step": 71,
+   "where": "首步前",
+   "evidence_zh": "观察：视频约 0.60–0.68 s 机器人松开开局夹持，球向脚前下落，视频约 0.96–1.00 s 触地后仍在身前反弹；直到视频 40.00 s 结束，未见手推球穿过两腿或重新控球。log step 71 的交付控制在 HOLD 后持续给固定 RELEASE，log step 176 确认最终动作与此前交付相同，log step 177 的回放只达到原生 dribble 阶段。推断：交付是仅含静态松手的基线，胯下运球的操作从未启动；重力造成的首次触地不足以认定本语义步骤已开始。"
+  },
+  "unmet": {
+   "between_legs_bounce": false,
+   "ball_gathered": false,
+   "dunk_landed": false
+  },
+  "phenomena": [
+   {
+    "cat": "没尝试",
+    "t": 1.0,
+    "log_step": 71,
+    "evidence_zh": "观察：log step 71 的交付控制先给 HOLD，再持续给固定 RELEASE；log step 176 确认最终交付动作相同。视频约 0.60–1.44 s 只见球离手后在身前下落、反弹���手未向胯下通道推球；视频 5.00–40.00 s 机器人空手站在原地，球在脚前。推断：胯下运球这一步在交付轨迹中从未启动，符合 docs/03_trajectory.md Step 3.1–3.2 的基线首步前与没尝试定义。",
+    "item": "没尝试"
+   }
+  ],
+  "root_causes": [],
+  "gap_zh": "交付只含开局夹持后的静态松手：inputs/run.json log step 71 的控制函数先给 HOLD，随后持续给固定 RELEASE，没有瞄准胯下通道的摆臂、推球或后续跟球动作；log step 176 确认最终动作与此前交付逐项相同，视频约 0.60–4.80 s 显示球在身前下落、反弹，视频 5.00–40.00 s 机器人空手原地站立。原生 dribble 阶段及 progress=0.2 记录了离手后的首次触地，但被动落球不证明胯下运球操作已启动，依 docs/03_trajectory.md:253–258、280–281 标首步前和没尝试。开局持球及站立是初态接触，不作为从未尝试节点的 contact_actual（Step 2）。log step 105、109 的摆臂和释放时机测试属于会话探索，未进入被评分交付；这些测试不足以确定本次静态松手之后未实施胯下动作的具体技术根因，故根因层留空。log step 173、179 的 NOTES.md 中 agent 自述交付不是解，不据此将停止本身归为原因（CASES.md FAIL-22），也不为交付中不存在的定向运动标位姿不准（FAIL-23）。没有逐帧模拟器状态日志；远景视频也不能精确核验掌指接触点及接触力。"
+ },
+ "summary_zh": "交付仅完成开局夹持后的静态松手，篮球在身前触地反弹后停在脚前；三个语义节点均未完成，原生 progress=0.2 只对应原生开始运球阶段。交付没有推球穿越胯下或后续控球动作，按从未启动该步的基线标为胯下运球／首步前／没尝试；被动反弹不标位姿不准。会话测试未进入交付，现有证据不足以确定具体技术根因，根因层留空并在 gap_zh 说明。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 3,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "fix",
+    "conf": 7,
+    "summary_zh": "节点值（三节点均未达）、时间轴与 native progress=0.2 一致；但交付轨迹只是开局夹持后的被动松手（log step 73/177、NOTES.md），不含任何瞄准胯下的运球动作，因此“位姿不准”和“控制”根因声称了交付里并不存在的动作（CASES.md FAIL-23）。",
+    "n_issues": 2,
+    "n_major": 2
+   },
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 2,
+    "verdict": "fix",
+    "conf": 7,
+    "summary_zh": "节点值、时间与 native progress=0.2 一致，位姿不准/控制根因的删除也正确；但交付只是松手让球自由落下，手从未推球运球，按 03 Step 3.1 属于“交付是从不启动该步的基线”，应为 首步前 + 没尝试，而不是 操作中 + 空现象层。",
+    "n_issues": 1,
+    "n_major": 1
+   },
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 3,
+    "kind": "confirm",
+    "verdict": "agree",
+    "summary_zh": "I1 已按提议完整落实：where=首步前，phenomena=没尝试，根因留空，证据与 gap_zh 一致，validate 通过。",
+    "n_issues": 0,
+    "n_checks": 1
+   }
+  ],
+  "open_zh": [],
+  "confirm": {
+   "verdict": "confirmed",
+   "summary_zh": "I1 已按提议完整落实：where=首步前，phenomena=没尝试，根因留空，证据与 gap_zh 一致，validate 通过。",
+   "checks_ok": "1/1",
+   "n_factual_errors": 0
+  }
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T01:14:15Z",
+  "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/embodied-olympics-fun-dunk-between-legs-i00-privileged--nteudmz/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-dunk-between-legs-i00-privileged--nteudmz/04_review_r1"
+   },
+   {
+    "stage": "reconcile",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-dunk-between-legs-i00-privileged--nteudmz/05_reconcile_r1"
+   },
+   {
+    "stage": "traj_review",
+    "round": 2,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-dunk-between-legs-i00-privileged--nteudmz/04_review_r2"
+   },
+   {
+    "stage": "reconcile",
+    "round": 2,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-dunk-between-legs-i00-privileged--nteudmz/05_reconcile_r2"
+   },
+   {
+    "stage": "confirm",
+    "round": 3,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-dunk-between-legs-i00-privileged--nteudmz/06_confirm_r3"
+   }
+  ],
+  "confirm_prompt": "prompts/06_confirm.md",
+  "confirm_prompt_sha256": "b9f56c095f7ea5e609e58bb21e076f565766c370333d71c7a3512cffb4d78267",
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "845327a45b6605915eea5724e0fb4b01e635d58c2cdc6406aae8829737567c61",
+   "video_sha256": "9232d5270beed36be2094018cba1e2d8a5ed89e255ecb01e18add21a9445d620",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/dunk_between_legs/trajectories/embodied-olympics-fun-dunk-between-legs-i00-privileged--t63onfs.json

@@ -0,0 +1,220 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-dunk-between-legs-i00-privileged--t63onfs",
+ "taskKey": "embodied-olympics-fun/dunk_between_legs",
+ "dag_sha256": "b3a9022ee9aaa233f14f035ce0b249f6a6cd4ac0723730a08b64571a639c20ad",
+ "outcome": {
+  "success": 0,
+  "reward": 0.2,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "n_actions": 452,
+   "sim_steps": 452,
+   "horizon": 4000,
+   "episode_over": 0,
+   "progress": 0.2,
+   "stages_reached": 1,
+   "stages_total": 5,
+   "fell": 0,
+   "replay_s": 5.0,
+   "bad_trajectory": 0,
+   "metric_ball_height_m": 0.143,
+   "metric_basket_touch_s": 0.0,
+   "metric_dribbles": 1.0,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 390,
+   "agent_resets": 132,
+   "agent_replays": 3,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 1.059,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.268,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-dunk-between-legs-i00-privileged--t63onfs/replay.mp4",
+  "duration_s": 4.56,
+  "fps": 25.0,
+  "frames": 114,
+  "sha256": "5407adff3fc9fbe3c62ea52958907bf6af69a0b18a44b8e18abdeecd9651986d"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 3,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "between_legs_bounce": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.5,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–0.16 s 机器人松开身前篮球，视频约 0.50 s 球在双脚前方首次触地反弹；视频 0.68–0.84 s 右手再次触球并向下拨球，球随后向身前弹离，未见从腿间穿到身体另一侧。"
+     },
+     {
+      "t": 4.52,
+      "log_step": 364,
+      "text_zh": "观察：视频 0.88–4.52 s 篮球持续在机器人前方低弹并逐渐远离双手，未再被手接触确认胯下穿越。log step 364 检查已保存文件为452行动行；评分器字段 stages_reached=1/5、metric_dribbles=1.0，与仅完成普通运球的画面一致，不能将该原生阶段算成本节点完成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 0.00–0.08 s 左右手在身前共同夹持篮球，随后分开释放；候选接触名“双掌夹持”，单手未包围整球，不能标为抓握，现有词表无双掌侧夹项。"
+     },
+     {
+      "who": "手指/夹爪",
+      "how": "推/扫",
+      "note_zh": "观察：视频 0.68–0.84 s 机器人右手掌/手指接触反弹到身前的篮球并向下拨送；左手在身后，未同时参与这次触球。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–0.84 s 双足接触球场支撑躯干及下蹲拨球动作，之后机器人起身并小幅踏步；未见篮球实际穿过腿间，故不追加腿部避开球的接触项。"
+     }
+    ]
+   },
+   "ball_gathered": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.76,
+      "log_step": null,
+      "text_zh": "观察：视频 0.68–0.84 s 只有右手在反弹球附近接触，另一手在身后；此后至视频 4.52 s 球在地面附近自由弹跳，两手未重新共同接住球。开局双手持球发生在胯下运球前，不满足本节点。"
+     }
+    ]
+   },
+   "dunk_landed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 4.52,
+      "log_step": null,
+      "text_zh": "观察：全片视频 0.00–4.52 s 机器人始终在远离篮圈的起始区域，篮球在其身前地面附近弹跳，没有被举至篮圈上方或由手送穿篮圈；最后一秒视频 3.52–4.52 s 也未发生扣篮及其后的站稳验收。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "basket_touch_s",
+   "value": 0.0,
+   "unit": "s",
+   "threshold": 0.3,
+   "direction": "lower_better"
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "between_legs_bounce",
+   "t": 0.84,
+   "log_step": 355,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 0.68–0.84 s 机器人下蹲用右手拨动反弹球，视频 0.88–1.36 s 球却继续向身前落地反弹，没有穿过腿间；到视频 4.52 s 仍未收回或补成胯下穿越。log step 355 的交付生成脚本对应松手、下蹲、右手下拨及随后步行动作，log step 364 确认保存452行，与评分器 n_actions=452 一致。"
+  },
+  "unmet": {
+   "between_legs_bounce": false,
+   "ball_gathered": false,
+   "dunk_landed": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 0.84,
+    "log_step": null,
+    "evidence_zh": "观察：视频 0.72–0.84 s 右手触球后，篮球在视频 0.88–1.36 s 向机器人身前弹出，落点和后续路径仍在双脚前方，没有进入所需的腿间通道；视频 1.40–4.52 s 继续在前方低弹，未修复。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "source_zh": "日志 / 视频",
+    "evidence_zh": "推断：log step 355 的已保存轨迹生成程序用固定25步下蹲和12步右臂关节插值执行拨球，然后直接过渡起身和步行，未按球的位置或腿间穿越结果修正拨球。视频 0.68–0.84 s 已有右手触球，随后球向身前弹离而非穿过腿间，表明这次动作未控制出所需球路；只能定位到控制，现有证据不足以进一步区分接触几何、关节跟踪或参数问题。"
+   }
+  ],
+  "gap_zh": "没有状态日志或成功参考视频，单一远景无法量化腿门法向速度、手球接触力和关节跟踪误差，故控制根因不细分。log step 130 等会话试验及后续人工设置收球阶段的扣篮试验不是交付轨迹，不用于补计节点。log step 496 记录模型过滤提示，但log step 364此前已确认交付文件有452行，log step 367已启动该文件回放，且现有评分视频明确显示已保存尝试的物理失败，按该尝试正常标注。"
+ },
+ "summary_zh": "评分轨迹仅完成身前的普通反弹及一次右手拨球，胯下运球、合法双手收球和扣篮落地站稳均未完成，语义进度为0/3。视频约 0.84 s 右手拨球后球向身前弹离，没有穿过腿间，断点为胯下运球，根因定位到控制。没有状态日志可细分控制误差；log step 496的模型过滤发生在452行交付文件已保存之后，后续会话调试中的收球或扣篮阶段不计入本次评分进度。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "标注与视频一致：视频 0–1.4 s 球松手、在身前触地反弹、机器人下蹲右手拨球后球继续在身前弹离，全程未穿腿间；三节点均未达成，断点 between_legs_bounce 与评分器 stages_reached=1/5 一致，位姿不准+控制（已保守不细分）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": "2026-10-31T00:54:11Z",
+  "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/embodied-olympics-fun-dunk-between-legs-i00-privileged--t63onfs/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-dunk-between-legs-i00-privileged--t63onfs/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "f2f60ade62fb42dbf0efbbb78a9d95289ddd62f2caa0c24ab3c313e48166d9ba",
+   "video_sha256": "5407adff3fc9fbe3c62ea52958907bf6af69a0b18a44b8e18abdeecd9651986d",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}