Add 521 Claude Fable 5.2 trajectories (phaseF, batch F01,F02,F05,F06,F07; writer GPT-6.1 Sol)

phaseF/ambench/cabinet_pick_place/trajectories/ambench-cabinet-pick-place-fahexa-i00-privileged--kjz3mdi.json

@@ -0,0 +1,220 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ambench-cabinet-pick-place-fahexa-i00-privileged--kjz3mdi",
+ "taskKey": "ambench/cabinet_pick_place",
+ "dag_sha256": "cc292a4bde462daa148ce8a9af3a7752d9d5f14850434c7531da2c92498b8c56",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "all_success": 1,
+   "deterministic": 1,
+   "state_identical": 1,
+   "determinism_checked": 0,
+   "instance_ok": 1,
+   "n_actions": 2364,
+   "replays": 1,
+   "replays_max": 3,
+   "replays_succeeded": 1,
+   "success_replay": 1,
+   "sim_steps": 2015,
+   "sim_time_s": 16.7917,
+   "timed_out": 0,
+   "replay_wall_s": 86.9,
+   "replay_timeout_s": 2091.0,
+   "grader_error": 0,
+   "infra_error": 0,
+   "replay_retries": 0,
+   "replay_gpu_fault": 0,
+   "agent_sessions": 2,
+   "agent_resets": 3,
+   "agent_sim_steps": 0,
+   "agent_records": 0,
+   "agent_replays": 0,
+   "agent_cost_self_reported": 1,
+   "video_rendered": 1,
+   "video_episode_success": 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,
+   "peak_cpus": 0.0,
+   "peak_pids": 0.0,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "media",
+  "url": "media:ambench-cabinet-pick-place-fahexa-i00-privileged--kjz3mdi.mp4",
+  "duration_s": 16.834,
+  "fps": 30.0,
+  "frames": 505,
+  "sha256": "177e86eaa823349408d4db9bbefa91bf1fab43f7ff2342fdcc04fa9ad4c28da0"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "sliding_door_opened": {
+    "method": "visual",
+    "intervals": [
+     [
+      6.6,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频起始时浅木色滑门遮住左半格，红色汽水罐不可见。"
+     },
+     {
+      "t": 6.6,
+      "log_step": null,
+      "text_zh": "观察：视频5.4—7.4 s，机器人的闭合夹爪抵着滑门竖直把手侧方横移，门向机器人右侧滑动；视频6.6 s两视图已显露红罐及其前方格口，随后视频8—11 s夹爪由此开口伸入。推断：取罐通路在视频约6.6 s已让开，视觉时刻误差约0.3 s。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "推/扫",
+      "note_zh": "观察：机器人用单臂闭合二指夹爪抵住滑门竖直把手侧部，横向推移把手和门板；未见两指张开后对夹把手。"
+     }
+    ]
+   },
+   "can_at_cabinet_top": {
+    "method": "visual",
+    "intervals": [
+     [
+      16.8,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 12.4,
+      "log_step": null,
+      "text_zh": "观察：视频11.6 s红罐仍在格内，二指夹爪张开；视频12.0—12.4 s两指向罐身合拢。视频13.2—14.4 s红罐随夹爪后撤，原先放罐的格内位置露出，罐始终留在两指之间。"
+     },
+     {
+      "t": 16.8,
+      "log_step": null,
+      "text_zh": "观察：视频15.9—16.8 s机器人持罐上升，机身视图中罐升至上层面板上方，末帧仍夹在柜前；末端视图中罐保持在两指之间，没有落下或松开。"
+     },
+     {
+      "t": 16.8,
+      "log_step": 20,
+      "text_zh": "观察：评分器reward.success=1、success_replay=1、video_episode_success=1，sim_steps=2015；log step 20对已保存交付轨迹的新进程回放也报告success=true、control_steps=2015、sim_time_s=16.7917。推断：结合视频16.8 s持罐到柜顶高度的末帧，目标的高度带及线速度联合条件在终止时满足，将首次达成锚定为视频约16.8 s；此前仅升高不提前记为达成，画面不能直接测得原生高度和速度。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "观察：单臂二指夹爪从两侧夹住红色汽水罐罐身，后撤取出并持续夹持，随机体升高至柜顶高度，直到视频16.8 s末帧。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "can_height_above_drawer",
+   "unit": "m",
+   "threshold": {
+    "gte": 0,
+    "lt": 0.05
+   },
+   "direction": "target",
+   "value": null
+  },
+  {
+   "name": "can_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.1,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人以闭合夹爪侧推滑门，视��约6.6 s取罐通路打开，随后夹住红罐、拉出格口并升高，视频约16.8 s按评分器成功结果完成目标，终态为2/2。任务设计问题：目标称为到柜顶，但验收只检查罐心高度带及线速度，不检查柜顶水平范围、顶板支撑或松手；本次末帧仍在柜前持罐，与成功参考的终态相符。没有状态日志，原生高度和速度数值未在评分字段报告，两项指标记null，精确数值达标依据成功验收推断。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "滑门约6.6 s打开、末帧持罐升至柜顶高度、成功与评分器一致，节点、接触与任务设计说明均有据；无需修改。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-11-12T11:23:04Z",
+  "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/ambench-cabinet-pick-place-fahexa-i00-privileged--kjz3mdi/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ambench-cabinet-pick-place-fahexa-i00-privileged--kjz3mdi/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "233bdc8a776b97dc77bb9e2c48e02de4cd7a778087702033e458f4ca0a46cf46",
+   "video_sha256": "177e86eaa823349408d4db9bbefa91bf1fab43f7ff2342fdcc04fa9ad4c28da0",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/ambench/cabinet_pick_place/trajectories/ambench-cabinet-pick-place-fahexa-i00-standard--jpsxagy.json

@@ -0,0 +1,228 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ambench-cabinet-pick-place-fahexa-i00-standard--jpsxagy",
+ "taskKey": "ambench/cabinet_pick_place",
+ "dag_sha256": "cc292a4bde462daa148ce8a9af3a7752d9d5f14850434c7531da2c92498b8c56",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "live_success": 1,
+   "graded_by_live": 1,
+   "replays": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "n_actions": 3273,
+   "sim_steps": 3273,
+   "physics_steps": 3273,
+   "steps": 3273,
+   "elapsed_s": 449.1,
+   "episode_timeout": 0,
+   "time_up": 0,
+   "live_gpu_faults": 0,
+   "hold_pad_rows": 0,
+   "determinism_verdict_mode": 0,
+   "video_rendered": 1,
+   "video_matches_live": 1,
+   "video_outcome_matches_live": 1,
+   "sim_error": 0,
+   "instance_mismatch": 0,
+   "instance_exact": 1,
+   "grader_error": 0,
+   "infra_error": 0,
+   "gave_up": 0,
+   "sim_crashed_on_call": 0,
+   "sim_startups": 1,
+   "resumes": 0,
+   "resume_state_mismatch": 0,
+   "resume_max_tcp_dev_m": 0.0,
+   "resume_max_joint_dev": 0.0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 0.337,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.425,
+   "peak_cpus": 1.0,
+   "peak_pids": 24,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "media",
+  "url": "media:ambench-cabinet-pick-place-fahexa-i00-standard--jpsxagy.mp4",
+  "duration_s": 27.334,
+  "fps": 30.0,
+  "frames": 820,
+  "sha256": "7182bb5c3fb4e822e035a2ecdce9fa0e99cdf04212cbacc728659d2cd5b2338e"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "sliding_door_opened": {
+    "method": "visual",
+    "intervals": [
+     [
+      10.0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s，机器人前方柜体的左半格被浅木色滑门遮住，红罐不可见。"
+     },
+     {
+      "t": 10.0,
+      "log_step": 14,
+      "text_zh": "观察：视频 6.7–11.3 s，机器人的张开夹爪抵着竖直把手侧向移动，滑门向机器人右侧滑移并与右面板重叠；视频 9.7 s 红罐开始显露，视频 10.0 s 罐与门缘之间已有供夹爪接近的空隙。"
+     },
+     {
+      "t": 16.0,
+      "log_step": 20,
+      "text_zh": "观察：视频 14.0–18.5 s，机器人末端相机清楚显示敞开的格口及红罐，张开的两指随后进入格口接近罐身；视频 22.0–23.5 s 罐确实被取出。推断：取罐通路首次达到约在视频 10.0 s，视觉估时误差约 0.3 s。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "推/扫",
+      "note_zh": "观察：视频 6.7–11.3 s，空中机械臂的二指夹爪保持张开，以指侧抵住滑门竖直把手并横向推门；未闭合夹紧把手。"
+     }
+    ]
+   },
+   "can_at_cabinet_top": {
+    "method": "visual",
+    "intervals": [
+     [
+      27.3,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 19.5,
+      "log_step": 23,
+      "text_zh": "观察：视频 18.5 s 红罐仍在格内，视频 19.0–19.5 s 机器人两指从罐身两侧闭合；视频 21.0–23.5 s 罐随夹爪抬离原支承并退出格口，机身视角可见罐悬在闭合夹爪下方，原放罐位置空出。"
+     },
+     {
+      "t": 27.3,
+      "log_step": 24,
+      "text_zh": "观察：视频 24.0–27.3 s，机器人持续夹持红罐向上运动；已每隔约 0.1 s 抽查视频 26.3–27.3 s，末帧机身相机可见罐位于上层面板及柜顶边缘上方，末端相机可见罐身仍夹在两指之间，未释放。"
+     },
+     {
+      "t": 27.3,
+      "log_step": 25,
+      "text_zh": "观察：log step 25 中机器人服务返回 control_steps=3273、done=True、reason=success、resumes=0；run.json 的评分器字段 live_success=1、video_matches_live=1。推断：按 task_text 及 log step 7 的 120 Hz，控制步 3273 对应约视频 27.28 s，与末帧完成抬升核对一致；因此将高度 1.67≤z<1.72 米且线速度<0.1 米/秒的首次同时达标记为约视频 27.3 s。无状态日志，不能从图像单独测定这些精确数值。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "观察：视频 19.0–19.5 s，单臂二指夹爪从相对两侧夹住红色汽水罐罐身；视频 21.0–27.3 s 持续夹持并将罐取出、升高，末���仍未松手。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "can_height_above_drawer",
+   "unit": "m",
+   "threshold": {
+    "gte": 0,
+    "lt": 0.05
+   },
+   "direction": "target",
+   "value": null
+  },
+  {
+   "name": "can_linear_speed",
+   "unit": "m/s",
+   "threshold": 0.1,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人约在视频 10.0 s 推开滑门，随后夹住红罐取出，并在视频 27.3 s 仍持罐升至柜顶高度；服务终止原因为 success，两节点最终均完成。任务设计问题：任务称将罐移到柜顶，但验收只检查罐心高度带与线速度，不检查水平范围、顶板支撑或释放，本次成功时罐仍夹在柜前。无状态日志且评分器未报告两项原生指标数值，目标精确达标时刻依据实况成功终止与末帧估定。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "两节点时间与接触标注与视频一致：约 10 s 红罐显露，末帧罐被夹持升至柜顶上方；成功、无失败层，任务设计问题说明合理。",
+    "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": "2026-11-12T11:20:46Z",
+  "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/ambench-cabinet-pick-place-fahexa-i00-standard--jpsxagy/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ambench-cabinet-pick-place-fahexa-i00-standard--jpsxagy/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "66d5d8b6f5866fed6c0b2135782bc7a220b771297d278eaa0d5ed72d0ab08642",
+   "video_sha256": "7182bb5c3fb4e822e035a2ecdce9fa0e99cdf04212cbacc728659d2cd5b2338e",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/ambench/peg_in_hole/trajectories/ambench-peg-in-hole-fahexa-i00-privileged--ulwjsbn.json

@@ -0,0 +1,178 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ambench-peg-in-hole-fahexa-i00-privileged--ulwjsbn",
+ "taskKey": "ambench/peg_in_hole",
+ "dag_sha256": "4ce14549cf9f097f094786c0e57e61d08ef25b8811c7d32b3ac24624a7503b3a",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "all_success": 1,
+   "deterministic": 1,
+   "state_identical": 1,
+   "determinism_checked": 0,
+   "instance_ok": 1,
+   "n_actions": 1590,
+   "replays": 1,
+   "replays_max": 3,
+   "replays_succeeded": 1,
+   "success_replay": 1,
+   "sim_steps": 1116,
+   "sim_time_s": 9.3,
+   "timed_out": 0,
+   "replay_wall_s": 68.6,
+   "replay_timeout_s": 1897.5,
+   "grader_error": 0,
+   "infra_error": 0,
+   "replay_retries": 0,
+   "replay_gpu_fault": 0,
+   "agent_sessions": 1,
+   "agent_resets": 1,
+   "agent_sim_steps": 0,
+   "agent_records": 0,
+   "agent_replays": 0,
+   "agent_cost_self_reported": 1,
+   "video_rendered": 1,
+   "video_episode_success": 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,
+   "peak_cpus": 0.0,
+   "peak_pids": 0.0,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "media",
+  "url": "media:ambench-peg-in-hole-fahexa-i00-privileged--ulwjsbn.mp4",
+  "duration_s": 9.334,
+  "fps": 30.0,
+  "frames": 280,
+  "sha256": "3897a74e7a7b26b143c7ffa6a56c79cce1d5d50e9d6e59eeb3b4a2f467de750c"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "peg_inserted": {
+    "method": "visual",
+    "intervals": [
+     [
+      9.3,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s，金属销位于机器人二指夹爪之间，墙上蓝色方孔仍在远处，插入条件尚未成立；视频 0.5–5.5 s，机器人夹持销向方孔靠近并调整朝向。"
+     },
+     {
+      "t": 9.3,
+      "log_step": null,
+      "text_zh": "观察：视频 5.5–7.5 s，机器人夹持的深色销对准蓝框内红色背板；视频 8.3–9.3 s 的每隔约 0.1 s 画面中，机器人继续向孔内推进，销与夹爪一起移动，未见松脱或回撤。视频 9.3 s 末帧中销位于方孔开口范围内，前端被销身遮挡，不能直接量测插入深度。"
+     },
+     {
+      "t": 9.3,
+      "log_step": 13,
+      "text_zh": "观察：log step 13，基准回放 CLI 对已保存并最终交付的 trajectory.npz 在新进程中回放，报告 instance_matches=true、success=true、end_reason=success、control_steps=1116、sim_time_s=9.3；本次评分器 reward 同样给出 success=1、success_replay=1、video_episode_success=1、sim_steps=1116。"
+     },
+     {
+      "t": 9.3,
+      "log_step": 13,
+      "text_zh": "推断：以交付文件回放的控制计数定位成功时刻：log step 13 的 control_steps=1116，按任务 120 Hz 得 1116/120=9.3 s，与视频 9.3 s 推进后的终止帧核对一致，故将插销入孔区间起点记为视频约 9.3 s。末帧没有直接显示未达条件，结合评分器成功判定确认深度及横向范围达标；参考视频 5.6 s 的成功终态仅用于对照可见外观，不作为本次运行达标的证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "观察：视频 0.0–9.3 s，机器人单臂末端的二指夹爪从两侧夹持深色金属销，带销靠近、对准并推进蓝色方孔，未见释放或销脱离夹爪。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [],
+ "failure": null,
+ "summary_zh": "机器人从初始夹销状态靠近墙面、对准蓝色方孔并沿孔轴推进，视频约 9.3 s 达成唯一节点“插销入孔”，终态完成 1/1。评分器及 log step 13 的交付文件新进程回放均报告成功，无失败断点。没有逐帧状态日志，末段销前端受遮挡，精确深度及达标时刻由评分结果和回放控制计数结合视频末帧推断。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "单节点插销入孔在视频约9.3 s达成，评分器成功，控制步数1116/120=9.3 s换算有证据，接触标注抓握合理，无失败断点。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-11-12T11:19:44Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/ambench-peg-in-hole-fahexa-i00-privileged--ulwjsbn/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ambench-peg-in-hole-fahexa-i00-privileged--ulwjsbn/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "23e101a088dd89459dffdd24f1f04ab4a6156a6d3a5c1c125754c1e2d34087a9",
+   "video_sha256": "3897a74e7a7b26b143c7ffa6a56c79cce1d5d50e9d6e59eeb3b4a2f467de750c",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/ambench/peg_in_hole/trajectories/ambench-peg-in-hole-fahexa-i00-standard--5hhn3ug.json

@@ -0,0 +1,181 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "ambench-peg-in-hole-fahexa-i00-standard--5hhn3ug",
+ "taskKey": "ambench/peg_in_hole",
+ "dag_sha256": "4ce14549cf9f097f094786c0e57e61d08ef25b8811c7d32b3ac24624a7503b3a",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "live_success": 1,
+   "graded_by_live": 1,
+   "replays": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "n_actions": 1149,
+   "sim_steps": 1149,
+   "physics_steps": 1149,
+   "steps": 1143,
+   "elapsed_s": 373.4,
+   "episode_timeout": 0,
+   "time_up": 0,
+   "live_gpu_faults": 0,
+   "hold_pad_rows": 0,
+   "determinism_verdict_mode": 0,
+   "video_rendered": 1,
+   "video_matches_live": 1,
+   "video_outcome_matches_live": 1,
+   "sim_error": 0,
+   "instance_mismatch": 0,
+   "instance_exact": 1,
+   "grader_error": 0,
+   "infra_error": 0,
+   "gave_up": 0,
+   "sim_crashed_on_call": 0,
+   "sim_startups": 1,
+   "resumes": 0,
+   "resume_state_mismatch": 0,
+   "resume_max_tcp_dev_m": 0.0,
+   "resume_max_joint_dev": 0.0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 0.217,
+   "peak_processes": 10,
+   "peak_disk_gb": 0.261,
+   "peak_cpus": 1.0,
+   "peak_pids": 31,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "media",
+  "url": "media:ambench-peg-in-hole-fahexa-i00-standard--5hhn3ug.mp4",
+  "duration_s": 9.634,
+  "fps": 30.0,
+  "frames": 289,
+  "sha256": "05f0e159e41356b8ba3e8c3dc04fe4e0e17c0d1e900603436a2ad7a199c6b101"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "peg_inserted": {
+    "method": "visual",
+    "intervals": [
+     [
+      9.6,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.1,
+      "log_step": 13,
+      "text_zh": "观察：视频 0.000–0.333 s 的末端视角中，深色金属销位于两根白色夹指之间，夹指短暂张开后重新闭合；之后销随夹爪一起运动。视频起点的销与墙上蓝色方孔相距明显，插入条件尚未成立。"
+     },
+     {
+      "t": 8.5,
+      "log_step": 21,
+      "text_zh": "观察：视频 0.5–7.3 s 机器人持销向墙靠近并调整方向；视频 7.3 s、8.5 s 的双视角中，销已朝向蓝框内红色背板，视频 8.600–9.600 s 逐帧画面显示末端继续向前推进，销仍在夹爪之间，未见掉落或退出。销前端被自身和孔口遮挡，二维重叠本身不能确认插入深度。"
+     },
+     {
+      "t": 9.6,
+      "log_step": 23,
+      "text_zh": "观察：log step 23 中实时机器人服务的 status 返回 control_steps=1149、done=True、reason=success、resumes=0；log step 22 的服务响应同样报告成功。inputs/run.json 的评分器字段为 success=1、live_success=1、graded_by_live=1、video_matches_live=1、video_outcome_matches_live=1，说明评分与本段实时回合视频一致。"
+     },
+     {
+      "t": 9.6,
+      "log_step": 22,
+      "text_zh": "推断：本回合控制频率为 120 Hz，log step 22 的成功控制步 1149 对应仿真 1149/120=9.575 s，落在视频约 9.6 s 的结束处；已核对 log step 20 的控制步 876/120=7.3 s、log step 21 的控制步 1020/120=8.5 s 对应视频中的靠近及继续推进阶段，并检查视频 9.567 s、9.600 s 的终帧。依据实时服务在首次成功时终止及评分器成功结果，将节点首次达成锚定在视频 9.6 s（约一帧精度），终态为真；严格深度阈值由服务验收确认，无法从遮挡画面独立量出。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "观察：视频 0.10–9.60 s，机器人机械臂末端的二指夹爪重新闭合后从两侧夹住深色金属销，持销靠近墙面并向蓝色方孔推进；飞行基座悬空移动，未见落地支承接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [],
+ "failure": null,
+ "summary_zh": "机器人重新闭合夹爪稳住销后，持销靠近蓝色方孔、调整方向并继续推进，在视频约 9.6 s 完成唯一的插销入孔节点，终态为 1/1，实时服务及评分器均判成功。末段销前端与孔口相互遮挡，且没有状态日志，严格超过 2 厘米的深度不能由画面独立测量；该条件依据同一实时回合的服务成功结果确认，时间以控制步计数并结合视频终帧核对。",
+ "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 peg_inserted 在视频约 9.6 s 达成，终态真，与服务 done=True/success、评分器成功及终帧所见（销在蓝框红背板内）一致；接触为抓握，时间换算有控制步映射与终帧核对，无问题。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-11-12T11:25:29Z",
+  "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/ambench-peg-in-hole-fahexa-i00-standard--5hhn3ug/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/ambench-peg-in-hole-fahexa-i00-standard--5hhn3ug/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "8b5b31bb501961efb14fbdb1fe8e19b128b1738487a7f754feaadbcde032df2a",
+   "video_sha256": "05f0e159e41356b8ba3e8c3dc04fe4e0e17c0d1e900603436a2ad7a199c6b101",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/dexjoco/bimanual_photograph/trajectories/dexjoco-bimanual-photograph-i00-privileged--iqkfhkl.json

@@ -0,0 +1,244 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexjoco-bimanual-photograph-i00-privileged--iqkfhkl",
+ "taskKey": "dexjoco/bimanual_photograph",
+ "dag_sha256": "47a64f383b78b6f8ab77576cf14aefb72daf5acdc73438c6710a560c48f0cd22",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "deterministic": 1,
+   "grader_error": 0,
+   "infra_error": 0,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "all_success": 1,
+   "determinism_checked": 0,
+   "replays": 1,
+   "instance_ok": 1,
+   "n_actions": 1809,
+   "sim_steps": 1809,
+   "task_failure": 0,
+   "replay_wall_s": 19.5,
+   "video_rendered": 1,
+   "visible_cpus": 32,
+   "openblas_num_threads_env": 4,
+   "blas_threads": 4,
+   "threads_ok": 1,
+   "agent_sessions": 18,
+   "agent_resets": 21,
+   "agent_sim_steps": 20117,
+   "agent_records": 8,
+   "agent_replays": 5,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 1,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 2.348,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.266,
+   "peak_cpus": 5.54,
+   "peak_pids": 99,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "over_cpus": 1,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "media",
+  "url": "media:dexjoco-bimanual-photograph-i00-privileged--iqkfhkl.mp4",
+  "duration_s": 36.24,
+  "fps": 25.0,
+  "frames": 906,
+  "sha256": "1e30df0445cb29d9db2b573d6e1330377e14aa1dd4a65a5f4ab7c609b4498597"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "camera_aimed": {
+    "method": "visual",
+    "intervals": [
+     [
+      25.5,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 相机斜靠在桌面上，尚未抬到标志前方；视频 3.0–7.0 s 右手手指推动机身，使相机翻到镜头朝下的姿态，随后退开。"
+     },
+     {
+      "t": 18.0,
+      "log_step": null,
+      "text_zh": "观察：视频 12.0–16.0 s 左手在相机握持端合拢手指；视频 18.0–21.5 s 相机离开原桌面位置，随左手抬高并转动，右手此时未持机。"
+     },
+     {
+      "t": 25.5,
+      "log_step": 21,
+      "text_zh": "推断：视频 20.0–28.0 s 每隔约 0.5 s 检查左手调姿，相机先明显倾斜，后在标志前方转正；首次同时满足位置与朝向约记为视频 25.5 s，边界约有 0.5 s 不确定性。log step 21 的同一 1809 步生成方案记录末次粗调后取景点约为 [-0.249,0.036,1.361] m、取景轴约为 [0.999,0.044,-0.028]，与随后可见稳定持机相符；该日志没有中间控制步编号，不把 agent 时间换算为视频时间。"
+     },
+     {
+      "t": 36.2,
+      "log_step": 22,
+      "text_zh": "观察：视频 28.0–36.2 s 左手持续持机，末秒视频 35.0、35.4、35.6、35.8、36.0、36.12、36.16、36.2 s 未见相机掉落或大幅偏转。log step 22 重新生成并保存 1809 步交付，服务末态报告 region_pass=true、angle_pass=true，评分 CLI 重放 success=true；inputs/run.json 的评分器字段 success=1、all_success=1、n_actions=1809。"
+     },
+     {
+      "t": 36.2,
+      "log_step": null,
+      "text_zh": "推断：终态成功按本任务的联合验收定义保证相机位置及朝向达标。标志和持机手遮住部分机身，不能仅凭本次画面独立测量目标区边界或 10 度阈值；成功参考视频 7.0–8.24 s 也显示标志前方持机及近景遮挡，仅作外观比较，不作为本次进度证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "视频 12.0–18.0 s 左 Allegro 手的拇指与弯曲手指夹握相机握持端，随后单手抬高、旋转并持续持机。准备阶段视频 3.0–7.0 s 为右手手指推/扫机身使其翻到镜头朝下，随后右手撤离；两手并未共同抓持相机完成调姿。"
+     }
+    ]
+   },
+   "shutter_touched": {
+    "method": "visual",
+    "intervals": [
+     [
+      36.1,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 32.0,
+      "log_step": null,
+      "text_zh": "观察：视频 28.0–32.0 s 右臂从停放位置移向左手所持相机，右手伸出一根手指；视频 34.0–36.2 s 以约 0.2 s 间隔检查近景和全景，右手继续向相机端部缓慢推进，左手保持抓握。精确快门接触面被机身、手和标志���挡。"
+     },
+     {
+      "t": 36.2,
+      "log_step": 22,
+      "text_zh": "观察：log step 22 的最终交付生成记录报告 1809 步结束时 shutter_pressed=true、shutter_contact_count=1，保存后评分 CLI 重放同一 1809 步轨迹得到 success=true；inputs/run.json 的最终评分器字段 success=1。视频 36.12–36.2 s 为该次推进结束的持机终态。"
+     },
+     {
+      "t": 36.1,
+      "log_step": 21,
+      "text_zh": "推断：视频末段右手推进、评分器联合成功和 log step 22 的最终交付状态共同确认末态快门已接触，首次接触约锚定视频 36.1 s（约 0.2 s 不确定性），而非把此前靠近视为接触。log step 21 的相同生成方案打印接触对 mf_tip_right/fingertip_mf_right 与 camera_body_top_cap，支持右中指尖触按的身份判断；这条接触对是生成方案的补充证据，视频不能独立辨认几何体。未见此前右手触按快门的可确认片段，不额外添加早期接触区间。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "抓握",
+      "note_zh": "视频 30.0–36.2 s 右手接近并触按时，左 Allegro 手以拇指和弯曲手指夹住相机机身，固定相机；左手实际抓握，并非两掌对挤。"
+     },
+     {
+      "who": "双臂协作",
+      "how": "压/擦",
+      "note_zh": "视频末段左手持机，右手伸指向快门推进并触按；终态接触由最终评分成功及 log step 22 的 shutter_pressed=true 确认，右中指尖身份由 log step 21 相同生成方案的接触对补充支持。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "radial_dist",
+   "unit": "m",
+   "threshold": 0.16,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "region_height_dist",
+   "unit": "m",
+   "threshold": 0.1,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "view_angle_deg",
+   "unit": "deg",
+   "threshold": 10,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人先用右手推转桌上的相机，再用左手抓起并调姿，约视频 25.5 s 对准标志，约视频 36.1 s 在左手持机时由右手触按快门，终态完成 2/2 节点，无失败断点。最终交付在 log step 22 保存并成功重放，评分器亦判 success=1；无状态日志，精确角度、目标区边界及首次快门接触被遮挡，区间起点为视觉与最终交付记录结合的估计，不能复原所有短暂接触；评分器 reward 未单列三个原生指标数值，native_metrics 因而留空值。",
+ "conf": 6,
+ "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，两个节点均达成，无失败断点；抽查帧与标注（左手持机对准、末段右手推进触按）一致，未发现需改动的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-11-12T11:18:29Z",
+  "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/dexjoco-bimanual-photograph-i00-privileged--iqkfhkl/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexjoco-bimanual-photograph-i00-privileged--iqkfhkl/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "31c89706b8076a4c455b2561b0b7ed2a31ce70f86391a50ddcc10401824f24e6",
+   "video_sha256": "1e30df0445cb29d9db2b573d6e1330377e14aa1dd4a65a5f4ab7c609b4498597",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/dexjoco/bimanual_photograph/trajectories/dexjoco-bimanual-photograph-i00-standard--2c4ckrv.json

@@ -0,0 +1,238 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexjoco-bimanual-photograph-i00-standard--2c4ckrv",
+ "taskKey": "dexjoco/bimanual_photograph",
+ "dag_sha256": "47a64f383b78b6f8ab77576cf14aefb72daf5acdc73438c6710a560c48f0cd22",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "live_success": 1,
+   "graded_by_live": 1,
+   "replays": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "no_verdict": 0,
+   "task_failure": 0,
+   "n_actions": 5415,
+   "sim_steps": 5415,
+   "physics_steps": 54150,
+   "steps": 1065,
+   "calls": 189,
+   "elapsed_s": 2020.3,
+   "video_rendered": 1,
+   "instance_mismatch": 0,
+   "too_few_cpus": 0,
+   "visible_cpus": 32,
+   "blas_threads": 4,
+   "gave_up": 0,
+   "sim_startups": 1,
+   "resumes": 0,
+   "resume_state_mismatch": 0,
+   "resume_max_joint_dev": 0.0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 0.406,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.416,
+   "peak_cpus": 3.04,
+   "peak_pids": 90,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "media",
+  "url": "media:dexjoco-bimanual-photograph-i00-standard--2c4ckrv.mp4",
+  "duration_s": 54.24,
+  "fps": 25.0,
+  "frames": 1356,
+  "sha256": "176069116aed56eb5621d6f36573bc379bb0816a421731e623d95bfbc3287758"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "camera_aimed": {
+    "method": "visual",
+    "intervals": [
+     [
+      39.0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 相机斜放在白色桌面上，未抬到圆形标志前的目标高度；视频 8.0–10.0 s 左手的早期抓取使相机翻倒，松手后相机仍在桌面，不能计为对准。"
+     },
+     {
+      "t": 29.6,
+      "log_step": 37,
+      "text_zh": "观察：视频 28.0–30.0 s，画面右侧的机器人左手弯指围住相机一端并将整机抬离桌面，旧位置空出；视频 31.6–40.4 s 左手继续转动和移动相机，右手尚未触机。"
+     },
+     {
+      "t": 39.0,
+      "log_step": 56,
+      "text_zh": "推断：视频 38.4–39.8 s 左手完成最后一轮明显转向，相机进入与末态相近的标志前持机姿态，估计视频 39.0 s 开始满足对准，边界约有 0.5 s 不确定性。log step 50 的 agent 图像估计轴误差为 14.732 度，log step 56 修正后的估计为 3.626 度、取景点约为 (-0.233, 0.103, 1.277) 米，仅作为这一可见转向的辅助依据，并非评分器原生读数。"
+     },
+     {
+      "t": 54.2,
+      "log_step": 70,
+      "text_zh": "推断：视频 40.4–54.2 s 左手持续持机，右手接近和触按时相机仅小幅晃动，未见离开末态位置；视频 54.2 s 仍持机于标志前。log step 70 中在线机器人服务返回 done=True、reason=success，reward 的 live_success=1、graded_by_live=1，联合验收说明末态位置和朝向均达标。精确区域和角度阈值不能从这些遮挡画面独立复测。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "观察：视频 28.0–40.4 s，机器人左手（正面画面右侧）以弯曲手指和手掌围持相机一端，抬起并转动相机；视频 40.4–54.2 s 继续持机。相机调姿由左手完成，未见右手共同抓握调姿。"
+     }
+    ]
+   },
+   "shutter_touched": {
+    "method": "visual",
+    "intervals": [
+     [
+      54.2,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 44.4,
+      "log_step": 62,
+      "text_zh": "观察：视频 42.4–44.8 s，画面左侧的机器人右手移向左手持住的相机，近景可见银色机身顶板与白色指尖；随后手指逐渐遮住顶板，靠近本身不足以证明接触快门。"
+     },
+     {
+      "t": 49.6,
+      "log_step": 65,
+      "text_zh": "推断：视频 45.2–49.6 s 右手伸指试探机身顶端后有退回再接近的动作；log step 65 的第一次候选触按以及 log step 70 的候选 A 均返回 done=False。结合该阶段左手持机姿态与最终达标姿态接近，未将这些被遮挡的试探计为已触及快门。"
+     },
+     {
+      "t": 54.2,
+      "log_step": 70,
+      "text_zh": "推断：视频 53.2–54.2 s，机器人右手伸出的指尖逐渐贴近相机顶端，左手始终围���机身，视频 54.2 s 为最后持机触按姿态。log step 70 中右手中指执行候选 C 的触按后，在线服务首次返回 done=True、reason=success；结合 reward 的 success=1，按同一步联合验收规则判定末态已接触快门。将到达锚点记为视频 54.2 s，精确接触点被标志和手部遮挡，不能仅靠近景分辨接触前沿。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "抓握",
+      "note_zh": "观察：视频 44.4–54.2 s，机器人左手弯指围持相机一端，在右手伸指触按期间固定机身。"
+     },
+     {
+      "who": "双臂协作",
+      "how": "压/擦",
+      "note_zh": "推断：视频 53.2–54.2 s 机器人右手伸指接近并触按银色顶板的快门区域，左手同时持机；log step 70 的脚本指定伸直右手中指，服务随候选 C 的触按返回成功，支持按压式接触。右手未围握相机。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "radial_dist",
+   "unit": "m",
+   "threshold": 0.16,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "region_height_dist",
+   "unit": "m",
+   "threshold": 0.1,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "view_angle_deg",
+   "unit": "deg",
+   "threshold": 10,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人左手在早期抓取松开后重新围持并抬起相机，估计视频 39.0 s 起对准标志；右手最终伸指触按快门，视频 54.2 s 的末态两个节点均成立，在线服务和评分器均判成功。无失败断点，早期偏差已由后续动作修复。没有状态日志或评分器报告的三项原生指标数值，且标志和手部遮住精确接触点；对准起点是结合可见调姿与日志图像估计的近似，末态联合成立由成功判定确认。",
+ "conf": 7,
+ "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，两个节点末态成立（服务 done=True、success=1），failure 为 null 合理；左手持机、右手触按的接触标注与帧一致，对准起点 39.0 s 为已标注的推断。",
+    "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": "2026-11-12T11:20:45Z",
+  "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/dexjoco-bimanual-photograph-i00-standard--2c4ckrv/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexjoco-bimanual-photograph-i00-standard--2c4ckrv/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "f96dca05f4ec4a05289e563ab67be3c4545d77c2ee4283ef7fafdb4efeffc0ea",
+   "video_sha256": "176069116aed56eb5621d6f36573bc379bb0816a421731e623d95bfbc3287758",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/dexjoco/fold_glasses/trajectories/dexjoco-fold-glasses-i00-privileged--cregdnj.json

@@ -0,0 +1,284 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexjoco-fold-glasses-i00-privileged--cregdnj",
+ "taskKey": "dexjoco/fold_glasses",
+ "dag_sha256": "e5d0096234bd74e7a31510d7ca64f54fc36c278269d57718abe3ae2e9ee5f0f4",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "deterministic": 1,
+   "grader_error": 0,
+   "infra_error": 0,
+   "missing_trajectory": 0,
+   "bad_trajectory": 0,
+   "all_success": 1,
+   "determinism_checked": 0,
+   "replays": 1,
+   "instance_ok": 1,
+   "n_actions": 2876,
+   "sim_steps": 2876,
+   "task_failure": 0,
+   "replay_wall_s": 15.9,
+   "video_rendered": 1,
+   "visible_cpus": 32,
+   "openblas_num_threads_env": 4,
+   "blas_threads": 4,
+   "threads_ok": 1,
+   "agent_sessions": 13,
+   "agent_resets": 14,
+   "agent_sim_steps": 23157,
+   "agent_records": 4,
+   "agent_replays": 2,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 1,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 1.782,
+   "peak_processes": 8,
+   "peak_disk_gb": 0.255,
+   "peak_cpus": 4.45,
+   "peak_pids": 97,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "over_cpus": 1,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "media",
+  "url": "media:dexjoco-fold-glasses-i00-privileged--cregdnj.mp4",
+  "duration_s": 57.6,
+  "fps": 25.0,
+  "frames": 1440,
+  "sha256": "8679d466675307f983a849a4064170cb4e4009575f5c1c865f69fc7f21e19fe0"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 3,
+  "base": 0,
+  "peak": 3,
+  "final": 3,
+  "nodes": {
+   "temples_closed_1": {
+    "method": "visual",
+    "intervals": [
+     [
+      33.0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 33.0,
+      "log_step": 19,
+      "text_zh": "推断：视频 28.0–35.2 s，右侧 Allegro 手的单根弯曲手指在盒内持续横向拨动镜腿，镜框靠盒壁，随后手指抬起；视频 36.0 s 的腕部视角可见一条镜腿横向收在镜框后，另一条仍展开。将第一条折合的时刻估计为视频 33.0 s，因盒壁遮挡，跨越 1.1 rad 的时刻约有 2 s 误差。"
+     },
+     {
+      "t": 35.2,
+      "log_step": 19,
+      "text_zh": "观察：log step 19 的最终生成运行在 control step 1768 打印两铰链角为 0.3259、1.2128 rad；同一步记录保存 2876 行轨迹，log step 20 对该交付文件复播成功且终态哈希一致。视频 35.2–36.0 s 可见第一次拨合结束、手指离开，支持该记录属于交付轨迹。"
+     },
+     {
+      "t": 57.56,
+      "log_step": 20,
+      "text_zh": "推断：视频 36.0–57.56 s 眼镜始终在盒内，机器人改拨另一条镜腿；未见第一条重新展开。评分器 success=1、all_success=1，且 log step 19 交付运行的终态两角为 1.1469、1.1592 rad，支持第一条折合保持到结束。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "推/扫",
+      "note_zh": "右侧 Allegro 手以弯曲中指的指端抵住一条镜腿侧面横向拨合；镜框靠木盒壁承受反力，手指没有与另一手指对夹该镜腿。"
+     }
+    ]
+   },
+   "temples_closed_2": {
+    "method": "visual",
+    "intervals": [
+     [
+      56.5,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 45.0,
+      "log_step": 19,
+      "text_zh": "观察：视频 36.0–45.0 s，机器人先将弯曲手指移到另一条仍展开的镜腿旁，作一次短推后抬指换位置；视频 46.4–57.56 s 再次在盒内持续横拨，外部视角可见眼镜留在盒壁之间，末尾手指仍在接触。"
+     },
+     {
+      "t": 56.5,
+      "log_step": 20,
+      "text_zh": "推断：log step 20 对交付轨迹复播，报告首次成功结束于 control step 2876；任务指令规定两角均大于 1.1 rad 且中心入盒连续 50 个 50 Hz 控制步，故最后有效窗口约始于 (2876-49)/50=56.54 s，对应视频约 56.5 s。2876/50=57.52 s 与视频 57.6 s 时长吻合，并已用 log step 19 的 control step 1768/50=35.36 s 对照视频 35.2–36.0 s 首次拨合后抬指核对；据此标记两条都折合约在视频 56.5 s 达成。"
+     },
+     {
+      "t": 57.56,
+      "log_step": 19,
+      "text_zh": "观察：视频 56.4–57.56 s，右臂在盒内完成末段拨动，眼镜没有离盒；log step 19 的交付运行终态打印两角为 1.1468879、1.1591551 rad、success=True，log step 20 复播相同终态哈希并通过。盒壁遮住镜腿的部分轮廓，精确角��依据交付运行读数与评分成功，不能单凭末帧测出。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "推/扫",
+      "note_zh": "右侧 Allegro 手用同一弯曲中指先短推眼镜调整位置，再从另一侧横向拨合剩余镜腿；利用盒壁限制镜框平移，没有对镜腿建立对向夹持。"
+     }
+    ]
+   },
+   "glasses_in_case": {
+    "method": "visual",
+    "intervals": [
+     [
+      16.6,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 6.0,
+      "log_step": 19,
+      "text_zh": "观察：视频 2.4–4.4 s 右手手指合拢夹住镜框，视频 4.8–6.0 s 眼镜随手抬起且原桌面位置空出；视频 9.0–15.6 s 右臂将眼镜横移到木盒上方并下降，搬运过程中眼镜保持在手中。"
+     },
+     {
+      "t": 16.6,
+      "log_step": 19,
+      "text_zh": "观察：视频 16.4–17.2 s，机器人松开镜框并抬手，眼镜落低到盒壁包围的高度；视频 17.6–18.0 s 可清楚看到眼镜留在盒内。以视频约 16.6 s 标记入盒，位置阈值的过界时刻有约 0.4 s 误差；此前悬在盒口上方的阶段不计入。"
+     },
+     {
+      "t": 57.56,
+      "log_step": 20,
+      "text_zh": "推断：视频 18.0–57.56 s 的外部视角显示眼镜虽被手指在盒内推移，但始终未离盒或再次抬出；log step 19 的交付运行在 control step 921 打印眼镜中心 [-0.3065,0.2896,1.0164] m，末步为 [-0.3266,0.2533,1.0164] m。评分器 success=1 和 log step 20 的交付文件复播成功共同确认结束时中心满足入盒范围。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "右侧 Allegro 手以拇指与对侧手指夹持镜框，抬起、搬到敞口盒上方并下降，随后张指释放；后续折合时用指端拨动眼镜，其中心持续留在盒内。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "temple_0_angle",
+   "unit": "rad",
+   "threshold": 1.1,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "temple_1_angle",
+   "unit": "rad",
+   "threshold": 1.1,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "abs_glass_case_dx",
+   "unit": "m",
+   "threshold": 0.145,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "abs_glass_case_dy",
+   "unit": "m",
+   "threshold": 0.145,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "glass_case_dz_lower_bound",
+   "unit": "m",
+   "threshold": -0.045,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "glass_case_dz_upper_bound",
+   "unit": "m",
+   "threshold": 0.0275,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "评分轨迹先由右手夹持眼镜并于视频约 16.6 s 入盒，再借盒壁支撑，用单根弯曲手指先后拨合两条镜腿，最终三个节点均成立。视频约 56.5 s 起满足联合条件，交付文件在 log step 20 的复播及评分器均报告成功，末尾仍接触眼镜符合本任务允许持物或接触的验收规则。没有状态日志，盒壁遮挡了部分镜腿；第一条折合时刻约有 2 s 误差，第二条依据交付轨迹控制步与连续 50 步验收推断，六项原生指标未在评分器字段中单独报告，故值留空。",
+ "conf": 6,
+ "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，三个节点终态均为真，与评分器 success=1 一致。入盒时刻（视频约 16.6 s）与帧相符。镜腿折合时刻被盒壁遮挡，帧无法分辨，文件已如实标为推断并写明约 2 s 误差，依据交付轨迹控制步，未发现需要修改的内容。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-11-12T11:20:47Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/dexjoco-fold-glasses-i00-privileged--cregdnj/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexjoco-fold-glasses-i00-privileged--cregdnj/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "892aaf9ea99ccd052e67acd0a8817f2e5982cc06b1c31f48adf229e6322c145b",
+   "video_sha256": "8679d466675307f983a849a4064170cb4e4009575f5c1c865f69fc7f21e19fe0",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/dexjoco/fold_glasses/trajectories/dexjoco-fold-glasses-i00-standard--n6efhng.json

@@ -0,0 +1,283 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexjoco-fold-glasses-i00-standard--n6efhng",
+ "taskKey": "dexjoco/fold_glasses",
+ "dag_sha256": "e5d0096234bd74e7a31510d7ca64f54fc36c278269d57718abe3ae2e9ee5f0f4",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "live_success": 1,
+   "graded_by_live": 1,
+   "replays": 0,
+   "infra_error": 0,
+   "sim_error": 0,
+   "grader_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "no_verdict": 0,
+   "task_failure": 0,
+   "n_actions": 7518,
+   "sim_steps": 7518,
+   "physics_steps": 75180,
+   "steps": 1504,
+   "calls": 138,
+   "elapsed_s": 2259.4,
+   "video_rendered": 1,
+   "instance_mismatch": 0,
+   "too_few_cpus": 0,
+   "visible_cpus": 32,
+   "blas_threads": 4,
+   "gave_up": 0,
+   "sim_startups": 1,
+   "resumes": 0,
+   "resume_state_mismatch": 0,
+   "resume_max_joint_dev": 0.0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 0.343,
+   "peak_processes": 10,
+   "peak_disk_gb": 0.387,
+   "peak_cpus": 1.7,
+   "peak_pids": 58,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "media",
+  "url": "media:dexjoco-fold-glasses-i00-standard--n6efhng.mp4",
+  "duration_s": 50.2,
+  "fps": 25.0,
+  "frames": 1255,
+  "sha256": "2745a0208057959c20d0c8f9eb93230ef22e9c3fb1bc8f39bf7a1f63f9b0276e"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 3,
+  "base": 0,
+  "peak": 3,
+  "final": 3,
+  "nodes": {
+   "temples_closed_1": {
+    "method": "visual",
+    "intervals": [
+     [
+      31.2,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–1.0 s，眼镜在盒外桌面上，两条镜腿均向后展开。"
+     },
+     {
+      "t": 31.2,
+      "log_step": 51,
+      "text_zh": "观察：视频 24.0–26.4 s，机器人以拇指与其余手指夹住镜框中部并将整副眼镜提离桌面；视频 27.4–31.6 s，机器人持镜翻转、滚动手腕，腕部近景中一条镜腿向镜框内侧转动，其弯曲末端出现在镜片背面。log step 49、51 记录了相应的持镜滚腕动作。"
+     },
+     {
+      "t": 31.2,
+      "log_step": 67,
+      "text_zh": "推断：根据视频 30.8–31.6 s 的收拢变化，将至少一条镜腿超过折合阈值的起点近似记为视频 31.2 s，误差约半秒；后续搬运和盒内调整未见两条镜腿同时重新展开。视频 50.16 s 眼镜仍在盒中，且实时机器人服务在 log step 67 返回 done=True、reason=success，评分器 live_success=1，支持本节点终态为真；精确关节角未记录。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "观察：机器人右侧 Allegro 手的拇指与其余手指夹住镜框中部，提起后翻转、滚腕，使镜腿向镜框内侧收拢；此前曾用伸出的手指扫拨镜腿，带动整副眼镜在桌面上滑转。并非直接捏住该镜腿铰链。"
+     }
+    ]
+   },
+   "temples_closed_2": {
+    "method": "visual",
+    "intervals": [
+     [
+      49.8,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 46.8,
+      "log_step": null,
+      "text_zh": "观察：视频 44.0–49.2 s，眼镜框已在盒壁内，一条黑色镜腿仍斜向上伸出；视频 46.8–49.2 s 的腕部视角也可见它尚未贴近镜框背面，不能提前记为两条均折合。"
+     },
+     {
+      "t": 49.8,
+      "log_step": 66,
+      "text_zh": "观察：视频 49.2–50.16 s，机器人伸直手指在盒中横向拨动突出镜腿，该镜腿逐渐降低并向镜框靠拢；最后一秒逐帧查看到视频 50.16 s，眼镜没有被带出盒。log step 66 的这一拨动调用因服务返回 EpisodeOver: success 而结束。"
+     },
+     {
+      "t": 49.8,
+      "log_step": 67,
+      "text_zh": "推断：结合视频 49.6–50.16 s 的拨合变化和 log step 67 的实时成功状态，将两条镜腿均折合的起点近似记为视频 49.8 s，误差约半秒；终态由原生成功判定确认。手指遮挡了铰链附近，不能从画面独立读出严格大于 1.1 rad 的精确越界帧。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "推/扫",
+      "note_zh": "观察：机器人右侧 Allegro 手以伸直的手指在盒中抵住仍突出的黑色镜腿，横向扫拨使其降低并向镜框内侧折合；没有合手夹住这条镜腿。此前机器���曾夹持镜框滚腕尝试使另一条镜腿也收拢。"
+     }
+    ]
+   },
+   "glasses_in_case": {
+    "method": "visual",
+    "intervals": [
+     [
+      36.6,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 36.6,
+      "log_step": 58,
+      "text_zh": "观察：视频 34.8–36.4 s，机器人夹着镜框移到浅棕色敞口盒上方并下降；视频 36.48–36.72 s，眼镜随松指进一步落低，镜片进入盒壁围成的区域，镜框上缘也降到接近盒沿的高度，之后手撤离而眼镜留在盒中。log step 57、58 分别记录了搬运下降和松指撤手。"
+     },
+     {
+      "t": 44,
+      "log_step": 63,
+      "text_zh": "观察：视频 37.2–43.6 s，镜框保持在盒壁围成的区域内，突出镜腿挂在盒沿；视频 43.6–44.4 s，机器人用伸出的食指从该镜腿下方抬拨，眼镜转落至盒底附近。此过程眼镜中心未见被抬出盒口，不能因挂沿或仍有接触而取消入盒进度。"
+     },
+     {
+      "t": 50.16,
+      "log_step": 67,
+      "text_zh": "推断：依据视频 36.48–36.72 s 的下降，将眼镜中心进入盒体接受区域的起点近似记为视频 36.6 s，误差约半秒；视频 44.4–50.16 s 的外部视角与腕部视角均显示镜框仍在盒内。log step 67 的服务成功状态及评分器 success=1 确认终态符合中心位置条件；没有状态日志，不能逐帧量测世界轴边界。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "观察：机器人右侧 Allegro 手的拇指与其余手指夹住镜框中部，将眼镜搬到敞口木盒上方并下降，随后张指释放；眼镜留在盒内后，机器人又以伸出食指从突出镜腿下方抬拨，使挂沿的眼镜转落至盒底附近。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "temple_0_angle",
+   "unit": "rad",
+   "threshold": 1.1,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "temple_1_angle",
+   "unit": "rad",
+   "threshold": 1.1,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "abs_glass_case_dx",
+   "unit": "m",
+   "threshold": 0.145,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "abs_glass_case_dy",
+   "unit": "m",
+   "threshold": 0.145,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "glass_case_dz_lower_bound",
+   "unit": "m",
+   "threshold": -0.045,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "glass_case_dz_upper_bound",
+   "unit": "m",
+   "threshold": 0.0275,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人夹持并翻转眼镜，使第1条镜腿收拢，再将眼镜降入盒内，处理镜腿挂沿后在盒内拨合第2条镜腿，最终三个节点均成立。视频末段与实时机器人服务在 log step 67 返回的 done=True、reason=success 以及评分器 live_success=1 一致，因此无失败断点。缺少关节角、眼镜中心和联合保持计数的状态日志，严格阈值的越界时间由视频近似判断，终态由原生成功结果确认；参考成功视频仅用于辨认折合和入盒的可见形态。",
+ "conf": 7,
+ "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，三个节点终态为真、无失败断点；时间与接触标注与帧基本一致（联合保持50个控制步约合视频0.3 s，与49.8 s 起点相容）。未发现需改之处。",
+    "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": "2026-11-12T11:19:10Z",
+  "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/dexjoco-fold-glasses-i00-standard--n6efhng/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexjoco-fold-glasses-i00-standard--n6efhng/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "021eac41b568368f5d315e2c297fdb37078f562438889146c238ab2513f234f5",
+   "video_sha256": "2745a0208057959c20d0c8f9eb93230ef22e9c3fb1bc8f39bf7a1f63f9b0276e",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/alpine_slalom/trajectories/embodied-olympics-fun-alpine-slalom-i00-privileged--jeydurz.json

@@ -0,0 +1,981 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-alpine-slalom-i00-privileged--jeydurz",
+ "taskKey": "embodied-olympics-fun/alpine_slalom",
+ "dag_sha256": "0382103c21732bd9a43317e1385d02d1f88b24ff2896d1d6062c6a895ff37a48",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "determinism_checked": 0,
+   "replays": 1,
+   "n_actions": 8472,
+   "sim_steps": 8472,
+   "horizon": 9000,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 34,
+   "stages_total": 34,
+   "fell": 0,
+   "replay_s": 79.99,
+   "bad_trajectory": 0,
+   "metric_gates": 32.0,
+   "metric_speed_m_s": 4.375,
+   "metric_time_s": 82.709,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 218,
+   "agent_resets": 213,
+   "agent_replays": 10,
+   "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": 4.504,
+   "peak_processes": 19,
+   "peak_disk_gb": 0.259,
+   "peak_cpus": 8.02,
+   "peak_pids": 42,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "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": "media",
+  "url": "media:embodied-olympics-fun-alpine-slalom-i00-privileged--jeydurz.mp4",
+  "duration_s": 84.76,
+  "fps": 25.0,
+  "frames": 2119,
+  "sha256": "4bcd5d5f3a5710503a37795fd48341863e36046237deea1bda9c04a7a75a81a3"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 33,
+  "base": null,
+  "peak": null,
+  "final": 33,
+  "nodes": {
+   "gate_1_passed": {
+    "method": "visual",
+    "intervals": [
+     [
+      7.4,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 7.4,
+      "log_step": null,
+      "text_zh": "观察：视频 6.8—7.6 s，机器人保持蹲姿、双脚连同雪板贴着雪面滑行，从第1对旗杆附近的上坡侧移动到下坡侧，随后转向下一门。远景不能逐点辨清两脚及两板尖。"
+     },
+     {
+      "t": 7.4,
+      "log_step": 51,
+      "text_zh": "推断：结合上述视频位置变化，以及 log step 51 对交付文件的官方回放结果（success=True、阶段包含 start、gate 1 至 gate 32、finish），第1门约在视频 7.4 s 达成，时间误差约 0.5 s；通过后的事件锁存。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 6.8—7.6 s，机器人双腿屈曲，以固定在双脚的雪板支承于雪面并滑过第1门附近；双臂未撑地。"
+     }
+    ]
+   },
+   "gate_2_passed": {
+    "method": "visual",
+    "intervals": [
+     [
+      10.0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 10.0,
+      "log_step": null,
+      "text_zh": "观察：视频 9.2—10.2 s，机器人保持蹲姿、双脚连同雪板贴着雪面滑行，从第2对旗杆附近的上坡侧移动到下坡侧，随后转向下一门。远景不能逐点辨清两脚及两板尖。"
+     },
+     {
+      "t": 10.0,
+      "log_step": 51,
+      "text_zh": "推断：结合上述视频位置变化，以及 log step 51 对交付文件的官方回放结果（success=True、阶段包含 start、gate 1 至 gate 32、finish），第2门约在视频 10.0 s 达成，时间误差约 0.5 s；通过后的事件锁存。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 9.2—10.2 s，机器人双腿屈曲，以固定在双脚的雪板支承于雪面并滑过第2门附近；双臂未撑地。"
+     }
+    ]
+   },
+   "gate_3_passed": {
+    "method": "visual",
+    "intervals": [
+     [
+      12.4,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 12.4,
+      "log_step": null,
+      "text_zh": "观察：视频 11.8—12.6 s，机器人保持蹲姿、双脚连同雪板贴着雪面滑行，从第3对旗杆附近的上坡侧移动到下坡侧，随后转向下一门。远景不能逐点辨清两脚及两板尖。"
+     },
+     {
+      "t": 12.4,
+      "log_step": 51,
+      "text_zh": "推断：结合上述视频位置变化，以及 log step 51 对交付文件的官方回放结果（success=True、阶段包含 start、gate 1 至 gate 32、finish），第3门约在视频 12.4 s 达成，时间误差约 0.5 s；通过后的事件锁存。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 11.8—12.6 s，机器人双腿屈曲，以固定在双脚的雪板支承于雪面并滑过第3门附近；双臂未撑地。"
+     }
+    ]
+   },
+   "gate_4_passed": {
+    "method": "visual",
+    "intervals": [
+     [
+      14.8,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 14.8,
+      "log_step": null,
+      "text_zh": "观察：视频 14.2—15.0 s，机器人保持蹲姿、双脚连同雪板贴着雪面滑行，从第4对旗杆附近的上坡侧移动到下坡侧，随后转向下一门。远景不能逐点辨清两脚及两板尖。"
+     },
+     {
+      "t": 14.8,
+      "log_step": 51,
+      "text_zh": "推断：结合上述视频位置变化，以及 log step 51 对交付文件的官方回放结果（success=True、阶段包含 start、gate 1 至 gate 32、finish），第4门约在视频 14.8 s 达成，时间误差约 0.5 s；通过后的事件锁存。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 14.2—15.0 s，机器人双腿屈曲，以固定在双脚的雪板支承于雪面并滑过第4门附近；双臂未撑地。"
+     }
+    ]
+   },
+   "gate_5_passed": {
+    "method": "visual",
+    "intervals": [
+     [
+      17.2,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 17.2,
+      "log_step": null,
+      "text_zh": "观察：视频 16.6—17.6 s，机器人保持蹲姿、双脚连同雪板贴着雪面滑行，从第5对旗杆附近的上坡侧移动到下坡侧，随后转向下一门。远景不能逐点辨清两脚及两板尖。"
+     },
+     {
+      "t": 17.2,
+      "log_step": 51,
+      "text_zh": "推断：结合上述视频位置变化，以及 log step 51 对交付文件的官方回放结果（success=True、阶段包含 start、gate 1 至 gate 32、finish），第5门约在视频 17.2 s 达成，时间误差约 0.5 s；通过后的事件锁存。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 16.6—17.6 s，机器人双腿屈曲，以固定在双脚的雪板支承于雪面并滑过第5门附近；双臂未撑地。"
+     }
+    ]
+   },
+   "gate_6_passed": {
+    "method": "visual",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 51,
+      "text_zh": "观察：log step 51 中，官方回放交付文件得到 success=True，stages 明列 gate 6，共34阶段；inputs/run.json 的评分器字段也报告 replay_success=1、metric_gates=32、stages_reached=34。"
+     },
+     {
+      "t": 84.72,
+      "log_step": null,
+      "text_zh": "推断：据评分器及交付回放的阶段清单，第6门最终已达成。已查看视频全程粗抽帧及视频 83.76—84.72 s 末段；固定远景中后段机器人及门序无法辨清，没有足以定位本门四个受检点越线的画面，故达成区间留 null，不能从会话试跑的时间代入。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本录像仅有固定远景，第6门的机器人脚、雪板与雪面或旗杆之间接触不能辨认，且无状态日志；不据任务典型滑行方案补写接触。"
+     }
+    ]
+   },
+   "gate_7_passed": {
+    "method": "visual",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 51,
+      "text_zh": "观察：log step 51 中，官方回放交付文件得到 success=True，stages 明列 gate 7，共34阶段；inputs/run.json 的评分器字段也报告 replay_success=1、metric_gates=32、stages_reached=34。"
+     },
+     {
+      "t": 84.72,
+      "log_step": null,
+      "text_zh": "推断：据评分器及交付回放的阶段清单，第7门最终已达成。已查看视频全程粗抽帧及视频 83.76—84.72 s 末段；固定远景中后段机器人及门序无法辨清，没有足以定位本门四个受检点越线的画面，故达成区间留 null，不能从会话试跑的时间代入。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本录像仅有固定远景，第7门的机器人脚、雪板与雪面或旗杆之间接触不能辨认，且无状态日志；不据任务典型滑行方案补写接触。"
+     }
+    ]
+   },
+   "gate_8_passed": {
+    "method": "visual",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 51,
+      "text_zh": "观察：log step 51 中，官方回放交付文件得到 success=True，stages 明列 gate 8，共34阶段；inputs/run.json 的评分器字段也报告 replay_success=1、metric_gates=32、stages_reached=34。"
+     },
+     {
+      "t": 84.72,
+      "log_step": null,
+      "text_zh": "推断：据评分器及交付回放的阶段清单，第8门最终已达成。已查看视频全程粗抽帧及视�� 83.76—84.72 s 末段；固定远景中后段机器人及门序无法辨清，没有足以定位本门四个受检点越线的画面，故达成区间留 null，不能从会话试跑的时间代入。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本录像仅有固定远景，第8门的机器人脚、雪板与雪面或旗杆之间接触不能辨认，且无状态日志；不据任务典型滑行方案补写接触。"
+     }
+    ]
+   },
+   "gate_9_passed": {
+    "method": "visual",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 51,
+      "text_zh": "观察：log step 51 中，官方回放交付文件得到 success=True，stages 明列 gate 9，共34阶段；inputs/run.json 的评分器字段也报告 replay_success=1、metric_gates=32、stages_reached=34。"
+     },
+     {
+      "t": 84.72,
+      "log_step": null,
+      "text_zh": "推断：据评分器及交付回放的阶段清单，第9门最终已达成。已查看视频全程粗抽帧及视频 83.76—84.72 s 末段；固定远景中后段机器人及门序无法辨清，没有足以定位本门四个受检点越线的画面，故达成区间留 null，不能从会话试跑的时间代入。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本录像仅有固定远景，第9门的机器人脚、雪板与雪面或旗杆之间接触不能辨认，且无状态日志；不据任务典型滑行方案补写接触。"
+     }
+    ]
+   },
+   "gate_10_passed": {
+    "method": "visual",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 51,
+      "text_zh": "观察：log step 51 中，官方回放交付文件得到 success=True，stages 明列 gate 10，共34阶段；inputs/run.json 的评分器字段也报告 replay_success=1、metric_gates=32、stages_reached=34。"
+     },
+     {
+      "t": 84.72,
+      "log_step": null,
+      "text_zh": "推断：据评分器及交付回放的阶段清单，第10门最终已达成。已查看视频全程粗抽帧及视频 83.76—84.72 s 末段；固定远景中后段机器人及门序无法辨清，没有足以定位本门四个受检点越线的画面，故达成区间留 null，不能从会话试跑的时间代入。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本录像仅有固定远景，第10门的机器人脚、雪板与雪面或旗杆之间接触不能辨认，且无状态日志；不据任务典型滑行方案补写接触。"
+     }
+    ]
+   },
+   "gate_11_passed": {
+    "method": "visual",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 51,
+      "text_zh": "观察：log step 51 中，官方回放交付文件得到 success=True，stages 明列 gate 11，共34阶段；inputs/run.json 的评分器字段也报告 replay_success=1、metric_gates=32、stages_reached=34。"
+     },
+     {
+      "t": 84.72,
+      "log_step": null,
+      "text_zh": "推断：据评分器及交付回放的阶段清单，第11门最终已达成。已查看视频全程粗抽帧及视频 83.76—84.72 s 末段；固定远景中后段机器人及门序无法辨清，没有足以定位本门四个受检点越线的画面，故达成区间留 null，不能从会话试跑的时间代入。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本录像仅有固定远景，第11门的机器人脚、雪板与雪面或旗杆之间接触不能辨认，且无状态日志；不据任务典型滑行方案补写接触。"
+     }
+    ]
+   },
+   "gate_12_passed": {
+    "method": "visual",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 51,
+      "text_zh": "观察：log step 51 中，官方回放交付文件得到 success=True，stages 明列 gate 12，共34阶段；inputs/run.json 的评分器字段也报告 replay_success=1、metric_gates=32、stages_reached=34。"
+     },
+     {
+      "t": 84.72,
+      "log_step": null,
+      "text_zh": "推断：据评分器及交付回放的阶段清单，第12门最终已达成。已查看视频全程粗抽帧及视频 83.76—84.72 s 末段；固定远景中后段机器人及门序无法辨清，没有足以定位本门四个受检点越线的画面，故达成区间留 null，不能从会话试跑的时间代入。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本录像仅有固定远景，第12门的机器人脚、雪板与雪面或旗杆之间接触不能辨认，且无状态日志；不据任务典型滑行方案补写接触。"
+     }
+    ]
+   },
+   "gate_13_passed": {
+    "method": "visual",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 51,
+      "text_zh": "观察：log step 51 中，官方回放交付文件得到 success=True，stages 明列 gate 13，共34阶段；inputs/run.json 的评分器字段也报告 replay_success=1、metric_gates=32、stages_reached=34。"
+     },
+     {
+      "t": 84.72,
+      "log_step": null,
+      "text_zh": "推断：据评分器及交付回放的阶段清单，第13门最终已达成。已查看视频全程粗抽帧及视频 83.76—84.72 s 末段；固定远景中后段机器人及门序无法辨清，没有足以定位本门四个受检点越线的画面，故达成区间留 null，不能从会话试跑的时间代入。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本录像仅有固定远景，第13门的机器人脚、雪板与雪面或旗杆之间接触不能辨认，且无状态日志；不据任务典型滑行方案补写接触。"
+     }
+    ]
+   },
+   "gate_14_passed": {
+    "method": "visual",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 51,
+      "text_zh": "观察：log step 51 中，官方回放交付文件得到 success=True，stages 明列 gate 14，共34阶段；inputs/run.json 的评分器字段也报告 replay_success=1、metric_gates=32、stages_reached=34。"
+     },
+     {
+      "t": 84.72,
+      "log_step": null,
+      "text_zh": "推断：据评分器及交付回放的阶段清单，第14门最终已达成。已查看视频全程粗抽帧及视频 83.76—84.72 s 末段；固定远景中后段机器人及门序无法辨清，没有足以定位本门四个受检点越线的画面，故达成区间留 null，不能从会话试跑的时间代入。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本录像仅有固定远景，第14门的机器人脚、雪板与雪面或旗杆之间接触不能辨认，且无状态日志；不据任务典型滑行方案补写接触。"
+     }
+    ]
+   },
+   "gate_15_passed": {
+    "method": "visual",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 51,
+      "text_zh": "观察：log step 51 中，官方回放交付文件得到 success=True，stages 明列 gate 15，共34阶段；inputs/run.json 的评分器字段也报告 replay_success=1、metric_gates=32、stages_reached=34。"
+     },
+     {
+      "t": 84.72,
+      "log_step": null,
+      "text_zh": "推断：据评分器及交付回放的阶段清单，第15门最终已达成。已查看视频全程粗抽帧及视频 83.76—84.72 s 末段；固定远景中后段机器人及门序无法辨清，没有足以定位本门四个受检点越线的画面，故达成区间留 null，不能从会话试跑的时间代入。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本录像仅有固定远景，第15门的机器人脚、雪板与雪面或旗杆之间接触不能辨认，且无状态日志；不据任务典型滑行方案补写接触。"
+     }
+    ]
+   },
+   "gate_16_passed": {
+    "method": "visual",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 51,
+      "text_zh": "观察：log step 51 中，官方回放交付文件得到 success=True，stages 明列 gate 16，共34阶段；inputs/run.json 的评分器字段也报告 replay_success=1、metric_gates=32、stages_reached=34。"
+     },
+     {
+      "t": 84.72,
+      "log_step": null,
+      "text_zh": "推断：据评分器及交付回放的阶段清单，第16门最终已达成。已查看视频全程粗抽帧及视频 83.76—84.72 s 末段；固定远景中后段机器人及门序无法辨清，没有足以定位本门四个受检点越线的画面，故达成区间留 null，不能从会话试跑的时间代入。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本录像仅有固定远景，第16门的机器人脚、雪板与雪面或旗杆之间接触不能辨认，且无状态日志；不据任务典型滑行方案补写接触。"
+     }
+    ]
+   },
+   "gate_17_passed": {
+    "method": "visual",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 51,
+      "text_zh": "观察：log step 51 中，官方回放交付文件得到 success=True，stages 明列 gate 17，共34阶段；inputs/run.json 的评分器字段也报告 replay_success=1、metric_gates=32、stages_reached=34。"
+     },
+     {
+      "t": 84.72,
+      "log_step": null,
+      "text_zh": "推断：据评分器及交付回放的阶段清单，第17门最终已达成。已查看视频全程粗抽帧及视频 83.76—84.72 s 末段；固定远景中后段机器人及门序无法辨清，没有足以定位本门四个受检点越线的画面，故达成区间留 null，不能从���话试跑的时间代入。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本录像仅有固定远景，第17门的机器人脚、雪板与雪面或旗杆之间接触不能辨认，且无状态日志；不据任务典型滑行方案补写接触。"
+     }
+    ]
+   },
+   "gate_18_passed": {
+    "method": "visual",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 51,
+      "text_zh": "观察：log step 51 中，官方回放交付文件得到 success=True，stages 明列 gate 18，共34阶段；inputs/run.json 的评分器字段也报告 replay_success=1、metric_gates=32、stages_reached=34。"
+     },
+     {
+      "t": 84.72,
+      "log_step": null,
+      "text_zh": "推断：据评分器及交付回放的阶段清单，第18门最终已达成。已查看视频全程粗抽帧及视频 83.76—84.72 s 末段；固定远景中后段机器人及门序无法辨清，没有足以定位本门四个受检点越线的画面，故达成区间留 null，不能从会话试跑的时间代入。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本录像仅有固定远景，第18门的机器人脚、雪板与雪面或旗杆之间接触不能辨认，且无状态日志；不据任务典型滑行方案补写接触。"
+     }
+    ]
+   },
+   "gate_19_passed": {
+    "method": "visual",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 51,
+      "text_zh": "观察：log step 51 中，官方回放交付文件得到 success=True，stages 明列 gate 19，共34阶段；inputs/run.json 的评分器字段也报告 replay_success=1、metric_gates=32、stages_reached=34。"
+     },
+     {
+      "t": 84.72,
+      "log_step": null,
+      "text_zh": "推断：据评分器及交付回放的阶段清单，第19门最终已达成。已查看视频全程粗抽帧及视频 83.76—84.72 s 末段；固定远景中后段机器人及门序无法辨清，没有足以定位本门四个受检点越线的画面，故达成区间留 null，不能从会话试跑的时间代入。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本录像仅有固定远景，第19门的机器人脚、雪板与雪面或旗杆之间接触不能辨认，且无状态日志；不据任务典型滑行方案补写接触。"
+     }
+    ]
+   },
+   "gate_20_passed": {
+    "method": "visual",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 51,
+      "text_zh": "观察：log step 51 中，官方回放交付文件得到 success=True，stages 明列 gate 20，共34阶段；inputs/run.json 的评分器字段也报告 replay_success=1、metric_gates=32、stages_reached=34。"
+     },
+     {
+      "t": 84.72,
+      "log_step": null,
+      "text_zh": "推断：据评分器及交付回放的阶段清单，第20门最终已达成。已查看视频全程粗抽帧及视频 83.76—84.72 s 末段；固定远景中后段机器人及门序无法辨清，没有足以定位本门四个受检点越线的画面，故达成区间留 null，不能从会话试跑的时间代入。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本录像仅有固定远景，第20门的机器人脚、雪板与雪面或旗杆之间接触不能辨认，且无状态日志；不据任务典型滑行方案补写接触。"
+     }
+    ]
+   },
+   "gate_21_passed": {
+    "method": "visual",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 51,
+      "text_zh": "观察：log step 51 中，官方回放交付文件得到 success=True，stages 明列 gate 21，共34阶段；inputs/run.json 的评分器字段也报告 replay_success=1、metric_gates=32、stages_reached=34。"
+     },
+     {
+      "t": 84.72,
+      "log_step": null,
+      "text_zh": "推断：据评分器及交付回放的阶段清单，第21门最终已达成。已查看视频全程粗抽帧及视频 83.76—84.72 s 末段；固定远景中后段机器人及门序无法辨清，没有足以定位本门四个受检点越线的画面，故达成区间留 null，不能从会话试跑的时间代入。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本录像仅有固定远景，第21门的机器人脚、雪板与雪面或旗杆之间接触不能辨认，且无状态日志；不据任务典型滑行方案补写接触。"
+     }
+    ]
+   },
+   "gate_22_passed": {
+    "method": "visual",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 51,
+      "text_zh": "观察：log step 51 中，官方回放交付文件得到 success=True，stages 明列 gate 22，共34阶段；inputs/run.json 的评分器字段也报告 replay_success=1、metric_gates=32、stages_reached=34。"
+     },
+     {
+      "t": 84.72,
+      "log_step": null,
+      "text_zh": "推断：据评分器及交付回放的阶段清单，第22门最终已达成。已查看视频全程粗抽帧及视频 83.76—84.72 s 末段；固定远景中后段机器人及门序无法辨清，没有足以定位本门四个受检点越线的画面，故达成区间留 null，不能从会话试跑的时间代入。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本录像仅有固定远景，第22门的机器人脚、雪板与雪面或旗杆之间接触不能辨认，且无状态日志；不据任务典型滑行方案补写接触。"
+     }
+    ]
+   },
+   "gate_23_passed": {
+    "method": "visual",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 51,
+      "text_zh": "观察：log step 51 中，官方回放交付文件得到 success=True，stages 明列 gate 23，共34阶段；inputs/run.json 的评分器字段也报告 replay_success=1、metric_gates=32、stages_reached=34。"
+     },
+     {
+      "t": 84.72,
+      "log_step": null,
+      "text_zh": "推断：据评分器及交付回放的阶段清单，第23门最终已达成。已查看视频全程粗抽帧及视频 83.76—84.72 s 末段；固定远景中后段机器人及门序无法辨清，没有足以定位本门四个受检点越线的画面，故达成区间留 null，不能从会话试跑的时间代入。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本录像仅有固定远景，第23门的机器人脚、雪板与雪面或旗杆之间接触不能辨认，且无状态日志；不据任务典型滑行方案补写接触。"
+     }
+    ]
+   },
+   "gate_24_passed": {
+    "method": "visual",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 51,
+      "text_zh": "观察：log step 51 中，官方回放交付文件得到 success=True，stages 明列 gate 24，共34阶段；inputs/run.json 的评分器字段也报告 replay_success=1、metric_gates=32、stages_reached=34。"
+     },
+     {
+      "t": 84.72,
+      "log_step": null,
+      "text_zh": "推断：据评分器及交付回放的阶段清单，第24门最终已达成。已查看视频全程粗抽帧及视频 83.76—84.72 s 末段；固定远景中后段机器人及门序无法辨清，没有足以定位本门四个受检点越线的画面，故达成区间留 null，不能从会话试跑的时间代入。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本录像仅有固定远景，第24门的机器人脚、雪板与雪面或旗杆之间接触不能辨认，且无状态日志；不据任务典型滑行方案补写接触。"
+     }
+    ]
+   },
+   "gate_25_passed": {
+    "method": "visual",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 51,
+      "text_zh": "观察：log step 51 中，官方回放交付文件得到 success=True，stages 明列 gate 25，共34阶段；inputs/run.json 的评分器字段也报告 replay_success=1、metric_gates=32、stages_reached=34。"
+     },
+     {
+      "t": 84.72,
+      "log_step": null,
+      "text_zh": "推断：据评分器及交付回放的阶段清单，第25门最终已达成。已查看视频全程粗抽帧及视频 83.76—84.72 s 末段；固定远景中后段机器人及门序无法辨清，没有足以定位本门四个受检点越线的画面，故达成区间留 null，不能从会话试跑的时间代入。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本录像仅有固定远景，第25门的机器人脚、雪板与雪面或旗杆之间接触不能辨认，且无状态日志；不据任务典型滑行方案补写接触。"
+     }
+    ]
+   },
+   "gate_26_passed": {
+    "method": "visual",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 51,
+      "text_zh": "观察：log step 51 中，官方回放交付文件得到 success=True，stages 明列 gate 26，共34阶段；inputs/run.json 的评分器字段也报告 replay_success=1、metric_gates=32、stages_reached=34。"
+     },
+     {
+      "t": 84.72,
+      "log_step": null,
+      "text_zh": "推断：据评分器及交付回放的阶段清单，第26门最终已达成。已查看视频全程粗抽帧及视频 83.76—84.72 s 末段；固定远景中后段机器人及门序无法辨清，没有足以定位本门四个受检点越线的画面，故达成区间留 null，不能从会话试跑的时间代入。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本录像仅���固定远景，第26门的机器人脚、雪板与雪面或旗杆之间接触不能辨认，且无状态日志；不据任务典型滑行方案补写接触。"
+     }
+    ]
+   },
+   "gate_27_passed": {
+    "method": "visual",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 51,
+      "text_zh": "观察：log step 51 中，官方回放交付文件得到 success=True，stages 明列 gate 27，共34阶段；inputs/run.json 的评分器字段也报告 replay_success=1、metric_gates=32、stages_reached=34。"
+     },
+     {
+      "t": 84.72,
+      "log_step": null,
+      "text_zh": "推断：据评分器及交付回放的阶段清单，第27门最终已达成。已查看视频全程粗抽帧及视频 83.76—84.72 s 末段；固定远景中后段机器人及门序无法辨清，没有足以定位本门四个受检点越线的画面，故达成区间留 null，不能从会话试跑的时间代入。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本录像仅有固定远景，第27门的机器人脚、雪板与雪面或旗杆之间接触不能辨认，且无状态日志；不据任务典型滑行方案补写接触。"
+     }
+    ]
+   },
+   "gate_28_passed": {
+    "method": "visual",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 51,
+      "text_zh": "观察：log step 51 中，官方回放交付文件得到 success=True，stages 明列 gate 28，共34阶段；inputs/run.json 的评分器字段也报告 replay_success=1、metric_gates=32、stages_reached=34。"
+     },
+     {
+      "t": 84.72,
+      "log_step": null,
+      "text_zh": "推断：据评分器及交付回放的阶段清单，第28门最终已达成。已查看视频全程粗抽帧及视频 83.76—84.72 s 末段；固定远景中后段机器人及门序无法辨清，没有足以定位本门四个受检点越线的画面，故达成区间留 null，不能从会话试跑的时间代入。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本录像仅有固定远景，第28门的机器人脚、雪板与雪面或旗杆之间接触不能辨认，且无状态日志；不据任务典型滑行方案补写接触。"
+     }
+    ]
+   },
+   "gate_29_passed": {
+    "method": "visual",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 51,
+      "text_zh": "观察：log step 51 中，官方回放交付文件得到 success=True，stages 明列 gate 29，共34阶段；inputs/run.json 的评分器字段也报告 replay_success=1、metric_gates=32、stages_reached=34。"
+     },
+     {
+      "t": 84.72,
+      "log_step": null,
+      "text_zh": "推断：据评分器及交付回放的阶段清单，第29门最终已达成。已查看视频全程粗抽帧及视频 83.76—84.72 s 末段；固定远景中后段机器人及门序无法辨清，没有足以定位本门四个受检点越线的画面，故达成区间留 null，不能从会话试跑的时间代入。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本录像仅有固定远景，第29门的机器人脚、雪板与雪面或旗杆之间接触不能辨认，且无状态日志；不据任务典型滑行方案补写接触。"
+     }
+    ]
+   },
+   "gate_30_passed": {
+    "method": "visual",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 51,
+      "text_zh": "观察：log step 51 中，官方回放交付文件得到 success=True，stages 明列 gate 30，共34阶段；inputs/run.json 的评分器字段也报告 replay_success=1、metric_gates=32、stages_reached=34。"
+     },
+     {
+      "t": 84.72,
+      "log_step": null,
+      "text_zh": "推断：据评分器及交付回放的阶段清单，第30门最终已达成。已查看视频全程粗抽帧及视频 83.76—84.72 s 末段；固定远景中后段机器人及门序无法辨清，没有足以定位本门四个受检点越线的画面，故达成区间留 null，不能从会话试跑的时间代入。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本录像仅有固定远景，第30门的机器人脚、雪板与雪面或旗杆之间接触不能辨认，且无状态日志；不据任务典型滑行方案补写接触。"
+     }
+    ]
+   },
+   "gate_31_passed": {
+    "method": "visual",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 51,
+      "text_zh": "观察：log step 51 中，官方回放交付文件得到 success=True，stages 明列 gate 31，共34阶段；inputs/run.json 的评分器字段也报告 replay_success=1、metric_gates=32、stages_reached=34。"
+     },
+     {
+      "t": 84.72,
+      "log_step": null,
+      "text_zh": "推断：据评分器及交付回放的阶段清单，第31门最终已达成。已查看视频全程粗抽帧及视频 83.76—84.72 s 末段；固定远景中后段机器人及门序无法辨清，没有足以定位本门四个受检点越线的画面，故达成区间留 null，不能从会话试跑的时间代入。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本录像仅有固定远景，第31门的机器人脚、雪板与雪面或旗杆之间接触不能辨认，且无状态日志；不据任务典型滑行方案补写接触。"
+     }
+    ]
+   },
+   "gate_32_passed": {
+    "method": "visual",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 51,
+      "text_zh": "观察：log step 51 中，官方回放交付文件得到 success=True，stages 明列 gate 32，共34阶段；inputs/run.json 的评分器字段也报告 replay_success=1、metric_gates=32、stages_reached=34。"
+     },
+     {
+      "t": 84.72,
+      "log_step": null,
+      "text_zh": "推断：据评分器及交付回放的阶段清单，第32门最终已达成。已查看视频全程粗抽帧及视频 83.76—84.72 s 末段；固定远景中后段机器人及门序无法辨清，没有足以定位本门四个受检点越线的画面，故达成区间留 null，不能从会话试跑的时间代入。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本录像仅有固定远景，第32门的机器人脚、雪板与雪面或旗杆之间接触不能辨认，且无状态日志；不据任务典型滑行方案补写接触。"
+     }
+    ]
+   },
+   "finish_crossed": {
+    "method": "visual",
+    "intervals": [
+     [
+      84.72,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 84.72,
+      "log_step": 65,
+      "text_zh": "观察：log step 65 对最终交付文件的官方回放输出 success=True、failed=None、34阶段、time_s=84.72、executed_rows=8472；log step 51 的同一交付生成和回放清单含 finish。评分器记录 replay_success=1、metric_gates=32。"
+     },
+     {
+      "t": 84.72,
+      "log_step": 65,
+      "text_zh": "推断：交付文件8472个控制步，每步0.01 s，对应视频约84.72 s 的终点验收（视频总长84.76 s）；以视频约2.0 s 双脚滑过起点红线，与84.72减评分器净比赛用时82.709得到的回合起滑时间约2.01 s 核对。末段视频83.76—84.72 s 仍是空的固定远景，终点未在可辨认视野内，越线姿态依评分器成功判定。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：视频83.76—84.72 s 的固定远景不能看见机器人双脚及终点，未提供状态日志，无法直接辨认终点步骤的实际接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "gates",
+   "unit": "count",
+   "threshold": 32,
+   "direction": "higher_better",
+   "value": 32.0
+  },
+  {
+   "name": "speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.1,
+   "direction": "higher_better",
+   "value": 4.375
+  },
+  {
+   "name": "time_s",
+   "unit": "s",
+   "threshold": null,
+   "direction": "lower_better",
+   "value": 82.709
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人自行起滑并连续左右回转，评分器确认依次通过全部32道旗门、越过终点，33/33个目标节点最终达成，且 fell=0；最终交付文件的官方回放在8472个控制步后成功（log step 65），对应视频约84.72 s。早段过门时间由视频位置变化近似定位；固定远景无法分辨后段机器人、四个受检点和终点，无状态日志，第6—32门的达成区间及后段实际接触保留未知，成功参考录像具有同样的视野限制；会话试跑时间未冒充交付录像时间。",
+ "conf": 6,
+ "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，33/33 节点 final=true 与评分器（stages 34/34、metric_gates=32、replay_success=1、fell=0）一致；第1–5门时间与视频中机器人越过各对旗杆的位置吻合（约±0.5 s），第6–32门及终点区间留 null、接触未知，符合固定远景的限制；84.72 s 由控制步8472×0.01 s 换算并以起滑时间交叉核对，已说明映射。无失败层，无需断点。",
+    "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": "2026-11-12T11:23:43Z",
+  "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-alpine-slalom-i00-privileged--jeydurz/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-alpine-slalom-i00-privileged--jeydurz/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "f7fdabe558ffe4cf35f2d184c67c3af7098f946e784e834d2f7437a409b1566f",
+   "video_sha256": "4bcd5d5f3a5710503a37795fd48341863e36046237deea1bda9c04a7a75a81a3",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/alpine_slalom/trajectories/embodied-olympics-fun-alpine-slalom-i00-standard--8agcflw.json

@@ -0,0 +1,633 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-alpine-slalom-i00-standard--8agcflw",
+ "taskKey": "embodied-olympics-fun/alpine_slalom",
+ "dag_sha256": "0382103c21732bd9a43317e1385d02d1f88b24ff2896d1d6062c6a895ff37a48",
+ "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": 961,
+   "sim_steps": 5368,
+   "calls": 1177,
+   "rows": 5368,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 4790.0,
+   "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": 0.648,
+   "peak_processes": 58,
+   "peak_disk_gb": 1.04,
+   "peak_cpus": 4.07,
+   "peak_pids": 1074,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "media",
+  "url": "media:embodied-olympics-fun-alpine-slalom-i00-standard--8agcflw.mp4",
+  "duration_s": 9.64,
+  "fps": 25.0,
+  "frames": 241,
+  "sha256": "0175ea073df4f48cf92efae119b893a134c6c67f007518de76ec6820faaaa990"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 33,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "gate_1_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 5.8,
+      "log_step": 91,
+      "text_zh": "观察：视频5.0–6.4 s，机器人以双脚所固定的雪板支承、向黑色起点拱架下坡侧滑出；视频6.4–8.4 s继续向首对旗杆滑行。起滑是首门步骤的一部分，尚不等于通过首门。"
+     },
+     {
+      "t": 9.6,
+      "log_step": 90,
+      "text_zh": "观察：视频8.4–9.6 s逐帧检查，机器人在首门附近转向，末段躯干倾斜下沉，未见双脚和两板尖全部通过首门。log step 89启动最后一次reset后的a6，log step 90记录该次服务裁判outcome=fell、stages_reached=1/34；此原生阶段为起滑，未包含首门。"
+     },
+     {
+      "t": 9.6,
+      "log_step": 91,
+      "text_zh": "推断：log step 91服务状态为episode=6、control_steps=961、resets_left=0；服务control_dt=0.01 s（log step 26），961步对应视频约9.61 s，与视频5.8–6.4 s离开起点以及视频9.6 s末段失衡核对一致。这是最终评分回合，其他重置和--sim本地替身测试不计进展。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频0–9.6 s机器人双脚固定于雪板，双腿经雪板在起点平台和坡面承重、蹬滑及转向；末段躯干倾斜下沉。画面不足以确定触发摔倒判定的具体身体接触点。"
+     }
+    ]
+   },
+   "gate_2_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 9.6,
+      "log_step": 90,
+      "text_zh": "推断：视频0–9.6 s机器人只从起点滑至首门附近，末段失衡；log step 90该回合服务裁判为fell、stages_reached=1/34（仅起滑阶段），首门未通过。第2门的顺序前置未满足，且录像无继续驶向该门的过程，故第2门未达成、未尝试。"
+     }
+    ]
+   },
+   "gate_3_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 9.6,
+      "log_step": 90,
+      "text_zh": "推断：视频0–9.6 s机器人只从起点滑至首门附近，末段失衡；log step 90该回合服务裁判为fell、stages_reached=1/34（仅起滑阶段），首门未通过。第3门的顺序前置未满足，且录像无继续驶向该门的过程，故第3门未达成、未尝试。"
+     }
+    ]
+   },
+   "gate_4_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 9.6,
+      "log_step": 90,
+      "text_zh": "推断：视频0–9.6 s机器人只从起点滑至首门附近，末段失衡；log step 90该回合服务裁判为fell、stages_reached=1/34（仅起滑阶段），首门未通过。第4门的顺序前置未满足，且录像无继续驶向该门的过程，故第4门未达成、未尝试。"
+     }
+    ]
+   },
+   "gate_5_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 9.6,
+      "log_step": 90,
+      "text_zh": "推断：视频0–9.6 s机器人只从起点滑至首门附近，末段失衡；log step 90该回合服务裁判为fell、stages_reached=1/34（仅起滑阶段），首门未通过。第5门的顺序前置未满足，且录像无继续驶向该门的过程，故第5门未达成、未尝试。"
+     }
+    ]
+   },
+   "gate_6_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 9.6,
+      "log_step": 90,
+      "text_zh": "推断：视频0–9.6 s机器人只从起点滑至首门附近，末段失衡；log step 90该回合服务裁判为fell、stages_reached=1/34（仅起滑阶段），首门未通过。第6门的顺序前置未满足，且录像无继续驶向该门的过程，故第6门未达成、未尝试。"
+     }
+    ]
+   },
+   "gate_7_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 9.6,
+      "log_step": 90,
+      "text_zh": "推断：视频0–9.6 s机器人只从起点滑至首门附近，末段失衡；log step 90该回合服务裁判为fell、stages_reached=1/34（仅起滑阶段），首门未通过。第7门的顺序前置未满足，且录像无继续驶向该门的过程，故第7门未达成、未尝试。"
+     }
+    ]
+   },
+   "gate_8_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 9.6,
+      "log_step": 90,
+      "text_zh": "推断：视频0–9.6 s机器人只从起点滑至首门附近，末段失衡；log step 90该回合服务裁判为fell、stages_reached=1/34（仅起滑阶段），首门未通过。第8门的顺序前置未满足，且录像无继续驶向该门的过程，故第8门未达成、未尝试。"
+     }
+    ]
+   },
+   "gate_9_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 9.6,
+      "log_step": 90,
+      "text_zh": "推断：视频0–9.6 s机器人只从起点滑至首门附近，末段失衡；log step 90该回合服务裁判为fell、stages_reached=1/34（仅起滑阶段），首门未通过。第9门的顺序前置未满足，且录像无继续驶向该门的过程，故第9门未达成、未尝试。"
+     }
+    ]
+   },
+   "gate_10_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 9.6,
+      "log_step": 90,
+      "text_zh": "推断：视频0–9.6 s机器人只从起点滑至首门附近，末段失衡；log step 90该回合服务裁判为fell、stages_reached=1/34（仅起滑阶段），首门未通过。第10门的顺序前置未满足，且录像无继续驶向该门的过程，故第10门未达成、未尝试。"
+     }
+    ]
+   },
+   "gate_11_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 9.6,
+      "log_step": 90,
+      "text_zh": "推断：视频0–9.6 s机器人只从起点滑至首门附近，末段失衡；log step 90该回合服务裁判为fell、stages_reached=1/34（仅起滑阶段），首门未通过。第11门的顺序前置未满足，且录像无继续驶向该门的过程，故第11门未达成、未尝试。"
+     }
+    ]
+   },
+   "gate_12_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 9.6,
+      "log_step": 90,
+      "text_zh": "推断：视频0–9.6 s机器人只从起点滑至首门附近，末段失衡；log step 90该回合服务裁判为fell、stages_reached=1/34（仅起滑阶段），首门未通过。第12门的顺序前置未满足，且录像无继续驶向该门的过程，故第12门未达成、未尝试。"
+     }
+    ]
+   },
+   "gate_13_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 9.6,
+      "log_step": 90,
+      "text_zh": "推断：视频0–9.6 s机器人只从起点滑至首门附近，末段失衡；log step 90该回合服务裁判为fell、stages_reached=1/34（仅起滑阶段），首门未通过。第13门的顺序前置未满足，且录像无继续驶向该门的过程，故第13门未达成、未尝试。"
+     }
+    ]
+   },
+   "gate_14_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 9.6,
+      "log_step": 90,
+      "text_zh": "推断：视频0–9.6 s机器人只从起点滑至首门附近，末段失衡；log step 90该回合服务裁判为fell、stages_reached=1/34（仅起滑阶段），首门未通过。第14门的顺序前置未满足，且录像无继续驶向该门的过程，故第14门未达成、未尝试。"
+     }
+    ]
+   },
+   "gate_15_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 9.6,
+      "log_step": 90,
+      "text_zh": "推断：视频0–9.6 s机器人只从起点滑至首门附近，末段失衡；log step 90该���合服务裁判为fell、stages_reached=1/34（仅起滑阶段），首门未通过。第15门的顺序前置未满足，且录像无继续驶向该门的过程，故第15门未达成、未尝试。"
+     }
+    ]
+   },
+   "gate_16_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 9.6,
+      "log_step": 90,
+      "text_zh": "推断：视频0–9.6 s机器人只从起点滑至首门附近，末段失衡；log step 90该回合服务裁判为fell、stages_reached=1/34（仅起滑阶段），首门未通过。第16门的顺序前置未满足，且录像无继续驶向该门的过程，故第16门未达成、未尝试。"
+     }
+    ]
+   },
+   "gate_17_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 9.6,
+      "log_step": 90,
+      "text_zh": "推断：视频0–9.6 s机器人只从起点滑至首门附近，末段失衡；log step 90该回合服务裁判为fell、stages_reached=1/34（仅起滑阶段），首门未通过。第17门的顺序前置未满足，且录像无继续驶向该门的过程，故第17门未达成、未尝试。"
+     }
+    ]
+   },
+   "gate_18_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 9.6,
+      "log_step": 90,
+      "text_zh": "推断：视频0–9.6 s机器人只从起点滑至首门附近，末段失衡；log step 90该回合服务裁判为fell、stages_reached=1/34（仅起滑阶段），首门未通过。第18门的顺序前置未满足，且录像无继续驶向该门的过程，故第18门未达成、未尝试。"
+     }
+    ]
+   },
+   "gate_19_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 9.6,
+      "log_step": 90,
+      "text_zh": "推断：视频0–9.6 s机器人只从起点滑至首门附近，末段失衡；log step 90该回合服务裁判为fell、stages_reached=1/34（仅起滑阶段），首门未通过。第19门的顺序前置未满足，且录像无继续驶向该门的过程，故第19门未达成、未尝试。"
+     }
+    ]
+   },
+   "gate_20_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 9.6,
+      "log_step": 90,
+      "text_zh": "推断：视频0–9.6 s机器人只从起点滑至首门附近，末段失衡；log step 90该回合服务裁判为fell、stages_reached=1/34（仅起滑阶段），首门未通过。第20门的顺序前置未满足，且录像无继续驶向该门的过程，故第20门未达成、未尝试。"
+     }
+    ]
+   },
+   "gate_21_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 9.6,
+      "log_step": 90,
+      "text_zh": "推断：视频0–9.6 s机器人只从起点滑至首门附近，末段失衡；log step 90该回合服务裁判为fell、stages_reached=1/34（仅起滑阶段），首门未通过。第21门的顺序前置未满足，且录像无继续驶向该门的过程，故第21门未达成、未尝试。"
+     }
+    ]
+   },
+   "gate_22_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 9.6,
+      "log_step": 90,
+      "text_zh": "推断：视频0–9.6 s机器人只从起点滑至首门附近，末段失衡；log step 90该回合服务裁判为fell、stages_reached=1/34（仅起滑阶段），首门未通过。第22门的顺序前置未满足，且录像无继续驶向该门的过程，故第22门未达成、未尝试。"
+     }
+    ]
+   },
+   "gate_23_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 9.6,
+      "log_step": 90,
+      "text_zh": "推断：视频0–9.6 s机器人只从起点滑至首门附近，末段失衡；log step 90该回合服务裁判为fell、stages_reached=1/34（仅起滑阶段），首门未通过。第23门的顺序前置未满足，且录像无继续驶向该门的过程，故第23门未达成、未尝试。"
+     }
+    ]
+   },
+   "gate_24_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 9.6,
+      "log_step": 90,
+      "text_zh": "推断：视频0–9.6 s机器人只从起点滑至首门附近，末段失衡；log step 90该回合服务裁判为fell、stages_reached=1/34（仅起滑阶段），首门未通过。第24门的顺序前置未满足，且录像无继续驶向该门的过程，故第24门未达成、未尝试。"
+     }
+    ]
+   },
+   "gate_25_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 9.6,
+      "log_step": 90,
+      "text_zh": "推断：视频0–9.6 s机器人只从起点滑至首门附近，末段失衡；log step 90该回合服务裁判为fell、stages_reached=1/34（仅起滑阶段），首门未通过。第25门的顺序前置未满足，且录像无继��驶向该门的过程，故第25门未达成、未尝试。"
+     }
+    ]
+   },
+   "gate_26_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 9.6,
+      "log_step": 90,
+      "text_zh": "推断：视频0–9.6 s机器人只从起点滑至首门附近，末段失衡；log step 90该回合服务裁判为fell、stages_reached=1/34（仅起滑阶段），首门未通过。第26门的顺序前置未满足，且录像无继续驶向该门的过程，故第26门未达成、未尝试。"
+     }
+    ]
+   },
+   "gate_27_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 9.6,
+      "log_step": 90,
+      "text_zh": "推断：视频0–9.6 s机器人只从起点滑至首门附近，末段失衡；log step 90该回合服务裁判为fell、stages_reached=1/34（仅起滑阶段），首门未通过。第27门的顺序前置未满足，且录像无继续驶向该门的过程，故第27门未达成、未尝试。"
+     }
+    ]
+   },
+   "gate_28_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 9.6,
+      "log_step": 90,
+      "text_zh": "推断：视频0–9.6 s机器人只从起点滑至首门附近，末段失衡；log step 90该回合服务裁判为fell、stages_reached=1/34（仅起滑阶段），首门未通过。第28门的顺序前置未满足，且录像无继续驶向该门的过程，故第28门未达成、未尝试。"
+     }
+    ]
+   },
+   "gate_29_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 9.6,
+      "log_step": 90,
+      "text_zh": "推断：视频0–9.6 s机器人只从起点滑至首门附近，末段失衡；log step 90该回合服务裁判为fell、stages_reached=1/34（仅起滑阶段），首门未通过。第29门的顺序前置未满足，且录像无继续驶向该门的过程，故第29门未达成、未尝试。"
+     }
+    ]
+   },
+   "gate_30_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 9.6,
+      "log_step": 90,
+      "text_zh": "推断：视频0–9.6 s机器人只从起点滑至首门附近，末段失衡；log step 90该回合服务裁判为fell、stages_reached=1/34（仅起滑阶段），首门未通过。第30门的顺序前置未满足，且录像无继续驶向该门的过程，故第30门未达成、未尝试。"
+     }
+    ]
+   },
+   "gate_31_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 9.6,
+      "log_step": 90,
+      "text_zh": "推断：视频0–9.6 s机器人只从起点滑至首门附近，末段失衡；log step 90该回合服务裁判为fell、stages_reached=1/34（仅起滑阶段），首门未通过。第31门的顺序前置未满足，且录像无继续驶向该门的过程，故第31门未达成、未尝试。"
+     }
+    ]
+   },
+   "gate_32_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 9.6,
+      "log_step": 90,
+      "text_zh": "推断：视频0–9.6 s机器人只从起点滑至首门附近，末段失衡；log step 90该回合服务裁判为fell、stages_reached=1/34（仅起滑阶段），首门未通过。第32门的顺序前置未满足，且录像无继续驶向该门的过程，故第32门未达成、未尝试。"
+     }
+    ]
+   },
+   "finish_crossed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 9.6,
+      "log_step": 90,
+      "text_zh": "推断：视频0–9.6 s机器人仅从起点滑至首门附近，末段失衡；log step 90该回合裁判返回fell、仅1/34原生阶段。32门均未完成，机器人未到达终点，故越过终点未达成；本地替身测试不属于评分录像。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "gates",
+   "unit": "count",
+   "threshold": 32,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.1,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "time_s",
+   "unit": "s",
+   "threshold": null,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "gate_1_passed",
+   "t": 9.4,
+   "log_step": 90,
+   "where": "操作中",
+   "evidence_zh": "观察：视频8.4–9.6 s机器人仍在首门附近滑行转向，视频9.4–9.6 s躯干明显倾斜下沉；log step 89启动最终a6回合，log step 90该回合的服务裁判返回fell、stages_reached=1/34。推断：机器人在通过首门的行进过程中失衡，首门从未完成，亦无后续修复。"
+  },
+  "unmet": {
+   "gate_1_passed": false,
+   "gate_2_passed": false,
+   "gate_3_passed": false,
+   "gate_4_passed": false,
+   "gate_5_passed": false,
+   "gate_6_passed": false,
+   "gate_7_passed": false,
+   "gate_8_passed": false,
+   "gate_9_passed": false,
+   "gate_10_passed": false,
+   "gate_11_passed": false,
+   "gate_12_passed": false,
+   "gate_13_passed": false,
+   "gate_14_passed": false,
+   "gate_15_passed": false,
+   "gate_16_passed": false,
+   "gate_17_passed": false,
+   "gate_18_passed": false,
+   "gate_19_passed": false,
+   "gate_20_passed": false,
+   "gate_21_passed": false,
+   "gate_22_passed": false,
+   "gate_23_passed": false,
+   "gate_24_passed": false,
+   "gate_25_passed": false,
+   "gate_26_passed": false,
+   "gate_27_passed": false,
+   "gate_28_passed": false,
+   "gate_29_passed": false,
+   "gate_30_passed": false,
+   "gate_31_passed": false,
+   "gate_32_passed": false,
+   "finish_crossed": false
+  },
+  "phenomena": [
+   {
+    "cat": "全身",
+    "item": "平衡",
+    "t": 9.4,
+    "log_step": 90,
+    "evidence_zh": "观察：视频9.0–9.6 s逐帧放大可见机器人随转向改变躯干姿态，视频9.4–9.6 s上身倾斜下沉、原有直立滑行姿态破坏；log step 90最终回合的服务裁判outcome=fell，回合结束。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "evidence_zh": "推断：agent脚本在log step 87配置双腿逆解、犁式制动与左右板刃差分转向，并在log step 89将此脚本用于最终实机服务回合。log step 91该次脚本输出的骨盆IMU航向psi由仿真8.59 s的1.25 rad升至仿真9.09 s的1.72 rad、仿真9.59 s的2.98 rad，转向输入u同时由-0.02降至-1.00；结合视频9.0–9.6 s旋转并倾斜下沉及log step 90服务裁判fell，控制未能稳定滑行和转向。现有证据只支持控制层，不能确定某个增益、板刃接触布局或逆解是具体原因。",
+    "source_zh": "日志",
+    "log_step": 91,
+    "t": 9.4
+   }
+  ],
+  "gap_zh": "没有逐帧状态日志；固定远景中机器人及雪板较小，无法直接量出四个受检点的过门坐标，也无法确认触发fell的具体身体接触点，首门未达成由画面与最终回合服务裁判阶段合判。日志中的位置、速度及转向输入为agent脚本的估计或计算，不能据此细分失稳机制。run.json的reward未报告gates、speed_m_s、time_s原生指标数值，故这三项保留null。"
+ },
+ "summary_zh": "机器人在最终评分回合自行起滑并向首门滑行，但在视频约9.4–9.6 s转向失衡，服务裁判判fell；32道旗门及终点均未达成，最终进度0/33。断点为通过第1门，现象为全身平衡失败，日志与画面支持控制未能稳定转向，不能确定具体控制子因。没有状态日志，固定远景无法直接核实精细过门几何及摔倒接触点；原生1/34阶段仅为起滑，不是通过了一门。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "首门未过、32门及终点未达成、断点gate_1、现象全身平衡与控制根因，均与视频8.4–9.6 s帧及961步≈9.61 s的核对一致；固定远景中机器人很小，细节无法进一步核实，未发现需要改动的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-11-12T11:22:10Z",
+  "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-alpine-slalom-i00-standard--8agcflw/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-alpine-slalom-i00-standard--8agcflw/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "d5a5209c194d013d93078fed1acf64bf84750afb2ec72aa7ab5f5ce2da0c6da7",
+   "video_sha256": "0175ea073df4f48cf92efae119b893a134c6c67f007518de76ec6820faaaa990",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/backflip/trajectories/embodied-olympics-fun-backflip-i00-privileged--hjqfxmm.json

@@ -0,0 +1,190 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-backflip-i00-privileged--hjqfxmm",
+ "taskKey": "embodied-olympics-fun/backflip",
+ "dag_sha256": "a2dce89dd7f0391d67651e778f82fd48e6fb3121b94be3c140496cb25fc4241d",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "determinism_checked": 0,
+   "replays": 1,
+   "n_actions": 1990,
+   "sim_steps": 464,
+   "horizon": 2000,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 4,
+   "stages_total": 4,
+   "fell": 0,
+   "replay_s": 3.17,
+   "bad_trajectory": 0,
+   "metric_air_s": 0.689,
+   "metric_turn_deg": 306.7,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 12,
+   "agent_resets": 0,
+   "agent_replays": 16,
+   "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": 7.914,
+   "peak_processes": 27,
+   "peak_disk_gb": 0.257,
+   "peak_cpus": 8.06,
+   "peak_pids": 50,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "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": "media",
+  "url": "media:embodied-olympics-fun-backflip-i00-privileged--hjqfxmm.mp4",
+  "duration_s": 4.68,
+  "fps": 25.0,
+  "frames": 117,
+  "sha256": "eae02f26060a0692b4a41357bccaa80d465892015c0bf4b07445a68bd17004b5"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "backflip_landed_stable": {
+    "method": "visual",
+    "intervals": [
+     [
+      4.64,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.96,
+      "log_step": null,
+      "text_zh": "观察：视频0.72–0.92 s，机器人屈膝后伸腿、躯干后仰，起跳前可见支撑在脚部；视频0.96–1.56 s机器人整体离地，头向后、脚向前翻过倒立位置，再展开双腿准备落地。开局站立尚未完成后空翻，本节点不从视频0 s计入。"
+     },
+     {
+      "t": 1.64,
+      "log_step": null,
+      "text_zh": "观察：逐帧检查视频1.56、1.60、1.64、1.68 s，机器人双脚下降并接地承重，手、膝和躯干仍离地；随后屈膝缓冲、调整躯干，没有倒地。"
+     },
+     {
+      "t": 4.64,
+      "log_step": null,
+      "text_zh": "观察：视频2.00–2.56 s机器人继续摆臂和调整脚步，视频2.64–4.64 s躯干基本直立、位置变化很小；已逐帧检查视频3.60–4.64 s末段，机器人始终以脚支撑，没有手、膝或躯干撑地。评分器reward报告success=1、replay_success=1、stages_reached=4/4、fell=0，metric_turn_deg=306.7、metric_air_s=0.689。"
+     },
+     {
+      "t": 4.64,
+      "log_step": 131,
+      "text_zh": "推断：log step 131对最终交付文件进行独立CLI回放，回放器报告执行464个控制步、仿真4.64 s成功，落地事件在仿真1.615 s，站稳事件在仿真4.64 s；log step 138再次生成并确认与交付动作数组逐项一致。按每控制步0.01 s，464×0.01对应视频约4.64 s；落地事件已与视频1.60–1.64 s的脚接地核对，且执行时长与视频末帧4.64 s一致。因此将完整后翻及连续站稳验收的节点首次达成锚定在视频4.64 s，method仍为visual。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频0.72–0.92 s机器人双脚在蓝色场地上支撑并发力起跳；视频0.96–1.56 s腾空阶段与地面分离；视频约1.60–1.64 s双脚落地承重，随后脚部支撑、调整步态并站稳至视频4.64 s。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "turn_deg",
+   "unit": "deg",
+   "threshold": 300.0,
+   "direction": "higher_better",
+   "value": 306.7
+  },
+  {
+   "name": "air_s",
+   "unit": "s",
+   "threshold": 0.2,
+   "direction": "higher_better",
+   "value": 0.689
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人完成唯一节点“后空翻并落地站稳”：视频约0.96 s已腾空后翻，视频约1.60–1.64 s脚先落地，调整姿态后保持直立，按交付回放与裁判结果在视频约4.64 s完成站稳验收，最终1/1。评分器记录后转306.7°、腾空0.689秒、四阶段全部完成且未摔倒，与录像动作一致，无失败断点。缺少状态日志，25帧/秒侧视录像不能独立核实逐采样接触、骨盆速度及精确站稳计数，相关数值验收依裁判记录，参考��功录像仅用于对照动作外观。",
+ "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，唯一节点在视频4.64 s达成，与评分器464控制步、四阶段全部完成、未摔倒一致；视频帧可见脚起跳、腾空后翻、脚先落地并保持直立，无失败断点，标注无误。",
+    "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": "2026-11-12T11:21:59Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-backflip-i00-privileged--hjqfxmm/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-backflip-i00-privileged--hjqfxmm/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "7b43d46898f23040b8975a3142567571ffb499ebdaa8ad36a8ccbf09b9b0fe4f",
+   "video_sha256": "eae02f26060a0692b4a41357bccaa80d465892015c0bf4b07445a68bd17004b5",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/backflip/trajectories/embodied-olympics-fun-backflip-i00-standard--ksyxepy.json

@@ -0,0 +1,194 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-backflip-i00-standard--ksyxepy",
+ "taskKey": "embodied-olympics-fun/backflip",
+ "dag_sha256": "a2dce89dd7f0391d67651e778f82fd48e6fb3121b94be3c140496cb25fc4241d",
+ "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": 397,
+   "sim_steps": 956,
+   "calls": 962,
+   "rows": 956,
+   "resets": 2,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 2858.1,
+   "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": 5.82,
+   "peak_processes": 15,
+   "peak_disk_gb": 0.557,
+   "peak_cpus": 4.01,
+   "peak_pids": 143,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "media",
+  "url": "media:embodied-olympics-fun-backflip-i00-standard--ksyxepy.mp4",
+  "duration_s": 4.0,
+  "fps": 25.0,
+  "frames": 100,
+  "sha256": "28a4b716d1a3b3d8ed365d17395e6e81d815240d2b07c12356805574fa766e08"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "backflip_landed_stable": {
+    "method": "visual",
+    "intervals": [
+     [
+      3.96,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.76,
+      "log_step": null,
+      "text_zh": "观察：视频0.56–0.72 s机器人以双脚支撑屈膝、摆臂，视频0.76–0.84 s伸腿蹬地并向后仰；视频0.88 s可清楚看到双脚离地。"
+     },
+     {
+      "t": 1.04,
+      "log_step": null,
+      "text_zh": "观察：视频0.96–1.20 s机器人头向后、脚向前翻过倒立位置，全身与蓝色地面分离；视频1.36 s双脚仍在地面上方，视频约1.40 s脚先接地，视频1.44–1.60 s屈膝承重，未见手、膝或躯干撑地。"
+     },
+     {
+      "t": 3.96,
+      "log_step": null,
+      "text_zh": "观察：视频1.56–2.36 s机器人逐渐伸腿直身；视频3.00–3.96 s逐帧检查均见机器人保持近竖直站立、双脚在落地点附近，未倒地。"
+     },
+     {
+      "t": 3.96,
+      "log_step": 61,
+      "text_zh": "观察：log step 61记录agent重置后运行最终第3回合，机器人服务返回reason=success、outcome=success、stages_reached=4/4、control_steps=397、sim_time_s=3.97；log step 62再次读取服务状态仍为第3回合成功。评分记录live_success=1、graded_by_live=1、finalized=1。"
+     },
+     {
+      "t": 3.96,
+      "log_step": 61,
+      "text_zh": "推断：log step 3服务规格的control_dt=0.01 s，最终第3回合397个控制步对应仿真3.97 s，与4.00 s录像及视频3.96 s末帧的站稳终态吻合；结合服务已验收4/4阶段，将完整后翻并站稳的首次达成锚定在视频约3.96 s（末帧精度）。起跳和落地只是这一节点的动作阶段，不能提前计作完成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：机器人双脚与蓝色场地地面接触，承担视频0.56–0.84 s的蹬地起跳和视频约1.40 s之后的落地、站立支撑；腾空期间无地面接触，手臂仅摆动，未见手或膝撑地。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "turn_deg",
+   "unit": "deg",
+   "threshold": 300.0,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "air_s",
+   "unit": "s",
+   "threshold": 0.2,
+   "direction": "higher_better",
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "受评的最终第3回合中，机器人在视频约0.84–1.40 s完成腾空后翻并双脚先落地，随后直身站稳至视频3.96 s末帧，完成1/1节点。log step 61的机器人服务及评分记录均确认该回合成功，节点完整验收时间取视频约3.96 s；此前调试回合的摔倒不属于本段受评视频。没有状态日志，25帧/秒录像不能独立核验采样级接触、精确转角和骨盆速度；评分器未报告turn_deg、air_s，故两项原生指标留空。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "标注与视频一致：约0.8–1.4 s腾空后翻、脚先落地，之后直身站稳到末帧；成功由评分记录确认，节点final=true，无失败层，契约校验通过。",
+    "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": "2026-11-12T11:22:08Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-backflip-i00-standard--ksyxepy/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-backflip-i00-standard--ksyxepy/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "a0448bad393fa17ac0e84f78696c1923bf5f167ce161141d4dfc4847e6c1f2d9",
+   "video_sha256": "28a4b716d1a3b3d8ed365d17395e6e81d815240d2b07c12356805574fa766e08",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/badminton_serve/trajectories/embodied-olympics-fun-badminton-serve-i00-privileged--7ezfgdq.json

@@ -0,0 +1,225 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-badminton-serve-i00-privileged--7ezfgdq",
+ "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,
+   "determinism_checked": 0,
+   "replays": 1,
+   "n_actions": 427,
+   "sim_steps": 427,
+   "horizon": 1500,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 4,
+   "stages_total": 4,
+   "fell": 0,
+   "replay_s": 6.22,
+   "bad_trajectory": 0,
+   "metric_hit_height_m": 0.6476,
+   "metric_landing_x_m": 7.085,
+   "metric_landing_y_m": 1.678,
+   "metric_shuttle_angle_deg": 35.79,
+   "metric_shuttle_speed_m_s": 14.491,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 914,
+   "agent_resets": 1756,
+   "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": 9.316,
+   "peak_processes": 18,
+   "peak_disk_gb": 0.244,
+   "peak_cpus": 8.0,
+   "peak_pids": 85,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "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": "media",
+  "url": "media:embodied-olympics-fun-badminton-serve-i00-privileged--7ezfgdq.mp4",
+  "duration_s": 4.28,
+  "fps": 25.0,
+  "frames": 107,
+  "sha256": "a356eba39c53c5035a5f3a5e5aa7a0e13a3094362fda9fce7451eea35f1d030f"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "serve_in_target": {
+    "method": "visual",
+    "intervals": [
+     [
+      4.24,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 2.64,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–2.64 s，机器人双脚支撑地面，右手持拍由身前调整到低位，左臂将白色羽毛球移到身前；视频 2.64–2.88 s 白色球离开左手并下落，右手随后挥拍。远景不能分辨手指与羽毛的精细接触。"
+     },
+     {
+      "t": 3.0,
+      "log_step": null,
+      "text_zh": "观察：视频 2.88–2.96 s，机器人右手球拍由低位快速向前上方挥动；视频 3.00 s 白色小球已与球拍分离、朝球网方向上升，视频 3.08–3.12 s 可见小球在网顶附近向远侧飞行。视频 3.24–4.24 s 机器人仍站立持拍，球过小且远侧场地受球网遮挡，不能直接辨认首次触地点。"
+     },
+     {
+      "t": 4.24,
+      "log_step": 88,
+      "text_zh": "观察：inputs/run.json 的评分器字段 success=1、replay_success=1、stages_reached=4/4、fell=0，首次落点指标为 x=7.085 m、y=1.678 m，均在目标区间内；log step 79 记录最终候选已保存为交付轨迹并成功回放，log step 88 对同一交付文件的官方回放输出 success=True、failed=None，阶段依次为 release、hit、over the net、in，执行 427 行，指标与本次评分一致。"
+     },
+     {
+      "t": 4.24,
+      "log_step": 88,
+      "text_zh": "推断：本次评分与已保存文件的官方回放结果一致，视频中的放球、挥拍和向远侧飞行也与之相符，故合法发球入区节点完成并锁存。完成锚点近似置于视频末帧 4.24 s（视频总长 4.28 s），并非直接拍到触地的精确时刻；毫秒级球托初触、脚步阈值及击球速度等验收依据裁判成功结果确认。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手持工具",
+      "how": "其他",
+      "note_zh": "观察：视频 2.88–3.00 s 机器人右手握持球拍向前上方挥动，白色羽毛球随即离拍飞出；候选：击打，现有接触方式没有对应取值，不能用推/扫或压/擦替代。左手此前持球并先放开，再由右手球拍击球；握拍链路不重复标注。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–4.24 s 机器人双脚在近侧发球区支撑地面，准备和挥拍时保持站立，未见迈步或倒地；远景不能量出脚部毫米级位移与短暂离地间隙。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "hit_height_m",
+   "unit": "m",
+   "threshold": 1.15,
+   "direction": "lower_better",
+   "value": 0.6476
+  },
+  {
+   "name": "shuttle_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": 14.491
+  },
+  {
+   "name": "shuttle_angle_deg",
+   "unit": "deg",
+   "threshold": 0,
+   "direction": "higher_better",
+   "value": 35.79
+  },
+  {
+   "name": "landing_x_m",
+   "unit": "m",
+   "threshold": [
+    4.43,
+    9.15
+   ],
+   "direction": "target",
+   "value": 7.085
+  },
+  {
+   "name": "landing_y_m",
+   "unit": "m",
+   "threshold": [
+    0.43,
+    3.04
+   ],
+   "direction": "target",
+   "value": 1.678
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人站立完成左手放球、右手低位挥拍，羽毛球向网及远侧飞出；评分器与交付文件的官方回放均确认四个原生阶段通过，合法发球入区节点完成，最终为 1/1。落点指标为 (7.085, 1.678) m，击球高度指标为 0.6476 m；无状态日志，远景不能独立核验首次触地和毫秒级接触条件，完成时刻仅近似锚定视频末帧 4.24 s。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "成功运行，评分器 success=1、4/4 阶段、fell=0；视频可见左手放球、右手挥拍、球向网飞出，节点完成时间锚定末帧 4.24 s 且已说明为近似，contact_actual 合理，无失败标注，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": "2026-11-12T11:24:10Z",
+  "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--7ezfgdq/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-badminton-serve-i00-privileged--7ezfgdq/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "01de051fbb3c4804d4501c5c4ef8ab2878b4a828a218cc00a43defad3faf5e4a",
+   "video_sha256": "a356eba39c53c5035a5f3a5e5aa7a0e13a3094362fda9fce7451eea35f1d030f",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/badminton_serve/trajectories/embodied-olympics-fun-badminton-serve-i00-standard--5dxrejy.json

@@ -0,0 +1,216 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-badminton-serve-i00-standard--5dxrejy",
+ "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": 853,
+   "sim_steps": 1892,
+   "calls": 1901,
+   "rows": 1892,
+   "resets": 2,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 2727.9,
+   "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": 2.964,
+   "peak_processes": 13,
+   "peak_disk_gb": 0.421,
+   "peak_cpus": 4.0,
+   "peak_pids": 524,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "media",
+  "url": "media:embodied-olympics-fun-badminton-serve-i00-standard--5dxrejy.mp4",
+  "duration_s": 8.56,
+  "fps": 25.0,
+  "frames": 214,
+  "sha256": "f87d5a9eb21a84bde6ae87fab6a94ee9eb3b9044e9f51bb13e4121bb1de599cd"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "serve_in_target": {
+    "method": "visual",
+    "intervals": [
+     [
+      8.53,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 6.84,
+      "log_step": 94,
+      "text_zh": "观察：视频 6.80–6.96 s，机器人左手前方的白色羽毛球脱离手部并下落，左臂随后上收；视频 7.00–7.08 s，机器人右手持球拍由低处向前上方挥出，羽毛球随后向球网方向升起。视频 1.00–2.16 s、3.00–3.96 s 的较早挥拍没有把球发出，属于本次尝试内的空挥准备。"
+     },
+     {
+      "t": 8.52,
+      "log_step": 94,
+      "text_zh": "观察：视频 7.20–8.08 s 的放大连续帧可见白色羽毛球沿高于网顶的弧线向远侧移动并开始下降；视频 8.16–8.52 s，球在球网及远侧场地背景中难以分辨，机器人仍站立并持拍。log step 94 中，机器人实时控制服务（执行动作并返回原生裁判结果的接口）报告 episode=3、done=true、reason=success、outcome=success、stages_reached=4、stages_total=4；run.json 的 reward 同时给出 graded_by_live=1、live_success=1、finalized=1。"
+     },
+     {
+      "t": 8.53,
+      "log_step": 94,
+      "text_zh": "推断：本次计分为 log step 85 重置后的 episode 3，log step 94 的原生裁判 4/4 与成功结果确认放球→击球→过网→首次落地区内均通过验收。完成时刻取视频约 8.53 s：log step 3/35 给出 control_dt=0.01 s，log step 94 的本局 control_steps=853，对应 853×0.01=8.53 s；log step 91 的 control_steps=667 对应视频约 6.68 s 的持球待发姿态，已与视频 6.64–6.72 s 核对，终止计数也与视频末帧 8.52 s、总长 8.56 s 一致。末帧不能直接解析首次触地点或毫秒级接触细节，合法性依据本局裁判结果，未以 agent 的完成自述代替观察。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手持工具",
+      "how": "其他",
+      "note_zh": "观察：视频 7.00–7.08 s，机器人右手持球拍从低位向前上方挥出，羽毛球随即飞离；结合 log step 94 本局裁判的击球阶段已通过，确认球拍拍面实际击打羽毛球。候选：击打；现有接触方式无击打，故用其他。左手此前持住羽毛球后先放开，见视频 6.80–6.96 s；持拍链路不重复标注。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 6.56–8.52 s，机器人双脚支撑球场地面，在放球、挥拍和等待球落下期间保持站立，未见迈步或摔倒。此处仅描述可见支撑，不用远景估计毫米级脚位移。"
+     }
+    ]
+   }
+  },
+  "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": "第三次实时尝试完成合法发球入区，最终节点为 1/1：视频约 6.8 s 放球、7.0 s 挥拍击出，随后可见羽毛球向远侧飞行，原生裁判在对应视频约 8.53 s 报告成功及阶段 4/4。日志中的前两次失败已在重置前结束，不是本次计分视频的失败，failure 为 null。无状态日志，远景及末帧不能独立核实球托初触、瞬时速度、接触时长和首次触地点，故这些合法性条件由本局裁判结果确认；五项原生指标均未在评分字段中报告，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": "成功运行，单节点 serve_in_target 在视频约 8.53 s 达成，与 log step 94 的 sim_time_s=8.53、control_steps=853、stages 4/4 一致；接触标注合理，failure 为 null，无需修改。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-11-12T11:25:01Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-badminton-serve-i00-standard--5dxrejy/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-badminton-serve-i00-standard--5dxrejy/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "bb66171f0543498034cbd5a412678ed24e876cd812bf9ce6cb5d7bdc5ba42c7c",
+   "video_sha256": "f87d5a9eb21a84bde6ae87fab6a94ee9eb3b9044e9f51bb13e4121bb1de599cd",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/balance_beam/trajectories/embodied-olympics-fun-balance-beam-i00-privileged--bnnk7bs.json

@@ -0,0 +1,190 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-balance-beam-i00-privileged--bnnk7bs",
+ "taskKey": "embodied-olympics-fun/balance_beam",
+ "dag_sha256": "85f86504f4e58cb42766772925fcd26453cfbc2255676f62e21bacd7398ab6ac",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "determinism_checked": 0,
+   "replays": 1,
+   "n_actions": 6995,
+   "sim_steps": 6995,
+   "horizon": 9000,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 3,
+   "stages_total": 3,
+   "fell": 0,
+   "replay_s": 18.69,
+   "bad_trajectory": 0,
+   "metric_distance_m": 4.543,
+   "metric_time_s": 69.95,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 15,
+   "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": 1.173,
+   "peak_processes": 11,
+   "peak_disk_gb": 0.25,
+   "peak_cpus": 5.0,
+   "peak_pids": 27,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "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": "media",
+  "url": "media:embodied-olympics-fun-balance-beam-i00-privileged--bnnk7bs.mp4",
+  "duration_s": 69.96,
+  "fps": 25.0,
+  "frames": 1749,
+  "sha256": "1ecf1727e56698d94b0631b133dcc7a2cec591b30f8f874fa002865fc0afeda4"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "beam_crossed_stable": {
+    "method": "visual",
+    "intervals": [
+     [
+      69.95,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频0.0 s机器人双脚前后排列支撑在梁的左侧近端；视频3.0–66.0 s的全程粗采样显示机器人交替迈腿向右侧黄色远端前进，未见跌落。初态尚未过梁，节点不从视频0.0 s成立。"
+     },
+     {
+      "t": 67.96,
+      "log_step": null,
+      "text_zh": "观察：视频65.0–68.6 s密集采样及视频67.6–68.2 s逐帧图中，机器人完成最后一次迈步，两脚在梁右端前后支撑，躯干接近竖直；视频68.96–69.92 s逐帧图显示其保持该姿势，没有继续向前迈步或跌落。"
+     },
+     {
+      "t": 69.92,
+      "log_step": 11,
+      "text_zh": "观察：log step 11中，官方回放器对已保存的交付轨迹报告success=true、executed_rows=6995、steps=6995，事件依次为仿真38.36 s的half way、仿真67.96 s的far end及仿真69.95 s的stand和success；log step 13重新生成轨迹后文件哈希相同且再次回放成功。评分器字段replay_success=1、stages_reached=3/3、metric_distance_m=4.543、metric_time_s=69.95与这一交付相符。"
+     },
+     {
+      "t": 69.95,
+      "log_step": 11,
+      "text_zh": "推断：交付回放6995个控制步×0.01 s（log step 3的CONTROL_DT及任务DAG给出的控制周期）=69.95 s，与视频69.96 s总长度相符；回放far end事件仿真67.96 s也与视频67.96–68.04 s最后收步进入双足支撑相符，故将该回放事件时钟按1:1映射到视频时钟。结合视频末端持续站姿和裁判stand事件，将节点达成锚点定为视频69.95 s；精确接触点、躯干角度和速度阈值由裁判接受结果支持，非远景像素测量。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频0.0–67.96 s机器人左右脚在窄梁顶面交替支撑并向远端迈步；视频68.0–69.92 s两脚在梁右侧黄色远端前后支撑，保持直立。双手位于身体两侧，未见手或躯干接触梁。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "distance_m",
+   "unit": "m",
+   "threshold": 4.5,
+   "direction": "higher_better",
+   "value": 4.543
+  },
+  {
+   "name": "time_s",
+   "unit": "s",
+   "threshold": 90.0,
+   "direction": "lower_better",
+   "value": 69.95
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人从近端交替迈步走过平衡木，末段双脚在黄色远端支撑并站稳，唯一节点于视频约69.95 s达成，最终1/1，failure为null。评分器确认三个阶段均完成，原生距离4.543米、成功仿真时刻69.95 s；日志中的早期摔倒及加噪试跑并非交付回放。没有状态日志，远景录像不能独立精测全部足梁接触点、躯干角度及骨盆速度，严格阈值与站稳窗口由���式裁判接受结果补足。",
+ "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，唯一节点于视频约69.95 s达成，final=1，failure为null，与视频末帧（机器人在梁右端直立）及评分器字段（stages 3/3、success=1）一致；时钟换算已说明并标为推断，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": "2026-11-12T11:21:58Z",
+  "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-balance-beam-i00-privileged--bnnk7bs/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-balance-beam-i00-privileged--bnnk7bs/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "aaa7fac9cb75f69c519b2c5da509e5621515e501802351077e160ef8c7aec5c3",
+   "video_sha256": "1ecf1727e56698d94b0631b133dcc7a2cec591b30f8f874fa002865fc0afeda4",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/balance_beam/trajectories/embodied-olympics-fun-balance-beam-i00-standard--qgs5gc6.json

@@ -0,0 +1,184 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-balance-beam-i00-standard--qgs5gc6",
+ "taskKey": "embodied-olympics-fun/balance_beam",
+ "dag_sha256": "85f86504f4e58cb42766772925fcd26453cfbc2255676f62e21bacd7398ab6ac",
+ "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": 7513,
+   "sim_steps": 8509,
+   "calls": 4169,
+   "rows": 8309,
+   "resets": 2,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 2055.6,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 3,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 2.453,
+   "peak_processes": 19,
+   "peak_disk_gb": 0.319,
+   "peak_cpus": 4.0,
+   "peak_pids": 408,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "media",
+  "url": "media:embodied-olympics-fun-balance-beam-i00-standard--qgs5gc6.mp4",
+  "duration_s": 75.16,
+  "fps": 25.0,
+  "frames": 1879,
+  "sha256": "9e7edb45f2adad41ef1abdbaff901790f677d11eea66b2e5f2592012158be50e"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "beam_crossed_stable": {
+    "method": "visual",
+    "intervals": [
+     [
+      75.1,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 机器人双脚串列站在画面左侧近端，尚未到达右侧黄色远端；视频 0–75 s 的每5秒画面及视频 0–8 s、36–44 s 的每秒画面显示机器人沿梁逐步前进，脚底交替支撑梁面。"
+     },
+     {
+      "t": 72.8,
+      "log_step": null,
+      "text_zh": "观察：视频 70.0–75.0 s 每0.2秒放大画面中，机器人将最后一只摆动脚放到黄色远端附近；视频约 72.8 s 起双脚均落在梁面，随后重心前移，视频 74.2–75.12 s 双脚位置基本不变、身体近乎直立。另已查看成功参考视频 60.0–63.6 s 的末端双足站姿，用于辨认验收区域。"
+     },
+     {
+      "t": 75.1,
+      "log_step": 55,
+      "text_zh": "推断：log step 55 的机器人服务裁判返回 episode=3、outcome=success、stages_reached=3/3、control_steps=7513、sim_time_s=75.13；log step 28 报告控制周期0.01秒，因此本评分回合终结为7513×0.01=仿真75.13 s，与视频75.12 s末帧的远端双足站姿及75.16 s总长核对一致，按约视频75.1 s标记完整站稳窗口首次通过。远景不能精测足梁接触点、躯干角度及骨盆速度边界，连续窗口通过依据服务裁判结果；节点保持至末帧。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频0–75.12 s，机器人左右脚在窄梁顶面交替承重行进，末尾双脚在黄色远端支撑身体；手臂未用于扶梁。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "distance_m",
+   "unit": "m",
+   "threshold": 4.5,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "time_s",
+   "unit": "s",
+   "threshold": 90.0,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "评分回合为最后实跑的episode 3：机器人从近端沿梁前行，在黄色远端双足支撑，约视频75.1 s通过连续站稳验收，完成1/1节点（log step 55服务裁判success、3/3阶段）。此前重置前的跌落不属于本段评分录像，本次无失败断点。无状态日志，远景无法精测接触点、角度和速度阈值，验收窗口由服务判定补足；reward未单独报告distance_m、time_s，故两项原生指标留空。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "评分回合视频0–75.12 s机器人沿梁前行并在黄色远端站定，节点beam_crossed_stable终值true、约视频75.1 s（控制步7513×0.01与视频总长核对）、接触腿/躯干支撑均合理；无失败断点，未发现需改之处。",
+    "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": "2026-11-12T11:24:07Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-balance-beam-i00-standard--qgs5gc6/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-balance-beam-i00-standard--qgs5gc6/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "35189010fd9e440c6fd6633dcf4bafafa6de556d554184831429ee9f3982b9af",
+   "video_sha256": "9e7edb45f2adad41ef1abdbaff901790f677d11eea66b2e5f2592012158be50e",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/balance_board/trajectories/embodied-olympics-fun-balance-board-i00-privileged--dzxnkah.json

@@ -0,0 +1,190 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-balance-board-i00-privileged--dzxnkah",
+ "taskKey": "embodied-olympics-fun/balance_board",
+ "dag_sha256": "33a43b602711eb1f0ce8accf150479b69fb8b84456d1d75783d478a23c0f1d05",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "determinism_checked": 0,
+   "replays": 1,
+   "n_actions": 3000,
+   "sim_steps": 3000,
+   "horizon": 3050,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 3,
+   "stages_total": 3,
+   "fell": 0,
+   "replay_s": 10.41,
+   "bad_trajectory": 0,
+   "metric_max_tilt_deg": 6.48,
+   "metric_time_on_board_s": 30.0,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 35,
+   "agent_resets": 92,
+   "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": 2.299,
+   "peak_processes": 17,
+   "peak_disk_gb": 0.25,
+   "peak_cpus": 8.0,
+   "peak_pids": 30,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "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": "media",
+  "url": "media:embodied-olympics-fun-balance-board-i00-privileged--dzxnkah.mp4",
+  "duration_s": 30.04,
+  "fps": 25.0,
+  "frames": 751,
+  "sha256": "161549273ab063779a95f2da45a902e0cbcc63904521a08ca3c578d9307e99bd"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "balanced_30s": {
+    "method": "visual",
+    "intervals": [
+     [
+      30.0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频0.00、0.20、0.40、0.60、0.80、1.00、1.20、1.40、1.60、1.80、2.00秒，机器人开局已有双脚踩在棕色板面上，随后屈膝、侧倾调整；蓝色滚筒承托板底，板端仍离开地面，未见机器人跌落或用手撑板。"
+     },
+     {
+      "t": 10.0,
+      "log_step": null,
+      "text_zh": "观察：本次视频全程按约1秒间隔检查，并核对视频10.00及20.00秒原帧；机器人始终留在滚筒板上，以脚支撑并作小幅姿态调整，未见板端落地或身体倒下。成功参考视频0.00、0.20、0.48、1.00、2.00、30.00秒呈现相同的脚—板—滚筒支撑关系；参考只用于辨认场景，不作为本次完成证据。"
+     },
+     {
+      "t": 30.0,
+      "log_step": null,
+      "text_zh": "观察：视频29.00、29.20、29.40、29.60、29.80及30.00秒，机器人双脚仍在棕色板面上，板底由蓝色滚筒承托，两端可见离地间隙，手臂未撑板；末帧没有跌倒。inputs/run.json的评分器reward记录success=1、replay_success=1、stages_reached=3、metric_time_on_board_s=30.0、metric_max_tilt_deg=6.48。"
+     },
+     {
+      "t": 30.0,
+      "log_step": 34,
+      "text_zh": "推断：结合本次视频30.00秒末帧仍保持脚在板上的支撑、评分器成功与板上30.0秒记录，将持续平衡事件记为视频约30.0秒首次达成并锁存。log step 30及34对最终交付动作文件的官方重放均报告成功、无失败原因及30秒完成，与录像相符；日志中的其他调试跌落不属于本次受评录像。未将agent工作时间换算为视频时间。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频0.00—30.00秒所检查帧中，机器人的脚踩棕色板面承重，腿和躯干调整姿态维持滚筒板上的平衡；开局侧倾调整较明显，之后小幅晃动。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "time_on_board_s",
+   "value": 30.0,
+   "unit": "s",
+   "threshold": 30.0,
+   "direction": "higher_better"
+  },
+  {
+   "name": "max_tilt_deg",
+   "value": 6.48,
+   "unit": "deg",
+   "threshold": null,
+   "direction": "lower_better"
+  }
+ ],
+ "failure": null,
+ "summary_zh": "本次受评录像中，机器人开局晃动后维持双脚在滚筒板上的支撑，至视频30.00秒仍未见板端触地或身体跌落，唯一节点“平衡满30秒”达成。评分器记录成功、板上时间30.0秒、最大倾角6.48°，与最终交付文件在log step 30及34的成功重放结果一致；会话中的其他失败测试不计入本次进度。无状态日志，25帧/秒录像不能排除每个毫秒物理步的瞬时接触，完整规则通过由本次评分器���功结果支持。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 9,
+    "summary_zh": "唯一节点balanced_30s在视频30.0秒达成，与评分器success=1、time_on_board=30.0、fell=0及抽帧所见（全程机器人站在滚筒板上、板端离地）一致；无失败层，contact_actual（腿/躯干 支撑）合理，未发现错误。",
+    "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": "2026-11-12T11:22:04Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-balance-board-i00-privileged--dzxnkah/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-balance-board-i00-privileged--dzxnkah/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "3c1b2e46086426c0a5875e5cfc450b42d54882ed93894012d2ebf5ed1aea738a",
+   "video_sha256": "161549273ab063779a95f2da45a902e0cbcc63904521a08ca3c578d9307e99bd",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/balance_board/trajectories/embodied-olympics-fun-balance-board-i00-standard--jfsz3xu.json

@@ -0,0 +1,205 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-balance-board-i00-standard--jfsz3xu",
+ "taskKey": "embodied-olympics-fun/balance_board",
+ "dag_sha256": "33a43b602711eb1f0ce8accf150479b69fb8b84456d1d75783d478a23c0f1d05",
+ "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": 164,
+   "sim_steps": 877,
+   "calls": 892,
+   "rows": 877,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 6188.6,
+   "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": 6.411,
+   "peak_processes": 23,
+   "peak_disk_gb": 0.335,
+   "peak_cpus": 4.07,
+   "peak_pids": 655,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "media",
+  "url": "media:embodied-olympics-fun-balance-board-i00-standard--jfsz3xu.mp4",
+  "duration_s": 1.68,
+  "fps": 25.0,
+  "frames": 42,
+  "sha256": "5f74eab9dfc294b4888e2eb8c4670ec6602929b1c2d871525dbb21822fc11b91"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "balanced_30s": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：已检查视频 0.00–1.64 s 的全部42帧；机器人开局双脚站在棕色板上，蓝色滚筒承托板底，随后机器人侧倾调节，脚仍接触板面。此节点要求完成30秒，开局站立本身不算达成。"
+     },
+     {
+      "t": 1.64,
+      "log_step": 153,
+      "text_zh": "观察：视频 1.48–1.64 s 画面右侧板端快速下沉，末帧板端降至蓝色垫面；log step 153 明确重置到最后的 episode 6，现场服务裁判返回 done=true、outcome=\"the deck touched the floor\"、sim_time_s=1.64、stages_reached=0。录像在该事件结束，未持续到30秒，节点未达成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–1.64 s，机器人的脚直接接触棕色板面承重，双腿与躯干侧倾调节；末段脚仍在板上，手臂没有接触板或滚筒。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "time_on_board_s",
+   "unit": "s",
+   "threshold": 30.0,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "max_tilt_deg",
+   "unit": "deg",
+   "threshold": null,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "balanced_30s",
+   "t": 1.6,
+   "log_step": 153,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 1.48–1.64 s 机器人仍在用脚支撑板面时，画面右侧板端下沉至地面；log step 153 的最后一次现场尝试由服务裁判以“the deck touched the floor”终止，仿真时间仅1.64秒。失败发生在持续平衡步骤中。"
+  },
+  "unmet": {
+   "balanced_30s": false
+  },
+  "phenomena": [
+   {
+    "cat": "全身",
+    "item": "平衡",
+    "t": 1.6,
+    "log_step": 153,
+    "evidence_zh": "观察：视频 1.48–1.64 s 板体向画面右侧倾落，板端与垫面的间隙消失，机器人脚仍站在倾斜板面上；log step 153 服务裁判确认板体触地。机器人与板未维持滚筒上的平衡，录像结束前机器人身体尚未倒地。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "t": 1.6,
+    "log_step": 153,
+    "source_zh": "日志 / 视频",
+    "evidence_zh": "推断：agent 的平衡控制未能补偿滚筒与板的运动。log step 153 中 agent 控制脚本的腿侧倾目标 bt 在控制索引150及163均达到0.255的限幅，服务同时报告 left_ankle_roll_q 超出允许范围而被裁剪；视频 1.48–1.64 s 板端仍继续下沉直至触地。这支持控制层面的失败，但不足以独立确定模型误差、状态估计误差或关节饱和中哪项起决定作用。"
+   }
+  ],
+  "gap_zh": "没有状态日志，板端首次触地的精确物理步与接触力无法逐步核实；触地判定由视频末段和 log step 153 的服务裁判结果共同支持。最后一次尝试��有完整的真实板倾角与控制误差记录，根因仅标到控制层；log step 155 关于重心行程余量和倾角估计偏差的解释是 agent 自述，未作为已验证的具体根因。评分器 reward 未报告 time_on_board_s、max_tilt_deg，故两项原生指标值保留null，不将仿真终止时间替代为指标值。"
+ },
+ "summary_zh": "观察：本次评分对应 log step 153 的最后一次现场尝试；视频 0.00–1.64 s 显示机器人以脚支撑滚筒板并尝试调节，随后板端触地，“平衡满30秒”未完成，最终0/1节点。推断：平衡控制在侧倾目标达到限幅且踝关节命令被裁剪时未能稳住系统；缺少状态与误差记录，无法进一步锁定具体控制或估计机制。",
+ "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、断点在 balanced_30s、log step 153 的服务裁判结果（deck touched the floor，sim_time 1.64 s）与视频长度1.68 s一致；平衡现象与控制根因（bt限幅、踝关节被裁剪，已标为推断）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": "2026-11-12T11:23:56Z",
+  "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-balance-board-i00-standard--jfsz3xu/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-balance-board-i00-standard--jfsz3xu/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "b0b5643237966187c8d7015ca9677957ee330dfebfc2d617bea234865e007ee1",
+   "video_sha256": "5f74eab9dfc294b4888e2eb8c4670ec6602929b1c2d871525dbb21822fc11b91",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/basketball_dribble/trajectories/embodied-olympics-fun-basketball-dribble-i00-privileged--fxsgasg.json

@@ -0,0 +1,254 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-basketball-dribble-i00-privileged--fxsgasg",
+ "taskKey": "embodied-olympics-fun/basketball_dribble",
+ "dag_sha256": "d5a4c6c70a883e3bab379df3e40c39ce5d4c704f8f7d56d8b3a445f58ada085b",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "determinism_checked": 0,
+   "replays": 1,
+   "n_actions": 3078,
+   "sim_steps": 3078,
+   "horizon": 6000,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 3,
+   "stages_total": 3,
+   "fell": 0,
+   "replay_s": 21.08,
+   "bad_trajectory": 0,
+   "metric_ball_height_m": 0.105,
+   "metric_distance_m": 14.468,
+   "metric_dribbles": 41.0,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 315,
+   "agent_resets": 391,
+   "agent_replays": 19,
+   "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": 2.938,
+   "peak_processes": 21,
+   "peak_disk_gb": 0.246,
+   "peak_cpus": 8.0,
+   "peak_pids": 40,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "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": "media",
+  "url": "media:embodied-olympics-fun-basketball-dribble-i00-privileged--fxsgasg.mp4",
+  "duration_s": 30.8,
+  "fps": 25.0,
+  "frames": 770,
+  "sha256": "bcf1345db86782ca80ef618bbb52ef622d92e5ba74d424889ee5835b1a0dab96"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 3,
+  "base": 0,
+  "peak": 3,
+  "final": 3,
+  "nodes": {
+   "dribble_started": {
+    "method": "visual",
+    "intervals": [
+     [
+      3.72,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 2.8,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–2.7 s，机器人在起始篮下先双手托扶篮球并抬高；视频约 2.8 s 后球离手下落，视频 3.28–3.36 s 球到达地面并反弹。"
+     },
+     {
+      "t": 3.72,
+      "log_step": 118,
+      "text_zh": "观察：视频 3.60–3.72 s，反弹篮球回到机器人伸出的单手下方，该手下拍，视频 3.80–4.00 s 球再次下降并反弹；首次续拍锚点取视频约 3.72 s。log step 118 对已保存交付文件的独立重放输出包含 dribble 阶段，且 steps=3078、dribbles=41，与本次评分字段一致。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "托/舀",
+      "note_zh": "视频 0.0–2.7 s 的准备阶段，两手共同托扶并抬高篮球，右手从下承托、左手在侧面扶球，随后分别撤开；这是单手续拍前的双手协作，并非双手同时运球。"
+     },
+     {
+      "who": "手指/夹爪",
+      "how": "其他",
+      "note_zh": "候选：击打。视频约 3.72 s 起，机器人以单只手掌向下短促拍击反弹的篮球，未见合手抓握；现有方式没有击打项。日志中交付脚本说明使用右手，与画面单侧下拍相符。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "机器人双脚接触球场地面，双腿与躯干支撑站立；释放并建立续拍时未倒地。"
+     }
+    ]
+   },
+   "free_throw_line_passed": {
+    "method": "visual",
+    "intervals": [
+     [
+      16.6,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 16.6,
+      "log_step": 117,
+      "text_zh": "观察：视频 7.0–16.6 s，机器人边单手拍球边向画面左侧行走；视频 16.4–16.6 s，球下落到右侧红色禁区前端白色罚球线的左侧地面，视频 16.8–17.0 s 随后反弹。log step 117 的最终生成及保存输出列出 past the free-throw line 阶段，log step 118 对该交付文件独立重放也包含此阶段。节点到达时间取视频约 16.6 s，远景不能核对全部球地接触点的厘米级边缘。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "其他",
+      "note_zh": "候选：击打。机器人继续用单手短促向下拍篮球，球在手与球场地面之间反复弹跳；无工具，现有方式没有击打项。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "双脚交替接触球场地面，腿与躯干支撑机器人向远离起始篮架的方向行走，通过罚球线附近。"
+     }
+    ]
+   },
+   "frontcourt_reached": {
+    "method": "visual",
+    "intervals": [
+     [
+      30.76,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 30.76,
+      "log_step": 118,
+      "text_zh": "观察：视频 29.6–30.76 s 的逐帧检查中，机器人在中圈旁继续交替落脚并单手拍球，向画面左侧跨越中线；视频 30.76 s 末帧仍站立，篮球处于低位，足底边缘与球地接触点较小，不能仅凭像素精确判线。log step 118 对已保存交付轨迹的独立重放返回 success=true、三个阶段全部出现、executed_rows=3078；本次评分 reward 同样 success=1、stages_reached=3、metric_dribbles=41、fell=0。"
+     },
+     {
+      "t": 30.76,
+      "log_step": 120,
+      "text_zh": "推断：结合视频末段的跨线动作与本次评分三个阶段全部通过，运球进入前场在视频末尾完成，导航锚点取最后可见帧视频 30.76 s（约 30.8 s，非毫秒级接触时刻）。log step 120 再次生成的 3078 行轨迹与成功备份逐值相同，支持所查重放结果对应最终交付，而非早期调试尝试；不以骨盆距离单独判定球和双脚已过线。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "其他",
+      "note_zh": "候选：击打。机器人在走向中线及跨线期间持续用单手下拍篮球；该手接触篮球后撤离让球落地，现有方式没有击打项。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "双脚交替落地支撑，机器人沿球场向前行走并跨越中线；没有轮式底盘，末段未见跌倒。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "distance_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": 14.468
+  },
+  {
+   "name": "dribbles",
+   "unit": "count",
+   "threshold": 1,
+   "direction": "higher_better",
+   "value": 41.0
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人完成开始运球、运球过罚球线、运球进入前场三个节点，视频锚点分别为约 3.72 s、16.6 s 和末帧 30.76 s；评分记录成功并计入 41 次运球。视频显示机器人先双手托扶抬球再释放，以单手连续下拍、双脚交替支撑前进，最终跨越中线，已保存轨迹的独立重放结果也通过全部阶段。没有逐帧状态日志，远景无法独立核对 0.03 秒触球间隔和球地、足底接触点的精确判线，最终合法性结合本次评分及交付重放判定，末节点时间仅为视频末尾的近似导航锚点。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "成功运行，三个节点均达成且与评分（stages 3/3）一致；抽查视频帧，时间锚点合理，接触标注合理，无失败层。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-11-12T11:25:56Z",
+  "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-basketball-dribble-i00-privileged--fxsgasg/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-basketball-dribble-i00-privileged--fxsgasg/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "8b6e1e783f993da1f70b35db9764cb385505565d8d1083aa8e8123e8c953c3ba",
+   "video_sha256": "bcf1345db86782ca80ef618bbb52ef622d92e5ba74d424889ee5835b1a0dab96",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/basketball_dribble/trajectories/embodied-olympics-fun-basketball-dribble-i00-standard--nqgbi3q.json

@@ -0,0 +1,258 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-basketball-dribble-i00-standard--nqgbi3q",
+ "taskKey": "embodied-olympics-fun/basketball_dribble",
+ "dag_sha256": "d5a4c6c70a883e3bab379df3e40c39ce5d4c704f8f7d56d8b3a445f58ada085b",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 6000,
+   "sim_steps": 18105,
+   "calls": 913,
+   "rows": 18105,
+   "resets": 3,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 3757.8,
+   "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,
+   "peak_cpus": 0.0,
+   "peak_pids": 0.0,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "media",
+  "url": "media:embodied-olympics-fun-basketball-dribble-i00-standard--nqgbi3q.mp4",
+  "duration_s": 60.04,
+  "fps": 25.0,
+  "frames": 1501,
+  "sha256": "6c19a952b6d813620eca2ea43f01ae690de1a6dbb554013c817a521e17ae961a"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 3,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "dribble_started": {
+    "method": "visual",
+    "intervals": [
+     [
+      1.08,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.64,
+      "log_step": null,
+      "text_zh": "观察：视频 0.32–0.64 s，机器人撤开原先托扶篮球的双手，球下落到地面；视频 0.72–1.00 s，球重新反弹到右手附近。"
+     },
+     {
+      "t": 1.08,
+      "log_step": 76,
+      "text_zh": "观察：视频 1.04–1.12 s，机器人右手向前推反弹中的篮球，随后视频 1.20–1.44 s 球离开右手并再次落地；左手未同时夹持球。log step 76 中，机器人服务（执行控制并返回裁判状态的接口）在最后一次重置后报告 stages_reached=1、stages_total=3，与这次续拍一致。首次完成时间取视频约 1.08 s，远景判时有约 0.1 s 误差。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "托/舀",
+      "note_zh": "视频开局，右手从下方托球、左手从侧面扶球，两手随后撤开完成释放；这是续拍前的托扶阶段。"
+     },
+     {
+      "who": "手指/夹爪",
+      "how": "推/扫",
+      "note_zh": "初始释放并落地反弹后，右手张开，从球的侧后方向前短促推球；视频约 1.04–1.12 s 可见球随右手动作向前离开，没有合手抓握。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "释放和首次右手推球期间，机器人双脚接触球场地面、腿与躯干支撑站立。"
+     }
+    ]
+   },
+   "free_throw_line_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 4.8,
+      "log_step": 76,
+      "text_zh": "观察：视频 1.20–4.80 s，篮球向远离起始篮架的方向反弹、减速，机器人随后向前跟行；视频 4.80–6.40 s，球停在红色禁区前端白色罚球线的篮架一侧，机器人也停止持续前进。"
+     },
+     {
+      "t": 60.0,
+      "log_step": 76,
+      "text_zh": "观察：全片粗查以及视频 59.04–60.00 s 的末秒密查，球始终未到罚球线另一侧，末秒仍停在原处、机器人直立在球后。log step 76 的最后一次现场尝试返回 outcome=time、stages_reached=1/3，支持罚球线检查点未完成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "首次续拍后，机器人双脚交替接触球场地面向罚球线方向跟球行走，随后在停住的球后小幅调整脚步；未见这一阶段再次发生手球接触。"
+     }
+    ]
+   },
+   "frontcourt_reached": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 60.0,
+      "log_step": 76,
+      "text_zh": "观察：视频全程机器人和篮球均留在起始篮架附近，末秒视频 59.04–60.00 s 球仍在罚球线的篮架一侧，机器人双脚也远未到画面左侧中线。成功参考视频 40.00–42.00 s 可见机器人与球到达��侧中线附近，本次没有出现该位置变化；log step 76 的最后一次服务裁判结果仅达到 1/3 阶段。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "distance_m",
+   "value": null,
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better"
+  },
+  {
+   "name": "dribbles",
+   "value": null,
+   "unit": "count",
+   "threshold": 1,
+   "direction": "higher_better"
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "free_throw_line_passed",
+   "t": 4.8,
+   "log_step": 76,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 2.40–4.80 s，机器人跟随首次推球后的篮球向前走；球反弹幅度逐渐减小并在罚球线前停住。视频 5.00–6.40 s、后续全片粗查及视频 59.04–60.00 s，机器人均未再次用手推球或将球送过罚球线。log step 76 显示该次重置后的尝试在服务仿真 60.0 s 结束，裁判 outcome=time、stages_reached=1/3。"
+  },
+  "phenomena": [
+   {
+    "cat": "其他",
+    "item": "运球中断",
+    "t": 4.8,
+    "log_step": 76,
+    "evidence_zh": "观察：视频 1.04–1.12 s 的右手推球后，篮球继续自行反弹；视频 2.40–4.80 s 反弹逐渐衰减并停住，机器人没有续拍。视频 5.00–6.40 s 及视频 59.04–60.00 s，球一直停在手前方的地面上，未继续向罚球线移动。",
+    "note_zh": "候选：运球中断。已有条目没有描述单次推球后不再续拍、球自行停住的动态操作中断；没有看到失败的抓握、对移动球的挥空动作或跌倒，故不强套没保持住、动态时机或平衡。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "认知",
+    "item": "推理/规划",
+    "log_step": 75,
+    "source_zh": "日志",
+    "evidence_zh": "推断：log step 69 的控制脚本只规划释放后的一次右掌推球，随后转入跟球行走；log step 75 将后续策略明确改为停在球后约 0.85 m，轨迹用完后手臂保持 NEUT，没有第二次推球或球停住后的续拍分支。log step 76 执行这一版本作为最后一次现场尝试；视频 4.80–60.00 s 的停球与跟随位置保持符合该脚本。该跨阶段策略缺少继续把球推进到罚球线所需的动作，解释了进度停在首次续拍之后。"
+   }
+  ],
+  "gap_zh": "没有逐帧状态日志；远景视频只有 25 fps，无法独立核对 0.03 s 接触合并、接触点及作用力，开始运球的合法性由画面中的释放、反弹、右手续推与最后一次服务裁判的阶段数共同支持。run.json 的 reward 未报告 distance_m 或 dribbles 原生数值，二者保留 null；不把 agent 脚本的视觉估计距离或 stages_reached 代填为原生指标。",
+  "unmet": {
+   "dribble_started": true,
+   "free_throw_line_passed": false,
+   "frontcourt_reached": false
+  }
+ },
+ "summary_zh": "评分对应最后一次重置后的现场尝试：机器人释放篮球，约视频 1.08 s 右手合法续推一次，完成“开始运球”，最终进度为 1/3。此后机器人仅跟随球走，约视频 4.8 s 球在罚球线前停住，直到视频 60.00 s 都未续拍；脚本只安排一次推球后保持跟随距离，缺少继续推进的规划。无逐帧状态日志，细微接触时序依靠画面与最后一次服务裁判的 1/3 阶段结果共同判断。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "视频 0.4–59.8 s 抽查：机器人与球全程停在篮架附近的红色禁区内，球未过罚球线，与 1/3 进度、断点和因果标签一致；未发现需要修改的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-11-12T11:27:59Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-basketball-dribble-i00-standard--nqgbi3q/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-basketball-dribble-i00-standard--nqgbi3q/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "14d6e2647bb82cf8b30023195b82fb2495ae58e9d9816f9ebc62dbdac4b8a409",
+   "video_sha256": "6c19a952b6d813620eca2ea43f01ae690de1a6dbb554013c817a521e17ae961a",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/bowling/trajectories/embodied-olympics-fun-bowling-i00-privileged--thybm8q.json

@@ -0,0 +1,189 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-bowling-i00-privileged--thybm8q",
+ "taskKey": "embodied-olympics-fun/bowling",
+ "dag_sha256": "958dbc6f1b15aea49b7fbb6f67f6fe11ebb2ebd0b8c66ca9ec74face659152c2",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "determinism_checked": 0,
+   "replays": 1,
+   "n_actions": 1088,
+   "sim_steps": 1088,
+   "horizon": 4000,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 3,
+   "stages_total": 3,
+   "fell": 0,
+   "replay_s": 12.46,
+   "bad_trajectory": 0,
+   "metric_entry_angle_deg": -0.53,
+   "metric_entry_speed_m_s": 3.43,
+   "metric_entry_y_m": -0.052,
+   "metric_pins_down": 10.0,
+   "metric_release_speed_m_s": 4.96,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 2118,
+   "agent_resets": 0,
+   "agent_replays": 9,
+   "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,
+   "peak_cpus": 0.0,
+   "peak_pids": 0.0,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "media",
+  "url": "media:embodied-olympics-fun-bowling-i00-privileged--thybm8q.mp4",
+  "duration_s": 10.92,
+  "fps": 25.0,
+  "frames": 273,
+  "sha256": "7ecaa1fdd7e0771b53408277e6fe21c7c90e6bfb21829f37ae4ae952a75cd829"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "strike_completed": {
+    "method": "visual",
+    "intervals": [
+     [
+      10.88,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.64,
+      "log_step": null,
+      "text_zh": "观察：视频 0.56–0.84 s，机器人转动上身，紫球随两掌移至身体侧前方；视频 1.08–1.28 s，紫球已与掌部分离并朝球道前进，双脚仍支承在犯规线后的助走区。"
+     },
+     {
+      "t": 5.88,
+      "log_step": null,
+      "text_zh": "观察：视频 5.72–6.16 s，紫球抵达白色球瓶阵列前部并进入瓶区；视频 6.40–8.80 s，球瓶陆续倾倒、散开。视频 9.88–10.88 s逐帧检查中机器人保持站立，瓶台上主要可见低伏的白色瓶体；远景不足以逐瓶量出倾角。"
+     },
+     {
+      "t": 10.88,
+      "log_step": 101,
+      "text_zh": "观察：log step 101 的官方轨迹回放工具针对最终交付文件报告 success=true、stages=[delivery,pins,strike]、pins_down=10、steps=1088、time_s=10.88；log step 109 再次确认最终文件与 S5_391 轨迹逐项一致。inputs/run.json 的评分器字段亦为 replay_success=1、metric_pins_down=10、stages_reached=3、fell=0，支持本次视频的全中终态。"
+     },
+     {
+      "t": 10.88,
+      "log_step": 101,
+      "text_zh": "推断：最终交付回放在仿真 10.88 s 完成点数，且本次视频约 5.9 s 的球入瓶区、末帧视频 10.88 s 的倒瓶终态与碰瓶后等待 5 秒的验收顺序一致。将完成锚定于已检查的末帧视频 10.88 s（约数，远景时刻误差约 0.3 s）；此前即使瓶已倒下也不提前记完成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "候选：夹持。视频 0.56–0.84 s 两掌共同保持紫球并随上身转动，随后共同送出，视频 1.08 s 球已离掌。结合任务所述无工作手指的平掌垫及 log step 106 对最终控制脚本的夹压说明，推断为两掌夹压和摩擦持球；没有单手合拢包住球的抓握。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–1.28 s 出手期间及视频 9.88–10.88 s 末段，机器人双脚接触犯规线后的木质助走区承重，支撑转体送球与出手后的站立。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "pins_down",
+   "value": 10.0,
+   "unit": "个",
+   "threshold": 10,
+   "direction": "higher_better"
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人以双掌夹压紫球并转体送出，球在视频约 5.9 s 进入瓶区，随后球瓶连锁倾倒；按碰瓶后五秒的点数验收，在视频约 10.88 s 完成唯一节点“投球全中”，最终为 1/1。最终交付文件的官方回放与本次评分器均确认十瓶倒下、三阶段完成且未跌倒，无失败断点。没有状态日志，本次与成功参照均为场地远景，���瓶倾角及细微接触无法独立量化，十瓶终态由评分器与交付回放记录补足。",
+ "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，节点 strike_completed 于视频 10.88 s 达成，与评分器 pins_down=10、stages 3/3、fell=0 及碰瓶(约5.9 s)+5 s 一致；contact_actual 合理，无失败断点。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-11-12T11:26:53Z",
+  "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-bowling-i00-privileged--thybm8q/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-bowling-i00-privileged--thybm8q/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "0e6f772c6a21f30a8306e6e74807ea42b5f4f21f45d2ec28abbbb478fc035f04",
+   "video_sha256": "7ecaa1fdd7e0771b53408277e6fe21c7c90e6bfb21829f37ae4ae952a75cd829",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/bowling/trajectories/embodied-olympics-fun-bowling-i00-standard--zj8uhzf.json

@@ -0,0 +1,207 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-bowling-i00-standard--zj8uhzf",
+ "taskKey": "embodied-olympics-fun/bowling",
+ "dag_sha256": "958dbc6f1b15aea49b7fbb6f67f6fe11ebb2ebd0b8c66ca9ec74face659152c2",
+ "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": 930,
+   "sim_steps": 3416,
+   "calls": 30,
+   "rows": 1220,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 3945.5,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 3,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 1,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.0,
+   "peak_processes": 0.0,
+   "peak_disk_gb": 0.0,
+   "peak_cpus": 0.0,
+   "peak_pids": 0.0,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "media",
+  "url": "media:embodied-olympics-fun-bowling-i00-standard--zj8uhzf.mp4",
+  "duration_s": 9.32,
+  "fps": 25.0,
+  "frames": 233,
+  "sha256": "38ffa7f6e52f13061ffea8064220c26fd6382d429aa610c391b1d8cd6a95b2c9"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "strike_completed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.8,
+      "log_step": 82,
+      "text_zh": "观察：视频 0.00–0.80 s 机器人双臂在髋前持球，随后向前上方挥臂；视频 0.88–1.36 s 紫球与双手分离并落向线前球道，双脚始终留在助走区。log step 82 明确重置并运行最后一次 episode 6，故本视频不包含前五次测试。"
+     },
+     {
+      "t": 9.28,
+      "log_step": 82,
+      "text_zh": "观察：视频 2.00–8.00 s 球向瓶区前进并逐渐偏向机器人左侧；视频 8.32–9.28 s 球沿左侧沟边前行，尚未抵达瓶阵，远端白色球瓶仍直立。log step 82 的在线服务（执行机器人命令并返回裁判状态）报告 episode 6、outcome=\"a gutter ball\"、stages_reached=1/3，未完成碰瓶及全中验收。"
+     },
+     {
+      "t": 9.28,
+      "log_step": 82,
+      "text_zh": "推断：log step 37 的服务控制周期为 0.01 s，log step 82 最后一次尝试共 930 个 control_steps，对应仿真 9.30 s；这与视频 9.28 s 最后画面球在沟边、视频总长 9.32 s 相符。该控制时钟仅用于核对末段判负，节点全程未达到。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "推断：候选“夹持”。视频 0.00–0.80 s 双臂在髋前共同保持紫球，结合任务所述开局双平掌夹球，记录左右掌垫夹压球；没有闭合手指包围球，不能标抓握。随后双臂挥动送球，视频 0.88–1.36 s 可见球已脱离双手；掌面转成承托的具体接触角度被身体遮挡，未另记托/舀。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–9.28 s 机器人双脚接触木质助走区，挥臂送球及等待球前行时持续支撑身体，末帧仍站立。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "pins_down",
+   "value": null,
+   "unit": "个",
+   "threshold": 10,
+   "direction": "higher_better"
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "strike_completed",
+   "t": 9.28,
+   "log_step": 82,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 8.32–9.28 s 紫球持续沿机器人左侧沟边前进，未到达仍直立的瓶阵；机器人站在犯规线后等待。log step 82 确认最后一次在线投球在 control_steps=930 时因“a gutter ball”结束，只达到 delivery 阶段，因此“投球全中”未完成且未修复。"
+  },
+  "unmet": {
+   "strike_completed": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 9.28,
+    "log_step": 82,
+    "evidence_zh": "观察：视频 2.00–8.00 s 离手后的紫球逐渐向机器人左侧偏离球道中央，视频 8.32–9.28 s 已沿左沟边滚动，未击中远端瓶阵；log step 82 的裁判结果为“a gutter ball”。这次投球的方向偏差使球在碰瓶前入沟。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "参数调试",
+    "log_step": 82,
+    "evidence_zh": "推断：agent 脚本用左右肩摆动速度差参数 dy 调整球路。log step 75、78 分别以 dy=-0.27、-0.172 测试，log step 77、79 的 agent 图像拟合得到向右的球路斜率约 -0.126、-0.050；这些仅为会话调参证据。log step 82 将最终 dy 设为 -0.137 后执行评分尝试，log step 83 对该次球路的拟合仍显示向左偏，且视频 2.00–9.28 s 直接显示球向左沟偏离并终止。前后调参及最终画面支持瞄准修正参数未校准到可沿球道抵达瓶区的范围；不把 agent 自建仿真的速度与倒瓶预测当作本次失败事实。",
+    "source_zh": "日志 / 视频"
+   }
+  ],
+  "gap_zh": "未提供状态日志；本次视频只有场地远景，髋前掌球接触部分被身体遮挡，无法精确判定夹压转承托的接触细节、离手瞬间和沟槽首次接触时刻。末段入沟判负由视频沟边球路与最后一次在线服务结果共同确定；run.json 的评分字段未报告 pins_down 数值，因此原生指标保留 null。"
+ },
+ "summary_zh": "评分轨迹是 log step 82 的第六次在线尝试：机器人站在犯规线后挥臂送出紫球，球未抵达瓶阵，“投球全中”全程未完成，最终 0/1。视频 8.32–9.28 s 球沿机器人左侧沟边前行，服务以入沟判负；瞄准速度差参数的调试未消除左偏。无状态日志且掌球接触部分遮挡，无法细分出手接触过程，评分字段也未报告逐瓶指标。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "节点未达成、断点与服务读数（930 control_steps、gutter ball、stages 1/3）一致，视频中球离手后沿左侧偏行；标签未发现需要改动的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-11-12T11:28:42Z",
+  "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-bowling-i00-standard--zj8uhzf/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-bowling-i00-standard--zj8uhzf/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "1aaa7cf3dac29d8f7b802a3c09d2979c13856f65684fc8160e823434564f8138",
+   "video_sha256": "38ffa7f6e52f13061ffea8064220c26fd6382d429aa610c391b1d8cd6a95b2c9",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/box_jump/trajectories/embodied-olympics-fun-box-jump-i00-privileged--yvnt96z.json

@@ -0,0 +1,203 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-box-jump-i00-privileged--yvnt96z",
+ "taskKey": "embodied-olympics-fun/box_jump",
+ "dag_sha256": "94c6447a4b03c43587761e86e851339c4316a0cf5855b294978de845b6362c39",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "determinism_checked": 0,
+   "replays": 1,
+   "n_actions": 592,
+   "sim_steps": 392,
+   "horizon": 2000,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 3,
+   "stages_total": 3,
+   "fell": 0,
+   "replay_s": 4.14,
+   "bad_trajectory": 0,
+   "metric_air_s": 0.325,
+   "metric_pelvis_above_box_m": 1.014,
+   "metric_pelvis_rise_m": 0.373,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 184,
+   "agent_resets": 22244,
+   "agent_replays": 5,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 3.36,
+   "peak_processes": 29,
+   "peak_disk_gb": 0.242,
+   "peak_cpus": 8.06,
+   "peak_pids": 52,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "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": "media",
+  "url": "media:embodied-olympics-fun-box-jump-i00-privileged--yvnt96z.mp4",
+  "duration_s": 3.96,
+  "fps": 25.0,
+  "frames": 99,
+  "sha256": "0705d8e801a1eab4486b15a8ed68529f163ad24f9be790922012edf095c09628"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "box_jump_stood": {
+    "method": "visual",
+    "intervals": [
+     [
+      3.92,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.6,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–0.84 s，机器人在棕色箱前的地面上由站立转为双腿屈曲，再伸腿蹬地；尚未站到箱顶，节点起初不成立。"
+     },
+     {
+      "t": 0.96,
+      "log_step": null,
+      "text_zh": "观察：视频 0.88–1.08 s，机器人双脚离开地面，身体向前上方移动并收腿越过箱沿；视频约 1.12 s 双脚落到箱顶，随后屈膝缓冲，尚未完成站直保持。"
+     },
+     {
+      "t": 3.92,
+      "log_step": null,
+      "text_zh": "观察：逐帧检查视频 1.40–3.92 s，机器人双脚留在箱顶，逐渐伸腿起身，视频约 2.00 s 起保持近直立；视频 2.80–3.92 s 的每帧及末帧仍见脚底支撑于箱顶，未见跌落或手、膝触箱。"
+     },
+     {
+      "t": null,
+      "log_step": 65,
+      "text_zh": "观察：log step 65 记录 agent 的部署脚本生成并回放交付轨迹，裁判输出 success=true、done=true，完成 take-off、landing、stand 三阶段；其 air_s=0.325、pelvis_rise_m=0.373、pelvis_above_box_m=1.014 与本次 run.json 的 reward 字段一致。后续 log step 75 为另一个候选的测试，不能代替本次交付。"
+     },
+     {
+      "t": 3.92,
+      "log_step": null,
+      "text_zh": "推断：本次评分器 reward 的 success=1、replay_success=1、stages_reached=3/3、fell=0 与上述视频过程一致，确认合规起跳、落箱及连续站立窗口均获验收；以已检查的末帧视频 3.92 s 作为节点完成锚点，之后至视频结束仍成立。没有逐步状态日志，精确接触高度、法线、骨盆速度和站立计数依赖裁判，完成时刻为近似视频锚点，并非把 agent 工作时间换算为视频时间。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：机器人双脚在箱前地面承重并蹬起，腾空后双脚落在棕色箱顶承重，伸腿并维持站立（视频 0.48–0.84 s、1.12–3.92 s）；双臂摆动，未见手部支承箱体。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "air_s",
+   "unit": "s",
+   "threshold": 0.1,
+   "direction": "higher_better",
+   "value": 0.325
+  },
+  {
+   "name": "pelvis_rise_m",
+   "unit": "m",
+   "threshold": 0.1,
+   "direction": "higher_better",
+   "value": 0.373
+  },
+  {
+   "name": "pelvis_above_box_m",
+   "unit": "m",
+   "threshold": 1.0,
+   "direction": "higher_better",
+   "value": 1.014
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人从地面蹬起并腾空跃上箱顶，屈膝缓冲后伸腿站稳，至视频 3.92 s 完成“跳上箱并站直”，最终 1/1 节点达成，评分器回放成功且未跌倒。无状态日志，精确接触几何、速度和���续站立计数由裁判成功结果补充确认，完成时间取视频末帧的近似锚点。log step 80、82 的会话额度提示发生在 log step 65 保存并验证交付之后，不改变已固定的成功结果。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "视频显示机器人从箱前蹬起、腾空、双脚落箱顶并站直保持至末帧；成功节点、接触（腿/躯干 支撑）、无 failure 均与视频及评分器 reward 一致。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-11-12T11:26:30Z",
+  "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-box-jump-i00-privileged--yvnt96z/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-box-jump-i00-privileged--yvnt96z/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "29b86ab0b97bc0e1a47c9decfe5fa3b515490dd093f753b6b58d81b0cc6e92e1",
+   "video_sha256": "0705d8e801a1eab4486b15a8ed68529f163ad24f9be790922012edf095c09628",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/box_jump/trajectories/embodied-olympics-fun-box-jump-i00-standard--cd23ibi.json

@@ -0,0 +1,216 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-box-jump-i00-standard--cd23ibi",
+ "taskKey": "embodied-olympics-fun/box_jump",
+ "dag_sha256": "94c6447a4b03c43587761e86e851339c4316a0cf5855b294978de845b6362c39",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 902,
+   "sim_steps": 5047,
+   "calls": 1594,
+   "rows": 5047,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 5791.4,
+   "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,
+   "peak_cpus": 0.0,
+   "peak_pids": 0.0,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "media",
+  "url": "media:embodied-olympics-fun-box-jump-i00-standard--cd23ibi.mp4",
+  "duration_s": 9.04,
+  "fps": 25.0,
+  "frames": 226,
+  "sha256": "f6b05fd3c39b940a93f82b2f048039a65c390eb5c5553d12772a268e27fa9d00"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "box_jump_stood": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": 127,
+      "text_zh": "观察：视频 0.00–7.28 s，机器人在箱前地面小步前移并站定；log step 127 明确重置并启动最后第 6 回合。视频随后只有一次深蹲起跳，与 log step 128 的第 6 回合结束记录相符；前五次尝试不计入本次节点。"
+     },
+     {
+      "t": 8.88,
+      "log_step": null,
+      "text_zh": "观察：视频 7.80–8.60 s 机器人在箱前屈膝蓄力；视频 8.64–8.80 s 伸腿蹬地，视频 8.88–8.96 s 双脚离地、收腿并向箱顶近沿移动。"
+     },
+     {
+      "t": 9.0,
+      "log_step": 128,
+      "text_zh": "观察：视频 9.00 s 最后一帧机器人仍屈膝，脚位于箱顶近沿，脚后部尚在顶面外，未出现箱顶伸腿站直及保持。机器人服务在 log step 128 返回 episode=6、control_steps=902、sim_time_s=9.02、outcome=\"landed on the side of the box\"、stages_reached=1/3，确认起跳阶段后因箱侧接触判负；这三个原生阶段不是三个 DAG 节点。完整“跳上箱并站直”节点从未满足。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–8.80 s，机器人左右脚在地面承重、小步前移、深蹲和蹬地；视频 8.88–8.96 s 双脚腾空，视频 9.00 s 脚到箱体近上沿。log step 128 的机器人服务确认脚触箱侧面，未建立合规双脚顶面支撑；未见上肢触箱。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "air_s",
+   "unit": "s",
+   "threshold": 0.1,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "pelvis_rise_m",
+   "unit": "m",
+   "threshold": 0.1,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "pelvis_above_box_m",
+   "unit": "m",
+   "threshold": 1.0,
+   "direction": "higher_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "box_jump_stood",
+   "t": 9.0,
+   "log_step": 128,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 8.88–9.00 s，机器人腾空收腿后双脚只到箱体近上沿，脚后部仍在顶面外，未站直；log step 128 的机器人服务在第 6 回合报告“landed on the side of the box”并立即结束失败，之后没有修复。"
+  },
+  "unmet": {
+   "box_jump_stood": false
+  },
+  "phenomena": [
+   {
+    "cat": "全身",
+    "item": "步态",
+    "t": 9.0,
+    "log_step": 128,
+    "evidence_zh": "观察：视频 8.88–9.00 s，机器人完成腾空和收腿，但向前跳的落脚位置不足，脚到达箱体近沿时后部仍在顶面外；机器人服务在 log step 128 确认箱侧落脚，回合结束前未见翻倒。所需双脚箱顶落脚未完成。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "log_step": 127,
+    "evidence_zh": "推断：log step 127 显示 agent 以 wd=0.15 的步行接近和自编关节控制脚本执行最后跳跃；视频 8.88–9.00 s 显示该执行未把双脚送入合规顶面区域，log step 128 确认箱侧接触。agent 在 log step 133 自述本地模型预测脚跟能进入箱顶，而实跑落在近沿，结合画面可归为跳跃控制未获得足够落脚余量；不足以细分为某个增益、轨迹跟踪或感知错误。",
+    "source_zh": "日志 / 视频 / agent 自述"
+   }
+  ],
+  "gap_zh": "无状态日志，无法逐控制步复核脚箱接触点与法线，亦无 air_s、pelvis_rise_m、pelvis_above_box_m 的评分器数值，三个原生指标均留 null。log step 129 的相机位移估计为 0.081 m、里程计为 0.154 m，脚本选择保留里程计；缺少独立位置真值，无法判定哪种估计错误，也无法验证 agent 在 log step 133 自述的约 8 cm 偏差与约 5 cm 余量，故不据此增加感知根因。"
+ },
+ "summary_zh": "本次评分为最后第 6 回合：机器人在地面接近箱子，深蹲后双脚蹬地腾空，但未完成唯一的“跳上箱并站直”节点，最终进度为 0/1。视频 9.00 s 双脚只到箱体近沿，机器人服务在 log step 128 以箱侧落脚判负；现象为步态，根因保留为控制。缺少状态日志及独立位置真值，无法进一步区分模型、位移估计与关节执行误差。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "标签与视频及日志一致：视频 9.00 s 双脚仅到箱体近沿，log step 128 服务返回 landed on the side of the box、1/3 阶段；唯一节点未达成、断点、步态现象、控制根因均有证据支持，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": "2026-11-12T11:27:28Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-box-jump-i00-standard--cd23ibi/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-box-jump-i00-standard--cd23ibi/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "dbde247c9bd712e02f35b7bd115934a224fda4d54d2136964b8f8e1e3afe84b4",
+   "video_sha256": "f6b05fd3c39b940a93f82b2f048039a65c390eb5c5553d12772a268e27fa9d00",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/boxing/trajectories/embodied-olympics-fun-boxing-i00-privileged--zm89ghy.json

@@ -0,0 +1,236 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-boxing-i00-privileged--zm89ghy",
+ "taskKey": "embodied-olympics-fun/boxing",
+ "dag_sha256": "5fe95f4fd88e6dbc49fa1e5ba7230e5fd61571eeb504fd1c0835f0ca7840e98d",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "determinism_checked": 0,
+   "replays": 1,
+   "n_actions": 1065,
+   "sim_steps": 465,
+   "horizon": 1500,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 2,
+   "stages_total": 2,
+   "fell": 0,
+   "replay_s": 3.74,
+   "bad_trajectory": 0,
+   "metric_contact_impulse_n_s": 24.08,
+   "metric_contact_s": 0.054,
+   "metric_hit_height_m": 1.304,
+   "metric_impulse_n_s": 17.78,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 287,
+   "agent_resets": 611,
+   "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": 4.359,
+   "peak_processes": 19,
+   "peak_disk_gb": 0.262,
+   "peak_cpus": 8.0,
+   "peak_pids": 39,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "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": "media",
+  "url": "media:embodied-olympics-fun-boxing-i00-privileged--zm89ghy.mp4",
+  "duration_s": 4.68,
+  "fps": 25.0,
+  "frames": 117,
+  "sha256": "78bcd99923302679cc10b5fc2020d31d2e397b57072f7bb30db1645ea9a8cac2"
+ },
+ "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 0.96–1.00 s 机器人右拳尚在袋前留有间隙；视频 1.04–1.12 s 右拳伸出，在红色沙袋白带下部接触；视频 1.16–1.32 s 拳头收回，沙袋向远离机器人的方向摆动。机器人双脚支撑出拳，左手保持在身前，未见左手触袋。"
+     },
+     {
+      "t": 1.16,
+      "log_step": 37,
+      "text_zh": "观察：log step 37 对最终交付文件的官方重放报告 strike 事件，仿真 1.141 s 的 impulse_n_s=17.78；contact_s=0.054、hit_height_m=1.304。inputs/run.json 的评分器字段报告完全相同的三个数值，且 replay_success=1、stages_reached=2。"
+     },
+     {
+      "t": 1.16,
+      "log_step": 37,
+      "text_zh": "推断：log step 2 的实例输出给出控制步长 0.01 s，log step 37 交付文件重放执行 465 控制步，即仿真 4.65 s，与视频 4.68 s 的长度相差不足一帧；仿真 1.141 s 的 strike 结算与已逐帧核对的视频 1.04–1.16 s 击中、摆离相符。因此在视频 1.16 s 标记有效击打完成；冲量与接触时长的达标依据为裁判记录，并非从摆幅估算。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "其他",
+      "note_zh": "观察：视频 1.04–1.12 s 机器人右手末端固定拳垫击中沙袋白带，随后迅速收回；候选接触名为“击打”，现有方式没有短促冲击的对应条目，不能用抓握或推/扫替代。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.96–1.32 s 机器人左右脚接地，双腿和躯干支承右拳出击与回收。"
+     }
+    ]
+   },
+   "guard_held": {
+    "method": "visual",
+    "intervals": [
+     [
+      4.64,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 1.4,
+      "log_step": null,
+      "text_zh": "观察：视频 1.16–1.40 s 机器人右肘弯曲，右拳从沙袋前收回身体附近；视频 1.40–2.60 s 机器人向后调整脚步，沙袋独立来回摆动，未见机器人再次触袋或跌倒。"
+     },
+     {
+      "t": 4.64,
+      "log_step": null,
+      "text_zh": "观察：视频 2.64–4.64 s 机器人保持右肘弯曲、右拳在肩部附近的护架姿态，身体站立、位置变化较小；已逐帧检查视频 3.60–4.64 s，末帧仍仅脚接地，拳头与摆动沙袋之间有明显间隙。"
+     },
+     {
+      "t": 4.64,
+      "log_step": 37,
+      "text_zh": "推断：log step 37 最终交付文件的裁判重放在仿真 4.65 s 报告 guard 和 success，log step 40 再次官方重放报告相同终态哈希、465 控制步和成功；评分器亦报告两个阶段完成、fell=0。按 465×0.01 s 与视频时长及击打事件的对照，护架连续计时完成在视频末帧附近，取已检查的视频 4.64 s 为近似锚点，误差约一帧。开局护架和刚收拳均不提前计入此节点。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 1.40–2.60 s 机器人双脚交替落地、后撤调整站位；视频 2.64–4.64 s 由脚部接地支撑身体保持护架，右拳留在身体附近且不接触沙袋。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "impulse_n_s",
+   "unit": "N*s",
+   "threshold": 12.0,
+   "direction": "higher_better",
+   "value": 17.78
+  },
+  {
+   "name": "contact_s",
+   "unit": "s",
+   "threshold": 0.1,
+   "direction": "lower_better",
+   "value": 0.054
+  },
+  {
+   "name": "hit_height_m",
+   "unit": "m",
+   "threshold": [
+    1.2,
+    1.6
+   ],
+   "direction": "target",
+   "value": 1.304
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人在视频约 1.16 s 完成右拳击中白带并分离，裁判记录有效冲量 17.78 N·s、接触时长 0.054 s；随后收拳、后撤调整站位，在视频约 4.64 s 完成护架站稳，最终两个节点均完成。交付文件的官方重放结果与本次评分器的阶段数、指标和结束步数一致。没有状态日志，拳峰接触区域、精确冲量和连续护架阈值依赖裁判记录确认；节点时刻以已检查视频帧及已核对的控制步时钟近似定位。",
+ "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，两个节点标注与视频、评分器字段（465步、17.78 N·s、stages 2/2、fell=0）一致；时间、contact_actual、无 failure 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": "2026-11-12T11:24:25Z",
+  "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-boxing-i00-privileged--zm89ghy/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-boxing-i00-privileged--zm89ghy/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "19464b0c5e11e90f4302d846de1468f7489f553aeee3836ecd368e6c5e3fcb59",
+   "video_sha256": "78bcd99923302679cc10b5fc2020d31d2e397b57072f7bb30db1645ea9a8cac2",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/boxing/trajectories/embodied-olympics-fun-boxing-i00-standard--35depcn.json

@@ -0,0 +1,233 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-boxing-i00-standard--35depcn",
+ "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": 630,
+   "sim_steps": 3834,
+   "calls": 3850,
+   "rows": 3834,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 5711.0,
+   "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,
+   "peak_memory_gb": 6.273,
+   "peak_processes": 36,
+   "peak_disk_gb": 0.702,
+   "peak_cpus": 4.0,
+   "peak_pids": 1290,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "media",
+  "url": "media:embodied-olympics-fun-boxing-i00-standard--35depcn.mp4",
+  "duration_s": 6.32,
+  "fps": 25.0,
+  "frames": 158,
+  "sha256": "401e22076730970c4c2121a3fdb766d430375ce788baa9c8e10dae71ed563d4d"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "bag_struck": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.4,
+      "log_step": 123,
+      "text_zh": "观察：视频 0.0–5.8 s 中机器人双脚支撑站立，右臂反复前伸、回摆，沙袋未出现参考成功录像中的击后摆离；log step 123 记录本次为 reset 后的 episode 6，并打印三次 shadow 空挥及随后实际出拳。"
+     },
+     {
+      "t": 6.28,
+      "log_step": 123,
+      "text_zh": "观察：视频 6.12–6.28 s 右手末端向沙袋白带下部靠拢，末帧仍在袋前，未见完成击打分离。log step 123 的现场机器人服务（执行动作并返回裁判状态的 API）在本回合 control_steps=630、sim_time_s=6.3 时返回 done=true、stages_reached=0/2，原因为“the bag was touched by something other than the knuckle part of the right fist”；据此未完成合法有效击打。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "其他",
+      "note_zh": "观察：视频 6.24–6.28 s 机器人右手末端伸向沙袋白带；log step 123 裁判确认实际触袋且接触部位不符合右拳拳峰要求。候选名称“击打”：此次出拳在首次触袋即终止，现有方式无击打条目；未发生抓握，也无持续推袋证据。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.0–6.28 s 机器人左右脚接地支撑摆臂和出拳，腿与躯干保持站立，未见跌倒。"
+     }
+    ]
+   },
+   "guard_held": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 6.28,
+      "log_step": 123,
+      "text_zh": "观察：视频 5.28–6.28 s 机器人右手仍在袋前，片段在首次违规触袋时结束，未出现有效击打后收右拳并连续护架站稳的阶段；log step 123 现场裁判返回 stages_reached=0/2。推断：此前空挥间的收臂不满足本节点的有效击打前置条件，本节点未完成且未进入击后护架操作。"
+     }
+    ]
+   }
+  },
+  "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": 6.28,
+   "log_step": 123,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 6.12–6.28 s 机器人正在向白色环带出右手，仍直立；log step 123 确认最后一次现场回合 episode 6 在首次触袋时因非右拳拳峰接触被终止，stages_reached=0/2。有效击袋这一步未完成，后续没有修复。"
+  },
+  "unmet": {
+   "bag_struck": false,
+   "guard_held": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 6.28,
+    "log_step": 123,
+    "evidence_zh": "观察：视频 6.12–6.28 s 右手以横转姿态向白带靠拢，未见合格击打后拳袋分离及沙袋明显摆离；log step 123 现场裁判明确判定触袋者不是右拳的拳峰部位。推断：出拳末端接触姿态未让规定拳峰成为唯一触袋区域；本现象针对接触部位失配，不推断冲量不足。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "抓取和接触设计",
+    "alt_feasible": "未知",
+    "log_step": 123,
+    "evidence_zh": "推断：本次击袋接触几何设计不适配规定拳峰。log step 123 的实际运行命令选择手背侧切点 xt=0.168，脚本打印躯干修正后切点 x_t=0.168，但同一次现场执行仍因非拳峰触袋终止；log step 125 的笔记及 log step 126 的最终自述说明 agent 未定位拳垫，且本地模拟的拳垫位置为猜测。故本地模拟的通过不能验证该接触设计；另一个掌侧方案仅解过逆运动学，未现场验证。",
+    "source_zh": "日志 / agent 自述"
+   }
+  ],
+  "gap_zh": "没有状态日志或本回合接触对、精确接触点、strike 事件及冲量明细；侧视录像不能分辨拳垫末端 0.03 m 内的具体触点，非拳峰结论依据最后一次现场裁判反馈。评分记录未报告 impulse_n_s、contact_s、hit_height_m，均填 null；log step 120–122 的本地模拟数值不属于评分回合，未移用。根因限于有日志支持的接触几何设计，无法进一步区分实际碰撞几何与模型估计的误差。"
+ },
+ "summary_zh": "评分回合完成 0/2 节点：机器人站立并多次空挥，最后在视频约 6.3 s 出拳触袋时，现场裁判因非右拳拳峰接触立即判负，尚未进入击后护架。日志支持接触几何设计不适配，agent 自述本地模拟拳垫位置为猜测；无状态日志及本回合精确接触、冲量数据，不能从模拟结果补记有效击打。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "标注与视频和日志一致：评分回合（episode 6，630 control_steps≈6.3 s）0/2，首次触袋即被判非右拳拳峰接触；断点 bag_struck、位姿不准、抓取和接触设计均有证据支持，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": "2026-11-12T11:29:41Z",
+  "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--35depcn/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-boxing-i00-standard--35depcn/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "815dba36ff98520b23f0bead16f3c9ebcb1a32a302d6039aa934942569b8f96c",
+   "video_sha256": "401e22076730970c4c2121a3fdb766d430375ce788baa9c8e10dae71ed563d4d",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/catch/trajectories/embodied-olympics-fun-catch-i00-privileged--puxtzuy.json

@@ -0,0 +1,202 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-i00-privileged--puxtzuy",
+ "taskKey": "embodied-olympics-fun/catch",
+ "dag_sha256": "61132849363e640170264bbd47c4c4cebc23f7ecfe683cae11a4abb07d10d44b",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "determinism_checked": 0,
+   "replays": 1,
+   "n_actions": 647,
+   "sim_steps": 497,
+   "horizon": 900,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 2,
+   "stages_total": 2,
+   "fell": 0,
+   "replay_s": 5.07,
+   "bad_trajectory": 0,
+   "metric_catch_height_m": 0.997,
+   "metric_catch_speed_m_s": 7.995,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 239,
+   "agent_resets": 4998,
+   "agent_replays": 13,
+   "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": 2.895,
+   "peak_processes": 18,
+   "peak_disk_gb": 0.242,
+   "peak_cpus": 8.0,
+   "peak_pids": 38,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "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": "media",
+  "url": "media:embodied-olympics-fun-catch-i00-privileged--puxtzuy.mp4",
+  "duration_s": 5.0,
+  "fps": 25.0,
+  "frames": 125,
+  "sha256": "359edea4cb0509891b6f030b6de363e4da36bd2d3e55f1581b89245f9116664b"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "fly_ball_caught": {
+    "method": "visual",
+    "intervals": [
+     [
+      4.96,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–3.60 s，机器人从初始站位迈步调整位置并伸出右手，尚未持球。"
+     },
+     {
+      "t": 3.96,
+      "log_step": null,
+      "text_zh": "观察：视频 3.88、3.92、3.96 s，白球逐帧下降接近机器人伸出的右手；视频 4.00–4.96 s 的每帧中，球留在右手末端附近，随该手的小幅动作移动，机器人保持站立。此时尚不能把首次接球直接算作完成一秒保持验收。"
+     },
+     {
+      "t": 4.96,
+      "log_step": 76,
+      "text_zh": "观察：inputs/run.json 的 reward 报告 success=1、replay_success=1、stages_reached=2/2、sim_steps=497、fell=0；log step 76 对最终交付文件重放同样报告 catch、hold 两阶段成功，执行497行，仿真结束时间为 4.97 s。视频 4.96 s 末帧仍见机器人站立并将球留在右手。"
+     },
+     {
+      "t": 4.96,
+      "log_step": 72,
+      "text_zh": "推断：视频约 3.96–4.00 s 的右手接球及片尾保持与本次评分器的两阶段成功一致，故将包含完整保持窗口的“接稳飞球”首次达成时间近似标在视频 4.96 s，末态为真。log step 72 的最终候选生成记录注明接球手为 right，且保存文件的重放指标与本次评分器一致；远景不能独立验证接触夹角和毫秒级中断，微观验收依赖原生成功结果，时间精度约一帧。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "推断：视频约 3.96–4.00 s 白球进入机器人右手，随后留在该手至视频 4.96 s；结合本次评分器 catch、hold 均成功及 log step 72 的 right 接球记录，右手以手指与拇指或掌部相对接触包住棒球。左手未参与持球。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–3.60 s 机器人双腿迈步调整接球位置，双脚交替接触场地地面承重；视频 4.00–4.96 s 双脚落地支撑站立，右手持球期间未见摔倒。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "catch_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": null,
+   "value": 0.997
+  },
+  {
+   "name": "catch_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": null,
+   "value": 7.995
+  },
+  {
+   "name": "landed_from_start_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": null,
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人迈步调整站位后，在视频约 3.96–4.00 s 用右手接住下降的棒球，并站立持球至片尾；包含完整保持窗口的接稳节点在视频约 4.96 s 达成，最终完成1/1节点，与评分器 catch、hold 两阶段成功一致。无状态日志，远景无法独立分辨包围接触的夹角及毫秒级中断，微观验收依据本次评分器成功结果，完成时刻为末帧附近的近似值。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "标签与视频、评分器成功结果一致：右手约3.96–4.00 s接球，持球至4.96 s，节点达成、final为真；远景无法独立证实接触细节，已在证据中如实说明为推断。无失败，failure为null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": "2026-11-12T11:29:24Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-privileged--puxtzuy/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-privileged--puxtzuy/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "d130fcd6fd1929e71fb35f58dcae1bc713d59ab443dcf76d7423609f91ea0a65",
+   "video_sha256": "359edea4cb0509891b6f030b6de363e4da36bd2d3e55f1581b89245f9116664b",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/catch/trajectories/embodied-olympics-fun-catch-i00-realtime--p8dimbv.json

@@ -0,0 +1,238 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-i00-realtime--p8dimbv",
+ "taskKey": "embodied-olympics-fun/catch",
+ "dag_sha256": "61132849363e640170264bbd47c4c4cebc23f7ecfe683cae11a4abb07d10d44b",
+ "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": 400,
+   "sim_steps": 2431,
+   "calls": 26,
+   "rows": 5400,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 4431.2,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "realtime": 1,
+   "slowdown": 1.0,
+   "clock_s": 38.93,
+   "max_lag_s": 0.5165,
+   "stalls": 26,
+   "stalled_s": 13.169,
+   "stalled_share": 0.3383,
+   "held_steps": 0,
+   "late_steps": 0,
+   "requests_refused": 0,
+   "request_busy_s": 0.513,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 5.916,
+   "peak_processes": 17,
+   "peak_disk_gb": 2.72,
+   "peak_cpus": 4.04,
+   "peak_pids": 221,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "media",
+  "url": "media:embodied-olympics-fun-catch-i00-realtime--p8dimbv.mp4",
+  "duration_s": 4.04,
+  "fps": 25.0,
+  "frames": 101,
+  "sha256": "04abf47ff315b84c1234baf38d32f3fddf54a82e298c3a42b7fa12f9352d2598"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "fly_ball_caught": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00 s 机器人站在场地上，双手未持球；视频 1.20–3.60 s 机器人向落点移动并伸出右手，尚未出现球被该手持续控制的状态。"
+     },
+     {
+      "t": 3.96,
+      "log_step": 117,
+      "text_zh": "观察：已逐帧检查视频 3.00–4.00 s；视频 3.80–3.96 s 小球从机器人右前上方下降至伸出的右手附近，视频 4.00 s 右手下移，机器人仍站立。画面未显示满一秒的稳定持球。"
+     },
+     {
+      "t": 4.0,
+      "log_step": 117,
+      "text_zh": "观察：log step 117 提交最终 att6，机器人现场服务返回 episode=6、control_steps=400、sim_time_s=4.0、stages_reached=1/2，outcome 为“the ball came out of the hand after the catch”。这是原生裁判对最终评分局的接住后脱手结论，不是 agent 对图像的自述；log step 118 再查服务仍为同一局终止状态。"
+     },
+     {
+      "t": 4.0,
+      "log_step": 117,
+      "text_zh": "推断：log step 25 的提交脚本使用 dt=0.01，最终局 log step 117 的 control step 400 对应约视频 4.00 s，已与视频 3.96–4.00 s 的球到手及片尾核对。原生接住阶段虽曾成立，但随后脱手且保持阶段未达成，故包含完整保持窗口的“接稳飞球”节点从未成立，终值为假。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "观察：视频 3.96 s 球到达机器人右手附近；log step 117 原生裁判确认先接住后脱手，支持右手曾短暂包住棒球。抓握未维持到规定保持窗口，远景无法分辨各指接触细节。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–4.00 s 机器人双脚在场地地面支撑身体，移动到接球位置并在伸右手接球时保持站立，未见摔倒。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "catch_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": null,
+   "value": null
+  },
+  {
+   "name": "catch_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": null,
+   "value": null
+  },
+  {
+   "name": "landed_from_start_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": null,
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "fly_ball_caught",
+   "t": 4.0,
+   "log_step": 117,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 3.96–4.00 s 小球到达机器人右手附近，机器人继续站立但未形成持续持球；log step 117 最终评分局 att6 的原生裁判报告“the ball came out of the hand after the catch”、stages_reached=1/2，并在 control step 400 终止。接球动作已开始，失败发生在接住后的保持过程中。"
+  },
+  "unmet": {
+   "fly_ball_caught": false
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没保持住",
+    "t": 4.0,
+    "log_step": 117,
+    "evidence_zh": "观察：视频 3.92–3.96 s 球接近伸出的右手，视频 4.00 s 已到片尾；log step 117 的现场裁判明确记录接住后球脱手、只到达 1/2 阶段，log step 118 确认该局已经终止。故已有短暂接住，随后丢失保持，并未完成规定的一秒窗口。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "t": 4.0,
+    "log_step": 117,
+    "source_zh": "日志",
+    "evidence_zh": "推断：log step 112 的脚本检查及 log step 117 的最终参数显示，agent 预先提交固定的右手闭合时序（t_close=393、g1=0.85）和手臂缓冲切换（t_sw=402），没有接住后主动放球的指令；原生裁判却在 control step 400 确认接住后脱手，预定缓冲切换尚未执行。结合视频 3.96–4.00 s 的接球动作，本次闭手与手臂控制未能维持已经建立的包围接触，归为控制；不足以确定具体接触几何、控制参数或反馈子类。"
+   }
+  ],
+  "gap_zh": "观察：本目录没有状态日志；视频为 640×360、25 fps 的场地远景，接球手和球只占少量像素，无法独立分辨接触方向、手指受力和 0.02 秒中断。接住后脱手由最终评分局 log step 117 的原生裁判记录佐证，但没有逐控制步的接触与保持计数，不能精确给出首次包围时刻或细分控制根因；agent 在 log step 122 对本地模型与裁判差异的解释未经验证。评分器未报告 DAG 所列三个原生指标的数值。"
+ },
+ "summary_zh": "观察：最终评分局是 log step 117 的第六次尝试；视频显示机器人移动到落点、伸出右手并保持站立，原生裁判确认短暂接住后脱手，仅到达 1/2 个裁判阶段，因此 DAG 的“接稳飞球”节点未完成，最终为 0/1。推断：断点约在视频 4.00 s 的接球保持过程中，现象为没保持住，固定闭手与手臂控制未能维持包围接触；没有状态日志，远景与汇总裁判记录不足以确定具体的接触或力控机制。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "终局 att6 的服务读数（control_steps=400、stages 1/2、'ball came out of the hand after the catch'）支持节点为假、断点在 fly_ball_caught、现象 接触·没保持住；根因停在 控制 符合分类规则。时钟（control step 400≈视频 4.00 s）与视频长度一致，未发现需改动的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-11-12T11:31:25Z",
+  "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-catch-i00-realtime--p8dimbv/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-realtime--p8dimbv/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "3eda86d13c9bd952fbdb9a7b1201126c12de251c5cd95d35f498ad355ad78a13",
+   "video_sha256": "04abf47ff315b84c1234baf38d32f3fddf54a82e298c3a42b7fa12f9352d2598",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/catch/trajectories/embodied-olympics-fun-catch-i00-standard--q3cbmrg.json

@@ -0,0 +1,201 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-i00-standard--q3cbmrg",
+ "taskKey": "embodied-olympics-fun/catch",
+ "dag_sha256": "61132849363e640170264bbd47c4c4cebc23f7ecfe683cae11a4abb07d10d44b",
+ "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": 500,
+   "sim_steps": 1304,
+   "calls": 1218,
+   "rows": 1304,
+   "resets": 2,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 1743.3,
+   "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": 3.413,
+   "peak_processes": 14,
+   "peak_disk_gb": 0.671,
+   "peak_cpus": 3.97,
+   "peak_pids": 187,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "media",
+  "url": "media:embodied-olympics-fun-catch-i00-standard--q3cbmrg.mp4",
+  "duration_s": 5.04,
+  "fps": 25.0,
+  "frames": 126,
+  "sha256": "3158f014e9352f7b950c2ca215c9841b4ff437bd0c948e77738319944c663e4c"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "fly_ball_caught": {
+    "method": "visual",
+    "intervals": [
+     [
+      5.0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0 s 机器人站在场地圆区内，双手未持球；视频 1.2–3.6 s 机器人迈步调整位置并伸出右手，尚未完成接稳。"
+     },
+     {
+      "t": 4.0,
+      "log_step": null,
+      "text_zh": "观察：逐帧检查视频 3.4–5.0 s；视频 3.88–3.96 s 白球从机器人右前上方下降，视频约 4.0 s 到达伸出的右手，此后小球留在右手边并随手轻微移动，至视频 5.0 s 未见球继续落向地面或机器人摔倒。"
+     },
+     {
+      "t": 5.0,
+      "log_step": 74,
+      "text_zh": "观察：log step 73 执行第三次现场尝试并返回 success；log step 74 的机器人服务终态为 episode=3、control_steps=500、sim_time_s=5.0、outcome=success、stages_reached=2/2；评分器字段 live_success=1、graded_by_live=1。这里的服务是仿真机器人控制及原生裁判接口，成功结果属于该最后现场尝试。"
+     },
+     {
+      "t": 5.0,
+      "log_step": 74,
+      "text_zh": "推断：接稳节点包含同手包住球满100个控制步，不能在初次触球时计为完成。log step 3 给出控制步长0.01 s，最后现场尝试的500步对应仿真 5.0 s；其结束与视频 5.0 s 片尾一致，且视频约 4.0 s 的截球与随后约一秒保持相符，故以视频约 5.0 s 为完成锚点。包围接触夹角与毫秒级中断由原生成功判定佐证，录像本身不能逐项测量。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "观察：视频约 4.0–5.0 s 机器人右手接住白球并持在手边，左手未参与接球。推断：结合 log step 74 原生裁判通过接球及保持两阶段，右手形成包围抓握；远景无法直接分辨各指接触面。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.0–5.0 s 机器人双足在地面支撑身体，接球前迈步调整位置，接球和保持期间继续站立。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "catch_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": null,
+   "value": null
+  },
+  {
+   "name": "catch_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": null,
+   "value": null
+  },
+  {
+   "name": "landed_from_start_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": null,
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "第三次现场尝试为本次评分对象：机器人移动至接球位置，视频约 4.0 s 以右手接住下降的白球，保持至视频约 5.0 s，接稳飞球节点完成，最终为1/1。log step 74 的原生服务记录成功且阶段2/2，与评分器 live_success=1 相符；前两次会话尝试的失败不属于该评分片段。无状态日志，远景不能分辨90°包围接触及0.02秒中断，微观接触���保持窗口依据原生成功判定，完成时间为近似锚点；评分器未报告三项原生指标数值。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "成功运行，接稳飞球节点 final=true、time 约5.0 s、无失败层，与视频及 log step 74 服务读数一致，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": "2026-11-12T11:28:35Z",
+  "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-catch-i00-standard--q3cbmrg/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard--q3cbmrg/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "3cb84686a6688ea728278d76658f1f9eaeb7256c4bfdea2db91cf51c9891d7ea",
+   "video_sha256": "3158f014e9352f7b950c2ca215c9841b4ff437bd0c948e77738319944c663e4c",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/catch_and_shoot/trajectories/embodied-olympics-fun-catch-and-shoot-i00-privileged--yckjj2e.json

@@ -0,0 +1,198 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-and-shoot-i00-privileged--yckjj2e",
+ "taskKey": "embodied-olympics-fun/catch_and_shoot",
+ "dag_sha256": "f6a6d84d578d7c1084b9432c694b8d43d8fa3421ae8d0142dbb9a316794fb326",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "determinism_checked": 0,
+   "replays": 1,
+   "n_actions": 633,
+   "sim_steps": 633,
+   "horizon": 1500,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 3,
+   "stages_total": 3,
+   "fell": 0,
+   "replay_s": 6.59,
+   "bad_trajectory": 0,
+   "metric_ball_height_m": 2.781,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 326,
+   "agent_resets": 315,
+   "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": 4.593,
+   "peak_processes": 16,
+   "peak_disk_gb": 0.24,
+   "peak_cpus": 8.0,
+   "peak_pids": 36,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "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": "media",
+  "url": "media:embodied-olympics-fun-catch-and-shoot-i00-privileged--yckjj2e.mp4",
+  "duration_s": 6.36,
+  "fps": 25.0,
+  "frames": 159,
+  "sha256": "a34eee26ff9435d86919228571687cbcf70755170350060afde711ba600de7b6"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "ball_caught": {
+    "method": "visual",
+    "intervals": [
+     [
+      2.68,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 2.44,
+      "log_step": null,
+      "text_zh": "观察：视频 2.20–2.44 s，橙色篮球从左侧传球机方向飞到机器人胸前，两臂迎球，篮球未先落地；视频 2.48–2.68 s 双手随球回收，随后至视频 3.20 s 球持续受控停在胸腹前。"
+     },
+     {
+      "t": 2.68,
+      "log_step": 94,
+      "text_zh": "观察：log step 94 保存最终 633 行动作并用重放工具播放该交付文件，裁判返回 catch、release、through the ring 三阶段和 success=true；其中 catch 事件为仿真 2.666 s。视频 2.68 s 已见两手控制篮球，节点在此附近记为达到，精确连续接触窗口由裁判结果佐证。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "推断：候选“夹持”。视频 2.44–3.20 s 机器人双手在篮球相对两侧共同承接并控制球，右手承托、左手从上方收合；结合 agent 对最终动作的说明（log step 98），采用双掌相向夹压，未见单手以手指和拇指形成独立包握，故不标抓握。"
+     }
+    ]
+   },
+   "shot_through_ring": {
+    "method": "visual",
+    "intervals": [
+     [
+      6.32,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 4.92,
+      "log_step": null,
+      "text_zh": "观察：视频 3.20–4.80 s 机器人两臂在胸腹前调整持球方向并将球降到髋前；视频 4.84–4.92 s 双臂向前上方摆动送球，视频 4.96 s 球与双手已有可见间隔，随后沿弧线飞向右侧篮筐。"
+     },
+     {
+      "t": 6.32,
+      "log_step": 94,
+      "text_zh": "观察：视频 5.92–6.32 s 篮筐裁剪逐帧显示篮球从篮圈上方向下接近，视频 6.28 s 球在圈口附近，视频 6.32 s 末帧球进入圈下网区；机器人仍保持站立。log step 94 的最终文件重放裁判返回 through the ring、success=true，inputs/run.json 的 reward 也报告 success=1、replay_success=1、stages_reached=3/3。"
+     },
+     {
+      "t": 6.32,
+      "log_step": null,
+      "text_zh": "推断：结合视频末尾向下穿圈的轨迹和评分器成功结果，投篮命中事件在视频约 6.32 s 达到并锁存；无状态日志，远景不能独立测量球心与圈心距离或圈下验收高度。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "推断：候选“夹持”。视频 3.20–4.80 s 双掌共同夹持篮球、换为左右相向接触并下移；视频 4.84–4.92 s 双臂共同摆送后分开释放。双掌靠相向压力和摩擦送球，未见单手独立包握；视频 4.96 s 后为离手飞行。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [],
+ "failure": null,
+ "summary_zh": "机器人在视频约 2.68 s 双手接住左侧来球，��球换姿后于视频约 4.92 s 下手投出，视频约 6.32 s 球向下穿圈，两个事件节点均完成。视频与最终交付文件的重放结果及评分器 success=1 一致，无失败断点。无状态日志，双手连续接触时长及穿圈几何阈值不能由远景独立精确测量，依裁判成功结果佐证。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "成功运行；两个事件节点时间与裁判结果（catch 仿真 2.666 s、success=1）一致，接触标注符合 DAG 候选，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": "2026-11-12T11:27:00Z",
+  "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-catch-and-shoot-i00-privileged--yckjj2e/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-privileged--yckjj2e/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "f889df12dd593bfd7db82a861a9bb3cb25ea0e4bfd4252545f3af99fec777ab6",
+   "video_sha256": "a34eee26ff9435d86919228571687cbcf70755170350060afde711ba600de7b6",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/catch_and_shoot/trajectories/embodied-olympics-fun-catch-and-shoot-i00-realtime--zchzcq5.json

@@ -0,0 +1,224 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-and-shoot-i00-realtime--zchzcq5",
+ "taskKey": "embodied-olympics-fun/catch_and_shoot",
+ "dag_sha256": "f6a6d84d578d7c1084b9432c694b8d43d8fa3421ae8d0142dbb9a316794fb326",
+ "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": 725,
+   "sim_steps": 5021,
+   "calls": 4048,
+   "rows": 8500,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 5522.4,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "realtime": 1,
+   "slowdown": 1.0,
+   "clock_s": 53.244,
+   "max_lag_s": 0.5055,
+   "stalls": 4,
+   "stalled_s": 2.008,
+   "stalled_share": 0.0377,
+   "held_steps": 0,
+   "late_steps": 0,
+   "requests_refused": 0,
+   "request_busy_s": 2.301,
+   "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,
+   "peak_cpus": 0.0,
+   "peak_pids": 0.0,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "media",
+  "url": "media:embodied-olympics-fun-catch-and-shoot-i00-realtime--zchzcq5.mp4",
+  "duration_s": 7.28,
+  "fps": 25.0,
+  "frames": 182,
+  "sha256": "43d50c87d8fe29013162977b39d155061178c63504fa89a6844653f8f2ed7058"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "ball_caught": {
+    "method": "visual",
+    "intervals": [
+     [
+      2.6,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 2.6,
+      "log_step": null,
+      "text_zh": "观察：视频 2.00–2.40 s，橙色篮球从机器人左侧传球机飞向胸前，全程未触地；视频 2.44–2.72 s，球停在两手之间，随后视频 2.80–3.20 s 仍随双手保持在胸前，没有弹走。"
+     },
+     {
+      "t": 2.6,
+      "log_step": 161,
+      "text_zh": "推断：结合视频 2.44–3.20 s 的共同控球和 log step 161 中机器人现场服务对最终第六次尝试返回的 stages_reached=3、stages_total=3、reason=success，双手接球事件成立；达到时刻取视频约 2.6 s，远景不能直接测出每手有效接触连续 0.10 秒的边界。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 2.44–3.20 s，机器人两只手从球的不同侧面共同夹持篮球。候选：夹持；画面未显示单只手以拇指和其他手指对向合拢包住局部球面，双掌共同夹压不归抓握，现有方式无准确对应项（D12）。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 2.00–3.20 s，机器人双脚接触球场地面，支撑躯干完成接球。"
+     }
+    ]
+   },
+   "shot_through_ring": {
+    "method": "visual",
+    "intervals": [
+     [
+      7.24,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 6.12,
+      "log_step": null,
+      "text_zh": "观察：视频 3.40–5.88 s，机器人把已接住的篮球降至腰腹前，双手在球下方承托；视频 5.96–6.08 s，双臂向前上方送球，视频 6.12 s 球已与双手分离，继续向右侧篮筐上方飞行。"
+     },
+     {
+      "t": 7.24,
+      "log_step": null,
+      "text_zh": "观察：视频 6.28–6.64 s，篮球上升至篮圈上方后转为下降；视频 7.08–7.24 s，球沿篮板前方向下进入篮圈，末帧落至圈下篮网区域，机器人仍站立。"
+     },
+     {
+      "t": 7.24,
+      "log_step": 161,
+      "text_zh": "观察：log step 161 中，机器人现场服务在最终第六次尝试返回 control_steps=725、done=True、reason=success、stages_reached=3、stages_total=3、outcome=success；本次 reward 另有 live_success=1、graded_by_live=1、n_actions=725。该服务状态是受评现场尝试的裁判反馈。"
+     },
+     {
+      "t": 7.24,
+      "log_step": 161,
+      "text_zh": "推断：任务文字规定每控制步为 0.01 秒，log step 161 的最终尝试 control_steps=725 对应视频约 7.25 s；已与视频 7.20–7.24 s 的入圈末段及视频总长 7.28 s 核对。完整向下穿圈并下降一球直径由成功���定确认，节点导航时刻取最末可见帧视频约 7.24 s，精度约 0.1 秒。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "托/舀",
+      "note_zh": "观察：视频 3.40–5.88 s，机器人两手转至球下方共同承托并降低篮球；视频 5.96–6.08 s，双手由下向前上方送球，随后张开分离。视频 6.12 s 后篮球自由飞行。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 3.40–7.24 s，机器人双脚在地面支撑身体完成下手投篮及出手后的回臂。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [],
+ "failure": null,
+ "summary_zh": "观察：最终受评的第六次现场尝试完成双手接球和投篮命中两节点；视频约 2.6 s 双手控制住来球，视频约 6.12 s 球已离手，视频约 7.24 s 球向下穿圈，与 log step 161 的现场服务成功判定一致。此前五次尝试属于会话调试，不计入本段进度。推断：没有状态日志，远景无法独立测量逐指接触、0.10 秒接球窗口及精确穿圈几何，节点完成由可见动作和本次裁判成功结果共同确认，接触方式按可见双掌夹持及后续托送标注。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "成功运行，两节点均达成，时间与接触标注与视频及 log step 161 的成功判定一致，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": "2026-11-12T11:31:02Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-realtime--zchzcq5/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-realtime--zchzcq5/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "61c716566db56437341713ebe9b84b54a515c62a864d0fc75c675bc16fb3eb6f",
+   "video_sha256": "43d50c87d8fe29013162977b39d155061178c63504fa89a6844653f8f2ed7058",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/catch_and_shoot/trajectories/embodied-olympics-fun-catch-and-shoot-i00-standard--jtl9cby.json

@@ -0,0 +1,229 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-and-shoot-i00-standard--jtl9cby",
+ "taskKey": "embodied-olympics-fun/catch_and_shoot",
+ "dag_sha256": "f6a6d84d578d7c1084b9432c694b8d43d8fa3421ae8d0142dbb9a316794fb326",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 650,
+   "sim_steps": 3093,
+   "calls": 1604,
+   "rows": 3093,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 5247.2,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "resource_enforced": 0,
+   "limits_violation_observed": 1,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 4.805,
+   "peak_processes": 15,
+   "peak_disk_gb": 2.575,
+   "peak_cpus": 8.54,
+   "peak_pids": 230,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "over_cpus": 1
+  }
+ },
+ "video": {
+  "kind": "media",
+  "url": "media:embodied-olympics-fun-catch-and-shoot-i00-standard--jtl9cby.mp4",
+  "duration_s": 6.52,
+  "fps": 25.0,
+  "frames": 163,
+  "sha256": "d9668c9f2c3d7e3139f19a631260ac34aa4fd6fcf8f174451e0eb65eb9987b60"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "ball_caught": {
+    "method": "visual",
+    "intervals": [
+     [
+      2.7,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 2.48,
+      "log_step": null,
+      "text_zh": "观察：视频 2.00–2.40 s 橙色篮球从左侧传球机飞向机器人，未先落地；视频 2.48–2.64 s 球进入胸腹前双手之间并随手下移，视频 2.72–4.32 s 球持续留在双手承接处，没有弹走或掉地。"
+     },
+     {
+      "t": 2.7,
+      "log_step": 153,
+      "text_zh": "推断：以视频约 2.7 s 为双手接住球的完成锚点（目测误差约 0.2 s）；log step 153 的机器人服务报告最后一次 episode 6 达到 stages_reached=2/3，随后因球触地结束，与画面中的接球后出手一致。原生阶段包含接球和释放，不能把其 2/3 直接当作此 DAG 的节点数；接球事件完成后锁存，离手不取消本节点。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 2.48–4.32 s 机器人两手共同承接、夹托篮球，球留在两手之间。候选：夹持；可见双掌形成承接槽，未见单只手以对向手指包握球面，故不标抓握；词表没有双掌相向夹持项。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 2.48–4.32 s 机器人双脚着地，双腿支撑身体完成接球和持球。"
+     }
+    ]
+   },
+   "shot_through_ring": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 4.6,
+      "log_step": 153,
+      "text_zh": "观察：视频 4.40–4.56 s 机器人双臂将球从腹前向上送出，视频 4.60 s 球已明显离开双手；视频 5.68–6.20 s 球从高处下降，始终在篮圈前方，未出现从上方穿圈的轨迹。"
+     },
+     {
+      "t": 6.48,
+      "log_step": 153,
+      "text_zh": "观察：视频 6.24–6.48 s 球继续落向篮圈前方的场地并在末帧到达地面，机器人仍站立；log step 153 的服务裁判给出 done=True、reason=failed、outcome=the ball touched the floor，评分记录 success=0。成功参照视频 6.52–6.60 s 可见球落入篮圈并到达圈下，本次视频没有这一现象。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "托/舀",
+      "note_zh": "观察：视频 4.40–4.56 s 机器人双手在球的下侧共同向上承托送球，随后双手与球分离，球独自飞行。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 4.40–4.72 s 机器人双脚保持着地，双腿支撑身体完成双臂上送投篮。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [],
+ "failure": {
+  "breakpoint": {
+   "ref": "shot_through_ring",
+   "t": 4.6,
+   "log_step": 153,
+   "where": "操作中",
+   "evidence_zh": "观察：视频约 4.6 s 机器人已接住的球随双臂上送而离手，飞行弧线很高；视频 6.00–6.48 s 球下降到篮圈前方并落地，没有穿圈，之后没有修复机会。log step 153 明确对应最后一次 episode 6，服务裁判以 the ball touched the floor 结束该次尝试。"
+  },
+  "unmet": {
+   "ball_caught": true,
+   "shot_through_ring": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 4.6,
+    "log_step": 153,
+    "evidence_zh": "观察：视频 4.56–4.72 s 机器人把球向很陡的上方送出；视频 5.68–6.48 s 球在到达篮圈前便下降，落在篮圈前方地板上，未从圈上方进入。推断：这是投篮出手形成的飞行轨迹未对准篮筐，属于位姿不准。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "source_zh": "日志",
+    "evidence_zh": "推断：最终投篮的挥臂与释放控制没有产生所需的出手方向。log step 104 的投篮执行代码按预先生成的动作序列逐步发送，log step 152 将本地模型搜索得到的挥臂时长、展臂释放时机等参数写入最终控制脚本，log step 153 用该脚本执行最后一次尝试；视频 4.56–6.48 s 显示实际高抛却提前落地。log step 154 中 agent 的像素轨迹拟合估算此次仰角约 75.6°、下落到圈面高度时仍短约 1.8 米（这些是 agent 脚本的估计，非原生状态量），与可见落点方向一致；log step 157 自述本地手臂模型未能预测最后一次出手。证据支持控制层，尚不足以断定是接触、跟踪还是某一具体参数导致。"
+   }
+  ],
+  "gap_zh": "没有状态日志；远景只能辨认双手夹托及球的运动，不能逐一核实手指接触、裁判的 0.10 秒接球窗口和确切释放速度。接球时间由视频估计并由最终现场裁判阶段结果支持；agent 对出手角度和动力学模型偏差的诊断未被独立状态记录核实，因此根因只标控制，不细分。"
+ },
+ "summary_zh": "本次评分对象是最后一次（第六次）现场尝试：机器人在视频约 2.7 s 双手接住球，接球节点锁存，最终完成 1/2 个 DAG 节点。视频约 4.6 s 出手后球高抛但在篮圈前方落地，断点为投篮命中，现象为位姿不准，日志支持投篮控制未达到所需出手方向。没有状态日志，接球精确时刻及控制失配的具体机制仍有不确定性。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "接球约 2.7 s 达成、投篮未达成、断点在投篮、现象位姿不准均与视频和 log step 153/154 一致；未发现需要修改的内容。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-11-12T11:30:20Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-standard--jtl9cby/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-and-shoot-i00-standard--jtl9cby/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "2eb175026e89532af767d7cafecb4070cf7de86ab6886682ccc308c98c4ad11f",
+   "video_sha256": "d9668c9f2c3d7e3139f19a631260ac34aa4fd6fcf8f174451e0eb65eb9987b60",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/clean_and_jerk/trajectories/embodied-olympics-fun-clean-and-jerk-i00-privileged--ufigegw.json

@@ -0,0 +1,253 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-clean-and-jerk-i00-privileged--ufigegw",
+ "taskKey": "embodied-olympics-fun/clean_and_jerk",
+ "dag_sha256": "f530c1d45bf82223dabd69c9d30cf24e240db1d588782640954595693a7015ca",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "determinism_checked": 0,
+   "replays": 1,
+   "n_actions": 3647,
+   "sim_steps": 3047,
+   "horizon": 6000,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 4,
+   "stages_total": 4,
+   "fell": 0,
+   "replay_s": 27.6,
+   "bad_trajectory": 0,
+   "metric_bar_height_m": 2.185,
+   "metric_bar_max_height_m": 2.185,
+   "metric_elbow_flex_deg": 2.0,
+   "metric_knee_flex_deg": 3.4,
+   "metric_time_s": 30.47,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 154,
+   "agent_resets": 0,
+   "agent_replays": 14,
+   "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": 4.15,
+   "peak_processes": 20,
+   "peak_disk_gb": 0.259,
+   "peak_cpus": 8.0,
+   "peak_pids": 42,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "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": "media",
+  "url": "media:embodied-olympics-fun-clean-and-jerk-i00-privileged--ufigegw.mp4",
+  "duration_s": 30.48,
+  "fps": 25.0,
+  "frames": 762,
+  "sha256": "6bf7e49ed7075a6858487639ed5b7b2b5fe1858af2c891f48ac9143e77aa5b12"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "bar_at_shoulders": {
+    "method": "visual",
+    "intervals": [
+     [
+      20.5,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 10.0,
+      "log_step": null,
+      "text_zh": "观察：视频 7.0–8.75 s 机器人俯身将左右手分别合在杠杆两侧；视频 9.5–10.75 s 双手随杠上移，杠铃离开两只棕色支架，原支架顶面露出，双脚仍支撑在木台上。"
+     },
+     {
+      "t": 20.5,
+      "log_step": null,
+      "text_zh": "观察：视频 18.0–19.5 s 机器人将杠铃从腰前抬至胸肩前；视频 20.0、20.25、20.5、20.75 s 及随后至视频 23.0 s，杠杆保持在肩前，机器人双手分握、躯干直立、双膝近伸直，双脚位置稳定。完成锚点取视频约 20.5 s，精度约半秒；上举后本事件保持已完成。"
+     },
+     {
+      "t": 20.5,
+      "log_step": 107,
+      "text_zh": "观察：log step 107 的最后一次生成、保存及回放输出列出 clean、overhead、down signal 阶段并报告成功；该文件的回放步数为 3047，终态杠高 2.185 m、肘屈曲 2.0 度、膝屈曲 3.4 度，与本次评分器字段一致。此日志用于支持交付回合的验收，节点时间取自视频。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "抓握",
+      "note_zh": "观察：视频 8.0–10.75 s 左右灵巧手分别合握杠杆两侧，双手共同持杠离架；视频 18.0–23.0 s 两手继续持杠抬至肩前并停稳。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 7.0–23.0 s 机器人的双脚接触木色举重台，俯身取杠、起身及上肩时双腿和躯干承重，未见迈步转场。"
+     }
+    ]
+   },
+   "bar_overhead": {
+    "method": "visual",
+    "intervals": [
+     [
+      30.4,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 28.5,
+      "log_step": null,
+      "text_zh": "观察：视频 23.5–28.25 s 机器人双手将杠铃从肩前缓慢上举；视频 28.5 s 杠铃已在头顶上方，两臂近伸直，双腿近伸直、躯干直立，双脚保持在木台原处。此时仅进入保持阶段，尚不据此提前计为完成。"
+     },
+     {
+      "t": 30.44,
+      "log_step": null,
+      "text_zh": "观察：视频 28.5、28.75、29.0、29.25、29.5、29.75、30.0、30.25 s 和末帧视频 30.44 s 中，机器人持续双手持杠过顶，双肘和双膝近伸直，杠铃与本体位置基本不变，片尾仍保持举杠姿态。"
+     },
+     {
+      "t": 30.4,
+      "log_step": 107,
+      "text_zh": "推断：结合视频约 28.5 s 至片尾的过顶稳定保持，以及评分器 success=1、replay_success=1、stages_reached=4/4，过顶保持及下放信号验收在片尾完成，节点从视频约 30.4 s 成立。log step 107 显示交付文件保存���回放成功并含 down signal 阶段；无需机器人继续下放。无状态日志，精确速度、细小几何阈值及保持计数由成功评分支持，完成时间为约半秒精度的视觉锚点。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "抓握",
+      "note_zh": "观察：视频 23.5–30.44 s 左右手持续分别握住杠杆两侧，双臂共同将杠铃从肩前上举并在头顶保持至片尾。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 23.5–30.44 s 双脚持续接触木色举重台，双腿近伸直、躯干直立以支撑上举和过顶保持。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "bar_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": 2.185
+  },
+  {
+   "name": "bar_max_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": 2.185
+  },
+  {
+   "name": "elbow_flex_deg",
+   "unit": "deg",
+   "threshold": 10.0,
+   "direction": "lower_better",
+   "value": 2.0
+  },
+  {
+   "name": "knee_flex_deg",
+   "unit": "deg",
+   "threshold": 10.0,
+   "direction": "lower_better",
+   "value": 3.4
+  },
+  {
+   "name": "time_s",
+   "unit": "s",
+   "threshold": 60.0,
+   "direction": "lower_better",
+   "value": 30.47
+  }
+ ],
+ "failure": null,
+ "summary_zh": "评分回合中，机器人双手持杠离架，在视频约 20.5 s 完成上肩停稳，再缓慢上举并保持至片尾，视频约 30.4 s 完成过顶验收，两个节点最终均完成；评分器记录成功及四个原生阶段全部通过。两步实际接触均为双手抓握杠杆及双脚在木台上支撑；log step 108 的会话额度提示发生在 log step 107 保存并成功回放交付文件之后，不影响本次成功。无状态日志，视频无法独立精确量测全部距离、速度和接触宽限，细项验收由成功评分支持，节点时间为视觉近似。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "成功回合，两节点均达成、时间（上肩约20.5 s、过顶约30.4 s，视频时钟）与帧一致，接触（双臂协作抓握+腿/躯干支撑）正确，无失败层，推断已标为推断。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-11-12T11:30:14Z",
+  "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-clean-and-jerk-i00-privileged--ufigegw/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-clean-and-jerk-i00-privileged--ufigegw/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "0df655780c39db039b08279cb852ff11ebf843037e36810515562ea0b6205670",
+   "video_sha256": "6bf7e49ed7075a6858487639ed5b7b2b5fe1858af2c891f48ac9143e77aa5b12",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/clean_and_jerk/trajectories/embodied-olympics-fun-clean-and-jerk-i00-standard--dxks9pw.json

@@ -0,0 +1,249 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-clean-and-jerk-i00-standard--dxks9pw",
+ "taskKey": "embodied-olympics-fun/clean_and_jerk",
+ "dag_sha256": "f530c1d45bf82223dabd69c9d30cf24e240db1d588782640954595693a7015ca",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "live_success": 1,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 3911,
+   "sim_steps": 9661,
+   "calls": 3236,
+   "rows": 9661,
+   "resets": 3,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 4070.8,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 5.15,
+   "peak_processes": 19,
+   "peak_disk_gb": 0.601,
+   "peak_cpus": 4.0,
+   "peak_pids": 661,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "media",
+  "url": "media:embodied-olympics-fun-clean-and-jerk-i00-standard--dxks9pw.mp4",
+  "duration_s": 39.12,
+  "fps": 25.0,
+  "frames": 978,
+  "sha256": "12c6f08a230afa42f29ee5c60f3b269ffd46aca5af24ed167c10baa7b9cd68f2"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "bar_at_shoulders": {
+    "method": "visual",
+    "intervals": [
+     [
+      27.0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 17.52,
+      "log_step": null,
+      "text_zh": "观察：视频 16.0 s 时杠铃仍搁在两只棕色支架上；视频 17.0–18.48 s，机器人左右手各握杠杆一侧，杠铃随双手抬起，两个套筒均离开原支架，机器人双脚仍支撑在木台上。"
+     },
+     {
+      "t": 27.0,
+      "log_step": null,
+      "text_zh": "观察：视频 24.0–26.6 s，机器人将杠杆从胸腹前升至肩前并逐渐站直；视频 26.8–30.0 s，杠杆停在肩关节附近，双手持杠、双腿近直，双脚保持并列，杠铃不再明显升降。"
+     },
+     {
+      "t": 27.0,
+      "log_step": 124,
+      "text_zh": "推断：结合视频 26.4–28.2 s 的密集帧和 log step 124 对本次第四回合记录的 success、4/4 阶段通过，将上肩完成时刻估为视频 27.0 s（约有0.5秒误差）。无状态日志，不能从画面精测肩距、膝角、速度及连续控制步计数；上肩成功后按事件模式保留。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "抓握",
+      "note_zh": "观察：视频 16.0–30.0 s，机器人两只灵巧手分别合拢握住杠杆两侧，协同将杠铃举离支架、抬至肩前并持稳。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.0–30.0 s，机器人双脚接触木色举重台，在俯身抓杠、起身及肩前持杠过程中支撑本体。"
+     }
+    ]
+   },
+   "bar_overhead": {
+    "method": "visual",
+    "intervals": [
+     [
+      39.0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 33.0,
+      "log_step": null,
+      "text_zh": "观察：视频 30.0–37.0 s，机器人从肩前继续双手上举杠铃，杠杆越过头部，双臂逐渐展开；双脚始终留在木台上，杠铃未脱手。"
+     },
+     {
+      "t": 39.08,
+      "log_step": null,
+      "text_zh": "观察：视频 37.0–39.08 s，机器人持续将银色杠铃保持在头顶上方，躯干直立、双腿近直、双脚并列；已检查视频 38.24、38.48、38.76、39.0 和 39.08 s，结束前未见下放或掉杠。"
+     },
+     {
+      "t": 39.0,
+      "log_step": 125,
+      "text_zh": "推断：视频末段的过顶保持与 log step 124 的 success、4/4 阶段通过及 log step 125 的服务状态 episode=4、done=True、reason=success 一致，判定过顶保持及下放信号验收已完成，将完整验收锚定在视频约 39.0 s（约有0.5秒误差）。画面不显示裁判计数或信号，细小关节角与速度阈值依现场成功结论确认。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "抓握",
+      "note_zh": "观察：视频 30.0–39.08 s，机器人左右灵巧手分别持续握住杠杆两侧，双臂共同将其从肩前上举并保持在头部上方。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 30.0–39.08 s，机器人双脚持续落在木色举重台上，腿与躯干支撑上举和过顶保持。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "bar_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "bar_max_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "elbow_flex_deg",
+   "unit": "deg",
+   "threshold": 10.0,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "knee_flex_deg",
+   "unit": "deg",
+   "threshold": 10.0,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "time_s",
+   "unit": "s",
+   "threshold": 60.0,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "评分对象是第四次现场回合：机器人双手将杠铃举离支架，在视频约 27.0 s 完成上肩停稳，再上举并保持至视频末段，约在视频 39.0 s 完成过顶验收，两节点最终均完成。log step 124–125 的现场服务结果为成功，与视频动作顺序和终态一致；前三次重置前的失败不属于本次评分回合。无状态日志，节点时刻为视觉近似，细小角度、距离、速度及保持计数不能独立精测，reward 也未报告 DAG 所列五项原生指标，故这些指标值留空。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "成功回合，两节点均达成。视频：16 s 杠铃仍在支架，17–20 s 双手抬离支架，26–28 s 杠在肩前停稳站直，30–39 s 杠举过头顶并保持到结束；时间与接触（双臂协作抓握、腿/躯干支撑）和标注一致，推断已标明，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": "2026-11-12T11:33:41Z",
+  "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-clean-and-jerk-i00-standard--dxks9pw/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-clean-and-jerk-i00-standard--dxks9pw/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "9ca5f6424e32918f0a02bf966eb96d48b088144826f77071ab235c071e1abc60",
+   "video_sha256": "12c6f08a230afa42f29ee5c60f3b269ffd46aca5af24ed167c10baa7b9cd68f2",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/discus/trajectories/embodied-olympics-fun-discus-i00-privileged--vedpojb.json

@@ -0,0 +1,228 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-discus-i00-privileged--vedpojb",
+ "taskKey": "embodied-olympics-fun/discus",
+ "dag_sha256": "ffb63c003bdbc581b50872ffe890e6ce579309876e24b407597d79050cb48e95",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "determinism_checked": 0,
+   "replays": 1,
+   "n_actions": 2927,
+   "sim_steps": 2527,
+   "horizon": 6000,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 3,
+   "stages_total": 3,
+   "fell": 0,
+   "replay_s": 17.45,
+   "bad_trajectory": 0,
+   "metric_distance_m": 17.809,
+   "metric_release_height_m": 2.454,
+   "metric_release_speed_m_s": 13.03,
+   "metric_release_spin_rad_s": 3.7,
+   "metric_release_tilt_deg": 12.2,
+   "metric_release_time_s": 13.695,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 320,
+   "agent_resets": 377,
+   "agent_replays": 68,
+   "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": 3.753,
+   "peak_processes": 18,
+   "peak_disk_gb": 0.379,
+   "peak_cpus": 8.0,
+   "peak_pids": 90,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "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": "media",
+  "url": "media:embodied-olympics-fun-discus-i00-privileged--vedpojb.mp4",
+  "duration_s": 25.28,
+  "fps": 25.0,
+  "frames": 632,
+  "sha256": "88a220b4a5d293f6622877ab212db1a1ae10a645e346c202501d24fa4c326a33"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "discus_landed_qualified": {
+    "method": "visual",
+    "intervals": [
+     [
+      15.5,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 13.72,
+      "log_step": null,
+      "text_zh": "观察：视频 13.4–13.8 s，机器人由俯身持饼转为伸髋、右臂向上越过头部，橙红色铁饼脱离右手向白线之间的扇形方向飞出；视频 13.72 s 铁饼已与右手分离。"
+     },
+     {
+      "t": 15.5,
+      "log_step": null,
+      "text_zh": "观察：视频 15.28–16.0 s 的远场裁剪连续帧显示铁饼下降到草地、随后沿地面运动，首次触地约在视频 15.5 s（远处目标仅数像素，时刻约有 0.2 s 误差）；机器人此时仍在圈内。"
+     },
+     {
+      "t": 15.5,
+      "log_step": 107,
+      "text_zh": "推断：结合视频约 15.5 s 的落地与评分器 reward 的 metric_distance_m=17.809、success=1、stages_reached=3/3，首次落点满足界内且距圈内沿至少15米。log step 107 对最终候选文件的原生裁判回放也报告 landing 阶段、distance_m=17.809；log step 108–109 将该候选保留并重新验证，非前期其他试验的成绩。画面不能独立量出精确距离。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "机器人右侧五指手持住橙红色铁饼；视频 0 s 铁饼在右掌与弯曲手指处，视频 13.0–13.6 s 俯身时铁饼仍随右手运动，随后摆臂投出。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "机器人双脚在圈内接地，腿和躯干支撑持饼行走、转身、俯身及伸髋摆臂投掷；投出后在圈内恢复直立。"
+     }
+    ]
+   },
+   "rear_exit_stood": {
+    "method": "visual",
+    "intervals": [
+     [
+      25.24,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 22.4,
+      "log_step": null,
+      "text_zh": "观察：视频 20.0–23.6 s，机器人背离扇形方向行进，约在视频 22.4 s 脚跨到灰圈后侧的绿色地台；此后继续向后移动，身体始终由脚支承，没有手膝撑地或倒地。"
+     },
+     {
+      "t": 25.24,
+      "log_step": null,
+      "text_zh": "观察：视频 24.24–25.24 s 已密查，机器人在圈后绿色地台上逐渐减速，片尾保持直立，视频 25.24 s 最后一帧双脚位于圈外后侧。"
+     },
+     {
+      "t": 25.24,
+      "log_step": 107,
+      "text_zh": "推断：评分器 reward 的 success=1、episode_over=1、stages_reached=3/3 证实合法后圈接触及连续站稳窗口均通过。log step 107 的最终文件回放报告 release、landing、exit 阶段及成功，控制步2527、仿真 25.27 s；其脱手诊断仿真 13.695 s 与视频 13.72 s 分离画面吻合，终点亦与视频末尾 25.24 s 相���，因此把验收锚定到视频末帧（约有0.1秒采样误差）。无逐帧状态日志，不能仅凭画面精确测出骨盆速度和躯干方向余弦。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "视频 20.0–25.24 s，机器人双腿交替迈步，脚从圈内跨过后半圈边沿踏上绿色地台，随后由双脚支撑减速站立。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "distance_m",
+   "unit": "m",
+   "threshold": 15.0,
+   "direction": "higher_better",
+   "value": 17.809
+  },
+  {
+   "name": "retouch_dv_m_s",
+   "unit": "m/s",
+   "threshold": 0.5,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人在圈内持饼前行并转身，以右手向后上方投掷，视频约 15.5 s 首次落地；评分器确认界内落点距圈内沿17.809米，铁饼落区达标。随后机器人向后半圈外行进，视频末尾约 25.24 s 完成离圈站稳，两个节点最终均达成，failure 为 null。没有逐帧状态日志，远处首次落地时间为视觉近似，精确距离、合法接触判据及站稳窗口由最终文件回放和评分器成功结果确认。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "成功运行，两个节点时间与接触标注与视频和评分器字段（release 13.695 s、距离17.809、3/3阶段）一致，failure 为 null 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": "2026-11-12T11:34:29Z",
+  "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-discus-i00-privileged--vedpojb/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-discus-i00-privileged--vedpojb/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "b30214f07489cea6c2c9a5fffdc6320524df3de5c7950c9f6ba186d28ea48d02",
+   "video_sha256": "88a220b4a5d293f6622877ab212db1a1ae10a645e346c202501d24fa4c326a33",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/discus/trajectories/embodied-olympics-fun-discus-i00-standard--cyqp43o.json

@@ -0,0 +1,243 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-discus-i00-standard--cyqp43o",
+ "taskKey": "embodied-olympics-fun/discus",
+ "dag_sha256": "ffb63c003bdbc581b50872ffe890e6ce579309876e24b407597d79050cb48e95",
+ "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": 707,
+   "sim_steps": 2912,
+   "calls": 334,
+   "rows": 2912,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 4304.5,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 5.133,
+   "peak_processes": 11,
+   "peak_disk_gb": 0.378,
+   "peak_cpus": 4.01,
+   "peak_pids": 154,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "media",
+  "url": "media:embodied-olympics-fun-discus-i00-standard--cyqp43o.mp4",
+  "duration_s": 7.08,
+  "fps": 25.0,
+  "frames": 177,
+  "sha256": "6c2f1af7236bdcef9d91bd0f84bbb986d0b43599980aaa98cc5a44d652d9742b"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "discus_landed_qualified": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 3.48,
+      "log_step": 90,
+      "text_zh": "观察：视频 0.0–6.12 s，橙红色铁饼随机器人右手从身前抬至头侧，log step 90 的本次试验本体读数仍报 holding_right=1；视频 6.20–6.32 s 铁饼已离手向前上方飞出。"
+     },
+     {
+      "t": 7.04,
+      "log_step": 90,
+      "text_zh": "观察：视频约 6.20 s 铁饼脱手，视频 6.20–6.32 s 可见其向前上方飞行，视频 6.36–7.04 s 画面内不再可见铁饼；视频 6.64–7.04 s 机器人投后迈步至圈后缘。log step 90 的机器人服务裁判返回“铁饼落地前触及圈沿顶面或圈外地面”，stages_reached=1/3。推断：依据本次服务裁判的 outcome 和阶段计数，本次只有脱手阶段，服务判负时铁饼尚未首次落地，故铁饼落区达标未完成；不以总分为零代替动作判断。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "观察：视频 0.0–6.12 s 右掌及屈曲手指持着橙红色铁饼，铁饼随右手抬举和摆臂；log step 90 本次试验本体读数 holding_right=1 支持持握。视频约 6.20 s 后铁饼脱离右手。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.0–7.04 s 机器人以脚接地支撑持饼、摆臂和投后恢复步，投后脚步移向圈后缘，未见躯干倒地。"
+     }
+    ]
+   },
+   "rear_exit_stood": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 7.04,
+      "log_step": 90,
+      "text_zh": "观察：视频 6.20–7.04 s 机器人仍在投后恢复脚步，末尾脚部接近圈后缘，没有出现有效落地后合法离圈并连续站稳的过程；log step 90 服务在收臂恢复段 post r1 即判提前触圈失败。推断：前置的有效落地未完成，本节点未成立；该脚步属于投后恢复，不能当成已开始执行合法离圈。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "distance_m",
+   "unit": "m",
+   "threshold": 15.0,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "retouch_dv_m_s",
+   "unit": "m/s",
+   "threshold": 0.5,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "discus_landed_qualified",
+   "t": 7.04,
+   "log_step": 90,
+   "where": "操作中",
+   "evidence_zh": "观察：视频约 6.20 s 铁饼脱手，视频 6.64–7.04 s 机器人保持直立但脚步移向后侧圈沿；log step 90 的机器人服务裁判在本次试验 post r1 返回“落地前触及圈沿顶面或圈外地面”并终止。推断：依据 log step 90 的 outcome 与 stages_reached=1/3，服务判负时铁饼尚未首次落地，仍在完成投掷及等待首次落地时犯规，断点是铁饼落区达标，而非尚未执行的后圈离圈站稳。log step 91 确认最后评分局为 episode=6、control_steps=707；log step 54 给出控制步长 0.01 s，707×0.01=7.07 s 与视频末尾相符，且已用视频约 6.20 s 的脱手核对本次动作。"
+  },
+  "unmet": {
+   "discus_landed_qualified": false,
+   "rear_exit_stood": false
+  },
+  "phenomena": [
+   {
+    "cat": "约束",
+    "item": "违反约束",
+    "t": 7.04,
+    "log_step": 90,
+    "evidence_zh": "观察：视频 6.64–7.04 s 机器人脚部逐渐落至灰色投掷圈后缘，身体仍直立；log step 90 的本次服务裁判明确记录铁饼落地前已触及圈沿顶面或圈外地面，立即判负。推断：违反落地前必须留在圈内的规则；没有把仍直立的恢复步标作跌倒。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "source_zh": "日志 / 视频",
+    "evidence_zh": "观察：log step 88 将本次投后恢复段改为向前 0.15 m/s、向左 0.10 m/s 的 walk 速度命令；log step 90 执行最终修改后的脚本，仍在 post r1 收臂恢复段就触圈终止，尚未进入脚本的后退离圈段。视频 6.20–7.04 s 显示机器人投掷时躯干后仰并迈出恢复步，脚部向圈后缘移动。推断：该投掷与恢复动作的全身控制未能将支撑脚保持在圈内；现有证据不足以细分为具体反馈环节或某个已证实错误的参数。"
+   }
+  ],
+  "gap_zh": "没有状态日志；单一场地斜视录像无法分辨触圈的精确接触点及圈沿顶面与圈外地面的区别，该事实以本次实时服务裁判记录为据。视频在铁饼首次落地前结束，评分字段未报告 distance_m 或 retouch_dv_m_s，因此两项原生指标留空；不采用前五次试验的估算落点和速度作为本次结果。"
+ },
+ "summary_zh": "观察：第六次实时试验是本次评分局，机器人用右手持饼摆臂，视频约 6.20 s 铁饼脱手，但两个 DAG 节点均未完成。观察：视频末尾机器人仍直立、恢复脚步移向圈后缘，log step 90 的服务裁判以铁饼落地前触圈终止；推断：投掷及恢复动作的全身控制未能保持圈内支撑，断在铁饼落区达标。观察：无状态日志且视频没有记录到首次落地，精确触点、落点距离和触饼冲量无法独立量出。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 2,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "fix",
+    "conf": 7,
+    "summary_zh": "WHERE 标签（两节点未达成、断点 discus_landed_qualified、操作中、违反约束 circle_foul 对应 DAG 约束）与视频和 log step 90/91 一致，时钟换算（707×0.01≈7.07 s，释放 6.20 s 与视频吻合）可核对。只有一处观察性陈述与视频不符：铁饼在视频 6.64–7.04 s 并不可见。",
+    "n_issues": 1,
+    "n_major": 0
+   },
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 2,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "两节点未达成、断点 discus_landed_qualified（操作中）、违反约束现象与视频（6.20 s 脱手、末尾脚步移向圈后缘、直立）及 log 末步服务裁判（stages 1/3、落地前触圈）一致；上一轮的铁饼可见性问题已修正；时钟换算（707×0.01≈7.07 s）可核对。",
+    "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": "2026-11-12T12:15:24Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-discus-i00-standard--cyqp43o/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-discus-i00-standard--cyqp43o/04_review_r1"
+   },
+   {
+    "stage": "reconcile",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-discus-i00-standard--cyqp43o/05_reconcile_r1"
+   },
+   {
+    "stage": "traj_review",
+    "round": 2,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-discus-i00-standard--cyqp43o/04_review_r2"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "63dddb98b45795945daba4980e285cea84394aca2a14bf95515efc10346ae261",
+   "video_sha256": "6c2f1af7236bdcef9d91bd0f84bbb986d0b43599980aaa98cc5a44d652d9742b",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/dodgeball/trajectories/embodied-olympics-fun-dodgeball-i00-privileged--ajfmeqj.json

@@ -0,0 +1,313 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-dodgeball-i00-privileged--ajfmeqj",
+ "taskKey": "embodied-olympics-fun/dodgeball",
+ "dag_sha256": "a13a84ada7b474ef0b05b1dd7e90325e49ac88a53ef3656b957cb843db7034f2",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "determinism_checked": 0,
+   "replays": 1,
+   "n_actions": 1550,
+   "sim_steps": 1288,
+   "horizon": 1550,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 5,
+   "stages_total": 5,
+   "fell": 0,
+   "replay_s": 10.13,
+   "bad_trajectory": 0,
+   "metric_dodged": 5.0,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 120,
+   "agent_resets": 451,
+   "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": 2.873,
+   "peak_processes": 17,
+   "peak_disk_gb": 0.244,
+   "peak_cpus": 8.0,
+   "peak_pids": 38,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "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": "media",
+  "url": "media:embodied-olympics-fun-dodgeball-i00-privileged--ajfmeqj.mp4",
+  "duration_s": 12.92,
+  "fps": 25.0,
+  "frames": 323,
+  "sha256": "d954c76bda9d2241fde1d3ae46f826629f1c0e0a02622dce0f8655781b2297be"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 5,
+  "base": 0,
+  "peak": 5,
+  "final": 5,
+  "nodes": {
+   "throw_1_dodged": {
+    "method": "visual",
+    "intervals": [
+     [
+      3.04,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 3.04,
+      "log_step": null,
+      "text_zh": "观察：视频 2.48–2.88 s，第1个红球从发球机飞向机器人，机器人侧身连续退步，球从机器人画面左侧经过；视频 2.96–3.20 s，球下降至木地板后弹起，机器人仍直立并继续退行。首次触地约在视频 3.04 s 附近，时间为目测近似。"
+     },
+     {
+      "t": null,
+      "log_step": 36,
+      "text_zh": "观察：log step 36 重新生成的动作数组与已保存交付文件完全相同；随后对该交付文件的新进程重放由基准裁判返回 success=true，stages 包含“dodge 1”。本次评分器字段 replay_success=1、stages_reached=5、metric_dodged=5，与视频中五次来球及末态一致；上述视频锚点取自画面，未用 agent 工作时间换算。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "避开",
+      "note_zh": "视频 2.48–3.20 s，机器人用双腿带动躯干侧身退行，第1个红球从身体旁经过；画面及裁判成功结果支持机器人未与此活球接触。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "视频 2.48–3.20 s，机器人双脚交替接触木质球场地面，支撑身体完成第1次躲球的侧身退行。"
+     }
+    ]
+   },
+   "throw_2_dodged": {
+    "method": "visual",
+    "intervals": [
+     [
+      5.52,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 5.52,
+      "log_step": null,
+      "text_zh": "观察：视频 4.72–5.36 s，第2个红球飞向机器人，机器人继续侧身退行，球从其画面左侧经过；视频 5.44–5.76 s，球下降至木地板后弹起，与机器人分离，机器人未倒地。首次触地约在视频 5.52 s 附近，时间为目测近似。"
+     },
+     {
+      "t": null,
+      "log_step": 36,
+      "text_zh": "观察：log step 36 重新生成的动作数组与已保存交付文件完全相同；随后对该交付文件的新进程重放由基准裁判返回 success=true，stages 包含“dodge 2”。本次评分器字段 replay_success=1、stages_reached=5、metric_dodged=5，与视频中五次来球及末态一致；上述视频锚点取自画面，未用 agent 工作时间换算。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "避开",
+      "note_zh": "视频 4.72–5.76 s，机器人用双腿带动躯干侧身退行，第2个红球从身体旁经过；画面及裁判成功结果支持机器人未与此活球接触。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "视频 4.72–5.76 s，机器人双脚交替接触木质球场地面，支撑身体完成第2次躲球的侧身退行。"
+     }
+    ]
+   },
+   "throw_3_dodged": {
+    "method": "visual",
+    "intervals": [
+     [
+      8.32,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 8.32,
+      "log_step": null,
+      "text_zh": "观察：视频 7.12–8.00 s，机器人已反向跑动，第3个红球从其画面右侧飞过；视频 8.08–8.32 s，球继续远离机器人，在画面右侧边线附近落地并弹起，机器人保持直立。完成锚点取视频 8.32 s，时间为目测近似。"
+     },
+     {
+      "t": null,
+      "log_step": 36,
+      "text_zh": "观察：log step 36 重新生成的动作数组与已保存交付文件完全相同；随后对该交付文件的新进程重放由基准裁判返回 success=true，stages 包含“dodge 3”。本次评分器字段 replay_success=1、stages_reached=5、metric_dodged=5，与视频中五次来球及末态一致；上述视频锚点取自画面，未用 agent 工作时间换算。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "避开",
+      "note_zh": "视频 7.12–8.32 s，机器人用双腿带动躯干反向跑动，第3个红球从身体旁经过；画面及裁判成功结果支持机器人未与此活球接触。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "视频 7.12–8.32 s，机器人双脚交替接触木质球场地面，支撑身体完成第3次躲球的反向跑动。"
+     }
+    ]
+   },
+   "throw_4_dodged": {
+    "method": "visual",
+    "intervals": [
+     [
+      10.48,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 10.48,
+      "log_step": null,
+      "text_zh": "观察：视频 9.76–10.32 s，第4个红球飞来，机器人继续向画面左侧跑动，球从其画面右侧经过；视频 10.40–10.72 s，球落到机器人后方的木地板并弹起，机器人继续跑动。首次触地约在视频 10.48 s 附近，时间为目测近似。"
+     },
+     {
+      "t": null,
+      "log_step": 36,
+      "text_zh": "观察：log step 36 重新生成的动作数组与已保存交付文件完全相同；随后对该交付文件的新进程重放由基准裁判返回 success=true，stages 包含“dodge 4”。本次评分器字段 replay_success=1、stages_reached=5、metric_dodged=5，与视频中五次来球及末态一致；上述视频锚点取自画面，未用 agent 工作时间换算。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "避开",
+      "note_zh": "视频 9.76–10.72 s，机器人用双腿带动躯干反向跑动，第4个红球从身体旁经过；画面及裁判成功结果支持机器人未与此活球接触。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "视频 9.76–10.72 s，机器人双脚交替接触木质球场地面，支撑身体完成第4次躲球的反向跑动。"
+     }
+    ]
+   },
+   "throw_5_dodged": {
+    "method": "visual",
+    "intervals": [
+     [
+      12.88,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 12.88,
+      "log_step": null,
+      "text_zh": "观察：视频 12.20–12.84 s，第5个红球由远端飞来，机器人继续侧身跑动，球从其画面右侧经过并下降；视频 12.88 s 末帧，球已到达木地板表面、与机器人分离，机器人仍直立且脚在白色边线以内。完成锚点取视频 12.88 s。"
+     },
+     {
+      "t": null,
+      "log_step": 36,
+      "text_zh": "观察：log step 36 重新生成的动作数组与已保存交付文件完全相同；随后对该交付文件的新进程重放由基准裁判返回 success=true，stages 包含“dodge 5”。本次评分器字段 replay_success=1、stages_reached=5、metric_dodged=5，与视频中五次来球及末态一致；上述视频锚点取自画面，未用 agent 工作时间换算。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "避开",
+      "note_zh": "视频 12.20–12.88 s，机器人用双腿带动躯干反向跑动，第5个红球从身体旁经过；画面及裁判成功结果支持机器人未与此活球接触。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "视频 12.20–12.88 s，机器人双脚交替接触木质球场地面，支撑身体完成第5次躲球的反向跑动。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "dodged",
+   "value": 5.0,
+   "unit": "count",
+   "threshold": 5,
+   "direction": "higher_better"
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人先转为侧身退行躲过前两球，再反向跑动躲过后三球，五个事件节点依次完成，最终为5/5。视频 12.88 s 末帧显示第5球已到达地面、机器人仍直立，评分器同时记录 replay_success=1、stages_reached=5、metric_dodged=5。没有状态日志，远景红球较小，逐球完成时间按落地及弹起画面作近似定位，瞬时无活球接触由画面结合裁判成功结果确认。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "五个节点时间与 log step 36 重放事件（dodge 1–5 于 2.99/5.48/8.27/10.41/12.88 s）及视频帧一致，末态直立在线内，contact_actual 合理，无失败故无断点。",
+    "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": "2026-11-12T11:32:40Z",
+  "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-dodgeball-i00-privileged--ajfmeqj/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-dodgeball-i00-privileged--ajfmeqj/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "e006f7ec67c7c5cad976168fede4d146732f40c1dde3800e92fd21ab9b5442f2",
+   "video_sha256": "d954c76bda9d2241fde1d3ae46f826629f1c0e0a02622dce0f8655781b2297be",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/dodgeball/trajectories/embodied-olympics-fun-dodgeball-i00-realtime--rwpkmjp.json

@@ -0,0 +1,324 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-dodgeball-i00-realtime--rwpkmjp",
+ "taskKey": "embodied-olympics-fun/dodgeball",
+ "dag_sha256": "a13a84ada7b474ef0b05b1dd7e90325e49ac88a53ef3656b957cb843db7034f2",
+ "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": 1330,
+   "sim_steps": 1330,
+   "calls": 667,
+   "rows": 3320,
+   "resets": 0,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 311.2,
+   "video_rendered": 1,
+   "first_success": 1,
+   "first_reason_code": 1,
+   "realtime": 1,
+   "slowdown": 1.0,
+   "clock_s": 13.3,
+   "max_lag_s": 0.0058,
+   "stalls": 0,
+   "stalled_s": 0.0,
+   "stalled_share": 0.0,
+   "held_steps": 2,
+   "late_steps": 0,
+   "requests_refused": 1,
+   "request_busy_s": 0.504,
+   "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,
+   "peak_cpus": 0.0,
+   "peak_pids": 0.0,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "media",
+  "url": "media:embodied-olympics-fun-dodgeball-i00-realtime--rwpkmjp.mp4",
+  "duration_s": 13.32,
+  "fps": 25.0,
+  "frames": 333,
+  "sha256": "e57925489f25c3972afa4ae74b59a626144a7b73323c8e44e22ca1c9fa893010"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 5,
+  "base": 0,
+  "peak": 5,
+  "final": 5,
+  "nodes": {
+   "throw_1_dodged": {
+    "method": "visual",
+    "intervals": [
+     [
+      3.1,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 3.1,
+      "log_step": null,
+      "text_zh": "观察：视频 2.48–3.04 s，第1球从远端飞向机器人原先所在位置，机器人转身并持续向侧方迈步，球与其身体分离；视频 3.08–3.16 s，球在机器人后侧落地并反弹，机器人仍直立。"
+     },
+     {
+      "t": 3.1,
+      "log_step": 8,
+      "text_zh": "推断：第1球按上述落地画面在视频约 3.1 s 完成躲避；这是同一未重置的实时评分回合。log step 8 中基准服务（机器人 API 的回合裁判反馈）报告 outcome=success、stages_reached=5、stages_total=5，支持该球已记账；日志未逐球报告记账时刻，因此此处为视频近似时间，并非按计划供球时刻计完成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "避开",
+      "note_zh": "机器人通过转身及向左侧移步，让腿部、躯干与第1个活球的飞行路线分离。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "躲避第1球期间，机器人左右脚交替接触木质球场地面，支撑躯干完成行走。"
+     }
+    ]
+   },
+   "throw_2_dodged": {
+    "method": "visual",
+    "intervals": [
+     [
+      5.8,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 5.8,
+      "log_step": null,
+      "text_zh": "观察：视频 5.00–5.72 s，第2球飞过机器人身旁，机器人继续沿场宽方向迈步；视频 5.76–5.88 s，球已在脚旁外侧触地、弹起并远离，机器人没有倒地。"
+     },
+     {
+      "t": 5.8,
+      "log_step": 8,
+      "text_zh": "推断：第2球按上述落地画面在视频约 5.8 s 完成躲避；这是同一未重置的实时评分回合。log step 8 中基准服务（机器人 API 的回合裁判反馈）报告 outcome=success、stages_reached=5、stages_total=5，支持该球已记账；日志未逐球报告记账时刻，因此此处为视频近似时间，并非按计划供球时刻计完成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "避开",
+      "note_zh": "机器人继续沿场宽方向行走并在随后换向，腿部、躯干避开第2个活球。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "躲避第2球期间，机器人左右脚交替接触木质球场地面，支撑躯干完成行走。"
+     }
+    ]
+   },
+   "throw_3_dodged": {
+    "method": "visual",
+    "intervals": [
+     [
+      8.1,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 8.1,
+      "log_step": null,
+      "text_zh": "观察：视频 7.40–8.00 s，第3球从远端飞来，机器人已反向退行，球从身体另一侧经过；视频 8.08–8.20 s，球在机器人后侧落地并反弹。"
+     },
+     {
+      "t": 8.1,
+      "log_step": 8,
+      "text_zh": "推断：第3球按上述落地画面在视频约 8.1 s 完成躲避；这是同一未重置的实时评分回合。log step 8 中基准服务（机器人 API 的回合裁判反馈）报告 outcome=success、stages_reached=5、stages_total=5，支持该球已记账；日志未逐球报告记账时刻，因此此处为视频近似时间，并非按计划供球时刻计完成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "避开",
+      "note_zh": "机器人反向退行，腿部、躯干离开第3球瞄准的位置，与活球无接触。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "躲避第3球期间，机器人左右脚交替接触木质球场地面，支撑躯干完成行走。"
+     }
+    ]
+   },
+   "throw_4_dodged": {
+    "method": "visual",
+    "intervals": [
+     [
+      10.8,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 10.8,
+      "log_step": null,
+      "text_zh": "观察：视频 10.12–10.68 s，第4球接近时机器人持续退行，红球与身体轮廓保持分离；视频 10.76–10.84 s，球在机器人身旁外侧触地并反弹，机器人仍在场内站立。"
+     },
+     {
+      "t": 10.8,
+      "log_step": 8,
+      "text_zh": "推断：第4球按上述落地画面在视频约 10.8 s 完成躲避；这是同一未重置的实时评分回合。log step 8 中基准服务（机器人 API 的回合裁判反馈）报告 outcome=success、stages_reached=5、stages_total=5，支持该球已记账；日志未逐球报告记账时刻，因此此处为视频近似时间，并非按计划供球时刻计完成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "避开",
+      "note_zh": "机器人持续退行，腿部、躯干避开第4个活球，没有伸手挡球。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "躲避第4球期间，机器人左右脚交替接触木质球场地面，支撑躯干完成行走。"
+     }
+    ]
+   },
+   "throw_5_dodged": {
+    "method": "visual",
+    "intervals": [
+     [
+      13.28,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 13.28,
+      "log_step": null,
+      "text_zh": "观察：视频 12.64–13.20 s，第5球从发球机飞来，从持续退行的机器人身旁通过；视频 13.24–13.28 s，红球下降至地面，末帧中球与机器人分离、机器人双脚支撑身体，未倒地。"
+     },
+     {
+      "t": 13.28,
+      "log_step": 8,
+      "text_zh": "推断：第5球按上述落地画面在视频约 13.28 s 完成躲避；这是同一未重置的实时评分回合。log step 8 中基准服务（机器人 API 的回合裁判反馈）报告 outcome=success、stages_reached=5、stages_total=5，支持该球已记账；日志未逐球报告记账时刻，因此此处为视频近似时间，并非按计划供球时刻计完成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "避开",
+      "note_zh": "机器人继续退行，腿部、躯干避开第5个活球，末帧红球与双脚分离。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "躲避第5球期间，机器人左右脚交替接触木质球场地面，支撑躯干完成行走。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "dodged",
+   "value": 5,
+   "unit": "count",
+   "threshold": 5,
+   "direction": "higher_better"
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人先转身沿场宽方向行走，随后反向退行，五球依次从身旁通过并落地，最终五个躲球节点全部完成。log step 8 的基准服务报告 success、stages_reached=5/5，与视频末段机器人直立且末球落地的画面一致，原生 dodged 计数取该五阶段完成数。没有状态日志及逐球裁判记账记录，各节点时刻按视频落地画面近似标注；agent 自述录像线程失败（log step 10）不代表本次提供的评分视频缺失。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "评分回合成功（success=1，step 8 服务报告 5/5），视频帧显示机器人全程直立、五球先后落地且未接触机器人，节点值、时间、contact_actual 均与证据一致；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": "2026-11-12T11:35:19Z",
+  "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-dodgeball-i00-realtime--rwpkmjp/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-dodgeball-i00-realtime--rwpkmjp/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "28dc69a0d3167846eb51672b485d641acef3c4e83469cd7696cdaf452b98007b",
+   "video_sha256": "e57925489f25c3972afa4ae74b59a626144a7b73323c8e44e22ca1c9fa893010",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/dodgeball/trajectories/embodied-olympics-fun-dodgeball-i00-standard--yrmvyiq.json

@@ -0,0 +1,313 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-dodgeball-i00-standard--yrmvyiq",
+ "taskKey": "embodied-olympics-fun/dodgeball",
+ "dag_sha256": "a13a84ada7b474ef0b05b1dd7e90325e49ac88a53ef3656b957cb843db7034f2",
+ "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": 1325,
+   "sim_steps": 1966,
+   "calls": 200,
+   "rows": 1966,
+   "resets": 1,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 950.2,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.626,
+   "peak_processes": 10,
+   "peak_disk_gb": 0.954,
+   "peak_cpus": 1.54,
+   "peak_pids": 146,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "media",
+  "url": "media:embodied-olympics-fun-dodgeball-i00-standard--yrmvyiq.mp4",
+  "duration_s": 13.28,
+  "fps": 25.0,
+  "frames": 332,
+  "sha256": "8a9656d2369e3077ed3a8a2f597ef7f6ca1eb98564dad2b56594599058ada672"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 5,
+  "base": 0,
+  "peak": 5,
+  "final": 5,
+  "nodes": {
+   "throw_1_dodged": {
+    "method": "visual",
+    "intervals": [
+     [
+      2.88,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 2.88,
+      "log_step": null,
+      "text_zh": "观察：视频 2.00–2.88 s，机器人沿弧线向初始朝向的右侧跑动；第1球从远端中间的发球机飞出，在视频 2.64–2.88 s 从机器人身旁通过并落到木地板上，视频 2.96–3.12 s 可见球已反弹，机器人继续站立跑动。"
+     },
+     {
+      "t": 13.24,
+      "log_step": 17,
+      "text_zh": "推断：log step 17 中机器人服务（基准提供的在线控制及裁判接口）在重置后的 episode 2 返回 outcome=success、stages_reached=5、stages_total=5，且评分器 live_success=1；结合第1球的逐帧飞行与触地画面，判定本节点完成。完成时间按视频触地帧近似标为视频 2.88 s，逐球原生记账时刻未记录，误差约 0.1–0.2 s。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "避开",
+      "note_zh": "观察：机器人双腿带动躯干沿右转弧线跑动，使身体避开第1次来球。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：机器人左右脚交替接触木质球场地面，支撑躯干完成本次躲球。"
+     }
+    ]
+   },
+   "throw_2_dodged": {
+    "method": "visual",
+    "intervals": [
+     [
+      5.28,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 5.28,
+      "log_step": null,
+      "text_zh": "观察：视频 4.48–5.28 s，机器人继续朝右侧边线方向跑动，第2球从画面最右侧发球机射来，在视频 5.12–5.28 s 从机器人身旁穿过并触地，视频 5.36–5.44 s 可见反弹；机器人仍在白色边线内侧。"
+     },
+     {
+      "t": 13.24,
+      "log_step": 17,
+      "text_zh": "推断：log step 17 中机器人服务（基准提供的在线控制及裁判接口）在重置后的 episode 2 返回 outcome=success、stages_reached=5、stages_total=5，且评分器 live_success=1；结合第2球的逐帧飞行与触地画面，判定本节点完成。完成时间按视频触地帧近似标为视频 5.28 s，逐球原生记账时刻未记录，误差约 0.1–0.2 s。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "避开",
+      "note_zh": "观察：机器人双腿带动躯干继续向右侧跑动，使身体避开第2次来球。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：机器人左右脚交替接触木质球场地面，支撑躯干完成本次躲球。"
+     }
+    ]
+   },
+   "throw_3_dodged": {
+    "method": "visual",
+    "intervals": [
+     [
+      8.08,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 8.08,
+      "log_step": null,
+      "text_zh": "观察：视频 5.60–7.20 s，机器人在右侧白色边线内侧减速、转身后开始折返；视频 7.36–8.08 s 第3球从画面最左侧发球机飞来，在机器人身旁通过，视频 8.08 s 球在侧边线附近触地，视频 8.12–8.24 s 反弹远离，机器人继续移动。"
+     },
+     {
+      "t": 13.24,
+      "log_step": 17,
+      "text_zh": "推断：log step 17 中机器人服务（基准提供的在线控制及裁判接口）在重置后的 episode 2 返回 outcome=success、stages_reached=5、stages_total=5，且评分器 live_success=1；结合第3球的逐帧飞行与触地画面，判定本节点完成。完成时间按视频触地帧近似标为视频 8.08 s，逐球原生记账时刻未记录，误差约 0.1–0.2 s。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "避开",
+      "note_zh": "观察：机器人在右侧边线内转身后折返，腿部与躯干避开第3次来球。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：机器人左右脚交替接触木质球场地面，支撑躯干完成本次躲球。"
+     }
+    ]
+   },
+   "throw_4_dodged": {
+    "method": "visual",
+    "intervals": [
+     [
+      10.52,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 10.52,
+      "log_step": null,
+      "text_zh": "观察：视频 9.52–10.52 s，机器人向场地后方及初始朝向的左侧跑动，第4球从画面右起第2台发球机飞来；视频 10.40–10.52 s 球在机器人身旁下降并触地，视频 10.60–10.64 s 可见反弹，机器人保持站立并继续离开球路。"
+     },
+     {
+      "t": 13.24,
+      "log_step": 17,
+      "text_zh": "推断：log step 17 中机器人服务（基准提供的在线控制及裁判接口）在重置后的 episode 2 返回 outcome=success、stages_reached=5、stages_total=5，且评分器 live_success=1；结合第4球的逐帧飞行与触地画面，判定本节点完成。完成时间按视频触地帧近似标为视频 10.52 s，逐球原生记账时刻未记录，误差约 0.1–0.2 s。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "避开",
+      "note_zh": "观察：机器人双腿带动躯干向初始朝向的左后方跑动，避开第4次来球。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：机器人左右脚交替接触木质球场地面，支撑躯干完成本次躲球。"
+     }
+    ]
+   },
+   "throw_5_dodged": {
+    "method": "visual",
+    "intervals": [
+     [
+      13.24,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 13.24,
+      "log_step": null,
+      "text_zh": "观察：视频 12.00–13.24 s，机器人继续向初始朝向的左后方跑动，第5球从画面左起第2台发球机飞来；视频 13.08–13.20 s 球从机器人身旁下降，视频 13.24 s 末帧球已贴近地面、球影合拢，机器人仍站在场内。"
+     },
+     {
+      "t": 13.24,
+      "log_step": 17,
+      "text_zh": "推断：log step 17 中机器人服务（基准提供的在线控制及裁判接口）在重置后的 episode 2 返回 outcome=success、stages_reached=5、stages_total=5，且评分器 live_success=1；结合第5球的逐帧飞行与触地画面，判定本节点完成。完成时间按视频触地帧近似标为视频 13.24 s，逐球原生记账时刻未记录，误差约 0.1–0.2 s。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "避开",
+      "note_zh": "观察：机器人继续向初始朝向的左后方跑动，腿部与躯干避开第5次来球。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：机器人左右脚交替接触木质球场地面，支撑躯干完成本次躲球。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "dodged",
+   "value": 5,
+   "unit": "count",
+   "threshold": 5,
+   "direction": "higher_better"
+  }
+ ],
+ "failure": null,
+ "summary_zh": "被评分的是重置后的第二回合：机器人先沿右转弧线跑动躲过前两球，在右侧边线内减速转身，再向左后方折返躲过后三球，五个节点最终均完成。视频约 2.88、5.28、8.08、10.52、13.24 s 显示各球触地，机器人保持站立，log step 17 的在线裁判返回 success、5/5 阶段，与评分器 live_success=1 一致。无状态日志或逐球裁判事件时间，完成时刻按视频触地画面近似标注，瞬时接触是否有效由成功裁决辅助确认；会话中的第一回合越界不属于本段被评分视频。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "评分回合为重置后的第二回合，视频13.28 s与log step 17的success（5/5，13.25 s）一致；五个节点均完成，时间与接触标注合理，无失败故障层，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": "2026-11-12T11:32:05Z",
+  "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-dodgeball-i00-standard--yrmvyiq/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-dodgeball-i00-standard--yrmvyiq/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "d41d5d0321365bcad08442a0c3ff767dc54fd0fe68496b36cd7188979e5c4a11",
+   "video_sha256": "8a9656d2369e3077ed3a8a2f597ef7f6ca1eb98564dad2b56594599058ada672",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/dribble_slalom/trajectories/embodied-olympics-fun-dribble-slalom-i00-privileged--kcmg9xv.json

@@ -0,0 +1,653 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-dribble-slalom-i00-privileged--kcmg9xv",
+ "taskKey": "embodied-olympics-fun/dribble_slalom",
+ "dag_sha256": "f4cfa2a0be22cf22f3ad536a52e6a1b08a34617ec27db20275bd0d73953e16cd",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "determinism_checked": 0,
+   "replays": 1,
+   "n_actions": 8480,
+   "sim_steps": 8480,
+   "horizon": 9000,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 13,
+   "stages_total": 13,
+   "fell": 0,
+   "replay_s": 44.43,
+   "bad_trajectory": 0,
+   "metric_max_ball_distance_m": 1.257,
+   "metric_planes_ball": 12.0,
+   "metric_planes_robot": 12.0,
+   "metric_touches": 17.0,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 40,
+   "agent_resets": 50,
+   "agent_replays": 3,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 1.59,
+   "peak_processes": 228,
+   "peak_disk_gb": 0.352,
+   "peak_cpus": 8.05,
+   "peak_pids": 246,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "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": "media",
+  "url": "media:embodied-olympics-fun-dribble-slalom-i00-privileged--kcmg9xv.mp4",
+  "duration_s": 84.84,
+  "fps": 25.0,
+  "frames": 2121,
+  "sha256": "30c4c6f4f95281f18c093d38823aa024b6496d901251285b7f548ecddf53ceda"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 13,
+  "base": 0,
+  "peak": 13,
+  "final": 13,
+  "nodes": {
+   "out_1_passed": {
+    "method": "visual",
+    "intervals": [
+     [
+      5.3,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 5.76,
+      "log_step": null,
+      "text_zh": "观察：视频 4.80–5.76 s，白球在机器人脚边或前方滚动，机器人保持站立行走；两者沿规定方向经过去程第1锥的开局左侧（黄色标记一侧，y > 0），随后继续前进。"
+     },
+     {
+      "t": 5.3,
+      "log_step": 35,
+      "text_zh": "推断：依据上述连续画面，去程第1锥的球心与骨盆均越面约在视频 5.3 s，时刻估读误差约0.5 s；评分器 reward 的 stages_reached=13、metric_planes_ball=12、metric_planes_robot=12，与 log step 35 对已保存交付文件重新回放得到的完整阶段列表一致，支持该检查点有效并按历史保留。"
+     },
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：视频 0.96–1.44 s，机器人脚部碰到白球，白球随后向第1锥滚动；机器人同时以双脚交替着地支撑行走。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频 0.96–1.44 s，机器人脚部碰到白球，白球随后向第1锥滚动。候选：击打；脚部踢球属于短时击打，现有接触方式中无踢球条目。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 4.80–5.76 s，机器人双脚交替接触草地，支撑本体通过去程第1锥。"
+     }
+    ]
+   },
+   "out_2_passed": {
+    "method": "visual",
+    "intervals": [
+     [
+      10.8,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 11.4,
+      "log_step": null,
+      "text_zh": "观察：视频 10.40–11.40 s，白球在机器人脚边或前方滚动，机器人保持站立行走；两者沿规定方向经过去程第2锥的开局右侧（黄色标记一侧，y < 0），随后继续前进。"
+     },
+     {
+      "t": 10.8,
+      "log_step": 35,
+      "text_zh": "推断：依据上述连续画面，去程第2锥的球心与骨盆均越面约在视频 10.8 s，时刻估读误差约0.5 s；评分器 reward 的 stages_reached=13、metric_planes_ball=12、metric_planes_robot=12，与 log step 35 对已保存交付文件重新回放得到的完整阶段列表一致，支持该检查点有效并按历史保留。"
+     },
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：视频 7.68–8.64 s，机器人脚部靠球踢送，白球向第2锥右侧滚动；机器人同时以双脚交替着地支撑行走。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频 7.68–8.64 s，机器人脚部靠球踢送，白球向第2锥右侧滚动。候选：击打；脚部踢球属于短时击打，现有接触方式中无踢球条目。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 10.40–11.40 s，机器人双脚交替接触草地，支撑本体通过去程第2锥。"
+     }
+    ]
+   },
+   "out_3_passed": {
+    "method": "visual",
+    "intervals": [
+     [
+      19.5,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 20.16,
+      "log_step": null,
+      "text_zh": "观察：视频 18.72–20.16 s，白球在机器人脚边或前方滚动，机器人保持站立行走；两者沿规定方向经过去程第3锥的开局左侧（黄色标记一侧，y > 0），随后继续前进。"
+     },
+     {
+      "t": 19.5,
+      "log_step": 35,
+      "text_zh": "推断：依据上述连续画面，去程第3锥的球心与骨盆均越面约在视频 19.5 s，时刻估读误差约0.5 s；评分器 reward 的 stages_reached=13、metric_planes_ball=12、metric_planes_robot=12，与 log step 35 对已保存交付文件重新回放得到的完整阶段列表一致，支持该检查点有效并按历史保留。"
+     },
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：视频 12.96–14.40 s，机器人脚部踢送白球，由第2锥旁转向第3锥左侧；机器人同时以双脚交替着地支撑行走。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频 12.96–14.40 s，机器人脚部踢送白球，由第2锥旁转向第3锥左侧。候选：击打；脚部踢球属于短时击打，现有接触方式中无踢球条目。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 18.72–20.16 s，机器人双脚交替接触草地，支撑本体通过去程第3锥。"
+     }
+    ]
+   },
+   "out_4_passed": {
+    "method": "visual",
+    "intervals": [
+     [
+      23.2,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 24.0,
+      "log_step": null,
+      "text_zh": "观察：视频 22.56–24.00 s，白球在机器人脚边或前方滚动，机器人保持站立行走；两者沿规定方向经过去程第4锥的开局右侧（黄色标记一侧，y < 0），随后继续前进。"
+     },
+     {
+      "t": 23.2,
+      "log_step": 35,
+      "text_zh": "推断：依据上述连续画面，去程第4锥的球心与骨盆均越面约在视频 23.2 s，时刻估读误差约0.5 s；评分器 reward 的 stages_reached=13、metric_planes_ball=12、metric_planes_robot=12，与 log step 35 对已保存交付文件重新回放得到的完整阶段列表一致，支持该检查点有效并按历史保留。"
+     },
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：视频 19.68–20.64 s，机器人脚部将白球踢向第4锥右侧；机器人同时以双脚交替着地支撑行走。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频 19.68–20.64 s，机器人脚部将白球踢向第4锥右侧。候选：击打；脚部踢球属于短时击打，现有接触方式中无踢球条目。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 22.56–24.00 s，机器人双脚交替接触草地，支撑本体通过去程第4锥。"
+     }
+    ]
+   },
+   "out_5_passed": {
+    "method": "visual",
+    "intervals": [
+     [
+      39.5,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 40.32,
+      "log_step": null,
+      "text_zh": "观察：视频 38.88–40.32 s，白球在机器人脚边或前方滚动，机器人保持站立行走；两者沿规定方向经过去程第5锥的开局左侧（黄色标记一侧，y > 0），随后继续前进。"
+     },
+     {
+      "t": 39.5,
+      "log_step": 35,
+      "text_zh": "推断：依据上述连续画面，去程第5锥的球心与骨盆均越面约在视频 39.5 s，时刻估读误差约0.5 s；评分器 reward 的 stages_reached=13、metric_planes_ball=12、metric_planes_robot=12，与 log step 35 对已保存交付文件重新回放得到的完整阶段列表一致，支持该检查点有效并按历史保留。"
+     },
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：视频 29.28–30.24 s 及视频 38.40–39.36 s，机器人在白球后方用脚踢控，白球先横移再向第5锥左侧前进；机器人同时以双脚交替着地支撑行走。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频 29.28–30.24 s 及视频 38.40–39.36 s，机器人在白球后方用脚踢控，白球先横移再向第5锥左侧前进。候选：击打；脚部踢球属于短时击打，现有接触方式中无踢球条目。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 38.88–40.32 s，机器人双脚交��接触草地，支撑本体通过去程第5锥。"
+     }
+    ]
+   },
+   "out_6_passed": {
+    "method": "visual",
+    "intervals": [
+     [
+      43.5,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 44.16,
+      "log_step": null,
+      "text_zh": "观察：视频 42.72–44.16 s，白球在机器人脚边或前方滚动，机器人保持站立行走；两者沿规定方向经过去程第6锥的开局右侧（黄色标记一侧，y < 0），随后继续前进。"
+     },
+     {
+      "t": 43.5,
+      "log_step": 35,
+      "text_zh": "推断：依据上述连续画面，去程第6锥的球心与骨盆均越面约在视频 43.5 s，时刻估读误差约0.5 s；评分器 reward 的 stages_reached=13、metric_planes_ball=12、metric_planes_robot=12，与 log step 35 对已保存交付文件重新回放得到的完整阶段列表一致，支持该检查点有效并按历史保留。"
+     },
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：视频 42.24–43.68 s，机器人脚部靠球后将白球踢过第6锥右侧；机器人同时以双脚交替着地支撑行走。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频 42.24–43.68 s，机器人脚部靠球后将白球踢过第6锥右侧。候选：击打；脚部踢球属于短时击打，现有接触方式中无踢球条目。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 42.72–44.16 s，机器人双脚交替接触草地，支撑本体通过去程第6锥。"
+     }
+    ]
+   },
+   "back_6_passed": {
+    "method": "visual",
+    "intervals": [
+     [
+      55.5,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 56.16,
+      "log_step": null,
+      "text_zh": "观察：视频 54.72–56.16 s，白球在机器人脚边或前方滚动，机器人保持站立行走；两者沿规定方向经过回程第6锥的开局左侧（黄色标记相反一侧，y > 0），随后继续前进。"
+     },
+     {
+      "t": 55.5,
+      "log_step": 35,
+      "text_zh": "推断：依据上述连续画面，回程第6锥的球心与骨盆均越面约在视频 55.5 s，时刻估读误差约0.5 s；评分器 reward 的 stages_reached=13、metric_planes_ball=12、metric_planes_robot=12，与 log step 35 对已保存交付文件重新回放得到的完整阶段列表一致，支持该检查点有效并按历史保留。"
+     },
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：视频 48.00–49.44 s 及视频 52.80–54.24 s，机器人绕到末锥外侧，用脚踢控白球并转为向起点方向运球；机器人同时以双脚交替着地支撑行走。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频 48.00–49.44 s 及视频 52.80–54.24 s，机器人绕到末锥外侧，用脚踢控白球并转为向起点方向运球。候选：击打；脚部踢球属于短时击打，现有接触方式中无踢球条目。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 54.72–56.16 s，机器人双脚交替接触草地，支撑本体通过回程第6锥。"
+     }
+    ]
+   },
+   "back_5_passed": {
+    "method": "visual",
+    "intervals": [
+     [
+      61.6,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 62.4,
+      "log_step": null,
+      "text_zh": "观察：视频 60.96–62.40 s，白球在机器人脚边或前方滚动，机器人保持站立行走；两者沿规定方向经过回程第5锥的开局右侧（黄色标记相反一侧，y < 0），随后继续前进。"
+     },
+     {
+      "t": 61.6,
+      "log_step": 35,
+      "text_zh": "推断：依据上述连续画面，回程第5锥的球心与骨盆均越面约在视频 61.6 s，时刻估读误差约0.5 s；评分器 reward 的 stages_reached=13、metric_planes_ball=12、metric_planes_robot=12，与 log step 35 对已保存交付文件重新回放得到的完整阶段列表一致，支持该检查点有效并按历史保留。"
+     },
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：视频 59.52–60.96 s，机器人脚部接近白球后将其踢向第5锥右侧；机器人同时以双脚交替着地支撑行走。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频 59.52–60.96 s，机器人脚部接近白球后将其踢向第5锥右侧。候选：击打；脚部踢球属于短时击打，现有接触方式中无踢球条目。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 60.96–62.40 s，机器人双脚交替接触草地，支撑本体通过回程第5锥。"
+     }
+    ]
+   },
+   "back_4_passed": {
+    "method": "visual",
+    "intervals": [
+     [
+      67.2,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 67.68,
+      "log_step": null,
+      "text_zh": "观察：视频 66.72–67.68 s，白球在机器人脚边或前方滚动，机器人保持站立行走；两者沿规定方向经过回程第4锥的开局左侧（黄色标记相反一侧，y > 0），随后继续前进。"
+     },
+     {
+      "t": 67.2,
+      "log_step": 35,
+      "text_zh": "推断：依据上述连续画面，回程第4锥的球心与骨盆均越面约在视频 67.2 s，时刻估读误差约0.5 s；评分器 reward 的 stages_reached=13、metric_planes_ball=12、metric_planes_robot=12，与 log step 35 对已保存交付文件重新回放得到的完整阶段列表一致，支持该检查点有效并按历史保留。"
+     },
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：视频 65.76–67.20 s，机器人脚部踢送白球，由第4锥左侧继续向起点前进；机器人同时以双脚交替着地支撑行走。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频 65.76–67.20 s，机器人脚部踢送白球，由第4锥左侧继续向起点前进。候选：击打；脚部踢球属于短时击打，现有接触方式中无踢球条目。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 66.72–67.68 s，机器人双脚交替接触草地，支撑本体通过回程第4锥。"
+     }
+    ]
+   },
+   "back_3_passed": {
+    "method": "visual",
+    "intervals": [
+     [
+      72.5,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 72.96,
+      "log_step": null,
+      "text_zh": "观察：视频 72.00–72.96 s，白球在机器人脚边或前方滚动，机器人保持站立行走；两者沿规定方向经过回程第3锥的开局右侧（黄色标记相反一侧，y < 0），随后继续前进。"
+     },
+     {
+      "t": 72.5,
+      "log_step": 35,
+      "text_zh": "推断：依据上述连续画面，回程第3锥的球心与骨盆均越面约在视频 72.5 s，时刻估读误差约0.5 s；评分器 reward 的 stages_reached=13、metric_planes_ball=12、metric_planes_robot=12，与 log step 35 对已保存交付文件重新回放得到的完整阶段列表一致，支持该检查点有效并按历史保留。"
+     },
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：视频 71.52–72.96 s，机器人在第3锥右侧用脚踢送白球；机器人同时以双脚交替着地支撑行走。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频 71.52–72.96 s，机器人在第3锥右侧用脚踢送白球。候选：击打；脚部踢球属于短时击打，现有接触方式中无踢球条目。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 72.00–72.96 s，机器人双脚交替接触草地，支撑本体通过回程第3锥。"
+     }
+    ]
+   },
+   "back_2_passed": {
+    "method": "visual",
+    "intervals": [
+     [
+      77.5,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 78.24,
+      "log_step": null,
+      "text_zh": "观察：视频 77.28–78.24 s，白球在机器人脚边或前方滚动，机器人保持站立行走；两者沿规定方向经过回程第2锥的开局左侧（黄色标记相反一侧，y > 0），随后继续前进。"
+     },
+     {
+      "t": 77.5,
+      "log_step": 35,
+      "text_zh": "推断：依据上述连续画面，回程第2锥的球心与骨盆均越面约在视频 77.5 s，时刻估读误差约0.5 s；评分器 reward 的 stages_reached=13、metric_planes_ball=12、metric_planes_robot=12，与 log step 35 对已保存交付文件重新回放得到的完整阶段列表一致，支持该检查点有效并按历史保留。"
+     },
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：视频 76.80–78.24 s，机器人在第2锥左侧用脚踢送白球；机器人同时以双脚交替着地支撑行走。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频 76.80–78.24 s，机器人在第2锥左侧用脚踢送白球。候选：击打；脚部踢球属于短时击打，现有接触方式中无踢球条目。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 77.28–78.24 s，机器人双脚交替接触草地，支撑本体通过回程第2锥。"
+     }
+    ]
+   },
+   "back_1_passed": {
+    "method": "visual",
+    "intervals": [
+     [
+      81.8,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 82.08,
+      "log_step": null,
+      "text_zh": "观察：视频 81.12–82.08 s，白球在机器人脚边或前方滚动，机器人保持站立行走；两者沿规定方向��过回程第1锥的开局右侧（黄色标记相反一侧，y < 0），随后继续前进。"
+     },
+     {
+      "t": 81.8,
+      "log_step": 35,
+      "text_zh": "推断：依据上述连续画面，回程第1锥的球心与骨盆均越面约在视频 81.8 s，时刻估读误差约0.5 s；评分器 reward 的 stages_reached=13、metric_planes_ball=12、metric_planes_robot=12，与 log step 35 对已保存交付文件重新回放得到的完整阶段列表一致，支持该检查点有效并按历史保留。"
+     },
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：视频 79.68–81.12 s，机器人脚部踢控白球，白球通过第1锥右侧；机器人同时以双脚交替着地支撑行走。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频 79.68–81.12 s，机器人脚部踢控白球，白球通过第1锥右侧。候选：击打；脚部踢球属于短时击打，现有接触方式中无踢球条目。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 81.12–82.08 s，机器人双脚交替接触草地，支撑本体通过回程第1锥。"
+     }
+    ]
+   },
+   "finish_crossed": {
+    "method": "visual",
+    "intervals": [
+     [
+      84.8,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 84.8,
+      "log_step": null,
+      "text_zh": "观察：视频 83.84–84.80 s，机器人跟随白球朝白色起点线前进；视频 84.64–84.80 s，白球整体滚至线后，机器人骨盆也到达线后，仍保持站立，录像随即结束。"
+     },
+     {
+      "t": 84.8,
+      "log_step": 35,
+      "text_zh": "观察：log step 35 对保存的交付轨迹进行新进程回放，服务返回 success=true、executed_rows=8480、完整13阶段及 finish；log step 36 再生并回放交付轨迹成功，动作数组与先前保存文件逐项一致。评分器 reward 同样为 success=1、replay_success=1、stages_reached=13。"
+     },
+     {
+      "t": 84.8,
+      "log_step": null,
+      "text_zh": "推断：结合末段返线画面及评分器成功结论，将两者均已完成返回越线的节点锚定在视频 84.80 s；不将开局在线后算作返回。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频 82.08–82.56 s，机器人脚部踢送白球向起点线滚动；视频 84.48–84.80 s，脚部再次靠球踢送，白球继续滚至线后。候选：击打；现有接触方式中无踢球条目。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 82.08–84.80 s，机器人以双脚交替着地支撑行走，跟球返回起点线后。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "max_ball_distance_m",
+   "unit": "m",
+   "threshold": 2.0,
+   "direction": "lower_better",
+   "value": 1.257
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人用脚分段踢控白球，依次完成六锥去程、绕末锥折返和六锥回程，并在视频 84.80 s 将整个球及骨盆带回起点线后，13/13个节点完成。评分器记录 success=1、双方各通过12平面、17次触球及最大球距1.257 m；log step 35–36 已保存并验证成功交付，log step 56 的后续会话额度中断发生在此后，不改变本次成功。无逐帧状态日志，球和脚偶有遮挡，越面时刻依据视频连续画面估读约有0.5 s误差，毫米级侧向及接触边界以原生成功验收作补充。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "评分器 success=1、stages_reached=13，视频末帧机器人与球越过起点线，13 个节点全部到达且无失败；校验通过，未发现需要修改的内容。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-11-12T11:35:22Z",
+  "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-dribble-slalom-i00-privileged--kcmg9xv/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-dribble-slalom-i00-privileged--kcmg9xv/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "942262d5cf25aa0d081d4f67502cca283e2cb9ad003219a12856bfc7e868b048",
+   "video_sha256": "30c4c6f4f95281f18c093d38823aa024b6496d901251285b7f548ecddf53ceda",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/dribble_slalom/trajectories/embodied-olympics-fun-dribble-slalom-i00-standard--dyx3zod.json

@@ -0,0 +1,512 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-dribble-slalom-i00-standard--dyx3zod",
+ "taskKey": "embodied-olympics-fun/dribble_slalom",
+ "dag_sha256": "f4cfa2a0be22cf22f3ad536a52e6a1b08a34617ec27db20275bd0d73953e16cd",
+ "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": 9000,
+   "sim_steps": 31932,
+   "calls": 3183,
+   "rows": 3275,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 4863.4,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "resource_enforced": 0,
+   "limits_violation_observed": 1,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 0.401,
+   "peak_processes": 17,
+   "peak_disk_gb": 0.264,
+   "peak_cpus": 7.96,
+   "peak_pids": 82,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "over_cpus": 1
+  }
+ },
+ "video": {
+  "kind": "media",
+  "url": "media:embodied-olympics-fun-dribble-slalom-i00-standard--dyx3zod.mp4",
+  "duration_s": 90.04,
+  "fps": 25.0,
+  "frames": 2251,
+  "sha256": "77cfb410188843f917bb3994d2e46786c0cea23c98d40d40ac6ab9678f1bf4f0"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 13,
+  "base": 0,
+  "peak": 9,
+  "final": 9,
+  "nodes": {
+   "out_1_passed": {
+    "method": "visual",
+    "intervals": [
+     [
+      7.2,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 7.2,
+      "log_step": null,
+      "text_zh": "观察：视频5.2–8.0 s，白球先从第1锥黄色标记侧（开局左侧）滚过，机器人随后沿同侧走过；视频7.6–8.0 s双方已在该检查点去程后方。推断：结合视频6.8–7.6 s的相邻帧估计联合越面约在视频7.2 s，时间误差约0.5–0.8 s；以球和骨盆均通过为准。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频0–7.6 s，机器人脚部踢拨白球向第1锥前进；候选：击打，现有接触方式没有踢球条目，不限定左右脚。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：机器人双脚交替接触草地，支撑身体行进、转向及踢球。"
+     }
+    ]
+   },
+   "out_2_passed": {
+    "method": "visual",
+    "intervals": [
+     [
+      14.0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 14.0,
+      "log_step": null,
+      "text_zh": "观察：视频12.0–14.8 s，机器人和白球转到第2锥黄色标记侧（开局右侧），白球先过，机器人随后通过；视频14.4–14.8 s双方已向第3锥行进。推断：结合视频13.2–14.4 s的相邻帧估计联合越面约在视频14.0 s，时间误差约0.5–0.8 s；以球和骨盆均通过为准。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频8.0–14.4 s，机器人脚部踢拨白球绕到第2锥右侧；候选：击打，现有接触方式没有踢球条目，不限定左右脚。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：机器人双脚交替接触草地，支撑身体行进、转向及踢球。"
+     }
+    ]
+   },
+   "out_3_passed": {
+    "method": "visual",
+    "intervals": [
+     [
+      19.6,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 19.6,
+      "log_step": null,
+      "text_zh": "观察：视频18.0–20.8 s，白球先经第3锥黄色标记侧（开局左侧）前进，机器人跟随；视频20.4–20.8 s双方均已过该锥。推断：结合视频19.2–20.4 s的相邻帧估计联合越面约在视频19.6 s，时间误差约0.5–0.8 s；以球和骨盆均通过为准。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频14.8–20.4 s，机器人脚部踢拨白球绕到第3锥左侧；候选：击打，现有接触方式没有踢球条目，不限定左右脚。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：机器人双脚交替接触草地，支撑身体行进、转向及踢球。"
+     }
+    ]
+   },
+   "out_4_passed": {
+    "method": "visual",
+    "intervals": [
+     [
+      33.6,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 33.6,
+      "log_step": null,
+      "text_zh": "观察：视频30.0–34.4 s，白球和机器人从第4锥黄色标记侧（开局右侧）绕过；白球先过，机器人在锥旁继续小步调整后前进，视频34.0–34.4 s双方已过该锥。推断：结合视频33.2–34.4 s的相邻帧估计联合越面约在视频33.6 s，时间误差约0.5–0.8 s；以球和骨盆均通过为准。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频22.0–34.4 s，机器人脚部多次踢拨白球绕第4锥；候选：击打，现有接触方式没有踢球条目，不限定左右脚。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：机器人双脚交替接触草地，支撑身体行进、转向及踢球。"
+     }
+    ]
+   },
+   "out_5_passed": {
+    "method": "visual",
+    "intervals": [
+     [
+      40.4,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 40.4,
+      "log_step": null,
+      "text_zh": "观察：视频37.2–41.6 s，白球和机器人在第5锥黄色标记侧（开局左侧）前进；白球先过，机器人随后走过，视频41.2–41.6 s双方已朝第6锥移动。推断：结合视频40.0–40.8 s的相邻帧估计联合越面约在视频40.4 s，时间误差约0.5–0.8 s；以球和骨盆均通过为准。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频34.4–41.2 s，机器人脚部踢拨白球绕第5锥；候选：击打，现有接触方式没有踢球条目，不限定左右脚。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：机器人双脚交替接触草地，支撑身体行进、转向及踢球。"
+     }
+    ]
+   },
+   "out_6_passed": {
+    "method": "visual",
+    "intervals": [
+     [
+      44.0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 44.0,
+      "log_step": null,
+      "text_zh": "观察：视频42.0–45.2 s，白球和机器人从第6锥黄色标记侧（开局右侧）去程越面；视频44.4–45.2 s双方已在最末锥外侧，开始准备折返。推断：结合视频43.6–44.4 s的相邻帧估计联合越面约在视频44.0 s，时间误差约0.5–0.8 s；以球和骨盆均通过为准。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频41.2–44.4 s，机器人脚部踢拨白球绕第6锥去程侧；候选：击打，现有接触方式没有踢球条目，不限定左右脚。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：机器人双脚交替接触草地，支撑身体行进、转向及踢球。"
+     }
+    ]
+   },
+   "back_6_passed": {
+    "method": "visual",
+    "intervals": [
+     [
+      60.4,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 60.4,
+      "log_step": null,
+      "text_zh": "观察：视频44.4–58.8 s，机器人带球绕到第6锥另一侧；视频59.2–61.2 s，白球先沿回程方向通过第6锥，机器人随后从开局左侧（黄色标记相反侧）跟过。推断：结合视频60.0–60.8 s的相邻帧估计联合越面约在视频60.4 s，时间误差约0.5–0.8 s；以球和骨盆均通过为准。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频44.4–61.2 s，机器人用脚多次踢拨白球并绕球转位，完成最末锥折返；候选：击打，现有接触方式没有踢球条目，不限定左右脚。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：机器人双脚交替接触草地，支撑身体行进、转向及踢球。"
+     }
+    ]
+   },
+   "back_5_passed": {
+    "method": "visual",
+    "intervals": [
+     [
+      66.0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 66.0,
+      "log_step": null,
+      "text_zh": "观察：视频64.0–68.0 s，白球先从第5锥开局右侧（黄色标记相反侧）返回，机器人随后沿同侧通过；视频68.0–70.0 s双方已在该锥起点方向一侧。推断：结合视频65.6–66.4 s的相邻帧估计联合越面约在视频66.0 s，时间误差约0.5–0.8 s；以球和骨盆均通过为准。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频61.2–68.0 s，机器人用脚踢拨白球向第5锥回程侧前进；候选：击打，现有接触方式没有踢球条目，不限定左右脚。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：机器人双脚交替接触草地，支撑身体行进��转向及踢球。"
+     }
+    ]
+   },
+   "back_4_passed": {
+    "method": "visual",
+    "intervals": [
+     [
+      77.6,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 77.6,
+      "log_step": null,
+      "text_zh": "观察：视频74.0–78.8 s，白球先从第4锥开局左侧（黄色标记相反侧）返回，机器人随后同侧跟过；视频78.0–80.0 s双方已向第3锥前进。推断：结合视频77.2–78.0 s的相邻帧估计联合越面约在视频77.6 s，时间误差约0.5–0.8 s；以球和骨盆均通过为准。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频68.0–78.0 s，机器人脚部踢拨白球绕第4锥回程侧；候选：击打，现有接触方式没有踢球条目，不限定左右脚。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：机器人双脚交替接触草地，支撑身体行进、转向及踢球。"
+     }
+    ]
+   },
+   "back_3_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 87.0,
+      "log_step": null,
+      "text_zh": "观察：视频78.0–90.0 s，机器人带球接近第3锥，在锥的远端反复转位、缓慢向开局右侧运球；视频89.0–90.0 s白球和机器人仍在第3锥的远端，尚未沿回程方向越过该锥平面。"
+     },
+     {
+      "t": 90.0,
+      "log_step": 96,
+      "text_zh": "观察：log step 96执行最后一次重置后的ep6脚本，机器人服务的终局元数据为outcome=time、stages_reached=9、stages_total=13、sim_time_s=90.0；该原生裁判结果经机器人API返回，与视频90.0 s停在第3锥前的终态一致。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频78.0–90.0 s，机器人用脚踢拨白球接近第3锥并调整绕行方向；候选：击打，现有接触方式没有踢球条目。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频78.0–90.0 s，机器人双脚交替支撑草地，以小步和转向重新接近白球。"
+     }
+    ]
+   },
+   "back_2_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 90.0,
+      "log_step": 96,
+      "text_zh": "观察：视频89.0–90.0 s，机器人与球尚在第3锥远端，未到达第2锥回程平面；log step 96原生裁判终局只达到9/13阶段。推断：此有序后续节点未达到。"
+     }
+    ]
+   },
+   "back_1_passed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 90.0,
+      "log_step": 96,
+      "text_zh": "观察：视频89.0–90.0 s，机器人与球尚在第3锥远端，未返回第1锥；log step 96原生裁判终局只达到9/13阶段。推断：此有序后续节点未达到。"
+     }
+    ]
+   },
+   "finish_crossed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 90.0,
+      "log_step": 96,
+      "text_zh": "观察：视频89.0–90.0 s，机器人与球尚在场地中段第3锥附近，未完成十二次绕锥，也未返回画面左端白色起点线；log step 96原生裁判终局只达到9/13阶段。推断：此有序后续节点未达到。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "max_ball_distance_m",
+   "value": null,
+   "unit": "m",
+   "threshold": 2.0,
+   "direction": "lower_better"
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "back_3_passed",
+   "t": 87.0,
+   "log_step": 96,
+   "where": "操作中",
+   "evidence_zh": "观察：视频78.0–90.0 s，机器人正将白球从第4锥带向第3锥回程侧；视频85.0–90.0 s仍在第3锥远端以小步调整和踢拨，未完成球心与骨盆的联合越面。log step 96的最后实时回合原生裁判报告9/13、outcome=time；前九个已完成检查点无需恢复，首个未完成且仍在操作的步骤是回程过3锥。"
+  },
+  "unmet": {
+   "out_1_passed": true,
+   "out_2_passed": true,
+   "out_3_passed": true,
+   "out_4_passed": true,
+   "out_5_passed": true,
+   "out_6_passed": true,
+   "back_6_passed": true,
+   "back_5_passed": true,
+   "back_4_passed": true,
+   "back_3_passed": false,
+   "back_2_passed": false,
+   "back_1_passed": false,
+   "finish_crossed": false
+  },
+  "phenomena": [
+   {
+    "cat": "全身",
+    "item": "步态",
+    "t": 87.0,
+    "log_step": 96,
+    "evidence_zh": "观察：视频78.0–90.0 s，机器人带球向第3锥前进，但多次转位、短步推进，视频85.0–90.0 s仍未把球及骨盆带过该锥回程平面；最后一秒机器人仍由双脚支撑，未见翻倒。推断：可见失败表现是运球行进过慢、距离推进不足；log step 96的回合时限判负说明终止条件。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "认知",
+    "item": "推理/规划",
+    "log_step": 94,
+    "source_zh": "日志 / agent 自述 / 视频",
+    "evidence_zh": "推断：agent保留了每次先绕球对齐、从约0.7 m助跑位置踢球、再追球重新准备的分段策略，整体推进效率不足。log step 94写入的笔记在最终回合前已明确记载该策略在其粗略二维替身中通常需约100 s，并指出约20次踢球各需约5 s；log step 95最终候选的160次替身测试仅4次成功，log step 96仍用该脚本启动最后回合。视频44.0–60.4 s的多次转位折返及视频78.0–90.0 s反复准备、缓慢推进与此策略相符。替身测试是agent选择跨阶段策略的证据，不作为本回合物理状态或必然不可行的证明。"
+   }
+  ],
+  "gap_zh": "没有逐帧状态日志；足球较小，部分触球被脚遮住，联合越面时间根据场地标记和相邻视频帧估读，误差约0.5–0.8 s。log step 97列出的是脚本g球计数变化时刻，不能直接当作球与骨盆共同完成的时间，故节点时间取视频估计。inputs/run.json的评分器字段未报告max_ball_distance_m数值；现有材料不足以进一步确定具体踢球参数或底层跟踪是否有独立缺陷。"
+ },
+ "summary_zh": "最后一次实时回合完成去程六锥及回程第6、5、4锥，共9/13个节点，结束时仍在操作“回程过3锥”。视频85.0–90.0 s显示机器人反复转位、缓慢运球而未越面，原生裁判在回合时限判负；日志支持分段踢球、追球、重新准备的策略整体效率不足。无逐帧状态日志，越面时间约有0.5–0.8 s估读误差，历史最大球距未报告。",
+ "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9/13、断点back_3_passed(操作中)、步态现象与推理/规划根因与视频、log step 97门计数(g6 43.0, g7 59.3, g8 63.8, g9 75.0 s)及原生裁判9/13一致；联合越面时间比球先过的log时刻晚约1-2.5 s，符合球先过、骨盆随后的判据，未发现需改的内容。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-11-12T11:39:19Z",
+  "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-dribble-slalom-i00-standard--dyx3zod/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-dribble-slalom-i00-standard--dyx3zod/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "399a5c5347e975e3346ab49905591cc1930ec744007de46066bad46a6fd3fe8d",
+   "video_sha256": "77cfb410188843f917bb3994d2e46786c0cea23c98d40d40ac6ab9678f1bf4f0",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/dunk/trajectories/embodied-olympics-fun-dunk-i00-privileged--jkrykke.json

@@ -0,0 +1,205 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-dunk-i00-privileged--jkrykke",
+ "taskKey": "embodied-olympics-fun/dunk",
+ "dag_sha256": "41266fca43525d52c45613fe2a45c7e2c37be81bdeaab1e09fc71cbf2428713c",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "determinism_checked": 0,
+   "replays": 1,
+   "n_actions": 479,
+   "sim_steps": 329,
+   "horizon": 1500,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 2,
+   "stages_total": 2,
+   "fell": 0,
+   "replay_s": 4.65,
+   "bad_trajectory": 0,
+   "metric_ball_height_m": 0.131,
+   "metric_basket_touch_s": 0.113,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 531,
+   "agent_resets": 522,
+   "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": 6.338,
+   "peak_processes": 19,
+   "peak_disk_gb": 0.249,
+   "peak_cpus": 8.0,
+   "peak_pids": 91,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "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": "media",
+  "url": "media:embodied-olympics-fun-dunk-i00-privileged--jkrykke.mp4",
+  "duration_s": 3.32,
+  "fps": 25.0,
+  "frames": 83,
+  "sha256": "e90712d4a949511c237c788d331e240ca2a998b39fe8dce3ab01f5fd2e5ae312"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "dunk_landed": {
+    "method": "visual",
+    "intervals": [
+     [
+      3.28,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.6,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–1.12 s，机器人双手从篮球两侧共同持球，屈膝下蹲后将球上举；视频 1.28–1.44 s 可见双脚腾空、球随双手升到篮圈上方。"
+     },
+     {
+      "t": 1.52,
+      "log_step": null,
+      "text_zh": "观察：逐帧查看视频 1.40–1.60 s，机器人双手随球向下运动，橙球由篮圈上方进入圈内并出现在网下；视频 1.68–1.80 s，球继续下落，双手与球明显分离。"
+     },
+     {
+      "t": 2.28,
+      "log_step": null,
+      "text_zh": "观察：视频约 1.80 s，机器人双脚回到地面，随后屈膝缓冲；逐帧查看视频 2.28–3.28 s，机器人保持脚部着地、躯干近竖直且位姿基本不变，双手仍举起，球已弹离机器人，未见跌倒。"
+     },
+     {
+      "t": null,
+      "log_step": 83,
+      "text_zh": "观察：log step 83 中，agent 生成并保存交付轨迹，重放服务返回 success=true、failed=null、stages=[dunk, stand]、steps=329、仿真 time_s=3.29 s；轨迹共有 479 行动作，与所选候选动作逐项相同。评分器 reward 同样报告 replay_success=1、stages_reached=2/2、fell=0，原生指标与该重放一致。"
+     },
+     {
+      "t": 3.28,
+      "log_step": 83,
+      "text_zh": "推断：视频的双手扣球、松球和落地站立顺序与 log step 83 的交付重放及评分器成功结果一致，因此合并事件已完成；达成时刻约取末帧视频 3.28 s，不能把球在视频约 1.56 s 穿圈时就记为整个节点完成。无状态日志，精确站稳计数与接触间隙由评分结果支持。"
+     },
+     {
+      "t": 1.4,
+      "log_step": 81,
+      "text_zh": "推断：视频 0.00–1.48 s 的局部放大显示两只手从相对两侧夹持球，未见手指闭合包握；log step 11、48 的控制代码将双手闭合命令设为零，log step 81 增加的是两掌间距收紧，log step 83 将该候选生成并验证为交付轨迹。因此实际双手接触记为开掌夹持，而非抓握。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "候选：夹持。视频 0.00–1.48 s，机器人左右手从篮球两侧相向夹住球，共同举球并向篮圈下压；两掌收紧后球向下脱离。未见单手手指闭合包握，因此抓握不适用；托/舀也不能概括两侧相向夹持。视频 1.68–1.80 s 可见两手与球分离。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "视频 0.00–1.12 s，左右脚支撑机器人持球下蹲、蹬地；腾空后视频约 1.80 s 双脚落地承重，机器人屈膝缓冲，并在视频 2.28–3.28 s 保持站立。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "basket_touch_s",
+   "value": 0.113,
+   "unit": "s",
+   "threshold": 0.3,
+   "direction": "lower_better"
+  },
+  {
+   "name": "ball_height_m",
+   "value": 0.131,
+   "unit": "m",
+   "threshold": null,
+   "direction": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人完成双手持球起跳、从上方压球穿圈、松球和落地站稳，唯一节点“双手扣篮站稳”在视频约 3.28 s 达成，最终为 1/1。log step 83 的交付轨迹已成功重放，且其阶段与指标和评分器结果一致；log step 90 的会话限制提示出现在保存验证之后，不改变本次成功。无状态日志，视频分辨率与采样率不足以独立复核毫秒级双手接触间隙、穿圈几何阈值及精确站稳计数，这些细节以评分器成功结果为依据，节点时刻为近似锚点。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "成功运行，唯一节点 dunk_landed 达成、final=1，与视频（起跳持球、球穿圈、松手、落地站立）和评分器一致；contact_actual 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": "2026-11-12T11:39:29Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-dunk-i00-privileged--jkrykke/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-dunk-i00-privileged--jkrykke/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "fe9778baa4e21a61870b9dcd492f9d512bdc5fe95d142ca25ea23a9a98a3c4f0",
+   "video_sha256": "e90712d4a949511c237c788d331e240ca2a998b39fe8dce3ab01f5fd2e5ae312",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/dunk/trajectories/embodied-olympics-fun-dunk-i00-standard--zybvham.json

@@ -0,0 +1,215 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-dunk-i00-standard--zybvham",
+ "taskKey": "embodied-olympics-fun/dunk",
+ "dag_sha256": "41266fca43525d52c45613fe2a45c7e2c37be81bdeaab1e09fc71cbf2428713c",
+ "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": 182,
+   "sim_steps": 783,
+   "calls": 799,
+   "rows": 783,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 4737.6,
+   "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": 7.383,
+   "peak_processes": 16,
+   "peak_disk_gb": 2.024,
+   "peak_cpus": 4.01,
+   "peak_pids": 337,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "media",
+  "url": "media:embodied-olympics-fun-dunk-i00-standard--zybvham.mp4",
+  "duration_s": 1.84,
+  "fps": 25.0,
+  "frames": 46,
+  "sha256": "30448ff4bd17d2555fc811b7931f380d6bc8b7eb4a05eead4fb1a3607369a0c2"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "dunk_landed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–1.20 s，机器人双手从篮球两侧夹持并举球，同时屈膝下蹲；视频 1.36–1.60 s 双脚离地、篮球随双手升到篮圈前方，已开始扣篮尝试。"
+     },
+     {
+      "t": 1.8,
+      "log_step": 123,
+      "text_zh": "观察：视频 1.64–1.80 s，篮球仍在篮圈前沿外侧的双手附近，未出现从圈上方向下穿出篮网的动作，视频终止时机器人尚在空中。log step 123 的服务回报明确对应 reset 后的 episode 6：done=True、outcome=the ball left the hands before the dunk、stages=0、control_steps=182；log step 124 再次记录该回合仿真终点 1.82 s。"
+     },
+     {
+      "t": 1.8,
+      "log_step": 123,
+      "text_zh": "推断：本次评分采用第六次实时回合（reward.graded_by_live=1，log step 123），其视频与服务的扣篮前失球判负一致，因此“双手扣篮站稳”从未完成；此前测试和本地模拟的入圈结果不计入进度。成功参考在参考视频 1.64–1.92 s 可见球由圈上方压入并穿出，本次视频没有这一过程。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 0.00–1.60 s，机器人左右张开手掌从篮球两侧相向夹持并共同举球，未见单手手指合拢包握；候选：夹持。随后在篮圈前失去接触，服务判据见 log step 123。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–1.24 s，机器人双脚接地支承持球下蹲和蹬地；视频 1.36 s 已腾空，视频结束前未见落地站稳。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "basket_touch_s",
+   "unit": "s",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "ball_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": null,
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "dunk_landed",
+   "t": 1.8,
+   "log_step": 123,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 1.36–1.80 s，机器人持球起跳后把球举至篮圈前方，尚未压球穿圈便结束；log step 123 的第六回合服务结果为 the ball left the hands before the dunk、stages=0。断点是正在执行而未完成的“双手扣篮站稳”。"
+  },
+  "unmet": {
+   "dunk_landed": false
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没保持住",
+    "t": 1.8,
+    "log_step": 123,
+    "evidence_zh": "推断：视频 0.00–1.60 s 篮球随机器人两手共同上移，说明此前已建立并保持持球接触；视频 1.64–1.80 s 球未进圈而停留在圈前的手掌附近。log step 123 的本回合服务判定为扣篮前双手均离球，故为已有持球接触随后丢失；远景无法直接分辨最后的毫米级间隙，失球时刻以视频约 1.80 s 为导航锚点。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "t": 1.8,
+    "log_step": 124,
+    "source_zh": "日志 / 视频",
+    "evidence_zh": "推断：当前控制器在腾空送球阶段未能协调手臂跟踪与持球接触。log step 113 的驱动代码用加速度阈值触发腾空轨迹，并在触发前冻结计划索引；log step 123 实际打印 takeoff detected at step 138 planned 129 shift 6，log step 124 的末段记录执行索引持续落后计划六步，右肩滚转目标与实测差在记录索引 176 达 0.17 rad。视频 1.36–1.80 s 显示机器人虽起跳举球，却在球进入圈前失去持球接触（log step 123 服务判负）；这些证据支持控制执行问题，但不足以唯一确定是触发阈值、跟踪饱和还是夹持载荷失配。"
+   }
+  ],
+  "gap_zh": "无状态日志，固定远景不能精确复核双手末次接触及 0.03 秒容差，失球依据为已确认对应第六评分回合的服务判负。log step 124 的中段本体数据在输入中已截断；agent 在 log step 127、129 自述起跳检测晚十二步且撞到肩部限位，但缺少完整传感与限位证据，未将该具体因果链当作已证实结论。两项 DAG 原生指标未见于评分器字段，均填 null。"
+ },
+ "summary_zh": "评分回合为第六次实时尝试：机器人双手夹球下蹲、蹬地腾空并将球举到篮圈前方，但未完成压球穿圈及落地站稳，节点最终为 0/1。视频约 1.80 s 的断点位于扣篮操作中，服务确认扣篮前失球，现象为没保持住；日志中的轨迹重定时偏移与肩部跟踪偏差支持控制问题。无状态日志且部分本体记录截断，无法进一步确定失球的具体控制机制。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "评分回合为第六次实时回合（服务 done=True、control_steps=182、sim_time 1.82 s≈视频 1.84 s），视频显示持球起跳升至篮圈前、未穿圈，dunk_landed=false、断点、没保持住及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": "2026-11-12T11:38:49Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-dunk-i00-standard--zybvham/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-dunk-i00-standard--zybvham/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "b38d3ab6b8ef6c65fd880ff09db6656f60be1460c540a43a83424cd264a570c2",
+   "video_sha256": "30448ff4bd17d2555fc811b7931f380d6bc8b7eb4a05eead4fb1a3607369a0c2",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/dunk_between_legs/trajectories/embodied-olympics-fun-dunk-between-legs-i00-privileged--kvkrh5l.json

@@ -0,0 +1,257 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-dunk-between-legs-i00-privileged--kvkrh5l",
+ "taskKey": "embodied-olympics-fun/dunk_between_legs",
+ "dag_sha256": "b3a9022ee9aaa233f14f035ce0b249f6a6cd4ac0723730a08b64571a639c20ad",
+ "outcome": {
+  "success": 0,
+  "reward": 0.6,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "determinism_checked": 0,
+   "replays": 1,
+   "n_actions": 443,
+   "sim_steps": 443,
+   "horizon": 4000,
+   "episode_over": 0,
+   "progress": 0.6,
+   "stages_reached": 3,
+   "stages_total": 5,
+   "fell": 0,
+   "replay_s": 8.93,
+   "bad_trajectory": 0,
+   "metric_ball_height_m": 0.971,
+   "metric_basket_touch_s": 0.0,
+   "metric_between_the_legs": 1.0,
+   "metric_dribbles": 1.0,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 147,
+   "agent_resets": 156,
+   "agent_replays": 3,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 1.771,
+   "peak_processes": 17,
+   "peak_disk_gb": 0.24,
+   "peak_cpus": 5.8,
+   "peak_pids": 38,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "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": "media",
+  "url": "media:embodied-olympics-fun-dunk-between-legs-i00-privileged--kvkrh5l.mp4",
+  "duration_s": 4.44,
+  "fps": 25.0,
+  "frames": 111,
+  "sha256": "ffaa705cb3ca00187add3f6a07df2ba128a22cd0d1c83ee8fc06673f0da8c63e"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 3,
+  "base": 0,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "between_legs_bounce": {
+    "method": "visual",
+    "intervals": [
+     [
+      2.92,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 2.6,
+      "log_step": null,
+      "text_zh": "观察：视频 2.32–2.52 s，机器人双臂把胸前篮球向下、向后摆送并放开；视频 2.56–2.68 s 球下降至双足之间的地面附近，随后从机器人身后向上反弹，视频 2.88–2.96 s 手重新接触球。"
+     },
+     {
+      "t": 2.92,
+      "log_step": 132,
+      "text_zh": "观察：log step 132 生成并回放最终交付文件，裁判记录 dribble、between the legs、gather 三阶段；穿越记录 inside=true、speed=1.838、z=0.218，评分器 metric_between_the_legs=1。推断：结合视频 2.60 s 附近的反弹和视频 2.92 s 附近的重新触球，胯下运球在后者得到确认；按视频帧取约视频 2.92 s，单帧不能分开相隔很短的胯下确认与双手收球。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 2.32–2.52 s，左右手在球两侧共同夹持并摆送篮球，随后张开放球。候选接触名：双掌夹持；没有单手包住球，现有方式没有双掌侧向挤持项，不能标抓握。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–2.92 s，机器人双脚在地面支撑躯干，保持站姿完成摆球与反弹。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "避开",
+      "note_zh": "推断：视频 2.56–2.88 s 球经两腿下方到达身后，未见球撞腿改变轨迹；log step 132 最终交付回放确认合法胯下反弹且没有判负，支持双腿避开篮球。"
+     }
+    ]
+   },
+   "ball_gathered": {
+    "method": "visual",
+    "intervals": [
+     [
+      2.92,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 2.92,
+      "log_step": 132,
+      "text_zh": "观察：视频 2.76–2.88 s，机器人双臂伸到身后迎接上升篮球；视频 2.92–3.20 s，篮球停在背后双手位置，不再自由反弹；逐帧检查视频 3.40–4.40 s，球持续留在那里。log step 132 的最终交付回放由裁判明确确认 gather，支持侧视中部分重叠的左右手同时接球；时间取约视频 2.92 s。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 2.92–4.40 s，双手在机器人背后从篮球两侧共同夹住球；log step 132 的裁判 gather 确认双手接触。候选接触名：双掌夹持；没有单手包住球，现有词表没有双掌侧向夹持项，不能标抓握。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 2.92–4.40 s，双脚承重，机器人站在原位置完成接球和后续持球���"
+     }
+    ]
+   },
+   "dunk_landed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 3.2,
+      "log_step": 132,
+      "text_zh": "观察：视频 0.00–4.40 s 全程机器人留在起始区域；视频 2.92 s 收球后至视频 4.40 s，球保持在背后腰部附近，机器人没有向篮筐行进、起跳或把球送到篮圈。log step 132 的交付生成脚本在背后接球及保持后直接保存，回放仅有 dribble、between the legs、gather，success=false、failed=null、episode_over=false。扣篮及其落地验收从未完成。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "basket_touch_s",
+   "value": 0.0,
+   "unit": "s",
+   "threshold": 0.3,
+   "direction": "lower_better"
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "dunk_landed",
+   "t": 3.0,
+   "log_step": 132,
+   "where": "首步前",
+   "evidence_zh": "观察：视频约 2.92 s 完成收球后，机器人到视频 4.40 s 仍站在原处把球夹在背后，没有开始向篮筐移动或起跳；log step 132 保存的最终轨迹只执行到接球后保持，log step 134 明确记载扣篮控制器未加入交付。断点为尚未启动的扣篮落地站稳步骤，视频 3.0 s 为收球后停留的导航锚点。"
+  },
+  "unmet": {
+   "between_legs_bounce": true,
+   "ball_gathered": true,
+   "dunk_landed": false
+  },
+  "phenomena": [
+   {
+    "cat": "没尝试",
+    "item": "没尝试",
+    "t": 3.0,
+    "log_step": 132,
+    "evidence_zh": "观察：视频 3.00–4.40 s，机器人保持背后持球站姿，篮筐仍在前方远处，没有扣篮动作；log step 132 最终生成代码在接球保持后保存，log step 134 写明扣篮未成功所以未放入轨迹。此前会话中的近篮测试不属于这段交付录像。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "evidence_zh": "推断：会话中扣篮动作的控制未能实现，导致最终交付只保留收球以前的部分。独立证据是 log step 81 的两次近篮测试都由裁判报告收球后、扣篮前脱手，log step 93 的正常模式复测再次同样失败；这些是另设近篮起点及持球阶段的会话测试，仅支持控制能力归因，不能当作本录像发生脱手。log step 134 的交付说明明确把失败的扣篮控制器排除，log step 136 自述扣篮从未成功且尚未建立向篮筐运球的控制流程；证据不足以细分为接触设计、轨迹或反馈控制。",
+    "source_zh": "日志"
+   }
+  ],
+  "gap_zh": "没有状态日志或成功参考录像；侧视中左右手与双腿局部重叠，精确穿越几何、双手接触先后由最终交付的裁判阶段记录辅助确认。扣篮仅在会话的修改初态测试中出现，交付没有扣篮段，因此不能把测试中的脱手、篮圈碰撞或落地表现标为本次失败现象，也不能确定控制根因的具体子类。"
+ },
+ "summary_zh": "机器人在视频约 2.92 s 完成合法胯下反弹并在背后双手收球，三个语义节点最终完成两个；原生裁判的五阶段完成三个，进度为0.6。交付随后保持原地持球，扣篮落地站稳从未启动；会话近篮测试反复在扣篮前脱手，支持笼统的控制能力不足，但这些测试不属于交付录像。没有状态日志和成功参考录像，局部接触细节依靠最终交付的裁判记录补充，控制子类无法确定。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "视频帧与标签一致：约2.6 s球从胯下反弹，约2.9 s背后收球，之后站立持球至4.4 s，扣篮未开始；断点与首步前/没尝试合理。",
+    "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": "2026-11-12T11:37:03Z",
+  "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-dunk-between-legs-i00-privileged--kvkrh5l/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-dunk-between-legs-i00-privileged--kvkrh5l/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "29b7ce938841b7b1a60fe07df3296155a38fe9e8fe2d5f27d4e5240e7cbc204e",
+   "video_sha256": "ffaa705cb3ca00187add3f6a07df2ba128a22cd0d1c83ee8fc06673f0da8c63e",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/dunk_between_legs/trajectories/embodied-olympics-fun-dunk-between-legs-i00-standard--6sjrual.json

@@ -0,0 +1,260 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-dunk-between-legs-i00-standard--6sjrual",
+ "taskKey": "embodied-olympics-fun/dunk_between_legs",
+ "dag_sha256": "b3a9022ee9aaa233f14f035ce0b249f6a6cd4ac0723730a08b64571a639c20ad",
+ "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": 606,
+   "sim_steps": 3686,
+   "calls": 19,
+   "rows": 3686,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 5610.7,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "resource_enforced": 0,
+   "limits_violation_observed": 1,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 6.415,
+   "peak_processes": 26,
+   "peak_disk_gb": 2.023,
+   "peak_cpus": 11.27,
+   "peak_pids": 349,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "over_cpus": 1
+  }
+ },
+ "video": {
+  "kind": "media",
+  "url": "media:embodied-olympics-fun-dunk-between-legs-i00-standard--6sjrual.mp4",
+  "duration_s": 6.08,
+  "fps": 25.0,
+  "frames": 152,
+  "sha256": "a759c0d0d3da75e150090c7cdbe3ddc6485c8ae174d88f8c62a73b64668cea7d"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 3,
+  "base": 0,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "between_legs_bounce": {
+    "method": "visual",
+    "intervals": [
+     [
+      6.0,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 5.72,
+      "log_step": 130,
+      "text_zh": "观察：视频 5.60–5.76 s，机器人将胸前篮球向下、向后送出，双臂随后分开；视频 5.88–6.00 s，球从双腿低处反弹至髋后，机器人双手向该处合拢。log step 130 的最终第6局返回阶段数3及收球后脱手的终止原因。"
+     },
+     {
+      "t": 6.0,
+      "log_step": 130,
+      "text_zh": "推断：视频 5.88–6.04 s 的反弹、穿腿后再触球与 log step 130 最终第6局已越过胯下运球阶段的服务结果一致，故记首次再触球确认约在视频 6.00 s；膝部附近有遮挡，具体穿越速度和门面积不能从画面直接量测。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 0.00–5.72 s，左右张开的手从篮球两侧夹持，预摆后一起下摆送球并放开。候选接触名：双掌侧夹送球；单手没有包住球，不能标为抓握，现有表没有双掌侧夹项（D12）。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：机器人先横向迈开站距，随后双脚在地板上支撑身体完成下摆和反弹动作（视频 0.00–6.04 s）。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "避开",
+      "note_zh": "推断：视频 5.88–6.00 s 球经两腿低处反弹至身后，log step 130 最终第6局继续进入收球阶段；结合该阶段验收，双腿在这次穿越中让开篮球，未以腿击球。"
+     }
+    ]
+   },
+   "ball_gathered": {
+    "method": "visual",
+    "intervals": [
+     [
+      6.04,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 6.04,
+      "log_step": 130,
+      "text_zh": "观察：视频 5.96–6.04 s，反弹球升至机器人髋后，左右手向球两侧合拢，末帧球仍在髋后手部附近；log step 130 最终第6局服务结果为阶段数3，终止原因为“the ball left your hands after the gather, before the dunk”。"
+     },
+     {
+      "t": 6.04,
+      "log_step": 130,
+      "text_zh": "推断：结合视频 6.00–6.04 s 的合掌接球与 log step 130 明确的收球后脱手结果，双手曾同时触球并触发收球；以视频约 6.04 s 作近似锚点。此节点是锁存事件，随后失控不取消已经发生的合法收球；开局胸前持球不计入。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 5.96–6.04 s，机器人左右手从两侧向髋后的篮球合拢；log step 130 确认曾进入双手收球阶段。候选接触名：双掌侧夹接球；没有单手包住球的可见证据，现有表缺少双掌侧夹项（D12），且夹持未持续。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察��视频 5.96–6.04 s，机器人双脚着地支撑身体，同时在髋后接球，未发生起跳或跌倒。"
+     }
+    ]
+   },
+   "dunk_landed": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 6.04,
+      "log_step": 130,
+      "text_zh": "观察：视频 0.00–6.04 s，机器人始终在罚球线附近，球只在胸前、腿间及髋后运动，始终未被送到篮圈上方；视频末段没有向篮筐助跑、起跳或扣穿动作。log step 130 最终第6局在收球后脱手时结束，未进入扣篮阶段。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "basket_touch_s",
+   "value": null,
+   "unit": "s",
+   "threshold": 0.3,
+   "direction": "lower_better"
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "ball_gathered",
+   "t": 6.04,
+   "log_step": 130,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 5.96–6.04 s，机器人仍在髋后合掌接反弹球，尚未开始向篮圈搬运或起跳；log step 130 最终第6局报告收球后、扣篮前完全脱手并终止。推断：失败发生在这次收球动作未能转为稳定持球时，故断点取双手收球，尽管瞬时双手触球已使该事件节点锁存为真。"
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没保持住",
+    "t": 6.04,
+    "log_step": 130,
+    "evidence_zh": "观察：视频 6.00–6.04 s，机器人在髋后合掌触及反弹球；log step 130 最终第6局的服务结果明确为“the ball left your hands after the gather, before the dunk”。推断：接球接触已建立，随后完全失去手球接触，属于没保持住；画面在此附近结束，没有后续重新接住。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "source_zh": "日志",
+    "evidence_zh": "推断：log step 129 将预先生成的髋后接球目标与闭合时序固定到最终脚本，log step 130 执行后虽触发双手收球却立即脱手；视频 5.96–6.04 s 也显示机器人在执行该合掌接球动作。该接球控制未能把瞬时触球转成持续持球；现有证据不足以在接触几何、夹持力和轨迹跟踪之间进一步归因。"
+   }
+  ],
+  "gap_zh": "没有状态日志或成功参考视频。已逐帧检查视频 5.00–6.04 s，但场馆远景与身体遮挡使双手同时触球及随后完全脱手的准确物理时刻不可直接分辨，阶段判定结合 log step 130 的最终第6局服务结果，视频锚点为近似值。没有最终接球的接触力和逐步手球位姿记录，根因只标控制；basket_touch_s 未报告。",
+  "unmet": {
+   "between_legs_bounce": true,
+   "ball_gathered": true,
+   "dunk_landed": false
+  }
+ },
+ "summary_zh": "评分局为 log step 130 的最终第6局：机器人横向开步后把球向下送过两腿，在视频约 6.00–6.04 s 完成胯下反弹确认及双手收球，按事件锁存保留2/3节点。收球动作随即完全脱手，未开始扣篮，断点为双手收球，现象为没保持住，根因保守标为控制。远景遮挡且没有状态日志，瞬时手球接触和精确脱手时刻依赖该局服务结果，无法细分控制失效机制。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "评分局（log step 130，606 个 100Hz 控制步≈6.06 s，与 6.08 s 视频吻合）的节点进度 2/3、dunk 未达成、现象没保持住与根因控制均与视频和服务结果一致；时钟与 log 引用合规。唯一可商榷处：断点落在已锁存为真的 ball_gathered，但该节点判据含“球心入圈前不得完全脱手”，标注者的理由自洽，不作为错误。",
+    "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": "2026-11-12T11:36:27Z",
+  "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-dunk-between-legs-i00-standard--6sjrual/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-dunk-between-legs-i00-standard--6sjrual/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "ffad3c88cb679d8439ae21f3e53018ab91ce727698c0a4485bec23ef31499878",
+   "video_sha256": "a759c0d0d3da75e150090c7cdbe3ddc6485c8ae174d88f8c62a73b64668cea7d",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/free_kick_wall/trajectories/embodied-olympics-fun-free-kick-wall-i00-privileged--9aue7f9.json

@@ -0,0 +1,206 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-free-kick-wall-i00-privileged--9aue7f9",
+ "taskKey": "embodied-olympics-fun/free_kick_wall",
+ "dag_sha256": "6749223ae6b8a425b1f7d86513bee960c64b42ed756a390e3a358baebdc119ab",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "determinism_checked": 0,
+   "replays": 1,
+   "n_actions": 582,
+   "sim_steps": 282,
+   "horizon": 3000,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 2,
+   "stages_total": 2,
+   "fell": 0,
+   "replay_s": 3.44,
+   "bad_trajectory": 0,
+   "metric_ball_x_m": 20.557,
+   "metric_cross_y_m": 1.788,
+   "metric_cross_z_m": 0.14,
+   "metric_fell_after_kick": 1.0,
+   "metric_kick_speed_m_s": 13.24,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 49,
+   "agent_resets": 6243,
+   "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": 3.168,
+   "peak_processes": 17,
+   "peak_disk_gb": 0.24,
+   "peak_cpus": 7.91,
+   "peak_pids": 38,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "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": "media",
+  "url": "media:embodied-olympics-fun-free-kick-wall-i00-privileged--9aue7f9.mp4",
+  "duration_s": 2.84,
+  "fps": 25.0,
+  "frames": 71,
+  "sha256": "6e057dabc740b0e89ec772aa198c2526cb26c6945400ea03340ed5ca22fcc5ce"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "free_kick_scored": {
+    "method": "visual",
+    "intervals": [
+     [
+      2.8,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.76,
+      "log_step": null,
+      "text_zh": "观察：视频 0.56–0.72 s，机器人左脚着地，右腿摆向原地足球；视频 0.76–0.80 s，足球离开右脚向前飞出，随后机器人倾倒，视频约 1.2 s 已倒在草地上。"
+     },
+     {
+      "t": 2.8,
+      "log_step": 22,
+      "text_zh": "观察：视频 1.4–1.76 s，白球从画面中橙色人墙的左侧向后方行进；墙体随后遮住部分球路，视频 2.56–2.80 s 白球在墙体右侧、球门下方附近再次可见。最终交付文件的重放输出（log step 22）为 success=True、executed_rows=282，球心 x=20.557 m、越线横向偏差 1.788 m、高度 0.14 m；inputs/run.json 的评分器字段与这些数值一致，stages_reached=2/2、replay_success=1。"
+     },
+     {
+      "t": 2.8,
+      "log_step": 22,
+      "text_zh": "推断：评分器的越线偏差和高度均在 DAG 接受范围内，任意球入门事件已完成，标在视频约 2.8 s。log step 22 是最终交付文件的重放，282 个控制步×0.01 s=仿真 2.82 s，与视频末段相符；机器人触球在视频 0.72–0.76 s，与 agent 在 log step 24 自述的仿真触球约 0.74 s 相符。远景及人墙遮挡使精确越线帧难以单独目测，时间取近似值；未将 agent 工作时间换算为视频时间。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频 0.72–0.80 s，机器人右脚摆向足球，白球随即离脚向前运动。候选名称：击打；现有接触方式没有踢击对应项，故用其他。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.56–0.76 s，机器人左脚接触草地，在右腿摆动踢球时支承身体；踢出后机器人失去支撑并倒地。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "ball_x_m",
+   "unit": "m",
+   "threshold": 20.49,
+   "direction": "higher_better",
+   "value": 20.557
+  },
+  {
+   "name": "cross_y_m",
+   "unit": "m",
+   "threshold": [
+    -3.66,
+    3.66
+   ],
+   "direction": "target",
+   "value": 1.788
+  },
+  {
+   "name": "cross_z_m",
+   "unit": "m",
+   "threshold": [
+    0,
+    2.44
+   ],
+   "direction": "target",
+   "value": 0.14
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人以左脚支撑、右脚踢出足球，白球绕过人墙后完成任意球入门，唯一节点在视频约 2.8 s 达成，终态为 1/1。机器人在踢球后倒地，但按本任务规则不撤销进球，评分器重放成功且越线横向偏差为 1.788 m、高度为 0.14 m。无状态日志，远景和人墙遮挡限制了精确越线时刻的目测，节点时间���末段画面结合最终交付重放记录近似确定。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "单节点任意球入门，评分器成功、越线数值在范围内；视频中脚踢、球飞向人墙方向、踢后倒地与标注相符，节点时间约2.8 s（视频时钟）可接受，接触标注符合S16。",
+    "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": "2026-11-12T11:36:47Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-free-kick-wall-i00-privileged--9aue7f9/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-free-kick-wall-i00-privileged--9aue7f9/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "1207877c8b8a23ca160d0be810bc42361fcac4e9279eec0556d8e222e75992ec",
+   "video_sha256": "6e057dabc740b0e89ec772aa198c2526cb26c6945400ea03340ed5ca22fcc5ce",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/free_kick_wall/trajectories/embodied-olympics-fun-free-kick-wall-i00-standard--vqdavrh.json

@@ -0,0 +1,229 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-free-kick-wall-i00-standard--vqdavrh",
+ "taskKey": "embodied-olympics-fun/free_kick_wall",
+ "dag_sha256": "6749223ae6b8a425b1f7d86513bee960c64b42ed756a390e3a358baebdc119ab",
+ "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": 1194,
+   "sim_steps": 1194,
+   "calls": 663,
+   "rows": 1009,
+   "resets": 0,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 790.6,
+   "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": 5.274,
+   "peak_processes": 13,
+   "peak_disk_gb": 0.468,
+   "peak_cpus": 4.0,
+   "peak_pids": 121,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "media",
+  "url": "media:embodied-olympics-fun-free-kick-wall-i00-standard--vqdavrh.mp4",
+  "duration_s": 11.96,
+  "fps": 25.0,
+  "frames": 299,
+  "sha256": "910d2d308a5423ae4de2795b3b5da38cdd059ec4ecec37f80cdfa6aa567f6182"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "free_kick_scored": {
+    "method": "visual",
+    "intervals": [
+     [
+      11.9,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 9.76,
+      "log_step": null,
+      "text_zh": "观察：视频 9.60–9.72 s 机器人右脚向白球前摆，视频 9.76–9.84 s 白球离开原位并向前上方飞出，左脚仍落在草地上；此前粗查视频 0–9.5 s 未见白球离开原位。"
+     },
+     {
+      "t": 10.44,
+      "log_step": null,
+      "text_zh": "观察：视频 10.04–10.44 s 白球从画面中橙色人墙左侧飞过并继续朝白色球门运动；机器人在踢球后倒地，视频 10.44–11.92 s 仍躺在草地上。"
+     },
+     {
+      "t": 11.92,
+      "log_step": 28,
+      "text_zh": "观察：视频 10.40–10.64 s 白球位于门框的二维投影内，但视频 10.40 s 随球移动的深色地面阴影仍在人墙底座的画面下方；视频 11.20 s 白球出现在人墙右侧、门框投影下方，随后视频 11.40–11.92 s 白球再次上升至门框投影内。逐帧检查视频 11.60–11.92 s，球仍向画面右上方移动。log step 27 的在线机器人服务裁判返回 outcome=success、stages=2/2；log step 28 查询同一 episode=1，服务返回 done=True、reason=success、control_steps=1194、sim_time_s=11.94（仿真秒）；inputs/run.json 的 reward.live_success=1、graded_by_live=1、resets=0。该服务状态是本次在线仿真回合的裁判反馈。"
+     },
+     {
+      "t": 11.9,
+      "log_step": 28,
+      "text_zh": "推断：视频 10.40 s 球与地面阴影的位置及视频 11.20 s 球在人墙右侧、门框投影下方的后续画面，说明球较早进入门框投影不能作为已越过球门平面的证据。结合视频 11.60–11.92 s 球继续朝门区运动和 log step 27–28 的本回合裁判成功记录，保留首次达成约在视频 11.9 s 的估计。仿真 11.94 s 是服务报告的成功终止时刻，不直接当作视频越线帧；无状态日志，无法独立确定精确越线帧，也无法量化时间误差。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：视频 9.72–9.80 s 机器人右脚前摆击打足球，随后足球与脚分离并飞向球门。候选名称：击打；接触方式词表没有踢击项，故用其他。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 9.60–9.84 s 机器人左脚接触草地，在右脚摆动踢球时支撑身体；随后机器人在视频约 10.4 s 倒地。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "ball_x_m",
+   "unit": "m",
+   "threshold": 20.49,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "cross_y_m",
+   "unit": "m",
+   "threshold": [
+    -3.66,
+    3.66
+   ],
+   "direction": "target",
+   "value": null
+  },
+  {
+   "name": "cross_z_m",
+   "unit": "m",
+   "threshold": [
+    0,
+    2.44
+   ],
+   "direction": "target",
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人在视频约 9.7 s 用右脚踢出足球，球绕过人墙后朝球门运动；结合末段画面及 log step 27–28 的在线裁判成功记录，任意球入门暂估在视频约 11.9 s 完成，最终为 1/1 节点。视频约 10.4 s 球进入门框二维投影时，地面阴影仍在人墙前侧，视频约 11.2 s 球又出现在人墙右侧、门框投影下方，故不能据投影重叠将入门提前至视频约 10.5 s。视频约 10.4 s 的踢后倒地不否决入门，failure 为 null。置信度由 8 降为 7：无状态日志，无法独立确定精确越线帧，保留的末段达成时间仍为推断；撤回原先约半秒误差的无依据量化。仿真 11.94 s 仅为服务报告的成功终止时刻。评分器未报告 ball_x_m、cross_y_m、cross_z_m，三个原生指标均保留 null。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 2,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "fix",
+    "conf": 8,
+    "summary_zh": "节点值（达成、final=1）和接触标注正确，无 failure 正确；但达成时间 11.9 s 错：逐帧可见白球在视频约 10.4–10.5 s 就已进入球门框，11.9 s 只是回合结束时间（done、sim_time 11.94）而非越线时间。",
+    "n_issues": 1,
+    "n_major": 1
+   },
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 2,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "节点达成、final=1、接触标注、failure=null 均正确；达成时间约视频 11.9 s 有依据：服务在 sim 11.94 s 以 success 终止回合且视频止于 11.96 s，视频 10.4 s 球与门框的重叠只是二维投影（11.2 s 球又回到门框下方），故不应提前到 10.5 s。",
+    "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": "2026-11-12T12:17:52Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-free-kick-wall-i00-standard--vqdavrh/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-free-kick-wall-i00-standard--vqdavrh/04_review_r1"
+   },
+   {
+    "stage": "reconcile",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-free-kick-wall-i00-standard--vqdavrh/05_reconcile_r1"
+   },
+   {
+    "stage": "traj_review",
+    "round": 2,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-free-kick-wall-i00-standard--vqdavrh/04_review_r2"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "d08c787494e2e0c89ebbf3a3e3c7d93aa0d71f5114727be395cbb77dc3f28114",
+   "video_sha256": "910d2d308a5423ae4de2795b3b5da38cdd059ec4ecec37f80cdfa6aa567f6182",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/free_throw/trajectories/embodied-olympics-fun-free-throw-i00-privileged--qpubbhs.json

@@ -0,0 +1,187 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-free-throw-i00-privileged--qpubbhs",
+ "taskKey": "embodied-olympics-fun/free_throw",
+ "dag_sha256": "c2d4b07a9ca0ff585b81cafb3eecf9f2484164969c066ba3cd5ef111abe016a5",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "determinism_checked": 0,
+   "replays": 1,
+   "n_actions": 481,
+   "sim_steps": 481,
+   "horizon": 1200,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 2,
+   "stages_total": 2,
+   "fell": 0,
+   "replay_s": 5.76,
+   "bad_trajectory": 0,
+   "metric_ball_height_m": 2.792,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 373,
+   "agent_resets": 132,
+   "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": 5.184,
+   "peak_processes": 18,
+   "peak_disk_gb": 0.251,
+   "peak_cpus": 8.04,
+   "peak_pids": 289,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "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": "media",
+  "url": "media:embodied-olympics-fun-free-throw-i00-privileged--qpubbhs.mp4",
+  "duration_s": 4.84,
+  "fps": 25.0,
+  "frames": 121,
+  "sha256": "ca59bc891b490a34b260a48b0aea41b914c878fcd9e6b81f57a62aa0d30c6235"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "free_throw_scored": {
+    "method": "visual",
+    "intervals": [
+     [
+      4.8,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00 s，机器人站在罚球区，橙色篮球位于两只手之间；视频 0.00–2.80 s，双手逐渐转到球的下侧共同承托，球未离开机器人双手。"
+     },
+     {
+      "t": 3.6,
+      "log_step": null,
+      "text_zh": "观察：视频 3.20–3.44 s，机器人将双手间的篮球降到大腿前方；视频 3.52–3.64 s，双臂向前上方摆动，随后篮球与双手分离并向篮筐飞行，机器人双脚仍着地。"
+     },
+     {
+      "t": 4.8,
+      "log_step": null,
+      "text_zh": "观察：逐帧检查视频 3.60–4.80 s，篮球先上升再下降；视频 4.56–4.76 s，球从篮圈上方接近并向下穿过圈口，视频 4.80 s 位于篮圈下方的篮网内。节点到达时间取视频约 4.80 s；成功参考视频末段呈现相同的自上而下入网结果。"
+     },
+     {
+      "t": 4.8,
+      "log_step": 92,
+      "text_zh": "观察：inputs/run.json 的评分器字段 success=1、replay_success=1、stages_reached=2、stages_total=2；log step 90 生成并保存最终 481 行轨迹，log step 91 检查重新生成文件一致，log step 92 对该最终文件再次重放，基准重放工具返回 success=True，并记录离手于仿真 3.584 s、穿圈及成功于仿真 4.807 s。此处以视频帧确定视频时间，重放结果补足穿圈精确几何验收证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "托/舀",
+      "note_zh": "观察：视频 0.00 s 两只张开的灵巧手从篮球两侧相向夹持；随后双手转到球的下侧共同承托，视频 3.20–3.60 s 双臂先降球再向前上方托送并放开。按主要送球接触记托/舀，初始相向夹持是此前依次发生的接触；画面未见单手合拢包住篮球，不记抓握。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–4.80 s，机器人双脚接触球场地面、双腿支撑躯干，在双手调整持球、降球和投出期间维持站姿。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "ball_height_m",
+   "value": 2.792,
+   "unit": "m",
+   "threshold": null,
+   "direction": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "观察：本次评分轨迹中，机器人双手调整持球并从下方托送，篮球在视频约 3.60 s 离手，随后沿弧线下降，在视频约 4.80 s 穿圈进入篮网，罚球命中节点最终完成 1/1；评分器报告成功。观察：log step 90–92 对最终交付文件的保存、再生成一致性检查和重放均与该入篮结果相符。观察：本次没有状态日志，视频分辨率为 640×360、帧率为 25 fps，不能独立核验毫秒级接触间断、各手指接触点及精确穿圈交点，精确验收由评分器和最终文件重放结果支持。",
+ "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：罚球命中节点在视频约4.8 s到达，帧中球在3.6 s后离手、4.2 s在高处、4.8 s入网，与评分器success=1一致；contact_actual（双臂协作托/舀 + 腿/躯干支撑）符合画面，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": "2026-11-12T11:38:24Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-free-throw-i00-privileged--qpubbhs/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-free-throw-i00-privileged--qpubbhs/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "83cca2730768c7534fca147a7dcbea92600829850eabe0bbb99cb4ae8005b970",
+   "video_sha256": "ca59bc891b490a34b260a48b0aea41b914c878fcd9e6b81f57a62aa0d30c6235",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/free_throw/trajectories/embodied-olympics-fun-free-throw-i00-standard--qumb5xv.json

@@ -0,0 +1,187 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-free-throw-i00-standard--qumb5xv",
+ "taskKey": "embodied-olympics-fun/free_throw",
+ "dag_sha256": "c2d4b07a9ca0ff585b81cafb3eecf9f2484164969c066ba3cd5ef111abe016a5",
+ "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": 347,
+   "sim_steps": 1016,
+   "calls": 15,
+   "rows": 730,
+   "resets": 2,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 2020.0,
+   "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": 1.436,
+   "peak_processes": 12,
+   "peak_disk_gb": 2.444,
+   "peak_cpus": 2.76,
+   "peak_pids": 213,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "media",
+  "url": "media:embodied-olympics-fun-free-throw-i00-standard--qumb5xv.mp4",
+  "duration_s": 3.48,
+  "fps": 25.0,
+  "frames": 87,
+  "sha256": "cad255bedfe5535f8df8c83a571220b760c66f7b306e7128fd291ed18b8418ff"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "free_throw_scored": {
+    "method": "visual",
+    "intervals": [
+     [
+      3.47,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.8,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–1.80 s，机器人双脚着地，左右张开的手从球两侧共同夹持篮球，并将球从胸前举至头上、移向头后；视频 0.80 s 的两手位于球的相对两侧，未见单手合指包握。"
+     },
+     {
+      "t": 2.12,
+      "log_step": null,
+      "text_zh": "观察：逐帧检查视频 1.84–2.24 s，机器人双臂向上前方挥动，视频 2.08 s 球已升至伸出的手上方，视频 2.12 s 双手张开且球已离开；随后篮球独立飞向篮筐。"
+     },
+     {
+      "t": 3.44,
+      "log_step": null,
+      "text_zh": "观察：逐帧检查视频 2.40–3.44 s，篮球先上升至篮板上方，再向篮圈下降；视频 3.32–3.44 s 球从篮圈上方进入圈及球网所在区域。末段机器人前倾但双脚仍支撑地面，未见手、膝或躯干着地。"
+     },
+     {
+      "t": 3.47,
+      "log_step": 57,
+      "text_zh": "推断：log step 57 的第三回合从 control_steps=0 开始，服务裁判最终报告 control_steps=347、sim_time_s=3.47、done=true、reason=success、outcome=success、stages_reached=2/2；评分字段 live_success=1、graded_by_live=1 与之相符。log step 43 给出控制周期 0.01 s，故本回合 control_steps×0.01 对应视频时间；180 步的举球完成与视频 1.80 s 头后持球相符，347 步对应视频约 3.47 s，也与视频 3.32–3.44 s 的入圈末段及 3.48 s 片长相符。因此将完整命中验收锚定为视频约 3.47 s；这是本次现场回合裁判记录，并非 agent 自述或前两回合测试结果。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 0.00–2.08 s，左右张开的灵巧手在篮球两侧相向夹持，共同举球并向前上方送出，随后分开离球。候选：夹持；未见任一手合拢包住篮球，故不用抓握；相向夹紧也不能仅由托/舀描述。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–3.44 s，机器人左右脚与球场地面接触，支撑举球、挥臂及出手后的身体；末段身体前倾，双脚仍承担支撑。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "ball_height_m",
+   "value": null,
+   "unit": "m",
+   "threshold": null,
+   "direction": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "计分视频对应 log step 57 的第三回合：机器人双手夹持举球后从头上投出，篮球在视频末段自上方进入篮圈，现场裁判在约视频 3.47 s 确认两项原生阶段完成，罚球命中节点最终为 1/1。log step 46、54 的前两次跌倒属于重置前测试，不计入本段轨迹。无状态日志，25 fps 视频无法独立核验毫秒级手球接触及精确穿圈交点，完整验收时刻结合本回合控制步与服务���判成功记录确定；评分未报告 ball_height_m，其值留空。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "成功运行，单节点 free_throw_scored 终值 true，时间锚定于服务裁判 control_steps 与视频片尾，接触标注（双臂协作+其他、腿/躯干+支撑）与帧相符，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": "2026-11-12T11:45:39Z",
+  "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-free-throw-i00-standard--qumb5xv/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-free-throw-i00-standard--qumb5xv/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "99adf9e91bc44f20a0168f909c8d009a05295dd599ce6f22fcf23ce19680ddf1",
+   "video_sha256": "cad255bedfe5535f8df8c83a571220b760c66f7b306e7128fd291ed18b8418ff",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}

phaseF/embodied-olympics-fun/freeski_360/trajectories/embodied-olympics-fun-freeski-360-i00-privileged--eni8tqf.json

@@ -0,0 +1,204 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-freeski-360-i00-privileged--eni8tqf",
+ "taskKey": "embodied-olympics-fun/freeski_360",
+ "dag_sha256": "861efa8ce64a6c121798b9927345b65cbb761cb3b91579bd06a292aa528db572",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "determinism_checked": 0,
+   "replays": 1,
+   "n_actions": 1824,
+   "sim_steps": 1824,
+   "horizon": 4000,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 4,
+   "stages_total": 4,
+   "fell": 0,
+   "replay_s": 15.36,
+   "bad_trajectory": 0,
+   "metric_air_s": 1.492,
+   "metric_from_knoll_m": -1.77,
+   "metric_height_m": 2.21,
+   "metric_ride_turn_deg": 22.3,
+   "metric_speed_m_s": 16.442,
+   "metric_trick_flip_deg": -28.6,
+   "metric_trick_flips": 0.0,
+   "metric_trick_half_turns": 2.0,
+   "metric_trick_pelvis_spin_deg": 314.6,
+   "metric_trick_short_deg": 8.3,
+   "metric_trick_spin_deg": 351.7,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 82,
+   "agent_resets": 0,
+   "agent_replays": 44,
+   "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": 1.981,
+   "peak_processes": 19,
+   "peak_disk_gb": 0.26,
+   "peak_cpus": 8.0,
+   "peak_pids": 36,
+   "resource_memory_kills": 0,
+   "resource_pids_kills": 0,
+   "resource_disk_kills": 0,
+   "resource_gpu_memory_kills": 0,
+   "resource_monitor_kills": 0,
+   "resource_emergency_kills": 0,
+   "resource_limit_exceeded": 0,
+   "resource_cpu_throttles": 0,
+   "resource_cpu_throttled_sec": 0.0,
+   "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": "media",
+  "url": "media:embodied-olympics-fun-freeski-360-i00-privileged--eni8tqf.mp4",
+  "duration_s": 18.28,
+  "fps": 25.0,
+  "frames": 457,
+  "sha256": "cf4f2d6c2ce0b5f0ad0773e898b5a22a6dba276c5844433a834b58378a59f2b2"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "spin_360_ridden_away": {
+    "method": "visual",
+    "intervals": [
+     [
+      18.24,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 18.24,
+      "log_step": null,
+      "text_zh": "观察：已查看本次视频全程逐秒取样、视频12.5–14.7 s每0.2 s取样、视频17.2–18.2 s每0.1 s取样及视频18.24 s末帧；固定场地远景中机器人仅为极小的模糊痕迹，不能可靠分辨雪板航向、首次触雪或跨越终点。成功参考视频全程逐秒取样及视频17.60 s末帧也有同样可见性限制，不能据此空景判本次未完成。"
+     },
+     {
+      "t": null,
+      "log_step": 104,
+      "text_zh": "观察：inputs/run.json的reward报告success=1、replay_success=1、stages_reached=4/4、fell=0，空中雪板转角351.7°、空翻数0、滑出最大转向22.3°。log step 104中官方回放工具针对已保存的trajectory.npz报告success=true、start/take-off/landing/ride-away四阶段全部完成、1824步及仿真18.24 s；log step 105确认重新生成的文件与该文件逐值相同，log step 106再次回放成功。这些结果对应交付文件，区别于之前和之后的参数、噪声试验。"
+     },
+     {
+      "t": 18.24,
+      "log_step": null,
+      "text_zh": "推断：依据本次评分器成功和四阶段完成，完整的“转体落地滑出”事件在回放结束前成立，final=true。将已检视的视频18.24 s末帧作为完成的近似锚点；该时间是视频末帧位置，并非从agent时间换算，也不是从远景直接测出的首次过线时刻，实际首次达成时间无法独立视觉定位。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：视频12.5–14.7 s与视频17.2–18.24 s的远景无法辨认机器人双靴、雪板与雪面的实际接触，且无状态接触日志；不能由任务的典型支撑方案代填实际接触。评分器虽确认整次滑雪动作成功，画面仍不能独立核实作用端与触雪方式。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "trick_spin_deg",
+   "unit": "deg",
+   "threshold": 360,
+   "direction": "target",
+   "value": 351.7
+  },
+  {
+   "name": "landing_heading_deg",
+   "unit": "deg",
+   "threshold": 45,
+   "direction": "lower_better",
+   "value": [
+    -15.7,
+    -18.1
+   ]
+  },
+  {
+   "name": "ride_turn_deg",
+   "unit": "deg",
+   "threshold": 90,
+   "direction": "lower_better",
+   "value": 22.3
+  }
+ ],
+ "failure": null,
+ "summary_zh": "评分器确认本次交付回放完成“转体落地滑出”节点，最终1/1：四阶段全部通过，雪板空���转角351.7°、空翻数0，滑出最大转向22.3°，无失败断点。原生落地航向列表[-15.7°, -18.1°]来自log step 104中官方回放工具对交付文件的评分结果，其余原生指标也见reward。证据缺口是本次与成功参考均为固定远景，无法可靠辨认具体动作与接触，也无状态日志；视频18.24 s仅作末帧完成锚点，首次过线时刻不确定，实际接触保留未知。",
+ "conf": 6,
+ "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，评分器四阶段4/4、fell=0，节点final=true、接触未知均有依据，无失败断点；时间锚点已声明为末帧近似。",
+    "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": "2026-11-12T11:39:06Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-freeski-360-i00-privileged--eni8tqf/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-freeski-360-i00-privileged--eni8tqf/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "90bf8dc807b5c56c5fa1f3d6132134067089f933c52580a013cc12449b6be6d2",
+   "video_sha256": "cf4f2d6c2ce0b5f0ad0773e898b5a22a6dba276c5844433a834b58378a59f2b2",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; local graded video"
+  },
+  "writer_note": "Writer: GPT-6.1 Sol (gpt-6.1-sol, high effort) in Codex; reviewer: Claude Sonnet 5.5 (claude-sonnet-5-5) in Claude Code. The same writer and reviewer as the phase-2 labels; same prompts (v3.2), same phase-2 task DAGs, same validation.",
+  "writer_access": "ChatGPT subscription"
+ }
+}