Add 118 trajectories (phase 2, remaining runs; batch S42)

phase2/embodied-olympics-fun/badminton_serve/trajectories/embodied-olympics-fun-badminton-serve-i00-privileged--bce4c43.json

@@ -0,0 +1,225 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-badminton-serve-i00-privileged--bce4c43",
+ "taskKey": "embodied-olympics-fun/badminton_serve",
+ "dag_sha256": "e5772140aff5f054a271f90fd295419869c755e083b6526601e08a99bee573e1",
+ "outcome": {
+  "success": 0,
+  "reward": 0.25,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "n_actions": 101,
+   "sim_steps": 101,
+   "horizon": 1500,
+   "episode_over": 1,
+   "progress": 0.25,
+   "stages_reached": 1,
+   "stages_total": 4,
+   "fell": 0,
+   "replay_s": 5.91,
+   "bad_trajectory": 0,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 122,
+   "agent_resets": 117,
+   "agent_replays": 8,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 0.987,
+   "peak_processes": 12,
+   "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-badminton-serve-i00-privileged--bce4c43/replay.mp4",
+  "duration_s": 1.04,
+  "fps": 25.0,
+  "frames": 26,
+  "sha256": "901d60ad0f3ba3cf704e823baf0f62b2eb847b76e08b16ae619e735b00fd9f55"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "serve_in_target": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–1.00 s 的全部 26 帧及机器人局部放大图显示，机器人在近侧发球区站立，双臂调整持球、持拍位置，随后右臂在身前低处挥拍；全段没有羽毛球飞越球网并落入远侧对角区的过程。"
+     },
+     {
+      "t": 1.0,
+      "log_step": 211,
+      "text_zh": "观察：log step 211 对已保存 trajectory.npz 的两次独立回放均返回 stages=[release]、steps=101、success=false，并报告“the racket hit the feathers first, not the base of the shuttle”；这是基准回放裁判的输出，说明已放球，但首次拍球接触不合法。视频 0.92–1.00 s 的末段仍在近侧挥拍，与此终止阶段一致。"
+     },
+     {
+      "t": 1.0,
+      "log_step": 221,
+      "text_zh": "观察：log step 216 恢复原挥拍参数并重新生成 101 行交付文件，log step 218 核对两次回放末态哈希相同、仅回放耗时不同，log step 221 从最终交付目录再次运行生成脚本仍返回先触羽毛的同一失败；inputs/run.json 的 reward 也给出 deterministic=1、stages_reached=1/4。推断：会话中其他调参测试不属于本视频；本次合法发球入区从未完成，不能将原生放球阶段的 0.25 当成语义节点完成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手持工具",
+      "how": "其他",
+      "note_zh": "观察：视频 0.92–1.00 s 可见机器人右手所持球拍在身前挥动；log step 211 的交付回放裁判确认球拍拍面实际先碰到羽毛球羽毛。候选：击打；现有接触方式无击打，不能以推/扫或压/擦代替。左手先持球后放开，再由右手经球拍接触球；不重复标注右手握拍链路。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–1.00 s 双脚持续支撑球场地面，机器人未跌倒；这里记录可见的站立支撑，不据远景断言毫米级脚位与离地时长达标。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "hit_height_m",
+   "unit": "m",
+   "threshold": 1.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "shuttle_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "shuttle_angle_deg",
+   "unit": "deg",
+   "threshold": 0,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "landing_x_m",
+   "unit": "m",
+   "threshold": [
+    4.43,
+    9.15
+   ],
+   "direction": "target",
+   "value": null
+  },
+  {
+   "name": "landing_y_m",
+   "unit": "m",
+   "threshold": [
+    0.43,
+    3.04
+   ],
+   "direction": "target",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "serve_in_target",
+   "t": 1.0,
+   "log_step": 211,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 0.92–1.00 s 机器人正在完成低位挥拍，尚无过网飞行；log step 211 的已保存文件回放裁判在 101 个控制步后以“球拍先碰羽毛而非球托”终止，仅接受 release 阶段。log step 216、218、221 确认最终重新生成的文件仍重现同一失败。因此断点属于合法发球入区的击球操作。"
+  },
+  "unmet": {
+   "serve_in_target": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 1.0,
+    "log_step": 211,
+    "evidence_zh": "观察：视频 0.92–1.00 s 右手球拍在机器人身前低处与放球后的羽毛球位置汇合，远景不足以区分球托和羽毛；log step 211 对交付文件的回放裁判明确报告首先碰到羽毛。推断：拍面与羽毛球球托的相对对位未实现，击中错误部位；已有拍球接触，不标没建立接触，也无依据细分为击球过早或过晚。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "source_zh": "日志",
+    "evidence_zh": "推断：机器人未能把实际挥拍路径对准球托。log step 175 的生成脚本以固定腰部和双臂关节目标执行开环放球、撤手、挥拍；会话诊断 log step 179 记录接触附近球拍中心 y≈0.048 m、羽毛球 y≈0.175 m，log step 205 同一原始参数加随挥后的测试仍报告先碰羽毛、位置差约 0.147 m。log step 210、216 重新生成最终文件后，同一失败由 log step 211、218、221 的交付回放核实；这些结果支持挥拍对位的控制问题，但不足以确定是逆解、跟踪还是接触设计的单一子类。"
+   }
+  ],
+  "gap_zh": "没有状态日志；640×360 的远景视频仅以 25 fps 记录，放大后仍无法直接分辨球托首次接触及毫秒级接触时长，先触羽毛依据已核实最终交付的基准回放裁判。会话位置诊断用于支持控制归因，不作为逐帧状态路径或视频计时依据；不能进一步确定控制子类。评分器未报告 DAG 的五项原生指标，均记 null。"
+ },
+ "summary_zh": "机器人完成放球并实际挥拍触球，但未完成“合法发球入区”，语义节点最终为 0/1。最终交付回放在击球时因球拍先碰羽毛而非球托终止，现象为击球对位不准，日志支持控制问题。无状态日志且远景无法分辨细微接触，具体首触部位依赖交付回放裁判，控制子类与五项原生指标无法确定。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "标签与视频、log 一致：放球阶段达成（release，0.25），球拍先碰羽毛导致失败，serve_in_target 未达成；断点、位姿不准现象和控制根因均有 log 回放裁判与 step 179/205 的几何数据支持，未发现需要修改之处。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:54:32Z",
+  "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-badminton-serve-i00-privileged--bce4c43/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-badminton-serve-i00-privileged--bce4c43/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "b1a2e33424c1c8c270512479e9652625488bbb3e41491c89f392c6fa22b1ded5",
+   "video_sha256": "901d60ad0f3ba3cf704e823baf0f62b2eb847b76e08b16ae619e735b00fd9f55",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/badminton_serve/trajectories/embodied-olympics-fun-badminton-serve-i00-privileged--du3sucw.json

@@ -0,0 +1,211 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-badminton-serve-i00-privileged--du3sucw",
+ "taskKey": "embodied-olympics-fun/badminton_serve",
+ "dag_sha256": "e5772140aff5f054a271f90fd295419869c755e083b6526601e08a99bee573e1",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "n_actions": 298,
+   "sim_steps": 298,
+   "horizon": 1500,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 4,
+   "stages_total": 4,
+   "fell": 0,
+   "replay_s": 8.2,
+   "bad_trajectory": 0,
+   "metric_hit_height_m": 0.7396,
+   "metric_landing_x_m": 5.933,
+   "metric_landing_y_m": 1.554,
+   "metric_shuttle_angle_deg": 38.16,
+   "metric_shuttle_speed_m_s": 11.6684,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 115,
+   "agent_resets": 584,
+   "agent_replays": 9,
+   "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": 3.747,
+   "peak_processes": 31,
+   "peak_disk_gb": 0.843,
+   "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-badminton-serve-i00-privileged--du3sucw/replay.mp4",
+  "duration_s": 3.0,
+  "fps": 25.0,
+  "frames": 75,
+  "sha256": "078d7e00d016c6d43845b1169ad65b7f26cc6986893d549550f33c2fcbc04be7"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "serve_in_target": {
+    "method": "visual",
+    "intervals": [
+     [
+      2.98,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 1.68,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–1.52 s，机器人在近侧发球区原地调整手臂和球拍；视频 1.56–1.72 s 左手从放球处移开，右手球拍向前上方挥动；视频 1.76–2.00 s 白色羽毛球离开机器人向上飞向球网，两脚保持地面支撑。远景不能直接分辨球托初触及手指接触细节。"
+     },
+     {
+      "t": 2.0,
+      "log_step": 197,
+      "text_zh": "观察：视频 1.92–2.08 s 羽毛球由近侧升至网顶上方并飞向远侧淡黄色目标区方向；视频 2.24–2.96 s 机器人仍站立、右手仍持拍。log step 197 中最终选中轨迹的原生裁判输出依次记录 release、hit、over the net、in，success=True、failed=None，击球最高处 0.7396 m、出球速度 11.6684 m/s、仰角 38.16°、过网净空 0.575 m。"
+     },
+     {
+      "t": 2.98,
+      "log_step": 197,
+      "text_zh": "推断：log step 196 选择最终候选，log step 197 的生成输出为 298 个控制步、仿真 2.98 s，故控制步长为 0.01 s；298×0.01=2.98 s 与视频约 3.00 s 的长度相符，且裁判仿真 1.690 s 的击球与视频 1.68–1.72 s 挥拍出球相符，仿真 1.990 s 的过网与视频约 2.00 s 相符。据此将最终控制步对应约视频 2.98 s，作为节点完成锚点；裁判实际记录仿真 2.973 s 首次落地区内。视频末帧 2.96 s 尚未显示触地，完成时刻依赖这一已核对的交付回放时钟，而非 agent 工作时间。"
+     },
+     {
+      "t": 2.98,
+      "log_step": 202,
+      "text_zh": "观察：inputs/run.json 的评分器字段 success=1、replay_success=1、deterministic=1、stages_reached=4/4，落点为 (5.933,1.554) m，处于 DAG 指定的接发球区。log step 199 验证最终文件重新生成后字节一致且为 298 行；log step 200 启动该文件的两次新进程回放，log step 202 输出两次成功且末态哈希均为 8ad0c87716c429273574da561b752e29645f02dedfe0e43cb6ae80d442ae5be9，与 log step 197 一致，因此采用最终交付结果，未采用此前 302 步候选。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手持工具",
+      "how": "其他",
+      "note_zh": "候选：击打。观察：视频 1.68–1.80 s 机器人右手持球拍前上挥，随后羽毛球离拍向上飞行；结合 log step 197 的原生 hit 事件，判定实际发生球拍击球。球拍与右手的握持链路不重复计数；左手此前持球并先行放开，球托初触由裁判事件补充确认。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–2.96 s 机器人双脚支撑球场地面，发球期间身体保持站立，未见移步或跌倒。远景不足以独立量出毫米级脚部位移。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "hit_height_m",
+   "unit": "m",
+   "threshold": 1.15,
+   "direction": "lower_better",
+   "value": 0.7396
+  },
+  {
+   "name": "shuttle_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": 11.6684
+  },
+  {
+   "name": "shuttle_angle_deg",
+   "unit": "deg",
+   "threshold": 0,
+   "direction": "higher_better",
+   "value": 38.16
+  },
+  {
+   "name": "landing_x_m",
+   "unit": "m",
+   "threshold": [
+    4.43,
+    9.15
+   ],
+   "direction": "target",
+   "value": 5.933
+  },
+  {
+   "name": "landing_y_m",
+   "unit": "m",
+   "threshold": [
+    0.43,
+    3.04
+   ],
+   "direction": "target",
+   "value": 1.554
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人完成唯一节点“合法发球入区”：视频约 1.7 s 挥拍将球向上击出，视频约 2.0 s 越网，最终交付轨迹的裁判确认四阶段全部完成，落点 (5.933,1.554) m 在目标区内。最终 298 步文件经两次新进程成功回放，评分器确认成功且确定性一致，无失败断点。无状态日志，远景无法核验毫秒级接触和脚部位移，视频末帧 2.96 s 早于裁判记录的首次落地，故约视频 2.98 s 的完成锚点及精细合法性依赖已核对的最终交付回放输出与评分器字段；oracle 为本运行自身，没有独立参考。",
+ "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，唯一节点 serve_in_target 终态 true，完成锚点按已核对的控制步映射取视频约 2.98 s（裁判首次落地 sim 2.973 s），与 grader 字段一致；contact_actual（球拍击打、腿/躯干支撑）合理，无失败断点，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": "2026-10-31T00:47:36Z",
+  "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-badminton-serve-i00-privileged--du3sucw/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-badminton-serve-i00-privileged--du3sucw/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "33bff7f102240c4478b1d1ea0a2218ce9bb76f1f296a829415a350cbab27288a",
+   "video_sha256": "078d7e00d016c6d43845b1169ad65b7f26cc6986893d549550f33c2fcbc04be7",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/badminton_serve/trajectories/embodied-olympics-fun-badminton-serve-i00-privileged--kicwndu.json

@@ -0,0 +1,211 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-badminton-serve-i00-privileged--kicwndu",
+ "taskKey": "embodied-olympics-fun/badminton_serve",
+ "dag_sha256": "e5772140aff5f054a271f90fd295419869c755e083b6526601e08a99bee573e1",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "n_actions": 193,
+   "sim_steps": 193,
+   "horizon": 1500,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 4,
+   "stages_total": 4,
+   "fell": 0,
+   "replay_s": 5.96,
+   "bad_trajectory": 0,
+   "metric_hit_height_m": 0.5789,
+   "metric_landing_x_m": 5.174,
+   "metric_landing_y_m": 1.529,
+   "metric_shuttle_angle_deg": 36.37,
+   "metric_shuttle_speed_m_s": 9.7762,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 23,
+   "agent_resets": 2,
+   "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": 11.286,
+   "peak_processes": 28,
+   "peak_disk_gb": 0.982,
+   "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-badminton-serve-i00-privileged--kicwndu/replay.mp4",
+  "duration_s": 1.96,
+  "fps": 25.0,
+  "frames": 49,
+  "sha256": "871e2641bf05ab8210b66e63add0c983e715fc036912ae6c049ef018d39a4d70"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "serve_in_target": {
+    "method": "visual",
+    "intervals": [
+     [
+      1.93,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.64,
+      "log_step": null,
+      "text_zh": "观察：已逐帧检查视频 0.00–1.92 s。视频 0.00–0.48 s 机器人左手将白色羽毛球带到身前，右手球拍由竖起转至较低位置；视频 0.60–0.84 s 白色羽毛球离开左手下落，视频 0.84–0.92 s 机器人右臂挥拍后羽毛球转为向前上方飞行，双脚仍支撑地面。"
+     },
+     {
+      "t": 1.12,
+      "log_step": null,
+      "text_zh": "观察：视频 0.96–1.12 s 的逐帧放大图中，白色羽毛球向球网方向升高；视频 1.16–1.92 s 机器人保持站立并持拍，球网遮住后续球路和首次落地接触点。成功参考视频也使用同一远景视角，仅用于核对可见线索，未用其结果替代本次证据。"
+     },
+     {
+      "t": 1.92,
+      "log_step": 162,
+      "text_zh": "观察：log step 162 对最终保存的 trajectory.npz 运行官方回放，裁判返回 success=true、failed=null，阶段依次为 release、hit、over the net、in，执行 193 个控制步、仿真 1.93 s；击球高度 0.5789 m、出球速度 9.7762 m/s、仰角 36.37°、首次落点 (5.174,1.529) m，落在 DAG 指定区间内。log step 164 的再次回放末态哈希相同；本次 reward 同样给出 replay_success=1、deterministic=1 和全部相同指标。"
+     },
+     {
+      "t": 1.93,
+      "log_step": 162,
+      "text_zh": "推断：log step 162 的最终交付回放执行 193 个控制步共仿真 1.93 s，对应每控制步 0.01 s；结合视频 25 fps、49 帧及末帧视频 1.92 s，将完成时刻近似定位为视频 1.93 s。已用 log step 152 的交付脚本在控制步 75 开始挥拍（后续修改仅调整放球步和挥拍终点），对照视频 0.76–0.92 s 的实际挥拍核对该映射。合法落地与毫秒级击球条件由最终交付回放裁判和评分器确认，未声称画面直接拍到首次触地。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手持工具",
+      "how": "其他",
+      "note_zh": "候选：击打；视频 0.84–0.92 s 机器人右手握持球拍挥动，羽毛球随之向前上方飞出，最终交付回放裁判确认 hit。作用端为球拍拍面，对象为羽毛球；现有方式没有击打，不用推/扫或压/擦代替，也不重复标注握拍链路。此前左手持球并放开，随后才由球拍击球。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "视频 0.00–1.92 s 机器人双脚支撑球场地面，呈现球、放球、挥拍期间保持站立；不从远景推断毫米级脚位移。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "hit_height_m",
+   "unit": "m",
+   "threshold": 1.15,
+   "direction": "lower_better",
+   "value": 0.5789
+  },
+  {
+   "name": "shuttle_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": 9.7762
+  },
+  {
+   "name": "shuttle_angle_deg",
+   "unit": "deg",
+   "threshold": 0,
+   "direction": "higher_better",
+   "value": 36.37
+  },
+  {
+   "name": "landing_x_m",
+   "unit": "m",
+   "threshold": [
+    4.43,
+    9.15
+   ],
+   "direction": "target",
+   "value": 5.174
+  },
+  {
+   "name": "landing_y_m",
+   "unit": "m",
+   "threshold": [
+    0.43,
+    3.04
+   ],
+   "direction": "target",
+   "value": 1.529
+  }
+ ],
+ "failure": null,
+ "summary_zh": "本次交付完成“合法发球入区”节点，最终进度 1/1；视频可见机器人左手放球、右手挥拍后羽毛球向前上方飞出，最终交付回放及评分器确认四阶段全部通过，落点为 (5.174,1.529) m。节点在视频约 1.93 s 完成，时间由最终回放的控制步与视频挥拍核对后推断；无状态日志，后段球路和首次落地被球网遮挡，球托初触及毫秒级接触条件依赖裁判结果确认。",
+ "conf": 7,
+ "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，单节点 serve_in_target 达成、final=true，时间 1.93 s 由控制步映射推断并已标明；接触标注（手持工具·其他·击打、腿/躯干·支撑）合理，未发现需要修改的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:53:40Z",
+  "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-badminton-serve-i00-privileged--kicwndu/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-badminton-serve-i00-privileged--kicwndu/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "e2d62bf5cf9df4f6edb266181bf2395e240ab5d8f8bee4baf13f317801194839",
+   "video_sha256": "871e2641bf05ab8210b66e63add0c983e715fc036912ae6c049ef018d39a4d70",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/badminton_serve/trajectories/embodied-olympics-fun-badminton-serve-i00-privileged--mbmlydq.json

@@ -0,0 +1,221 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-badminton-serve-i00-privileged--mbmlydq",
+ "taskKey": "embodied-olympics-fun/badminton_serve",
+ "dag_sha256": "e5772140aff5f054a271f90fd295419869c755e083b6526601e08a99bee573e1",
+ "outcome": {
+  "success": 0,
+  "reward": 0.25,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "n_actions": 143,
+   "sim_steps": 143,
+   "horizon": 1500,
+   "episode_over": 1,
+   "progress": 0.25,
+   "stages_reached": 1,
+   "stages_total": 4,
+   "fell": 0,
+   "replay_s": 6.23,
+   "bad_trajectory": 0,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 145,
+   "agent_resets": 6847,
+   "agent_replays": 0,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 2.68,
+   "peak_processes": 17,
+   "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": 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-badminton-serve-i00-privileged--mbmlydq/replay.mp4",
+  "duration_s": 1.44,
+  "fps": 25.0,
+  "frames": 36,
+  "sha256": "c3e0adebe8b4c07d0dd6b338c601156bae469720e085e3bf1ce2082211b7275a"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "serve_in_target": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 1.0,
+      "log_step": 13,
+      "text_zh": "观察：视频 0.00–1.00 s 机器人双脚站在近侧场地、右手持拍，未见挥拍；视频 1.04–1.36 s 白色羽毛球从左手附近向身前下方下落，球拍仍留在右手前上方，未见羽毛球向球网飞出。"
+     },
+     {
+      "t": 1.4,
+      "log_step": 13,
+      "text_zh": "观察：视频 1.40 s 末帧机器人仍站立持拍，球没有飞到远侧目标区；log step 13 保存的占位轨迹返回“you missed the shuttle: it reached the floor without a hit”，共 143 个动作，与本次 reward 的 n_actions=143、stages_reached=1/4、replay_success=0 一致。推断：本次只完成放球，未完成合法击球、过网和入区，事件节点全程未达成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "观察：视频 0.00–1.00 s 左手在身前持羽毛球，随后放开，视频 1.04–1.36 s 球向下离开手部；右手始终握球拍拍柄。视频未见球拍拍面击中羽毛球，故不填写手持工具击打接触。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–1.40 s 机器人双脚在近侧球场地面支撑身体，放球过程中保持站立。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "hit_height_m",
+   "unit": "m",
+   "threshold": 1.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "shuttle_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "shuttle_angle_deg",
+   "unit": "deg",
+   "threshold": 0,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "landing_x_m",
+   "unit": "m",
+   "threshold": [
+    4.43,
+    9.15
+   ],
+   "direction": "target",
+   "value": null
+  },
+  {
+   "name": "landing_y_m",
+   "unit": "m",
+   "threshold": [
+    0.43,
+    3.04
+   ],
+   "direction": "target",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "serve_in_target",
+   "t": 1.12,
+   "log_step": 13,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 1.04–1.36 s 左手放出的羽毛球在机器人身前下落，右手球拍没有挥向球，未发生向网击出的过程。log step 13 的保存结果明确为球未击中便落地，保存动作数 143 与评分记录一致；该次发球失败后没有修复。"
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没建立接触",
+    "t": 1.12,
+    "log_step": 13,
+    "evidence_zh": "观察：视频 1.04–1.36 s 羽毛球从手部附近下落，右手球拍保持在身体右前上方，没有随球下挥或击出球；log step 13 保存轨迹的反馈为“you missed the shuttle: it reached the floor without a hit”。所需的球拍与羽毛球击打接触未建立。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "log_step": 13,
+    "source_zh": "日志",
+    "evidence_zh": "推断：交付的动作序列未实现放球后的挥拍控制。log step 13 的保存脚本逐步复制初始 hold，仅在循环 k≥100 时把左��开合量 row[17] 改为 0，右臂目标保持不变；视频 1.04–1.40 s 的持拍不挥与此一致。后续会话 log step 50 的带裁判轨迹网格仍多次返回漏球或击球前停挥，log step 53 的动态逆解修正返回“no valid contact candidate”，支持挥拍轨迹控制未解决；这些测试不计入本次节点进度。"
+   }
+  ],
+  "gap_zh": "观察：inputs/ 没有状态日志；视频仅有远景，羽毛球很小，视频 1.40 s 末帧不能精确核实首次触地时刻及接触细节，落地判负由 log step 13 的已保存轨迹反馈与评分记录支持。五项原生指标在 reward 中均未报告，保留 null；没有足够证据进一步区分控制子类。log step 38 只写出含占位内容的生成模板，后续会话搜索没有已执行的重新保存记录，不能将其测试结果视为本次交付。",
+  "unmet": {
+   "serve_in_target": false
+  }
+ },
+ "summary_zh": "机器人在视频约 1.04 s 放开羽毛球后保持持拍站姿，球在身前下落，未建立拍球接触；合法发球入区为 0/1，原生裁判仅完成四阶段中的放球阶段。交付对应 log step 13 的 143 动作占位轨迹，其脚本只打开左手而未加入挥拍；后续会话轨迹搜索不属于本次交付。无状态日志且球体在远景中很小，精确触地时刻与原生速度、击球高度等指标无法核实。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "标注与视频、log 一致：视频 0–1.44 s 机器人持拍不动，约 1.04 s 后左手放出小球下落，球拍没有挥动；log step 13 保存的是只开左手的占位轨迹（143 动作，与 n_actions 一致），之后会话从未重新保存。断点、现象（没建立接触）、事件节点未达成与 contact_actual 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": "2026-10-31T00:57:15Z",
+  "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-badminton-serve-i00-privileged--mbmlydq/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-badminton-serve-i00-privileged--mbmlydq/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "1986223e4c4c8870549710fff95682f026ce9f748cbf7eedcdd754f88fc6134e",
+   "video_sha256": "c3e0adebe8b4c07d0dd6b338c601156bae469720e085e3bf1ce2082211b7275a",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/badminton_serve/trajectories/embodied-olympics-fun-badminton-serve-i00-privileged--xhnfjzn.json

@@ -0,0 +1,209 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-badminton-serve-i00-privileged--xhnfjzn",
+ "taskKey": "embodied-olympics-fun/badminton_serve",
+ "dag_sha256": "e5772140aff5f054a271f90fd295419869c755e083b6526601e08a99bee573e1",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "n_actions": 196,
+   "sim_steps": 196,
+   "horizon": 1500,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 4,
+   "stages_total": 4,
+   "fell": 0,
+   "replay_s": 5.36,
+   "bad_trajectory": 0,
+   "metric_hit_height_m": 0.9299,
+   "metric_landing_x_m": 5.45,
+   "metric_landing_y_m": 1.536,
+   "metric_shuttle_angle_deg": 43.21,
+   "metric_shuttle_speed_m_s": 10.455,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 714,
+   "agent_resets": 277,
+   "agent_replays": 19,
+   "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-badminton-serve-i00-privileged--xhnfjzn/replay.mp4",
+  "duration_s": 2.0,
+  "fps": 25.0,
+  "frames": 50,
+  "sha256": "3c0b8e25d30ea1f05c91b3b69e9b7babdbda64492e103a2f59df198542f84bd3"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "serve_in_target": {
+    "method": "visual",
+    "intervals": [
+     [
+      1.96,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00 s，机器人站在近侧发球区，右手持球拍，双脚着地；视频 0.00–0.52 s 调整持拍手臂，没有完成发球。"
+     },
+     {
+      "t": 0.64,
+      "log_step": null,
+      "text_zh": "观察：视频 0.56–0.72 s，机器人右臂带球拍向前上方挥动，白色羽毛球随后离开机器人向上飞行；视频 0.84–1.60 s，白球继续向球网另一侧飞行并下降。视频 1.64–1.96 s 球很小且逐渐受球网遮挡，机器人仍站立持拍；画面不能直接分辨首次触地。"
+     },
+     {
+      "t": 1.96,
+      "log_step": 432,
+      "text_zh": "观察：log step 430 明确对交付的 trajectory.npz 启动两次独立回放；log step 432 的基准回放裁判（仿真中的规则检查器）输出均为 success=true、failed=null、196 控制步、time_s=1.96，并具有相同末态哈希。其事件依次为 release（仿真 0.580 s，脱离接触起点仿真 0.550 s）、hit（仿真 0.612 s）、over the net（仿真 0.967 s，越网余量 0.918 m）、in 和 success（仿真 1.959 s）；落点为 (5.450, 1.536) m。inputs/run.json 的 reward 同样报告 success=1、replay_success=1、deterministic=1、stages_reached=4/4 及完全相同的五项原生指标。"
+     },
+     {
+      "t": 1.96,
+      "log_step": 432,
+      "text_zh": "推断：交付回放的 196 控制步对应仿真 1.96 s，即每步 0.01 s；视频以 25 fps 呈现该次短发球，视频 0.60–0.64 s 的挥拍、随后白球飞离与裁判仿真 0.612 s 的击球吻合。结合末帧视频 1.96 s，将仿真 1.959 s 的入区成功近似锚定在视频 1.96 s，事件此后锁存。首次落地及球托初触、毫秒级接触/脚步规则依据上述交付回放裁判与本次评分确认，并非从远景直接量得。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手持工具",
+      "how": "其他",
+      "note_zh": "候选：击打；现有方式没有瞬时击打，不能用推/扫或压/擦代替。观察：视频 0.56–0.72 s 机器人右手持拍前挥后白色羽毛球上升飞离。推断：结合 log step 432 的 release→hit 裁判事件，左手先放球，随后右手经球拍拍面击打羽球；持拍的握持链路不另计。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–0.64 s，机器人双脚在球场上承重并完成挥拍，随后至视频 1.96 s 仍以双脚站立。远景不能量出毫米级脚部位移或毫秒级离地间隙。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "hit_height_m",
+   "unit": "m",
+   "threshold": 1.15,
+   "direction": "lower_better",
+   "value": 0.9299
+  },
+  {
+   "name": "shuttle_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": 10.455
+  },
+  {
+   "name": "shuttle_angle_deg",
+   "unit": "deg",
+   "threshold": 0,
+   "direction": "higher_better",
+   "value": 43.21
+  },
+  {
+   "name": "landing_x_m",
+   "unit": "m",
+   "threshold": [
+    4.43,
+    9.15
+   ],
+   "direction": "target",
+   "value": 5.45
+  },
+  {
+   "name": "landing_y_m",
+   "unit": "m",
+   "threshold": [
+    0.43,
+    3.04
+   ],
+   "direction": "target",
+   "value": 1.536
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人完成一次放球、挥拍、越网并落入对角接发球区的合法发球，唯一事件节点在视频约 1.96 s 达成并锁存，最终为 1/1。交付文件的两次独立裁判回放与本次评分均成功，击球高度 0.9299 m、落点 (5.450, 1.536) m，失败标注为空。已逐帧检查完整视频及末段，并与成功参考画面对照；本次无状态日志，首次触地受球网遮挡，精细接触规则和入区结果以交付裁判及评分字段确认，达成时间为经挥拍画面核对后的近似视频锚点。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "单节点成功标注与视频、评分字段（4/4、196步=1.96 s）一致，接触与时钟无误，failure 为空合理。",
+    "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": "2026-10-31T00:48:33Z",
+  "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-badminton-serve-i00-privileged--xhnfjzn/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-badminton-serve-i00-privileged--xhnfjzn/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "30e35e84da177f88f02dff8e9a3f3fb911e0132a5133b74c892016473799e3dd",
+   "video_sha256": "3c0b8e25d30ea1f05c91b3b69e9b7babdbda64492e103a2f59df198542f84bd3",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/badminton_serve/trajectories/embodied-olympics-fun-badminton-serve-i00-privileged--xnwja8r.json

@@ -0,0 +1,231 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-badminton-serve-i00-privileged--xnwja8r",
+ "taskKey": "embodied-olympics-fun/badminton_serve",
+ "dag_sha256": "e5772140aff5f054a271f90fd295419869c755e083b6526601e08a99bee573e1",
+ "outcome": {
+  "success": 0,
+  "reward": 0.5,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "n_actions": 105,
+   "sim_steps": 105,
+   "horizon": 1500,
+   "episode_over": 1,
+   "progress": 0.5,
+   "stages_reached": 2,
+   "stages_total": 4,
+   "fell": 0,
+   "replay_s": 5.87,
+   "bad_trajectory": 0,
+   "metric_hit_height_m": 0.7667,
+   "metric_landing_x_m": 1.532,
+   "metric_landing_y_m": -0.057,
+   "metric_shuttle_angle_deg": 36.45,
+   "metric_shuttle_speed_m_s": 2.7081,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 129,
+   "agent_resets": 3063,
+   "agent_replays": 9,
+   "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": 1.157,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.26,
+   "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-badminton-serve-i00-privileged--xnwja8r/replay.mp4",
+  "duration_s": 1.08,
+  "fps": 25.0,
+  "frames": 27,
+  "sha256": "7c83250e8029c82e87e3dbbad245d5e3c75811f3fdc444ae24aba89ac80962c0"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "serve_in_target": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.44,
+      "log_step": 393,
+      "text_zh": "观察：视频 0.28–0.44 s，机器人左手附近的白色羽毛球下落，右手球拍从低处向前上方挥动；视频 0.48–0.64 s 羽毛球离开机器人向前短飞。log step 388 对最终交付文件重新生成并回放，log step 393 的基准裁判输出确认 release、hit 两阶段，击球高度 0.7667 m、出球速度 2.7081 m/s、仰角 36.45°，与评分器 reward 字段相同。"
+     },
+     {
+      "t": 1.04,
+      "log_step": 393,
+      "text_zh": "观察：逐帧检查视频 0.00–1.04 s；视频 0.84–1.04 s 白球始终在网前近侧并下降至接近地面，没有越网进入远侧淡黄色目标区。log step 393 的最终交付回放裁判判为 the service landed on your own side，落点 x=1.532 m、y=-0.057 m，只有 release、hit，没有 over the net、in；reward 的 success=0、deterministic=1，落点与该回放一致。"
+     },
+     {
+      "t": 1.04,
+      "log_step": null,
+      "text_zh": "推断：结合视频末段网前下降的飞行和交付回放裁判落点，球未到 x∈[4.43,9.15] m、y∈[0.43,3.04] m 的接发球区，完整合法发球节点全程未达成。成功参考视频 1.00–2.96 s 中白球可飞越网并进入远侧区域，只用于比较视野中的完成线索。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手持工具",
+      "how": "其他",
+      "note_zh": "观察：视频 0.40–0.48 s 机器人右手持球拍向前上方挥动，随后白色羽毛球向前飞出；log step 393 的交付回放裁判确认球拍击球。候选：击打；现有方式没有瞬时击打，故用其他。左手此前持球并张开放球，握拍链路不重复计。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–1.04 s 机器人双脚在场地原位支撑身体，放球和挥拍期间没有可见移步或跌倒。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "hit_height_m",
+   "unit": "m",
+   "threshold": 1.15,
+   "direction": "lower_better",
+   "value": 0.7667
+  },
+  {
+   "name": "shuttle_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": 2.7081
+  },
+  {
+   "name": "shuttle_angle_deg",
+   "unit": "deg",
+   "threshold": 0,
+   "direction": "higher_better",
+   "value": 36.45
+  },
+  {
+   "name": "landing_x_m",
+   "unit": "m",
+   "threshold": [
+    4.43,
+    9.15
+   ],
+   "direction": "target",
+   "value": 1.532
+  },
+  {
+   "name": "landing_y_m",
+   "unit": "m",
+   "threshold": [
+    0.43,
+    3.04
+   ],
+   "direction": "target",
+   "value": -0.057
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "serve_in_target",
+   "t": 0.44,
+   "log_step": 393,
+   "where": "操作中",
+   "evidence_zh": "推断：视频约 0.44 s 的挥拍把羽毛球送成短而低的飞行，视频 0.84–1.04 s 球仍在己方网前下降，之后没有修复机会。log step 388–393 对最终交付轨迹的回放明确判为己方落地，故断点是正在执行的合法发球入区步骤。"
+  },
+  "unmet": {
+   "serve_in_target": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 0.44,
+    "log_step": 393,
+    "evidence_zh": "观察：视频 0.48–1.04 s，机器人击出的白球只在己方网前短距离飞行并下落，没有进入对面接发球区；log step 393 的交付回放裁判记录落点为 (1.532,-0.057) m，判为己方落地。推断：发球飞行距离不足、落点偏短，属于击出后的目标位置未到达。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "log_step": 317,
+    "evidence_zh": "推断：agent 的交付生成脚本采用固定的后摆、左手放球和余弦前挥关节指令（log step 317），其击球动作产生的出球速度及方向不足以把球送过网到目标区。最终交付回放（log step 393）已确认合法 release、hit，但出球仅 2.7081 m/s、仰角 36.45°，落点仍在己方；视频 0.48–1.04 s 的低而短飞行支持挥拍控制未实现所需发球轨迹。现有证据不能进一步区分拍面接触几何、关节轨迹或具体参数问题。",
+    "source_zh": "日志 / 视频 / 评分器字段"
+   }
+  ],
+  "gap_zh": "没有状态日志。视频为 25 fps 远景，能看见挥拍、球飞出及末段网前下降，但无法独立核实球托初触、毫秒级接触时长、脚部微移和首次触地瞬间；精细合法性与落点采用与 reward 一致的最终交付回放裁判记录（log step 393）。未将日志中的仿真事件时刻直接写作视频时刻，断点仅取已检查的视频约 0.44 s；控制根因的具体子类证据不足。"
+ },
+ "summary_zh": "机器人完成放球和击球两个原生阶段，但视频中羽毛球仅在己方网前短飞后下落，合法发球入区节点未完成（0/1）。最终交付回放裁判与评分器均确认己方落地，断点为发球操作中的出球轨迹偏短，根因归于控制。无状态日志且视频为远景，精细接触与首次落点依赖交付回放裁判，不能细分控制子类。",
+ "conf": 7,
+ "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 serve_in_target、位姿不准现象与控制根因均有视频帧及 log step 393 的交付回放裁判（stages release/hit，落点 1.532,-0.057，己方落地）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": "2026-10-31T00:50:16Z",
+  "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-badminton-serve-i00-privileged--xnwja8r/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-badminton-serve-i00-privileged--xnwja8r/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "c5d50a1feaf655a116f1ba713df24b38e9c3a90d8cf934646e8ded4bfa9107b4",
+   "video_sha256": "7c83250e8029c82e87e3dbbad245d5e3c75811f3fdc444ae24aba89ac80962c0",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/badminton_serve/trajectories/embodied-olympics-fun-badminton-serve-i00-privileged--xx6kkxy.json

@@ -0,0 +1,211 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-badminton-serve-i00-privileged--xx6kkxy",
+ "taskKey": "embodied-olympics-fun/badminton_serve",
+ "dag_sha256": "e5772140aff5f054a271f90fd295419869c755e083b6526601e08a99bee573e1",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "n_actions": 569,
+   "sim_steps": 569,
+   "horizon": 1500,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 4,
+   "stages_total": 4,
+   "fell": 0,
+   "replay_s": 9.9,
+   "bad_trajectory": 0,
+   "metric_hit_height_m": 0.8135,
+   "metric_landing_x_m": 5.227,
+   "metric_landing_y_m": 1.457,
+   "metric_shuttle_angle_deg": 33.69,
+   "metric_shuttle_speed_m_s": 9.6682,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 211,
+   "agent_resets": 2606,
+   "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": 4.177,
+   "peak_processes": 26,
+   "peak_disk_gb": 0.244,
+   "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-badminton-serve-i00-privileged--xx6kkxy/replay.mp4",
+  "duration_s": 5.72,
+  "fps": 25.0,
+  "frames": 143,
+  "sha256": "c05026e52e74142cd03ce7dae8b5161600438c6d200712a37fef75ab26263d01"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "serve_in_target": {
+    "method": "visual",
+    "intervals": [
+     [
+      5.69,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 4.52,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–4.28 s，机器人站在近侧发球区，调整左臂持球位置及右手球拍姿态；视频 4.32–4.64 s 左手回撤，右臂挥拍，视频 4.68–4.88 s 白色羽毛球与球拍分离后向上、向球网方向飞行。已逐帧检查视频 4.28–4.92 s；接触瞬间的球托朝向与毫秒级时长不能由该远景直接分辨。"
+     },
+     {
+      "t": 4.92,
+      "log_step": 96,
+      "text_zh": "观察：视频 4.84–4.92 s，白色羽毛球到达网顶上方并飞向远侧；视频 4.96–5.68 s 可见其继续朝远侧对角区域下落，机器人仍站立且球拍随右手运动。log step 96 中，agent 生成并保存最终提交后执行两次官方回放；仿真裁判（基准内判定发球阶段的程序）的输出均为 success=True、failed=None、release→hit→over the net→in，且末态哈希相同，与评分器 reward 的 replay_success=1、deterministic=1 和四阶段通过一致。"
+     },
+     {
+      "t": 5.69,
+      "log_step": 96,
+      "text_zh": "推断：节点在视频约 5.69 s 达成并锁存。log step 95 的生成器按每行 0.01 s 控制，log step 96 对所交付文件的两次正式回放均为 569 控制步，即仿真 5.69 s，与视频 5.72 s 的长度及末帧视频 5.68 s 一致；交付输出的过网事件为仿真 4.894 s，已在视频 4.88–4.92 s 的球越网处核对该时钟对应关系。该输出在仿真 5.689 s 记录 in 和 success，评分器报告首次落点 (5.227,1.457) m，落在 DAG 目标区内，因此以视频约 5.69 s 为达成锚点；末帧视频 5.68 s 未清楚显示实际触地，不把落地瞬间称为直接视觉观察。"
+     },
+     {
+      "t": 5.69,
+      "log_step": 96,
+      "text_zh": "观察：inputs/run.json 的评分器字段给出 success=1、stages_reached=4、stages_total=4、fell=0，击球高度 0.8135 m、出球速度 9.6682 m/s、仰角 33.69°、落点 x=5.227 m、y=1.457 m；log step 96 的最终提交正式回放报告相同指标。放球脱离时长、球托初触、脚步微小位移及接触持续时间由裁判通过结果支持，远景视频不能逐项独立量测。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手持工具",
+      "how": "其他",
+      "note_zh": "候选：击打；视频 4.52–4.72 s 机器人右手所持球拍上挥，羽毛球随后上升飞离，log step 96 最终提交回放的裁判确认 hit。球拍拍面瞬时作用于羽毛球，没有适用的既有接触方式，故用其他；此前左手持球并在视频约 4.32–4.44 s 回撤放球，右手握拍链路不另重复计为抓握。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "视频 0.00–4.64 s 可见机器人双脚在近侧球场支撑身体完成准备及挥拍，视频 4.68–5.68 s 身体略有转动但仍保持站立；双脚的毫米级移动与极短离地无法由远景量测。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "hit_height_m",
+   "unit": "m",
+   "threshold": 1.15,
+   "direction": "lower_better",
+   "value": 0.8135
+  },
+  {
+   "name": "shuttle_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": 9.6682
+  },
+  {
+   "name": "shuttle_angle_deg",
+   "unit": "deg",
+   "threshold": 0,
+   "direction": "higher_better",
+   "value": 33.69
+  },
+  {
+   "name": "landing_x_m",
+   "unit": "m",
+   "threshold": [
+    4.43,
+    9.15
+   ],
+   "direction": "target",
+   "value": 5.227
+  },
+  {
+   "name": "landing_y_m",
+   "unit": "m",
+   "threshold": [
+    0.43,
+    3.04
+   ],
+   "direction": "target",
+   "value": 1.457
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人完成左手放球、右手持拍上挥和羽毛球越网飞向远侧对角区域；最终提交的裁判回放于视频约 5.69 s 确认首次落点入区，唯一节点“合法发球入区”达成，最终进度 1/1。评分器判成功且确定性复现，击球高度 0.8135 m、落点 (5.227,1.457) m 均符合相应阈值。无状态日志，远景无法独立核验毫秒级接触及脚步细节，末帧视频 5.68 s 也未清楚拍到触地瞬间，这些验收由交付回放的裁判事件和评分器指标支持。",
+ "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，唯一节点 serve_in_target 达成并以交付回放的 569 控制步（仿真 5.69 s）映射到视频约 5.69 s，已在越网事件处核对时钟；contact_actual（手持工具·其他·击打、腿/躯干·支撑）与评分器指标一致，无失败层，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": "2026-10-31T00:53:25Z",
+  "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-badminton-serve-i00-privileged--xx6kkxy/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-badminton-serve-i00-privileged--xx6kkxy/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "99670daa365d815cab578285f2ea0a5b83c0d31bb84a894444a85e8f429b87e0",
+   "video_sha256": "c05026e52e74142cd03ce7dae8b5161600438c6d200712a37fef75ab26263d01",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/badminton_serve/trajectories/embodied-olympics-fun-badminton-serve-i00-standard--5atxlb5.json

@@ -0,0 +1,229 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-badminton-serve-i00-standard--5atxlb5",
+ "taskKey": "embodied-olympics-fun/badminton_serve",
+ "dag_sha256": "e5772140aff5f054a271f90fd295419869c755e083b6526601e08a99bee573e1",
+ "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": 193,
+   "sim_steps": 1821,
+   "calls": 67,
+   "rows": 912,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 2525.0,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "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-badminton-serve-i00-standard--5atxlb5/replay.mp4",
+  "duration_s": 1.96,
+  "fps": 25.0,
+  "frames": 49,
+  "sha256": "f3a7ee814671016637ce8dc4e7ae43cb4b38c9955bc03a1ab4b4fab4ad319dbf"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "serve_in_target": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 1.6,
+      "log_step": 192,
+      "text_zh": "观察：视频 0.00–1.60 s，机器人双脚站在近侧场内，左臂将羽毛球举到身前，右手持拍调整并开始挥动；log step 192 的最后一次 reset 后，服务在准备结束时报告 holding_left=1.0、holding_right=1.0。"
+     },
+     {
+      "t": 1.92,
+      "log_step": 192,
+      "text_zh": "观察：逐帧检查视频 1.64–1.92 s，机器人左手撤回、右手球拍向上挥过，白色羽毛球仍在机器人身前下落，没有向球网及远侧目标区飞行；末帧机器人仍站立、右手仍持拍。log step 192 的第六局服务返回 done=True、stages_reached=1，裁判原文为“the racket's forward movement stopped before the hit”；log step 194、198 均确认最后一局已经失败且未再重置。"
+     },
+     {
+      "t": 1.92,
+      "log_step": 192,
+      "text_zh": "推断：本次评分局只完成放球，未完成击球、过网和合法首次落地，因此“合法发球入区”全程未达成；此判断结合可见挥空、球留在近侧及本局裁判结果，并非仅由 success=0 得出。成功参考视频 1.80–2.96 s 可见球离开机器人飞向远侧，该视觉结果在本次视频中没有出现。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "抓握",
+      "note_zh": "观察：视频 0.00–1.60 s，左手抓持羽毛球并抬臂定位，右手抓握球拍拍柄并调整拍姿，两臂各持一物配合准备发球；log step 192 的准备结束读数也显示两手 holding 均为 1.0。随后左手张开放球、右手持拍挥动，裁判未确认拍球击打，故不把计划中的手持工具击打记作实际接触。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–1.92 s，机器人双脚在球场地面支撑站立，发球动作中没有可见移步或跌倒；没有状态日志，不能核验毫米级脚部位移或短暂离地。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "hit_height_m",
+   "value": null,
+   "unit": "m",
+   "threshold": 1.15,
+   "direction": "lower_better"
+  },
+  {
+   "name": "shuttle_speed_m_s",
+   "value": null,
+   "unit": "m/s",
+   "threshold": null,
+   "direction": "higher_better"
+  },
+  {
+   "name": "shuttle_angle_deg",
+   "value": null,
+   "unit": "deg",
+   "threshold": 0,
+   "direction": "higher_better"
+  },
+  {
+   "name": "landing_x_m",
+   "value": null,
+   "unit": "m",
+   "threshold": [
+    4.43,
+    9.15
+   ],
+   "direction": "target"
+  },
+  {
+   "name": "landing_y_m",
+   "value": null,
+   "unit": "m",
+   "threshold": [
+    0.43,
+    3.04
+   ],
+   "direction": "target"
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "serve_in_target",
+   "t": 1.9,
+   "log_step": 192,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 1.64–1.92 s，机器人在放球后挥拍，羽毛球未被向网击出而在身前下落；log step 192 的最后一局裁判在仿真 1.93 s 报告“the racket's forward movement stopped before the hit”、stages_reached=1、done=True。失败发生在发球动作内，后续 log step 194 的重置被拒绝，本局没有修复。"
+  },
+  "unmet": {
+   "serve_in_target": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "动态时机",
+    "t": 1.9,
+    "log_step": 192,
+    "evidence_zh": "推断：视频 1.64–1.92 s 的右手上挥与左手放出的下落羽毛球没有形成击球时刻的相遇，球仍在身前而球拍已挥至较高处；log step 192 的本局裁判明确报告击球前前挥停止，阶段停在放球。这里是对运动中羽毛球的挥空，不另将同一事件重复标为没建立接触。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "log_step": 192,
+    "evidence_zh": "推断：log step 192 的实际执行脚本用预设右臂关节插值挥拍，并按固定行序打开左手、撤回左臂，挥拍段未依据羽毛球观测重新调整；同一步服务报告击球前前挥停止，视频 1.64–1.92 s 可见球与球拍错过。证据支持放球与挥拍的控制配合未实现，但不能区分接触几何、实际关节跟踪或具体时间参数哪项主导，故保留在“控制”。",
+    "source_zh": "日志"
+   }
+  ],
+  "gap_zh": "无状态日志；远景中羽毛球及手指很小，不能量出球拍与羽毛球的三维最近距离、毫秒级接触和实际速度，也不能分离挥拍轨迹误差与放球时刻误差。log step 192 的裁判结果属于最后评分局，可支持未击球的判断；前五局的脱拍、触身或落地记录均不作为本局现象。DAG 所列五项原生指标未在本次评分器 reward 中报告，均保留 null。"
+ },
+ "summary_zh": "本次评分是最后一次重置后的第六局，机器人站立持拍、举球并放球挥拍，但球未被击向球网，“合法发球入区”最终为 0/1。视频末段可见羽毛球仍在身前下落，log step 192 的本局裁判报告击球前球拍前挥停止，断点为发球操作中的动态时机失配，根因保留为控制。缺少状态日志和精细拍球运动数据，无法再区分轨迹、跟踪与放球时间参数的责任。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "节点未达成、断点 serve_in_target、末局（第六局）视频 1.96 s 与 log step 192（ready 1.2 s、挥拍结束 sim 1.93 s、stages=1、裁判“前挥在击球前停止”）一致；现象（动态时机）与根因（控制，未细分）有证据支持且标为推断，未发现需要改的内容。球在远景中极小，视频只能佐证未飞向球网。",
+    "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": "2026-10-31T00:56:57Z",
+  "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-badminton-serve-i00-standard--5atxlb5/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-badminton-serve-i00-standard--5atxlb5/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "149251bc83ea81eb77dc3f90d65721b4ad151bcda3d0bbfc9e94b7c904f4b476",
+   "video_sha256": "f3a7ee814671016637ce8dc4e7ae43cb4b38c9955bc03a1ab4b4fab4ad319dbf",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/badminton_serve/trajectories/embodied-olympics-fun-badminton-serve-i00-standard--cshxrw7.json

@@ -0,0 +1,230 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-badminton-serve-i00-standard--cshxrw7",
+ "taskKey": "embodied-olympics-fun/badminton_serve",
+ "dag_sha256": "e5772140aff5f054a271f90fd295419869c755e083b6526601e08a99bee573e1",
+ "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": 473,
+   "sim_steps": 2592,
+   "calls": 291,
+   "rows": 2588,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 1930.0,
+   "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": 1.172,
+   "peak_processes": 10,
+   "peak_disk_gb": 0.329,
+   "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-badminton-serve-i00-standard--cshxrw7/replay.mp4",
+  "duration_s": 4.76,
+  "fps": 25.0,
+  "frames": 119,
+  "sha256": "ed985af1cc064ce689ef4b809aade8ef86fa04a28249a1094ca235294193c172"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "serve_in_target": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": 88,
+      "text_zh": "观察：视频 0.00–1.60 s，机器人在近侧球场站立，左手持白色羽毛球调整位置，右手将球拍降到身体右侧；视频 2.00–4.32 s 继续调整左臂，双脚支撑地面。log step 88 开始最后一次重置后的尝试；log step 102 的服务状态确认最终为 episode 6，control_steps=473。"
+     },
+     {
+      "t": 4.48,
+      "log_step": 101,
+      "text_zh": "观察：视频 4.44–4.60 s，白色羽毛球由左手旁向下移动；视频 4.68–4.72 s，机器人右手球拍开始向前挥动，末帧羽毛球仍在机器人近旁，未见越网或在远侧淡黄色目标区落地。log step 101 的服务裁判返回 done=true、reason=failed、stages_reached=1/4，明确指出“the racket hit the feathers first, not the base of the shuttle”；这是服务裁判的击球接触判定，不是 agent 自述。"
+     },
+     {
+      "t": 4.72,
+      "log_step": 101,
+      "text_zh": "推断：本次只完成原生放球阶段，击球因先触羽毛被裁判终止，因此整个“合法发球入区”事件从未成立。log step 3 给出控制周期 0.01 s，log step 101 的末控制步 473 对应仿真 4.73 s，与视频 4.68–4.72 s 的挥拍及视频总长 4.76 s 一致；断点以已检查的视频 4.72 s 为近似锚点，未将 agent 工作时间换算为视频时间。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手持工具",
+      "how": "其他",
+      "note_zh": "观察：视频 4.68–4.72 s 机器人右手握持球拍前挥；log step 101 的服务裁判确认球拍实际先击中羽毛球的羽毛。候选：击打；现有方式没有击打，不能用推/扫或压/擦替代。此前视频 0.00–4.40 s 左手持球、随后张开放球；右手握拍链路不重复记接触。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–4.72 s 机器人双脚在近侧球场地面承重，调整双臂并挥拍时仍站立，未见身体倒地。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "hit_height_m",
+   "unit": "m",
+   "threshold": 1.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "shuttle_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "shuttle_angle_deg",
+   "unit": "deg",
+   "threshold": 0,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "landing_x_m",
+   "unit": "m",
+   "threshold": [
+    4.43,
+    9.15
+   ],
+   "direction": "target",
+   "value": null
+  },
+  {
+   "name": "landing_y_m",
+   "unit": "m",
+   "threshold": [
+    0.43,
+    3.04
+   ],
+   "direction": "target",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "serve_in_target",
+   "t": 4.72,
+   "log_step": 101,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 4.44–4.60 s 羽毛球从左手旁下落，视频 4.68–4.72 s 机器人挥动右手球拍；log step 101 中，服务裁判在本次第 473 控制步判失败，原因是球拍先触羽毛而非球托，仅验收放球阶段 1/4��合法发球在击球环节中断，未发生后续合法越网入区。"
+  },
+  "unmet": {
+   "serve_in_target": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 4.72,
+    "log_step": 101,
+    "evidence_zh": "推断：视频 4.68–4.72 s 的挥拍与 log step 101 的服务裁判“先触羽毛、未先触球托”相互对应，说明击球瞬间球拍与羽毛球的相对位姿未让球托成为首次接触部位。这里已发生击打接触，错误是击中部位；仅凭远景与该裁判结果，不能确认挥拍究竟过早还是过晚。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "t": 4.72,
+    "log_step": 101,
+    "source_zh": "日志",
+    "evidence_zh": "推断：log step 68 定义的脚本在检测左手松球后按延迟启动预设八步挥拍序列；log step 86–87 保留这一发球控制逻辑，log step 101 最后一次执行使用延迟参数 0.226 s，服务裁判仍确认先触羽毛。目标及放球后挥拍的顺序正确，但这次放球与挥拍的执行未实现球托先接触，支持归于控制；没有证据将其细分为抓取设计、轨迹跟踪或参数错误。"
+   }
+  ],
+  "gap_zh": "观察：inputs/ 中有本次视频、完整运行日志和成功参考视频，没有本次状态日志。视频为 640×360、25 fps 的场地远景，羽毛球及手指很小，不能直接分辨球托与羽毛的首次接触、毫秒级接触持续、瞬时速度；先触羽毛依据最后一次实时尝试的服务裁判记录。log step 104 中 agent 将球拍滑转、张手推偏羽毛球列为猜测，现有帧和日志未证实，故不据此细分控制根因；DAG 五个原生指标均未由本次评分器输出，记为 null。"
+ },
+ "summary_zh": "本次评分对应最后的第六次发球：机器人站立调整持球及持拍姿态，放球后挥拍，但合法发球入区节点始终未完成，最终为 0/1。视频 4.68–4.72 s 的挥拍对应 log step 101 的服务裁判失败：球拍先触羽毛而非球托，仅完成原生放球阶段 1/4，归为击球位姿不准及控制问题。远景和缺少状态日志使球拍是否滑转、放球是否被手指推偏及具体时序误差无法区分。",
+ "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 serve_in_target 在击球环节、时钟（控制步 473≈4.73 s，release 4.41 s、挥拍 4.64 s）与服务裁判记录一致；现象位姿不准与根因控制有日志依据，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": "2026-10-31T00:58:26Z",
+  "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-badminton-serve-i00-standard--cshxrw7/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-badminton-serve-i00-standard--cshxrw7/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "70d9a0f66e14231fd07bdf0ae2c0ff2416f6f512b6f1d0bb94bde8bd5a12b354",
+   "video_sha256": "ed985af1cc064ce689ef4b809aade8ef86fa04a28249a1094ca235294193c172",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/badminton_serve/trajectories/embodied-olympics-fun-badminton-serve-i00-standard--lb6jxal.json

@@ -0,0 +1,231 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-badminton-serve-i00-standard--lb6jxal",
+ "taskKey": "embodied-olympics-fun/badminton_serve",
+ "dag_sha256": "e5772140aff5f054a271f90fd295419869c755e083b6526601e08a99bee573e1",
+ "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": 132,
+   "sim_steps": 7146,
+   "calls": 106,
+   "rows": 7146,
+   "resets": 50,
+   "reset_budget": 50,
+   "sim_startups": 1,
+   "elapsed_s": 6931.4,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 1.215,
+   "peak_processes": 10,
+   "peak_disk_gb": 0.42,
+   "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-badminton-serve-i00-standard--lb6jxal/replay.mp4",
+  "duration_s": 1.36,
+  "fps": 25.0,
+  "frames": 34,
+  "sha256": "f2be0da22fc01e405a88986828a675cb5cbbc2425cadc7caa33d18d0eaeb9304"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "serve_in_target": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0,
+      "log_step": 867,
+      "text_zh": "观察：视频 0.00–1.20 s，机器人站在近侧发球区，双臂由初始持球、持拍姿态移至准备姿态；视频 1.24–1.32 s 手臂开始变化，但没有羽毛球向远侧飞行、越网并落入接发区的过程。log step 867 的最终脚本先调整双臂，再打开左手、挥动右臂。"
+     },
+     {
+      "t": 1.32,
+      "log_step": 869,
+      "text_zh": "观察：log step 869 是最后一次实时执行，reset 后为 episode 51，服务返回 executed_rows=132、stages_reached=1 of 4、reason=failed，裁判原文为“the shuttle touched you after you let it go”（放球后羽毛球碰到机器人）。该裁判反馈由机器人服务返回，不是 agent 的完成自述；log step 870 确认同一 episode 已结束且 control_steps=132。"
+     },
+     {
+      "t": 1.32,
+      "log_step": 870,
+      "text_zh": "推断：log step 7 的 control_dt=0.01 s，最终 episode 的 132 个控制步对应视频约 1.32 s；已与视频 1.28–1.32 s 的末段手臂动作及终止画面核对，视频共 1.36 s（末帧为视频 1.32 s）。原生阶段仅确认放球且未出现合法越网落地，故完整发球事件从未达到。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "抓握",
+      "note_zh": "观察：视频 0.00–1.20 s，左手持羽毛球、右手持球拍，双臂共同调整至发球准备姿态；随后左手张开放球、右臂启动挥拍（log step 867、869）。此条记录准备阶段两手各持一物的配合，不据此认定球拍击中了球。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–1.32 s，机器人双脚支撑球场地面并保持站立，未见跌倒；远景不用于验证厘米级脚步阈值。"
+     },
+     {
+      "who": "未知",
+      "how": "其他",
+      "note_zh": "观察：log step 869 的本次裁判确认放球后羽毛球再次碰到机器人。候选：碰撞；具体身体接触部位及接触方式在视频 1.24–1.32 s 的远景中无法辨认，不能归为抓握、推/扫或压/擦。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "hit_height_m",
+   "unit": "m",
+   "threshold": 1.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "shuttle_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "shuttle_angle_deg",
+   "unit": "deg",
+   "threshold": 0,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "landing_x_m",
+   "unit": "m",
+   "threshold": [
+    4.43,
+    9.15
+   ],
+   "direction": "target",
+   "value": null
+  },
+  {
+   "name": "landing_y_m",
+   "unit": "m",
+   "threshold": [
+    0.43,
+    3.04
+   ],
+   "direction": "target",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "serve_in_target",
+   "t": 1.32,
+   "log_step": 869,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 1.24–1.32 s 机器人仍在原地进行放球、挥拍动作，未将球送至远侧；log step 869 的最终 episode 51 在执行 132 行后由裁判以“the shuttle touched you after you let it go”终止，仅确认 1/4 原生阶段。推断：断点是正在执行的合法发球入区，导航锚点取视频约 1.32 s，而不是前面重置测试中的失误。"
+  },
+  "unmet": {
+   "serve_in_target": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "被挡/碰撞",
+    "t": 1.32,
+    "log_step": 869,
+    "evidence_zh": "观察：log step 869 中机器人服务的最终裁判明确报告，羽毛球已被放开，随后碰到机器人，立即判负；视频 1.24–1.32 s 球仍处于机器人手臂附近，未出现远侧飞行。推断：此次放球、挥拍中的球与机器人发生了非预期碰撞；不把这个事件另记为违反约束，也不声称看清了具体指节。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "source_zh": "日志",
+    "evidence_zh": "观察：log step 867 的最终控制脚本在左手打开后仍以 A3[:,3:10]=q_left 保持左臂姿态，右臂按固定延迟和插值挥拍；log step 869 的同次执行反馈是放球后碰身、只确认放球阶段。推断：这套双臂放球与挥拍控制未实现羽毛球脱离身体的必要间隙，导致发球中断。没有本次逐步接触与跟踪数据，不能进一步确定是具体接触几何、实际跟踪偏差还是时序参数问题。"
+   }
+  ],
+  "gap_zh": "本目录没有状态日志；视频只有远景，已逐帧检查视频 0.00–1.32 s，尤其视频 1.00–1.32 s 的放大图，仍不能分辨羽毛球再接触的机器人部位、球拍是否短暂触球及毫秒级接触顺序。最终服务裁判足以确认放球后碰身失败，但不报告五项原生指标，故这些值均为 null；根因仅归至控制，不指定子类。log step 876 自述已击球且避免干涉，与 log step 869 的最终裁判结果不符，不能作为本次合法击球的证据。"
+ },
+ "summary_zh": "最后一次实时尝试（episode 51）中，机器人完成持球、持拍准备并放球，但在视频约 1.32 s 因羽毛球放开后碰到机器人而被裁判终止，合法发球入区未达到，最终为 0/1。最终脚本的双臂放球、挥拍控制未保证球与身体分离；远景及缺少状态日志使具体碰撞部位、瞬时球拍接触和控制子类无法核实。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "最终 episode 51 的视频（1.36 s，132 控制步，dt=0.01）与 log step 867/869/870 一致：放球后羽毛球碰到机器人，服务裁判终止，仅 1/4 阶段；serve_in_target 未达成、断点、接触与现象/根因均有证据支持，未发现需要修改的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:55:25Z",
+  "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-badminton-serve-i00-standard--lb6jxal/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-badminton-serve-i00-standard--lb6jxal/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "4768ab70d065edf7e127ee959a29ea937c92116a4a6a3fa53371f73979dbe313",
+   "video_sha256": "f2be0da22fc01e405a88986828a675cb5cbbc2425cadc7caa33d18d0eaeb9304",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/badminton_serve/trajectories/embodied-olympics-fun-badminton-serve-i00-standard--mvet3aa.json

@@ -0,0 +1,221 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-badminton-serve-i00-standard--mvet3aa",
+ "taskKey": "embodied-olympics-fun/badminton_serve",
+ "dag_sha256": "e5772140aff5f054a271f90fd295419869c755e083b6526601e08a99bee573e1",
+ "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": 783,
+   "sim_steps": 24174,
+   "calls": 98,
+   "rows": 24174,
+   "resets": 42,
+   "reset_budget": 50,
+   "sim_startups": 1,
+   "elapsed_s": 7028.4,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 1.155,
+   "peak_processes": 11,
+   "peak_disk_gb": 0.887,
+   "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-badminton-serve-i00-standard--mvet3aa/replay.mp4",
+  "duration_s": 7.84,
+  "fps": 25.0,
+  "frames": 196,
+  "sha256": "1071b66a3e94c2c00d523b6e86ca372ae2afbd4a8691301c7fedaae2f33dd86b"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "serve_in_target": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–6.80 s，机器人在近侧发球区双脚站立，左手持白色羽毛球调整放球姿态，右手握球拍逐渐收拍准备；尚无球飞向远侧接发球区。"
+     },
+     {
+      "t": 7.76,
+      "log_step": 366,
+      "text_zh": "观察：视频 6.80–7.80 s 已逐帧检查，机器人保持站立，右手球拍在视频约 7.76–7.80 s 快速挥动，未见羽毛球形成越网飞向远场的轨迹。log step 366 是最后一次重置后的实时提交，服务返回 executed_rows=783、done=True、outcome=\"the racket's forward movement stopped before the hit\"；reward.n_actions 同为 783，graded_by_live=1。该服务结果明确为击球前前挥停止，合法发球入区未完成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "抓握",
+      "note_zh": "观察：视频 0.00–7.60 s，左手包握羽毛球并调整放球姿态，右手握住球拍拍柄同时准备挥拍；随后左手放球、右手挥拍。log step 366 的服务结果未记录到击球，因此不将球拍与羽毛球的击打作为已发生接触；这里记录实际双手持物配合。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–7.80 s，机器人双脚在球场地面支撑身体，调整双臂及末段挥拍期间仍保持站立。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "hit_height_m",
+   "unit": "m",
+   "threshold": 1.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "shuttle_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "shuttle_angle_deg",
+   "unit": "deg",
+   "threshold": 0,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "landing_x_m",
+   "unit": "m",
+   "threshold": [
+    4.43,
+    9.15
+   ],
+   "direction": "target",
+   "value": null
+  },
+  {
+   "name": "landing_y_m",
+   "unit": "m",
+   "threshold": [
+    0.43,
+    3.04
+   ],
+   "direction": "target",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "serve_in_target",
+   "t": 7.76,
+   "log_step": 366,
+   "where": "操作中",
+   "evidence_zh": "观察：视频约 7.76–7.80 s，机器人正在持拍挥动，未出现球飞向对角接发球区的后续过程；最后一次实时发球提交在 log step 366 返回“the racket's forward movement stopped before the hit”，done=True。推断：此次已启动发球操作，但前挥停止时尚未击中羽毛球，断点为合法发球入区。"
+  },
+  "unmet": {
+   "serve_in_target": false
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没建立接触",
+    "t": 7.76,
+    "log_step": 366,
+    "evidence_zh": "观察：视频 7.48–7.80 s 的近身裁剪中可见右手球拍末段挥动，没有羽毛球被击出并越网的飞行；log step 366 的本次服务终止反馈明确为前挥在击球前停止。推断：本次发球所需的球拍—羽毛球击打接触尚未建立；不将同一失误另标为动态时机，因为现有证据不能区分时间偏差与空间偏差���"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "evidence_zh": "观察：log step 366 中 agent 的脚本修改左肩横向姿态并将 I_REL 从 11 改为 17，重建固定关节动作序列后重置、以 dt=0.01 提交执行；本次服务仍返回前挥在击球前停止。视频约 7.76–7.80 s 确有挥拍而无击出飞行。推断：放球与挥拍的运动配合未使球拍在有效前挥期间接到球，归为控制；没有本次精细轨迹与接触测量，不进一步指定参数调试、逆解或反馈控制。",
+    "source_zh": "日志 / 视频"
+   }
+  ],
+  "gap_zh": "无状态日志；远景中羽毛球及手指很小，无法仅凭帧图确定精确脱手时刻、拍球最近距离及毫秒级接触。未击球判断由本次最后提交的服务反馈与视频共同支持；log step 365 的几何测算属于前一次 v18 调试，不能当成本次 v19 的实际轨迹，因此具体空间偏差、时间偏差及控制子类仍不确定。评分器未报告 DAG 所列五项原生指标，均填 null。"
+ },
+ "summary_zh": "本次评分为 log step 366 重置后最后一次实时发球：机器人完成持球调整并挥拍，但合法发球入区节点始终未达到，最终 0/1。视频末段未见球被击出越网，服务明确反馈前挥在击球前停止；失败表现为没建立击打接触，根因保留为控制配合失败。无状态日志且球体很小，无法确定具体失配方向或精确释放时刻。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "标签与视频、log step 366 的服务反馈一致：合法发球入区未达到，断点、没建立接触、控制根因均有证据支持，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": "2026-10-31T00:53:26Z",
+  "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-badminton-serve-i00-standard--mvet3aa/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-badminton-serve-i00-standard--mvet3aa/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "eae716d05a914dc69c7079299310c13d1704b53de1e963b337cde1f00cbb9c8f",
+   "video_sha256": "1071b66a3e94c2c00d523b6e86ca372ae2afbd4a8691301c7fedaae2f33dd86b",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/badminton_serve/trajectories/embodied-olympics-fun-badminton-serve-i00-standard--sahqtl6.json

@@ -0,0 +1,196 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-badminton-serve-i00-standard--sahqtl6",
+ "taskKey": "embodied-olympics-fun/badminton_serve",
+ "dag_sha256": "e5772140aff5f054a271f90fd295419869c755e083b6526601e08a99bee573e1",
+ "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": 1,
+   "sim_steps": 1,
+   "calls": 31,
+   "rows": 1,
+   "resets": 0,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 1795.3,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 4,
+   "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-badminton-serve-i00-standard--sahqtl6/replay.mp4",
+  "duration_s": 0.04,
+  "fps": 25.0,
+  "frames": 1,
+  "sha256": "7f28443388c68edd77037540350bf291bc019550e5b1cbe4705cd6e929f008a5"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "serve_in_target": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：已查看本次完整视频，视频仅有 0.00 s 一帧（总长 0.04 s）；机器人直立于近侧球场并在右手处持拍，画面未显示发球、羽毛球越网或落入远侧对角区。"
+     },
+     {
+      "t": null,
+      "log_step": 57,
+      "text_zh": "观察：log step 57 的脚本只向实时机器人服务提交一行初始 stand 指令，用于读取站姿力矩；服务返回仿真 0.01 s、done=False、holding_left=1.0、holding_right=1.0。此后日志只有本地逆运动学等计算，没有继续提交发球控制；评分器记录 n_actions=1、sim_steps=1、resets=0、live_success=0，未形成合法发球入区事件。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "hit_height_m",
+   "unit": "m",
+   "threshold": 1.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "shuttle_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "shuttle_angle_deg",
+   "unit": "deg",
+   "threshold": 0,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "landing_x_m",
+   "unit": "m",
+   "threshold": [
+    4.43,
+    9.15
+   ],
+   "direction": "target",
+   "value": null
+  },
+  {
+   "name": "landing_y_m",
+   "unit": "m",
+   "threshold": [
+    0.43,
+    3.04
+   ],
+   "direction": "target",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 79,
+   "where": "首步前",
+   "evidence_zh": "观察：log step 79 明确返回 Gemini 模型过滤拦截消息；此前唯一实时控制在 log step 57 只维持初始站姿，服务返回 done=False，随后未提交其他控制。视频 0.00 s 唯一帧也未显示发球动作。推断：评分尝试尚未结束或形成已确定的物理失败，便被提供方过滤中断；依 CASES.md FAIL-5 与 DECISIONS.md D37，按外因中断处理，不能强选机器人失败步骤。"
+  },
+  "unmet": {
+   "serve_in_target": false
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：inputs/ 中有本次单帧视频与完整会话日志，无状态日志；log step 57 服务仍返回 done=False，log step 79 是提供方过滤拦截，没有已完成发球的交付或裁判事件记录。推断：这是结果确定前的外因中断，应重试并排除出机器人失败分析；不能将未启动发球标成“没尝试”，也没有证据支持任何机器人根因。单帧无法核实瞬时接触过程，评分器未报告击球高度、速度、角度或落点指标。"
+ },
+ "summary_zh": "本次记录的合法发球入区节点未达成，终态进度为 0/1；视频 0.00 s 唯一帧仅显示机器人持拍直立，log step 57 只执行了一行初始站姿诊断，服务仍返回 done=False。log step 79 随后发生 Gemini 提供方过滤拦截，属于评分结果确定前的外因中断，断点记 unknown／首步前，现象与根因留空并应重试。视频仅一帧且无状态日志，不能核实精细接触或原生击球、落点指标；本地仿真与逆解计算不计为本次节点进度。",
+ "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：视频仅一帧，节点未达成；整个会话只提交了一次 10 ms stand 指令（sim_total.submit=1），log step 79 为 Gemini 过滤拦截，发球结果未定，按外因中断（FAIL-5）处理，断点 unknown/首步前、两层为空且有 gap_zh，合理。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:56:41Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-badminton-serve-i00-standard--sahqtl6/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-badminton-serve-i00-standard--sahqtl6/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "c7c89b3fbc262a9b97df65c064e5a24a7b14a5e228c9813c6437fe6c3906def5",
+   "video_sha256": "7f28443388c68edd77037540350bf291bc019550e5b1cbe4705cd6e929f008a5",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/badminton_serve/trajectories/embodied-olympics-fun-badminton-serve-i00-standard--xsd7hrj.json

@@ -0,0 +1,214 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-badminton-serve-i00-standard--xsd7hrj",
+ "taskKey": "embodied-olympics-fun/badminton_serve",
+ "dag_sha256": "e5772140aff5f054a271f90fd295419869c755e083b6526601e08a99bee573e1",
+ "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": 494,
+   "sim_steps": 494,
+   "calls": 13,
+   "rows": 17,
+   "resets": 0,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 407.1,
+   "video_rendered": 1,
+   "first_success": 1,
+   "first_reason_code": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.311,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.868,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-badminton-serve-i00-standard--xsd7hrj/replay.mp4",
+  "duration_s": 4.96,
+  "fps": 25.0,
+  "frames": 124,
+  "sha256": "fbc297911e7d048129e60ec52deba503f1a12a8ec8b34ee7a4cd45810d0e2fd0"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "serve_in_target": {
+    "method": "visual",
+    "intervals": [
+     [
+      4.94,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–3.48 s，机器人站在近侧发球区，持球、持拍调整双臂姿态，尚未出现球飞向对场的发球结果；双脚持续在场地上支撑。"
+     },
+     {
+      "t": 3.72,
+      "log_step": 35,
+      "text_zh": "观察：逐帧检查视频 3.40–3.96 s，白色羽毛球先由左手附近下落，机器人在视频约 3.72 s 右臂挥拍，视频 3.76 s 白球已离开手拍附近并升高，随后机器人维持随挥姿态；log step 35 的实时服务记录执行击球指令后 control_steps=371、episode=1，log step 37 为 control_steps=373，右手持物传感值仍为 1。"
+     },
+     {
+      "t": 4.04,
+      "log_step": null,
+      "text_zh": "观察：视频 3.96–4.12 s 可见微小白球继续向网后飞行，视频 4.04 s 白球投影在网顶上方；已逐帧检查视频 3.96–4.92 s，后段白球难以与背景分辨，末帧未清楚显示首次触地接触点。"
+     },
+     {
+      "t": 4.92,
+      "log_step": 40,
+      "text_zh": "观察：log step 40 中，机器人实时服务（含本次发球的程序裁判）返回 episode=1、control_steps=494、sim_time_s=4.94、done=true、reason=success、outcome=success、stages_reached=4、stages_total=4；inputs/run.json 的 reward 同时记录 success=1、graded_by_live=1、resets=0。该结果来自本次唯一实时评分局，不是 agent 自述或另一次测试。"
+     },
+     {
+      "t": 4.94,
+      "log_step": 40,
+      "text_zh": "推断：log step 7 的服务规格给出 control_dt=0.01 s，以本次评分局控制步数映射视频时钟；log step 35–37 的 control_steps=371–373 对应视频 3.71–3.73 s，已与视频约 3.72 s 的挥拍核对，494 个控制步也与视频总长 4.96 s 相符。因此将实时裁判成功时刻 494×0.01=4.94 s 标为视频约 4.94 s 达成“合法发球入区”，之后事件锁存；该时刻比最后采样帧视频 4.92 s 晚约 0.02 s，合法落地及精细击球条件依据裁判验收，未声称画面直接拍到触地。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手持工具",
+      "how": "其他",
+      "note_zh": "候选：击打。视频约 3.72–3.76 s 机器人右手所持球拍挥向下落羽毛球，随后球上升离开；结合 log step 40 的四阶段合法发球成功，判定球拍实际击中羽毛球。左手此前持球并先行松开；右手握拍链路不重复标注。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "视频 0.00–4.92 s 机器人双脚支撑球场地面，调整双臂、挥拍和随挥时均未见移步或倒地。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "hit_height_m",
+   "unit": "m",
+   "threshold": 1.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "shuttle_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "shuttle_angle_deg",
+   "unit": "deg",
+   "threshold": 0,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "landing_x_m",
+   "unit": "m",
+   "threshold": [
+    4.43,
+    9.15
+   ],
+   "direction": "target",
+   "value": null
+  },
+  {
+   "name": "landing_y_m",
+   "unit": "m",
+   "threshold": [
+    0.43,
+    3.04
+   ],
+   "direction": "target",
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "观察：机器人完成左手放球、右手挥拍及羽毛球向网后飞行，log step 40 的本次实时裁判确认四阶段全部成功，唯一节点“合法发球入区”在视频约 4.94 s 达成，最终 1/1。观察：本次与参考视频均为远景，无状态日志；首次触地、球托初触及毫秒级接触条件无法由画面独立核实，故以本次裁判验收补足，五项原生指标未报告数值，均保留 null。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "单节点成功标注与证据一致：log step 40 的实时裁判 done=true、success、4/4、sim_time_s=4.94、control_steps=494，时钟映射(0.01 s/步)与 step 35/37 的 371/373 步对应视频约3.72 s 核对无误；contact_actual 合理，无失败层，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": "2026-10-31T01:01:05Z",
+  "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-badminton-serve-i00-standard--xsd7hrj/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-badminton-serve-i00-standard--xsd7hrj/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "6aa96cc29d0cb45c9beb63f3a3a82ab8f8f64c2e1424186fb953aa7d2944ca6e",
+   "video_sha256": "fbc297911e7d048129e60ec52deba503f1a12a8ec8b34ee7a4cd45810d0e2fd0",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/badminton_serve/trajectories/embodied-olympics-fun-badminton-serve-i00-standard--zyb6fty.json

@@ -0,0 +1,226 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-badminton-serve-i00-standard--zyb6fty",
+ "taskKey": "embodied-olympics-fun/badminton_serve",
+ "dag_sha256": "e5772140aff5f054a271f90fd295419869c755e083b6526601e08a99bee573e1",
+ "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": 218,
+   "sim_steps": 8015,
+   "calls": 350,
+   "rows": 7663,
+   "resets": 37,
+   "reset_budget": 50,
+   "sim_startups": 1,
+   "elapsed_s": 7060.3,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 1.351,
+   "peak_processes": 10,
+   "peak_disk_gb": 0.955,
+   "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-badminton-serve-i00-standard--zyb6fty/replay.mp4",
+  "duration_s": 2.2,
+  "fps": 25.0,
+  "frames": 55,
+  "sha256": "520ba939beaa65327f4820bb6aef5fc8ff32fc4d8db4aacdb54e8617dd267236"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "serve_in_target": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–2.16 s 的远景中，机器人双脚支撑在近侧发球区，先调整双臂和球拍，末段右手持拍向左手附近挥动；未见羽毛球飞越球网进入远侧目标区。成功参考视频 1.72–2.96 s 可见球离开机器人向远侧飞行，参考只用于识别该视觉线索。"
+     },
+     {
+      "t": 2.16,
+      "log_step": 460,
+      "text_zh": "观察：log step 459 启动最后一次 reset 发球，log step 460 返回 episode 38，末次 release/stroke 的裁判 outcome 为“the racket hit the feathers first, not the base of the shuttle”，stages=1、done=true；这是机器人服务返回的裁判判定，非 agent 自述。视频 2.12–2.16 s 可见末段挥拍，但分辨率不足以辨别羽毛与球托的首次接触。"
+     },
+     {
+      "t": 2.16,
+      "log_step": 461,
+      "text_zh": "推断：log step 461 的末次 episode 38 状态为 failed、control_steps=218，与评分 reward.n_actions=218 相同；服务 control_dt=0.01 s（log step 25），218 个控制步对应仿真 2.18 s，与视频 2.16 s 末帧及 2.20 s 总长吻合，并已核对视频 2.12–2.16 s 的挥拍末段。故本次评分局为最后一次发球，已放球但首次击球不合法，合法发球入区节点从未成立；会话中早先 reset 的结果不计入。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手持工具",
+      "how": "其他",
+      "note_zh": "观察：视频 0.00–2.16 s 机器人右手持球拍调整并挥动；log step 460 的末次裁判确认球拍实际击中羽毛球羽毛。候选：击打；现有方式无击打，不能以推/扫或压/擦代替。左手此前持球后张开放球，属先后动作；右手握拍链路不重复标注。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–2.16 s 机器人双脚持续支撑场地，发球调整和挥拍期间未见跌倒或明显移步；不据远景判断毫米级脚步阈值。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "hit_height_m",
+   "unit": "m",
+   "threshold": 1.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "shuttle_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "shuttle_angle_deg",
+   "unit": "deg",
+   "threshold": 0,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "landing_x_m",
+   "unit": "m",
+   "threshold": [
+    4.43,
+    9.15
+   ],
+   "direction": "target",
+   "value": null
+  },
+  {
+   "name": "landing_y_m",
+   "unit": "m",
+   "threshold": [
+    0.43,
+    3.04
+   ],
+   "direction": "target",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "serve_in_target",
+   "t": 2.16,
+   "log_step": 460,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 2.08–2.16 s 机器人右臂挥拍靠近放球位置，未出现随后越网飞行；最后一次评分局的服务裁判在 log step 460 返回“先击中羽毛而非球托”、stages=1、done=true，log step 461 确认 episode 38 已失败。断在正在执行的合法发球��区步骤，视频 2.16 s 为末段击球的近似导航锚点。"
+  },
+  "unmet": {
+   "serve_in_target": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 2.16,
+    "log_step": 460,
+    "evidence_zh": "观察：视频 2.08–2.16 s 可见机器人挥拍至左手附近，log step 460 的本次裁判明确判定球拍先接触羽毛、没有球托先触拍。推断：球拍与羽毛球的相对击球位置或朝向未对准要求的球托，属于击球部位偏差；没有证据进一步断言是挥拍过早或过晚。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "evidence_zh": "推断：agent 在 log step 457 自述要恢复放球对齐并小幅调整手腕以提高击球方向，随后 log step 459–460 实际执行发球，裁判仍检出羽毛先触拍；视频 2.08–2.16 s 也显示机器人已进入挥拍动作。目标与放球后击球的顺序明确，但实际挥拍未实现球托先接触的控制目标，因此归为控制；现有证据不足以区分接触设计、轨迹跟踪或具体参数问题。",
+    "source_zh": "日志 / 视频"
+   }
+  ],
+  "gap_zh": "本次及成功参考均无状态日志；视频远景只有 640×360、25 fps，不能直接分辨毫秒级球托/羽毛初触或重建接触点，首次接触部位依据已核对为最后评分局的服务裁判返回。缺少最终控制脚本完整修改内容与球拍、羽毛球接触轨迹，控制根因不能细分；DAG 的五项原生指标在评分 reward 中均未报告，value 均为 null。"
+ },
+ "summary_zh": "本次评分为最后的 episode 38：机器人完成发球准备并放球挥拍，但合法发球入区节点未达到，最终进度 0/1。视频 2.08–2.16 s 的末段挥拍与 log step 460 的裁判结果对应，球拍先击中羽毛而非球托，断点为操作中的合法发球入区，归为击球位姿不准及控制问题。远景无法分辨精细接触，控制子类与原生指标缺乏证据；早先 reset 的测试进展不计入本次评分。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "评分局为最后的 episode 38（control_steps=218，约 2.18 s，与视频 2.2 s 吻合）；服务裁判返回球拍先击中羽毛，serve_in_target 未达成，断点、位姿不准与控制根因均有证据支持，与先例一致，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": "2026-10-31T01:01:43Z",
+  "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-badminton-serve-i00-standard--zyb6fty/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-badminton-serve-i00-standard--zyb6fty/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "523e1c0c1167c57668c0e29c65f3f980354e2437ce3b1a34f86cb9411357540d",
+   "video_sha256": "520ba939beaa65327f4820bb6aef5fc8ff32fc4d8db4aacdb54e8617dd267236",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/boxing/trajectories/embodied-olympics-fun-boxing-i00-privileged--ebemxbr.json

@@ -0,0 +1,222 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-boxing-i00-privileged--ebemxbr",
+ "taskKey": "embodied-olympics-fun/boxing",
+ "dag_sha256": "5fe95f4fd88e6dbc49fa1e5ba7230e5fd61571eeb504fd1c0835f0ca7840e98d",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "n_actions": 224,
+   "sim_steps": 224,
+   "horizon": 1500,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 2,
+   "stages_total": 2,
+   "fell": 0,
+   "replay_s": 3.1,
+   "bad_trajectory": 0,
+   "metric_contact_impulse_n_s": 20.64,
+   "metric_contact_s": 0.049,
+   "metric_hit_height_m": 1.419,
+   "metric_impulse_n_s": 16.06,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 44,
+   "agent_resets": 7439,
+   "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": 1.188,
+   "peak_processes": 13,
+   "peak_disk_gb": 0.846,
+   "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-boxing-i00-privileged--ebemxbr/replay.mp4",
+  "duration_s": 2.28,
+  "fps": 25.0,
+  "frames": 57,
+  "sha256": "78fb7ce7961dfdc5cd0d2f6acf06541f8b0dedc21fd759ae4bcf7788bf539368"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "bag_struck": {
+    "method": "visual",
+    "intervals": [
+     [
+      0.24,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.12,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–0.08 s 机器人右臂抬起向沙袋伸出；视频 0.12–0.20 s 右拳末端抵达红色沙袋白色环带，沙袋开始向远离机器人的方向摆动。视频 0.24–0.28 s 右拳明显收回并与沙袋分离，未见左手击袋。"
+     },
+     {
+      "t": 0.24,
+      "log_step": 216,
+      "text_zh": "观察：log step 216 的最终交付验证结果记录两个新进程重放均成功、224 个动作行完全相同、终态哈希一致；裁判在仿真 0.213 s 记录 strike，impulse_n_s=16.06、contact_s=0.049、hit_height_m=1.419，与 inputs/run.json 的最终评分字段一致。"
+     },
+     {
+      "t": 0.24,
+      "log_step": 216,
+      "text_zh": "推断：结合视频 0.12–0.28 s 的一次触袋、摆离和收拳，以及最终交付裁判的有效击打记录，击中沙袋节点已完成；以视频 0.24 s 的分离帧作近似完成锚点。精确冲量及拳峰接触区域依据裁判验收，不能由画面独立测量。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "其他",
+      "note_zh": "观察：视频 0.12–0.20 s 机器人右手末端拳垫击中沙袋白带，随后迅速撤回；候选接触名“击打”，现有接触方式没有短促冲击的对应值，故用其他。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–0.28 s 机器人双脚接地，腿和躯干支承出拳，身体保持站立。"
+     }
+    ]
+   },
+   "guard_held": {
+    "method": "visual",
+    "intervals": [
+     [
+      2.24,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.28,
+      "log_step": null,
+      "text_zh": "观察：视频 0.24–0.32 s 机器人右肘弯曲，右拳从沙袋前收向右肩附近；随后至视频 2.24 s，双脚持续接地，躯干直立、位置基本不变，右拳保持在肩前附近，沙袋摆回时未见再次接触。已逐帧检查视频 1.28–2.24 s 的末段。"
+     },
+     {
+      "t": 2.24,
+      "log_step": 216,
+      "text_zh": "观察：log step 216 的最终交付及两次独立重放均在仿真 2.24 s、控制步 224 记录 guard 和 success，failed=null；inputs/run.json 的评分结果为 success=1、replay_success=1、deterministic=1、stages_reached=2、fell=0。视频 2.24 s 末帧仍呈收拳站立姿态。"
+     },
+     {
+      "t": 2.24,
+      "log_step": 216,
+      "text_zh": "推断：视频中的收拳站立与最终交付的 guard 验收一致，因此护架连续保持达标，节点在视频末帧 2.24 s 完成；不把视频约 0.28 s 开始呈护架姿态误记为已完成保持。无状态日志，右掌至肩距离、躯干余弦及骨盆速度的精确阈值由裁判成功结果支持。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.28–2.24 s 机器人以双脚支撑地面，腿与躯干保持站立；右拳收回肩前，未再接触沙袋。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "impulse_n_s",
+   "unit": "N*s",
+   "threshold": 12.0,
+   "direction": "higher_better",
+   "value": 16.06
+  },
+  {
+   "name": "contact_s",
+   "unit": "s",
+   "threshold": 0.1,
+   "direction": "lower_better",
+   "value": 0.049
+  },
+  {
+   "name": "hit_height_m",
+   "unit": "m",
+   "threshold": [
+    1.2,
+    1.6
+   ],
+   "direction": "target",
+   "value": 1.419
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人完成有效击袋和护架站稳两步：视频约 0.24 s 已击中白带并收拳分离，随后保持站立，视频 2.24 s 完成护架验收，最终 2/2 节点成立。评分器记录有效冲量 16.06 N·s、接触 0.049 s、首次接触高度 1.419 m，最终交付两次独立重放均成功且终态一致（log step 216）。没有状态日志，精确拳峰接触位置及护架距离、速度阈值不能从录像独立测量，依据最终裁判验收确认。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "两节点均成立：视频约 0.12–0.28 s 右拳击中白带并收回，final 在视频 2.24 s（控制步 224）完成，与 log step 216 的裁判事件（strike 仿真 0.213 s、guard/success 2.24 s）及 run.json 评分字段一致；时间钟明确，contact_actual 合理，无失败故 failure=null。",
+    "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": "2026-10-31T00:46:19Z",
+  "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-boxing-i00-privileged--ebemxbr/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-boxing-i00-privileged--ebemxbr/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "26a97cfd052c7111c362c52a56b278fc9e31f8402aec49cd6d9b5b3c22b25fd9",
+   "video_sha256": "78fb7ce7961dfdc5cd0d2f6acf06541f8b0dedc21fd759ae4bcf7788bf539368",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/boxing/trajectories/embodied-olympics-fun-boxing-i00-privileged--gvnh9sq.json

@@ -0,0 +1,222 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-boxing-i00-privileged--gvnh9sq",
+ "taskKey": "embodied-olympics-fun/boxing",
+ "dag_sha256": "5fe95f4fd88e6dbc49fa1e5ba7230e5fd61571eeb504fd1c0835f0ca7840e98d",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "n_actions": 1490,
+   "sim_steps": 425,
+   "horizon": 1500,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 2,
+   "stages_total": 2,
+   "fell": 0,
+   "replay_s": 3.11,
+   "bad_trajectory": 0,
+   "metric_contact_impulse_n_s": 25.86,
+   "metric_contact_s": 0.046,
+   "metric_hit_height_m": 1.358,
+   "metric_impulse_n_s": 21.16,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 122,
+   "agent_resets": 1130,
+   "agent_replays": 63,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 0.507,
+   "peak_processes": 11,
+   "peak_disk_gb": 1.049,
+   "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-boxing-i00-privileged--gvnh9sq/replay.mp4",
+  "duration_s": 4.28,
+  "fps": 25.0,
+  "frames": 107,
+  "sha256": "d86b330b0a2868c461729b12f0ce8de5201c5fe9d45239085e081a9cd3ad8fac"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "bag_struck": {
+    "method": "visual",
+    "intervals": [
+     [
+      0.96,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.8,
+      "log_step": null,
+      "text_zh": "观察：视频 0.56–0.84 s，机器人右臂由屈肘蓄势转为向前伸出，右手末端打到红色沙袋白带；左手留在身前，双脚支承身体，沙袋随后向远离机器人的方向摆动。"
+     },
+     {
+      "t": 0.96,
+      "log_step": null,
+      "text_zh": "观察：视频 0.96–1.04 s，右手末端与白带之间已出现可见间隙，沙袋继续摆离。inputs/run.json 的评分器字段报告 metric_impulse_n_s=21.16、metric_contact_s=0.046、metric_hit_height_m=1.358，stages_reached=2、success=1；冲量和接触时长均满足击打验收。"
+     },
+     {
+      "t": 0.96,
+      "log_step": 264,
+      "text_zh": "推断：结合视频中的一次击中、分离和评分器通过结果，击中沙袋节点已达成；以视频 0.96 s 的清晰分离帧作完成锚点，实际结算时刻有约 0.2 s 不确定性。log step 264 对最终交付文件重新生成并两次重放，输出均含 strike、guard 成功，指标与本次评分一致；未将早期试验中的击打计入。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "其他",
+      "note_zh": "观察：视频 0.72–0.84 s，机器人右手末端拳垫击中沙袋白色环带。候选接触名为“击打”；现有方式中无短促冲击条目，不属于抓握、推/扫或压/擦；具体拳峰末端接触区由评分器成功结果支持。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.56–1.04 s，机器人以左右脚接地支承出拳，腿和躯干维持直立，脚有原地调整。"
+     }
+    ]
+   },
+   "guard_held": {
+    "method": "visual",
+    "intervals": [
+     [
+      4.24,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 2.32,
+      "log_step": null,
+      "text_zh": "观察：视频 1.04–2.32 s，机器人右肘逐渐弯曲、右手从袋前收回身前；视频 2.28–4.24 s，右臂保持收回姿态，躯干直立，双脚原地小幅交替调整，未见手、膝或躯干落地，右手与沙袋保持分离。"
+     },
+     {
+      "t": 4.24,
+      "log_step": 267,
+      "text_zh": "观察：视频 3.24–4.24 s 的全部逐帧画面中，机器人继续保持屈肘站立至最后一帧。inputs/run.json 评分器报告 success=1、stages_reached=2/2、fell=0；log step 267 对最终交付文件的重放也输出 success，指标与评分记录一致。"
+     },
+     {
+      "t": 4.24,
+      "log_step": null,
+      "text_zh": "推断：评分器两阶段全部通过，且片尾姿态与护架站稳相符，因此连续保持验收完成、节点终值为真；将视频 4.24 s 最后一帧作为约略完成锚点，而非把开始收拳或刚进入护架的时刻算作完成。无状态日志，不能仅从画面独立量出右掌至肩距离、骨盆速度及连续 200 控制步。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察��视频 1.04–4.24 s，机器人收回右臂后以左右脚接地、原地小幅调整支承身体；右手不再触袋，腿和躯干完成站立平衡保持。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "impulse_n_s",
+   "unit": "N*s",
+   "threshold": 12.0,
+   "direction": "higher_better",
+   "value": 21.16
+  },
+  {
+   "name": "contact_s",
+   "unit": "s",
+   "threshold": 0.1,
+   "direction": "lower_better",
+   "value": 0.046
+  },
+  {
+   "name": "hit_height_m",
+   "unit": "m",
+   "threshold": [
+    1.2,
+    1.6
+   ],
+   "direction": "target",
+   "value": 1.358
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人完成一次右拳击白带并与沙袋分离，随后收拳护架站立至视频 4.24 s，两个节点最终均完成。评分器记录有效冲量 21.16 N·s、接触 0.046 s、命中高度 1.358 m，且两阶段通过；最终交付文件的重放核验与之相符。无状态日志，精确拳峰接触位置、连续护架阈值依赖评分器结果，节点时刻按视频帧近似定位；成功参考视频仅用于比较姿态线索。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "两个节点均成功，与评分器（success=1、stages 2/2、fell=0、冲量 21.16、接触 0.046 s）及视频帧一致；时刻按视频时钟，击打 0.96 s、护架 4.24 s（sim_steps 425≈4.25 s）合理，接触标注符合 S16，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": "2026-10-31T00:45:09Z",
+  "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-boxing-i00-privileged--gvnh9sq/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-boxing-i00-privileged--gvnh9sq/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "f94272e78c17ad40a7dc8d2e7409fdf2381ea415913ddba718ac917e6245ec08",
+   "video_sha256": "d86b330b0a2868c461729b12f0ce8de5201c5fe9d45239085e081a9cd3ad8fac",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/boxing/trajectories/embodied-olympics-fun-boxing-i00-privileged--iihc6lf.json

@@ -0,0 +1,222 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-boxing-i00-privileged--iihc6lf",
+ "taskKey": "embodied-olympics-fun/boxing",
+ "dag_sha256": "5fe95f4fd88e6dbc49fa1e5ba7230e5fd61571eeb504fd1c0835f0ca7840e98d",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "n_actions": 373,
+   "sim_steps": 373,
+   "horizon": 1500,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 2,
+   "stages_total": 2,
+   "fell": 0,
+   "replay_s": 3.66,
+   "bad_trajectory": 0,
+   "metric_contact_impulse_n_s": 25.51,
+   "metric_contact_s": 0.028,
+   "metric_hit_height_m": 1.401,
+   "metric_impulse_n_s": 21.03,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 10,
+   "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": 0.706,
+   "peak_processes": 7,
+   "peak_disk_gb": 0.246,
+   "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-boxing-i00-privileged--iihc6lf/replay.mp4",
+  "duration_s": 3.76,
+  "fps": 25.0,
+  "frames": 94,
+  "sha256": "f3912a413e02bdaa0a5caa489bd654b7cc4b0dd3414e45b8c2fc986b52d4fc43"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "bag_struck": {
+    "method": "visual",
+    "intervals": [
+     [
+      0.68,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.6,
+      "log_step": null,
+      "text_zh": "观察：视频 0.48–0.60 s，机器人右臂快速前伸，右拳末端到达红色沙袋的白色环带；视频 0.64–0.72 s，右肘弯曲收拳、拳袋分离，沙袋随后明显摆离。双脚在地面支撑，机器人未倒地。"
+     },
+     {
+      "t": 0.68,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的本次评分器（仿真裁判）记录 success=1、replay_success=1、stages_reached=2/2，metric_impulse_n_s=21.03 N·s、metric_contact_s=0.028 s、metric_hit_height_m=1.401 m，确认该次击打有效。"
+     },
+     {
+      "t": 0.68,
+      "log_step": null,
+      "text_zh": "推断：结合视频 0.60–0.72 s 的击中与分离过程及本次评分器通过结果，将有效击打结算完成粗定位于视频 0.68 s；25 fps 画面不能独立测出毫秒级接触起止或拳峰接触深度。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "其他",
+      "note_zh": "观察：视频 0.56–0.60 s，机器人右手末端击中沙袋白带后收回；候选接触名为“击打”，属于短促冲击，现有怎么碰条目无对应值，不是抓握或持续推压。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.40–0.72 s，机器人左右脚接地支承出拳，腿和躯干维持站立。"
+     }
+    ]
+   },
+   "guard_held": {
+    "method": "visual",
+    "intervals": [
+     [
+      3.72,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 1.8,
+      "log_step": null,
+      "text_zh": "观察：视频 0.68–0.84 s，机器人右肘弯曲、右拳从袋前收回至肩前附近；视频 0.80–1.60 s，机器人以脚步调整位置，之后明显趋于稳定。"
+     },
+     {
+      "t": 3.72,
+      "log_step": null,
+      "text_zh": "观察：视频 1.80–3.72 s，机器人保持弯肘护架与直立姿态，脚在地面支撑、位置基本稳定；逐帧检查视频 2.72–3.72 s 未见倒地或再次触袋。inputs/run.json 记录 stages_reached=2/2、progress=1.0、success=1、fell=0。"
+     },
+     {
+      "t": 3.72,
+      "log_step": null,
+      "text_zh": "推断：本次评分器成功结果确认击打后的护架距离、直立程度、骨盆速度及连续 200 个控制步保持条件均已通过；结合末尾视频帧，将护架保持完成定位于视频约 3.72 s，而非收拳初始时刻。无状态日志，不能逐控制步独立核对阈值及计时起点。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.80–1.60 s，机器人双腿通过脚步调整吸收击打后的运动，随后双脚接地支撑直立护架至视频 3.72 s；收回的右手与沙袋保持分离。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "impulse_n_s",
+   "unit": "N*s",
+   "threshold": 12.0,
+   "direction": "higher_better",
+   "value": 21.03
+  },
+  {
+   "name": "contact_s",
+   "unit": "s",
+   "threshold": 0.1,
+   "direction": "lower_better",
+   "value": 0.028
+  },
+  {
+   "name": "hit_height_m",
+   "unit": "m",
+   "threshold": [
+    1.2,
+    1.6
+   ],
+   "direction": "target",
+   "value": 1.401
+  }
+ ],
+ "failure": null,
+ "summary_zh": "本次评分轨迹完成击中沙袋与护架站稳两个节点：视频约 0.68 s 已击袋并分离，视频末尾约 3.72 s 完成护架连续保持，评分器确认成功且未跌倒。右拳对白带的短促击打及收拳、双脚调整后站稳均可见，原生指标为有效冲量 21.03 N·s、接触 0.028 s、击中高度 1.401 m。无状态日志，毫秒级接触、拳峰深度及护架连续阈值无法从画面独立测量，精确验收采用本次评分器结果，节点时刻为视频近似锚点。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "两个节点（击中沙袋约视频 0.68 s、护架站稳约视频 3.72 s）与视频和评分器结果一致，接触标注合理，无失败故而 failure 为 null，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": "2026-10-31T00:53:35Z",
+  "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-boxing-i00-privileged--iihc6lf/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-boxing-i00-privileged--iihc6lf/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "0b08fcfcdc5f6a90ea5866871fe1d235667c11bdd64b30adb2803a2d59064c03",
+   "video_sha256": "f3912a413e02bdaa0a5caa489bd654b7cc4b0dd3414e45b8c2fc986b52d4fc43",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/boxing/trajectories/embodied-olympics-fun-boxing-i00-privileged--mpenplb.json

@@ -0,0 +1,222 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-boxing-i00-privileged--mpenplb",
+ "taskKey": "embodied-olympics-fun/boxing",
+ "dag_sha256": "5fe95f4fd88e6dbc49fa1e5ba7230e5fd61571eeb504fd1c0835f0ca7840e98d",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "n_actions": 317,
+   "sim_steps": 317,
+   "horizon": 1500,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 2,
+   "stages_total": 2,
+   "fell": 0,
+   "replay_s": 3.64,
+   "bad_trajectory": 0,
+   "metric_contact_impulse_n_s": 21.17,
+   "metric_contact_s": 0.056,
+   "metric_hit_height_m": 1.563,
+   "metric_impulse_n_s": 14.88,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 34,
+   "agent_resets": 1607,
+   "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": 2.253,
+   "peak_processes": 18,
+   "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-boxing-i00-privileged--mpenplb/replay.mp4",
+  "duration_s": 3.2,
+  "fps": 25.0,
+  "frames": 80,
+  "sha256": "1c23b1c7ba19324690d7ca3f63177cbcc268a797400d4c3d742e0be3aaf24d13"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "bag_struck": {
+    "method": "visual",
+    "intervals": [
+     [
+      1.04,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.96,
+      "log_step": null,
+      "text_zh": "观察：视频 0.80–0.92 s 机器人转动躯干并向沙袋伸出右拳，视频 0.96–1.00 s 右拳末端抵达白色环带上部；视频 1.04–1.20 s 沙袋摆离，右肘弯曲、拳头后撤。双脚始终在地面支撑。"
+     },
+     {
+      "t": 1.04,
+      "log_step": 30,
+      "text_zh": "观察：log step 30 对刚保存的交付文件执行官方重放，裁判返回 strike 事件（仿真 1.028 s）、impulse_n_s=14.88、contact_s=0.056、hit_height_m=1.563，且 failed=null；inputs/run.json 的最终评分器字段报告相同指标，replay_success=1、deterministic=1。"
+     },
+     {
+      "t": 1.04,
+      "log_step": 30,
+      "text_zh": "推断：log step 6 记录控制周期为 0.01 s；log step 30 的交付重放共 317 控制步、仿真时长 3.17 s，与视频 3.20 s 相符，且其击打事件与视频 0.96–1.04 s 的拳袋接触及随后分离相符。因此将有效击打结算近似锚定在视频 1.04 s，精确冲量与接触时长采用裁判记录，不由画面摆幅估算。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "其他",
+      "note_zh": "机器人右手末端拳垫短促击打沙袋白色环带；候选名称“击打”，现有接触方式无短促冲击条目，不是抓握或持续推袋。视频 0.96–1.04 s 可见拳袋相遇，随后分离。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "出拳和躯干转动期间，机器人左右脚接地承重，双腿支撑身体，视频 0.80–1.20 s 未见跌倒。"
+     }
+    ]
+   },
+   "guard_held": {
+    "method": "visual",
+    "intervals": [
+     [
+      3.16,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 1.2,
+      "log_step": null,
+      "text_zh": "观察：视频 1.12–1.40 s 机器人右肘弯曲，右拳从袋前收至肩部附近，躯干逐渐转回；视频 1.60–3.16 s 右拳保持收回、身体基本直立，双脚留在原处支撑。已逐帧检查视频 2.16–3.16 s，未见跌倒或再次触袋。"
+     },
+     {
+      "t": 3.16,
+      "log_step": 30,
+      "text_zh": "观察：log step 30 对交付文件的官方重放返回 guard 与 success 事件，均在仿真 3.17 s，stages=[strike,guard]、steps=317；最终评分器字段也报告 stages_reached=2/2、success=1、fell=0，与视频末帧的收拳站姿一致。"
+     },
+     {
+      "t": 3.16,
+      "log_step": 30,
+      "text_zh": "推断：按 log step 6 的每控制步 0.01 s 和 log step 30 的交付重放 317 步，护架验收在回合末完成；采用视频最后可见帧 3.16 s 为近似到达时刻，不把刚开始收拳计为已完成连续保持。单视角不能精确测出掌肩距离、骨盆速度和连续 200 步计数，这些阈值由该交付文件的裁判成功记录确认。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "机器人收回右拳并保持护架时，左右脚���续与地面接触承重，双腿和躯干维持站立；视频 1.20–3.16 s 未见手或膝触地。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "impulse_n_s",
+   "unit": "N*s",
+   "threshold": 12.0,
+   "direction": "higher_better",
+   "value": 14.88
+  },
+  {
+   "name": "contact_s",
+   "unit": "s",
+   "threshold": 0.1,
+   "direction": "lower_better",
+   "value": 0.056
+  },
+  {
+   "name": "hit_height_m",
+   "unit": "m",
+   "threshold": [
+    1.2,
+    1.6
+   ],
+   "direction": "target",
+   "value": 1.563
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人完成有效击袋和护架站稳两步：约视频 1.04 s 完成击打结算，右拳随后收回并站立至视频 3.16 s 的回合末验收。交付文件的官方重放与最终评分器均确认成功，有效冲量 14.88 N·s、接触历时 0.056 s、首次接触高度 1.563 m。无状态日志，视频无法独立精测拳峰接触位置、冲量及护架数值阈值；这些依裁判记录确认，节点时间为与视频核对后的近似值。",
+ "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；bag_struck（视频约1.04 s）与 guard_held（视频3.16 s）与帧、log step 30 的官方重放（strike 仿真1.028 s，success 3.17 s，317步）及评分器字段一致，contact_actual 合理，无失败标签，推断/观察标注正确。",
+    "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": "2026-10-31T00:44:46Z",
+  "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-boxing-i00-privileged--mpenplb/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-boxing-i00-privileged--mpenplb/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "c474b21f98cf1ce65aa61a26bd3d676d5d09d64c7806d3c6f1a4b63692f936e7",
+   "video_sha256": "1c23b1c7ba19324690d7ca3f63177cbcc268a797400d4c3d742e0be3aaf24d13",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/boxing/trajectories/embodied-olympics-fun-boxing-i00-privileged--sswikn5.json

@@ -0,0 +1,220 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-boxing-i00-privileged--sswikn5",
+ "taskKey": "embodied-olympics-fun/boxing",
+ "dag_sha256": "5fe95f4fd88e6dbc49fa1e5ba7230e5fd61571eeb504fd1c0835f0ca7840e98d",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "n_actions": 287,
+   "sim_steps": 287,
+   "horizon": 1500,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 2,
+   "stages_total": 2,
+   "fell": 0,
+   "replay_s": 3.9,
+   "bad_trajectory": 0,
+   "metric_contact_impulse_n_s": 25.39,
+   "metric_contact_s": 0.042,
+   "metric_hit_height_m": 1.412,
+   "metric_impulse_n_s": 15.62,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 276,
+   "agent_resets": 10889,
+   "agent_replays": 38,
+   "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-boxing-i00-privileged--sswikn5/replay.mp4",
+  "duration_s": 2.88,
+  "fps": 25.0,
+  "frames": 72,
+  "sha256": "6fc8420c2ac1d3d87a574f6a2da3a34cff99b51ab8353a4686d88e5207449886"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "bag_struck": {
+    "method": "visual",
+    "intervals": [
+     [
+      0.68,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.6,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–0.52 s 机器人在沙袋前站立并预备右臂，视频 0.56–0.64 s 机器人转动躯干、右拳向前上方击向红色沙袋的白色环带；视频 0.68–0.80 s 右肘弯曲、拳袋分离，沙袋随后摆离。双脚提供地面支撑，未见机器人跌倒。"
+     },
+     {
+      "t": 0.68,
+      "log_step": 594,
+      "text_zh": "观察：log step 593 重新生成最终交付文件，log step 594 对该文件的两次独立新进程重放均报告 stages=[strike,guard]、success=true、相同终态哈希；strike 事件记录在仿真 0.663 s，有效冲量 15.62 N·s、接触时长 0.042 s、首次击中高度 1.412 m，与本次评分 reward 字段一致。"
+     },
+     {
+      "t": 0.68,
+      "log_step": 594,
+      "text_zh": "推断：结合视频 0.60–0.68 s 的白带击打及分离和最终交付的裁判 strike 事件，将视频 0.68 s 记为有效击打已结算的近似时刻；精确冲量和拳峰合法性由本次成功裁判记录确认，不能由摆幅独立测定。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "其他",
+      "note_zh": "观察：视频 0.60–0.64 s 机器人右手末端拳垫击中沙袋白带，随后收回；候选接触名为“击打”，现有怎么碰条目没有短促冲击，故用其他。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–0.80 s 机器人脚接地承重，腿和躯干在预备、转体出拳与收拳时维持站立。"
+     }
+    ]
+   },
+   "guard_held": {
+    "method": "visual",
+    "intervals": [
+     [
+      2.87,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.88,
+      "log_step": null,
+      "text_zh": "观察：视频 0.68–0.88 s 机器人弯曲右肘将拳从沙袋前收至右肩附近；视频 0.88–2.84 s 保持该护架姿态、身体直立，仅脚接地且无明显整体平移，沙袋继续摆动；已逐帧查看视频末段 1.88–2.84 s，未见再触袋或倒地。"
+     },
+     {
+      "t": 2.84,
+      "log_step": 594,
+      "text_zh": "观察：log step 594 的最终交付重放报告在控制步 287 完成 guard，success=true、fell 对应本次评分字段为 0；本次 reward 同时报告 stages_reached=2、stages_total=2、replay_success=1、deterministic=1。"
+     },
+     {
+      "t": 2.87,
+      "log_step": 594,
+      "text_zh": "推断：log step 594 是最终交付文件的控制步计数；log step 7 所读控制周期为 0.01 s，287×0.01=2.87 s，与录像时长 2.88 s 相符。其 strike 事件仿真 0.663 s 已与视频 0.60–0.68 s 的击打、分离核对，故以终末控制步对应的视频 2.87 s 标记护架连续保持验收完成；最后可见帧为视频 2.84 s，终末验收比该帧晚约 0.03 s。精确肩掌距离、骨盆速度和连续 200 步判据依据成功裁判记录。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.88–2.84 s 机器人脚与地面承重接触，腿和躯干支撑收拳后的直立护架；右拳保持离开沙袋。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "impulse_n_s",
+   "unit": "N*s",
+   "threshold": 12.0,
+   "direction": "higher_better",
+   "value": 15.62
+  },
+  {
+   "name": "contact_s",
+   "unit": "s",
+   "threshold": 0.1,
+   "direction": "lower_better",
+   "value": 0.042
+  },
+  {
+   "name": "hit_height_m",
+   "unit": "m",
+   "threshold": [
+    1.2,
+    1.6
+   ],
+   "direction": "target",
+   "value": 1.412
+  }
+ ],
+ "failure": null,
+ "summary_zh": "观察：本次评分录像中，机器人约在视频 0.60–0.64 s 用右拳击中沙袋白带，随后收拳至右肩附近并站立至视频末帧 2.84 s；最终交付的两次新进程重放及评分记录均确认击打、护架两节点完成，冲量 15.62 N·s、接触时长 0.042 s。推断：按最终交付控制步映射，视频约 0.68 s 完成击打结算，视频 2.87 s 完成护架连续保持验收，无失败断点。观察：没有状态日志，录像不能独立量出拳峰接触区、冲量及护架的精确距离和速度阈值，这些判据依据与录像一致的最终交付裁判记录。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "两个节点均达成，击打约视频 0.60–0.64 s（帧 15–16）、沙袋随后摆离，护架保持至末帧；控制步 287×0.01=2.87 s 的映射有说明且与视频时长相符；contact_actual 合理，无失败断点，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": "2026-10-31T00:53:39Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-boxing-i00-privileged--sswikn5/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-boxing-i00-privileged--sswikn5/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "520aeba035ed3162b9d1377c35a0046558d57d4c203412956325b8e431850ee4",
+   "video_sha256": "6fc8420c2ac1d3d87a574f6a2da3a34cff99b51ab8353a4686d88e5207449886",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/boxing/trajectories/embodied-olympics-fun-boxing-i00-privileged--sz2kvle.json

@@ -0,0 +1,222 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-boxing-i00-privileged--sz2kvle",
+ "taskKey": "embodied-olympics-fun/boxing",
+ "dag_sha256": "5fe95f4fd88e6dbc49fa1e5ba7230e5fd61571eeb504fd1c0835f0ca7840e98d",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "n_actions": 319,
+   "sim_steps": 319,
+   "horizon": 1500,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 2,
+   "stages_total": 2,
+   "fell": 0,
+   "replay_s": 4.43,
+   "bad_trajectory": 0,
+   "metric_contact_impulse_n_s": 18.3,
+   "metric_contact_s": 0.041,
+   "metric_hit_height_m": 1.392,
+   "metric_impulse_n_s": 15.23,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 42,
+   "agent_resets": 202,
+   "agent_replays": 12,
+   "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": 0.691,
+   "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-boxing-i00-privileged--sz2kvle/replay.mp4",
+  "duration_s": 3.2,
+  "fps": 25.0,
+  "frames": 80,
+  "sha256": "d32a2dce5d6465a205977a4e28508cd938aa894599a56894192066c65188b225"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "bag_struck": {
+    "method": "visual",
+    "intervals": [
+     [
+      1.16,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 1.08,
+      "log_step": null,
+      "text_zh": "观察：视频 1.00–1.12 s，机器人右臂向前快速伸展，右手末端到达红色沙袋白色环带；视频 1.16–1.28 s 拳袋分开，沙袋向远离机器人的方向摆动，右肘弯曲收拳。双脚持续接地支撑。"
+     },
+     {
+      "t": 1.16,
+      "log_step": 143,
+      "text_zh": "观察：inputs/run.json 的评分器字段 replay_success=1、deterministic=1、stages_reached=2/2，metric_impulse_n_s=15.23、metric_contact_s=0.041、metric_hit_height_m=1.392；log step 143 对最终交付文件的两次独立重放均由仿真裁判返回 strike、guard 阶段成功且终态哈希相同。"
+     },
+     {
+      "t": 1.16,
+      "log_step": null,
+      "text_zh": "推断：结合视频 1.08 s 的击袋及视频 1.16 s 后的拳袋分离与评分器记录，有效击打在视频约 1.16 s 完成；冲量和短接触是否达标以本次评分器字段为据，不能由摆幅独立测出。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "其他",
+      "note_zh": "视频 1.08 s 附近，机器人右手末端拳垫击中沙袋白色环带，随后收回并与沙袋分离。候选名称“击打”：短促冲击不属于抓握、推/扫或压/擦。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "视频 0.92–1.36 s 出拳与收拳过程中，机器人双脚接触地面承重，腿和躯干维持直立。"
+     }
+    ]
+   },
+   "guard_held": {
+    "method": "visual",
+    "intervals": [
+     [
+      3.19,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 1.28,
+      "log_step": null,
+      "text_zh": "观察：视频 1.20–1.36 s，机器人右肘弯曲，将出拳的右手收至右肩附近；视频 1.40–3.16 s 右臂保持回收，身体直立、双脚留在地面，沙袋往返摆动而未见再次触袋。视频 2.20–3.16 s 已逐帧检查，包括末帧。"
+     },
+     {
+      "t": 3.16,
+      "log_step": 143,
+      "text_zh": "观察：视频 3.16 s 末帧机器人仍维持护架站姿；本次评分器 success=1、stages_reached=2、fell=0。log step 143 的最终文件重放返回 stages=[strike, guard]、steps=319、time_s=3.19、failed=null，说明裁判已接受连续护架保持。"
+     },
+     {
+      "t": 3.19,
+      "log_step": 143,
+      "text_zh": "推断：log step 143 所报最终轨迹 319 个控制步对应仿真 3.19 s，即每步 0.01 s，与视频总长 3.20 s 相符；log step 139 对该文件诊断的接触控制步 107–110 对应视频约 1.07–1.10 s，已用视频 1.08 s 的白带接触核对。故将护架验收完成锚定为视频约 3.19 s，精度约一个视频帧；视频 1.20 s 附近开始收拳不等于已经完成连续保持验收。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "视频 1.20–3.16 s，机器人以双脚接地支撑护架站立，腿和躯干维持直立，右手收于肩部附近；未见手、���或躯干落地。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "impulse_n_s",
+   "unit": "N*s",
+   "threshold": 12.0,
+   "direction": "higher_better",
+   "value": 15.23
+  },
+  {
+   "name": "contact_s",
+   "unit": "s",
+   "threshold": 0.1,
+   "direction": "lower_better",
+   "value": 0.041
+  },
+  {
+   "name": "hit_height_m",
+   "unit": "m",
+   "threshold": [
+    1.2,
+    1.6
+   ],
+   "direction": "target",
+   "value": 1.392
+  }
+ ],
+ "failure": null,
+ "summary_zh": "本次评分轨迹完成击中沙袋与护架站稳两节点：视频约 1.16 s 完成击打分离，随后收拳，视频约 3.19 s 连续护架验收完成；评分器记录冲量 15.23 N·s、接触历时 0.041 s、击中高度 1.392 m，成功且未跌倒。成功参考视频中击打后沙袋摆动与收拳站立的线索同样可见，本次节点仅依据本次视频及本次裁判结果判定。没有状态日志，精确拳峰接触位置、护架距离与速度阈值无法从画面独立量测，阈值是否满足依据评分器及最终文件重放成功记录，完成时刻有约一帧误差。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "两节点均达成、成功运行无失败层；视频约1.08–1.16 s击袋、收拳后站稳至3.19 s，与评分器字段一致，控制步到视频时间的换算有证据。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": "2026-10-31T00:58:30Z",
+  "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-boxing-i00-privileged--sz2kvle/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-boxing-i00-privileged--sz2kvle/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "109b2e326044c9e16f196656b98ec7e07bb1e4dd67c7a1c7e9554a7e0d5cfdf0",
+   "video_sha256": "d32a2dce5d6465a205977a4e28508cd938aa894599a56894192066c65188b225",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/boxing/trajectories/embodied-olympics-fun-boxing-i00-privileged--zxqrdj3.json

@@ -0,0 +1,222 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-boxing-i00-privileged--zxqrdj3",
+ "taskKey": "embodied-olympics-fun/boxing",
+ "dag_sha256": "5fe95f4fd88e6dbc49fa1e5ba7230e5fd61571eeb504fd1c0835f0ca7840e98d",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "n_actions": 364,
+   "sim_steps": 364,
+   "horizon": 1500,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 2,
+   "stages_total": 2,
+   "fell": 0,
+   "replay_s": 3.3,
+   "bad_trajectory": 0,
+   "metric_contact_impulse_n_s": 15.41,
+   "metric_contact_s": 0.071,
+   "metric_hit_height_m": 1.456,
+   "metric_impulse_n_s": 13.87,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 80,
+   "agent_resets": 0,
+   "agent_replays": 4,
+   "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": 0.6,
+   "peak_processes": 7,
+   "peak_disk_gb": 0.25,
+   "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-boxing-i00-privileged--zxqrdj3/replay.mp4",
+  "duration_s": 3.68,
+  "fps": 25.0,
+  "frames": 92,
+  "sha256": "a806de4ff8c35f613b12bd58eef383edfad67be4ac316e6d0a2f8379fecb01df"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "bag_struck": {
+    "method": "visual",
+    "intervals": [
+     [
+      0.56,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.44,
+      "log_step": null,
+      "text_zh": "观察：逐帧检查视频 0.36–0.68 s，机器人转动躯干、右臂前伸，视频 0.44–0.48 s 右手末端在沙袋白色环带处击打；随后右肘弯曲收拳，视频 0.56 s 拳袋已分离，沙袋开始摆离。双脚支承身体，未见其他部位落地。"
+     },
+     {
+      "t": 0.56,
+      "log_step": 94,
+      "text_zh": "观察：log step 94 对最终交付文件的重放报告 stages=[strike, guard]、success=true，strike 事件在仿真 0.537 s；评分器 reward 同时报 metric_impulse_n_s=13.87、metric_contact_s=0.071、metric_hit_height_m=1.456，分别满足冲量至少 12.0 N·s、接触不超过 0.10 s 和白带高度要求。"
+     },
+     {
+      "t": 0.56,
+      "log_step": 94,
+      "text_zh": "推断：最终文件重放的 364 个控制步对应仿真 3.64 s，即每步 0.01 s，与视频末帧 3.64 s 及视频 0.44–0.56 s 的击袋分离动作相符；据此将仿真 0.537 s 的 strike 结算定位于视频约 0.54 s，取已检查的邻近帧视频 0.56 s 为节点锚点。精确冲量与拳峰接触合法性由评分器通过结果支持，非从摆幅测得。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "其他",
+      "note_zh": "观察：视频 0.44–0.48 s 机器人右手末端固定拳垫短促击中沙袋白带后收回；候选接触名为“击打”，现有怎么碰条目不能准确表示短促冲击，故用其他。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.36–0.68 s 双脚与地面支承机器人出拳和收拳，腿与躯干维持身体直立。"
+     }
+    ]
+   },
+   "guard_held": {
+    "method": "visual",
+    "intervals": [
+     [
+      3.64,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.68,
+      "log_step": null,
+      "text_zh": "观察：视频 0.52–0.80 s 机器人右肘弯曲、右手从袋前收至肩旁；视频 0.80–1.64 s 机器人仍有小幅足部与身体调整，未将刚收拳的一帧计为连续站稳完成。"
+     },
+     {
+      "t": 3.64,
+      "log_step": 94,
+      "text_zh": "观察：逐帧检查视频 1.64–3.64 s，机器人保持右肘弯曲、右手在右肩附近、躯干直立，足部支撑且身体位置基本稳定；视频最后一秒 2.68–3.64 s 未见跌倒或再次触袋。log step 94 最终交付文件重放的 guard 与 success 事件均在仿真 3.64 s，reward 的 stages_reached=2、replay_success=1、deterministic=1、fell=0。"
+     },
+     {
+      "t": 3.64,
+      "log_step": 94,
+      "text_zh": "推断：最终文件的 364 控制步×0.01 s 与视频末帧 3.64 s 一致，并已用视频 0.44–0.56 s 的击袋动作核对时序；护架连续保持完成以视频 3.64 s 为锚点。成功裁判结果支持掌肩距离、骨盆速度及连续 200 控制步等无法仅凭画面精确量出的验收条件。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.68–3.64 s 机器人通过双脚接地及腿部小幅调整支撑收拳后的身体，后段保持直立护架；右手收在肩旁，与沙袋分离。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "impulse_n_s",
+   "unit": "N*s",
+   "threshold": 12.0,
+   "direction": "higher_better",
+   "value": 13.87
+  },
+  {
+   "name": "contact_s",
+   "unit": "s",
+   "threshold": 0.1,
+   "direction": "lower_better",
+   "value": 0.071
+  },
+  {
+   "name": "hit_height_m",
+   "unit": "m",
+   "threshold": [
+    1.2,
+    1.6
+   ],
+   "direction": "target",
+   "value": 1.456
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人在视频约 0.44–0.48 s 用右手末端击中沙袋白带，分离后完成有效击打，随后收拳调整站姿，在视频 3.64 s 完成连续护架站稳，最终两节点均达成。最终交付文件重放及评分器均判成功，冲量 13.87 N·s、接触时长 0.071 s、击中高度 1.456 m 与视频动作一致。已检查本次全部 92 帧及参考成功视频；无状态日志，精确冲量、拳峰接触位置和护架阈值依评分器及最终重放记录确认，节点时间误差约一帧至 0.1 s。",
+ "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，两节点最终达成；视频 0.44 s 右拳触袋、3.64 s 护架站稳与评分器结果一致，接触标签合理，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": "2026-10-31T00:51:16Z",
+  "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-boxing-i00-privileged--zxqrdj3/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-boxing-i00-privileged--zxqrdj3/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "4fe3dc1993bcaa97c5d2a2bef70cde1c0c75a46b07e51bde4ab5e4bf4fa83162",
+   "video_sha256": "a806de4ff8c35f613b12bd58eef383edfad67be4ac316e6d0a2f8379fecb01df",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/boxing/trajectories/embodied-olympics-fun-boxing-i00-standard--hvtgpyv.json

@@ -0,0 +1,221 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-boxing-i00-standard--hvtgpyv",
+ "taskKey": "embodied-olympics-fun/boxing",
+ "dag_sha256": "5fe95f4fd88e6dbc49fa1e5ba7230e5fd61571eeb504fd1c0835f0ca7840e98d",
+ "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": 42,
+   "sim_steps": 257,
+   "calls": 15,
+   "rows": 257,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 487.9,
+   "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": 0.213,
+   "peak_processes": 7,
+   "peak_disk_gb": 0.281,
+   "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-boxing-i00-standard--hvtgpyv/replay.mp4",
+  "duration_s": 0.44,
+  "fps": 25.0,
+  "frames": 11,
+  "sha256": "9a636ff4f0fb4e49632b7dcc9d2c33caa89d4e0f4be89d61447c5f709400afec"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "bag_struck": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.4,
+      "log_step": 111,
+      "text_zh": "观察：已逐帧检查视频 0.00–0.40 s 的全部 11 帧；机器人双脚站立，先弯肘蓄拳，视频 0.36–0.40 s 右手向红色沙袋白带伸出，末帧右手仍贴近袋面。log step 111 显示最后一次 reset 后执行 episode 6，机器人服务返回 done=True、reason=failed、stages_reached=0/2，并由内置裁判明确判定“the bag was touched by something other than the knuckle part of the right fist”；此为服务裁判结果，非 agent 自述。有效拳峰击打未获验收。"
+     },
+     {
+      "t": 0.4,
+      "log_step": 112,
+      "text_zh": "观察：log step 112 确认最终 episode=6、control_steps=42、done=True；此前 episode 1–5 均被 reset，不将其击打或失败计入本视频。视频仅有最后一次蓄拳、伸拳，未显示成功参考视频中击后分离、收拳并持续护架的后续过程。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "其他",
+      "note_zh": "观察：视频 0.36–0.40 s 机器人右手末端向沙袋白带出拳，log step 111 的裁判结果确认发生非合法拳峰部位触袋；候选接触名“击打”，短促出拳接触不属于抓握、推/扫或压/擦。画面不能分辨具体违规接触斑块。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–0.40 s 机器人双脚持续接地支撑蓄拳和出拳，未见跌倒。"
+     }
+    ]
+   },
+   "guard_held": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.4,
+      "log_step": 111,
+      "text_zh": "观察：视频 0.00–0.40 s 全程未出现有效击打后收回右拳并护架保持的过程，末帧右臂仍伸向袋面；log step 111 的最终回合由裁判因违规触袋终止，stages_reached=0/2。推断：击打前置未完成，且整个视频不足要求的击后连续 2.0 s，护架站稳节点从未成立；脚本中尚未执行完成的护架计划不计为进度。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "impulse_n_s",
+   "unit": "N*s",
+   "threshold": 12.0,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "contact_s",
+   "unit": "s",
+   "threshold": 0.1,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "hit_height_m",
+   "unit": "m",
+   "threshold": [
+    1.2,
+    1.6
+   ],
+   "direction": "target",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "bag_struck",
+   "t": 0.4,
+   "log_step": 111,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 0.36–0.40 s 机器人正在把右手伸向白带，尚未完成拳袋分离和护架；log step 111 的最终 episode 6 被裁判以非右拳拳峰部位触袋判失败，stages_reached=0/2。断点属于正在进行的击中沙袋步骤；视频 0.40 s 为可见末段接触的近似导航锚点，未把 agent 工作时间换算成视频时间。"
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 0.4,
+    "log_step": 111,
+    "evidence_zh": "观察：视频 0.36–0.40 s 机器人右手���至沙袋白带附近；log step 111 的内置裁判报告实际触袋者包含合法右拳拳峰之外的部位。推断：这次出拳没有把实际接触限制在要求的拳峰区域，属于击打接触位置或姿态不准；并非完全没有触袋，也未见机器人失衡。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "t": 0.4,
+    "log_step": 110,
+    "source_zh": "日志 / 视频",
+    "evidence_zh": "观察：log step 105 的离线运动学计算报告目标拳峰位置及掌部间隙，log step 110 的最终脚本按固定关节插值执行蓄拳、快速伸拳与回收，并声称目标姿态可保证拳峰接触；视频 0.36–0.40 s 显示机器人确实伸向白带，但 log step 111 的实际执行被裁判判为非拳峰触袋。推断：接触控制未将目标点上的静态校验落实为合法的实际接触；证据不足以继续区分接触几何设计或动态跟踪问题，故停留在“控制”。"
+   }
+  ],
+  "gap_zh": "没有状态日志及逐接触部位记录；视频 0.36–0.40 s 的远景不足以确定具体是哪一块非拳峰表面触袋，违规事实依据最终 episode 6 的服务裁判反馈。评分器未报告 impulse_n_s、contact_s、hit_height_m，三项原生指标均留空；不能从 agent 在 log step 114 所称“沙袋变形”确定材料变形机制或控制子类。",
+  "unmet": {
+   "bag_struck": false,
+   "guard_held": false
+  }
+ },
+ "summary_zh": "评分采用最后一次现场 episode 6：机器人站立蓄拳并向沙袋白带出拳，但裁判因非右拳拳峰部位触袋终止回合，击中沙袋与护架站稳均未完成（0/2）。断点在击中沙袋操作中，表现为实际击打接触位置或姿态不准，根因保守标为控制。已检查视频全部 11 帧并对照成功参考；无状态日志，具体违规接触斑块和动态误差来源无法细分，裁判也未报告三项原生指标。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "两节点未达成、断点在击中沙袋（操作中）、位姿不准+控制（保守）均有证据支持：视频 0.00–0.40 s（11 帧）仅见蓄拳与伸向沙袋，末帧手掌贴袋；log step 111 服务裁判返回 reason=failed、stages_reached=0/2、非右拳拳峰部位触袋。时钟均为视频时间，未换算 agent 时间；未发现需修改项。",
+    "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": "2026-10-31T00:52:00Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-boxing-i00-standard--hvtgpyv/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-boxing-i00-standard--hvtgpyv/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "666579a6a337f9a36a63fcf30be515ae6dcd77bcf73c55fb810eaa4ed4bcdcfb",
+   "video_sha256": "9a636ff4f0fb4e49632b7dcc9d2c33caa89d4e0f4be89d61447c5f709400afec",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/boxing/trajectories/embodied-olympics-fun-boxing-i00-standard--nbxbqbe.json

@@ -0,0 +1,214 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-boxing-i00-standard--nbxbqbe",
+ "taskKey": "embodied-olympics-fun/boxing",
+ "dag_sha256": "5fe95f4fd88e6dbc49fa1e5ba7230e5fd61571eeb504fd1c0835f0ca7840e98d",
+ "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": 343,
+   "sim_steps": 1402,
+   "calls": 36,
+   "rows": 19,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 2073.3,
+   "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": 0.21,
+   "peak_processes": 12,
+   "peak_disk_gb": 1.93,
+   "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-boxing-i00-standard--nbxbqbe/replay.mp4",
+  "duration_s": 3.44,
+  "fps": 25.0,
+  "frames": 86,
+  "sha256": "648cc20be7fbf75154bfb47121c814d5ec3ca803eeaffaef9d9a4225cd356117"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "bag_struck": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": 136,
+      "text_zh": "观察：本次视频从机器人站在沙袋前开始，右臂先抬起、屈肘并分段调整；视频 0.00–3.40 s 的沙袋基本竖直，未见成功参考中击打后明显摆离、拳袋分离的过程。视频 2.32–3.40 s 已逐帧检查，机器人没有跌倒。"
+     },
+     {
+      "t": 3.4,
+      "log_step": 136,
+      "text_zh": "观察：log step 136 明确运行最后一次现场回合，重置后 resets_left=0；guard、wind1、wind2、windF 均返回未结束，punch 返回 done=True、reason=failed。log step 137 的服务状态为 episode=6、control_steps=343，未打印 strike 事件或冲量。"
+     },
+     {
+      "t": 3.4,
+      "log_step": 137,
+      "text_zh": "推断：log step 11 给出控制周期 0.01 s；log step 130 的最终流程在出拳前依次执行 1.2+0.9+0.6+0.7=3.40 s，且 log step 136 未执行额外 windG。该控制步时钟与视频约 1.20 s 的右臂姿态切换及视频末端吻合；343 步对应约视频 3.43 s，出拳仅起动约 0.03 s 即判负。结合未见沙袋摆离与拳袋分离，未完成符合冲量、接触及部位要求的有效击打，不能凭 agent 的本地模拟结果记为达成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–3.40 s，机器人双脚接地，腿和躯干支撑右臂准备及出拳起动，全程未见倒地。"
+     },
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：视频末端可见机器人右臂朝沙袋白带方向起动出拳，但固定拳垫的真实接触面细小，视频在起动处结束，无法确认是否发生触袋及首先触袋的具体部位；此处标的是目标接触未知，出拳步骤确已尝试。"
+     }
+    ]
+   },
+   "guard_held": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 3.4,
+      "log_step": 136,
+      "text_zh": "观察：视频 0.00–3.40 s 只显示击打前的抬臂、屈肘准备和末端出拳起动，未显示有效击打之后的收拳及连续站稳保持。log step 136 在 punch 即返回失败，没有执行后续 retract、guard_hold；开局的护架准备不能作为击打后的完成状态。"
+     },
+     {
+      "t": 3.4,
+      "log_step": 137,
+      "text_zh": "推断：前置 bag_struck 未达成，且视频结束前不存在击打后的连续 2.0 s 护架窗口，因此 guard_held 从未达成；本回合未开始该后续步骤。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "impulse_n_s",
+   "unit": "N*s",
+   "threshold": 12.0,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "contact_s",
+   "unit": "s",
+   "threshold": 0.1,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "hit_height_m",
+   "unit": "m",
+   "threshold": [
+    1.2,
+    1.6
+   ],
+   "direction": "target",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "bag_struck",
+   "t": 3.4,
+   "log_step": 136,
+   "where": "操作中",
+   "evidence_zh": "观察：视频约 3.40 s，机器人仍站立，右臂从���前屈肘姿态起动出拳，沙袋未出现明显摆离；log step 136 的最后现场回合在 punch 返回 failed，log step 137 确认回合已结束。推断：按控制周期及该回合命令长度，终止发生在出拳起动约 0.03 s 后；未修复的失败步骤为有效击中沙袋，随后护架尚未开始。"
+  },
+  "unmet": {
+   "bag_struck": false,
+   "guard_held": false
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：inputs/ 有完整视频和会话日志，但没有状态日志；最后回合的日志只打印 reason=failed，未打印裁判 outcome、stages_reached、strike、impulse_n_s、contact_s 或接触点。评分字段 first_reason_code=3 未附代码含义，不能自行解码。视频最后一秒已逐帧检查，但 25 fps 的场馆远景不能分辨出拳起动末端约 0.03 s 内的拳峰、掌侧或前臂触袋先后。推断：因此可确定断点与两节点未达成，不能可靠区分部位或位姿违规、力不足等具体失败现象，也不能在接触设计、轨迹跟踪或感知误差之间确定根因，现象与根因均留空。agent 在 log step 138、140 将最后回合描述为“可能推袋超时或接触姿态违规”，属于其推测；此前重置回合的推袋与本地模拟冲量不是本次回合的事实。三个原生指标均未由评分器报告，故 value=null。"
+ },
+ "summary_zh": "观察：最后现场回合对应本次视频，机器人完成右臂抬起及分段准备，在视频约 3.40 s 起动出拳后被服务判负，沙袋未见明显摆离，有效击袋与击打后的护架保持均未达成，最终为 0/2。推断：断点为 bag_struck 的操作中；裁判细项与接触状态未记录，末端短暂接触亦无法从远景分辨，因此未采纳 agent 关于推袋超时的猜测，现象及根因留空。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "两节点未达成、断点 bag_struck 操作中、控制步时钟（343 步≈3.43 s，与视频 3.44 s 吻合）均有帧和 log step 136/137 支持；现象与根因留空并写明 gap_zh，因裁判细项缺失而无法区分，符合证据纪律。",
+    "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": "2026-10-31T00:51:32Z",
+  "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-boxing-i00-standard--nbxbqbe/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-boxing-i00-standard--nbxbqbe/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "02b490c94d0c0b583a2571a969999f09e0f58afc511d095b5a7f67ea71ad28e3",
+   "video_sha256": "648cc20be7fbf75154bfb47121c814d5ec3ca803eeaffaef9d9a4225cd356117",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/boxing/trajectories/embodied-olympics-fun-boxing-i00-standard--odhv9c9.json

@@ -0,0 +1,216 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-boxing-i00-standard--odhv9c9",
+ "taskKey": "embodied-olympics-fun/boxing",
+ "dag_sha256": "5fe95f4fd88e6dbc49fa1e5ba7230e5fd61571eeb504fd1c0835f0ca7840e98d",
+ "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": 417,
+   "sim_steps": 771,
+   "calls": 21,
+   "rows": 771,
+   "resets": 2,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 370.6,
+   "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": 0.0,
+   "peak_memory_gb": 0.33,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.842,
+   "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-boxing-i00-standard--odhv9c9/replay.mp4",
+  "duration_s": 4.2,
+  "fps": 25.0,
+  "frames": 105,
+  "sha256": "f8356ef83e80eb0e2afa5caecb1659d425d3ffd7112c33095b21bd1304d717a2"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "bag_struck": {
+    "method": "visual",
+    "intervals": [
+     [
+      1.2,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 1.04,
+      "log_step": 38,
+      "text_zh": "观察：视频 0.92–1.12 s，机器人右臂迅速向前伸出，右手末端在沙袋白色环带处接触；视频 1.16–1.28 s 右肘弯曲、拳袋分离，沙袋随后摆离。视频全程只见这一次击袋动作。"
+     },
+     {
+      "t": 1.2,
+      "log_step": 45,
+      "text_zh": "推断：以视频约 1.20 s 作为拳袋分离后的击打结算锚点，误差约 0.1 s；log step 45 中机器人服务的末回合裁判返回 episode=3、outcome=success、stages_reached=2、stages_total=2，结合视频可判此击被裁判接受，包括带内拳峰、冲量和接触时长判据；不能由摆幅独立测出精确冲量。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "其他",
+      "note_zh": "观察：视频约 1.04–1.12 s，机器人右手末端拳垫击打红色沙袋的白色环带后迅速分离。候选名称“击打”；现有方式条目没有短促冲击，不属于抓握或持续推扫。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0–1.20 s，机器人双脚落地支撑出拳，腿和躯干保持站立，未见膝或躯干落地。"
+     }
+    ]
+   },
+   "guard_held": {
+    "method": "visual",
+    "intervals": [
+     [
+      4.17,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 1.28,
+      "log_step": 38,
+      "text_zh": "观察：视频 1.16–1.28 s，机器人右拳从袋前收回肩前附近；视频 1.28–2.16 s 躯干有小幅前后调整，仍保持站立。视频 2.20–4.16 s 右肘持续弯曲、右拳保持肩前，身体趋于直立、位置变化很小，未见再次触袋或跌倒；末秒视频 3.20–4.16 s 已逐帧核对。"
+     },
+     {
+      "t": 4.17,
+      "log_step": 45,
+      "text_zh": "观察：log step 45 的机器人服务返回末回合 episode=3、control_steps=417、sim_time_s=4.17、done=true、reason=success、stages_reached=2/2；这是裁判现场结果。log step 7 的服务规格给出 control_dt=0.01 s。"
+     },
+     {
+      "t": 4.17,
+      "log_step": 45,
+      "text_zh": "推断：采用本次评分回合的控制步映射，417×0.01=视频约 4.17 s，作为连续护架站稳验收完成时刻；log step 38 的第 3 回合出拳结束 control_steps=106 对应视频约 1.06 s 的伸拳、随后收拳，且末回合 4.17 s 与视频末帧 4.16 s 对齐，已核对映射。精确连续保持由成功裁判确认，不能仅以视频中开始收拳的时刻提前记为完成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 1.28–4.16 s，机器人双脚支撑地面，腿和躯干调整姿态并保持直立；右拳收于肩前，不再接触沙袋。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "impulse_n_s",
+   "unit": "N*s",
+   "threshold": 12.0,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "contact_s",
+   "unit": "s",
+   "threshold": 0.1,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "hit_height_m",
+   "unit": "m",
+   "threshold": [
+    1.2,
+    1.6
+   ],
+   "direction": "target",
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "观察：评分视频对应 log step 37 重置后的第 3 回合；机器人约在视频 1.04–1.12 s 以右拳击中白带，约在视频 1.20 s 完成击打分离，随后收拳并站立保持，log step 45 的现场裁判确认 2/2 成功。推断：按末回合控制步数与视频核对，将护架站稳验收记在视频约 4.17 s；前两回合的带外击打和跌倒仅属会话调试。观察：本次没有状态日志，评分记录和服务输出未报告 impulse_n_s、contact_s、hit_height_m 数值，故原生指标均留 null，精确冲量、拳峰接触位置及连续保持判据依赖末回合裁判成功结果。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "评分回合成功（2/2），两节点时间和接触标注与视频及 log step 38/45 的控制步映射一致（417×0.01=4.17 s，视频 4.2 s），未发现需修改之处。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:59:50Z",
+  "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-boxing-i00-standard--odhv9c9/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-boxing-i00-standard--odhv9c9/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "ec194e76eef3e88e8a8ff3b88400f1c7c99a86e66a1c2354cbdbbcbfd1253eae",
+   "video_sha256": "f8356ef83e80eb0e2afa5caecb1659d425d3ffd7112c33095b21bd1304d717a2",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/boxing/trajectories/embodied-olympics-fun-boxing-i00-standard--pucvb24.json

@@ -0,0 +1,258 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-boxing-i00-standard--pucvb24",
+ "taskKey": "embodied-olympics-fun/boxing",
+ "dag_sha256": "5fe95f4fd88e6dbc49fa1e5ba7230e5fd61571eeb504fd1c0835f0ca7840e98d",
+ "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": 208,
+   "sim_steps": 287,
+   "calls": 6,
+   "rows": 287,
+   "resets": 2,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 7055.3,
+   "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": 0.0,
+   "peak_memory_gb": 1.223,
+   "peak_processes": 8,
+   "peak_disk_gb": 0.311,
+   "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-boxing-i00-standard--pucvb24/replay.mp4",
+  "duration_s": 2.12,
+  "fps": 25.0,
+  "frames": 53,
+  "sha256": "a3ce80b1b6493f6d86672a889bc4804af998a0e6cf75922672ecb49dccec03b4"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "bag_struck": {
+    "method": "visual",
+    "intervals": [
+     [
+      0.72,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.56,
+      "log_step": null,
+      "text_zh": "观察：视频 0.36–0.52 s 机器人右手末端在画面中已贴靠沙袋白色环带，视频 0.56–0.64 s 沙袋摆离、右手撤开，视频 0.72 s 拳袋已明确分离；双脚接地支撑。视频不能独立测量实际接触历时与冲量，是否达到击打判据以 log step 251 的末回合裁判 stages_reached=1 结合视频判定。"
+     },
+     {
+      "t": 0.72,
+      "log_step": 251,
+      "text_zh": "推断：log step 251 的现场服务记录为最后一次重置后的 episode 3，裁判返回 stages_reached=1、stages_total=2、outcome=fell，与视频中的击袋后后倒一致，故有效击打已经通过裁判验收；首次完成约取视频 0.72 s，视觉定位误差约 0.1 s。该记录的 208 个控制步、dt=0.01 s 对应仿真 2.08 s，已与视频 2.08 s 的倒地末帧核对；不使用 agent 工作时间换算。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "其他",
+      "note_zh": "观察：视频 0.52–0.60 s 机器人右手末端拳垫击中沙袋白带，随后撤离。候选接触名“击打”，现有接触方式无短促冲击条目，故用其他。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.48–0.72 s 机器人左右脚接地，双腿支承身体完成出拳和撤拳。"
+     }
+    ]
+   },
+   "guard_held": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.92,
+      "log_step": 250,
+      "text_zh": "观察：视频 0.76–0.92 s 机器人右肘弯曲，右手从袋前收至右肩附近，开始护架；log step 250 的脚本也在击打及撤拳之后切换到护架关节目标。"
+     },
+     {
+      "t": 1.6,
+      "log_step": 251,
+      "text_zh": "观察：视频 1.20–1.56 s 机器人身体逐渐向后倾斜，视频 1.60–2.08 s 后仰明显加剧直至倒下，未在击打后连续站稳 2.0 s。log step 251 的现场裁判仅计 1/2 阶段，终止原因为 fell。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.92–1.56 s 机器人双脚接地，腿和躯干尝试维持收拳护架；视频 1.60–2.08 s 脚部支承未能维持平衡，身体向后倒下。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "impulse_n_s",
+   "unit": "N*s",
+   "threshold": 12.0,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "contact_s",
+   "unit": "s",
+   "threshold": 0.1,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "hit_height_m",
+   "unit": "m",
+   "threshold": [
+    1.2,
+    1.6
+   ],
+   "direction": "target",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "guard_held",
+   "t": 1.6,
+   "log_step": 251,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 0.92 s 机器人已收拳护架，此后持续后仰，视频 1.60 s 起明显失稳且未恢复，至视频 2.08 s 倒下；log step 251 的最后现场回合由裁判以 fell 终止，护架阶��未通过。"
+  },
+  "unmet": {
+   "bag_struck": true,
+   "guard_held": false
+  },
+  "phenomena": [
+   {
+    "cat": "全身",
+    "item": "平衡",
+    "t": 1.6,
+    "log_step": 251,
+    "evidence_zh": "观察：视频 1.60–2.08 s 机器人上体与骨盆向后下方运动，身体由站立转为向后倒下；log step 251 返回裁判 outcome=fell。失稳发生在击袋完成后的护架保持期间。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "evidence_zh": "推断：log step 250 的实际执行脚本一次性提交按时间插值的 stand 关节目标，击打后转至固定护架姿态，没有根据回传姿态调整后续指令；视频 0.92–2.08 s 机器人在此阶段持续后倾且未纠正。log step 251 确认该脚本的最后回合以跌倒结束，支持击打后站立控制未能稳定身体；尚不能细分为具体增益、支承布局或反馈环节的问题。",
+    "source_zh": "日志"
+   }
+  ],
+  "gap_zh": "本次输入没有状态日志；评分器 reward 与最后回合服务结果未报告 impulse_n_s、contact_s、hit_height_m 数值，三个原生指标留空，有效击打以末回合裁判 stages_reached=1 结合视频判定。录像可见后倒，但不足以分辨具体失稳力矩、关节增益或支承控制缺陷，根因仅标控制。log step 86、134 属于此前两个已重置回合，其弱击打和违规触袋不是本段视频中的失败。"
+ },
+ "summary_zh": "最后评分回合约在视频 0.72 s 完成有效击袋，右拳随后收至肩旁，最终完成 1/2 节点。机器人在护架期间逐渐后仰，视频 1.60–2.08 s 失稳倒下，断于护架站稳；执行脚本的击打后站立控制未能维持平衡。无状态日志及精确冲量等原生读数，击打合格由末回合裁判阶段数与视频共同确认，具体失稳机制仍不明。",
+ "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、时间、断点（guard_held，倒地）、平衡现象和控制根因均与视频及末回合裁判读数（stages_reached=1/2, outcome=fell）一致；仅 bag_struck 的第一条观察描述与帧不符。",
+    "n_issues": 1,
+    "n_major": 0
+   },
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 2,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "bag_struck 约视频 0.72 s 完成（末回合裁判 stages_reached=1/2），guard_held 未完成；断点 guard_held、现象平衡、根因控制均与视频 1.2–2.08 s 后倒及 log step 251 outcome=fell 相符；时钟换算（208 控制步=2.08 s=视频末帧）已核对，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": "2026-10-31T01:02:41Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-boxing-i00-standard--pucvb24/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-boxing-i00-standard--pucvb24/04_review_r1"
+   },
+   {
+    "stage": "reconcile",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-boxing-i00-standard--pucvb24/05_reconcile_r1"
+   },
+   {
+    "stage": "traj_review",
+    "round": 2,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-boxing-i00-standard--pucvb24/04_review_r2"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "979f57abe23d6748fba9a34cdec3d6211d85019fb5d5ee2c18249510cd11c452",
+   "video_sha256": "a3ce80b1b6493f6d86672a889bc4804af998a0e6cf75922672ecb49dccec03b4",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/boxing/trajectories/embodied-olympics-fun-boxing-i00-standard--q2mzaun.json

@@ -0,0 +1,225 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-boxing-i00-standard--q2mzaun",
+ "taskKey": "embodied-olympics-fun/boxing",
+ "dag_sha256": "5fe95f4fd88e6dbc49fa1e5ba7230e5fd61571eeb504fd1c0835f0ca7840e98d",
+ "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": 100,
+   "sim_steps": 3056,
+   "calls": 134,
+   "rows": 1642,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 2252.7,
+   "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": 0.265,
+   "peak_processes": 11,
+   "peak_disk_gb": 1.07,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-boxing-i00-standard--q2mzaun/replay.mp4",
+  "duration_s": 1.04,
+  "fps": 25.0,
+  "frames": 26,
+  "sha256": "c7e93ecc649d688efea30bae2f901a134f304c9bf92ff323b81fb05297c7cf07"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "bag_struck": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 1.0,
+      "log_step": 170,
+      "text_zh": "观察：视频 0.00–0.88 s 机器人站立、右臂准备，视频 0.92–1.00 s 右手向沙袋挥出，末帧右手到达白带下缘附近；未见击打后分离、沙袋明显摆离的后续过程。log step 170 的最后现场回合 episode=6 在 control_steps=100 时由服务裁判返回“the bag was touched by something other than the knuckle part of the right fist”，stages_reached=0；这是服务裁判结果，不是 agent 自述。"
+     },
+     {
+      "t": 1.0,
+      "log_step": 10,
+      "text_zh": "推断：log step 10 的服务规则明确仅最后一个现场回合受评；log step 170 的最后回合出拳终止与视频末尾动作一致。非法部位先触袋使有效击打未被接受，故本节点全程未达成，而非由总分为零单独推得。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "其他",
+      "note_zh": "观察：视频 0.92–1.00 s 右手末端挥向沙袋并接触，log step 170 的裁判确认存在非右拳拳峰的违规触袋；候选接触名“击打”，短促挥击不适用抓握或推/扫。画面不能细分究竟是拳垫哪一面或邻近部位先接触。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–1.00 s 机器人双脚持续接地，双腿与躯干支撑准备和出拳，未见跌倒。"
+     }
+    ]
+   },
+   "guard_held": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 1.0,
+      "log_step": 170,
+      "text_zh": "观察：视频 0.92–1.00 s 机器人仍在向袋挥右臂，视频结束前没有击打后收拳至肩旁并连续保持的阶段。log step 170 的服务在出拳阶段即返回 failed、stages_reached=0；脚本仅在未终止时才执行后续收拳和护架，此分支没有执行。"
+     },
+     {
+      "t": 1.0,
+      "log_step": 170,
+      "text_zh": "推断：前置有效击打未达成，且本回合在收拳护架前终止，无法满足有效击打后的连续 2.0 s 护架验收。开局站立不算完成此 final 节点。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "impulse_n_s",
+   "value": null,
+   "unit": "N*s",
+   "threshold": 12.0,
+   "direction": "higher_better"
+  },
+  {
+   "name": "contact_s",
+   "value": null,
+   "unit": "s",
+   "threshold": 0.1,
+   "direction": "lower_better"
+  },
+  {
+   "name": "hit_height_m",
+   "value": null,
+   "unit": "m",
+   "threshold": [
+    1.2,
+    1.6
+   ],
+   "direction": "target"
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "bag_struck",
+   "t": 1.0,
+   "log_step": 170,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 0.92–1.00 s 机器人右手向袋前上方挥出并到达袋面，身体仍站立；log step 170 的最后现场回合裁判在出拳时以“非右拳拳峰部位触袋”判负，stages_reached=0。log step 172 确认该回合已结束且不能再次重置，故击打步骤没有修复。"
+  },
+  "unmet": {
+   "bag_struck": false,
+   "guard_held": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 1.0,
+    "log_step": 170,
+    "evidence_zh": "推断：视频 0.92–1.00 s 的右手挥击到达沙袋，但 log step 170 的现场服务裁判判定接触不属于要求的右拳拳峰。问题是触袋时的接触部位/姿态未对准指定拳峰区；画面支持挥击触袋，不能独立分辨违规接触面的细节。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "抓取和接触设计",
+    "alt_feasible": "未知",
+    "evidence_zh": "推断：log step 170 的脚本预先将躯干偏航设为 0.30 rad，随后主要通过右肩俯仰从 0.4 rad 向 −0.90 rad 挥臂，其余右臂目标保持零；该接触姿态在视频 1.00 s 到达袋面时，被现场服务裁判判为非右拳拳峰触袋。所选挥击的接触几何未使规定拳峰独占触袋；具体应如何调整腕/臂姿态及可行替代未在本回合验证。",
+    "source_zh": "日志"
+   }
+  ],
+  "gap_zh": "本回合无状态日志；已逐帧检查视频 0.00–1.00 s，并对视频 0.84–1.00 s 放大检查，仍不能精确识别裁判判负的接触面，也不能测得冲量、接触时长或首次接触点高度。run.json 未报告 impulse_n_s、contact_s、hit_height_m，故原生指标留空。成功参考视频仅用于辨认击打后袋摆动和收拳护架的外观，不作为本回合证据；不采用 agent 关于“手指先碰”或“握拳无效”的猜测作为已证实的机械原因。"
+ },
+ "summary_zh": "评分对应 log step 170 的最后现场回合，视频中机器人站立准备后挥右手触袋，裁判以非右拳拳峰部位触袋终止；有效击打与护架站稳均未完成，最终进度 0/2。断点为击中沙袋，所选挥击接触姿态未满足拳峰限定；无状态日志，具体违规接触面及冲量等精确指标不可核实，先前重置回合的表现不计入本轨迹。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "标签与视频（仅1.04 s，最后现场回合）及 log step 170/172 一致：服务裁判以非右拳拳峰触袋判负，stages_reached=0，两节点未达成，断点 bag_struck 合理；现象与根因有日志依据且标为推断，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": "2026-10-31T00:56:51Z",
+  "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-boxing-i00-standard--q2mzaun/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-boxing-i00-standard--q2mzaun/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "9efdee9a73b99be97d55a49afade242d37213c0f2972f2222811114fdb9cf902",
+   "video_sha256": "c7e93ecc649d688efea30bae2f901a134f304c9bf92ff323b81fb05297c7cf07",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/boxing/trajectories/embodied-olympics-fun-boxing-i00-standard--uzi6vau.json

@@ -0,0 +1,220 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-boxing-i00-standard--uzi6vau",
+ "taskKey": "embodied-olympics-fun/boxing",
+ "dag_sha256": "5fe95f4fd88e6dbc49fa1e5ba7230e5fd61571eeb504fd1c0835f0ca7840e98d",
+ "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": 60,
+   "sim_steps": 378,
+   "calls": 12,
+   "rows": 378,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 5450.6,
+   "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": 0.0,
+   "peak_memory_gb": 1.248,
+   "peak_processes": 12,
+   "peak_disk_gb": 2.975,
+   "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-boxing-i00-standard--uzi6vau/replay.mp4",
+  "duration_s": 0.64,
+  "fps": 25.0,
+  "frames": 16,
+  "sha256": "309809651b52fddc3a00eaca44590e1eb9158c06d9ee3d8d0d37291fd185e566"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "bag_struck": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.52,
+      "log_step": null,
+      "text_zh": "观察：已逐帧检查视频 0.00–0.60 s 的全部 16 帧；机器人双脚支撑地面，先抬臂蓄力，再在视频 0.52–0.60 s 将右手迅速伸向沙袋白带，末帧右臂仍在袋前，未出现完成击打后分离的过程。"
+     },
+     {
+      "t": 0.6,
+      "log_step": 156,
+      "text_zh": "观察：log step 156 的机器人服务返回 reset 后 episode=6、control_steps=60、done=True、stages=0/2，并由现场裁判报告“the bag was touched by something other than the knuckle part of the right fist”；log step 158 确认最后仍是 episode=6、failed。视频 0.60 s 的右臂前伸触袋与此失败回合相符，因此这是违规击袋尝试，未完成有效击中沙袋节点；前五次重置测试不计入本视频进度。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "其他",
+      "note_zh": "观察：视频 0.52–0.60 s 机器人右手末端朝沙袋白带快速击出；log step 156 的现场裁判确认触袋含右拳拳峰之外部位。候选接触名为“击打”，现有怎么碰条目没有短促冲击，不能标抓握；录像不足以辨明具体违规接触点。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–0.60 s 机器人双脚接地，腿和躯干支撑蓄力与出拳，未见跌倒。"
+     }
+    ]
+   },
+   "guard_held": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.6,
+      "log_step": 156,
+      "text_zh": "观察：视频 0.52–0.60 s 机器人右臂继续向袋伸出，最后一帧未收回右肩附近，整段视频没有击打后连续 2.0 s 的护架保持。log step 156 的现场裁判在违规触袋时终止回合，stages=0/2；有效击打这一前置也未完成，护架步骤尚未开始。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "impulse_n_s",
+   "unit": "N*s",
+   "threshold": 12.0,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "contact_s",
+   "unit": "s",
+   "threshold": 0.1,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "hit_height_m",
+   "unit": "m",
+   "threshold": [
+    1.2,
+    1.6
+   ],
+   "direction": "target",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "bag_struck",
+   "t": 0.6,
+   "log_step": 156,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 0.52–0.60 s 机器人正向沙袋白带出拳，尚未收拳；log step 156 的最后一次现场回合 episode=6 在 control_steps=60 因右拳拳峰之外部位触袋而立即失败，stages=0/2。视频止于这次击打，未有后续修复。"
+  },
+  "unmet": {
+   "bag_struck": false,
+   "guard_held": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 0.6,
+    "log_step": 156,
+    "evidence_zh": "观察：视频 0.56–0.60 s 机器人右手及前臂伸至沙袋白带，末帧右臂仍伸出；log step 156 的现场裁判明确报告触袋部位不全是右拳拳峰。推断：击打时手臂与袋的相对位姿未把接触限制在���定拳峰区，属于击打接触位姿不准；不再将同一次违规接触重复标为违反约束。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "log_step": 156,
+    "evidence_zh": "推断：agent 的关键帧出拳控制未能在现场执行中维持规定的接触部位。log step 151、155 将出拳姿态、跟随段及驱动力矩写入脚本，log step 156 实际执行该脚本后，机器人服务报告非右拳拳峰触袋并终止，视频 0.52–0.60 s 同时显示右臂向袋击出。agent 在 log step 159 将原因解释为较强驱动和离线手部模型差异，但该细分解释缺少真实接触坐标验证，因此仅归为控制。",
+    "source_zh": "日志"
+   }
+  ],
+  "gap_zh": "无状态日志或逐接触记录；视频仅有场馆侧视且为 25 fps，无法分辨拳垫末端 0.03 m 的接触边界、具体违规部位及毫秒级接触起止。inputs/run.json 的奖励字段及最后一次服务输出未报告 impulse_n_s、contact_s、hit_height_m，故原生指标留 null；离线副本的数值不是本次现场回合的指标。尚不能区分接触几何设计、轨迹跟踪或力矩参数的具体贡献，agent 对模型差异和接触软硬的解释待核实。"
+ },
+ "summary_zh": "评分视频对应最后一次现场回合 episode=6：机器人站立蓄力后向沙袋白带出拳，但现场裁判因右拳拳峰之外部位触袋立即判负，有效击打与后续护架均未完成，最终为 0/2。断点在击打操作中，现象为接触位姿不准，根因归于出拳控制；录像无法定位具体违规接触点，原生冲量、接触时长和击中高度未报告。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "标签与证据一致：视频仅 0.64 s（16 帧）对应 episode 6 的 60 个控制步，末帧右臂已触及沙袋；log 步骤 id 156 的服务读数为 failed、stages 0/2，两个节点均为 false，断点 bag_struck 操作中合理，现象和根因均已标为推断并说明了不确定性。",
+    "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-boxing-i00-standard--uzi6vau/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-boxing-i00-standard--uzi6vau/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "7d8cf706b07a19778b4ccc2b34ca6b9f6a35a0fac825856a0cf677039b2b67f2",
+   "video_sha256": "309809651b52fddc3a00eaca44590e1eb9158c06d9ee3d8d0d37291fd185e566",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/boxing/trajectories/embodied-olympics-fun-boxing-i00-standard--vwthikh.json

@@ -0,0 +1,226 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-boxing-i00-standard--vwthikh",
+ "taskKey": "embodied-olympics-fun/boxing",
+ "dag_sha256": "5fe95f4fd88e6dbc49fa1e5ba7230e5fd61571eeb504fd1c0835f0ca7840e98d",
+ "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": 226,
+   "sim_steps": 1677,
+   "calls": 296,
+   "rows": 1677,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 3903.2,
+   "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": 4.445,
+   "peak_processes": 26,
+   "peak_disk_gb": 0.722,
+   "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-boxing-i00-standard--vwthikh/replay.mp4",
+  "duration_s": 2.28,
+  "fps": 25.0,
+  "frames": 57,
+  "sha256": "b23ee25775893e00fd1103b51c8c8470d4abfd35cad9215bbd3fa4ff8dd7060f"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "bag_struck": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.24,
+      "log_step": 150,
+      "text_zh": "观察：逐帧检查视频 0.00–2.24 s，机器人先抬起并调整右手，视频 2.00–2.24 s 右手向沙袋白带加速前伸，末帧手部到达袋面附近；没有随后拳袋分离、沙袋明显摆离的完整有效击打过程。log step 150 的最后一局 episode 6 由现场裁判返回 outcome=\"the bag was touched by something other than the knuckle part of the right fist\"、stages_reached=0、stages_total=2，故该次接触未被接纳为有效击打。"
+     },
+     {
+      "t": 0,
+      "log_step": 37,
+      "text_zh": "观察：log step 37 的服务规则明确只有最后一局受评；log step 150 重置进入 episode 6，log step 151 确认该局在 control_steps=226 时 failed。视频仅显示一次准备和末尾出拳，与该局相符；更早几局及本地仿真测试不计为本视频的进度。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "其他",
+      "note_zh": "候选名称：击打。视频 2.16–2.24 s 右手末端向沙袋白带出击，log step 150 的现场裁判确认发生包含非右拳拳峰部位的触袋；不能从该视角区分具体是手指还是拳垫其他区域。属于单手短促击袋尝试，现有接触方式无击打项，不是抓握。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–2.24 s，机器人双脚接地支承抬臂和出拳，腿、躯干保持站立，未见跌倒。"
+     }
+    ]
+   },
+   "guard_held": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.24,
+      "log_step": 150,
+      "text_zh": "观察：视频 2.00–2.24 s 机器人右手仍向袋前伸，视频在这次接触附近结束，没有击打后收回右拳并连续护架站稳的过程。log step 150 中脚本发送恢复段后立即收到 failed、stages_reached=0/2，未进入有效击打后的保持阶段。"
+     },
+     {
+      "t": 1.6,
+      "log_step": null,
+      "text_zh": "推断：视频 0.80–2.00 s 的屈肘站立是击打前准备，bag_struck 尚未达成，不能满足 guard_held 的前置条件；即使该准备姿态接近护架，也不提前计入此节点。成功参考录像在其视频约 0.24 s 后出现击打后的收拳保持，本次视频没有相应后续过程；参考录像仅用于辨认可见动作。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "impulse_n_s",
+   "unit": "N*s",
+   "threshold": 12.0,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "contact_s",
+   "unit": "s",
+   "threshold": 0.1,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "hit_height_m",
+   "unit": "m",
+   "threshold": [
+    1.2,
+    1.6
+   ],
+   "direction": "target",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "bag_struck",
+   "t": 2.24,
+   "log_step": 150,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 2.16–2.24 s 机器人右手前伸至白带袋面附近，身体仍站立；log step 150 对最后一局 episode 6 返回非右拳拳峰部位触袋的现场裁判判词，stages_reached=0/2。失败发生在这次击袋操作中，未完成有效击打，视频内也没有修复；视频 2.24 s 为末尾接触附近的粗略导航锚点。"
+  },
+  "unmet": {
+   "bag_struck": false,
+   "guard_held": false
+  },
+  "phenomena": [
+   {
+    "cat": "约束",
+    "item": "违反约束",
+    "t": 2.24,
+    "log_step": 150,
+    "evidence_zh": "观察：视频 2.16–2.24 s 右手末端到达沙袋白带附近；log step 150 中现场裁判明确返回“the bag was touched by something other than the knuckle part of the right fist”，随即结束该局，违反 wrong_bag_contact 所述只允许右拳拳峰触袋的规则。画面未见跌倒或被固定障碍阻挡，不另推断冲量不足、持续推袋或具体违规接触部位。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "抓取和接触设计",
+    "alt_feasible": "已验证有",
+    "log_step": 148,
+    "evidence_zh": "推断：最终击袋姿态的接触几何未能只让拳峰触袋。log step 148 的最终脚本把右腕偏航设在约 +1.27 rad，以侧向拳垫边缘先行；log step 149 的本地模型测试对最终采用的 acc=5、ex=1 仍报告拳垫接触后约 0.015 s 出现 right_finger_index 接触，log step 150 随后在现场得到非拳峰触袋判词。该模型是 agent 自建近似模型，不能证明真实接触的是哪根手指，但与视频末尾右手触袋及裁判结果共同支持接触布局不合适。可行替代仅由 inputs/oracle/oracle.json 的 success=1 和参考录像中其视频约 0.24 s 后收拳保持验证；不把参考动作计入本次运行。",
+    "source_zh": "日志"
+   }
+  ],
+  "gap_zh": "没有状态日志；视频仅有场馆侧视，手部像素不足以区分拳垫拳峰、侧缘与手指的具体接触，也不能独立测量冲量、接触历时或触点高度。inputs/run.json 的评分字段未报告 impulse_n_s、contact_s、hit_height_m，故三项原生指标均留 null；本地测试的约 14.9 N·s 不是现场裁判测量。agent 在 log step 150、152 所述手指突出尺寸和碰撞机理仍待核实，根因仅归到接触设计，不据此认定任务无合法解。"
+ },
+ "summary_zh": "受评的是最后一局 episode 6：机器人站立调整右臂后向沙袋白带出拳，视频约 2.24 s 的接触附近因非右拳拳峰部位触袋被现场裁判终止，有效击打和击后护架均未达成（0/2）。最终侧向出拳的接触布局未能避开违规触袋，与自建模型中仍出现手指接触的测试相符。缺少状态日志和现场冲量等数值，具体违规接触部位及细小几何尺寸无法独立核实。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "受评的最后一局 episode 6（log control_steps=226，约 2.26 s，与视频 2.28 s 吻合）：视频中右手贴近沙袋白带出拳，现场裁判返回非右拳拳峰部位触袋、stages_reached=0/2。bag_struck 与 guard_held 均为 false、断点 bag_struck/操作中、违反约束、接触设计根因均有 log 与评分字段支持；未发现需要改内容的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:48:55Z",
+  "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-boxing-i00-standard--vwthikh/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-boxing-i00-standard--vwthikh/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "03ad0022b2ff24f84e2ffc57f1aad263c39e7722a62d13a553b5936bc8023cc9",
+   "video_sha256": "b23ee25775893e00fd1103b51c8c8470d4abfd35cad9215bbd3fa4ff8dd7060f",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/golf_putt/trajectories/embodied-olympics-fun-golf-putt-i00-privileged--63uqpxj.json

@@ -0,0 +1,239 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-golf-putt-i00-privileged--63uqpxj",
+ "taskKey": "embodied-olympics-fun/golf_putt",
+ "dag_sha256": "74a1c00f893c3316873e4cc2017a91bda4a2926d26f9ae1a009d7dbb3a3e0854",
+ "outcome": {
+  "success": 0,
+  "reward": 0.5,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "n_actions": 1038,
+   "sim_steps": 1038,
+   "horizon": 3000,
+   "episode_over": 0,
+   "progress": 0.5,
+   "stages_reached": 1,
+   "stages_total": 2,
+   "fell": 0,
+   "replay_s": 8.71,
+   "bad_trajectory": 0,
+   "metric_ball_speed_m_s": 0.004,
+   "metric_ball_to_hole_m": 2.949,
+   "metric_grip_break_max_s": 0.004,
+   "metric_head_contact_s": 0.012,
+   "metric_head_hits": 1.0,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 139,
+   "agent_resets": 21,
+   "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": 0.943,
+   "peak_processes": 10,
+   "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-golf-putt-i00-privileged--63uqpxj/replay.mp4",
+  "duration_s": 10.4,
+  "fps": 25.0,
+  "frames": 260,
+  "sha256": "6fb3cb25bd57fecf88b70db14b09aef02aeda08794c70123aa9294967589f368"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "ball_holed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00-9.40 s，机器人双脚着地，右手持推杆缓慢转动躯干并调整杆头，白球仍在脚前，球洞位于画面右上方浅色条带上。"
+     },
+     {
+      "t": 9.52,
+      "log_step": null,
+      "text_zh": "观察：视频 9.44-10.36 s，机器人挥动推杆后白球向画面右侧（略偏下）短距离滚动，仍留在机器人脚前，未向右上方洞口滚去，更未落入洞中；已逐帧检查视频 9.28-10.36 s。"
+     },
+     {
+      "t": 10.36,
+      "log_step": 212,
+      "text_zh": "观察：log step 212 确认最终保存并重放的是 1038 行轨迹，仅有 stroke 阶段，球距洞 2.949 米；inputs/run.json 评分器字段同样为 stages_reached=1/2、metric_head_hits=1、metric_head_contact_s=0.012、metric_ball_to_hole_m=2.949、success=0。视频 10.36 s 白球仍在脚前，与该结果一致。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手持工具",
+      "how": "其他",
+      "note_zh": "观察：视频约 9.5 s，机器人右手挥动所持推杆，杆头经过白球后白球开始滚动；评分器记录杆头触球 1 次、累计 0.012 秒。候选接触名为“击打”，现有接触方式无击打项；按工具至白球链路记录，不重复标握柄。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00-10.36 s，机器人双脚在果岭上支撑身体，转动躯干并用右手调整和挥动推杆；击球及后续球滚动时未见跌倒。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "ball_to_hole_m",
+   "unit": "m",
+   "threshold": 0.054,
+   "direction": "lower_better",
+   "value": 2.949
+  },
+  {
+   "name": "ball_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": "lower_better",
+   "value": 0.004
+  },
+  {
+   "name": "head_contact_s",
+   "unit": "s",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": 0.012
+  },
+  {
+   "name": "grip_break_max_s",
+   "unit": "s",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": 0.004
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "ball_holed",
+   "t": 9.5,
+   "log_step": null,
+   "where": "操作中",
+   "evidence_zh": "观察：视频约 9.5 s，机器人挥杆触球后，白球向画面右侧（略偏下）短距离滚动，偏离通往右上方球洞的路线，仍在脚前；视频 10.36 s 仍未入洞。评分器记录一次杆头触球而仅达到 stroke 阶段，末态球距洞 2.949 米；失败发生在一杆推球入洞的击球操作中。"
+  },
+  "unmet": {
+   "ball_holed": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 9.5,
+    "log_step": null,
+    "evidence_zh": "观察：视频 9.44-10.36 s，机器人挥杆后白球向画面右侧（略偏下）短距离滚动，球的路线没有朝向画面右上方浅色条带上的球洞；至视频 10.36 s，球仍在机器人附近。评分器记录末态球距��� 2.949 米，未完成入洞。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "source_zh": "日志",
+    "evidence_zh": "推断：log step 209 的交付脚本先调整右手及杆面朝向，再命令右手沿 +y 挥动；会话中同配置试验的输出显示击球准备时杆面中心 x=0.3544 米，而初始球心 x=0.44 米，击出后的球速主要沿 +x（约 [0.55,0.14,0] 米/秒），而球洞在 +y 方向。log step 211-212 将该配置生成并保存为最终 1038 行轨迹，重放仅达到 stroke、球距洞 2.949 米；视频 9.44-10.36 s 同样显示侧向短滚。因此可归因于击球对位及挥杆控制未能产生朝洞的有效击球，具体控制子类证据不足。"
+   }
+  ],
+  "gap_zh": "没有逐帧状态日志，远景无法独立核实杆头具体哪一部位触球，也无法核实 agent 在 log step 251 自述的关节饱和、承载下沉及杆头趾部擦球机制，因此根因停在“控制”。视频为 25 fps，毫秒级接触时长取自评分器字段，断点仅锚定视频约 9.5 s。"
+ },
+ "summary_zh": "机器人完成了持杆调整及一次触球，但白球只在脚前侧向短滚，未入洞，唯一语义节点最终为 0/1。断点位于视频约 9.5 s 的击球操作，现象为位姿不准；交付脚本及同配置试验支持击球对位与挥杆控制不足，评分器末态球距洞 2.949 米。缺少逐帧状态和杆头近景，不能进一步确认具体接触或跟踪机制；会话后续滚得更远的测试版本不属于最终交付。",
+ "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(final=false)、断点 ball_holed/操作中、位姿不准加控制根因与评分器字段及帧一致；仅球滚动方向的描述与视频不符，需小改。",
+    "n_issues": 1,
+    "n_major": 0
+   },
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 2,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "节点未达成、断点 ball_holed/操作中、位姿不准与评分器字段（stages 1/2、球距洞 2.949 m）及视频一致；上一轮方向描述已修正，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": "2026-10-31T00:59:45Z",
+  "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-golf-putt-i00-privileged--63uqpxj/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-golf-putt-i00-privileged--63uqpxj/04_review_r1"
+   },
+   {
+    "stage": "reconcile",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-golf-putt-i00-privileged--63uqpxj/05_reconcile_r1"
+   },
+   {
+    "stage": "traj_review",
+    "round": 2,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-golf-putt-i00-privileged--63uqpxj/04_review_r2"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "357f10b2224cc60fb23f4cad56cfe07a0b838b72bebff42ee7c488f6f3db917e",
+   "video_sha256": "6fb3cb25bd57fecf88b70db14b09aef02aeda08794c70123aa9294967589f368",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/golf_putt/trajectories/embodied-olympics-fun-golf-putt-i00-privileged--btzgldk.json

@@ -0,0 +1,201 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-golf-putt-i00-privileged--btzgldk",
+ "taskKey": "embodied-olympics-fun/golf_putt",
+ "dag_sha256": "74a1c00f893c3316873e4cc2017a91bda4a2926d26f9ae1a009d7dbb3a3e0854",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "n_actions": 664,
+   "sim_steps": 664,
+   "horizon": 3000,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 2,
+   "stages_total": 2,
+   "fell": 0,
+   "replay_s": 5.28,
+   "bad_trajectory": 0,
+   "metric_ball_speed_m_s": 0.0,
+   "metric_ball_to_hole_m": 0.002,
+   "metric_grip_break_max_s": 0.007,
+   "metric_head_contact_s": 0.025,
+   "metric_head_hits": 2.0,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 961,
+   "agent_resets": 260,
+   "agent_replays": 52,
+   "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-golf-putt-i00-privileged--btzgldk/replay.mp4",
+  "duration_s": 6.68,
+  "fps": 25.0,
+  "frames": 167,
+  "sha256": "9ef647976e496b781f1cf3ebf11162adafece52ebe6a2710f05507711fd49433"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "ball_holed": {
+    "method": "visual",
+    "intervals": [
+     [
+      6.64,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00 s，白球位于机器人脚前，远处浅色条带上可见洞口；机器人右手已持推杆，双脚落在果岭上，入洞条件尚未成立。"
+     },
+     {
+      "t": 2.36,
+      "log_step": null,
+      "text_zh": "观察：逐帧检查视频 2.08–2.68 s，机器人右臂带动推杆挥过脚前白球，随后白球离开原位向画面右上方的球洞运动；视频 2.60–4.48 s 可连续追踪白球沿果岭接近洞口，机器人双脚保持支撑。"
+     },
+     {
+      "t": 4.96,
+      "log_step": null,
+      "text_zh": "观察：视频 4.48–4.96 s 的洞口放大序列中，白球从洞口近侧滚至洞沿后消失；视频 5.60–6.64 s 的密集末段帧中，洞口及附近果岭不再露出白球，机器人仍站立持杆。远景看不到洞内球体，不能直接量出球顶高度或毫米级静止。"
+     },
+     {
+      "t": 6.64,
+      "log_step": 330,
+      "text_zh": "观察：log step 329 重新运行交付文件的生成脚本；log step 330 对最终 trajectory.npz 的两次独立重放均由仿真裁判返回 success=true、failed=null、stages=[stroke, holed]，执行 664 个控制步，time_s=6.64，末态哈希一致。两次事件记录均为仿真 2.358 s 的 stroke、仿真 6.64 s 的 holed；run.json 的评分器字段 success=1、replay_success=1、deterministic=1 与之相符。"
+     },
+     {
+      "t": 6.64,
+      "log_step": 330,
+      "text_zh": "推断：log step 330 中最终交付的 664 控制步对应仿真 6.64 s，即每步 0.01 s；视频共 167 帧、25 fps，末帧为视频 6.64 s（封装时长 6.68 s）。重放的仿真 2.358 s 击球事件与视频约 2.36–2.60 s 的挥杆及白球离开原位相符，据此将末次 holed 验收定位到视频 6.64 s；结合白球在洞口消失并一直未再出现，节点自该时刻成立且终值为真。洞内停稳 1.0 秒及细粒度接触合规由最终重放裁判确认，并非仅凭白球消失提前记为完成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手持工具",
+      "how": "其他",
+      "note_zh": "观察：视频约 2.36–2.68 s，机器人右手持推杆挥过白球，白球随后离开脚前并滚向球洞。作用端为推杆杆头、对象为白球；候选接触名为“击打”，现有方式无击打项，因此用“其他”；握杆属于同一工具链，不重复计为抓握。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–6.64 s，机器人双脚接触果岭支撑身体，挥杆击球及随后等待期间均保持站立。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "ball_to_hole_m",
+   "unit": "m",
+   "threshold": 0.054,
+   "direction": "lower_better",
+   "value": 0.002
+  },
+  {
+   "name": "ball_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": "lower_better",
+   "value": 0.0
+  },
+  {
+   "name": "head_contact_s",
+   "unit": "s",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": 0.025
+  },
+  {
+   "name": "grip_break_max_s",
+   "unit": "s",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": 0.007
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人完成“一杆推球入洞”，终态为 1/1：视频约 2.36–2.60 s 右手持杆击球，白球约在视频 4.9 s 到洞口后消失，机器人保持站立持杆至结束。最终交付的两次重放在 log step 330 均记录 stroke 后 holed、成功且末态哈希一致，完整停稳验收对应视频 6.64 s，未发现失败断点。无状态日志，远景无法直接核验洞内毫米级静止、球顶高度和毫秒级接触，相关判据采用最终重放裁判及评分器结果；没有另用独立成功参考视频。",
+ "conf": 7,
+ "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，单节点 ball_holed 终值为真；时钟映射（664 步×0.01 s=6.64 s）、stroke 2.358 s 与视频挥杆相符，接触标签合理，无失败断点，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": "2026-10-31T00:45:04Z",
+  "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-golf-putt-i00-privileged--btzgldk/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-golf-putt-i00-privileged--btzgldk/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "2f804a776c08be9ce2182374a377b3a057b0376015bd42f75f8cb9c80b2ce8ab",
+   "video_sha256": "9ef647976e496b781f1cf3ebf11162adafece52ebe6a2710f05507711fd49433",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/golf_putt/trajectories/embodied-olympics-fun-golf-putt-i00-privileged--cbn3ffv.json

@@ -0,0 +1,198 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-golf-putt-i00-privileged--cbn3ffv",
+ "taskKey": "embodied-olympics-fun/golf_putt",
+ "dag_sha256": "74a1c00f893c3316873e4cc2017a91bda4a2926d26f9ae1a009d7dbb3a3e0854",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "n_actions": 782,
+   "sim_steps": 782,
+   "horizon": 3000,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 2,
+   "stages_total": 2,
+   "fell": 0,
+   "replay_s": 7.39,
+   "bad_trajectory": 0,
+   "metric_ball_speed_m_s": 0.003,
+   "metric_ball_to_hole_m": 0.026,
+   "metric_grip_break_max_s": 0.001,
+   "metric_head_contact_s": 0.013,
+   "metric_head_hits": 1.0,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 72,
+   "agent_resets": 56,
+   "agent_replays": 5,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 0.664,
+   "peak_processes": 10,
+   "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-golf-putt-i00-privileged--cbn3ffv/replay.mp4",
+  "duration_s": 7.84,
+  "fps": 25.0,
+  "frames": 196,
+  "sha256": "91413cc97a9c432b42b89a3231a0671ba73a4a20b55cc78eb33c06b9eca8649e"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "ball_holed": {
+    "method": "visual",
+    "intervals": [
+     [
+      7.8,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 4.12,
+      "log_step": null,
+      "text_zh": "观察：视频 3.76–4.08 s，机器人右手持推杆将杆头挥向脚前白球；视频 4.12–4.96 s，白球离开原位并沿果岭向浅色条带上的洞口滚动，机器人双脚持续着地支撑。"
+     },
+     {
+      "t": 6.32,
+      "log_step": null,
+      "text_zh": "观察：视频 5.84–6.16 s 白球接近洞沿，视频约 6.2–6.4 s 白球在洞口处落下；视频 6.4–7.80 s 洞外及洞沿没有可辨识的露出白球，机器人仍站立持杆。远景无法辨明球在洞内的毫米级运动。"
+     },
+     {
+      "t": 7.8,
+      "log_step": 74,
+      "text_zh": "观察：inputs/run.json 的评分器字段 success=1、replay_success=1、deterministic=1、stages_reached=2/2、metric_head_hits=1、fell=0；log step 74 中 agent 重新生成最终交付文件后，两次官方重放均返回 success 且末态相同，阶段为 stroke、holed，指标与最终评分一致。"
+     },
+     {
+      "t": 7.8,
+      "log_step": 74,
+      "text_zh": "推断：结合视频约 6.3 s 落洞及视频 7.80 s 末帧仍无洞外白球、最终评分成功和 log step 74 的最终文件重放结果，将包含洞内停稳 1.0 秒的完整节点标为在视频约 7.8 s 达成并保持至结束；这是验收完成的近似末帧锚点，远景不能确定毫米级停稳窗口的精确起点。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手持工具",
+      "how": "其他",
+      "note_zh": "观察：视频约 4.08–4.16 s，机器人右手持推杆挥过白球原位，白球随即滚离；评分器 metric_head_hits=1、metric_head_contact_s=0.013 支持杆头击球。候选接触名为“击打”，现有方式无对应条目；仅记录推杆杆头至白球的工具链。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0–7.80 s，机器人双脚着地，原地站立完成调杆、击球和等待白球入洞，未见跌倒。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "ball_to_hole_m",
+   "unit": "m",
+   "threshold": 0.054,
+   "direction": "lower_better",
+   "value": 0.026
+  },
+  {
+   "name": "ball_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": "lower_better",
+   "value": 0.003
+  },
+  {
+   "name": "head_contact_s",
+   "unit": "s",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": 0.013
+  },
+  {
+   "name": "grip_break_max_s",
+   "unit": "s",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": 0.001
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人右手持杆、双脚原地支撑，在视频约 4.1 s 击出白球，白球沿果岭滚动并于视频约 6.3 s 落入洞口；完整的一杆入洞及停稳节点在视频约 7.8 s 标为完成，最终达到 1/1。评分器判定成功，最终交付文件的两次官方重放结果一致，单次杆头击球累计接触 0.013 秒。没有状态日志，远景无法独立核验全球���于表面及毫米级停稳窗口，精细验收依据评分器与最终文件重放结果。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "成功运行，单节点 ball_holed 达成、final=true，与评分器 success=1、stages 2/2、metric_ball_to_hole 0.026 一致；远景下击球与滚入洞口的描述与帧相符，接触标注（手持工具/其他、腿/躯干/支撑）合理，无失败层。7.8 s 为近似锚点，证据已明示其为推断。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:44:50Z",
+  "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-golf-putt-i00-privileged--cbn3ffv/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-golf-putt-i00-privileged--cbn3ffv/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "5bf0706f638da880ac0ceb71545099a54b5870247fc4080e306f68b9fcc054de",
+   "video_sha256": "91413cc97a9c432b42b89a3231a0671ba73a4a20b55cc78eb33c06b9eca8649e",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/golf_putt/trajectories/embodied-olympics-fun-golf-putt-i00-privileged--hvt294e.json

@@ -0,0 +1,203 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-golf-putt-i00-privileged--hvt294e",
+ "taskKey": "embodied-olympics-fun/golf_putt",
+ "dag_sha256": "74a1c00f893c3316873e4cc2017a91bda4a2926d26f9ae1a009d7dbb3a3e0854",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "n_actions": 413,
+   "sim_steps": 413,
+   "horizon": 3000,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 2,
+   "stages_total": 2,
+   "fell": 0,
+   "replay_s": 4.89,
+   "bad_trajectory": 0,
+   "metric_ball_speed_m_s": 0.0,
+   "metric_ball_to_hole_m": 0.022,
+   "metric_grip_break_max_s": 0.001,
+   "metric_head_contact_s": 0.012,
+   "metric_head_hits": 1.0,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 86,
+   "agent_resets": 309,
+   "agent_replays": 5,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 0.743,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.253,
+   "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-golf-putt-i00-privileged--hvt294e/replay.mp4",
+  "duration_s": 4.16,
+  "fps": 25.0,
+  "frames": 104,
+  "sha256": "d0e9341a8275b37152baaca7a59bdfbc3783d68f6e675230b83154d54296bbbd"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "ball_holed": {
+    "method": "visual",
+    "intervals": [
+     [
+      4.12,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.96,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–0.92 s，机器人双脚支撑于果岭，右手持推杆调整并回摆；视频 0.96–1.04 s 杆头向前扫过白球原位，白球离开脚前位置，随后沿果岭朝远处浅色条带上的洞口滚动。"
+     },
+     {
+      "t": 2.6,
+      "log_step": null,
+      "text_zh": "观察：视频 2.24–2.60 s，白球逐渐接近圆形洞口，随后在洞口处不再可辨；逐帧检查视频 3.12–4.12 s，洞口及洞沿均未再见白球，机器人继续站立持杆。成功参考视频使用相同远景视角，入洞后也呈现洞口无白球的外观；参考只用于辨认视觉线索。"
+     },
+     {
+      "t": 4.12,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器 reward 记录 success=1、replay_success=1、deterministic=1、stages_reached=2/2、episode_over=1，末态球距洞心 0.022 m、球速 0 m/s，杆头触球累计 0.012 s、head_hits=1。"
+     },
+     {
+      "t": 4.12,
+      "log_step": null,
+      "text_zh": "推断：视频中的击球、滚向洞口并消失与评分器成功结果一致，故一杆入洞且洞内停稳验收成立，final=true。远景不能测量全球高度、毫米级位移或洞内停稳计时起点，达成时间保守取视频 4.12 s 末帧作近似锚点，精确首次成立时间未知；未把 agent 工作时间或重放耗时换算为视频时间。"
+     },
+     {
+      "t": null,
+      "log_step": 153,
+      "text_zh": "观察：log step 147 的脚本生成并保存击球轨迹；log step 153 写出的最终生成器包含准备、回摆、前挥及等待回合结束四段，log step 154 执行生成器，log step 157 发起交付文件重放检查，log step 163 读出交付轨迹为 413 行有限动作。agent 在 log step 166 自述已一杆入洞；节点结论依据本次视频和评分器结果，未把会话调试回合当作评分回合。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手持工具",
+      "how": "其他",
+      "note_zh": "观察：视频 0.92–1.04 s，机器人右手持推杆前挥，杆头经过白球原位后白球向洞口运动；评分器记录一次杆头接触。候选接触名为“击打”，现有方式无对应项；只记录杆头至白球的工具接触链。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–4.12 s，机器人双脚接触果岭承重，挥杆及等待球滚动时持续站立。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "ball_to_hole_m",
+   "unit": "m",
+   "threshold": 0.054,
+   "direction": "lower_better",
+   "value": 0.022
+  },
+  {
+   "name": "ball_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": "lower_better",
+   "value": 0.0
+  },
+  {
+   "name": "head_contact_s",
+   "unit": "s",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": 0.012
+  },
+  {
+   "name": "grip_break_max_s",
+   "unit": "s",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": 0.001
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人右手持推杆击出白球，白球沿果岭滚至洞口并消失，评分器确认一杆入洞成功，唯一节点完成，终态为 1/1。视频 3.12–4.12 s 洞口及洞沿未再见白球，评分器同时记录确定性重放成功、末态球速为零；无失败断点。没有状态日志，远景无法精确复核洞内高度、毫米级停稳及毫秒级接触，节点达成时间以视频 4.12 s 末帧为保守近似锚点。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "单节点 ball_holed 成功，视频 2.5 s 左右白球入洞消失，之后洞口无球，评分器 success=1；4.12 s 作为含 1 s 停稳窗口的达成锚点与我的判断（约 3.5 s）相差不到一秒，接触标注与 DAG 一致且符合画面，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": "2026-10-31T00:44:53Z",
+  "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-golf-putt-i00-privileged--hvt294e/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-golf-putt-i00-privileged--hvt294e/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "63993beb9638d28408d2acc3895e64aaaa0f925c41108f931c965c654c3e627d",
+   "video_sha256": "d0e9341a8275b37152baaca7a59bdfbc3783d68f6e675230b83154d54296bbbd",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/golf_putt/trajectories/embodied-olympics-fun-golf-putt-i00-privileged--uawrqwe.json

@@ -0,0 +1,198 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-golf-putt-i00-privileged--uawrqwe",
+ "taskKey": "embodied-olympics-fun/golf_putt",
+ "dag_sha256": "74a1c00f893c3316873e4cc2017a91bda4a2926d26f9ae1a009d7dbb3a3e0854",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "n_actions": 597,
+   "sim_steps": 597,
+   "horizon": 3000,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 2,
+   "stages_total": 2,
+   "fell": 0,
+   "replay_s": 9.49,
+   "bad_trajectory": 0,
+   "metric_ball_speed_m_s": 0.0,
+   "metric_ball_to_hole_m": 0.025,
+   "metric_grip_break_max_s": 0.003,
+   "metric_head_contact_s": 0.013,
+   "metric_head_hits": 1.0,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 34,
+   "agent_resets": 29,
+   "agent_replays": 4,
+   "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": 0.702,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.251,
+   "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-golf-putt-i00-privileged--uawrqwe/replay.mp4",
+  "duration_s": 6.0,
+  "fps": 25.0,
+  "frames": 150,
+  "sha256": "a770514f7c3c43f2242e6806b24252395211af938e1fca3a493387609cfa0471"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "ball_holed": {
+    "method": "visual",
+    "intervals": [
+     [
+      5.96,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 2.4,
+      "log_step": null,
+      "text_zh": "观察：视频 2.24–2.68 s 的逐帧画面中，机器人双脚站在果岭上，右手持推杆向白球挥动；视频 2.40 s 白球仍在杆头前，视频 2.44–2.68 s 白球离开原位并向远处洞口滚动，杆头留在机器人附近。"
+     },
+     {
+      "t": 4.2,
+      "log_step": null,
+      "text_zh": "观察：视频 3.20–4.16 s 白球持续靠近浅色条带上的圆形洞口；视频约 4.20–4.24 s 白球在洞口处消失。逐帧检查视频 4.40–5.96 s，洞口及洞沿未见白球重新露出；末帧机器人仍站立持杆。参考成功视频的末尾同样只见洞口，远景均不能显示洞底细节。"
+     },
+     {
+      "t": 5.96,
+      "log_step": null,
+      "text_zh": "推断：本次交付的评分器 reward 记录 success=1、replay_success=1、deterministic=1、stages_reached=2/2，终态 ball_to_hole_m=0.025、ball_speed_m_s=0，结合白球在洞口消失且未再露出的画面，判定入洞停稳验收已通过。视频不能精确显示洞内静止窗口的起止，故将达成时间近似锚定于已检查的末帧视频 5.96 s（约视频 6.0 s），不把白球初入洞时刻当作完成停稳。"
+     },
+     {
+      "t": null,
+      "log_step": 73,
+      "text_zh": "观察：log step 59 修正生成脚本的保存调用，log step 60 调用修正后的脚本生成并重放交付文件；log step 73 显示最终脚本仍采用抬杆中间姿态、击球前停留及单次挥杆方案。log step 76 中 agent 自述交付含 597 个控制步并已独立重放；这些会话记录用于核对交付内容，节点值依据本次视频和评分器结果。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手持工具",
+      "how": "其他",
+      "note_zh": "观察：视频 2.40–2.68 s，机器人右手握持的推杆杆头击出白球，白球随即滚离；候选接触名为“击打”，现有接触方式没有对应项，短促击球不属推/扫。评分器记录 head_hits=1、head_contact_s=0.013 秒；工具链仅记录杆头与白球，不重复记录握杆。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–5.96 s，机器人双脚在果岭上支撑身体，右臂完成移杆、挥杆及随挥，击球和等待白球入洞期间均保持站立。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "ball_to_hole_m",
+   "unit": "m",
+   "threshold": 0.054,
+   "direction": "lower_better",
+   "value": 0.025
+  },
+  {
+   "name": "ball_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": "lower_better",
+   "value": 0.0
+  },
+  {
+   "name": "head_contact_s",
+   "unit": "s",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": 0.013
+  },
+  {
+   "name": "grip_break_max_s",
+   "unit": "s",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": 0.003
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人右手持推杆约���视频 2.4 s 击出白球，白球约在视频 4.2 s 于洞口消失，至视频 5.96 s 末帧未再露出；评分器确认入洞停稳成功，唯一节点最终完成。节点完成时间近似标于视频末尾；没有状态日志，远景无法独立精确复核全球高度、毫米级静止窗口及毫秒级接触，相关验收由本次交付的评分器结果补足。",
+ "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，唯一节点 ball_holed 终值 true；视频 4.24 s 球在洞口消失、末帧未再露出，与评分器结果一致。时间锚点 5.96 s 略晚于洞内停稳约 5.2 s，但在容差内；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": "2026-10-31T00:44:50Z",
+  "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-golf-putt-i00-privileged--uawrqwe/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-golf-putt-i00-privileged--uawrqwe/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "6b4652d4b4296feaf7a9d117baebb77c55e3fc68d3fedaf1de1ea17c5c9ffc45",
+   "video_sha256": "a770514f7c3c43f2242e6806b24252395211af938e1fca3a493387609cfa0471",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/golf_putt/trajectories/embodied-olympics-fun-golf-putt-i00-standard--ac3ubxs.json

@@ -0,0 +1,222 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-golf-putt-i00-standard--ac3ubxs",
+ "taskKey": "embodied-olympics-fun/golf_putt",
+ "dag_sha256": "74a1c00f893c3316873e4cc2017a91bda4a2926d26f9ae1a009d7dbb3a3e0854",
+ "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": 1157,
+   "sim_steps": 3341,
+   "calls": 35,
+   "rows": 1501,
+   "resets": 3,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 7064.3,
+   "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": 0.0,
+   "peak_memory_gb": 1.304,
+   "peak_processes": 12,
+   "peak_disk_gb": 2.542,
+   "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-golf-putt-i00-standard--ac3ubxs/replay.mp4",
+  "duration_s": 11.6,
+  "fps": 25.0,
+  "frames": 290,
+  "sha256": "740b20fe0bc3fa7d36ed9a369b7066e524438ac1eb1a4c03954cb687e76fa5eb"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "ball_holed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00 s，机器人右手持推杆站在果岭上，白球在脚前，球洞位于远处浅色条带上，初始未入洞。"
+     },
+     {
+      "t": 3.8,
+      "log_step": 62,
+      "text_zh": "观察：视频 2.80–4.40 s 的逐帧画面中，机器人转动腰部、右手带杆先回摆再前挥，白球始终留在脚前，没有随挥杆滚向球洞。log step 62 的服务记录为 episode=4，提交 557 个控制步后又观察 600 个控制步，总计 1157 步，现场尚未成功。"
+     },
+     {
+      "t": 11.56,
+      "log_step": 64,
+      "text_zh": "观察：视频 10.60–11.56 s 的末秒画面及末帧，白球仍在脚前原处，机器人持杆站立，未出现白球滚到洞口或落洞的过程。log step 64 中 agent 自述“Ball never moved — a whiff”，与画面一致。"
+     },
+     {
+      "t": 5.6,
+      "log_step": 62,
+      "text_zh": "推断：log step 3 给出控制周期 0.01 s，log step 62 的本回合 1157 个控制步对应 11.57 s，与视频 11.60 s 长度吻合；557 步对应视频约 5.57 s，已用视频 5.60 s 挥杆结束后的持杆静止姿态核对。该视频对应最后的 episode=4；log step 54 中 episode=3 的击球及洞外停球属于重置前测试，不计入本次进度。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "观察：视频 0.00–11.56 s，机器人右手围握推杆握柄并带杆挥动；本次杆头未击中白球，因此不填写手持工具击球接触。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–11.56 s，机器人双脚接触果岭支撑身体，腰部转动挥杆期间保持站立。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "ball_to_hole_m",
+   "unit": "m",
+   "threshold": 0.054,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "ball_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "head_contact_s",
+   "unit": "s",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "grip_break_max_s",
+   "unit": "s",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "ball_holed",
+   "t": 3.8,
+   "log_step": 62,
+   "where": "操作中",
+   "evidence_zh": "观察：视频约 3.60–4.00 s，机器人右手持推杆完成前挥，但杆头从白球旁经过，白球没有移动；至视频 11.56 s 仍未入洞，后续没有补救动作。log step 62 是最后一次重置后的现场回合，log step 54 的前一回合击球结果不属于本次。"
+  },
+  "unmet": {
+   "ball_holed": false
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没建立接触",
+    "t": 3.8,
+    "log_step": 62,
+    "evidence_zh": "观察：视频 2.80–4.40 s 逐帧可见推杆回摆和前挥，白球保持在脚前同一位置；视频 10.60–11.56 s 仍在原处。推断：这次挥杆未建立将白球击出的杆头接触，��非击球后未能入洞。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "t": 3.8,
+    "log_step": 64,
+    "source_zh": "日志",
+    "evidence_zh": "推断：log step 61 计算对准球后的右臂目标，log step 62 将该目标用于实际挥杆；log step 64 读取这次提交保存的本体感知，腰部、右肘及右肩的采样实际角度接近对应指令，但视频 3.60–4.00 s 的实际杆头仍未触球，说明挥杆控制没有使实际球杆路径与球匹配。agent 在 log step 65 将其解释为握柄内滑动，此解释没有独立接触或球杆姿态数据验证，故根因仅标控制，不细分。"
+   }
+  ],
+  "gap_zh": "没有状态日志；远景画面能确认挥杆后白球不动，但不能精确区分杆头偏离是由握柄内转动、实际球杆几何与局部模型不一致，还是其他控制误差引起。agent 在 log step 65 所称的滑动尚未独立核实；评分器未报告 DAG 的四项原生指标，均填 null。"
+ },
+ "summary_zh": "本次评分视频对应最后一次现场回合 episode=4：机器人保持站立、右手持杆完成回摆和前挥，但白球始终留在脚前，一杆推球入洞节点未达，最终为 0/1。断点在视频约 3.80 s 的挥杆操作，现象为没建立接触，根因归控制；远景及缺少状态日志使具体杆头偏离机制无法核实。日志中此前重置回合的击球不计入本次。",
+ "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、log 一致：视频 11.6 s 对应 episode 4 的 1157 个控制步（0.01 s/步），白球全程留在原处，ball_holed 未达，断点在挥杆，现象没建立接触、根因控制（不细分）均有证据且已声明不确定性。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": "2026-10-31T01:00:27Z",
+  "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-golf-putt-i00-standard--ac3ubxs/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-golf-putt-i00-standard--ac3ubxs/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "4e8bbf4c093d304a867ee2cb1d63a29954b9aa4eb998fd09593a502b83bc147d",
+   "video_sha256": "740b20fe0bc3fa7d36ed9a369b7066e524438ac1eb1a4c03954cb687e76fa5eb",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/golf_putt/trajectories/embodied-olympics-fun-golf-putt-i00-standard--ekj9qc2.json

@@ -0,0 +1,191 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-golf-putt-i00-standard--ekj9qc2",
+ "taskKey": "embodied-olympics-fun/golf_putt",
+ "dag_sha256": "74a1c00f893c3316873e4cc2017a91bda4a2926d26f9ae1a009d7dbb3a3e0854",
+ "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": 647,
+   "sim_steps": 1409,
+   "calls": 6,
+   "rows": 1409,
+   "resets": 1,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 3154.1,
+   "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": 0.895,
+   "peak_processes": 7,
+   "peak_disk_gb": 0.361,
+   "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-golf-putt-i00-standard--ekj9qc2/replay.mp4",
+  "duration_s": 6.48,
+  "fps": 25.0,
+  "frames": 162,
+  "sha256": "36afd362c723d0208d5f1d935c30f420fb8ee262bea94939c65088a26592eb5a"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "ball_holed": {
+    "method": "visual",
+    "intervals": [
+     [
+      6.47,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 2.8,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–2.6 s，机器人双脚站在果岭上，右手持推杆调整至白球后方；视频 2.72–2.84 s，右臂挥杆，杆头经过白球，白球离开脚前原位并朝远处浅色条带上的洞口滚动。"
+     },
+     {
+      "t": 5.0,
+      "log_step": null,
+      "text_zh": "观察：视频 4.20–4.92 s 的逐帧洞口裁剪中，白球从洞口近侧接近并与洞口重合；视频约 5.0 s 后球不再露出，直至末帧视频 6.44 s 未见球重新出现在洞沿或果岭。远景能显示滚入洞口的过程，不能分辨洞内球顶高度与毫米级位移。"
+     },
+     {
+      "t": 6.47,
+      "log_step": 130,
+      "text_zh": "推断：log step 126–127 执行重置后的第二回合，log step 130 的机器人服务末态为 episode=2、control_steps=647、done=true、reason=success，评分器 reward 的 live_success=1、graded_by_live=1。log step 3/6 给出 control_dt=0.01 s，因此本回合终止对应视频约 647×0.01=6.47 s，与视频总长 6.48 s 一致；log step 94/124 的脚本在本回合控制时钟 2.5 s 后挥杆，已与视频 2.6–2.84 s 的实际挥杆核对。结合入洞后不再露出的画面，以视频约 6.47 s 作为含洞内停稳一秒在内的首次完成时刻；不是把 agent 工作时间换算为视频时间。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手持工具",
+      "how": "其他",
+      "note_zh": "观察：视频 2.72–2.84 s，机器人右手所持推杆的杆头经过白球并将球击出，随后球独立滚向洞口；候选接触名为“击打”，现有方式无对应项。按工具至球的链路记录，不另列握杆接触。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.0–6.44 s，机器人双脚支撑于果岭，站立调整右臂、挥杆并保持随挥姿态，未见跌倒。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "ball_to_hole_m",
+   "unit": "m",
+   "threshold": 0.054,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "ball_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "head_contact_s",
+   "unit": "s",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "grip_break_max_s",
+   "unit": "s",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "受评的是重置后的第二回合：机器人右手持杆击出白球，白球滚入洞口后未再露出，服务与评分器均判成功，唯一节点“一杆推球入洞”在视频约 6.47 s 完成。第一回合未进洞的调试结果不计入本次进度；本次无失败断点。没有状态日志，远景无法独立核验洞内毫米级停稳、球顶高度及毫秒级接触阈值，这些细节依据本回合成功裁决；四项原生指标未在评分器字段报告，均记 null。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "成功回合，唯一节点 ball_holed 在视频约 6.47 s 完成，与视频、控制步数 647×0.01 s 及评分器一致；接触与无失败断点均合理。",
+    "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": "2026-10-31T00:45:42Z",
+  "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-golf-putt-i00-standard--ekj9qc2/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-golf-putt-i00-standard--ekj9qc2/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "e2844cbc9479f5e90a9ab96e0ad6b7383b1410a5a1bb2605a5b7ac22441b2891",
+   "video_sha256": "36afd362c723d0208d5f1d935c30f420fb8ee262bea94939c65088a26592eb5a",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/golf_putt/trajectories/embodied-olympics-fun-golf-putt-i00-standard--he4xdrj.json

@@ -0,0 +1,238 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-golf-putt-i00-standard--he4xdrj",
+ "taskKey": "embodied-olympics-fun/golf_putt",
+ "dag_sha256": "74a1c00f893c3316873e4cc2017a91bda4a2926d26f9ae1a009d7dbb3a3e0854",
+ "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": 1503,
+   "sim_steps": 13580,
+   "calls": 55,
+   "rows": 13580,
+   "resets": 14,
+   "reset_budget": 50,
+   "sim_startups": 1,
+   "elapsed_s": 5149.6,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 1.296,
+   "peak_processes": 10,
+   "peak_disk_gb": 1.955,
+   "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-golf-putt-i00-standard--he4xdrj/replay.mp4",
+  "duration_s": 15.04,
+  "fps": 25.0,
+  "frames": 376,
+  "sha256": "1daade181cb7928775e7a6aa5d0a02fcd3509b2885d1f6439704cb9f0ad5408f"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "ball_holed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00 s 白球在机器人脚前，球洞在远处浅色条带上；机器人右手已持推杆。"
+     },
+     {
+      "t": 6.4,
+      "log_step": null,
+      "text_zh": "观察：视频 5.60–8.20 s 机器人右手带动推杆摆动，杆头在白球的图像左侧（起点侧）摆动，未到达白球，白球未见位移；视频 8.40–8.88 s 收杆时杆头投影靠近白球后移至其图像右侧，白球仍留在原处；视频 12.00–15.00 s 球杆垂在手下，白球仍未滚向球洞。"
+     },
+     {
+      "t": 15.0,
+      "log_step": 312,
+      "text_zh": "观察：视频 14.00–15.00 s 白球仍在原起点附近，远离洞口，未出现滚向洞口并落下的过程。log step 312 的末次现场回合输出为 OUTCOME: None、control_steps=1503；评分字段 live_success=0、graded_by_live=1 与未入洞画面一致。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "观察：视频 0.00–15.00 s 机器人右手握持推杆并带动它摆动；这里记录实际持杆接触，未确认杆头对白球形成击打，不把计划中的工具击球接触记为实际发生。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–15.00 s 机器人以双脚在果岭上支撑，挥杆及恢复姿态时有踏步和身体位移。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "ball_to_hole_m",
+   "unit": "m",
+   "threshold": 0.054,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "ball_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "head_contact_s",
+   "unit": "s",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "grip_break_max_s",
+   "unit": "s",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "ball_holed",
+   "t": 6.4,
+   "log_step": 312,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 5.60–8.20 s 机器人已持杆挥动，但杆头在白球的图像左侧摆动，未到达白球，白球没有被送出；视频 8.40–8.88 s 收杆时杆头投影靠近球后移至图像右侧，白球仍未见位移，视频 15.00 s 仍在起点附近。log step 312 是最后一次重置后的现场尝试，输出 OUTCOME: None steps 1503；reward 的 graded_by_live=1、n_actions=1503、live_success=0 表明评分取此末次回合，较早试验的推球、非法触球和跌倒结果不用于本回合现象。"
+  },
+  "unmet": {
+   "ball_holed": false
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没建立接触",
+    "t": 6.4,
+    "log_step": null,
+    "evidence_zh": "推断：视频 5.60–8.20 s 机器人右手挥动时杆头在白球的图像左侧摆动，未到达球；视频 8.40–8.88 s 收杆时虽有杆头投影靠近球并移至其图像右侧，白球仍未见位移，视频 12.00–15.00 s 也仍在原处。据此判断本次挥杆未建立所需的有效杆头击球接触；远景不足以排除没有可��位移的极短擦碰。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "log_step": 312,
+    "t": 6.4,
+    "source_zh": "日志 / 视频",
+    "evidence_zh": "推断：log step 303 的脚本以手部位置目标生成右臂逆解及恒加速度挥杆，log step 312 将手部高度目标从 0.885 m 降至 0.80 m 后执行末次回合；视频 5.60–8.20 s 机器人右手和推杆确实摆动，但杆头未到达白球，视频 15.00 s 白球仍在原处。该手部轨迹及其实际执行未使杆头到达所需击球位置，支持控制层归因；未记录实际手部及关节跟踪误差、杆头三维轨迹和接触诊断，不能区分手部路径与杆摆动设计、轨迹或逆解、反馈跟踪问题，故保留“控制”而不指定子类。是否存在可行替代：未知；不把 agent 自述的松摆机制当成已验证根因。",
+    "alt_feasible": "未知"
+   }
+  ],
+  "gap_zh": "没有状态日志；远景中球和杆头仅占数个像素，无法排除未引起可见位移的极短擦碰，也无法量出握持接触、杆头速度或接触时长。视频 8.40–8.88 s 收杆时杆头投影靠近球后移至其图像右侧，因此审查意见所称视频 5.60–12.00 s “始终在球左侧”不适用于整段；主挥杆未到达球与收杆投影经过球附近分开记录。末次回合服务输出未报告 stroke 或接触诊断，reward 也未报告 DAG 的四项原生指标，均保留 null；缺少实际关节跟踪及杆头三维轨迹，无法进一步区分手部路径与杆摆动设计、持杆柔顺性、轨迹或逆解和反馈控制，故根因仅标控制，可行替代为未知。"
+ },
+ "summary_zh": "本次评分回合中机器人右手持杆并进行了挥杆和姿态恢复，但白球始终留在起点附近，唯一节点“一杆推球入洞”未达到，最终为 0/1。断点在视频约 6.40 s 的挥杆：主要挥杆段杆头未到达球，后续收杆时杆头投影移过球附近也未使球发生可见位移，所需的有效击球接触未建立；较早重置回合的违规和跌倒不属于本次评分画面。根因保留控制层，因缺少状态、接触及跟踪诊断，无法确定具体机制或毫秒级擦碰，置信度维持 7。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 2,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "fix",
+    "conf": 7,
+    "summary_zh": "节点未达成、断点 ball_holed、现象“没建立接触”与时钟均成立（视频 15.04 s = 1503 控制步，挥杆约 5.9 s 起与 log step 303/312 一致）。但证据把杆头描述为“从白球旁摆过/经过”，画面并不支持：杆头全程停在白球的起点侧（图中左侧），始终未到达球。需改写观察文字，并据此重新审视根因归类。",
+    "n_issues": 2,
+    "n_major": 1
+   },
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 2,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "节点 ball_holed 未达成、断点、现象“没建立接触”、根因“控制”（alt_feasible 未知）及时钟均有证据支持；上一轮意见已按画面修正，未发现新的错误。",
+    "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": "2026-10-31T01:00:58Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-golf-putt-i00-standard--he4xdrj/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-golf-putt-i00-standard--he4xdrj/04_review_r1"
+   },
+   {
+    "stage": "reconcile",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-golf-putt-i00-standard--he4xdrj/05_reconcile_r1"
+   },
+   {
+    "stage": "traj_review",
+    "round": 2,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-golf-putt-i00-standard--he4xdrj/04_review_r2"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "f7ee66836cf6656de5a35385ae427c4a9f6dcb949a74ac4d143c15bbc71df201",
+   "video_sha256": "1daade181cb7928775e7a6aa5d0a02fcd3509b2885d1f6439704cb9f0ad5408f",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/golf_putt/trajectories/embodied-olympics-fun-golf-putt-i00-standard--jq97mv4.json

@@ -0,0 +1,201 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-golf-putt-i00-standard--jq97mv4",
+ "taskKey": "embodied-olympics-fun/golf_putt",
+ "dag_sha256": "74a1c00f893c3316873e4cc2017a91bda4a2926d26f9ae1a009d7dbb3a3e0854",
+ "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": 1473,
+   "sim_steps": 3263,
+   "calls": 32,
+   "rows": 2575,
+   "resets": 2,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 1469.1,
+   "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,
+   "peak_memory_gb": 0.309,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.867,
+   "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-golf-putt-i00-standard--jq97mv4/replay.mp4",
+  "duration_s": 14.76,
+  "fps": 25.0,
+  "frames": 369,
+  "sha256": "64b3466ad465cf5bf9380ddda4f93beac705d6a379ed8b0873bd2803c2bd902f"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "ball_holed": {
+    "method": "visual",
+    "intervals": [
+     [
+      14.7,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00 s 白球位于机器人脚前，球洞在远处浅色条带上，入洞条件尚未成立；机器人开局右手持杆，双脚站在果岭上。"
+     },
+     {
+      "t": 10.12,
+      "log_step": 96,
+      "text_zh": "观察：视频 9.60–10.00 s 机器人将杆头置于白球后方；视频 10.04–10.16 s 右手持杆向前挥动，杆头经过白球，视频 10.24–10.40 s 白球已离开原位并向洞口滚动。视频 10.60–12.32 s 白球继续接近球洞，机器人维持双脚支撑。"
+     },
+     {
+      "t": 13.2,
+      "log_step": null,
+      "text_zh": "观察：视频 12.32–13.12 s 白球到达洞口附近并在洞沿处短暂可见；视频 13.20–13.76 s 及视频 13.80–14.72 s 的密集帧中，洞口及洞沿已不再露出白球，机器人仍站立持杆。远景无法直接分辨洞内球心的毫米级运动。"
+     },
+     {
+      "t": 14.72,
+      "log_step": 102,
+      "text_zh": "观察：log step 102 的机器人现场服务返回 done=True、reason=success、outcome=success、stages_reached=2、control_steps=1473，传感读数 holding_right=1.0；inputs/run.json 的评分器字段 live_success=1、graded_by_live=1、finalized=1。该终局与视频 14.72 s 所见的持杆站立、洞口不露球一致。"
+     },
+     {
+      "t": 14.7,
+      "log_step": 103,
+      "text_zh": "推断：log step 80 重置后进行的第三回合才是视频中的评分回合；log step 96–97 记录挥杆前后控制步 975–1035，log step 103 保存同一成功动作并明确每控制步 dt=0.01 s，故对应视频 9.75–10.35 s，已用视频 10.04–10.40 s 的击球及白球离开原位核对。log step 102 的终局控制步 1473 对应视频约 14.73 s，与末帧视频 14.72 s 相差一个帧间隔以内；将包含洞内停稳验收的节点达成锚点取为视频约 14.7 s。前两回合的跌倒和推压判罚不属于此评分回合。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手持工具",
+      "how": "其他",
+      "note_zh": "观察：视频 10.04–10.16 s 机器人右手所持推杆的杆头击打白球，随后白球独立滚向球洞。候选方式为“击打”，现有接触方式没有对应项；仅记录杆头对白球的工具链，不重复记录右手握杆。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–14.72 s 机器人由双脚接触果岭支撑身体，击球及等待白球入洞期间保持站立。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "ball_to_hole_m",
+   "unit": "m",
+   "threshold": 0.054,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "ball_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "head_contact_s",
+   "unit": "s",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "grip_break_max_s",
+   "unit": "s",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "评分回合中，机器人右手持杆于视频约 10.1 s 击球，白球滚至洞口后于视频约 13.2 s 起不再露出洞沿；现场裁判在 log step 102 确认成功，唯一节点“一杆推球入洞”于视频约 14.7 s 完成，终态为 1/1。视频对应第三回合，重置前两次失败不计入本次进度或失败标签；成功参考视频也呈现白球在洞口消失的远景外观，仅用于比较。缺少状态日志和洞内近景，毫米级停稳及毫秒级接触无法从视频独立核验，依据现场裁判验收；四项原生指标均未报告数值，保留 null。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "唯一目标节点 ball_holed 于视频约 14.7 s 达成、终态 1/1，与评分器字段（live_success=1）及帧所见一致；控制步映射（dt=0.01 s）有击球事件核验，接触标注合理，未发现需改动的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:55:22Z",
+  "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-golf-putt-i00-standard--jq97mv4/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-golf-putt-i00-standard--jq97mv4/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "fd6bd6eec0eca6c20a9a7148584f59df97c4aad2c6eb8f74d642079eb4b802ce",
+   "video_sha256": "64b3466ad465cf5bf9380ddda4f93beac705d6a379ed8b0873bd2803c2bd902f",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/golf_putt/trajectories/embodied-olympics-fun-golf-putt-i00-standard--ngkcwly.json

@@ -0,0 +1,214 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-golf-putt-i00-standard--ngkcwly",
+ "taskKey": "embodied-olympics-fun/golf_putt",
+ "dag_sha256": "74a1c00f893c3316873e4cc2017a91bda4a2926d26f9ae1a009d7dbb3a3e0854",
+ "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": 3000,
+   "sim_steps": 19921,
+   "calls": 374,
+   "rows": 348,
+   "resets": 50,
+   "reset_budget": 50,
+   "sim_startups": 1,
+   "elapsed_s": 4630.1,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.825,
+   "peak_processes": 10,
+   "peak_disk_gb": 1.011,
+   "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-golf-putt-i00-standard--ngkcwly/replay.mp4",
+  "duration_s": 30.04,
+  "fps": 25.0,
+  "frames": 751,
+  "sha256": "380536d68a95225c4767c939e700cd793bf00fb31dac77537efb478fe9d882ea"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "ball_holed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.68,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–1.56 s 逐帧画面中，机器人双脚支撑、右手持推杆转腰挥杆并抬起杆头，白球在脚前原处保持可见，未出现被击出后滚向远处球洞的过程。"
+     },
+     {
+      "t": 30.0,
+      "log_step": null,
+      "text_zh": "观察：视频 1–29 s 的逐秒画面及视频 29.04–30.00 s 的密集末秒画面中，白球仍在机器人脚前，远处浅色条带上的洞口未接收到球；入洞条件全程未成立。"
+     },
+     {
+      "t": null,
+      "log_step": 385,
+      "text_zh": "观察：log step 385 的服务状态为 episode=51、control_steps=3000、done=True、reason=failed、resets_left=0；同一步输出的最终控制器说明记载末次伸肘试杆未击中球。agent 在 log step 384、388 也自述末次回合未记录 stroke、阶段为 0/2；此前会话中的合法击球不计入本评分视频。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "右手包握推杆握柄，持杆完成转腰挥动及随后悬持；左手下垂未辅助握杆。仅记录实际持杆接触，未把未发生的杆头击球记为实际接触。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "机器人双脚接触果岭、支撑身体完成挥杆和随后的站立，未见倒地。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "ball_to_hole_m",
+   "unit": "m",
+   "threshold": 0.054,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "ball_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "head_contact_s",
+   "unit": "s",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "grip_break_max_s",
+   "unit": "s",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "ball_holed",
+   "t": 0.68,
+   "log_step": 377,
+   "where": "操作中",
+   "evidence_zh": "观察：视频约 0.52–0.80 s，机器人右手带动推杆随转腰从白球附近挥过，随后抬杆；白球没有离开脚前原位，后续直到视频 30.00 s 仍未被送向球洞。log step 377 执行末次重置后的控制器，log step 385 的最终服务状态确认该回合已经失败。"
+  },
+  "unmet": {
+   "ball_holed": false
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没建立接触",
+    "t": 0.68,
+    "log_step": null,
+    "evidence_zh": "观察：视频 0.00–1.56 s 逐帧可见右手持杆挥动，杆头从白球附近经过后升起，而白球保持在原位；视频 29.04–30.00 s 白球仍留在脚前。推断：本次挥杆未建立所需的杆头与白球击打接触。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "log_step": 385,
+    "evidence_zh": "推断：log step 385 输出的最终脚本用预设转腰、腕部补偿和伸肘命令组合挥杆（ELBOW_PUSH=0.12）；视频约 0.52–0.80 s 实际挥杆没有使杆头接触静止白球。agent 自述伸肘改变了挥杆路径并造成末���挥空（log step 384），与视频相符，支持击球运动控制失败；不足以确认具体控制子类。",
+    "source_zh": "视频 / 日志 / agent 自述"
+   }
+  ],
+  "gap_zh": "没有状态日志，远景无法测出杆头相对球心的三维偏差及瞬时接触力，不能确定挥空主要来自预设轨迹、持杆几何变化还是跟踪误差。run.json 的评分字段未报告四项原生指标，故其 value 均为 null；末次调用的详细服务返回文本也未保留，阶段 0/2 与裁判 time 来自 agent 自述，仅作为视频判定的辅助。"
+ },
+ "summary_zh": "评分采用最后一次现场回合，机器人右手持杆并完成挥杆，但白球始终留在脚前，“一杆推球入洞”全程未达成，最终为 0/1。断点在视频约 0.68 s 的挥空，现象为没建立接触，证据支持控制失败；此前会话中击出白球的调试回合不计入本次进度。缺少状态日志及完整末次服务返回，无法进一步确认挥空的控制子类。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "评分回合为末次现场回合（视频 30.04 s，log step 385 为 failed/3000 步），视频中球始终在脚前、未入洞，ball_holed=false、断点与\"没建立接触\"·控制均有视频和服务读数支持；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": "2026-10-31T00:55:33Z",
+  "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-golf-putt-i00-standard--ngkcwly/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-golf-putt-i00-standard--ngkcwly/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "5e08f2de87cae157fac638c618adc986a3ee93beacb6edbd51b7d01f9b45dd46",
+   "video_sha256": "380536d68a95225c4767c939e700cd793bf00fb31dac77537efb478fe9d882ea",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/golf_putt/trajectories/embodied-olympics-fun-golf-putt-i00-standard--nrnbbmc.json

@@ -0,0 +1,220 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-golf-putt-i00-standard--nrnbbmc",
+ "taskKey": "embodied-olympics-fun/golf_putt",
+ "dag_sha256": "74a1c00f893c3316873e4cc2017a91bda4a2926d26f9ae1a009d7dbb3a3e0854",
+ "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": 926,
+   "sim_steps": 38582,
+   "calls": 124,
+   "rows": 38207,
+   "resets": 42,
+   "reset_budget": 50,
+   "sim_startups": 1,
+   "elapsed_s": 7151.3,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 1.64,
+   "peak_processes": 10,
+   "peak_disk_gb": 0.329,
+   "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-golf-putt-i00-standard--nrnbbmc/replay.mp4",
+  "duration_s": 9.28,
+  "fps": 25.0,
+  "frames": 232,
+  "sha256": "1b8d5a7cd6b5fab0d154b4f34b085a155034d0071061a713171d2cec10f7d7e8"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "ball_holed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 6.16,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–5.5 s，机器人右手持推杆抬起、移到白球后方并下放；视频 6.08–6.24 s，杆头挥过脚前白球处，白球离开原位向远处洞口滚动。"
+     },
+     {
+      "t": 9.24,
+      "log_step": null,
+      "text_zh": "观察：视频 6.4–8.2 s，白球滚向浅色条带上的洞口并逐渐减速；视频 8.2–9.24 s，白球停留在条带之前的果岭上，球与洞口明显分离，未落入洞中。"
+     },
+     {
+      "t": null,
+      "log_step": 236,
+      "text_zh": "观察：log step 236 的脚本先重置再提交最后一次 stand 控制序列，并从机器人服务 get_obs 的回合结果元数据读取到 FINAL ATTEMPT OUTCOME: the ball came to rest outside the hole；该结果与视频末态相符。"
+     },
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：成功参考视频 inputs/oracle/replay.mp4 的视频 4.0–5.6 s 显示白球到达浅色条带的洞口并消失，参考视频 6.64 s 洞外没有白球；本次视频 9.24 s 则仍能看到洞前白球。参考仅用于辨认入洞的可见表现。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手持工具",
+      "how": "其他",
+      "note_zh": "推断：视频 6.08–6.24 s，机器人右手持杆挥过白球，白球随即离开原位，支持推杆杆头实际击中白球；候选方式为“击打”，现有接触方式没有对应项。握杆不另重复记录；远景不能解析毫秒级接触。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.0–9.24 s，机器人以双脚在果岭上站立支撑，配合上身和右臂完成移杆与挥杆，末态仍站立。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "ball_to_hole_m",
+   "unit": "m",
+   "threshold": 0.054,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "ball_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "head_contact_s",
+   "unit": "s",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "grip_break_max_s",
+   "unit": "s",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "ball_holed",
+   "t": 6.12,
+   "log_step": 236,
+   "where": "操作中",
+   "evidence_zh": "推断：视频约 6.12 s 的这次挥杆击出白球，但视频 8.2–9.24 s 白球在到达洞口前停住，入洞步骤未完成且没有后续修复。log step 236 的最后一次回合由机器人服务明确判为白球停在洞外；锚点取击球时刻，未将日志的 agent 时间换算为视频时间。"
+  },
+  "unmet": {
+   "ball_holed": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 8.2,
+    "log_step": 236,
+    "evidence_zh": "观察：视频 6.16–8.2 s，白球被击出后朝洞口方向滚动，却在浅色洞口条带之前减速停下；视频 8.2–9.24 s 白球一直露在果岭上，与洞口有明显间隔。log step 236 的服务结果亦为白球停在洞外，表现为击���落点偏短。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "t": 6.12,
+    "log_step": 209,
+    "source_zh": "日志",
+    "evidence_zh": "推断：log step 194 恢复挥杆行程越过触球位置 0.12 米及手部高度补偿 0.035 米，log step 209 将脚本的加速度和加速时长设为 9.0 m/s²、0.20 秒；log step 236 用该脚本生成动作并提交最终回合。视频 6.16–9.24 s 显示它建立了击球接触，却未使白球滚到洞口，支持挥杆控制未能产生所需击球效果；这些脚本参数不是实测杆头速度，证据不足以细分控制子类。"
+   }
+  ],
+  "gap_zh": "没有状态日志，远景无法精确测出杆头和白球速度、触球部位、接触时长、右手接触中断及关节跟踪误差；无法确定偏短具体来自速度设定、触球几何还是跟踪误差。agent 在 log step 237 自述的速度饱和及行程限制未获独立测量验证，不据此细分根因。评分器未报告四项原生指标的数值，均记 null。"
+ },
+ "summary_zh": "评分采用 log step 236 的最后一次现场回合：机器人持杆站立、挥杆击出白球，但视频 8.2–9.24 s 白球停在洞前果岭上，一杆入洞节点始终未达成，最终为 0/1。断点在一杆推球入洞的操作中，现象是击球落点偏短，根因保守归为控制。缺少状态及接触、速度和跟踪测量，无法进一步确定具体控制问题。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "视频 6.1 s 挥杆击球、白球在洞前停下（视频 8.2–9.24 s），ball_holed 未达成，final=false、断点、位姿不准现象与保守的控制根因均与帧和 log step 236 的服务结果相符，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": "2026-10-31T00:47:31Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-golf-putt-i00-standard--nrnbbmc/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-golf-putt-i00-standard--nrnbbmc/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "cb567c97dd52033cfda53cd23064e15fa411a61a90196b9f03832a0eea79a9f5",
+   "video_sha256": "1b8d5a7cd6b5fab0d154b4f34b085a155034d0071061a713171d2cec10f7d7e8",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/golf_putt/trajectories/embodied-olympics-fun-golf-putt-i00-standard--qyhcyw7.json

@@ -0,0 +1,217 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-golf-putt-i00-standard--qyhcyw7",
+ "taskKey": "embodied-olympics-fun/golf_putt",
+ "dag_sha256": "74a1c00f893c3316873e4cc2017a91bda4a2926d26f9ae1a009d7dbb3a3e0854",
+ "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": 690,
+   "sim_steps": 5482,
+   "calls": 15,
+   "rows": 4282,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 864.3,
+   "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": 0.0,
+   "peak_memory_gb": 0.249,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.345,
+   "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-golf-putt-i00-standard--qyhcyw7/replay.mp4",
+  "duration_s": 6.92,
+  "fps": 25.0,
+  "frames": 173,
+  "sha256": "f94ee4e324aa849946b8f9042adbbe5c9d1d2aa0536c6c781523a6af7253c131"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "ball_holed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.76,
+      "log_step": 171,
+      "text_zh": "观察：视频 2.40–2.68 s 机器人右手持杆，白球仍在脚前；视频约 2.72–2.80 s 杆头经过白球位置后，白球离开原位向远处球洞滚动，机器人双脚持续支撑身体。"
+     },
+     {
+      "t": 5.8,
+      "log_step": 171,
+      "text_zh": "观察：视频 5.00–5.96 s 白球逐渐接近洞口近侧并停下，视频 5.80–6.88 s 的逐帧洞口裁剪仍能看到洞口近侧的亮白球体，没有完整落洞消失。log step 171 是最后现场回合 episode=6 的执行结果，服务返回 stages_reached=1、stages_total=2、outcome=\"the ball came to rest outside the hole\"；该回合 control_steps=690、sim_time_s=6.9，与本视频 6.92 s 时长相符。"
+     },
+     {
+      "t": 6.88,
+      "log_step": 194,
+      "text_zh": "观察：成功参考视频约 5.4–6.6 s 的洞口不再露出白球，本回合视频 6.88 s 洞前仍可见白球。log step 193 后来把默认挥杆时长改为 0.19 s，但 log step 194 运行时服务显示 episode=6 已结束且 resets_left=0，脚本直接退出，没有新的击球回合；该修改不属于本段评分视频。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手持工具",
+      "how": "其他",
+      "note_zh": "观察：视频约 2.72–2.80 s，机器人右手持推杆，杆头经过白球后白球向洞口滚出；工具链为推杆杆头作用于白球。候选接触名为“击打”，现有方式中没有对应项，故用其他，不重复记录握杆接触。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–6.88 s，机器人双脚接触果岭，站立支撑身体完成后摆、击球与随挥。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "ball_to_hole_m",
+   "unit": "m",
+   "threshold": 0.054,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "ball_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "head_contact_s",
+   "unit": "s",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "grip_break_max_s",
+   "unit": "s",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "ball_holed",
+   "t": 5.8,
+   "log_step": 171,
+   "where": "操作中",
+   "evidence_zh": "观察：机器人在视频约 2.76 s 击出白球，视频约 5.8 s 白球停在洞口近侧，直至视频 6.88 s 仍未落入洞内；log step 171 的现场服务结果明确为球在洞外停稳，仅完成击球阶段。因此失败步骤是一杆推球入洞，锚点取白球在洞前停下的位置。"
+  },
+  "unmet": {
+   "ball_holed": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 5.8,
+    "log_step": 171,
+    "evidence_zh": "观察：视频 2.76–5.8 s 白球沿果岭向洞口滚动后停在洞口近侧，视频 5.80–6.88 s 仍露出白球，没有进入洞内；log step 171 服务判定“the ball came to rest outside the hole”。推断：此次推球距离不��，球的终点未达到入洞位置。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "t": 2.76,
+    "log_step": 167,
+    "source_zh": "日志 / 视频",
+    "evidence_zh": "观察：log step 155、157 中 agent 的计算预测挥杆时长 0.20 s 可以滚过约三米，log step 167 的脚本据此生成固定关节轨迹，log step 171 确实以 T_swing=0.2s 执行；视频 2.76–5.8 s 显示球实际停在洞前，服务确认洞外停球。推断：实际击球控制未给白球足够的前进效果，预先估计的挥杆与滚动关系未达到目标；缺少杆头速度、接触冲量和关节跟踪记录，无法进一步区分力度参数标定与运动跟踪误差，根因保留在控制大项。"
+   }
+  ],
+  "gap_zh": "没有状态日志，现有评分视频仅含场地远景，无法独立核对毫米级洞内高度、停稳位移和毫秒级握杆/杆头接触；洞外停球有视频及最后现场回合的服务判定共同支持。run.json 的 reward 未报告 DAG 所列四项原生指标，均填 null；agent 在 log step 172 的像素反投影是其自算估计，未用作评分器指标。缺少实际击球速度、接触冲量及跟踪误差，控制子原因不能细分；修改为 0.19 s 后没有执行，无法验证修正效果。"
+ },
+ "summary_zh": "机器人右手持杆完成击球，白球滚向洞口后停在洞前，一杆推球入洞节点始终未达成，最终为 0/1。视频约 5.8 s 至结尾仍可见洞前白球，log step 171 的最后现场回合明确判定洞外停球；推断实际挥杆控制的前进效果不足。之后将挥杆时长改为 0.19 s 的脚本没有执行，且缺少状态和接触数据，不能验证修正效果或进一步细分控制原因。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "标签与证据一致：视频 5.0 s 起白球停在洞口近侧，log step 171 服务判定 the ball came to rest outside the hole（stages 1/2），ball_holed 未达成、断点与位姿不准/控制归因有证据支持且推断已标注。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:55:03Z",
+  "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-golf-putt-i00-standard--qyhcyw7/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-golf-putt-i00-standard--qyhcyw7/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "d67a00d625526e280172835926a17d0ec8640dce836e746698b98fb5fd31e4c8",
+   "video_sha256": "f94ee4e324aa849946b8f9042adbbe5c9d1d2aa0536c6c781523a6af7253c131",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/golf_putt/trajectories/embodied-olympics-fun-golf-putt-i00-standard--vcydxbu.json

@@ -0,0 +1,196 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-golf-putt-i00-standard--vcydxbu",
+ "taskKey": "embodied-olympics-fun/golf_putt",
+ "dag_sha256": "74a1c00f893c3316873e4cc2017a91bda4a2926d26f9ae1a009d7dbb3a3e0854",
+ "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": 1312,
+   "sim_steps": 8169,
+   "calls": 40,
+   "rows": 2695,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 2040.9,
+   "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": 1.076,
+   "peak_processes": 11,
+   "peak_disk_gb": 0.631,
+   "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-golf-putt-i00-standard--vcydxbu/replay.mp4",
+  "duration_s": 13.16,
+  "fps": 25.0,
+  "frames": 329,
+  "sha256": "f99f890a0cb35f5f532f7961f3086bd8a9f1905eb662844fa86bf199f0acb31c"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "ball_holed": {
+    "method": "visual",
+    "intervals": [
+     [
+      13.12,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 9.4,
+      "log_step": 82,
+      "text_zh": "观察：视频 9.28–9.64 s 中，机器人双脚站在果岭上，右手持推杆向前挥动；杆头经过脚前白球位置后，白球离开原位并向远处浅色条带上的洞口滚动。"
+     },
+     {
+      "t": 11.2,
+      "log_step": null,
+      "text_zh": "观察：视频 10.80–11.52 s 的洞口区域逐帧检查可见白球接近并进入洞口区域，随后不再能辨出洞外白球；视频 11.60–13.12 s 未见白球重新露出，机器人仍站立持杆。"
+     },
+     {
+      "t": 13.12,
+      "log_step": 82,
+      "text_zh": "观察：log step 82 在 reset 后明确进入 episode 6，最终打印的机器人服务返回值为 done=True、reason=success、outcome=success、stages_reached=2，回合仿真时间约 13.12 s；log step 83 的服务 status 再次报告 episode=6、control_steps=1312、done=True、reason=success。评分字段 live_success=1、graded_by_live=1，说明评分采用这次最终现场回合，前五次重置回合不属于本视频。"
+     },
+     {
+      "t": 13.12,
+      "log_step": 83,
+      "text_zh": "推断：log step 3 的服务规格给出控制周期 0.01 s，log step 83 的最终回合控制步 1312 对应视频约 13.12 s；视频总长 13.16 s，且脚本在回合仿真 9.0 s 开始挥杆、约 9.33 s 经过摆动底点，与视频 9.32–9.44 s 的击球相符，故采用视频 13.12 s 作为裁判完成入洞停稳验收的节点起点。远景只支持球进洞后不再露出的判断，全球低于果岭及洞内停稳 1.0 秒由本次服务成功结果确认。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手持工具",
+      "how": "其他",
+      "note_zh": "观察：视频 9.32–9.64 s，机器人右手持推杆，杆头经过白球后白球立即向洞口滚动；候选接触名为“击打”，现有接触方式没有对应项。此处记录杆头对白球的工具链，不重复记录握杆。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.0–13.12 s，机器人双脚在果岭上支撑身体，击球及等待白球滚入洞时均保持站立。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "ball_to_hole_m",
+   "unit": "m",
+   "threshold": 0.054,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "ball_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "head_contact_s",
+   "unit": "s",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "grip_break_max_s",
+   "unit": "s",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "本次评分视频对应最终第六回合：机器人右手持推杆，在视频约 9.4 s 击球，白球滚至洞口后不再露出；服务在视频约 13.12 s 对应的回合终点确认成功，唯一节点一杆推球入洞达成，最终为 1/1。前五次重置回合的失败不属于本次评分轨迹。没有状态日志，远景不能独立量测洞���高度、毫米级停稳或毫秒级接触，精细验收依据本次裁判成功结果；评分字段未报告四项原生指标数值，均保留为 null。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "单节点 ball_holed 达成、final=true 与评分成功一致；击球时间约 9.4 s、球入洞后不再露出（约 11 s 起）与帧相符，13.12 s 的节点起点由服务控制步 1312 的推断得出且标注为推断，contact_actual 合理。未发现需改动处。",
+    "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": "2026-10-31T00:48:50Z",
+  "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-golf-putt-i00-standard--vcydxbu/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-golf-putt-i00-standard--vcydxbu/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "f3bd5e263b6f5863b9e1eb6c098d5df7f54ca74f1a308536bc543ab2013d87da",
+   "video_sha256": "f99f890a0cb35f5f532f7961f3086bd8a9f1905eb662844fa86bf199f0acb31c",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/juggling_feet/trajectories/embodied-olympics-fun-juggling-feet-i00-privileged--8gsjzxu.json

@@ -0,0 +1,313 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-juggling-feet-i00-privileged--8gsjzxu",
+ "taskKey": "embodied-olympics-fun/juggling_feet",
+ "dag_sha256": "6fb5f868dd0ec1e7ac4f8d9035cd8456789adc45e6c5ceda5fec1ad9d3ba10f6",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "n_actions": 514,
+   "sim_steps": 514,
+   "horizon": 6000,
+   "episode_over": 1,
+   "progress": 0.0,
+   "stages_reached": 0,
+   "stages_total": 10,
+   "fell": 0,
+   "replay_s": 4.43,
+   "bad_trajectory": 0,
+   "metric_ball_height_m": 0.099,
+   "metric_bounces": 0.0,
+   "metric_touches": 0.0,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 476,
+   "agent_resets": 577,
+   "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": 9.16,
+   "peak_processes": 24,
+   "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-juggling-feet-i00-privileged--8gsjzxu/replay.mp4",
+  "duration_s": 5.16,
+  "fps": 25.0,
+  "frames": 129,
+  "sha256": "300dfccd482e55bd29b05ac0279cbe07864c04c2754f249c8b3184f3ecbd3f5f"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 10,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "juggles_counted_1": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 4.84,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–4.80 s 机器人将足球保持在双掌之间，约视频 4.20 s 起抬右脚，视频 4.84–4.96 s 双掌分开、足球下落；视频 5.00–5.12 s 球从抬起的右脚前方落向草地，未出现脚击球后离脚上升，机器人仍站立。"
+     },
+     {
+      "t": 5.12,
+      "log_step": 333,
+      "text_zh": "观察：评分器字段 stages_reached=0、metric_touches=0、metric_bounces=0、fell=0；log step 239、277、333 对交付的同一514步文件回放均返回 touches=0、stages=[]、足球触地结束，末态哈希一致，与视频 5.12 s 的落球相符。第1次有效颠球未完成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "视频 0.00–4.80 s 左右平掌共同夹持足球，随后分开让球落下；候选名称“双掌挤压”。两只平掌均未包围足球，现有方式无合适的双掌挤压项，不能记作抓握。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "准备首次颠球时双脚承重接触草地；视频约 4.20–5.12 s 右脚抬起，左脚继续在草地支撑身体。右脚没有击中下落足球，因此不记录脚对球的击打接触。"
+     }
+    ]
+   },
+   "juggles_counted_2": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 5.12,
+      "log_step": 333,
+      "text_zh": "推断：视频 0.00–5.12 s 只有双掌持球、抬脚和一次放球下落，至视频 5.12 s 未出现任何脚击后上升；评分器 stages_reached=0、metric_touches=0，log step 333 回放交付文件也为 touches=0、stages=[]。第1次尚未完成，第2次有效颠球未开始且未达到。"
+     }
+    ]
+   },
+   "juggles_counted_3": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 5.12,
+      "log_step": 333,
+      "text_zh": "推断：视频 0.00–5.12 s 只有双掌持球、抬脚和一次放球下落，至视频 5.12 s 未出现任何脚击后上升；评分器 stages_reached=0、metric_touches=0，log step 333 回放交付文件也为 touches=0、stages=[]。第1次尚未完成，第3次有效颠球未开始且未达到。"
+     }
+    ]
+   },
+   "juggles_counted_4": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 5.12,
+      "log_step": 333,
+      "text_zh": "推断：视频 0.00–5.12 s 只有双掌持球、抬脚和一次放球下落，至视频 5.12 s 未出现任何脚击后上升；评分器 stages_reached=0、metric_touches=0，log step 333 回放交付文件也为 touches=0、stages=[]。第1次尚未完成，第4次有效颠球未开始且未达到。"
+     }
+    ]
+   },
+   "juggles_counted_5": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 5.12,
+      "log_step": 333,
+      "text_zh": "推断：视频 0.00–5.12 s 只有双掌持球、抬脚和一次放球下落，至视频 5.12 s 未出现任何脚击后上升；评分器 stages_reached=0、metric_touches=0，log step 333 回放交付文件也为 touches=0、stages=[]。第1次尚未完成，第5次有效颠球未开始且未达到。"
+     }
+    ]
+   },
+   "juggles_counted_6": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 5.12,
+      "log_step": 333,
+      "text_zh": "推断：视频 0.00–5.12 s 只有双掌持球、抬脚和一次放球下落，至视频 5.12 s 未出现任何脚击后上升；评分器 stages_reached=0、metric_touches=0，log step 333 回放交付文件也为 touches=0、stages=[]。第1次尚未完成，第6次有效颠球未开始且未达到。"
+     }
+    ]
+   },
+   "juggles_counted_7": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 5.12,
+      "log_step": 333,
+      "text_zh": "推断：视频 0.00–5.12 s 只有双掌持球、抬脚和一次放球下落，至视频 5.12 s 未出现任何脚击后上升；评分器 stages_reached=0、metric_touches=0，log step 333 回放交付文件也为 touches=0、stages=[]。第1次尚未完成，第7次有效颠球未开始且未达到。"
+     }
+    ]
+   },
+   "juggles_counted_8": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 5.12,
+      "log_step": 333,
+      "text_zh": "推断：视频 0.00–5.12 s 只有双掌持球、抬脚和一次放球下落，至视频 5.12 s 未出现任何脚击后上升；评分器 stages_reached=0、metric_touches=0，log step 333 回放交付文件也为 touches=0、stages=[]。第1次尚未完成，第8次有效颠球未开始且未达到。"
+     }
+    ]
+   },
+   "juggles_counted_9": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 5.12,
+      "log_step": 333,
+      "text_zh": "推断：视频 0.00–5.12 s 只有双掌持球、抬脚和一次放球下落，至视频 5.12 s 未出现任何脚击后上升；评分器 stages_reached=0、metric_touches=0，log step 333 回放交付文件也为 touches=0、stages=[]。第1次尚未完成，第9次有效颠球未开始且未达到。"
+     }
+    ]
+   },
+   "juggles_counted_10": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 5.12,
+      "log_step": 333,
+      "text_zh": "推断：视频 0.00–5.12 s 只有双掌持球、抬脚和一次放球下落，至视频 5.12 s 未出现任何脚击后上升；评分器 stages_reached=0、metric_touches=0，log step 333 回放交付文件也为 touches=0、stages=[]。第1次尚未完成，第10次有效颠球未开始且未达到。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "touches",
+   "value": 0.0,
+   "unit": "count",
+   "threshold": 10,
+   "direction": "higher_better"
+  },
+  {
+   "name": "ball_height_m",
+   "value": 0.099,
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better"
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "juggles_counted_1",
+   "t": 5.0,
+   "log_step": 333,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 4.84–5.12 s 机器人已放开双掌并抬起右脚，但足球在右脚前方持续下落，视频 5.00 s 附近已错过脚部截击，未再升起；视频 5.12 s 球到达草地，机器人未倒下。log step 333 的交付文件回放返回足球触地结束、touches=0，故失败发生于第1次颠球操作。"
+  },
+  "unmet": {
+   "juggles_counted_1": false,
+   "juggles_counted_2": false,
+   "juggles_counted_3": false,
+   "juggles_counted_4": false,
+   "juggles_counted_5": false,
+   "juggles_counted_6": false,
+   "juggles_counted_7": false,
+   "juggles_counted_8": false,
+   "juggles_counted_9": false,
+   "juggles_counted_10": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "动态时机",
+    "t": 5.0,
+    "log_step": null,
+    "evidence_zh": "观察：视频 4.84–5.12 s 足球脱离双掌后向下运动，机器人右脚保持抬起但没有跟到球的下落位置；视频 5.00–5.12 s 球从右脚前方落到草地，整段没有向上的脚击反弹。机器人未能截击这个运动中的目标。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "轨迹或逆解",
+    "source_zh": "日志",
+    "evidence_zh": "推断：agent 的控制脚本未生成可执行并能把右脚送到落球处的关节轨迹。log step 245 对默认控制器的诊断复现514步触地失败，逆解欲将脚部送至[0.39,0.06,0.32]米，但实际右脚锚点约为[0.015,0.056,0.265]米；log step 248 的会话诊断在仿真 4.60–4.90 s 给出右踝滚转目标约−0.645至−0.692弧度，实际仅约−0.263至−0.269弧度，log step 250 读取模型右踝滚转限位为[−0.262,0.262]弧度，目标超出范围。log step 238 生成交付文件时在仿真 5.01 s 输出球心x=0.349米、右脚标记x=0.014米；视频 4.84–5.12 s 同样显示脚留在落球���方。上述诊断支持逆解及轨迹可执行性问题，不采用 agent 关于所有合法方案均不可行的自述。"
+   }
+  ],
+  "gap_zh": "没有状态日志；25帧每秒的固定视角不能独立解析毫秒级接触、精确速度和关节受力。节点未达由完整落球过程及交付回放的原生计数共同确定；根因的限位和关节跟踪依据是 log step 245、248、250 的会话诊断，不能据此确定交付回放每一帧的电机饱和或量化各控制环节的贡献。"
+ },
+ "summary_zh": "机器人完成了首次颠球的抬脚和放球准备，但足球在视频 4.84–5.12 s 从右脚前方直接落地，十个有效颠球节点均未达到，评分器计数为0且未跌倒。断点为第1次颠球，现象是未能截击下落足球；日志诊断支持逆解目标超出右踝限位、脚位未到落球处的控制问题。没有状态日志，无法独立核验毫秒接触及精确受力，会话中其他测试的跌倒和碰腿不计入本次交付。",
+ "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 0–5.12 s 球一直在双掌间，随后落地，无脚击上升；评分器 stages_reached=0、touches=0，log step 333 回放证实。十个节点均未达、断点 juggles_counted_1、现象与根因（逆解/轨迹，右踝滚转目标超限位）均有 log step 245/248/250 的数值支持，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": "2026-10-31T00:48:33Z",
+  "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-juggling-feet-i00-privileged--8gsjzxu/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-juggling-feet-i00-privileged--8gsjzxu/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "f46d6bee95747aead1012268df279698384da2271af232f6afc44507c551524f",
+   "video_sha256": "300dfccd482e55bd29b05ac0279cbe07864c04c2754f249c8b3184f3ecbd3f5f",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/juggling_feet/trajectories/embodied-olympics-fun-juggling-feet-i00-privileged--kihvqab.json

@@ -0,0 +1,324 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-juggling-feet-i00-privileged--kihvqab",
+ "taskKey": "embodied-olympics-fun/juggling_feet",
+ "dag_sha256": "6fb5f868dd0ec1e7ac4f8d9035cd8456789adc45e6c5ceda5fec1ad9d3ba10f6",
+ "outcome": {
+  "success": 0,
+  "reward": 0.1,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "n_actions": 85,
+   "sim_steps": 85,
+   "horizon": 6000,
+   "episode_over": 0,
+   "progress": 0.1,
+   "stages_reached": 1,
+   "stages_total": 10,
+   "fell": 0,
+   "replay_s": 6.01,
+   "bad_trajectory": 0,
+   "metric_ball_height_m": 0.851,
+   "metric_bounces": 0.0,
+   "metric_touches": 1.0,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 622,
+   "agent_resets": 16,
+   "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": 1.744,
+   "peak_processes": 10,
+   "peak_disk_gb": 0.277,
+   "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-juggling-feet-i00-privileged--kihvqab/replay.mp4",
+  "duration_s": 0.88,
+  "fps": 25.0,
+  "frames": 22,
+  "sha256": "46ee9802553ad40faa13e345ecc160403f8a0c677f46ca2d67fc9e673110e5c0"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 10,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "juggles_counted_1": {
+    "method": "visual",
+    "intervals": [
+     [
+      0.8,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.72,
+      "log_step": null,
+      "text_zh": "观察：已检查视频 0.00–0.84 s 全部22帧；机器人先以双掌夹持足球并张臂放球，球在视频 0.64–0.72 s 下落至右脚附近，在视频 0.76–0.84 s 离脚上升，未见球触地或机器人倒地。"
+     },
+     {
+      "t": 0.8,
+      "log_step": 439,
+      "text_zh": "推断：视频 0.80 s 球已明显离脚升高，取此帧为首次有效颠球的近似达成点（约有一帧误差）；log step 439 对所交付85行轨迹的回放返回 stages=[touch 1]、touches=1、ball_height_m=0.851，与最终评分器 stages_reached=1、metric_touches=1 一致。有效计数由评分器补证，未从25 fps画面独立测算毫秒接触或速度门槛。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 0.00–0.40 s 初始双掌从两侧夹住足球，随后张开释放；候选方式“对向挤压夹持”，双掌未各自包围足球，不属于抓握。此为首次颠球的放球准备。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频 0.72–0.80 s 右脚在足球下方摆动，足球随即离脚上升；候选方式“击打”，现有接触方式无击打项。log step 358 同参数会话测试的裁判事件也标记 right_foot，仅作为作用脚身份的辅助。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–0.72 s 双脚先接触草地，右脚摆动击球时左脚承担地面支撑。"
+     }
+    ]
+   },
+   "juggles_counted_2": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.84,
+      "log_step": 439,
+      "text_zh": "观察：视频 0.00–0.84 s 全部帧只显示一次下落和踢起；末帧足球仍处于首次击球后的上升段，没有第2次离脚上升。log step 439 对所交付轨迹的回放仅有 touch 1，最终评分器 stages_reached=1、metric_touches=1，故第2次有效颠球未达成。"
+     }
+    ]
+   },
+   "juggles_counted_3": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.84,
+      "log_step": 439,
+      "text_zh": "观察：视频 0.00–0.84 s 全部帧只显示一次下落和踢起；末帧足球仍处于首次击球后的上升段，没有第3次离脚上升。log step 439 对所交付轨迹的回放仅有 touch 1，最终评分器 stages_reached=1、metric_touches=1，故第3次有效颠球未达成。"
+     }
+    ]
+   },
+   "juggles_counted_4": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.84,
+      "log_step": 439,
+      "text_zh": "观察：视频 0.00–0.84 s 全部帧只显示一次下落和踢起；末帧足球仍处于首次击球后的上升段，没有第4次离脚上升。log step 439 对所交付轨迹的回放仅有 touch 1，最终评分器 stages_reached=1、metric_touches=1，故第4次有效颠球未达成。"
+     }
+    ]
+   },
+   "juggles_counted_5": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.84,
+      "log_step": 439,
+      "text_zh": "观察：视频 0.00–0.84 s 全部帧只显示一次下落和踢起；末帧足球仍处于首次击球后的上升段，没有第5次离脚上升。log step 439 对所交付轨迹的回放仅有 touch 1，最终评分器 stages_reached=1、metric_touches=1，故第5次有效颠球未达成。"
+     }
+    ]
+   },
+   "juggles_counted_6": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.84,
+      "log_step": 439,
+      "text_zh": "观察：视频 0.00–0.84 s 全部帧只显示一次下落和踢起；末帧足球仍处于首次击球后的上升段，没有第6次离脚上升。log step 439 对所交付轨迹的回放仅有 touch 1，最终评分器 stages_reached=1、metric_touches=1，故第6次有效颠球未达成。"
+     }
+    ]
+   },
+   "juggles_counted_7": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.84,
+      "log_step": 439,
+      "text_zh": "观察：视频 0.00–0.84 s 全部帧只显示一次下落和踢起；末帧足球仍处于首次击球后的上升段，没有第7次离脚上升。log step 439 对所交付轨迹的回放仅有 touch 1，最终评分器 stages_reached=1、metric_touches=1，故第7次有效颠球未达成。"
+     }
+    ]
+   },
+   "juggles_counted_8": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.84,
+      "log_step": 439,
+      "text_zh": "观察：视频 0.00–0.84 s 全部帧只显示一次下落和踢起；末帧足球仍处于首次击球后的上升段，没有第8次离脚上升。log step 439 对所交付轨迹的回放仅有 touch 1，最终评分器 stages_reached=1、metric_touches=1，故第8次有效颠球未达成。"
+     }
+    ]
+   },
+   "juggles_counted_9": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.84,
+      "log_step": 439,
+      "text_zh": "观察：视频 0.00–0.84 s 全部帧只显示一次下落和踢起；末帧足球仍处于首次击球后的上升段，没有第9次离脚上升。log step 439 对所交付轨迹的回放仅有 touch 1，最终评分器 stages_reached=1、metric_touches=1，故第9次有效颠球未达成。"
+     }
+    ]
+   },
+   "juggles_counted_10": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.84,
+      "log_step": 439,
+      "text_zh": "观察：视频 0.00–0.84 s 全部帧只显示一次下落和踢起；末帧足球仍处于首次击球后的上升段，没有第10次离脚上升。log step 439 对所交付轨迹的回放仅有 touch 1，最终评分器 stages_reached=1、metric_touches=1，故第10次有效颠球未达成。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "touches",
+   "unit": "count",
+   "threshold": 10,
+   "direction": "higher_better",
+   "value": 1.0
+  },
+  {
+   "name": "ball_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": 0.851
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "juggles_counted_2",
+   "t": null,
+   "log_step": 437,
+   "where": "首步前",
+   "evidence_zh": "观察：视频 0.76–0.84 s 仅出现首次击球后的上升，未出现第二次迎球击打。log step 437 的生成脚本只执行一次放球及右脚击球，在控制步85前结束并保存；log step 674 再运行该生成脚本，log step 680 最终报告仍为85个控制步、Touch 1。最终评分器 episode_over=0、fell=0、stages_reached=1，表明交付在第二次颠球启动前结束，不能把会话测试中的后续跌倒当作本次判负事件。"
+  },
+  "unmet": {
+   "juggles_counted_1": true,
+   "juggles_counted_2": false,
+   "juggles_counted_3": false,
+   "juggles_counted_4": false,
+   "juggles_counted_5": false,
+   "juggles_counted_6": false,
+   "juggles_counted_7": false,
+   "juggles_counted_8": false,
+   "juggles_counted_9": false,
+   "juggles_counted_10": false
+  },
+  "phenomena": [
+   {
+    "cat": "没尝试",
+    "item": "没尝试",
+    "t": 0.84,
+    "log_step": 437,
+    "evidence_zh": "观察：视频 0.84 s 末帧中足球仍从首次击球上升；全段未见第二次击球准备或新的触球周期。log step 437 脚本的最后一段只在首次击球姿态与上挑姿态之间切换，到控制步85即保存，没有第二次颠球段；log step 439 的交付回放仅计入 touch 1。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "log_step": 414,
+    "source_zh": "日志",
+    "evidence_zh": "推断：机器人首次击球后的恢复与持续支撑控制未解决，限制了连续颠球。依据是会话测试 log step 358 首次计入 touch 1 后骨盆持续下降并由服务返回 fell；log step 414 的平滑回收测试仍在控制步100附近跌倒；log step 488 改用张臂双足恢复仍跌倒；更晚的 log step 648 快速回双足支撑测试也只达 touch 1 后跌倒。最终 log step 437/674 交付仍停在首次击球后的85行。以上均为会话证据，不能据此声称视频内发生了跌倒；不足以确定具体控制子类。"
+   }
+  ],
+  "gap_zh": "没有仿真状态日志，25 fps视频不能独立确认接触簇时长、能量速度门槛及精确计数时刻，首次有效计数由最终评分器补证。交付止于首次击球后，未展示足球回落和第二次迎球；控制根因由多次会话续接测试推断，日志未明确说明最终为何仅保留85行，无法进一步确定具体控制子类。"
+ },
+ "summary_zh": "机器人放球后用右脚完成第1次有效颠球，视频约0.80 s 达成，最终为1/10；第2次颠球尚未启动，交付便结束，评分器未记录跌倒或回合终止。会话中多次首次击球后的恢复测试失稳，支持持续控制能力不足的推断，但这些跌倒不属于交付录像；无状态日志，精确触球验收依赖评分器计数。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "第1次颠球约视频0.8 s达成、其余未达成、终分1/10均与视频帧和评分器字段一致；断点在 juggles_counted_2（首步前/没尝试，交付仅85行且止于首次击球）与交付内容相符，会话中的跌倒仅作为控制根因的推断且已声明不属于交付录像，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": "2026-10-31T00:52:03Z",
+  "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-juggling-feet-i00-privileged--kihvqab/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-juggling-feet-i00-privileged--kihvqab/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "4e6722b2782bb12fa0c4ed0bf2c17f81dfe67c028c7cf9ee9d272fe5f715c6ea",
+   "video_sha256": "46ee9802553ad40faa13e345ecc160403f8a0c677f46ca2d67fc9e673110e5c0",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/juggling_feet/trajectories/embodied-olympics-fun-juggling-feet-i00-privileged--o6ebpor.json

@@ -0,0 +1,361 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-juggling-feet-i00-privileged--o6ebpor",
+ "taskKey": "embodied-olympics-fun/juggling_feet",
+ "dag_sha256": "6fb5f868dd0ec1e7ac4f8d9035cd8456789adc45e6c5ceda5fec1ad9d3ba10f6",
+ "outcome": {
+  "success": 0,
+  "reward": 0.3,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "n_actions": 337,
+   "sim_steps": 337,
+   "horizon": 6000,
+   "episode_over": 1,
+   "progress": 0.3,
+   "stages_reached": 3,
+   "stages_total": 10,
+   "fell": 0,
+   "replay_s": 3.79,
+   "bad_trajectory": 0,
+   "metric_ball_height_m": 0.088,
+   "metric_bounces": 0.0,
+   "metric_touches": 3.0,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 1372,
+   "agent_resets": 0,
+   "agent_replays": 11,
+   "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": 11.217,
+   "peak_processes": 10,
+   "peak_disk_gb": 0.844,
+   "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-juggling-feet-i00-privileged--o6ebpor/replay.mp4",
+  "duration_s": 3.4,
+  "fps": 25.0,
+  "frames": 85,
+  "sha256": "b0fbdace00fc6d2a208d72a7953cfe65d1507ac12da53c1f8972e86118af1e6a"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 10,
+  "base": 0,
+  "peak": 3,
+  "final": 3,
+  "nodes": {
+   "juggles_counted_1": {
+    "method": "visual",
+    "intervals": [
+     [
+      0.64,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.64,
+      "log_step": 499,
+      "text_zh": "观察：视频 0.24–0.56 s，机器人张开双掌放球，足球下落至左脚；视频 0.56–0.72 s，球从左脚附近转为上升并离脚，视频 0.84 s 已升至腰部附近。log step 499 对交付的337行轨迹回放，裁判列出 touch 1（left_foot）；评分器最终 stages_reached=3。首次达到时间按视频取约0.64 s，毫秒接触和速度合规由该回放裁判计数补充。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 0.00–0.24 s，机器人双掌相对挤压足球，随后分开双掌放落足球；候选名“双掌挤压”，平掌没有合围抓握，现有方式无准确对应项。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：机器人左脚短促击打足球后球离脚上升；候选接触名为“击打”，现有接触方式无对应项，故用其他。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：左脚击球时，机器人右脚接触草地承重支撑身体。"
+     }
+    ]
+   },
+   "juggles_counted_2": {
+    "method": "visual",
+    "intervals": [
+     [
+      1.24,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 1.24,
+      "log_step": 499,
+      "text_zh": "观察：视频 1.04–1.20 s，足球再次下落，机器人抬起右脚承接；视频 1.24–1.52 s，球离开右脚持续向上，升至头部附近。log step 499 对交付文件的回放裁判列出 touch 2（right_foot）；按画面取视频约1.24 s达到，已计入的第1次保留。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：机器人右脚短促击打足球后球离脚上升；候选接触名为“击打”，现有接触方式无对应项，故用其他。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：右脚击球时，机器人左脚接触草地承重支撑身体。"
+     }
+    ]
+   },
+   "juggles_counted_3": {
+    "method": "visual",
+    "intervals": [
+     [
+      2.2,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 2.2,
+      "log_step": 499,
+      "text_zh": "观察：视频 1.76–2.12 s，足球从高处下落到右脚前，机器人屈伸右膝击球；视频 2.16–2.40 s，球从右脚附近快速离脚上升并超过腰部。log step 499 对交付文件的回放裁判列出 touch 3（right_foot），最终 touches=3；按画面取视频约2.20 s达到。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：机器人右脚短促击打足球后球离脚上升；候选接触名为“击打”，现有接触方式无对应项，故用其他。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：右脚击球时，机器人左脚接触草地承重支撑身体。"
+     }
+    ]
+   },
+   "juggles_counted_4": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 3.2,
+      "log_step": 499,
+      "text_zh": "观察：视频 2.40–3.36 s，机器人放低击球腿并调整站姿，足球在其侧方下落；视频 3.20–3.36 s，双脚均未移到下落足球下方，球没有第四次离脚上升，最后接近并落到草地。log step 499 的交付回放以“the ball touched the ground”结束，只有 touch 1–3，最终评分器 metric_touches=3。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 2.40–3.36 s，机器人双脚落地并调整支撑站姿以接续颠球；下落的足球位于脚侧方，没有第四次脚球接触。"
+     }
+    ]
+   },
+   "juggles_counted_5": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 3.36,
+      "log_step": 499,
+      "text_zh": "观察：视频 3.20–3.36 s，足球在机器人脚侧落地，录像结束前没有后续击起。推断：结合交付回放（log step 499）仅列出 touch 1–3，以及最终评分器 stages_reached=3、metric_touches=3，顺序计数的第5次颠球未完成，且未进入该次击球。"
+     }
+    ]
+   },
+   "juggles_counted_6": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 3.36,
+      "log_step": 499,
+      "text_zh": "观察：视频 3.20–3.36 s，足球在机器人脚侧落地，录像结束前没有后续击起。推断：结合交付回放（log step 499）仅列出 touch 1–3，以及最终评分器 stages_reached=3、metric_touches=3，顺序计数的第6次颠球未完成，且未进入该次击球。"
+     }
+    ]
+   },
+   "juggles_counted_7": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 3.36,
+      "log_step": 499,
+      "text_zh": "观察：视频 3.20–3.36 s，足球在机器人脚侧落地，录像结束前没有后续击起。推断：结合交付回放（log step 499）仅列出 touch 1–3，以及最终评分器 stages_reached=3、metric_touches=3，顺序计数的第7次颠球未完成，且未进入该次击球。"
+     }
+    ]
+   },
+   "juggles_counted_8": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 3.36,
+      "log_step": 499,
+      "text_zh": "观察：视频 3.20–3.36 s，足球在机器人脚侧落地，录像结束前没有后续击起。推断：结合交付回放（log step 499）仅列出 touch 1–3，以及最终评分器 stages_reached=3、metric_touches=3，顺序计数的第8次颠球未完成，且未进入该次击球。"
+     }
+    ]
+   },
+   "juggles_counted_9": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 3.36,
+      "log_step": 499,
+      "text_zh": "观察：视频 3.20–3.36 s，足球在机器人脚侧落地，录像结束前没有后续击起。推断：结合交付回放（log step 499）仅列出 touch 1–3，以及最终评分器 stages_reached=3、metric_touches=3，顺序计数的第9次颠球未完成，且未进入该次击球。"
+     }
+    ]
+   },
+   "juggles_counted_10": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 3.36,
+      "log_step": 499,
+      "text_zh": "观察：视频 3.20–3.36 s，足球在机器人脚侧落地，录像结束前没有后续击起。推断：结合交付回放（log step 499）仅列出 touch 1–3，以及最终评分器 stages_reached=3、metric_touches=3，顺序计数的第10次颠球未完成，且未进入该次击球。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "touches",
+   "value": 3.0,
+   "unit": "count",
+   "threshold": 10,
+   "direction": "higher_better"
+  },
+  {
+   "name": "ball_height_m",
+   "value": 0.088,
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better"
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "juggles_counted_4",
+   "t": 3.2,
+   "log_step": 499,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 2.40–3.36 s，机器人在第三次击球后收腿、调整站姿，但没跟上足球的下落路径；视频 3.20 s 球已在双脚侧方下落，视频 3.36 s 球落到草地，机器人仍站立。交付文件回放（log step 499、501）均只有前三次计数并以球触地结束，最终评分器 stages_reached=3、fell=0。推断：未修复的断点为接续第4次颠球。"
+  },
+  "unmet": {
+   "juggles_counted_1": true,
+   "juggles_counted_2": true,
+   "juggles_counted_3": true,
+   "juggles_counted_4": false,
+   "juggles_counted_5": false,
+   "juggles_counted_6": false,
+   "juggles_counted_7": false,
+   "juggles_counted_8": false,
+   "juggles_counted_9": false,
+   "juggles_counted_10": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "动态时机",
+    "t": 3.2,
+    "log_step": 499,
+    "evidence_zh": "观察：视频 2.96–3.36 s，足球从高处落向机器人侧方，机器人调整双脚但没有跟随到球下方，球直接落地，未出现第四次脚部向上击球。推断：机器人错过了运动中足球的接续击球窗口；此为同一次未跟上的失误。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "t": 3.2,
+    "log_step": 477,
+    "source_zh": "日志 / 视频",
+    "evidence_zh": "观察：log step 477 的脚本替换内容显示，保存生成器在第三次计数后持续使用零平移、零转向目标的 walk_falcon 恢复命令直至回合结束；该步将追球命令写入另一个会话测试文件，未替换保存生成器。log step 499、501 对交付文件的回放均为相同337行、相同末态哈希、三次计数后球触地。视频 2.40–3.36 s 显示机器人恢复站姿但球与脚的水平间隔增大，始终未在下落时补上击球。推断：交付控制序列只实现了前三次击球及站姿恢复，未实现随球移动和接续触球；据此归为控制，证据不足以进一步确定控制子类。"
+   }
+  ],
+  "gap_zh": "没有状态日志；25帧每秒录像无法独立核验每次约0.01秒的触球历时及速度门槛，前三次有效性以交付文件回放的裁判事件和最终评分器计数补充。节点达到时刻按已检查的视频帧取近似值，未将agent工作时间或会话测试的时间轴换成视频时间；无法进一步区分接续控制的具体子类。"
+ },
+ "summary_zh": "机器人放球后先用左脚、再连续两次用右脚将球击起，完成第1–3次有效颠球，最终进度为3/10。第4次接续时，机器人恢复站姿却未跟上足球的下落路径，球在视频约3.36 s落地，机器人未跌倒；交付回放确认三次计数后以球触地结束，根因归为控制。没有状态日志，毫秒接触及速度合法性由同一交付文件的裁判回放补充，达到时刻为视频近似值。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "节点时间与交付回放裁判事件（touch 1/2/3 于 0.634/1.24/2.156 s）及视频帧一致；断点在第4次颠球、未跌倒，现象与根因（控制）有日志和画面支撑，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": "2026-10-31T00:55:43Z",
+  "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-juggling-feet-i00-privileged--o6ebpor/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-juggling-feet-i00-privileged--o6ebpor/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "3573e7f826423a57140d32ad42482364c6157254f1f81f3143b1d1042ea26273",
+   "video_sha256": "b0fbdace00fc6d2a208d72a7953cfe65d1507ac12da53c1f8972e86118af1e6a",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/juggling_feet/trajectories/embodied-olympics-fun-juggling-feet-i00-privileged--sfy2pjr.json

@@ -0,0 +1,403 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-juggling-feet-i00-privileged--sfy2pjr",
+ "taskKey": "embodied-olympics-fun/juggling_feet",
+ "dag_sha256": "6fb5f868dd0ec1e7ac4f8d9035cd8456789adc45e6c5ceda5fec1ad9d3ba10f6",
+ "outcome": {
+  "success": 0,
+  "reward": 0.4,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "n_actions": 1497,
+   "sim_steps": 1497,
+   "horizon": 6000,
+   "episode_over": 1,
+   "progress": 0.4,
+   "stages_reached": 4,
+   "stages_total": 10,
+   "fell": 0,
+   "replay_s": 9.09,
+   "bad_trajectory": 0,
+   "metric_ball_height_m": 0.354,
+   "metric_bounces": 1.0,
+   "metric_touches": 4.0,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 146,
+   "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": 2.802,
+   "peak_processes": 21,
+   "peak_disk_gb": 0.253,
+   "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-juggling-feet-i00-privileged--sfy2pjr/replay.mp4",
+  "duration_s": 15.0,
+  "fps": 25.0,
+  "frames": 375,
+  "sha256": "88cafa21ec8b0de751d89ec8a246406c7104b658eb9227489849f3593e37ef7e"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 10,
+  "base": 0,
+  "peak": 4,
+  "final": 4,
+  "nodes": {
+   "juggles_counted_1": {
+    "method": "visual",
+    "intervals": [
+     [
+      12.52,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 12.52,
+      "log_step": null,
+      "text_zh": "观察：视频 12.36–12.40 s 足球在右脚前低处改变运动方向，视频 12.44–12.64 s 离脚上升。机器人始终以左脚支撑，没有跌倒。"
+     },
+     {
+      "t": 12.52,
+      "log_step": 140,
+      "text_zh": "推断：log step 140 对交付文件的重建验证依次输出 strike 0 touches 0、strike 1 touches 1，直至 strike 4 touches 4；log step 141 的独立验证回放也只列出 touch 1 至 touch 4，且 reward.metric_touches=4、deterministic=1。结合视频 11.84–12.16 s 的先行低反弹和后续四次上升，将本次对应第1次有效颠球；视频 12.52 s 是上升达标的近似锚点，误差约0.1–0.2秒，毫秒接触及速度判据采用裁判计数。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 0–11.20 s 双掌从两侧挤持足球，视频 11.28–11.60 s 双臂前摆并分开释放足球，作为首次颠球的准备；候选接触名为“双掌挤压”，单掌没有包围球，不能标抓握，现有方式无对应项（D12）。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：机器人以右脚短促击打足球后球离脚上升；候选接触名为“击打”，现有接触方式无对应项。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：左脚接触草地承重，在右脚迎球和击球时支撑机器人。"
+     }
+    ]
+   },
+   "juggles_counted_2": {
+    "method": "visual",
+    "intervals": [
+     [
+      13.04,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 13.04,
+      "log_step": null,
+      "text_zh": "观察：视频 12.88–12.92 s 足球再次到达右脚处，视频 12.96–13.20 s 离脚上升。机器人始终以左脚支撑，没有跌倒。"
+     },
+     {
+      "t": 13.04,
+      "log_step": 140,
+      "text_zh": "推断：log step 140 对交付文件的重建验证依次输出 strike 0 touches 0、strike 1 touches 1，直至 strike 4 touches 4；log step 141 的独立验证回放也只列出 touch 1 至 touch 4，且 reward.metric_touches=4、deterministic=1。结合视频 11.84–12.16 s 的先行低反弹和后续四次上升，将本次对应第2次有效颠球；视频 13.04 s 是上升达标的近似锚点，误差约0.1–0.2秒，毫秒接触及速度判据采用裁判计数。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：机器人以右脚短促击打足球后球离脚上升；候选接触名为“击打”，现有接触方式无对应项。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：左脚接触草地承重，在右脚迎球和击球时支撑机器人。"
+     }
+    ]
+   },
+   "juggles_counted_3": {
+    "method": "visual",
+    "intervals": [
+     [
+      13.6,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 13.6,
+      "log_step": null,
+      "text_zh": "观察：视频 13.48–13.52 s 足球到达右脚处，视频 13.56–13.84 s 离脚上升至更高处。机器人始终以左脚支撑，没有跌倒。"
+     },
+     {
+      "t": 13.6,
+      "log_step": 140,
+      "text_zh": "推断：log step 140 对交付文件的重建验证依次输出 strike 0 touches 0、strike 1 touches 1，直至 strike 4 touches 4；log step 141 的独立验证回放也只列出 touch 1 至 touch 4，且 reward.metric_touches=4、deterministic=1。结合视频 11.84–12.16 s 的先行低反弹和后续四次上升，将本次对应第3次有效颠球；视频 13.60 s 是上升达标的近似锚点，误差约0.1–0.2秒，毫秒接触及速度判据采用裁判计数。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：机器人以右脚短促击打足球后球离脚上升；候选接触名为“击打”，现有接触方式无对应项。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：左脚接触草地承重，在右脚迎球和击球时支撑机器人。"
+     }
+    ]
+   },
+   "juggles_counted_4": {
+    "method": "visual",
+    "intervals": [
+     [
+      14.32,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 14.32,
+      "log_step": null,
+      "text_zh": "观察：视频 14.20–14.24 s 足球到达右脚处，视频 14.28–14.60 s 明显离脚上升，随后再下落。机器人始终以左脚支撑，没有跌倒。"
+     },
+     {
+      "t": 14.32,
+      "log_step": 140,
+      "text_zh": "推断：log step 140 对交付文件的重建验证依次输出 strike 0 touches 0、strike 1 touches 1，直至 strike 4 touches 4；log step 141 的独立验证回放也只列出 touch 1 至 touch 4，且 reward.metric_touches=4、deterministic=1。结合视频 11.84–12.16 s 的先行低反弹和后续四次上升，将本次对应第4次有效颠球；视频 14.32 s 是上升达标的近似锚点，误差约0.1–0.2秒，毫秒接触及速度判据采用裁判计数。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：机器人以右脚短促击打足球后球离脚上升；候选接触名为“击打”，现有接触方式无对应项。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：左脚接触草地承重，在右脚迎球和击球时支撑机器人。"
+     }
+    ]
+   },
+   "juggles_counted_5": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 14.96,
+      "log_step": 141,
+      "text_zh": "观察：视频 14.72–14.96 s 足球从高处下落至机器人右小腿附近，录像随即结束，没有后续合法离脚上升；log step 141 针对交付文件的验证回放报告 failed=\"the ball touched a part other than a foot\"、stages 仅 touch 1 至 touch 4，reward.stages_reached=4、metric_touches=4。推断：累计第5次有效颠球未达到。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频 14.96 s 足球到达右小腿前侧，交付文件验证回放在 log step 141 确认发生非脚部触球；agent 在 log step 142–143 称其为小腿犯规。推断：实际接触为右小腿与足球碰撞，候选接触名“碰撞”，现有方式无对应项；未形成第五次合法脚部击打。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 14.72–14.96 s 左脚继续接触草地承重，右腿准备迎接下落的足球，机器人保持站立。"
+     }
+    ]
+   },
+   "juggles_counted_6": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 14.96,
+      "log_step": 141,
+      "text_zh": "观察：视频 14.72–14.96 s 足球从高处下落至机器人右小腿附近，录像随即结束，没有后续合法离脚上升；log step 141 针对交付文件的验证回放报告 failed=\"the ball touched a part other than a foot\"、stages 仅 touch 1 至 touch 4，reward.stages_reached=4、metric_touches=4。推断：累计第6次有效颠球未达到。"
+     }
+    ]
+   },
+   "juggles_counted_7": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 14.96,
+      "log_step": 141,
+      "text_zh": "观察：视频 14.72–14.96 s 足球从高处下落至机器人右小腿附近，录像随即结束，没有后续合法离脚上升；log step 141 针对交付文件的验证回放报告 failed=\"the ball touched a part other than a foot\"、stages 仅 touch 1 至 touch 4，reward.stages_reached=4、metric_touches=4。推断：累计第7次有效颠球未达到。"
+     }
+    ]
+   },
+   "juggles_counted_8": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 14.96,
+      "log_step": 141,
+      "text_zh": "观察：视频 14.72–14.96 s 足球从高处下落至机器人右小腿附近，录像随即结束，没有后续合法离脚上升；log step 141 针对交付文件的验证回放报告 failed=\"the ball touched a part other than a foot\"、stages 仅 touch 1 至 touch 4，reward.stages_reached=4、metric_touches=4。推断：累计第8次有效颠球未达到。"
+     }
+    ]
+   },
+   "juggles_counted_9": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 14.96,
+      "log_step": 141,
+      "text_zh": "观察：视频 14.72–14.96 s 足球从高处下落至机器人右小腿附近，录像随即结束，没有后续合法离脚上升；log step 141 针对交付文件的验证回放报告 failed=\"the ball touched a part other than a foot\"、stages 仅 touch 1 至 touch 4，reward.stages_reached=4、metric_touches=4。推断：累计第9次有效颠球未达到。"
+     }
+    ]
+   },
+   "juggles_counted_10": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 14.96,
+      "log_step": 141,
+      "text_zh": "观察：视频 14.72–14.96 s 足球从高处下落至机器人右小腿附近，录像随即结束，没有后续合法离脚上升；log step 141 针对交付文件的验证回放报告 failed=\"the ball touched a part other than a foot\"、stages 仅 touch 1 至 touch 4，reward.stages_reached=4、metric_touches=4。推断：累计第10次有效颠球未达到。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "touches",
+   "value": 4.0,
+   "unit": "count",
+   "threshold": 10,
+   "direction": "higher_better"
+  },
+  {
+   "name": "ball_height_m",
+   "value": 0.354,
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better"
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "juggles_counted_5",
+   "t": 14.96,
+   "log_step": 141,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 14.20–14.60 s 右脚击球后足球上升，视频 14.72–14.96 s 足球又落向右小腿前侧，右腿正在迎球；末帧仍站立。log step 141 对已交付文件的验证回放确认仅完成四次，因“the ball touched a part other than a foot”终止。推断：失败发生在争取第5次有效颠球的迎击中，此后没有修复。"
+  },
+  "unmet": {
+   "juggles_counted_1": true,
+   "juggles_counted_2": true,
+   "juggles_counted_3": true,
+   "juggles_counted_4": true,
+   "juggles_counted_5": false,
+   "juggles_counted_6": false,
+   "juggles_counted_7": false,
+   "juggles_counted_8": false,
+   "juggles_counted_9": false,
+   "juggles_counted_10": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "动态时机",
+    "t": 14.96,
+    "log_step": 141,
+    "evidence_zh": "观察：视频 14.72–14.96 s 足球快速下落，右腿迎球时足球到达小腿前侧，没有像视频 14.20–14.60 s 那样从合法脚部弹起；log step 141 的交付回放报告非脚部触球。推断：右脚未能在下落球到达时以合法部位迎击，属于与运动中足球的迎击配合失败。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "t": 14.96,
+    "log_step": 140,
+    "evidence_zh": "推断：视频 12.36–14.60 s 已能连续用右脚将球击起，但视频 14.72–14.96 s 的下一次迎击没有把接触维持在合法脚部；log step 140 证实脚本重建的交付轨迹逐项相同，log step 141 的独立回放复现四次后非脚触球，支持迎球控制失败。agent 在 log step 142–143 也自述后续来球偏离落点或速度过快时会碰小腿，但其所称碰撞体尺寸和可行区域未独立验证，不能据此确定接触设计、逆解或反馈的具体子类。",
+    "source_zh": "视频 / 日志 / agent 自述"
+   }
+  ],
+  "gap_zh": "观察：inputs/ 没有本次状态日志，录像为25帧每秒且没有逐帧裁判计数。推断：四次计数与可见上升的配对依赖交付重建日志及确定性验证回放；视频 11.84–12.16 s 首次反弹为何未计数，无法独立区分高度、能量折算速度等门槛。末次非脚触球由验证回放确认，右小腿位置由视频和 agent 自述相互支持，但精确碰撞点、合法区域边界及过早/过晚或跟踪误差的细分机制不能从帧图独立量出。"
+ },
+ "summary_zh": "机器人以左脚支撑、右脚迎球，完成第1至第4次有效颠球，十个节点终态为4/10；视频 11.84–12.16 s 还有一次未计入的低反弹。第5次迎击在视频约14.96 s 因球触及非脚部而结束，机器人未跌倒，断点为第5次颠球，归因为迎球控制失败。没有状态日志及逐帧裁判读数，达标时刻取可见上升的近似时间，计数和非脚部犯规由交付文件的验证回放补足。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "四个节点已达成、第5次未达成，断点在第5次颠球（右小腿触球犯规），帧图与交付回放日志一致；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": "2026-10-31T00:59:26Z",
+  "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-juggling-feet-i00-privileged--sfy2pjr/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-juggling-feet-i00-privileged--sfy2pjr/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "e62213e2e24cdbd51927e7463a4418a5f59875f56e8bcc41fad2f231a660a14a",
+   "video_sha256": "88cafa21ec8b0de751d89ec8a246406c7104b658eb9227489849f3593e37ef7e",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/juggling_feet/trajectories/embodied-olympics-fun-juggling-feet-i00-privileged--uovpsef.json

@@ -0,0 +1,318 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-juggling-feet-i00-privileged--uovpsef",
+ "taskKey": "embodied-olympics-fun/juggling_feet",
+ "dag_sha256": "6fb5f868dd0ec1e7ac4f8d9035cd8456789adc45e6c5ceda5fec1ad9d3ba10f6",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "n_actions": 372,
+   "sim_steps": 372,
+   "horizon": 6000,
+   "episode_over": 1,
+   "progress": 0.0,
+   "stages_reached": 0,
+   "stages_total": 10,
+   "fell": 0,
+   "replay_s": 7.69,
+   "bad_trajectory": 0,
+   "metric_ball_height_m": 0.302,
+   "metric_bounces": 0.0,
+   "metric_touches": 0.0,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 132,
+   "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": 0.0,
+   "peak_memory_gb": 1.416,
+   "peak_processes": 10,
+   "peak_disk_gb": 0.98,
+   "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-juggling-feet-i00-privileged--uovpsef/replay.mp4",
+  "duration_s": 3.76,
+  "fps": 25.0,
+  "frames": 94,
+  "sha256": "c8b6fdce221814fb612aa9a7f0527b42e93407363397a1513204c0c6888d79d4"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 10,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "juggles_counted_1": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频初态及视频0.20–3.20 s，机器人以双掌从两侧夹住足球并调整腿部站姿；没有可见的脚部击球后离脚上升。"
+     },
+     {
+      "t": 3.72,
+      "log_step": 162,
+      "text_zh": "观察：视频3.32–3.72 s，机器人逐渐侧倾下沉，足球由腰前下落至大腿附近，未出现一次合法脚部击打后的升高。log step 162对交付文件的回放报告stages=[]、touches=0、bounces=0，并以“the ball touched a part other than a foot”结束；inputs/run.json评分字段stages_reached=0、metric_touches=0与此一致。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频0.00–3.32 s可见两只平掌从足球两侧挤压夹持，随后放球；候选接触名“对向挤压夹持”。单只手没有包围足球，因此不标抓握；这是首次颠球尝试中的放球准备。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频0.00–3.72 s，机器人双脚与草地承重接触，在调整击球站姿和放球时支撑身体；末段身体侧倾下沉。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频3.72 s足球下落至机器人右腿上部并与其相贴，交付回放log step 162确认非脚部触球；候选接触名“非预期碰撞”，不是可见的脚部击打。"
+     }
+    ]
+   },
+   "juggles_counted_2": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 3.72,
+      "log_step": 162,
+      "text_zh": "推断：第2次有效颠球未达到。视频0.00–3.72 s只有持球、站姿调整及末段球下落，未见连续脚部击球后升高；log step 162的交付回放与inputs/run.json评分均报告有效计数0、已达阶段0/10。首次计数未发生，因此后续累计第2次计数也未发生。"
+     }
+    ]
+   },
+   "juggles_counted_3": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 3.72,
+      "log_step": 162,
+      "text_zh": "推断：第3次有效颠球未达到。视频0.00–3.72 s只有持球、站姿调整及末段球下落，未见连续脚部击球后升高；log step 162的交付回放与inputs/run.json评分均报告有效计数0、已达阶段0/10。首次计数未发生，因此后续累计第3次计数也未发生。"
+     }
+    ]
+   },
+   "juggles_counted_4": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 3.72,
+      "log_step": 162,
+      "text_zh": "推断：第4次有效颠球未达到。视频0.00–3.72 s只有持球、站姿调整及末段球下落，未见连续脚部击球后升高；log step 162的交付回放与inputs/run.json评分均报告有效计数0、已达阶段0/10。首次计数未发生，因此后续累计第4次计数也未发生。"
+     }
+    ]
+   },
+   "juggles_counted_5": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 3.72,
+      "log_step": 162,
+      "text_zh": "推断：第5次有效颠球未达到。视频0.00–3.72 s只有持球、站姿调整及末段球下落，未见连续脚部击球后升高；log step 162的交付回放与inputs/run.json评分均报告有效计数0、已达阶段0/10。首次计数未发生，因此后续累计第5次计数也未发生。"
+     }
+    ]
+   },
+   "juggles_counted_6": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 3.72,
+      "log_step": 162,
+      "text_zh": "推断：第6次有效颠球未达到。视频0.00–3.72 s只有持球、站姿调整及末段球下落，未见连续脚部击球后升高；log step 162的交付回放与inputs/run.json评分均报告有效计数0、已达阶段0/10。首次计数未发生，因此后续累计第6次计数也未发生。"
+     }
+    ]
+   },
+   "juggles_counted_7": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 3.72,
+      "log_step": 162,
+      "text_zh": "推断：第7次有效颠球未达到。视频0.00–3.72 s只有持球、站姿调整及末段球下落，未见连续脚部击球后升高；log step 162的交付回放与inputs/run.json评分均报告有效计数0、已达阶段0/10。首次计数未发生，因此后续累计第7次计数也未发生。"
+     }
+    ]
+   },
+   "juggles_counted_8": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 3.72,
+      "log_step": 162,
+      "text_zh": "推断：第8次有效颠球未达到。视频0.00–3.72 s只有持球、站姿调整及末段球下落，未见连续脚部击球后升高；log step 162的交付回放与inputs/run.json评分均报告有效计数0、已达阶段0/10。首次计数未发生，因此后续累计第8次计数也未发生。"
+     }
+    ]
+   },
+   "juggles_counted_9": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 3.72,
+      "log_step": 162,
+      "text_zh": "推断：第9次有效颠球未达到。视频0.00–3.72 s只有持球、站姿调整及末段球下落，未见连续脚部击球后升高；log step 162的交付回放与inputs/run.json评分均报告有效计数0、已达阶段0/10。首次计数未发生，因此后续累计第9次计数也未发生。"
+     }
+    ]
+   },
+   "juggles_counted_10": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 3.72,
+      "log_step": 162,
+      "text_zh": "推断：第10次有效颠球未达到。视频0.00–3.72 s只有持球、站姿调整及末段球下落，未见连续脚部击球后升高；log step 162的交付回放与inputs/run.json评分均报告有效计数0、已达阶段0/10。首次计数未发生，因此后续累计第10次计数也未发生。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "touches",
+   "unit": "count",
+   "threshold": 10,
+   "direction": "higher_better",
+   "value": 0.0
+  },
+  {
+   "name": "ball_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": 0.302
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "juggles_counted_1",
+   "t": 3.4,
+   "log_step": 162,
+   "where": "操作中",
+   "evidence_zh": "观察：视频3.32–3.72 s，机器人正在放球及尝试首次颠球，身体持续向画面左侧侧倾下沉，球落向腿部，直到末帧仍未恢复直立或产生一次离脚上升；log step 162的交付回放确认非脚部触球终止、0/10阶段。log step 190再次生成交付文件所得校验和不变，说明后续会话测试没有替换这段交付。断点取首次颠球的失稳过程，非脚部触球是此步的终止事件。"
+  },
+  "unmet": {
+   "juggles_counted_1": false,
+   "juggles_counted_2": false,
+   "juggles_counted_3": false,
+   "juggles_counted_4": false,
+   "juggles_counted_5": false,
+   "juggles_counted_6": false,
+   "juggles_counted_7": false,
+   "juggles_counted_8": false,
+   "juggles_counted_9": false,
+   "juggles_counted_10": false
+  },
+  "phenomena": [
+   {
+    "cat": "全身",
+    "item": "平衡",
+    "t": 3.6,
+    "log_step": null,
+    "evidence_zh": "观察：视频3.32–3.72 s逐帧可见机器人侧倾不断加大，躯干下沉、双腿斜伸，至末帧没有恢复稳定站立；足球同时落至腿部附近，未被脚向上颠起。这里标可见的全身失稳；inputs/run.json的fell=0表示评分器在非脚部触球终止前尚未判定跌倒，不把会话中其他试验的摔倒算入本轨迹。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "evidence_zh": "推断：站姿变换及放球阶段的全身姿态控制未能稳住机器人。log step 132记录交付生成器对右腿关节作插值后保持目标，以骨盆偏航反馈修正左腿逆解并施加支撑前馈力矩；log step 162、190确认该生��器交付的是相同372行动的可重复失败轨迹。视频0.60–3.32 s可见机器人逐渐转身，视频3.32–3.72 s进一步侧倾下沉，表明实际全身运动没有保持生成器所需的稳定击球站姿。证据不足以进一步断定是接触布局、逆解目标、反馈律还是某个增益参数所致，故停在“控制”。",
+    "source_zh": "日志 / 视频"
+   }
+  ],
+  "gap_zh": "没有逐帧状态日志，固定远景不能精确分辨末段接触点及毫秒接触顺序；非脚部触球由交付文件回放结果确认，未凭会话测试的具体身体部位替代末帧证据。缺少全身受力和姿态跟踪诊断，无法可靠细分控制根因；未将25 fps视频用于估算0.05秒接触或0.95倍速度门槛。"
+ },
+ "summary_zh": "机器人完成持球状态下的站姿调整并开始放球，但首次有效颠球未发生，十个计数节点均未达到，评分器有效触球和反弹计数均为0。视频3.32–3.72 s显示机器人持续侧倾下沉、球落向腿部，交付回放以非脚部触球结束；日志中的生成器与可重复回放支持将根因保守归为控制。缺少逐帧状态和受力数据，无法精确定位碰撞点或细分失稳的控制原因。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "十个节点均未达到与评分器（stages_reached=0、touches=0）一致；视频3.2–3.7 s可见机器人侧倾下沉、球仍在腰腹处后落向腿部，断点、平衡现象和控制根因有证据支持，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": "2026-10-31T00:54:33Z",
+  "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-juggling-feet-i00-privileged--uovpsef/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-juggling-feet-i00-privileged--uovpsef/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "3335acbeaa990c68599b4ea15bbd4b3bc6f8d93af956aa799dae3e74f4bae1f0",
+   "video_sha256": "c8b6fdce221814fb612aa9a7f0527b42e93407363397a1513204c0c6888d79d4",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/juggling_feet/trajectories/embodied-olympics-fun-juggling-feet-i00-privileged--x6wdyax.json

@@ -0,0 +1,314 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-juggling-feet-i00-privileged--x6wdyax",
+ "taskKey": "embodied-olympics-fun/juggling_feet",
+ "dag_sha256": "6fb5f868dd0ec1e7ac4f8d9035cd8456789adc45e6c5ceda5fec1ad9d3ba10f6",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "n_actions": 573,
+   "sim_steps": 573,
+   "horizon": 6000,
+   "episode_over": 1,
+   "progress": 0.0,
+   "stages_reached": 0,
+   "stages_total": 10,
+   "fell": 1,
+   "replay_s": 3.57,
+   "bad_trajectory": 0,
+   "metric_ball_height_m": 0.208,
+   "metric_bounces": 0.0,
+   "metric_touches": 0.0,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 60,
+   "agent_resets": 89,
+   "agent_replays": 0,
+   "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": 0.496,
+   "peak_processes": 11,
+   "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-juggling-feet-i00-privileged--x6wdyax/replay.mp4",
+  "duration_s": 5.76,
+  "fps": 25.0,
+  "frames": 144,
+  "sha256": "df24ddb2ea0cdda1dd5a37189fe08e1b6568207d3dc19f27be7ac036955f37fb"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 10,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "juggles_counted_1": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 4.8,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–4.80 s，机器人双掌持续挤持足球，并转动上身、调整腿部姿态；球没有从脚部击打后离脚上升。"
+     },
+     {
+      "t": 5.72,
+      "log_step": 34,
+      "text_zh": "观察：视频 5.32–5.72 s，机器人持球倾倒、骨盆快速降低，足球随双手下降，未出现一次离脚上升的颠球；log step 34 保存的轨迹为573步、fell、touches=0，与评分器 stages_reached=0、metric_touches=0、fell=1一致。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 0.00–5.72 s，机器人两掌从两侧挤持足球，并在调整姿态时一起带动球；候选接触名“双掌挤持”，两只平掌均未包围足球，不属于抓握，现有方式没有对应项。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–5.32 s，机器人双脚接触草地承重，并在首次颠球的准备中调整腿部姿态；视频 5.32–5.72 s 支撑失稳。未见脚对足球完成击打。"
+     }
+    ]
+   },
+   "juggles_counted_2": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 5.72,
+      "log_step": null,
+      "text_zh": "推断：视频 0.00–5.72 s 的足球一直随双掌移动，末段随机器人倒下，未见连续离脚上升；评分器 stages_reached=0、metric_touches=0。第1次有效颠球未达成，故依次计数的第2次颠球亦未达成，机器人尚未进入该次尝试。"
+     }
+    ]
+   },
+   "juggles_counted_3": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 5.72,
+      "log_step": null,
+      "text_zh": "推断：视频 0.00–5.72 s 的足球一直随双掌移动，末段随机器人倒下，未见连续离脚上升；评分器 stages_reached=0、metric_touches=0。第1次有效颠球未达成，故依次计数的第3次颠球亦未达成，机器人尚未进入该次尝试。"
+     }
+    ]
+   },
+   "juggles_counted_4": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 5.72,
+      "log_step": null,
+      "text_zh": "推断：视频 0.00–5.72 s 的足球一直随双掌移动，末段随机器人倒下，未见连续离脚上升；评分器 stages_reached=0、metric_touches=0。第1次有效颠球未达成，故依次计数的第4次颠球亦未达成，机器人尚未进入该次尝试。"
+     }
+    ]
+   },
+   "juggles_counted_5": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 5.72,
+      "log_step": null,
+      "text_zh": "推断：视频 0.00–5.72 s 的足球一直随双掌移动，末段随机器人倒下，未见连续离脚上升；评分器 stages_reached=0、metric_touches=0。第1次有效颠球未达成，故依次计数的第5次颠球亦未达成，机器人尚未进入该次尝试。"
+     }
+    ]
+   },
+   "juggles_counted_6": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 5.72,
+      "log_step": null,
+      "text_zh": "推断：视频 0.00–5.72 s 的足球一直随双掌移动，末段随机器人倒下，未见连续离脚上升；评分器 stages_reached=0、metric_touches=0。第1次有效颠球未达成，故依次计数的第6次颠球亦未达成，机器人尚未进入该次尝试。"
+     }
+    ]
+   },
+   "juggles_counted_7": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 5.72,
+      "log_step": null,
+      "text_zh": "推断：视频 0.00–5.72 s 的足球一直随双掌移动，末段随机器人倒下，未见连续离脚上升；评分器 stages_reached=0、metric_touches=0。第1次有效颠球未达成，故依次计数的第7次颠球亦未达成，机器人尚未进入该次尝试。"
+     }
+    ]
+   },
+   "juggles_counted_8": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 5.72,
+      "log_step": null,
+      "text_zh": "推断：视频 0.00–5.72 s 的足球一直随双掌移动，末段随机器人倒下，未见连续离脚上升；评分器 stages_reached=0、metric_touches=0。第1次有效颠球未达成，故依次计数的第8次颠球亦未达成，机器人尚未进入该次尝试。"
+     }
+    ]
+   },
+   "juggles_counted_9": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 5.72,
+      "log_step": null,
+      "text_zh": "推断：视频 0.00–5.72 s 的足球一直随双掌移动，末段随机器人倒下，未见连续离脚上升；评分器 stages_reached=0、metric_touches=0。第1次有效颠球未达成，故依次计数的第9次颠球亦未达成，机器人尚未进入该次尝试。"
+     }
+    ]
+   },
+   "juggles_counted_10": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 5.72,
+      "log_step": null,
+      "text_zh": "推断：视频 0.00–5.72 s 的足球一直随双掌移动，末段随机器人倒下，未见连续离脚上升；评分器 stages_reached=0、metric_touches=0。第1次有效颠球未达成，故依次计数的第10次颠球亦未达成，机器人尚未进入该次尝试。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "touches",
+   "value": 0.0,
+   "unit": "count",
+   "threshold": 10,
+   "direction": "higher_better"
+  },
+  {
+   "name": "ball_height_m",
+   "value": 0.208,
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better"
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "juggles_counted_1",
+   "t": 5.3,
+   "log_step": 34,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 5.32–5.72 s，机器人在持球调整腿部、准备首次颠球时身体持续倾倒、骨盆降近草地，未恢复站立，也未完成脚部击球。log step 34 明确保存573步、fell、touches=0的轨迹，评分器同样记录 fell=1、stages_reached=0；失败位于第1次颠球的准备操作中。"
+  },
+  "unmet": {
+   "juggles_counted_1": false,
+   "juggles_counted_2": false,
+   "juggles_counted_3": false,
+   "juggles_counted_4": false,
+   "juggles_counted_5": false,
+   "juggles_counted_6": false,
+   "juggles_counted_7": false,
+   "juggles_counted_8": false,
+   "juggles_counted_9": false,
+   "juggles_counted_10": false
+  },
+  "phenomena": [
+   {
+    "cat": "全身",
+    "item": "平衡",
+    "t": 5.3,
+    "log_step": null,
+    "evidence_zh": "观察：视频 5.32–5.72 s，机器人持球向前侧方倾倒，双腿弯折，骨盆快速下降，末帧已接近草地且未恢复；评分器 fell=1。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "log_step": 34,
+    "evidence_zh": "推断：视频 4.72–5.72 s 显示机器人在调整腿部支撑时姿态不断偏离直立并倒下；log step 21 的同一run2方案在准备右脚悬停时报告失衡，log step 34 随后以该方案保存573步失败轨迹，log step 95 的交付说明确认保留此早期轨迹。首次脚触球之前的全身动作未得到稳定控制，导致本次失败；不足以区分轨迹设计、反馈跟踪或具体参数问题。",
+    "source_zh": "视频 / 日志"
+   }
+  ],
+  "gap_zh": "没有状态日志，固定场边视频不能确定精确接触力、重心和关节跟踪误差，故控制根因不细分。视频未显示脚部击球或离脚上升，裁判有效计数也为0，这些缺口不影响十个节点均未达成的判断。后续会话调试未覆盖log step 34保存的交付文件；节点及现象仅依据本次回放，不采用后续测试结果。"
+ },
+ "summary_zh": "机器人进行了持球转身和首次颠球的姿态准备，但有效颠球为0次，十个节点均未达成。视频约 5.3–5.72 s，机器人仍双掌夹球便失衡倒下，断在第1次颠球操作中，根因可归于全身控制未稳定。缺少状态日��，无法进一步区分具体轨迹、反馈或参数问题；交付为log step 34保存的早期失败轨迹。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "十个节点均未达成、断点在第1次颠球（平衡/全身倒下）、根因停在控制均有视频与评分器 fell=1、touches=0 支持；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": "2026-10-31T00:57:14Z",
+  "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-juggling-feet-i00-privileged--x6wdyax/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-juggling-feet-i00-privileged--x6wdyax/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "0a348762112ce5d8c7b081a3e248fe2d075c78f48e1aa6dd6f8fe6ee65cc2f60",
+   "video_sha256": "df24ddb2ea0cdda1dd5a37189fe08e1b6568207d3dc19f27be7ac036955f37fb",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/juggling_feet/trajectories/embodied-olympics-fun-juggling-feet-i00-privileged--zyxoa98.json

@@ -0,0 +1,488 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-juggling-feet-i00-privileged--zyxoa98",
+ "taskKey": "embodied-olympics-fun/juggling_feet",
+ "dag_sha256": "6fb5f868dd0ec1e7ac4f8d9035cd8456789adc45e6c5ceda5fec1ad9d3ba10f6",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "n_actions": 926,
+   "sim_steps": 926,
+   "horizon": 6000,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 10,
+   "stages_total": 10,
+   "fell": 0,
+   "replay_s": 5.43,
+   "bad_trajectory": 0,
+   "metric_ball_height_m": 0.508,
+   "metric_bounces": 0.0,
+   "metric_touches": 10.0,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 79,
+   "agent_resets": 21,
+   "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": 0.767,
+   "peak_processes": 10,
+   "peak_disk_gb": 0.055,
+   "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-juggling-feet-i00-privileged--zyxoa98/replay.mp4",
+  "duration_s": 9.28,
+  "fps": 25.0,
+  "frames": 232,
+  "sha256": "71ef3c97e05ef20d7bbbb4fac5b078b28d5a2828bef415e187cba0571d5f64e5"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 10,
+  "base": 0,
+  "peak": 10,
+  "final": 10,
+  "nodes": {
+   "juggles_counted_1": {
+    "method": "visual",
+    "intervals": [
+     [
+      3.0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 2.64,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00 s 足球位于机器人两掌之间；视频 2.40–2.52 s 仍由双掌夹持，视频 2.64–2.84 s 两掌分开，足球离开双手向左脚下落。放球属于首次颠球的准备动作。"
+     },
+     {
+      "t": 3.0,
+      "log_step": null,
+      "text_zh": "观察：视频 2.84–2.92 s，足球下落至机器人左脚（画面右侧前伸脚）后转为上升；至视频 3.12 s，球已离脚，机器人右脚仍支撑草地。这是放球后按视频顺序辨认的第1次击球。"
+     },
+     {
+      "t": 3.0,
+      "log_step": null,
+      "text_zh": "推断：以视频 3.00 s 附近的离脚上升作为第1次有效颠球达到高度条件的近似时刻，误差约0.1 s；评分器 reward.stages_reached=10、metric_touches=10、success=1 支持该次最终计入。25帧每秒画面不能独立精测接触历时或0.95倍离球速度门槛。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 0.00–2.52 s，机器人两侧平掌共同挤压并移动足球，随后分开双掌放球；候选名为“双掌挤压夹持”，两手均未包合足球，不标抓握；此接触先于首次脚部击打。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频 2.84–3.12 s，机器人前伸的左脚向上击打足球，球随后离脚上升；候选接触名为“击打”，现有方式没有击打项，推/扫或托/舀不能准确描述这次短促踢击。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 2.84–3.12 s，机器人右脚接触草地并承重，支撑左脚颠球时的身体。"
+     }
+    ]
+   },
+   "juggles_counted_2": {
+    "method": "visual",
+    "intervals": [
+     [
+      3.72,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 3.72,
+      "log_step": null,
+      "text_zh": "观察：视频 3.56–3.64 s，足球下落至机器人左脚（画面右侧前伸脚）后转为上升；至视频 3.84 s，球已离脚，机器人右脚仍支撑草地。这是放球后按视频顺序辨认的第2次击球。"
+     },
+     {
+      "t": 3.72,
+      "log_step": null,
+      "text_zh": "推断：以视频 3.72 s 附近的离脚上升作为第2次有效颠球达到高度条件的近似时刻，误差约0.1 s；评分器 reward.stages_reached=10、metric_touches=10、success=1 支持该次最终计入。25帧每秒画面不能独立精测接触历时或0.95倍离球速度门槛。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频 3.56–3.84 s，机器人前伸的左脚向上击打足球，球随后离脚上升；候选接触名为“击打”，现有方式没有击打项，推/扫或托/舀不能准确描述这次短促踢击。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 3.56–3.84 s，机器��右脚接触草地并承重，支撑左脚颠球时的身体。"
+     }
+    ]
+   },
+   "juggles_counted_3": {
+    "method": "visual",
+    "intervals": [
+     [
+      4.4,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 4.4,
+      "log_step": null,
+      "text_zh": "观察：视频 4.28–4.36 s，足球下落至机器人左脚（画面右侧前伸脚）后转为上升；至视频 4.56 s，球已离脚，机器人右脚仍支撑草地。这是放球后按视频顺序辨认的第3次击球。"
+     },
+     {
+      "t": 4.4,
+      "log_step": null,
+      "text_zh": "推断：以视频 4.40 s 附近的离脚上升作为第3次有效颠球达到高度条件的近似时刻，误差约0.1 s；评分器 reward.stages_reached=10、metric_touches=10、success=1 支持该次最终计入。25帧每秒画面不能独立精测接触历时或0.95倍离球速度门槛。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频 4.28–4.56 s，机器人前伸的左脚向上击打足球，球随后离脚上升；候选接触名为“击打”，现有方式没有击打项，推/扫或托/舀不能准确描述这次短促踢击。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 4.28–4.56 s，机器人右脚接触草地并承重，支撑左脚颠球时的身体。"
+     }
+    ]
+   },
+   "juggles_counted_4": {
+    "method": "visual",
+    "intervals": [
+     [
+      5.12,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 5.12,
+      "log_step": null,
+      "text_zh": "观察：视频 5.00–5.08 s，足球下落至机器人左脚（画面右侧前伸脚）后转为上升；至视频 5.28 s，球已离脚，机器人右脚仍支撑草地。这是放球后按视频顺序辨认的第4次击球。"
+     },
+     {
+      "t": 5.12,
+      "log_step": null,
+      "text_zh": "推断：以视频 5.12 s 附近的离脚上升作为第4次有效颠球达到高度条件的近似时刻，误差约0.1 s；评分器 reward.stages_reached=10、metric_touches=10、success=1 支持该次最终计入。25帧每秒画面不能独立精测接触历时或0.95倍离球速度门槛。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频 5.00–5.28 s，机器人前伸的左脚向上击打足球，球随后离脚上升；候选接触名为“击打”，现有方式没有击打项，推/扫或托/舀不能准确描述这次短促踢击。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 5.00–5.28 s，机器人右脚接触草地并承重，支撑左脚颠球时的身体。"
+     }
+    ]
+   },
+   "juggles_counted_5": {
+    "method": "visual",
+    "intervals": [
+     [
+      5.8,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 5.8,
+      "log_step": null,
+      "text_zh": "观察：视频 5.64–5.72 s，足球下落至机器人左脚（画面右侧前伸脚）后转为上升；至视频 5.96 s，球已离脚，机器人右脚仍支撑草地。这是放球后按视频顺序辨认的第5次击球。"
+     },
+     {
+      "t": 5.8,
+      "log_step": null,
+      "text_zh": "推断：以视频 5.80 s 附近的离脚上升作为第5次有效颠球达到高度条件的近似时刻，误差约0.1 s；评分器 reward.stages_reached=10、metric_touches=10、success=1 支持该次最终计入。25帧每秒画面不能独立精测接触历时或0.95倍离球速度门槛。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频 5.64–5.96 s，机器人前伸的左脚向上击打足球，球随后离脚上升；候选接触名为“击打”，现有方式没有击打项，推/扫或托/舀不能准确描述这次短促踢击。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 5.64–5.96 s，机器人右脚接触草地并承重，支撑左脚颠球时的身体。"
+     }
+    ]
+   },
+   "juggles_counted_6": {
+    "method": "visual",
+    "intervals": [
+     [
+      6.48,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 6.48,
+      "log_step": null,
+      "text_zh": "观察：视频 6.36–6.44 s，足球下落至机器人左脚（画面右侧前伸脚）后转为上升；至视频 6.68 s，球已离脚，机器人右脚仍支撑草地。这是放球后按视频顺序辨认的第6次击球。"
+     },
+     {
+      "t": 6.48,
+      "log_step": null,
+      "text_zh": "推断：以视频 6.48 s 附近的离脚上升作为第6次有效颠球达到高度条件的近似时刻，误差约0.1 s；评分器 reward.stages_reached=10、metric_touches=10、success=1 支持该次最终计入。25帧每秒画面不能独立精测接触历时或0.95倍离球速度门槛。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频 6.36–6.68 s，机器人前伸的左脚向上击打足球，球随后离脚上升；候选接触名为“击打”，现有方式没有击打项，推/扫或托/舀不能准确描述这次短促踢击。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 6.36–6.68 s，机器人右脚接触草地并承重，支撑左脚颠球时的身体。"
+     }
+    ]
+   },
+   "juggles_counted_7": {
+    "method": "visual",
+    "intervals": [
+     [
+      7.2,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 7.2,
+      "log_step": null,
+      "text_zh": "观察：视频 7.04–7.12 s，足球下落至机器人左脚（画面右侧前伸脚）后转为上升；至视频 7.36 s，球已离脚，机器人右脚仍支撑草地。这是放球后按视频顺序辨认的第7次击球。"
+     },
+     {
+      "t": 7.2,
+      "log_step": null,
+      "text_zh": "推断：以视频 7.20 s 附近的离脚上升作为第7次有效颠球达到高度条件的近似时刻，误差约0.1 s；评分器 reward.stages_reached=10、metric_touches=10、success=1 支持该次最终计入。25帧每秒画面不能独立精测接触历时或0.95倍离球速度门槛。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频 7.04–7.36 s，机器人前伸的左脚向上击打足球，球随后离脚上升；候选接触名为“击打”，现有方式没有击打项，推/扫或托/舀不能准确描述这次短促踢击。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 7.04–7.36 s，机器人右脚接触草地并承重，支撑左脚颠球时的身体。"
+     }
+    ]
+   },
+   "juggles_counted_8": {
+    "method": "visual",
+    "intervals": [
+     [
+      7.92,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 7.92,
+      "log_step": null,
+      "text_zh": "观察：视频 7.76–7.84 s，足球下落至机器人左脚（画面右侧前伸脚）后转为上升；至视频 8.08 s，球已离脚，机器人右脚仍支撑草地。这是放球后按视频顺序辨认的第8次击球。"
+     },
+     {
+      "t": 7.92,
+      "log_step": null,
+      "text_zh": "推断：以视频 7.92 s 附近的离脚上升作为第8次有效颠球达到高度条件的近似时刻，误差约0.1 s；评分器 reward.stages_reached=10、metric_touches=10、success=1 支持该次最终计入。25帧每秒画面不能独立精测接触历时或0.95倍离球速度门槛。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频 7.76–8.08 s，机器人前伸的左脚向上击打足球，球随后离脚上升；候选接触名为“击打”，现有方式没有击打项，推/扫或托/舀不能准确描述这次短促踢击。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 7.76–8.08 s，机器人右脚接触草地并承重，支撑左脚颠球时的身体。"
+     }
+    ]
+   },
+   "juggles_counted_9": {
+    "method": "visual",
+    "intervals": [
+     [
+      8.6,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 8.6,
+      "log_step": null,
+      "text_zh": "观察：视频 8.44–8.52 s，足球下落至机器人左脚（画面右侧前伸脚）后转为上升；至视频 8.76 s，球已离脚，机器人右脚仍支撑草地。这是放球后按视频顺序辨认的第9次击球。"
+     },
+     {
+      "t": 8.6,
+      "log_step": null,
+      "text_zh": "推断：以视频 8.60 s 附近的离脚上升作为第9次有效颠球达到高度条件的近似时刻，误差约0.1 s；评分器 reward.stages_reached=10、metric_touches=10、success=1 支持该次最终计入。25帧每秒画面不能独立精测接触历时或0.95倍离球速度门槛。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频 8.44–8.76 s，机器人前伸的左脚向上击打足球，球随后离脚上升；候选接触名为“击打”，现有方式没有击打项，推/扫或托/舀不能准确描述这次短促踢击。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 8.44–8.76 s，机器人右脚接触草地并承重，支撑左脚颠球时的身体。"
+     }
+    ]
+   },
+   "juggles_counted_10": {
+    "method": "visual",
+    "intervals": [
+     [
+      9.24,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 9.24,
+      "log_step": null,
+      "text_zh": "观察：视频 9.16–9.20 s，足球下落至机器人左脚（画面右侧前伸脚）后转为上升；至视频 9.24 s，球已离脚，机器人右���仍支撑草地。这是放球后按视频顺序辨认的第10次击球。"
+     },
+     {
+      "t": 9.24,
+      "log_step": null,
+      "text_zh": "推断：视频 9.16–9.24 s 可见最后一次左脚击球及离脚上升，末帧仍在上升；结合评分器 reward.stages_reached=10、metric_touches=10、replay_success=1，确认第10次最终计入。视频 9.24 s 是末段近似导航锚点，精确达到高度及速度条件的时刻在末帧附近，误差约0.1 s，不能由25帧每秒录像单独解析。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频 9.16–9.24 s，机器人前伸的左脚向上击打足球，球随后离脚上升；候选接触名为“击打”，现有方式没有击打项，推/扫或托/舀不能准确描述这次短促踢击。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 9.16–9.24 s，机器人右脚接触草地并承重，支撑左脚颠球时的身体。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "touches",
+   "unit": "count",
+   "threshold": 10,
+   "direction": "higher_better",
+   "value": 10.0
+  },
+  {
+   "name": "ball_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": 0.508
+  }
+ ],
+ "failure": null,
+ "summary_zh": "观察：机器人先将足球保持在双掌之间并准备单脚支撑，随后放球，用左脚连续完成十次可见的短促颠球，右脚持续支撑；视频 9.24 s 末帧足球正从最后一次击球后上升，未见跌倒或球落地，评分器记录有效颠球10次、成功且确定性回放通过。推断：十个事件节点均已完成，时间按逐帧画面给近似锚点；本次没有状态日志，25帧每秒录像无法精确验证毫秒级接触、离球速度和最后一次高度门槛的瞬时值，这些验收由评分器成功及计数补足，且没有独立于本次运行的参考视频。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "十个事件节点均达成，与视频和评分器（stages_reached=10、touches=10、fell=0）一致；时间为近似锚点，contact_actual（双臂协作其他、腿/躯干击打、支撑）合理，failure 为空正确。",
+    "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": "2026-10-31T00:46:20Z",
+  "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-juggling-feet-i00-privileged--zyxoa98/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-juggling-feet-i00-privileged--zyxoa98/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "6b81470f118f4fdfa8876d4a47084cfd8b6cec9ecca7e5944a98c94ec4bc2e44",
+   "video_sha256": "71ef3c97e05ef20d7bbbb4fac5b078b28d5a2828bef415e187cba0571d5f64e5",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/juggling_feet/trajectories/embodied-olympics-fun-juggling-feet-i00-standard--dwfhh6g.json

@@ -0,0 +1,311 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-juggling-feet-i00-standard--dwfhh6g",
+ "taskKey": "embodied-olympics-fun/juggling_feet",
+ "dag_sha256": "6fb5f868dd0ec1e7ac4f8d9035cd8456789adc45e6c5ceda5fec1ad9d3ba10f6",
+ "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": 263,
+   "sim_steps": 3129,
+   "calls": 44,
+   "rows": 3129,
+   "resets": 22,
+   "reset_budget": 50,
+   "sim_startups": 1,
+   "elapsed_s": 3282.6,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 1.238,
+   "peak_processes": 13,
+   "peak_disk_gb": 0.481,
+   "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-juggling-feet-i00-standard--dwfhh6g/replay.mp4",
+  "duration_s": 2.64,
+  "fps": 25.0,
+  "frames": 66,
+  "sha256": "8bc1d4ebbf66147bf82966eee59d3d49f530ff4885082cb0dbc51d7bf471b05a"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 10,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "juggles_counted_1": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频0.00 s机器人双掌挤压夹持足球，双脚站在草地上；视频0.48–2.20 s机器人屈膝并抬起一脚作准备，足球仍随双手运动。"
+     },
+     {
+      "t": 2.6,
+      "log_step": 181,
+      "text_zh": "观察：视频2.24–2.60 s机器人逐渐失去站立平衡，躯干倾斜、身体下沉，足球一直位于双掌之间，没有放落至脚部或离脚上升。log step 181最后一次实时提交返回done=true、outcome=fell、sim_time_s=2.63；与本段视频末尾失衡相符，第1次有效颠球未发生。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频0.00–2.60 s两只无可动手指的平掌从两侧挤压夹持足球并随身体调整移动；候选接触名“双掌挤压夹持”，没有单手包围或夹爪闭合，故不用抓握。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频0.00–2.60 s双脚先共同承重，随后机器人抬起一脚，另一脚接触草地支撑身体；准备阶段最终失衡，没有观察到脚与足球接触。"
+     }
+    ]
+   },
+   "juggles_counted_2": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.6,
+      "log_step": null,
+      "text_zh": "观察：检查本次视频0.00–2.60 s，足球始终在机器人双掌之间，未出现首次放球及脚部击球；视频2.24–2.60 s机器人失衡而结束，未进入第2次颠球。推断：第1次颠球未完成且没有任何后续颠球，第2次计数节点未达到。"
+     }
+    ]
+   },
+   "juggles_counted_3": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.6,
+      "log_step": null,
+      "text_zh": "观察：检查本次视频0.00–2.60 s，足球始终在机器人双掌之间，未出现首次放球及脚部击球；视频2.24–2.60 s机器人失衡而结束，未进入第3次颠球。推断：第1次颠球未完成且没有任何后续颠球，第3次计数节点未达到。"
+     }
+    ]
+   },
+   "juggles_counted_4": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.6,
+      "log_step": null,
+      "text_zh": "观察：检查本次视频0.00–2.60 s，足球始终在机器人双掌之间，未出现首次放球及脚部击球；视频2.24–2.60 s机器人失衡而结束，未进入第4次颠球。推断：第1次颠球未完成且没有任何后续颠球，第4次计数节点未达到。"
+     }
+    ]
+   },
+   "juggles_counted_5": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.6,
+      "log_step": null,
+      "text_zh": "观察：检查本次视频0.00–2.60 s，足球始终在机器人双掌之间，未出现首次放球及脚部击球；视频2.24–2.60 s机器人失衡而结束，未进入第5次颠球。推断：第1次颠球未完成且没有任何后续颠球，第5次计数节点未达到。"
+     }
+    ]
+   },
+   "juggles_counted_6": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.6,
+      "log_step": null,
+      "text_zh": "观察：检查本次视频0.00–2.60 s，足球始终在机器人双掌之间，未出现首次放球及脚部击球；视频2.24–2.60 s机器人失衡而结束，未进入第6次颠球。推断：第1次颠球未完成且没有任何后续颠球，第6次计数节点未达到。"
+     }
+    ]
+   },
+   "juggles_counted_7": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.6,
+      "log_step": null,
+      "text_zh": "观察：检查本次视频0.00–2.60 s，足球始终在机器人双掌之间，未出现首次放球及脚部击球；视频2.24–2.60 s机器人失衡而结束，未进入第7次颠球。推断：第1次颠球未完成且没有任何后续颠球，第7次计数节点未达到。"
+     }
+    ]
+   },
+   "juggles_counted_8": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.6,
+      "log_step": null,
+      "text_zh": "观察：检查本次视频0.00–2.60 s，足球始终在机器人双掌之间，未出现首次放球及脚部击球；视频2.24–2.60 s机器人失衡而结束，未进入第8次颠球。推断：第1次颠球未完成且没有任何后续颠球，第8次计数节点未达到。"
+     }
+    ]
+   },
+   "juggles_counted_9": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.6,
+      "log_step": null,
+      "text_zh": "观察：检查本次视频0.00–2.60 s，足球始终在机器人双掌之间，未出现首次放球及脚部击球；视频2.24–2.60 s机器人失衡而结束，未进入第9次颠球。推断：第1次颠球未完成且没有任何后续颠球，第9次计数节点未达到。"
+     }
+    ]
+   },
+   "juggles_counted_10": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.6,
+      "log_step": null,
+      "text_zh": "观察：检查本次视频0.00–2.60 s，足球始终在机器人双掌之间，未出现首次放球及脚部击球；视频2.24–2.60 s机器人失衡而结束，未进入第10次颠球。推断：第1次颠球未完成且没有任何后续颠球，第10次计数节点未达到。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "touches",
+   "unit": "count",
+   "threshold": 10,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "ball_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "juggles_counted_1",
+   "t": 2.4,
+   "log_step": 181,
+   "where": "操作中",
+   "evidence_zh": "观察：视频0.48–2.20 s机器人持球屈膝、移重并抬脚，为首次颠球调整支撑；视频2.24–2.60 s身体逐渐倾斜并下沉，球仍在双掌间，未放落、未踢球，也未恢复站稳。log step 181最后一次实时提交由机器人服务返回outcome=fell；之后log step 182–185只有说明及文件操作，没有新回合。推断：失败发生在第1次颠球的准备动作中，此步未被修复。"
+  },
+  "unmet": {
+   "juggles_counted_1": false,
+   "juggles_counted_2": false,
+   "juggles_counted_3": false,
+   "juggles_counted_4": false,
+   "juggles_counted_5": false,
+   "juggles_counted_6": false,
+   "juggles_counted_7": false,
+   "juggles_counted_8": false,
+   "juggles_counted_9": false,
+   "juggles_counted_10": false
+  },
+  "phenomena": [
+   {
+    "cat": "全身",
+    "item": "平衡",
+    "t": 2.4,
+    "log_step": 181,
+    "evidence_zh": "观察：视频2.16–2.60 s机器人一脚抬起，身体持续倾斜、下沉，支撑姿态不能维持；足球仍在手边而没有完成脚部击打。log step 181机器人服务对最后一次实时回合返回done=true、reason=failed、outcome=fell。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "evidence_zh": "推断：log step 124的控制脚本先按预定骨盆与踝部路径计算整段关节目标，再批量提交lowcmd；log step 179、181由该脚本派生并运行最终att5，最后一次提交在放球前即返回fell。log step 182的说明中agent自述没有尝试姿态反馈修正回路。结合视频2.24–2.60 s抬脚准备时的持续失稳，说明当前关节轨迹执行和支撑控制未能维持动态平衡；证据不足以进一步判定为逆解、承载布局或某个增益参数错误。",
+    "source_zh": "日志"
+   }
+  ],
+  "gap_zh": "观察：inputs/有本次视频、完整run.json及成功参考视频，但没有状态日志；reward没有报告touches或ball_height_m，故两项原生指标保留null。视频末帧为2.60 s，显示失衡过程而未完整展示倒地后的姿态，终止判定由log step 181机器人服务的fell结果补充。推断：无本次连续姿态、接触力和重心数据，不能把控制根因进一步细分；早期测试的逆解错误或agent关于向后跌倒的解释不直接当作本次的已证实原因。"
+ },
+ "summary_zh": "观察：本次视频0.00–2.60 s中，机器人持球屈膝移重并抬脚准备，足球始终在双掌间，十次有效颠球均未完成，最终进度为0/10。观察：视频2.24–2.60 s机器人失衡，log step 181最后一次实时回合返回fell；推断：断点为第1次颠球的准备动作，当前控制未能维持支撑平衡。观察：没有状态日志或裁判颠球计数、球高指标，控制的具体失稳机制无法进一步确定。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "视频0–2.6 s球始终在双掌间、机器人抬脚后失衡，十个节点均未达成，断点与平衡现象、控制根因均有证据支持。",
+    "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": "2026-10-31T00:57:23Z",
+  "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-juggling-feet-i00-standard--dwfhh6g/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-juggling-feet-i00-standard--dwfhh6g/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "68a7baada2268b27b6ab33ab385fd2f4d578b10387a2a0619a0b5295867f2ac4",
+   "video_sha256": "8bc1d4ebbf66147bf82966eee59d3d49f530ff4885082cb0dbc51d7bf471b05a",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/juggling_feet/trajectories/embodied-olympics-fun-juggling-feet-i00-standard--nkero7w.json

@@ -0,0 +1,342 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-juggling-feet-i00-standard--nkero7w",
+ "taskKey": "embodied-olympics-fun/juggling_feet",
+ "dag_sha256": "6fb5f868dd0ec1e7ac4f8d9035cd8456789adc45e6c5ceda5fec1ad9d3ba10f6",
+ "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": 748,
+   "sim_steps": 32852,
+   "calls": 15174,
+   "rows": 32852,
+   "resets": 50,
+   "reset_budget": 50,
+   "sim_startups": 1,
+   "elapsed_s": 3770.1,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 1.262,
+   "peak_processes": 10,
+   "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-juggling-feet-i00-standard--nkero7w/replay.mp4",
+  "duration_s": 7.52,
+  "fps": 25.0,
+  "frames": 188,
+  "sha256": "c98fc1edcdedc59553058e41a2b3b7424234210fdecd9b0702e2851e780aa4f2"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 10,
+  "base": 0,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "juggles_counted_1": {
+    "method": "visual",
+    "intervals": [
+     [
+      6.04,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 6.04,
+      "log_step": 124,
+      "text_zh": "观察：视频5.56–5.88 s机器人分开双掌，足球落向前置右脚；视频5.92 s球在右脚前端附近转为上升，视频6.04–6.28 s已离脚升至膝部以上。log step 124最终回合的服务返回stages=2，支持本次击球被计入；完成锚点取视频6.04 s，约有0.1 s误差。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频0.00–5.52 s两块平掌垫从两侧挤压足球，视频5.56–5.64 s分开使球下落；候选接触名为“双掌挤压夹持”，两只平掌各自没有包围或夹爪式合拢物体，故不标抓握。此为首次颠球的准备接触。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频5.92 s前置右脚前端将下落足球击起；候选接触名为“击打”，现有接触方式没有对应项。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频0.00–6.28 s机器人以双脚接触草地承重，准备时挪动右脚形成前后站姿，击球时左脚维持支撑。"
+     }
+    ]
+   },
+   "juggles_counted_2": {
+    "method": "visual",
+    "intervals": [
+     [
+      6.8,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 6.8,
+      "log_step": 124,
+      "text_zh": "观察：视频6.32–6.64 s足球回落，视频6.68 s在右脚前端附近再次改变运动方向，视频6.72–7.04 s离脚向上并向前飞；视频6.80 s已明显升过膝部。log step 124最终回合服务返回stages=2，与两次上升一致；完成锚点取视频6.80 s，约有0.1 s误差。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频6.68 s机器人右脚前端再次击打回落足球；候选接触名为“击打”，无对应的既有接触方式。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频6.32–7.04 s机器人保持前后站姿，双脚接触草地承重，左脚在右脚上翘击球时维持支撑。"
+     }
+    ]
+   },
+   "juggles_counted_3": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 7.36,
+      "log_step": 124,
+      "text_zh": "观察：视频6.80–7.48 s第二次击出的足球持续向机器人前方偏移，视频7.24–7.48 s在双脚前方落向草地；机器人保持原站位，右脚虽有上翘动作，未出现第三次球脚接触或新的离脚上升。log step 124最终回合服务返回stages=2及“the ball touched the ground”，log step 125确认该回合episode=51已失败。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频7.04–7.48 s机器人双脚继续接触草地承重，右脚上翘但未跟到前移足球下方；没有观察到本次续颠所需的球脚接触，因此不记第三次击打接触。"
+     }
+    ]
+   },
+   "juggles_counted_4": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 7.48,
+      "log_step": 124,
+      "text_zh": "观察：视频5.56–7.48 s仅见两次脚部击球后的上升，随后球飞离脚前并落地；log step 124最终回合服务返回stages=2、球触地结束，未发生第4次有效颠球。"
+     }
+    ]
+   },
+   "juggles_counted_5": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 7.48,
+      "log_step": 124,
+      "text_zh": "观察：视频5.56–7.48 s仅见两次脚部击球后的上升，随后球飞离脚前并落地；log step 124最终回合服务返回stages=2、球触地结束，未发生第5次有效颠球。"
+     }
+    ]
+   },
+   "juggles_counted_6": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 7.48,
+      "log_step": 124,
+      "text_zh": "观察：视频5.56–7.48 s仅见两次脚部击球后的上升，随后球飞离脚前并落地；log step 124最终回合服务返回stages=2、球触地结束，未发生第6次有效颠球。"
+     }
+    ]
+   },
+   "juggles_counted_7": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 7.48,
+      "log_step": 124,
+      "text_zh": "观察：视频5.56–7.48 s仅见两次脚部击球后的上升，随后球飞离脚前并落地；log step 124最终回合服务返回stages=2、球触地结束，未发生第7次有效颠球。"
+     }
+    ]
+   },
+   "juggles_counted_8": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 7.48,
+      "log_step": 124,
+      "text_zh": "观察：视频5.56–7.48 s仅见两次脚部击球后的上升，随后球飞离脚前并落地；log step 124最终回合服务返回stages=2、球触地结束，未发生第8次有效颠球。"
+     }
+    ]
+   },
+   "juggles_counted_9": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 7.48,
+      "log_step": 124,
+      "text_zh": "观察：视频5.56–7.48 s仅见两次脚部击球后的上升，随后球飞离脚前并落地；log step 124最终回合服务返回stages=2、球触地结束，未发生第9次有效颠球。"
+     }
+    ]
+   },
+   "juggles_counted_10": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 7.48,
+      "log_step": 124,
+      "text_zh": "观察：视频5.56–7.48 s仅见两次脚部击球后的上升，随后球飞离脚前并落地；log step 124最终回合服务返回stages=2、球触地结束，未发生第10次有效颠球。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "touches",
+   "unit": "count",
+   "threshold": 10,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "ball_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "juggles_counted_3",
+   "t": 7.36,
+   "log_step": 124,
+   "where": "操作中",
+   "evidence_zh": "观察：视频6.68 s第二次击球后，足球向前飞离；视频7.24–7.48 s机器人没有移脚追到落点下方，第三次续颠未接触球，球落向草地。log step 124明确执行最终配置，服务返回stages=2、球触地；log step 125确认episode=51结束。断点为正在续接的第3次颠球，前两次已完成。"
+  },
+  "unmet": {
+   "juggles_counted_1": true,
+   "juggles_counted_2": true,
+   "juggles_counted_3": false,
+   "juggles_counted_4": false,
+   "juggles_counted_5": false,
+   "juggles_counted_6": false,
+   "juggles_counted_7": false,
+   "juggles_counted_8": false,
+   "juggles_counted_9": false,
+   "juggles_counted_10": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "动态时机",
+    "t": 7.36,
+    "log_step": 124,
+    "evidence_zh": "观察：视频6.80–7.48 s足球上升并向前飞出后回落，机器人仍保持前后站姿，右脚的上翘没有跟上移动中的足球，球从脚前方落下，未形成第三次触球；log step 124最终回合以球触地结束。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "t": 7.36,
+    "log_step": 124,
+    "evidence_zh": "推断：视频6.80–7.48 s第二次触球后的前向偏移没有被脚部移动补偿，是续颠控制失败。log step 71定义按预设时刻上翘右踝的击球函数，log step 97、102将其接入根据本体姿态和关节读数修正重心的循环；log step 124最终配置仍使用三组预设踝部上翘时刻并返回仅2次有效计数。日志与视频共同支持控制未能把后续击球位置接到球路上，但不能唯一确定是接触几何、轨迹跟踪还是具体增益所致，故保留“控制”总项。",
+    "source_zh": "日志 / 视频"
+   }
+  ],
+  "gap_zh": "没有本次状态日志或逐帧裁判计数；25 fps视频不能独立核验0.05秒接触簇、精确速度及高度阈值，前两次有效性由最终回合服务stages=2与视频共同支持，节点时间为视觉近似。未观察到足够证据验证agent在log step 126所称胫部碰撞几何及毫秒级敏感性，不据此细分根因。reward字段未报告touches及ball_height_m，两个native_metrics值置null；服务最终回合的stages=2另作节点证据。"
+ },
+ "summary_zh": "评分视频显示机器人分开双掌放球，并以右脚完成两次有效颠球，分别在视频约6.04 s、6.80 s达到上升判据，最终为2/10。第二次击球后足球向前偏离，机器人未移脚跟随，第三次颠球在视频约7.36 s失接，随后球触地；最终回合log step 124的服务结果与视频一致。无状态日志，毫秒接触、精确阈值及控制子类无法独立确定。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "视频帧显示两次击球后球上升、第二次后前飘落地，与stages=2、第3次断点及动态时机/控制归因一致；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": "2026-10-31T00:55:13Z",
+  "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-juggling-feet-i00-standard--nkero7w/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-juggling-feet-i00-standard--nkero7w/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "c53c105a52a7786421b9084ab44a059f8bc41526fa6d00e59e5ad1050cad642d",
+   "video_sha256": "c98fc1edcdedc59553058e41a2b3b7424234210fdecd9b0702e2851e780aa4f2",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}