Add 123 trajectories (phase 2, remaining runs; batch S37)

phase2/embodied-olympics-fun/catch/trajectories/embodied-olympics-fun-catch-i00-privileged--3sxt4sq.json

@@ -0,0 +1,209 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-i00-privileged--3sxt4sq",
+ "taskKey": "embodied-olympics-fun/catch",
+ "dag_sha256": "61132849363e640170264bbd47c4c4cebc23f7ecfe683cae11a4abb07d10d44b",
+ "outcome": {
+  "success": 0,
+  "reward": 0.5,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "n_actions": 397,
+   "sim_steps": 397,
+   "horizon": 900,
+   "episode_over": 1,
+   "progress": 0.5,
+   "stages_reached": 1,
+   "stages_total": 2,
+   "fell": 0,
+   "replay_s": 4.94,
+   "bad_trajectory": 0,
+   "metric_catch_height_m": 1.387,
+   "metric_catch_speed_m_s": 9.398,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 38,
+   "agent_resets": 87,
+   "agent_replays": 1,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 0.751,
+   "peak_processes": 13,
+   "peak_disk_gb": 0.028,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-catch-i00-privileged--3sxt4sq/replay.mp4",
+  "duration_s": 4.0,
+  "fps": 25.0,
+  "frames": 100,
+  "sha256": "9c68b3fcefef1e8bf568d22fce8ed88ae0211195d201220026017952abff9699"
+ },
+ "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 机器人站在场地内，手中无球；视频 1.60–3.60 s 机器人双足移动并逐渐伸出右臂迎球，全段至视频 3.96 s 未见摔倒。"
+     },
+     {
+      "t": 3.96,
+      "log_step": 158,
+      "text_zh": "观察：逐帧检查视频 3.00–3.96 s，白球在视频 3.92 s 位于右手上方，在视频 3.96 s 到达右手处。log step 158 对实际交付 trajectory.npz 的独立进程重放由原生裁判返回 stages=[catch]、progress=0.5、success=false、failed=the ball came out of the hand after the catch，executed_rows=397；run.json 的 deterministic=1、stages_reached=1/2、n_actions=397 与此相符。"
+     },
+     {
+      "t": 3.96,
+      "log_step": 164,
+      "text_zh": "推断：右手曾短暂包住球，但随后失去包围接触，未完成同手保持100个控制步的验收，故完整节点从未成立；原生catch阶段不单独计作DAG节点。log step 164 再生成的交付文件哈希相同且仍报告接住后脱手；未把会话内其他试验当作本次视频进度。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "推断：视频 3.92–3.96 s 白球下降到机器人伸出的右手处；log step 158 对交付文件的原生裁判重放确认catch阶段，支持右手手指与拇指或掌部曾从相对两侧包住球，随后脱手。远景无法分辨具体指节。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 1.60–3.96 s 机器人双足在地面交替支撑并向落点移动，伸右手接球时仍直立，未见非足部触地。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "catch_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": null,
+   "value": 1.387
+  },
+  {
+   "name": "catch_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": null,
+   "value": 9.398
+  },
+  {
+   "name": "landed_from_start_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": null,
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "fly_ball_caught",
+   "t": 3.96,
+   "log_step": 158,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 3.92–3.96 s 机器人正伸右手接下降的白球；log step 158 的交付文件重放裁判确认接住后球脱手、只达到catch而未达到hold，评分在397个控制步后结束。本断点位于接稳飞球步骤的持球操作中，视频 3.96 s 为末段接球处的近似导航锚点。"
+  },
+  "unmet": {
+   "fly_ball_caught": false
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没保持住",
+    "t": 3.96,
+    "log_step": 158,
+    "evidence_zh": "推断：视频 3.92–3.96 s 白球到达右手；log step 158 原生裁判对实际交付文件明确报告catch后球从手中出来，且只有catch阶段、没有hold阶段。因此是建立包围抓握后未保持住，未见机器人摔倒；细小脱手过程由裁判结果佐证，不能从远景直接辨清。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "抓取和接触设计",
+    "alt_feasible": "未知",
+    "log_step": 165,
+    "source_zh": "agent 自述 / 日志",
+    "evidence_zh": "推断：agent 在 log step 165 自述最终拦截姿态使球浅入指缘，未深入预闭合的手内便弹出；log step 157 的交付说明也承认该接触几何不能保持球。log step 158 对交付文件的独立重放确认短暂catch后脱手，log step 161 的闭手时机和掌倾角变体仍未成功，支持所用包围接触布局不稳定这一解释。指缘浅入的具体几何仍仅来自agent自述，待核实；未验证本次模型与初态下可行的替代抓法。"
+   }
+  ],
+  "gap_zh": "观察：inputs/提供视频和完整运行记录，但没有状态日志。视频为640×360远景、25 fps，球与指节仅占少数像素；视频 3.92–3.96 s 只能看出白球到达右手，无法独立核实包围接触方向、0.02秒中断及具体脱手瞬间，短暂接住后丢失依据交付文件的原生裁判结果。指缘浅入的几何解释来自 log step 165 的agent自述，缺少该交付轨迹逐物理步接触与力记录；成功参考视频仅用于比较外观，不能验证当前轨迹的微观接触或替代抓法。"
+ },
+ "summary_zh": "机器人移动到落点附近并伸右手接球，原生裁判确认短暂接住，但随即脱手，未完成保持窗口，完整DAG节点为0/1。断点在视频约 3.96 s 的接稳飞球操作中，现象为没保持住；抓取和接触设计根因依据agent对球浅入指缘的自述及交付重放结果推断。无状态日志，远景无法独立核实指节接触、毫秒级失握及具体接触几何。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "节点 fly_ball_caught=false、断点、现象（没保持住）与根因均有 log step 158/164 的原生裁判重放佐证，视频帧显示机器人移动并伸臂、未摔倒，与标注一致；未发现需要修改的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T23:59:41Z",
+  "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-catch-i00-privileged--3sxt4sq/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-privileged--3sxt4sq/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "ad56d57f48b67a307704beff984962df42c23626cb26235da34f03f171f23144",
+   "video_sha256": "9c68b3fcefef1e8bf568d22fce8ed88ae0211195d201220026017952abff9699",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch/trajectories/embodied-olympics-fun-catch-i00-privileged--jmjt3tc.json

@@ -0,0 +1,193 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-i00-privileged--jmjt3tc",
+ "taskKey": "embodied-olympics-fun/catch",
+ "dag_sha256": "61132849363e640170264bbd47c4c4cebc23f7ecfe683cae11a4abb07d10d44b",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "n_actions": 494,
+   "sim_steps": 494,
+   "horizon": 900,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 2,
+   "stages_total": 2,
+   "fell": 0,
+   "replay_s": 10.77,
+   "bad_trajectory": 0,
+   "metric_catch_height_m": 1.343,
+   "metric_catch_speed_m_s": 6.906,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 23,
+   "agent_resets": 681,
+   "agent_replays": 2,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 0.727,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.251,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-catch-i00-privileged--jmjt3tc/replay.mp4",
+  "duration_s": 4.96,
+  "fps": 25.0,
+  "frames": 124,
+  "sha256": "d4ee3e745b26d8363a369a1b6b3ac7b3ec768fda871fe2a624c4094814395849"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "fly_ball_caught": {
+    "method": "visual",
+    "intervals": [
+     [
+      4.94,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–3.52 s 的顺序抽帧中，机器人从初始位置迈步移动，随后抬起右臂；起始没有持球，尚未完成接稳飞球。"
+     },
+     {
+      "t": 3.96,
+      "log_step": null,
+      "text_zh": "观察：逐帧检查视频 3.76–4.92 s；视频 3.88–3.92 s 白色小球仍在伸出手的上方，视频 3.96 s 到达手边，视频 4.00–4.92 s 持续留在该手末端并随手移动；机器人双脚支撑身体，没有倒地。"
+     },
+     {
+      "t": 4.92,
+      "log_step": null,
+      "text_zh": "观察：视频 4.92 s 末帧仍见小球留在伸出的手边。inputs/run.json 的原生评分器字段 success=1、replay_success=1、deterministic=1、stages_reached=2、stages_total=2、fell=0，确认本次交付重放通过接球及保持验收。"
+     },
+     {
+      "t": 4.94,
+      "log_step": 57,
+      "text_zh": "推断：log step 57 的交付脚本在控制步索引 394 下达右手全闭合命令；log step 6 的接口说明给出每控制步 0.01 s，故对应视频约 3.94 s，已用视频 3.92–3.96 s 小球到手的画面核对。评分器记录交付重放共 494 个控制步并成功终止，故接稳窗口完成约在视频 4.94 s（约一帧误差），与视频 4.92 s 末帧及 4.96 s 总长相符；此前只有接球，尚未完成整段保持。"
+     },
+     {
+      "t": 4.92,
+      "log_step": 62,
+      "text_zh": "观察：agent 在 log step 62 自述交付文件为 494 行动作，两次独立重放均通过 catch、hold；这与评分器记录及视频末段一致。远景不能逐一分辨四指、拇指或掌部接触及 0.02 s 中断，微观包围接触和保持计数以本次原生裁判成功验收佐证。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "推断：视频 3.96–4.92 s 小球到达并留在机器人伸出的右手末端；log step 57 的交付脚本控制右手闭合，结合本次原生裁判通过包围接触和保持验收，判为右手抓握棒球。远景不能分辨各手指的具体接触面。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–3.52 s 机器人双脚交替落地迈步到接球位置，视频 3.96–4.92 s 接球持球期间双脚继续支撑身体，未见摔倒。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "catch_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": null,
+   "value": 1.343
+  },
+  {
+   "name": "catch_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": null,
+   "value": 6.906
+  },
+  {
+   "name": "landed_from_start_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": null,
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人迈步到接球位置后，用右手接住下降的小球并保持，约在视频 4.94 s 完成接稳飞球节点，最终 1/1；本次原生评分器确认接球、保持均通过且未摔倒。无状态日志，远景只能看清球到手并持续留在��边，不能直接分辨接触夹角和毫秒级中断，相关细节依据本次原生裁判成功记录；完成时刻由交付控制步数映射并用视频到手事件核对。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "节点 fly_ball_caught 在视频约 4.94 s 达成、最终 1/1、成功无 failure，与原生评分器字段和视频末段一致；控制步映射已说明并用视频到手事件核对，接触标签合理。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:08:09Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-privileged--jmjt3tc/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-privileged--jmjt3tc/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "4b9cfc852c8086200bf1e44a97b33891e14c0d3a3be229cc679a35a5f53fb6c1",
+   "video_sha256": "d4ee3e745b26d8363a369a1b6b3ac7b3ec768fda871fe2a624c4094814395849",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch/trajectories/embodied-olympics-fun-catch-i00-privileged--p3lvwnd.json

@@ -0,0 +1,188 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-i00-privileged--p3lvwnd",
+ "taskKey": "embodied-olympics-fun/catch",
+ "dag_sha256": "61132849363e640170264bbd47c4c4cebc23f7ecfe683cae11a4abb07d10d44b",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "n_actions": 494,
+   "sim_steps": 494,
+   "horizon": 900,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 2,
+   "stages_total": 2,
+   "fell": 0,
+   "replay_s": 5.92,
+   "bad_trajectory": 0,
+   "metric_catch_height_m": 1.33,
+   "metric_catch_speed_m_s": 9.049,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 103,
+   "agent_resets": 308,
+   "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": 2.065,
+   "peak_processes": 20,
+   "peak_disk_gb": 0.988,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-catch-i00-privileged--p3lvwnd/replay.mp4",
+  "duration_s": 4.96,
+  "fps": 25.0,
+  "frames": 124,
+  "sha256": "809887032e910649cd55be616abe9213d05b202f235db7eb5bbabc5f49fcf872"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "fly_ball_caught": {
+    "method": "visual",
+    "intervals": [
+     [
+      4.92,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00 s 机器人站在场地上，手中没有球；视频 0.40–3.60 s 机器人迈步移动并将一臂伸向来球方向，此时尚未完成接稳。"
+     },
+     {
+      "t": 3.96,
+      "log_step": null,
+      "text_zh": "观察：逐帧检查视频 3.44–4.92 s，视频 3.92 s 小白球在伸出的手上方，视频 3.96 s 球到达该手，随后到视频 4.92 s 均留在手边并随手略微上抬；另一只手在腰侧，机器人保持站立。此处初次接到球还不是满保持窗口的完成时刻。"
+     },
+     {
+      "t": 4.92,
+      "log_step": 193,
+      "text_zh": "观察：inputs/run.json 的 reward 报告 success=1、replay_success=1、deterministic=1、stages_reached=2/2、fell=0；log step 193 中对最终交付文件的两次独立重放均报告 success=True、stages=[catch,hold]、494行且末态一致。log step 191 再生成的交付文件与此前成功文件逐项相同。"
+     },
+     {
+      "t": 4.92,
+      "log_step": 170,
+      "text_zh": "推断：结合视频 3.92 s 球尚在空中、视频 3.96–4.92 s 球一直随同一只手运动，以及原生裁判确认 catch、hold 两阶段全部通过，接稳节点在片尾约视频 4.92 s 达成并保持至结束；到达时间以所见末帧近似标记。log step 170 的交付重放报告仿真 4.94 s 结束，与片尾相符，但未将 agent 会话时间换算成视频时间。远景不能逐项辨认手指接触方向和0.02秒中断，完整验收依原生成功结果佐证。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "推断：视频 3.96–4.92 s 小球停留在机器人伸出的右手中并随手运动，结合 log step 194 对右臂接球的描述和原生 catch、hold 验收通过，右手手指与拇指或掌部包住棒球；另一只手在腰侧，未参与夹持。远景不足以单独辨认各指接触点。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.40–3.60 s 机器人用双腿迈步接近来球位置；视频 3.96–4.92 s 双脚在地面支撑身体，持球期间保持站立。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "catch_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": null,
+   "value": 1.33
+  },
+  {
+   "name": "catch_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": null,
+   "value": 9.049
+  },
+  {
+   "name": "landed_from_start_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": null,
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人移动到接球位置，视频 3.96 s 右手接到下降的小球，随后同手持球并保持站立至视频 4.92 s 末帧；接稳飞球节点在片尾约视频 4.92 s 完成，最终达到1/1节点。原生评分及最终文件的独立重放均确认接球和保持两阶段通过，没有失败断点。无状态日志，远景无法精确核验接触夹角、毫秒级中断和保持计数，完整验收由裁判成功结��佐证，节点完成时间为片尾近似值。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "成功运行，单节点接稳飞球在片尾约 4.92 s 达成，与视频 3.96 s 接到球加约1秒保持及原生 success/stages 2/2 一致；接触标签合理，无失败断点。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:05:07Z",
+  "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-catch-i00-privileged--p3lvwnd/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-privileged--p3lvwnd/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "f524b9aec55bb23e95a90559c8f973c10185a498dd4b9e6c9c771adbfc40c052",
+   "video_sha256": "809887032e910649cd55be616abe9213d05b202f235db7eb5bbabc5f49fcf872",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch/trajectories/embodied-olympics-fun-catch-i00-privileged--u8zdmbr.json

@@ -0,0 +1,188 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-i00-privileged--u8zdmbr",
+ "taskKey": "embodied-olympics-fun/catch",
+ "dag_sha256": "61132849363e640170264bbd47c4c4cebc23f7ecfe683cae11a4abb07d10d44b",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "n_actions": 499,
+   "sim_steps": 499,
+   "horizon": 900,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 2,
+   "stages_total": 2,
+   "fell": 0,
+   "replay_s": 5.88,
+   "bad_trajectory": 0,
+   "metric_catch_height_m": 1.09,
+   "metric_catch_speed_m_s": 0.909,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 180,
+   "agent_resets": 0,
+   "agent_replays": 5,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 1.536,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.251,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-catch-i00-privileged--u8zdmbr/replay.mp4",
+  "duration_s": 5.0,
+  "fps": 25.0,
+  "frames": 125,
+  "sha256": "5bf3c17b23796a28f1896fc4f960c73d89e533ac52d4bd072d1d89f455214abb"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "fly_ball_caught": {
+    "method": "visual",
+    "intervals": [
+     [
+      4.99,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频0.00 s机器人直立、双手未持球；视频0.40–3.68 s机器人向前侧方行走，伸出的手保持在身前。初态尚未满足接稳飞球条件。"
+     },
+     {
+      "t": 4.0,
+      "log_step": null,
+      "text_zh": "观察：逐帧检查视频3.76–4.08 s及每0.08秒检查视频4.12–4.92 s，白球从机器人前上方下降，在视频约3.96–4.00 s到达伸出的右手，此后停留在右手处随机器人移动；视频4.96 s末帧球仍在右手处，机器人没有摔倒。成功参考录像视频约3.9–4.92 s也呈现球留在伸出右手的外观，仅作视角比较。"
+     },
+     {
+      "t": 4.96,
+      "log_step": 137,
+      "text_zh": "观察：log step 135–136生成并保存交付轨迹；log step 137对该文件的正式重放报告success=true、stages=[catch,hold]、executed_rows=499，原生裁判事件为仿真3.998 s右手catch、仿真4.99 s完成hold并成功；log step 139的两个新进程重放均成功且末态哈希一致。评分器reward另报告success=1、replay_success=1、deterministic=1、stages_reached=2、fell=0。"
+     },
+     {
+      "t": 4.99,
+      "log_step": 137,
+      "text_zh": "推断：交付文件的499个10毫秒控制步共仿真4.99 s，与视频5.00 s时长吻合；log step 137的仿真3.998 s接住事件已在视频3.96–4.08 s球停留右手处核对。因此把原生hold完成映射至视频约4.99 s作为节点起点，末帧视频4.96 s提供可见锚点。远景不能分辨相对接触方向及0.02秒中断，单手包围接触满100步由正式重放裁判和最终评分佐证，未将最初触球当作完成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "观察：视频约4.00–4.96 s白球留在伸出的右手并随手移动。推断：结合log step 137正式重放的右手catch与hold成功记录，实际为右手四指与拇指或掌部包围抓握球；远景无法逐指辨认接触。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频0.40–4.96 s机器人以双足交替触地行走至接球处，接球后继续步行，腿脚持续支撑身体，未见摔倒。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "catch_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": null,
+   "value": 1.09
+  },
+  {
+   "name": "catch_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": null,
+   "value": 0.909
+  },
+  {
+   "name": "landed_from_start_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": null,
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人行走到飞球落点附近，在视频约4.00 s用右手接到下降的球，球随右手移动至视频4.96 s末帧，机器人保持站立。正式交付重放在仿真4.99 s完成原生保持窗口，映射到视频约4.99 s，接稳飞球节点最终为1/1，评分器确认成功。没有状态日志，远景无法独立分辨接触夹角和毫秒中断；这些细节由���式重放的catch、hold记录及确定性成功评分佐证。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "成功运行，接稳飞球节点终值1/1由评分器（success=1、stages_reached=2/2、fell=0）及正式重放catch/hold记录支持；时钟换算（499步×10 ms=4.99 s，对应视频5.00 s）有说明，接触标签合理，无failure。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:07:24Z",
+  "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--u8zdmbr/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-privileged--u8zdmbr/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "8f7d47ff94b0dbf52464388af93b74899fa2e020e72496fb81d5e435e638c561",
+   "video_sha256": "5bf3c17b23796a28f1896fc4f960c73d89e533ac52d4bd072d1d89f455214abb",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch/trajectories/embodied-olympics-fun-catch-i00-privileged--v7zjg7v.json

@@ -0,0 +1,183 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-i00-privileged--v7zjg7v",
+ "taskKey": "embodied-olympics-fun/catch",
+ "dag_sha256": "61132849363e640170264bbd47c4c4cebc23f7ecfe683cae11a4abb07d10d44b",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "n_actions": 494,
+   "sim_steps": 494,
+   "horizon": 900,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 2,
+   "stages_total": 2,
+   "fell": 0,
+   "replay_s": 4.61,
+   "bad_trajectory": 0,
+   "metric_catch_height_m": 1.338,
+   "metric_catch_speed_m_s": 2.45,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 166,
+   "agent_resets": 1448,
+   "agent_replays": 38,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.759,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.837,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-catch-i00-privileged--v7zjg7v/replay.mp4",
+  "duration_s": 4.96,
+  "fps": 25.0,
+  "frames": 124,
+  "sha256": "603f6d31386a5890c30645cfafb901607028d4d7f79933956406f269e4a34f32"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "fly_ball_caught": {
+    "method": "visual",
+    "intervals": [
+     [
+      4.92,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 3.92,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–3.92 s，机器人双足交替迈步移向落点，抬起并伸出右手；视频 3.92 s 白球仍在右手上方，尚未完成接稳。"
+     },
+     {
+      "t": 3.96,
+      "log_step": null,
+      "text_zh": "观察：逐帧检查视频 3.20–4.92 s；视频 3.92 s 白球向手部下降，视频 3.96 s 白球已到达伸出的右手，视频 4.00–4.92 s 留在该手处随手部小幅运动，未见再次飞离或落地，机器人始终直立。"
+     },
+     {
+      "t": 4.92,
+      "log_step": null,
+      "text_zh": "推断：inputs/run.json 的评分器字段 success=1、replay_success=1、deterministic=1、stages_reached=2、stages_total=2、fell=0，佐证交付轨迹通过原生接住和保持验收。结合视频 3.96–4.92 s 的持球画面，将包含100个控制步保持窗口的完整节点完成锚点记为视频 4.92 s（片尾约视频 4.9 s，非首次触球时刻）；远景不能独立测量包围接触夹角或0.02秒中断，完成时刻为近似定位。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "推断：视频 3.96–4.92 s 白球停留在机器人伸出的右手处，结合评分器接住及保持两阶段成功，判为右手手指与拇指或掌部包住棒球；远景无法分辨各指接触点。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–4.92 s，机器人双脚交替与场地地面接触，腿部承重，迈步移至落点并在右手持球期间保持站立。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "catch_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": null,
+   "value": 1.338
+  },
+  {
+   "name": "catch_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": null,
+   "value": 2.45
+  },
+  {
+   "name": "landed_from_start_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": null,
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人迈步移至落点，约在视频 3.96 s 用右手接到下降的白球，并保持至视频 4.92 s 片尾；结合评分器两阶段成功记录，接稳飞球节点在片尾约视频 4.9 s 完成，最终为1/1。实际接触为右手抓握棒球、双足及腿部支撑，未见脱球或摔倒。无状态日志，远景无法独立核验90°接触夹角及0.02秒中断，原生保持窗口由评分器成功记录佐证，节点完成时刻为近似定位。",
+ "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 fly_ball_caught 终值 true 与评分器 success=1、stages 2/2、fell=0 一致；完成锚点视频 4.92 s（约 3.96 s 接球 + 100 控制步）合理，接触标签与证据标注为推断，无需修改。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T23:59:38Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-privileged--v7zjg7v/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-privileged--v7zjg7v/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "d77b9ef7e00792e132db5f6a9ade966bd14133bfc64394e5880bb6c23725c654",
+   "video_sha256": "603f6d31386a5890c30645cfafb901607028d4d7f79933956406f269e4a34f32",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch/trajectories/embodied-olympics-fun-catch-i00-privileged--vvdbc3e.json

@@ -0,0 +1,208 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-i00-privileged--vvdbc3e",
+ "taskKey": "embodied-olympics-fun/catch",
+ "dag_sha256": "61132849363e640170264bbd47c4c4cebc23f7ecfe683cae11a4abb07d10d44b",
+ "outcome": {
+  "success": 0,
+  "reward": 0.5,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "n_actions": 398,
+   "sim_steps": 398,
+   "horizon": 900,
+   "episode_over": 1,
+   "progress": 0.5,
+   "stages_reached": 1,
+   "stages_total": 2,
+   "fell": 0,
+   "replay_s": 4.96,
+   "bad_trajectory": 0,
+   "metric_catch_height_m": 1.293,
+   "metric_catch_speed_m_s": 3.601,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 127,
+   "agent_resets": 478,
+   "agent_replays": 2,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 1.098,
+   "peak_processes": 13,
+   "peak_disk_gb": 0.241,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-catch-i00-privileged--vvdbc3e/replay.mp4",
+  "duration_s": 4.0,
+  "fps": 25.0,
+  "frames": 100,
+  "sha256": "29bf8dc2b0cd7a515b201c71646339690ce17c607d97e9da5b6da7c8aa78bf43"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "fly_ball_caught": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.8,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–2.8 s，机器人双足向前方落点附近移动并保持站立；视频 3.0–3.96 s，机器人伸出右臂，下降的白球在视频 3.84–3.92 s 接近右手，视频末帧没有可辨认的一秒稳定持球过程。"
+     },
+     {
+      "t": 3.96,
+      "log_step": 129,
+      "text_zh": "观察：log step 128 重新生成并写出最终交付的398行 trajectory.npz；log step 129 对该文件的重放报告“the ball came out of the hand after the catch”，stages=[catch]，catch_height_m=1.293、catch_speed_m_s=3.601，与评分器的 stages_reached=1/2、progress=0.5 一致。"
+     },
+     {
+      "t": 3.96,
+      "log_step": 102,
+      "text_zh": "推断：log step 102 对同一交付文件的原生重放记录为 steps=398、time_s=3.98、stages=[catch]，末态哈希前缀与最终 log step 129 一致；结合视频末段球到右手附近的画面，判定曾形成短暂接住，但没有完成规定的同手保持窗口，故整个“接稳飞球”节点从未成立。视频时间锚来自所查帧，未用 agent 工作时间换算。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "推断：右手四指与拇指或掌部曾短暂从两侧包住棒球，随后失去包围；依据视频 3.92–3.96 s 球到伸出的右手附近，以及最终交付重放 log step 129 的 catch 阶段与接住后脱手结果。远景无法分辨各指接触。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.0–3.96 s，机器人双脚交替接触场地地面、承重并向落点移动，在伸右手接球时仍站立；评分器 fell=0。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "catch_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": null,
+   "value": 1.293
+  },
+  {
+   "name": "catch_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": null,
+   "value": 3.601
+  },
+  {
+   "name": "landed_from_start_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": null,
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "fly_ball_caught",
+   "t": 3.96,
+   "log_step": 129,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 3.84–3.96 s 白球降至机器人伸出的右手附近，机器人仍站立，随后视频结束。log step 128 保存的最终交付在 log step 129 重放时仅到 catch 阶段，并明确报告“the ball came out of the hand after the catch”；接住后的保持失败未获修复，断点为接稳飞球。"
+  },
+  "unmet": {
+   "fly_ball_caught": false
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没保持住",
+    "t": 3.96,
+    "log_step": 129,
+    "evidence_zh": "观察：视频 3.84–3.96 s 机器人以伸出的右手迎接下降白球；最终交付重放 log step 129 报告接住后球脱手、stages=[catch]，与 log step 102 的原生重放失败原因相同，评分器也仅记录阶段进度0.5。短暂接住后保持失败，而非从未接住；远景不能单独显��脱离接触的毫秒过程。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "抓取和接触设计",
+    "alt_feasible": "未知",
+    "source_zh": "日志",
+    "evidence_zh": "推断：log step 122 对相同接球参数的会话诊断中，agent 脚本调用原生 enclosed 判据并统计右手两组接触：仿真 3.95 s 曾为四指组2处、拇指或掌组3处且 enclosed=True，随后两组均降至0，后续仅有零星四指接触、拇指或掌组保持0，球沿+x方向移出。该诊断的接住高度、速度和失败原因均与最终交付重放 log step 129 一致，支持所用手指与掌部包围布局未能阻止球在减速后逸出。诊断属于会话根因证据，不用其仿真时间直接替代视频时间；冲击推开手指是 agent 在 log step 123 的解释，具体驱动力尚待核实，未验证本次手部模型下的可行替代握法。"
+   }
+  ],
+  "gap_zh": "没有逐帧状态日志；640×360、25 fps 远景及放大图无法核验90°包围接触、0.02秒中断或具体驱动力，微观脱手依据最终交付的原生重放结果，接触设计归因辅以相同参数的会话诊断而非逐帧状态证据。成功参照仅说明可见持球形态，其手部模型版本可能不同，不能证明本次参数下的替代握法可行。"
+ },
+ "summary_zh": "机器人走到落点附近并伸出右手，最终交付的原生重放确认短暂接住，但随即脱手，未完成一秒同手保持，因此唯一节点“接稳飞球”始终未达成，最终为0/1。断点在视频约 3.96 s 的接球保持过程中，现象为没保持住；相同参数的会话接触诊断支持手指与掌部包围布局未能阻止球逸出的控制问题。远景无法分辨毫秒接触和驱动力，微观判断依赖已核对交付身份的重放及会话诊断。",
+ "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(final=false)、断点 fly_ball_caught、现象“没保持住”均与 log step 102/122/129 的原生重放(stages=[catch]、接住后脱手)及评分器 progress=0.5 一致；视频仅 4 s，远景无法分辨手部接触，标注已将微观判断标为推断。未发现需改内容的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:02:53Z",
+  "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-catch-i00-privileged--vvdbc3e/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-privileged--vvdbc3e/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "3bb0d98dc865fd5b16b4d86b6c91b84086905b5b230a08bfd3e02a67b0bd5e0d",
+   "video_sha256": "29bf8dc2b0cd7a515b201c71646339690ce17c607d97e9da5b6da7c8aa78bf43",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch/trajectories/embodied-olympics-fun-catch-i00-privileged--zbdkvfq.json

@@ -0,0 +1,186 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-i00-privileged--zbdkvfq",
+ "taskKey": "embodied-olympics-fun/catch",
+ "dag_sha256": "61132849363e640170264bbd47c4c4cebc23f7ecfe683cae11a4abb07d10d44b",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "n_actions": 497,
+   "sim_steps": 497,
+   "horizon": 900,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 2,
+   "stages_total": 2,
+   "fell": 0,
+   "replay_s": 5.91,
+   "bad_trajectory": 0,
+   "metric_catch_height_m": 0.933,
+   "metric_catch_speed_m_s": 6.633,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 48,
+   "agent_resets": 349,
+   "agent_replays": 2,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 1,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.0,
+   "peak_processes": 0.0,
+   "peak_disk_gb": 0.0,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-catch-i00-privileged--zbdkvfq/replay.mp4",
+  "duration_s": 5.0,
+  "fps": 25.0,
+  "frames": 125,
+  "sha256": "eb9a528e4c92b68f8172e139f5d219861883770b6f69ea9f4e95adebbd3fc2c9"
+ },
+ "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 0.0 s 机器人直立、双手尚未持球；视频 0.0–3.6 s 机器人在场地上向前移动并调整右臂，没有已接稳飞球的状态。"
+     },
+     {
+      "t": 4.0,
+      "log_step": null,
+      "text_zh": "观察：逐帧检查视频 3.76–4.16 s，小白球从机器人右上方下降，视频 3.96 s 仍在右手上方附近，视频 4.00–4.04 s 到达右手区域；随后小白点留在该手附近，机器人双脚交替支撑、身体直立。"
+     },
+     {
+      "t": 4.96,
+      "log_step": 36,
+      "text_zh": "观察：视频 4.04–4.96 s 的逐帧画面中，球留在右手附近，未见球另行落地或机器人倒下；log step 36 显示 agent 保存497行轨迹，官方重放返回 success=true、stages=[catch,hold]、executed_rows=497、time_s=4.97；run.json 的评分字段 success=1、replay_success=1、stages_reached=2、fell=0。"
+     },
+     {
+      "t": 4.96,
+      "log_step": 37,
+      "text_zh": "推断：视频所见接球并保持至片尾与交付文件的官方重放成功一致，log step 37 又报告第二次重放 success 且末态哈希相同，故接稳飞球节点在终态为真。判据包括完整持球窗口，不能把视频约4.00 s 的初次接球当作完成；将完成锚定于视频末帧4.96 s，精度约一帧。远景不能验证90°接触夹角或0.02秒中断，这些微观条件依据原生裁判成功结论，而非直接目测。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "机器人右手接住下降的棒球并持续持球；视频约4.00–4.96 s 可见球到达并留在右手附近，结合原生 catch、hold 阶段通过，推断右手手指与拇指或掌部形成包围抓握。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "机器人双脚与地面交替接触，腿部支撑身体向接球位置移动，并在右手持球时保持直立；已查看视频全程及4.00–4.96 s 逐帧画面。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "catch_height_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": null,
+   "value": 0.933
+  },
+  {
+   "name": "catch_speed_m_s",
+   "unit": "m/s",
+   "threshold": null,
+   "direction": null,
+   "value": 6.633
+  },
+  {
+   "name": "landed_from_start_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": null,
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人移动至落点附近，用右手接住下降的飞球并保持到片尾，完成接稳飞球节点（终态1/1）。视频约4.00 s 可见球到达右手，完整持球窗口的完成锚定在视频4.96 s；交付文件两次官方重放均成功且末态哈希一致，故没有失败断点。没有状态日志，640×360远景不能直接验证接触夹角和毫秒级中断，微观抓握与保持条件以原生裁判成功记录佐证，完成时刻约有一帧误差。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "成功运行，评分字段 success=1、stages 2/2、fell=0，接稳飞球节点终态为真与证据一致；远景帧无矛盾，完成时刻锚定片尾可接受，无失败断点。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:08:40Z",
+  "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-catch-i00-privileged--zbdkvfq/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-privileged--zbdkvfq/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "44e7b1f6af95b0a97deccc38c3135c70d23d82cd9677a40bfaa85caf08245c01",
+   "video_sha256": "eb9a528e4c92b68f8172e139f5d219861883770b6f69ea9f4e95adebbd3fc2c9",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch/trajectories/embodied-olympics-fun-catch-i00-standard--bptya9r.json

@@ -0,0 +1,208 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-i00-standard--bptya9r",
+ "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": 398,
+   "sim_steps": 802,
+   "calls": 39,
+   "rows": 802,
+   "resets": 1,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 6726.5,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.241,
+   "peak_processes": 13,
+   "peak_disk_gb": 2.791,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-catch-i00-standard--bptya9r/replay.mp4",
+  "duration_s": 4.0,
+  "fps": 25.0,
+  "frames": 100,
+  "sha256": "dcf681b8e1404d1675e1872b27628063b988173b274a96869de74af3da174058"
+ },
+ "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 机器人空手站在场地，尚未接球；视频 2.00–3.36 s 机器人向前侧移动，视频 3.40–3.88 s 抬起右手迎球。"
+     },
+     {
+      "t": 3.96,
+      "log_step": 141,
+      "text_zh": "观察：视频 3.92 s 白球下降到机器人头部前上方，视频 3.96 s 球已靠近头肩与上臂区域，右手仍伸在更靠画面右侧的位置，未见球留在手中；log step 141 的机器人现场服务返回 episode=2、control_steps=398、sim_time_s=3.98、stages_reached=0/2，outcome 为“the ball touched the body before it was caught”。该服务字段报告的是受评第二回合的原生裁判结果，明确未进入接住阶段。"
+     },
+     {
+      "t": null,
+      "log_step": 139,
+      "text_zh": "观察：log step 139 启动真实第二回合，log step 140 记录 reset 后执行迎球动作，log step 141 记录该回合终止；此前 log step 137 的 success 来自 agent 自建模拟服务测试，不是本次受评接球成功。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–3.96 s 机器人以双足在场地地面支撑，走向接球位置后抬臂迎球，片尾仍由脚支撑，未见倒地。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：log step 141 原生裁判报告球在接住前碰到禁触身体部位；视频 3.92–3.96 s 球下降至头肩附近。候选名称“被动受撞”：这是来球撞上机器人身体，非抓握、托举、推扫或承重支撑；远景不能分辨具体身体部位。右手虽伸出迎球，但没有可确认的包围抓握，不将典型抓握方案记为已发生。"
+     }
+    ]
+   }
+  },
+  "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": 3.96,
+   "log_step": 141,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 3.40–3.96 s 机器人正在抬右手拦接下降球，视频 3.92–3.96 s 球进入头肩附近而未停在伸出的手中；log step 141 现场服务报告第二回合在仿真 3.98 s 因接住前触身终止，阶段为 0/2。失败发生于接稳飞球步骤，尚未开始同手保持窗口。"
+  },
+  "unmet": {
+   "fly_ball_caught": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "动态时机",
+    "t": 3.96,
+    "log_step": 141,
+    "evidence_zh": "观察：视频 3.60–3.96 s 机器人右手抬在身前迎球，但视频 3.92 s 的下降白球位于头部前上方，随后进入身体附近，没有被右手截住；log step 141 原生裁判报告接住前触身、接住阶段未达成。运动中的球与迎接手没有在合适位置和时刻相遇。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "evidence_zh": "推断：log step 79 的接球控制代码在 soft 阶段后继续更新来球预测，但仅在 not soft 时重新求解手臂目标，触发闭手后将同一低层关节目标整段提交。log step 141 记录实际回合在仿真 3.67 s 进入 SOFT，仿真 3.79 s 与 3.80 s 预测接球点仍从 [2.1386,-0.83,1.6] 变到 [2.223,-0.7749,1.6]，仿真 3.80 s 触发 GRIP；视频 3.68–3.96 s 可见手臂位置及身体前倾继续变化，最终未拦住球。末段预测变化未反馈到新的手臂目标，支持接球控制未能补偿末段偏差；缺少实际末端轨迹，无法细分控制子类。",
+    "source_zh": "日志 / 视频"
+   }
+  ],
+  "gap_zh": "无状态日志，远景视频仅能看到球进入头肩附近，不能确认触球的具体身体连杆，也不能排除终止前曾短暂擦碰手或前臂；原生裁判明确报告接住前触身及阶段 0/2，因此节点未达成可以判定。缺少实际手掌三维轨迹和逐物理步接触，无法量化手球偏差或进一步区分估计误差、姿态跟踪及闭手参数的贡献；不据模拟服务测试的成功结果推定真实接触。"
+ },
+ "summary_zh": "受评的是重置后的第二次现场接球尝试：机器人移动到迎球位置并抬起右手，但接稳飞球节点始终未达成，最终为 0/1。视频 3.92–3.96 s 球进入头肩附近，log step 141 原生裁判报告接住前触身而终止；末段控制在更新来球预测后未继续更新手臂目标，是接球未对上的有据控制因素。无状态日志，具体触身部位及手指的瞬时接触不可辨，不能进一步细分控制原因。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "节点未达成、终止时刻（log step 141 仿真 3.98 s，视频总长 4.0 s）、断点和控制类根因均与原生裁判结果及日志一致；远景下球极小，球位置的帧描述不能独立证实，但有裁判字段支撑。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:15:16Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard--bptya9r/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard--bptya9r/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "b88cba10cdcffeb1b33fd13021a6b2d56a08bfde5913b23a704d582313fc3bb6",
+   "video_sha256": "dcf681b8e1404d1675e1872b27628063b988173b274a96869de74af3da174058",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch/trajectories/embodied-olympics-fun-catch-i00-standard--ffwthfj.json

@@ -0,0 +1,211 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-i00-standard--ffwthfj",
+ "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": 429,
+   "sim_steps": 2069,
+   "calls": 55,
+   "rows": 1665,
+   "resets": 4,
+   "reset_budget": 50,
+   "sim_startups": 1,
+   "elapsed_s": 3052.9,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 1.324,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.44,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-catch-i00-standard--ffwthfj/replay.mp4",
+  "duration_s": 4.32,
+  "fps": 25.0,
+  "frames": 108,
+  "sha256": "2487abd2573c31146c40ab65adf155265ef39379bee96ad7a03e384d34beadf0"
+ },
+ "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 0.00–3.84 s，机器人从初始站位移动并向来球方向伸出右手，双脚支撑身体；球尚未被任何一只手接住。"
+     },
+     {
+      "t": 3.96,
+      "log_step": null,
+      "text_zh": "观察：视频 3.92 s 白球仍在右手上方，视频 3.96 s 到达右手；视频 4.08–4.28 s 白球已离开右手，向画面左下方下降至地面附近，右手没有带球移动。逐帧检查视频 3.20–4.28 s，未见持续包住球的阶段。"
+     },
+     {
+      "t": 4.28,
+      "log_step": 91,
+      "text_zh": "观察：log step 91 的机器人服务返回最后一次重置后的 episode=5、done=True、outcome=\"the ball touched the ground\"、stages_reached=0/2；reward 的 calls=55、rows=1665、sim_steps=2069 与此次返回一致。该回合没有达到原生接住阶段，更未完成单手保持窗口。"
+     },
+     {
+      "t": 3.96,
+      "log_step": 92,
+      "text_zh": "推断：log step 3 给出控制步长 0.01 s，log step 91 的第 5 回合共 429 步，即仿真 4.29 s，与视频末帧 4.28 s 相符；log step 92 读取同一回合保存的服务本体数据，仿真 3.96 s 的到球时刻已在视频 3.96 s 核对，故可将这些回合内采样用于辅助判断。该输出中右手 holding 始终为 0。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "其他",
+      "note_zh": "推断：视频约 3.96 s，来球在机器人右手处改变飞行方向后离开，右手只发生短暂碰触，未形成包握。候选名称：碰触；这次被动碰球既无夹持，也非主动推扫、按擦或从下承托，故用其他；远景不能区分具体指节或掌面。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–4.28 s，机器人用双脚与腿支撑身体，移动至接球位置并在伸出右手时保持站立，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": 3.96,
+   "log_step": 91,
+   "where": "操作中",
+   "evidence_zh": "观察：视频约 3.96 s，机器人伸出的右手迎到飞球，但白球随后离开手部，视频 4.08–4.28 s 持续向地面下降，未被接回。log step 91 的第 5 回合原生裁判以球触地判负，接住及保持两阶段均未达到；断点为正在执行的接稳飞球步骤。"
+  },
+  "unmet": {
+   "fly_ball_caught": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "动态时机",
+    "t": 3.96,
+    "log_step": 92,
+    "evidence_zh": "推断：视频 3.96 s 白球到达右手，视频 4.08 s 已离开手部；log step 92 对该第 5 回合服务本体数据的读取显示，仿真 3.93 s 右手闭合量仍为 0.22，仿真 3.96 s 为 0.27，仿真 3.99 s 才到 0.34，holding 始终为 0。结合视频中球通过手部后偏离、下降的过程，右手合拢未赶上飞球，未建立有效接住状态。上述回合采样时钟由 0.01 s 控制步长及视频到球事件核对。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "source_zh": "日志",
+    "evidence_zh": "推断：log step 91 的实际受评脚本先将右手预闭合到 0.22，直到仿真 3.945 s 后才命令闭合量 1.0，并自仿真 3.965 s 起按预设时刻转腕收臂，未等待接住反馈。log step 92 的同回合采样显示到球时闭合量仅为 0.27、holding=0；视频 3.96–4.28 s 可见球未被包住并最终落下。命令与手部响应未形成及时的接球控制；现有证据不足以进一步分清接触几何、闭合时序和关节跟踪各自的贡献，保留控制大类。"
+   }
+  ],
+  "gap_zh": "没有状态日志或逐物理步的原生接触对、接触方向与保持计数；640×360、25 fps 远景可确认球离手及未接稳，但不能确定触球的具体指节、90°包围角或毫秒级中断。log step 105、111 的几何估计及后续 MuJoCo 参数测试均是 agent 对已结束回合的离线分析，不当作本次接球成功的证据，也不足以唯一确定控制子类。log step 194 的模型过滤提示发生在 log step 91 的受评回合已触地判负之后，按既有物理结果标注。"
+ },
+ "summary_zh": "受评的最后一次现场尝试为第 5 回合：机器人移动并伸出右手，视频约 3.96 s 迎到飞球，但未及时包住，球随后偏离手部并落地，接稳飞球节点全程未达成，最终为 0/1。原生裁判在 log step 91 报告球触地、阶段 0/2；同回合手部采样支持闭合与来球未同步的控制失败。缺少逐步接触记录，不能进一步确定控制子类；后续离线测试及片尾过滤提示不改变已结束回合的判定。",
+ "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、接触与现象/根因均与视频和 log step 91/92 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": "2026-10-31T00:05:48Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard--ffwthfj/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard--ffwthfj/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "e878d9ad59b6046a3975e7f02ca0515cf94f7194fc9dec7f27a20f21839dd6fe",
+   "video_sha256": "2487abd2573c31146c40ab65adf155265ef39379bee96ad7a03e384d34beadf0",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch/trajectories/embodied-olympics-fun-catch-i00-standard--hljesry.json

@@ -0,0 +1,214 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-i00-standard--hljesry",
+ "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": 406,
+   "sim_steps": 9314,
+   "calls": 80,
+   "rows": 9314,
+   "resets": 31,
+   "reset_budget": 50,
+   "sim_startups": 1,
+   "elapsed_s": 5859.2,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 0.528,
+   "peak_processes": 13,
+   "peak_disk_gb": 0.647,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-catch-i00-standard--hljesry/replay.mp4",
+  "duration_s": 4.08,
+  "fps": 25.0,
+  "frames": 102,
+  "sha256": "80ba158d9a6d68af3f8e10cf165f4b85155ef35110d42ecda252c86750dbb1b1"
+ },
+ "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，手中没有球；视频 0.40–3.44 s 机器人逐步前移，尚未接球。"
+     },
+     {
+      "t": 3.96,
+      "log_step": null,
+      "text_zh": "观察：逐帧检查视频 3.00–4.04 s；视频 3.80–3.92 s 机器人俯身并将双臂伸向来球，视频 3.92 s 白球仍在手前上方，视频 3.96 s 白球到达前伸手部附近，视频 4.00–4.04 s 白球已向手下方落去，没有形成随手保持的持球状态。"
+     },
+     {
+      "t": 4.04,
+      "log_step": 262,
+      "text_zh": "观察：log step 262 的机器人服务返回最后一次现场尝试 episode=32、control_steps=406、仿真 4.06 s、done=true、outcome=\"the ball touched the ground\"、stages_reached=0/2；该服务字段是原生裁判的接球/保持阶段结果。inputs/run.json 的评分字段 n_actions=406、calls=80、rows=9314 与这次尝试对应，log step 263、267 确认最后尝试为 R18，之后没有再次执行机器人动作。"
+     },
+     {
+      "t": 4.04,
+      "log_step": 262,
+      "text_zh": "推断：结合视频 3.96–4.04 s 球从手边继续下落及 log step 262 的未接住即落地裁判结果，接稳飞球从未达成，最终为 false；不是仅根据总分为零判断。成功参考在其视频 3.96–4.92 s 可见白球持续留在伸出的手边，本次未出现这一保持过程。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–3.80 s 机器人双足交替支撑并向来球方向前移；视频 3.84–4.04 s 机器人以足部支撑俯身伸手，片尾仍未见躯干触地。"
+     },
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "推断：候选“拦接时短暂碰触”。视频 3.88–3.96 s 机器人双臂同时前伸拦接，白球到达前伸手部附近后于视频 4.00–4.04 s 下落；结合 log step 259–262 的双手拦接及右手后撤程序和 log step 267 的触手反弹自述，推断至少一手短暂触球，未形成包围抓握。现有抓握、托/舀、推/扫、压/擦均不能准确表达这种短暂碰触；远景不能确认具体接触手指或另一手是否也触球。"
+     }
+    ]
+   }
+  },
+  "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": 3.96,
+   "log_step": 262,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 3.88–3.96 s 机器人正在俯身伸臂接球，白球到达手边后在视频 4.00–4.04 s 继续下落，没有随手保持；log step 262 对应最后现场尝试的原生裁判报告接球阶段 0/2，因球触地结束。断点为正在进行的“接稳飞球”。"
+  },
+  "unmet": {
+   "fly_ball_caught": false
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没建立接触",
+    "t": 3.96,
+    "log_step": 262,
+    "evidence_zh": "推断：视频 3.92–4.04 s 显示白球到达机器人前伸手部附近后继续下落，未出现被手包住并随手移动的过程；log step 262 原生裁判报告 stages_reached=0/2 且球触地。所需的相对两侧包围抓握没有建立，短暂触手不等于已经握住，因此不标“没保持住”。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "抓取和接触设计",
+    "alt_feasible": "未知",
+    "t": 3.96,
+    "log_step": 267,
+    "source_zh": "agent 自述",
+    "evidence_zh": "推断：log step 267 中 agent 自述，最终俯身姿态下手掌朝向与来球方向约偏离90°，位置与朝向目标冲突，球从手边反弹而未进入闭合手指形成的包围区域；这是 agent 的分析，具体角度待核实。日志中 log step 257–258 可见 agent 调整位置/朝向优化权重，log step 259–262 又加入左手拦接和右手沿球速方向后撤，最终仍未接住；视频 3.96–4.04 s 的触手附近后下落与此解释一致。证据支持接触几何不适配这一根因，不据此认定所有合法接球姿态不可行。"
+   }
+  ],
+  "gap_zh": "没有状态日志；视频为640×360远景、25 fps，不能分辨手指与掌部的接触组、90°方向阈值或0.02秒中断，也不能确认左手是否实际触球。至少一手短暂触球及接触设计归因含推断；最终约90°掌向偏差来自 log step 267 的 agent 自述，没有本次原生接触或掌向数据独立验证。三个原生数值指标在评分字段中均未报告，故保留 null。"
+ },
+ "summary_zh": "机器人前移后俯身伸出双臂拦接，视频 3.96 s 小球到达手边却继续下落；接稳飞球未达成，最终完成 0/1。log step 262 确认最后现场尝试接球阶段为0/2、因球触地结束；结合 agent 在 log step 267 对掌向不适配的自述，根因推断为抓取和接触设计。缺少状态日志，远景无法独立验证微观接触及自述的具体掌向偏差。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "标签与视频、log 一致：视频 3.96–4.04 s 球到手边后下落，未保持；log step 262 服务返回 stages_reached=0/2、球触地，节点 false、断点 fly_ball_caught 合理。现象“没建立接触”（所需包围抓握从未形成）与根因“抓取和接触设计”均已标注为推断/agent 自述并在 gap_zh 说明，可接受。远景过小无法独立验证掌向。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:00:12Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard--hljesry/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard--hljesry/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "b595ec6e7e8c350b1eda9b06e5189a0dc32d033230fbf4e2388fbfee0716b826",
+   "video_sha256": "80ba158d9a6d68af3f8e10cf165f4b85155ef35110d42ecda252c86750dbb1b1",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch/trajectories/embodied-olympics-fun-catch-i00-standard--mmxtfns.json

@@ -0,0 +1,208 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-i00-standard--mmxtfns",
+ "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": 434,
+   "sim_steps": 21230,
+   "calls": 446,
+   "rows": 20784,
+   "resets": 50,
+   "reset_budget": 50,
+   "sim_startups": 1,
+   "elapsed_s": 6707.7,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 0.595,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.876,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-catch-i00-standard--mmxtfns/replay.mp4",
+  "duration_s": 4.36,
+  "fps": 25.0,
+  "frames": 109,
+  "sha256": "038b3c485608d3d7fe062a80aa8c93e46a535f0e02f227abe3473377f1963858"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "fly_ball_caught": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 3.52,
+      "log_step": null,
+      "text_zh": "观察：视频 0.0–3.52 s，机器人从站立转为向前侧走动，并抬起右手迎接来球；尚无球停在手内。"
+     },
+     {
+      "t": 3.96,
+      "log_step": null,
+      "text_zh": "观察：视频 3.80–4.00 s 的逐帧画面中，白球下降经过机器人右手附近，随后出现在画面中机器人左侧的空中；视频 4.00–4.32 s 球继续远离右手并向地面下降，右手没有带球移动，机器人仍站立。没有出现球被单手控制并保持的状态。"
+     },
+     {
+      "t": 4.32,
+      "log_step": 352,
+      "text_zh": "观察：视频 4.32 s 球已接近地面且与右手明显分离；agent 最后自述“最终尝试在仿真 4.34 s 因球触地失败，裁判为 0/2 阶段”（log step 352）。评分器字段 success=0、live_success=0、graded_by_live=1。该自述与录像中的漏接及球下落相符，但不用于给视频时钟换算。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.0–4.32 s，机器人双脚交替接触场地地面，腿部支撑身体走向拦截位置，抬右手接球时未见摔倒。"
+     },
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：视频 3.80–4.00 s 机器人右手迎向白球，但远景无法确定球是否瞬时擦到手指、掌或前臂；未见合手包住球，故手球接触不能标为已建立抓握。"
+     }
+    ]
+   }
+  },
+  "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": 3.96,
+   "log_step": 346,
+   "where": "操作中",
+   "evidence_zh": "观察：最后一次重置并运行接球脚本的是 log step 346；视频 3.80–4.00 s 机器人正抬右手截球，球却经过手边继续下落，视频 4.00–4.32 s 始终未回到手内。agent 在 log step 348、352 将此次末局报告为球触地且未接住；前面的重置试验不是本段评分录像。"
+  },
+  "unmet": {
+   "fly_ball_caught": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "动态时机",
+    "t": 3.96,
+    "log_step": null,
+    "evidence_zh": "观察：视频 3.80–4.00 s 白球快速下降到机器人伸出的右手附近，随后越过手部并继续下落；视频 4.00–4.32 s 球与手逐渐分离，未出现球随手停留。推断：机器人没有在运动目标经过时完成有效拦截，属于动态接球未跟上；没有证据表明先握住再脱手。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "t": 3.96,
+    "log_step": 348,
+    "evidence_zh": "推断：视频 3.80–4.00 s 机器人已走近并伸右手迎球，但手部动作未与来球形成有效拦截，球在视频 4.00–4.32 s 继续落下。log step 345 记录 agent 为最后一局调整肘角并准备在触球窗口闭手，log step 346 执行脚本，log step 348 输出的脚本使用预设步行、臂姿及闭手序列；目标是接住并保持，没有错误任务目标的证据。结合录像将根因限于控制，现有证据不能再细分具体接触几何、轨迹或闭手时机。",
+    "source_zh": "视频"
+   }
+  ],
+  "gap_zh": "没有状态日志；场地远景分辨率不足以确定球是否瞬时擦碰右手或前臂，也无法测量手指包围接触方向、毫秒级中断及原生保持计数。末局的详细裁判输出未保留在可见工具返回中，仿真 4.34 s 触地及 0/2 阶段仅为 agent 自述（log step 348、352），但漏接及球下落在视频 3.80–4.32 s 可见。控制根因的子类不能可靠区分。"
+ },
+ "summary_zh": "评分录像中机器人走向来球并伸出右手，但白球在视频约 3.9 s 经过手边后继续下落，接稳飞球节点始终未完成，最终为 0/1。断点位于接球操作中的动态拦截，根因仅能归为控制；视频 4.32 s 机器人仍站立且球已远离手部。无状态日志，远景不能确定瞬时手球擦碰或进一步区分控制误差，不能将早先重置试验的触球算入本局。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "节点 fly_ball_caught 未达成、断点在该步骤的操作中、单一动态时机/控制归因与 0/1 结果及视频所见一致；视频 3.4–4.32 s 小球（微小白点）自上而下落到机器人侧下方，未见随手停留。远景分辨率有限，证据已如实标注为推断。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:14:28Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard--mmxtfns/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard--mmxtfns/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "6c03dd825da4070d4319674504e36dabb0b9404b077c3edc913b4d41feec3557",
+   "video_sha256": "038b3c485608d3d7fe062a80aa8c93e46a535f0e02f227abe3473377f1963858",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch/trajectories/embodied-olympics-fun-catch-i00-standard--nb96egp.json

@@ -0,0 +1,213 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-i00-standard--nb96egp",
+ "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": 429,
+   "sim_steps": 20503,
+   "calls": 2152,
+   "rows": 17675,
+   "resets": 50,
+   "reset_budget": 50,
+   "sim_startups": 1,
+   "elapsed_s": 2415.0,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 0.443,
+   "peak_processes": 10,
+   "peak_disk_gb": 1.099,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-catch-i00-standard--nb96egp/replay.mp4",
+  "duration_s": 4.32,
+  "fps": 25.0,
+  "frames": 108,
+  "sha256": "2beaab0a11aa12dd8db7b1ae06e13c8e7e3a2a8dab8722e1ffc70be513dcd7a2"
+ },
+ "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 机器人双脚站在场地上，手中无球；随后迈步靠近落点并伸出右手。"
+     },
+     {
+      "t": 3.96,
+      "log_step": null,
+      "text_zh": "观察：逐帧检查视频 3.20–4.28 s，白球在视频 3.92 s 位于右手上方，视频 3.96 s 到达右手附近，视频 4.00–4.28 s 向画面右下方离开手并落到地面；机器人始终站立，未见球被手包住后持续随手移动。"
+     },
+     {
+      "t": 4.28,
+      "log_step": 156,
+      "text_zh": "观察：log step 156 的最终配置运行返回原生裁判结果“the ball touched the ground”、stages 0、control steps 429；log step 161、164 的服务状态均为 episode 51、429 控制步、failed，后续重置已被拒绝。该结尾与视频 4.28 s 球到达地面相符，早先试验的留球或超时结果不计入本次进度。"
+     },
+     {
+      "t": 4.28,
+      "log_step": null,
+      "text_zh": "推断：本次未建立并维持合规单手包围抓握，接稳飞球节点从未完成；成功参考录像的视频 4.00–4.80 s 可见白球持续位于伸出的手边，本次在相同视角下出现的是球离手下落。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "其他",
+      "note_zh": "推断：视频 3.92–4.04 s 白球下降到伸出的右手处后改变运动方向并离开，符合右手与棒球短暂碰触。候选名称“接球碰挡”：未见手指对向夹住球，也没有持续承托、主动推扫或击打，现有方式不能准确描述这种短暂碰触。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–4.28 s 机器人以双足接触场地地面，迈步到接球位置后保持站立，未见其他身体部位触地。"
+     }
+    ]
+   }
+  },
+  "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": 3.96,
+   "log_step": 156,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 3.96 s 飞球到达机器人伸出的右手附近后未停留，视频 4.00–4.28 s 球离手向右下方运动至地面；log step 156 返回球触地、stages 0，log step 161、164 确认最后现场 episode 51 已失败。失败发生在接球操作中，没有完成接稳窗口。"
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没建立接触",
+    "t": 3.96,
+    "log_step": 156,
+    "evidence_zh": "推断：视频 3.92–4.28 s 球仅在右手附近短暂交会便离开，没有出现包住并随手移动的过程；log step 156 原生裁判报告 stages 0、球触地。未建立的是接球所需的对向包围抓握，不能将一次碰触算成已抓住后脱落。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "t": 3.96,
+    "log_step": 156,
+    "source_zh": "日志",
+    "evidence_zh": "推断：log step 126 的驱动脚本将右臂保持在固定接球姿态，以预设时刻切换右手闭合；log step 156 将预闭合设为 0.3、T_CLOSE 设为脚本仿真时钟 4.10 s 后运行，原生裁判仍返回 stages 0。视频 3.96–4.28 s 可见球已到手边但未被控制住，说明该手部接球控制未形成有效抓握；现有证据不足以进一步确定几何、时序或力控子类。"
+   }
+  ],
+  "gap_zh": "没有状态日志或逐控制步的手球接触、包围判定与保持计数；640×360 的远景只能支持短暂碰触及随后的离手下落，不能精确分辨手指接触方向或毫秒中断。log step 165 关于手指力量无法承受冲击的解释是 agent 自述，缺少接触力记录验证，未据此归为环境不可行。成功参考录像仅作可见状态对照，其手部参数版本差异不用于证明当前配置的力学可行性。",
+  "unmet": {
+   "fly_ball_caught": false
+  }
+ },
+ "summary_zh": "机器人迈步并伸出右手迎接飞球，但视频约 3.96 s 球到达手边后即离开，至视频 4.28 s 落地；接稳飞球节点未完成，最终为 0/1。原生裁判的 stages 0 和球触地结果与画面一致，失败表现为未建立包围抓握，根因归于控制；缺少微观接触与力记录，不能确定更细的控制原因。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "单节点未完成、最终0/1与视频及原生裁判结果（球触地、stages 0、429控制步）一致；断点在接球操作中，现象“没建立接触”和根因“控制”（未细分子类）有日志和画面支撑，证据标注观察/推断恰当，未发现需要修改的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:00:15Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard--nb96egp/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard--nb96egp/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "186c12ae80ce0b43a61d90866bbc43a0e446653c5c4b8dbe904de656c27aabc7",
+   "video_sha256": "2beaab0a11aa12dd8db7b1ae06e13c8e7e3a2a8dab8722e1ffc70be513dcd7a2",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch/trajectories/embodied-olympics-fun-catch-i00-standard--rm9ki8a.json

@@ -0,0 +1,206 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-i00-standard--rm9ki8a",
+ "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": 445,
+   "sim_steps": 21099,
+   "calls": 4647,
+   "rows": 20695,
+   "resets": 50,
+   "reset_budget": 50,
+   "sim_startups": 1,
+   "elapsed_s": 3832.5,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 1.559,
+   "peak_processes": 14,
+   "peak_disk_gb": 2.425,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-catch-i00-standard--rm9ki8a/replay.mp4",
+  "duration_s": 4.48,
+  "fps": 25.0,
+  "frames": 112,
+  "sha256": "4bdcdc1a0a08039b9925f102d91a6eca14cdc20ceff56e1ecbad82373a1c997e"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "fly_ball_caught": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.8,
+      "log_step": null,
+      "text_zh": "观察：视频0.0–2.8 s机器人双足向落球方向移动，随后抬起并伸出右臂；视频3.4–4.44 s机器人始终站立并用右手迎球，确已尝试接球。"
+     },
+     {
+      "t": 3.96,
+      "log_step": 151,
+      "text_zh": "观察：视频3.88、3.92 s白球向右手下降，视频3.96 s球到达右手附近；视频4.04–4.20 s白球向手的内侧、躯干方向离开，视频4.36–4.44 s已落到躯干侧下方，未留在右手随手移动。log step 151的最后一次现场调用返回裁判outcome=\"the ball was not caught in time after it first touched you\"、stages_reached=0，并由服务status确认episode=51、done=True、control_steps=445、resets_left=0；这与本段视频相符。"
+     },
+     {
+      "t": 4.44,
+      "log_step": 151,
+      "text_zh": "推断：依据视频3.96–4.44 s球离开右手的过程及最后一次现场裁判的未接住判定，本次未建立规定的包围接触，也未完成同手保持窗口；接稳飞球始终为假。先前会话试验的stage 1记录不属于此评分局。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "托/舀",
+      "note_zh": "观察：视频3.96 s机器人伸出的右手承接下降的棒球，随后球离开手掌区域；这里记录右手短暂承接触球，未观察到手指与拇指合拢夹住球，最后一次裁判也未认定包围抓握。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频0.0–2.8 s机器人双足在地面支撑并移动身体至接球位置；视频3.4–4.44 s双腿继续支撑身体迎球，未摔倒。"
+     }
+    ]
+   }
+  },
+  "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": 3.96,
+   "log_step": 151,
+   "where": "操作中",
+   "evidence_zh": "观察：视频3.96 s下降的白球到达机器人伸出的右手，视频4.04–4.20 s却离手向躯干方向运动，未被该手包住；机器人继续站立至视频4.44 s。log step 151记录评分所用最后一次现场尝试的裁判返回“触球后未及时接住”、stage 0，并确认episode 51已结束。"
+  },
+  "unmet": {
+   "fly_ball_caught": false
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没建立接触",
+    "t": 3.96,
+    "log_step": 151,
+    "evidence_zh": "观察：视频3.96 s球虽到达右手，视频4.04–4.44 s已从手边移向躯干侧并下降；最后一次现场裁判在log step 151返回触球后仍未接住、stages_reached=0。所缺的是四指与拇指或掌部从两侧包住球的抓握接触，不能把短暂碰手算作已握住再脱手。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "evidence_zh": "推断：log step 150将调整后的右臂目标写入最终���数，log step 151实际执行这些参数后，视频3.96 s球已到达右手而视频4.04–4.44 s球仍离手，裁判也未认定接住；失败落在迎球及闭手的接触控制。现有证据不能进一步区分手型几何、闭合时机或接触响应，故停在控制类。agent在log step 153关于手指过短、反弹系数及厘米级精度的解释未获独立验证，不据此归因环境或感知。",
+    "source_zh": "视频 / 日志"
+   }
+  ],
+  "gap_zh": "无状态日志或最后一次评分局的逐物理步接触对、接触方向与手指力记录；640×360、25 fps远景不能分辨90°包围接触或0.02秒中断，也不能确定右手具体哪一指先碰球。原生裁判明确报告最后一次未接住，可判节点为假，但不足以细分控制根因。DAG列出的三个原生数值指标均未在本次评分字段中报告，保留null。"
+ },
+ "summary_zh": "评分局是log step 151运行的最后一次现场尝试（episode 51）：机器人移动后伸出右手迎球，但视频3.96 s触球附近之后，白球向手内侧离开并下降，接稳飞球最终0/1。最后一次裁判报告触球后未及时握住、stage 0，断点为接稳飞球操作中，现象是未建立包围抓握接触，根因保守归为控制。无逐步接触日志，远景无法细分手型与闭手时机；成功参照中球留在右手的画面仅用于比较，不计入本次进度。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "评分局（log step 151，episode 51）未接住与视频一致：视频3.96 s球到右手，4.04–4.44 s球离手落向躯干侧；裁判返回\"not caught in time after it first touched you\"、stage 0，节点为假、断点与现象无明显错误；根因停在控制类有保守理由。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:08:37Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard--rm9ki8a/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard--rm9ki8a/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "5927539e5148953b9c643c36b72ed307b9957b380e6e7463318b033577aeb7ad",
+   "video_sha256": "4bdcdc1a0a08039b9925f102d91a6eca14cdc20ceff56e1ecbad82373a1c997e",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch/trajectories/embodied-olympics-fun-catch-i00-standard--ydqkrth.json

@@ -0,0 +1,189 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-i00-standard--ydqkrth",
+ "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": 492,
+   "sim_steps": 2195,
+   "calls": 50,
+   "rows": 1791,
+   "resets": 4,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 3913.9,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 1,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.0,
+   "peak_processes": 0.0,
+   "peak_disk_gb": 0.0,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-catch-i00-standard--ydqkrth/replay.mp4",
+  "duration_s": 4.96,
+  "fps": 25.0,
+  "frames": 124,
+  "sha256": "7b142f3abf0b2651c76103d17f2cc62536060f3577e9c641aa70173edc96ebd1"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "fly_ball_caught": {
+    "method": "visual",
+    "intervals": [
+     [
+      4.92,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00 s 机器人站在场地上，手中尚无球；视频 0.00–3.60 s 机器人向前移动并调整右臂，尚未完成接球保持。"
+     },
+     {
+      "t": 3.92,
+      "log_step": 151,
+      "text_zh": "观察：视频 3.80–3.96 s 机器人抬起右手，白色小球从右手上方下降至手边；视频 4.00–4.92 s 小球留在右手附近并随手的小幅运动保持相邻，机器人持续以双脚支撑，没有可见脱手或摔倒。log step 151 的第5次现场尝试记录 control_steps=393，所读右手 holding=1；这时只是接住，尚未完成规定保持窗口。"
+     },
+     {
+      "t": 4.92,
+      "log_step": 159,
+      "text_zh": "观察：log step 159 中，机器人服务（agent通过客户端获取的模拟器状态及原生裁判通道）报告 episode=5、control_steps=492、sim_time_s=4.92、done=true、reason=success、outcome=success、stages_reached=2、stages_total=2；run.json 的 reward.live_success=1、graded_by_live=1。视频 4.92 s 机器人仍站立并保持右手接球姿态。"
+     },
+     {
+      "t": 4.92,
+      "log_step": 159,
+      "text_zh": "推断：log step 132 重置后开始第5次现场尝试，log step 159 结束该次尝试，log step 160 保存其492个控制步；评分记录 n_actions=492，与这段录像一致。该尝试每控制步为0.01秒，log step 151 的393步对应视频约3.93 s，已用视频 3.88–3.96 s 球到达右手的可见事件核对；据此 log step 159 的492步对应视频4.92 s。结合持续留球画面及原生裁判两阶段成功，接稳飞球节点在视频4.92 s 完成并保持至片尾；不把首次接触时刻当作满窗口完成时刻。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "推断：视频约3.9 s 球下降到机器人抬起的右手，此后至视频4.92 s 留在右手边并随手运动，log step 159 原生裁判确认接住及保持均成功，支持右手手指与拇指或掌部包住棒球；左手没有参与接球。远景无法直接分辨各指接触方向。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频0.00–3.60 s 机器人用双脚在地面支撑并迈步向前接近接球位置，视频3.80–4.92 s 继续以双脚支撑身体，在右手接球及保持时没有可见摔倒。"
+     }
+    ]
+   }
+  },
+  "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": null,
+ "summary_zh": "评分对应第5次现场尝试：机器人向前移动并抬起右手，在视频约3.9 s 接住下降的球，保持站立和同手控球后于视频4.92 s 完成“接稳飞球”，最终1/1节点成立，无失败断点。原生机器人服务在 log step 159 报告���住与保持两阶段成功，评分字段亦确认现场成功；此前各次重置试验不计入本段轨迹。没有状态日志，远景无法独立核实90°接触夹角或0.02秒中断，微观接触与保持窗口依原生裁判结果佐证；三项原生数值指标未在评分字段中报告。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "评分为第5次现场尝试成功；log step 151/159 的 control_steps(393→3.93 s，492→4.92 s)与视频时钟映射一致，节点完成于视频 4.92 s、contact_actual 合理，无失败断点，未发现需改动之处。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:13:59Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard--ydqkrth/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard--ydqkrth/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "b054b7b82fb4f9c65c39eca5ea906b0da06989a561c44d4d73ad5eccc27db396",
+   "video_sha256": "7b142f3abf0b2651c76103d17f2cc62536060f3577e9c641aa70173edc96ebd1",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch/trajectories/embodied-olympics-fun-catch-i00-standard--yqsyzs4.json

@@ -0,0 +1,208 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-i00-standard--yqsyzs4",
+ "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": 447,
+   "sim_steps": 2467,
+   "calls": 18,
+   "rows": 1484,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 883.6,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.225,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.385,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-catch-i00-standard--yqsyzs4/replay.mp4",
+  "duration_s": 4.48,
+  "fps": 25.0,
+  "frames": 112,
+  "sha256": "e61477514b717e11980991b4fb156b90677499e4bfebeb5459ed6e45d266fa58"
+ },
+ "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 0.00–3.60 s，机器人从初始站位迈步移动并摆动右臂准备接球，尚未持球；视频 3.88–3.96 s 白球向右上肢下降。"
+     },
+     {
+      "t": 4.0,
+      "log_step": 136,
+      "text_zh": "观察：视频 3.96–4.00 s，白球到达机器人伸出的右手／前臂附近后改变运动方向，视频 4.04–4.44 s 白球与手分离，未见球留在手中随手移动，机器人仍站立。log step 136 的最后一次现场尝试为 episode 6，服务返回 done=True、reason=failed、outcome=\"the ball was not caught in time after it first touched you\"、stages_reached=0/2、sim_time_s=4.47；该仿真结束时刻与视频片尾相符。"
+     },
+     {
+      "t": 4.44,
+      "log_step": 136,
+      "text_zh": "推断：视频片尾球已离开接球上肢，且 log step 136 的原生裁判明确未进入接住阶段，因此本次评分尝试从未满足同手包围握持并保持规定窗口的节点判据；这不是仅由总分为零反推未完成。参考成功录像的参考视频 4.00–4.92 s 可见球持续留在伸出的手旁，本次录像没有该持续持球表现。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–4.44 s，机器人以双腿和脚在场地地面支撑身体，迈步移动后站立伸臂接球，未见摔倒。"
+     },
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：视频 3.96–4.00 s 白球在机器人右手／前臂附近改变运动方向后弹离，log step 136 的原生裁判确认曾首次触及机器人但未形成接住；远景无法确定实际接触的是手指、掌部还是前臂，也无法细分瞬时接触方式，不能记为抓握。"
+     }
+    ]
+   }
+  },
+  "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": 136,
+   "where": "操作中",
+   "evidence_zh": "观察：视频约 3.96–4.00 s，机器人正在伸右臂接下降的白球，球在右手／前臂附近弹离，随后至视频 4.44 s 都未停在手中；log step 136 的最后一次现场执行由原生裁判以首次触球后未及时接住结束，接住与保持阶段均未通过。"
+  },
+  "unmet": {
+   "fly_ball_caught": false
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没建立接触",
+    "t": 4.0,
+    "log_step": 136,
+    "evidence_zh": "观察：视频 3.96–4.04 s，球到达接球上肢后立即与手分离，至视频 4.44 s 没有随手保持；log step 136 的原生裁判报告首次触球后未及时接住、stages_reached=0/2。这里未建立的是接球所需的包围握持接触，已有瞬时碰触，未见先握稳再脱手。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "evidence_zh": "推断：log step 133 的实际执行脚本预先生成整段右臂关节目标与右手闭合指令，一次 submit 提交；右臂到预设姿态后维持，闭手按预设仿真时序渐增，触球后没有依据球运动重新调整。log step 136 执行此脚本，视频 3.96–4.44 s 显示球到达接球上肢却弹离，原生裁判未认可包围握持，支持接球控制未完成所需接触；现有证据不足以区分接触几何、闭手时序或执行跟踪等具体控制子类。",
+    "source_zh": "日志、视频"
+   }
+  ],
+  "gap_zh": "没有状态日志；640×360、25 fps 的场地远景不能分清首次触球的具体手部或前臂部位，也不能核实接触夹角、手指闭合程度、接触力及毫秒级接触变化。因此保留球接触的作用端与方式为未知，控制根因不细分；节点未达成由球弹离的画面与最后一次现场原生裁判结果共同确认。"
+ },
+ "summary_zh": "评分采用 log step 136 的第六次、最后一次现场尝试，机器人迈步就位并伸右臂迎球，但视频约 3.96–4.00 s 球在右手／前臂附近弹离，接稳飞球从未达成，最终为 0/1。原生裁判以首次触球后未及时接住结束，失败表现为未建立包围握持接触，证据支持归到控制。无状态日志且远景无法分辨具体触球部位、接触几何和闭手细节，不能进一步细分根因。",
+ "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 fly_ball_caught、时间（视频 4.0 s 与 sim_time 4.47 s 吻合）均有 log step 136 原生裁判（stages_reached 0/2、首次触球后未及时接住）支持；接触标注为未知合理，根因停在“控制”不细分有据。远景下球极小，我无法从帧中独立分辨弹离细节，但裁判结果足以决定。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:00:02Z",
+  "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-i00-standard--yqsyzs4/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard--yqsyzs4/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "476c008503fdb155b78ae20612c2b97473b91d9483a48fe778ba713e36b4fb86",
+   "video_sha256": "e61477514b717e11980991b4fb156b90677499e4bfebeb5459ed6e45d266fa58",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch/trajectories/embodied-olympics-fun-catch-i00-standard-realtime--2errrrd.json

@@ -0,0 +1,241 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-i00-standard-realtime--2errrrd",
+ "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": 446,
+   "sim_steps": 4533,
+   "calls": 287,
+   "rows": 5830,
+   "resets": 10,
+   "reset_budget": 10,
+   "sim_startups": 1,
+   "elapsed_s": 1479.8,
+   "video_rendered": 1,
+   "realtime": 1,
+   "slowdown": 1.0,
+   "clock_s": 45.393,
+   "max_lag_s": 0.0607,
+   "stalls": 0,
+   "stalled_s": 0.0,
+   "stalled_share": 0.0,
+   "held_steps": 404,
+   "late_steps": 0,
+   "requests_refused": 1,
+   "request_busy_s": 0.716,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 1,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.0,
+   "peak_processes": 0.0,
+   "peak_disk_gb": 0.0,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-catch-i00-standard-realtime--2errrrd/replay.mp4",
+  "duration_s": 4.48,
+  "fps": 25.0,
+  "frames": 112,
+  "sha256": "7fd42a933e1998852df91807c7a8f84b02ebf78cc80a00f2d7188edb2b596de1"
+ },
+ "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,
+      "log_step": null,
+      "text_zh": "观察：视频 0–3.6 s，机器人从初位向前侧移动，随后伸出右臂准备接球；双足支撑身体，未见摔倒。"
+     },
+     {
+      "t": 3.96,
+      "log_step": 69,
+      "text_zh": "观察：视频 3.6–4.44 s，机器人伸出右臂并保持站立；640×360 远景无法分辨球与手的具体接触及弹离过程。log step 69 的现场服务结果为“the ball was not caught in time after it first touched you”，stages_reached=0，control_steps=446；log step 75 确认最终为 episode 11、done=True、reason=failed。"
+     },
+     {
+      "t": 4.44,
+      "log_step": 77,
+      "text_zh": "推断：本次评分的是 episode 11；log step 69 运行 cfg11，log step 74 的后续配置因重置次数耗尽未能运行，log step 75 服务状态仍为 episode 11。结合 log step 69 首次触球后未及时接住的服务判词及 stages_reached=0，“接稳飞球”从未成立；此前会话中其他 episode 短暂接住的结果不计入。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：log step 69 的现场服务判词“the ball was not caught in time after it first touched you”确认首次触球，但未说明接触部位和方式；视频 3.84–4.44 s 的远景及放大帧不足以分辨掌部、手指或其他部位触球，也不能确认托/舀。所需抓握未由裁判注册，故不标抓握。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0–4.44 s，机器人双脚交替踏地移动到接球位置，随后由双腿支撑站立并伸右手接球，全程未见倒地。"
+     }
+    ]
+   }
+  },
+  "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": 69,
+   "where": "操作中",
+   "evidence_zh": "推断：log step 69 的本次现场服务判词为首次触球后未在规定的 0.5 秒窗口内接住，stages_reached=0；视频 3.6–4.44 s 机器人正伸右臂接球，故断点为接球操作中。视频约 4.0 s 仅为导航锚点，远景无法核实具体接触与弹离；本次 446 个 0.01 秒控制步约为视频 4.46 s，与视频 4.48 s 片长相符，控制步时钟还由脚本抬臂窗口 2.95–3.65 秒与视频约 3.0–3.6 s 的伸臂动作核对。"
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没建立接触",
+    "t": 4.0,
+    "log_step": 69,
+    "evidence_zh": "推断：log step 69 的原生裁判报告首次触球后未在规定窗口内接住，stages_reached=0，说明未建立本步骤所需的四指与拇指或掌部的相对包围接触；已有短暂碰触不等于抓握。视频 3.84–4.44 s 无法分辨具体触球部位及弹离过程，本现象依据裁判的具体失败判词，而非仅依据零分；本次并非其他 episode 的“接住后脱手”。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "t": 4.0,
+    "log_step": 69,
+    "source_zh": "日志 / 视频",
+    "evidence_zh": "推断：log step 69 的脚本将右手预闭合设为 0.3，至本次控制步 400 才改为 0.45，至控制步 412 才发出全闭合命令；右臂沿用预置接球姿态。log step 32 的生成脚本以 k/100 编排动作并以 dt=0.01、at=0 提交，log step 69 本次也从控制步 0 提交；446 个控制步约为视频 4.46 s，与视频 4.48 s 片长相符，脚本抬臂窗口 2.95–3.65 秒亦与视频约 3.0–3.6 s 的伸臂动作相符。因此全闭合命令约对应视频 4.12 s；结合本次服务的首次触球后未及时接住及 stages_reached=0，推断接球姿态与闭手控制未能把碰触转为所需抓握。画面不能证明全闭合时球已离手，也不足以进一步区分接触几何、撞击响应与跟踪误差。"
+   }
+  ],
+  "gap_zh": "没有状态日志；640×360、25 fps 的远景及放大帧不能可靠分辨具体触球部位、接触方式、球与手的接触及弹离过程，更不能核实接触指节、90°接触夹角、毫秒级中断及接触力。视频中的机器人姿态与本次原生裁判的具体失败判词支持未接稳，控制根因仅为日志与画面结合的推断，不能确认闭手晚于球离手，故置信度由 8 降为 7；agent 自建局部仿真及此前 episode 的诊断不作为本次物理过程的直接证据。评分器 reward 未报告三个原生指标数值，均记 null。",
+  "unmet": {
+   "fly_ball_caught": false
+  }
+ },
+ "summary_zh": "评分尝试为最后一次现场 episode 11：机器人移动后伸出右臂接球，原生裁判报告首次触球后未及时接住且 stages_reached=0，未完成“接稳飞球”，最终 0/1。结合本次脚本的预置姿态与闭手命令，推断接球控制配合失败，后续准备的 cfg12 未实际运行。没有状态日志，远景无法确认触球部位、承接方式及弹离过程，具体控制机制仍不确定，置信度降为 7。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 2,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "fix",
+    "conf": 6,
+    "summary_zh": "节点 fly_ball_caught=false、断点 fly_ball_caught/操作中、控制步→视频映射均与服务读数（446 步≈4.46 s ≈ 视频 4.48 s，episode 11）一致，无大错。但证据把 640×360 远景中看不清的球轨迹和“右手掌部承接”写成了观察，应改为推断。",
+    "n_issues": 2,
+    "n_major": 0
+   },
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 2,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "第二轮：节点 false、断点 fly_ball_caught/操作中、控制步时钟映射、contact_actual（触球部位未知）均与 log step 69/75 的服务读数（episode 11、446 控制步、stages_reached=0）一致；上轮观察/推断问题已改正，未发现新的实质错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:18:30Z",
+  "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-catch-i00-standard-realtime--2errrrd/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard-realtime--2errrrd/04_review_r1"
+   },
+   {
+    "stage": "reconcile",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard-realtime--2errrrd/05_reconcile_r1"
+   },
+   {
+    "stage": "traj_review",
+    "round": 2,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard-realtime--2errrrd/04_review_r2"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "1fe79440538d93d78b390da2342dca04e6f5cecfbbeec93dd887529326647a5a",
+   "video_sha256": "7fd42a933e1998852df91807c7a8f84b02ebf78cc80a00f2d7188edb2b596de1",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch/trajectories/embodied-olympics-fun-catch-i00-standard-realtime--ecvukv8.json

@@ -0,0 +1,238 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-i00-standard-realtime--ecvukv8",
+ "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": 397,
+   "sim_steps": 801,
+   "calls": 182,
+   "rows": 4530,
+   "resets": 1,
+   "reset_budget": 10,
+   "sim_startups": 1,
+   "elapsed_s": 931.5,
+   "video_rendered": 1,
+   "realtime": 1,
+   "slowdown": 1.0,
+   "clock_s": 8.01,
+   "max_lag_s": 0.0214,
+   "stalls": 0,
+   "stalled_s": 0.0,
+   "stalled_share": 0.0,
+   "held_steps": 404,
+   "late_steps": 190,
+   "requests_refused": 0,
+   "request_busy_s": 0.4,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 1,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.0,
+   "peak_processes": 0.0,
+   "peak_disk_gb": 0.0,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-catch-i00-standard-realtime--ecvukv8/replay.mp4",
+  "duration_s": 4.0,
+  "fps": 25.0,
+  "frames": 100,
+  "sha256": "abdf0b3bfcfdb802323ab38ef0bab4cd074aca8b73438b72611b16d7cf8549a8"
+ },
+ "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": 37,
+      "text_zh": "观察：视频 0.00–3.96 s 中机器人双足行走，片尾仍站立；视频 3.92 s 放大帧可见其头部右上方的白色小点，视频 3.96 s 已不能将球与机器人轮廓可靠分开，远景不能判读手指包握。log step 37 中机器人现场服务在 episode 2 返回 done=True、outcome=\"the ball touched the body before it was caught\"、control_steps=397。"
+     },
+     {
+      "t": 3.96,
+      "log_step": 37,
+      "text_zh": "推断：log step 37 唯一一次重置后的 episode 2 是最后评分局，其行走过程与视频相符，服务控制步数397与 inputs/run.json 的 reward.n_actions=397 一致；log step 4 给出控制周期0.01秒，397步对应仿真 3.97 s，与视频 3.96 s 末帧及该处录像终止相核对，仅作为片尾失败的粗略时间锚。原生服务明确判为接住前碰身体，故本节点从未达到，不能由远景不可辨改记未知。log step 37 脚本始终发送 r_arm_ready 和 grip_r=0.35，支持固定右臂目标的推断，不等于视频观察到实际手姿；成功参考不作本次接触证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–3.96 s 中，机器人双脚接触场地，双腿承重并行走至球路附近，片尾仍站立。"
+     },
+     {
+      "who": "未知",
+      "how": "其他",
+      "note_zh": "观察：log step 37 的机器人现场服务确认球在被接住前碰到机器人身体；视频 3.92 s 放大帧可见机器人头部右上方白色小点，视频 3.96 s 不能可靠分辨球与身体的接触。推断：白色小点为来球，但撞身事实依据服务 outcome，具体部位未知。候选名称：被动碰撞；该意外球体撞身不属于抓握、承托、推扫或支撑，不能据此记手指抓握。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "catch_height_m",
+   "value": null,
+   "unit": "m",
+   "threshold": null,
+   "direction": null
+  },
+  {
+   "name": "catch_speed_m_s",
+   "value": null,
+   "unit": "m/s",
+   "threshold": null,
+   "direction": null
+  },
+  {
+   "name": "landed_from_start_m",
+   "value": null,
+   "unit": "m",
+   "threshold": null,
+   "direction": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "fly_ball_caught",
+   "t": 3.96,
+   "log_step": 37,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 3.80–3.96 s 中机器人仍站立，远景不能辨认手指接持；log step 37 的机器人现场服务报告最后一局 episode 2 在控制步397因球接住前碰到身体而终止。推断：该控制步对应仿真 3.97 s，与视频 3.96 s 片尾核对，故以片尾作为接稳飞球失败的粗略锚点；该脚本固定右臂目标，不能把实际手姿或来球与手的精确相对位置写成视频观察。之后没有新的现场重置或成功修复。"
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "动态时机",
+    "t": 3.96,
+    "log_step": 37,
+    "evidence_zh": "观察：log step 37 的机器人现场服务判为球接住前碰身体；同一步脚本从初始到末段均发送 r_arm_ready、grip_r=0.35，末段只降低行走速度。视频 3.92 s 放大帧可见机器人头部右上方白色小点，但无法辨认手指与球接触。推断：固定手臂目标及闭合指令未能完成对运动中来球的拦截，暂标动态时机；这不是观察到闭手过早或过晚，无法排除手位或手型偏差，置信度降低。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "认知",
+    "item": "推理/规划",
+    "t": 3.96,
+    "log_step": 37,
+    "source_zh": "日志",
+    "alt_feasible": "未知",
+    "evidence_zh": "观察：log step 37 的调用标为“Running Episode 2 walk calibration and ball observation”；检查完整脚本可见，初始、循环和末段均只提交 walk 行，右臂目标固定为 r_arm_ready，grip_r始终为0.35，循环反馈只调整行走方向，末段只减速后读取图像，未安排接球时的手位调整、闭合或缓冲阶段。现场服务随后返回接住前碰身体。推断：本次评分局执行的是校准与观察方案，接球阶段的动作设计尚未纳入该局，依 docs/03_trajectory.md Step 3.3 的判别顺序归认知·推理/规划；无证据表明一个完整可行的接球方案仅因跟踪执行失败，亦不单凭恒定闭合值断言手型必然不可行。",
+    "note_zh": "可行替代方案在本次现场服务中未验证，alt_feasible记未知；成功参考及后续本地MuJoCo试验不证明该现场方案的接触可行性。"
+   }
+  ],
+  "gap_zh": "没有状态日志；640×360远景只能辨认机器人运动及片尾个别帧的白色小点，不能分辨实际手位、球撞到的具体身体部位、手指接触方向或毫秒级包围接触。撞身事实由 log step 37 的现场服务 outcome 补足；动态时机为低置信推断，无法区分手位/手型偏差与闭合时机，不能指定过早或过晚。log step 43 的球相对骨盆坐标由 agent 在假设骨盆位置[1.40,-0.52,1.02]下离线计算，并非现场球位读数，不能用来证明球已到手位。评分字段未报告三项原生指标，均填null。log step 47 之后的本地MuJoCo试验没有进入现场评分局；log step 78 的模型过滤发生在 log step 37 的物理失败已确定之后，不改变该局失败标签。",
+  "unmet": {
+   "fly_ball_caught": false
+  }
+ },
+ "summary_zh": "最后评分局中机器人行走但未接稳飞球，最终为0/1节点；log step 37 的现场服务在控制步397以接住前碰身体终止该局，失败锚点为视频约 3.96 s。完整现场脚本固定右臂目标和闭合指令，只安排校准行走与观察，没有接球阶段，根因改为认知·推理/规划；动态时机仅为拦截错失的推断，无法分清手位、手型与闭合时机，故整体置信度由8降至6。远景不能识别具体撞身部位；log step 78 的模型过滤发生在物理失败之后，后续本地试验不计入节点进度。",
+ "conf": 6,
+ "review": {
+  "status": "agreed",
+  "rounds": 2,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "fix",
+    "conf": 6,
+    "summary_zh": "节点未达成、断点 fly_ball_caught 与时钟（397步×0.01s≈3.97s≈视频3.96s）均成立。问题在于：视频证据里的手部与球的“观察”视频中并不可见（应改为日志推断）；根因只定“控制”而未按判别顺序考虑规划/接触设计——最后评分局的脚本本身就是“walk calibration and ball observation”，从未包含闭手或随球摆手的接球段。",
+    "n_issues": 3,
+    "n_major": 2
+   },
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 2,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "节点未达成、断点 fly_ball_caught（视频约 3.96 s，397步×0.01 s 核对）、根因（评分局脚本仅校准行走、无接球段，推理/规划）均有 log step 37 与服务 outcome 支持；观察与推断已区分，第一轮问题已修复。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:19:57Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard-realtime--ecvukv8/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard-realtime--ecvukv8/04_review_r1"
+   },
+   {
+    "stage": "reconcile",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard-realtime--ecvukv8/05_reconcile_r1"
+   },
+   {
+    "stage": "traj_review",
+    "round": 2,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard-realtime--ecvukv8/04_review_r2"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "a8a845e3f64d78a288474a811900427348610f329d47f94b35cb224514cf703c",
+   "video_sha256": "abdf0b3bfcfdb802323ab38ef0bab4cd074aca8b73438b72611b16d7cf8549a8",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch/trajectories/embodied-olympics-fun-catch-i00-standard-realtime--floeinz.json

@@ -0,0 +1,219 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-i00-standard-realtime--floeinz",
+ "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": 396,
+   "sim_steps": 2020,
+   "calls": 17,
+   "rows": 443,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 2932.0,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "realtime": 1,
+   "slowdown": 1.0,
+   "clock_s": 20.193,
+   "max_lag_s": 0.0353,
+   "stalls": 0,
+   "stalled_s": 0.0,
+   "stalled_share": 0.0,
+   "held_steps": 408,
+   "late_steps": 0,
+   "requests_refused": 0,
+   "request_busy_s": 0.364,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 1,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.0,
+   "peak_processes": 0.0,
+   "peak_disk_gb": 0.0,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-catch-i00-standard-realtime--floeinz/replay.mp4",
+  "duration_s": 4.0,
+  "fps": 25.0,
+  "frames": 100,
+  "sha256": "7d9dceeaea3cf9b3a089d3762c16f828760458b17acb0367c87b0e302a806dc7"
+ },
+ "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 机器人站在场地内，双手未持球；视频 0.00–3.20 s 机器人随后迈步靠近来球方向。"
+     },
+     {
+      "t": 3.96,
+      "log_step": 194,
+      "text_zh": "观察：视频 3.60–3.96 s 的逐帧图像显示机器人弯曲右臂准备接球，视频 3.92 s 白球仍在手臂上方，视频 3.96 s 白球到达右臂内侧附近，未出现球被包住后随手移动的过程。最后一次现场尝试的服务裁判在 log step 194 返回 control_steps=396、done=True、stages_reached=0，outcome 为“the ball touched the body before it was caught”。"
+     },
+     {
+      "t": 3.96,
+      "log_step": 194,
+      "text_zh": "推断：log step 5 的服务规则说明只评分最后一次 episode；log step 194 重置并完成 episode 6，且其 396 个控制步与评分字段 n_actions=396 相符，因此此前测试不计入本节点。按任务文本每控制步 0.01 s，396×0.01=视频 3.96 s，与已检查的白球到达右臂附近并终止的片尾相符；接稳及同手保持窗口均未完成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–3.96 s 机器人双足在地面交替支撑，迈步至接球位置，片尾仍站立。"
+     },
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：视频 3.92–3.96 s 白球下降至机器人弯曲的右臂附近；log step 194 的最后一次现场服务裁判明确记录球在接住前碰到禁触身体部位。远景不能区分具体部位，也不能确认右手是否曾先短暂触球或如何触球，故不标抓握。"
+     }
+    ]
+   }
+  },
+  "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": 3.96,
+   "log_step": 194,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 3.60–3.96 s 机器人正抬起并弯曲右臂接下降白球，视频 3.96 s 球到达右臂附近即结束；最后一次现场服务裁判在 log step 194 返回“the ball touched the body before it was caught”、stages_reached=0。失败发生在接球操作中，尚未接稳。"
+  },
+  "unmet": {
+   "fly_ball_caught": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "动态时机",
+    "t": 3.96,
+    "log_step": 194,
+    "evidence_zh": "推断：视频 3.92 s 白球在机器人伸出的右手上方，视频 3.96 s 已下降到弯曲右臂内侧附近；log step 194 的现场裁判记录接住前触身且��段为0。机器人伸手、闭手的接球动作未与下降来球配合形成捕获，球先碰到禁触身体部位；没有已经握住后再丢失的证据。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "source_zh": "日志",
+    "evidence_zh": "推断：log step 188 的脚本预先指定行走、右臂关节目标、闭手和回撤，log step 194 又将闭手命令改为仿真 3.55 s 起、回撤改为仿真 3.85 s 起后执行最后一次尝试；该脚本在接球期间按预定动作运行，没有按来球观测实时修正。结合视频 3.92–3.96 s 机器人伸臂但球到达右臂附近，以及同一 log step 194 的原生触身判罚，判为接球动作控制未形成有效捕获；现有证据不能把失败进一步归结为特定闭手参数、接触几何或跟踪误差。"
+   }
+  ],
+  "gap_zh": "本次没有状态日志；640×360 的场地远景及视频 3.96 s 的终止帧不能分辨具体禁触身体部位、是否先短暂碰手、相对两侧接触或手指闭合程度。log step 191 的手部读数属于此前 episode 5，不能移用于最后一次 episode 6；agent 在 log step 198 自述“弹开”和位置误差，但最后一次裁判只证实接住前触身，不能据此标为已握住后脱手，也不能确认更细的控制或感知根因。"
+ },
+ "summary_zh": "评分对象为 log step 194 的最后一次现场尝试：机器人迈步并抬起右臂接球，但视频 3.96 s 来球到达右臂附近时，原生裁判以接住前触身终止，接稳飞球未达成（0/1）。断点位于接球操作中，现象为动态时机，预定接球动作未形成有效捕获，根因仅标到控制。远景且无状态日志，无法确认具体触球部位及闭手、接触几何或轨迹误差的各自作用；此前测试和 agent 的“弹开”自述不作为本次曾握住球的证据。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "最后一次现场尝试（log step 194，control_steps=396，视频 3.96 s）服务裁判判定接住前触身、stages_reached=0，节点未达成、断点、时钟换算（396×0.01 s）与保守的现象/根因标注均有证据支持；远景帧过小，只能确认机器人站立伸臂，未发现需要修改的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:17:19Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard-realtime--floeinz/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard-realtime--floeinz/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "0b8d4e60deee70517d1cb995c34e078944d6971f70956c13cb22e2bc760134f1",
+   "video_sha256": "7d9dceeaea3cf9b3a089d3762c16f828760458b17acb0367c87b0e302a806dc7",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch/trajectories/embodied-olympics-fun-catch-i00-standard-realtime--m94r3sr.json

@@ -0,0 +1,234 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-i00-standard-realtime--m94r3sr",
+ "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": 404,
+   "sim_steps": 4438,
+   "calls": 169,
+   "rows": 9900,
+   "resets": 10,
+   "reset_budget": 10,
+   "sim_startups": 1,
+   "elapsed_s": 1598.7,
+   "video_rendered": 1,
+   "realtime": 1,
+   "slowdown": 1.0,
+   "clock_s": 119.829,
+   "max_lag_s": 0.684,
+   "stalls": 127,
+   "stalled_s": 73.426,
+   "stalled_share": 0.6128,
+   "held_steps": 0,
+   "late_steps": 0,
+   "requests_refused": 0,
+   "request_busy_s": 25.468,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 1,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.0,
+   "peak_processes": 0.0,
+   "peak_disk_gb": 0.0,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-catch-i00-standard-realtime--m94r3sr/replay.mp4",
+  "duration_s": 4.08,
+  "fps": 25.0,
+  "frames": 102,
+  "sha256": "294294ef582140921c79478e46205c41b5e122d31e4c42c1808602200f276172"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "fly_ball_caught": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.6,
+      "log_step": 134,
+      "text_zh": "观察：视频 0.0–3.8 s，机器人从初始站位迈步移动并举起接球手，未见球被手带动或持住；log step 134 显示脚本重置至 episode 11，提交从 control step 0 开始的900步动作计划。本片对应最后一次现场尝试，较早的十次尝试不计入节点进度。"
+     },
+     {
+      "t": 3.96,
+      "log_step": 136,
+      "text_zh": "观察：视频 3.92–4.04 s，机器人举手并保持站立，未见球留在手内随手移动；640×360远景中的小球仅为极少像素，不能据此逐帧确认手球接触或球从手旁到腿旁的具体路径。log step 136 的服务原生裁判结果为 episode 11、control_steps=404、stages=0、outcome=\"the ball touched the ground\"。"
+     },
+     {
+      "t": 4.04,
+      "log_step": 136,
+      "text_zh": "推断：log step 134 报告控制周期0.01 s，log step 136 的最终 control step 404 对应视频约 4.04 s，与视频 4.04 s 的片尾停下核对一致；视频总长4.08 s，末帧时间为视频 4.04 s，此映射不用于断言可见的精确落地瞬间。结合该次原生裁判的 stages=0、球触地结果及视频未见持球状态，判定接稳飞球全程未达成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.0–4.04 s，机器人双脚交替接触场地地面支撑身体并向接球位置移动，末段保持站立；举起的接球手未见包住棒球，不计抓握接触。"
+     }
+    ]
+   }
+  },
+  "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": 3.96,
+   "log_step": 136,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 3.92–4.04 s，机器人举手站立，未见接住后持球随手移动；log step 136 的 episode 11 服务裁判报告 stages=0、outcome=\"the ball touched the ground\"，并判失败。推断：本次在接稳飞球步骤中未完成拦截，断点仍为正在执行的接稳飞球；球从手旁掠过的精确路径不能由远景确认。"
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "动态时机",
+    "t": 3.96,
+    "log_step": 136,
+    "evidence_zh": "推断：视频 3.92–4.04 s 可见机器人举手站立，未见持球随手移动；log step 136 的 episode 11 原生裁判报告 stages=0、outcome=\"the ball touched the ground\"。结合正在执行的飞球接取动作，判断机器人未能在飞球落地前完成拦截，保留动态时机现象；不能从远景确认具体早晚误差、手球相对位置���“接球手未跟上”的细节。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "evidence_zh": "推断：log step 138 展示该次脚本以 make_scout_plan 预生成步行和手臂姿态，再用 r.submit 提交；log step 134 确认从控制步0提交900步计划。视频 3.92–4.04 s 可见机器人举手站立，而 log step 136 的原生裁判报告 stages=0、球触地，说明该次步行、摆臂与飞球拦截的配合未奏效。现有证据仅支持控制层级，不能确认手与球路径错开的精确关系，也不能判定具体是接触几何、轨迹跟踪还是闭手参数。",
+    "source_zh": "日志 / 视频"
+   }
+  ],
+  "gap_zh": "无状态日志；640×360远景中的小球仅为极少像素，视频 3.92–4.04 s 无法可靠逐帧分辨其与手、腿的相对运动及接触关系，不能把“从手旁掠过、落至腿旁”作为直接观察。远景也不能分辨手指间接触、90°包围角或0.02秒级接触中断，因此不排除未成抓握的瞬时擦碰，也不将举手动作记为抓握；动态时机是结合原生裁判落地结果与机器人接球动作的推断。最终控制器在 log step 138 的输出有截断，缺少完整关节命令与实测误差，根因无法进一步细分；三个原生指标均未在评分器字段中报告。",
+  "unmet": {
+   "fly_ball_caught": false
+  }
+ },
+ "summary_zh": "本次评分采用最后一次现场尝试 episode 11：机器人迈步并举手接球，但结合 log step 136 的 stages=0、球触地结果及视频未见持球状态，接稳飞球节点始终未达到，最终为0/1。动态时机和控制层级根因来自接球动作、控制脚本与原生裁判结果的综合推断。缺少状态日志，远景不能可靠逐帧辨清小球与手的相对路径，控制器日志又有截断，故置信度由8降为7，不能确定微观接触或具体控制子因。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 2,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "fix",
+    "conf": 7,
+    "summary_zh": "节点未达成、final=0、断点 fly_ball_caught、评分 stages=0 与落地判负均与视频和 log step 136 一致；问题在于证据文字把 640×360 远景中约1像素的小球轨迹写成“观察”，细节（球从手旁掠过、落至腿旁）视频分辨不出，现象“动态时机”仅由 log 的落地结果加推断支撑。",
+    "n_issues": 1,
+    "n_major": 0
+   },
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 2,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "节点 fly_ball_caught 未达成、final=0、断点与 log step 136 的 episode 11 原生裁判结果（stages=0、球触地、control_steps=404≈视频 4.04 s）一致；证据已区分观察与推断，第1轮问题已修正，未发现新的内容错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:19:50Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard-realtime--m94r3sr/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard-realtime--m94r3sr/04_review_r1"
+   },
+   {
+    "stage": "reconcile",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard-realtime--m94r3sr/05_reconcile_r1"
+   },
+   {
+    "stage": "traj_review",
+    "round": 2,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard-realtime--m94r3sr/04_review_r2"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "e318a301d9b4055bca605d2f66abbd61b066b4d37155ac773ca2e545542450fd",
+   "video_sha256": "294294ef582140921c79478e46205c41b5e122d31e4c42c1808602200f276172",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch/trajectories/embodied-olympics-fun-catch-i00-standard-realtime--mduez8w.json

@@ -0,0 +1,219 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-i00-standard-realtime--mduez8w",
+ "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": 398,
+   "sim_steps": 2412,
+   "calls": 860,
+   "rows": 4500,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 2477.2,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "realtime": 1,
+   "slowdown": 1.0,
+   "clock_s": 24.122,
+   "max_lag_s": 0.0335,
+   "stalls": 0,
+   "stalled_s": 0.0,
+   "stalled_share": 0.0,
+   "held_steps": 404,
+   "late_steps": 0,
+   "requests_refused": 2,
+   "request_busy_s": 5.93,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 1,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.0,
+   "peak_processes": 0.0,
+   "peak_disk_gb": 0.0,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-catch-i00-standard-realtime--mduez8w/replay.mp4",
+  "duration_s": 4.0,
+  "fps": 25.0,
+  "frames": 100,
+  "sha256": "a40d17b4ac09fc08120383b72daf10bdd72eb234e9f14b41ef3cdf11b1fe5fde"
+ },
+ "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 0.00–3.96 s 中，机器人从初始站位向前及侧方走动，并在视频约 3.4 s 后调整右臂迎球；视频 3.84、3.88、3.92、3.96 s 显示小白球下降至手臂和身体附近，未出现球留在同一只手中随手移动的接稳状态，机器人至末帧仍站立。"
+     },
+     {
+      "t": 3.96,
+      "log_step": 191,
+      "text_zh": "观察：log step 189 记录最后一次 reset 为 episode 6，提交900行计划后该局于 control_steps=398 失败；log step 191 的机器人服务原生裁判字段为 stages_reached=0、stages_total=2、outcome=\"the ball touched the body before it was caught\"，holding_right=0、hand_closure_right=0.162，明确没有进入接住阶段。"
+     },
+     {
+      "t": 3.96,
+      "log_step": 189,
+      "text_zh": "推断：log step 8 的控制周期为0.01秒，最终局的398个控制步对应仿真 3.98 s；其结束与视频 3.96 s 末帧处球到达身体的事件吻合，视频按25 fps取样、总长4.00秒。采用该局裁判结果佐证未接稳，不将前五次试验并入此节点；成功参考录像约视频 3.92–4.92 s 可见球停留在伸出的手边，本局没有这一持续持球段。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–3.96 s 中机器人双脚交替接触场地地面，双腿支撑身体向落球位置走动并迎球，未见摔倒。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "其他",
+      "note_zh": "观察：log step 191 的最终局原生裁判确认棒球在被接住前碰到身体；视频 3.92–3.96 s 球下降到机器人手臂及身体附近，远景不足以区分具体禁触身体部位。候选接触方式为“被动碰撞”：飞球撞到身体，既非抓握、托举，也非机器人主动推扫或击打。"
+     }
+    ]
+   }
+  },
+  "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": 3.96,
+   "log_step": 191,
+   "where": "操作中",
+   "evidence_zh": "观察：视频约 3.4–3.96 s 机器人调整右臂迎接下降的球，但末段球没有留在手中；log step 189、191 确认这是最终第六局，裁判在control step 398因球接住前触身终止该局，接住与保持阶段均未到达。断点为正在执行的“接稳飞球”。"
+  },
+  "unmet": {
+   "fly_ball_caught": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "动态时机",
+    "t": 3.96,
+    "log_step": 191,
+    "evidence_zh": "观察：视频 3.84–3.96 s 白球快速下降至机器人右臂及身体附近，右手未将球截停、包住；最终局原生裁判在log step 191报告球接住前碰身体、接住阶段未到达。推断：机器人迎接运动中的球时未实现手球同步拦截，属于动态时机失配。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "source_zh": "日志",
+    "evidence_zh": "推断：agent脚本在log step 187将右臂目标于仿真 3.30–3.80 s 插值到预设接球姿态，闭手指令到仿真 3.90 s 才开始、仿真 4.05 s 才达到1；log step 189实际提交该计划后，log step 191在仿真 3.98 s 的终局读数为hand_closure_right=0.162、holding_right=0。结合视频 3.84–3.96 s 球到达而未被截停，说明预设手臂运动和闭手控制未在球通过时建立包围接触。log step 219另有agent自行计算的末态手部深度估计与FK预测不一致，但其坐标和误差尚未独立核实，因此只归为控制，不细分子类。"
+   }
+  ],
+  "gap_zh": "没有状态日志；25 fps远景不能确定手指是否短暂擦到球、触身的具体部位、90°包围接触或0.02秒中断。原生裁判记录足以判定接住阶段未达成；agent在log step 219、227、232报告的手位误差来自其自行计算，不能据此确定精确偏差，也不足以将根因进一步区分为模型/逆解误差或闭手时序问题。三个原生数值指标在reward字段中均未报告，故填null。"
+ },
+ "summary_zh": "最终评分局中，机器人走向落球位置并伸右手迎球，但“接稳飞球”始终未达成，节点终态为0/1。视频约 3.96 s 球到达身体附近，最终局原生裁判确认接住前触身；预设手臂及闭手控制未实现同步拦截，标注动态时机与控制。远景和缺少状态日志限制了微观接触及具体控制子因的判断。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "节点 fly_ball_caught 未达成、断点、动态时机与控制（开环预设臂/闭手时序，无反馈）均有 log step 189/191 的服务原生裁判字段（control_steps=398、stages_reached=0、球接住前触身）支撑；视频 3.92 s 可见球在手臂上方附近，末帧球被遮挡，与标注的观察/推断划分一致。时间钟名清楚，未换算 agent 时间。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:15:36Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard-realtime--mduez8w/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard-realtime--mduez8w/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "8e31aaeca7ca9207b6c28652c21e1fddaea9bce4ff4cefdaaec10e11ae0f1e13",
+   "video_sha256": "a40d17b4ac09fc08120383b72daf10bdd72eb234e9f14b41ef3cdf11b1fe5fde",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch/trajectories/embodied-olympics-fun-catch-i00-standard-realtime--qhbhsou.json

@@ -0,0 +1,221 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-i00-standard-realtime--qhbhsou",
+ "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": 425,
+   "sim_steps": 2488,
+   "calls": 3334,
+   "rows": 19665,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 747.8,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "realtime": 1,
+   "slowdown": 1.0,
+   "clock_s": 24.865,
+   "max_lag_s": 0.0394,
+   "stalls": 0,
+   "stalled_s": 0.0,
+   "stalled_share": 0.0,
+   "held_steps": 404,
+   "late_steps": 2,
+   "requests_refused": 5,
+   "request_busy_s": 2.399,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 1,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.0,
+   "peak_processes": 0.0,
+   "peak_disk_gb": 0.0,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-catch-i00-standard-realtime--qhbhsou/replay.mp4",
+  "duration_s": 4.28,
+  "fps": 25.0,
+  "frames": 107,
+  "sha256": "a887c21995428c8b547af099de3bc14ff73c9eb23bba3c520d3d9442cd255738"
+ },
+ "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 机器人站在场地内，手中无球；视频 0.40–3.60 s 机器人向前侧移并伸出右手准备接球。"
+     },
+     {
+      "t": 3.96,
+      "log_step": 125,
+      "text_zh": "观察：视频 3.88–3.96 s 白球下降到机器人右手边；视频 4.00–4.24 s 球从右手边向下离开并到达地面，没有随手保持。log step 125 的机器人服务返回第六次尝试结束，outcome 为 the ball touched the ground，stages_reached 为 0、stages_total 为 2；这里的阶段值来自任务原生裁判，表示接住和保持两个验收阶段均未通过。"
+     },
+     {
+      "t": 4.24,
+      "log_step": 125,
+      "text_zh": "推断：视频 3.96–4.24 s 的球手分离和落地与 log step 125 的最后一次现场尝试判决一致，接稳飞球从未完成，末态为假；前五次重置前的调试不计入本段。成功参考在参考视频约 4.00–4.92 s 可见球留在右手边，本段没有这一持续持球表现。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "其他",
+      "note_zh": "推断：视频约 3.96–4.04 s 球到达右手边后改变运动并下落，右手与球发生短暂碰触，未形成可见包握。候选名称“短暂碰触”：画面不支持抓握、持续托举、推扫或压擦，故用其他；无法分辨具体指节或掌缘。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–4.24 s 机器人双脚在地面交替支撑、向接球位置移动，随后站立伸右手；片尾没有摔倒。"
+     }
+    ]
+   }
+  },
+  "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": 125,
+   "where": "操作中",
+   "evidence_zh": "观察：视频约 3.96–4.00 s 机器人正用伸出的右手承接下降的球，随后视频 4.04–4.24 s 球离开手边、下落至地面，接球没有完成或得到补救。log step 125 明确记录最后一次重置后的第六次现场尝试因球触地结束、阶段完成数 0/2。"
+  },
+  "unmet": {
+   "fly_ball_caught": false
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没建立接触",
+    "t": 4.0,
+    "log_step": 125,
+    "evidence_zh": "推断：视频 3.96–4.24 s 球虽经过右手边并发生短暂碰触，却没有停留在手内，随即下落；log step 125 的原生裁判记录接住阶段也未通过（0/2）。缺失的是四指组与拇指或掌部的包围接触，不能把短暂碰球算作已经抓住后脱手。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "t": 4.0,
+    "log_step": 123,
+    "source_zh": "日志",
+    "evidence_zh": "推断：log step 123 的最终控制脚本安排右臂到达承接姿态、右手预闭合后完全闭合，并安排手臂回撤缓冲；log step 125 确认实际执行该脚本。视频 3.96–4.24 s 中球已到达右手附近，机器人仍未形成包握而让球下落，说明这组接球动作未实现有效捕获。现有证据支持控制层失败，但不能进一步区分接触几何、闭手时机或关节跟踪误差。"
+   }
+  ],
+  "gap_zh": "无状态日志或逐步原生接触对、包握标记及手指实际轨迹；25 fps 场地远景只能定位球手相遇和随后落地，无法分辨具体接触部位、90°包围夹角或0.02秒中断。短暂右手触球由相遇前后的球运动推断；控制根因不细分子类，且不将前五次调试或 agent 在 log step 133 的动作总结当作本段已接住的证据。"
+ },
+ "summary_zh": "最后一次现场尝试中，机器人移动到接球位置并伸出右手，视频约 3.96–4.00 s 球到达手边，但未建立包握，随后落地，接稳飞球最终完成 0/1。log step 125 的原生裁判返回球触地、阶段完成数 0/2，与画面一致；失败归于接球控制，无法从远景和现有日志细分闭手时机、接触几何或跟踪误差。",
+ "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 fly_ball_caught 为假、断点在该步骤、log step 125 的原生裁判（球触地、阶段 0/2）与帧一致；标签保守，未发现需改内容的错误。右手与球是否真有短暂碰触在 25 fps 远景中无法确定，文件已标为推断并写入 gap_zh。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:16:24Z",
+  "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-standard-realtime--qhbhsou/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard-realtime--qhbhsou/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "8c28fd6f5c8e802d5f98f968993f5fa719d19eb40aeac945b55c18cf0a295931",
+   "video_sha256": "a887c21995428c8b547af099de3bc14ff73c9eb23bba3c520d3d9442cd255738",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/catch/trajectories/embodied-olympics-fun-catch-i00-standard-realtime--xvrrhhr.json

@@ -0,0 +1,200 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-catch-i00-standard-realtime--xvrrhhr",
+ "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": 496,
+   "sim_steps": 1700,
+   "calls": 680,
+   "rows": 13040,
+   "resets": 3,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 2037.0,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "realtime": 1,
+   "slowdown": 1.0,
+   "clock_s": 16.997,
+   "max_lag_s": 0.0307,
+   "stalls": 0,
+   "stalled_s": 0.0,
+   "stalled_share": 0.0,
+   "held_steps": 2,
+   "late_steps": 345,
+   "requests_refused": 1,
+   "request_busy_s": 1.175,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 1,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.0,
+   "peak_processes": 0.0,
+   "peak_disk_gb": 0.0,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-catch-i00-standard-realtime--xvrrhhr/replay.mp4",
+  "duration_s": 5.0,
+  "fps": 25.0,
+  "frames": 125,
+  "sha256": "b89501d580be7ad1fbac72ef474d86f5150980ed0fff81da832365d1a20c84ea"
+ },
+ "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：本次视频 0.00 s 机器人站在场地上，双手未持球；视频 1.60–3.60 s 机器人向前移动并伸出右手准备接球，双脚交替支撑。"
+     },
+     {
+      "t": 3.96,
+      "log_step": null,
+      "text_zh": "观察：本次视频 3.92 s 白球位于机器人右手前上方，视频 3.96 s 白球下降到右手处；视频 4.00–4.96 s 白色球点留在右手中并随手小幅移动，机器人始终直立，未见球继续落向地面。已逐帧检查视频 3.76–4.96 s。"
+     },
+     {
+      "t": 4.96,
+      "log_step": 70,
+      "text_zh": "观察：log step 68 启动第4次现场尝试，log step 70 的该次服务终局记录给出 episode=4、control_steps=496、sim_time_s=4.96、outcome=success、stages_reached=2、stages_total=2；inputs/run.json 的评分器字段 success=1、live_success=1、graded_by_live=1，最终评分采用此次现场成功。"
+     },
+     {
+      "t": 4.96,
+      "log_step": 70,
+      "text_zh": "推断：受评第4局的控制步长为0.01 s，log step 70 的496个控制步对应视频 4.96 s；视频约 3.96 s 接球及视频 4.96 s 仍持球的末帧与这一映射相符。节点判据包含100步持球验收，故在视频 4.96 s 记完成，不能在最初接触时即记完成；包围接触角度及容许中断由原生裁判成功记录佐证，远景不能逐项分辨。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "机器人用右侧Inspire裸手接球，随后同一只手闭合包住白球至片尾。视频约 3.96–4.96 s 的球随右手保持可见，结合log step 70 的原生接球及持球两阶段成功判定抓握；未见左手参与持球。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "机器人双足接触场地地面，腿部支撑其行进至接球位置，并在右手接球、持球期间保持身体直立；依据本次视频 0.00–4.96 s。"
+     }
+    ]
+   }
+  },
+  "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": "受评视频对应第4次现场尝试：机器人行进后在视频约 3.96 s 用右手接到下降白球，并在视频 4.96 s 完成接稳飞球节点，终态为1/1。log step 70 的现场裁判记录为success、两阶段全部通过，与本次末帧持球且机器人直立的画面一致；前三次失败尝试不计入本次节点轨迹。没有状态日志，远景无法分辨90°包围接触和0.02秒中断，相关验收依赖原生裁判记录；评分字段未报告三项原生指标，均保留为null。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "成功运行，节点 fly_ball_caught 在视频 4.96 s 完成（log step 70 的 control_steps=496 对应 4.96 s，服务终局记录 success），接触标注与右手持球一致；未发现需要修改的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:17:55Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard-realtime--xvrrhhr/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-catch-i00-standard-realtime--xvrrhhr/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "74f9effb394c03ed7f482848d233b86984a949b52a5b6df812f5ce3c1e9b2fa9",
+   "video_sha256": "b89501d580be7ad1fbac72ef474d86f5150980ed0fff81da832365d1a20c84ea",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/discus/trajectories/embodied-olympics-fun-discus-i00-privileged--2yxpvjl.json

@@ -0,0 +1,218 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-discus-i00-privileged--2yxpvjl",
+ "taskKey": "embodied-olympics-fun/discus",
+ "dag_sha256": "ffb63c003bdbc581b50872ffe890e6ce579309876e24b407597d79050cb48e95",
+ "outcome": {
+  "success": 0,
+  "reward": 0.3333,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "n_actions": 367,
+   "sim_steps": 367,
+   "horizon": 6000,
+   "episode_over": 1,
+   "progress": 0.3333,
+   "stages_reached": 1,
+   "stages_total": 3,
+   "fell": 0,
+   "replay_s": 5.99,
+   "bad_trajectory": 0,
+   "metric_distance_m": 3.897,
+   "metric_release_height_m": 1.683,
+   "metric_release_speed_m_s": 6.58,
+   "metric_release_spin_rad_s": 35.7,
+   "metric_release_tilt_deg": 68.7,
+   "metric_release_time_s": 2.418,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 752,
+   "agent_resets": 8196,
+   "agent_replays": 6,
+   "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": 18.789,
+   "peak_processes": 30,
+   "peak_disk_gb": 0.982,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "over_memory_gb": 1,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-discus-i00-privileged--2yxpvjl/replay.mp4",
+  "duration_s": 3.68,
+  "fps": 25.0,
+  "frames": 92,
+  "sha256": "69709503b5c66d3cc78fed2d1ebd4780126eb7818726f72f4b2b52b332235094"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "discus_landed_qualified": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–2.40 s，机器人右手持橙红色铁饼，转动躯干并摆右臂；视频 2.44–2.52 s，铁饼与右手分离并向圈前飞出。"
+     },
+     {
+      "t": 3.64,
+      "log_step": 295,
+      "text_zh": "观察：视频 3.28–3.64 s，铁饼在圈前近处持续下降，机器人仍在圈内。log step 295 对最终交付文件的原生重放报告 distance_m=3.897、failed=\"the throw is short of the qualifying mark\"、stages=[\"release\"]；与 inputs/run.json 的评分距离和367个执行步一致，未达到15.0米首次落点门槛。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "观察：视频 0.00–2.40 s，机器人右手掌与弯曲手指围持橙红色铁饼，铁饼随右手转动和摆动，随后脱手。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–3.64 s，机器人双腿支撑投掷动作，脚在灰色圈内接地，躯干旋转带动右臂。"
+     }
+    ]
+   },
+   "rear_exit_stood": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 3.64,
+      "log_step": 295,
+      "text_zh": "观察：视频 0.00–3.64 s，机器人没有从后半圈迈到绿色圈外地台；末段仅在圈内完成摆臂与身体倾斜，未出现后退离圈站稳的动作。log step 295 的最终交付重放只达到 release 阶段，投掷距离不足即结束。"
+     },
+     {
+      "t": 3.64,
+      "log_step": 298,
+      "text_zh": "推断：有效落地前置节点未达，后圈离圈站稳节点不成立。agent 自述离圈步法曾单独测试、但交付中未运行（log step 298），不将会话中的离圈测试计入受评轨迹；成功参考视频 4.00–6.60 s 可见后退跨出灰圈的动作，本次录像中没有该动作。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "distance_m",
+   "value": 3.897,
+   "unit": "m",
+   "threshold": 15.0,
+   "direction": "higher_better"
+  },
+  {
+   "name": "retouch_dv_m_s",
+   "value": null,
+   "unit": "m/s",
+   "threshold": 0.5,
+   "direction": "lower_better"
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "discus_landed_qualified",
+   "t": 2.44,
+   "log_step": 295,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 2.40–2.52 s，机器人摆右臂将铁饼投出，视频 3.28–3.64 s 铁饼在圈前近处下降；log step 295 对最终交付文件重放确认首次落点距离为3.897米，裁判以投掷不足15.0米结束本次尝试，仅 release 阶段成立。推断：未修复的失败步骤是铁饼落区达标，断点以出手画面作导航锚点。"
+  },
+  "unmet": {
+   "discus_landed_qualified": false,
+   "rear_exit_stood": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 2.44,
+    "log_step": 295,
+    "evidence_zh": "观察：视频 2.44–3.64 s，机器人投出的铁饼飞出后落向圈前近处，未送到远处目标区域；log step 295 的原生裁判重放给出首次落点距离3.897米，并明确判定 \"the throw is short of the qualifying mark\"。这是本次投掷落点过近，非目标处释放后失稳。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "source_zh": "日志",
+    "evidence_zh": "推断：投掷控制未能维持持饼加速至脚本预定释放。log step 261 的最终参数设置 t_go=2.761、t_snap=2.699；log step 272 的生成代码在仿真 t_go 之前保持 grip=1，并在仿真 t_snap 才开始躯干回摆，但 log step 295 的同一交付重放已在仿真 2.418 s 记录最后持握与6.58米/秒出手速度，随后首次落点仅3.897米。视频 2.40 s 饼仍贴近右手、视频 2.48 s 已分离，与提前脱手一致；具体是接触承载还是关节跟踪限制，现有记录不能区分。"
+   }
+  ],
+  "gap_zh": "本次没有状态日志。斜视录像不能精确测量首次接地点和距离，末帧视频 3.64 s 时铁饼仍接近地面，首次接地及3.897米结论由最终交付文件的原生重放结果（log step 295）与评分字段共同确认。缺少受评投掷的逐帧接触力、持握接触对和关节跟踪记录，故控制根因不细分；retouch_dv_m_s 未报告，保留 null。"
+ },
+ "summary_zh": "机器人在圈内右手持饼、转身摆臂并投出，视频约 2.44 s 可见脱手，但首次落点仅距圈内沿3.897米，铁饼落区达标与后圈离圈站稳均未完成（0/2）。断点为投掷操作，现象是落点过近；最终脚本预定松手之前已脱手，支持控制失败，但没有接触力和关节跟踪记录来细分原因。精确首次落点依照 log step 295 的最终交付重放及评分字段判定。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "节点值、时间、断点、现象（运动·位姿不准）与根因（控制·控制）均有视频与 log step 295 重放（distance_m=3.897、release 阶段成立）支持，且与 03 的投篮先例一致；未发现需要改动的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:04:11Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-discus-i00-privileged--2yxpvjl/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-discus-i00-privileged--2yxpvjl/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "0f9d065f69032b7e355c7632b5c4330a254ce811a4b36a8e2ee0a36388a37d80",
+   "video_sha256": "69709503b5c66d3cc78fed2d1ebd4780126eb7818726f72f4b2b52b332235094",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/discus/trajectories/embodied-olympics-fun-discus-i00-privileged--pxbblpf.json

@@ -0,0 +1,215 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-discus-i00-privileged--pxbblpf",
+ "taskKey": "embodied-olympics-fun/discus",
+ "dag_sha256": "ffb63c003bdbc581b50872ffe890e6ce579309876e24b407597d79050cb48e95",
+ "outcome": {
+  "success": 0,
+  "reward": 0.3333,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "n_actions": 199,
+   "sim_steps": 199,
+   "horizon": 6000,
+   "episode_over": 1,
+   "progress": 0.3333,
+   "stages_reached": 1,
+   "stages_total": 3,
+   "fell": 1,
+   "replay_s": 5.7,
+   "bad_trajectory": 0,
+   "metric_release_height_m": 1.767,
+   "metric_release_speed_m_s": 11.52,
+   "metric_release_spin_rad_s": 15.3,
+   "metric_release_tilt_deg": 55.9,
+   "metric_release_time_s": 1.492,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 475,
+   "agent_resets": 369,
+   "agent_replays": 5,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 1,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.0,
+   "peak_processes": 0.0,
+   "peak_disk_gb": 0.0,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-discus-i00-privileged--pxbblpf/replay.mp4",
+  "duration_s": 2.0,
+  "fps": 25.0,
+  "frames": 50,
+  "sha256": "1765980400ad76d749738e6ddc5f139c2487b4d40d82d595003f02b333b9de73"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "discus_landed_qualified": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–1.48 s，机器人站在圈内，右手持橙红色铁饼转体、摆臂；视频 1.52–1.56 s 铁饼与右手分离并向扇形方向飞出。"
+     },
+     {
+      "t": 1.8,
+      "log_step": null,
+      "text_zh": "观察：视频 1.60–1.96 s 机器人持续倾倒，视频 1.80 s 铁饼仍在画面上方空中，随后飞出视野；截至视频 1.96 s 没有看到首次触地。评分器 reward 记录 stages_reached=1、stages_total=3、fell=1，且仅报告脱手诊断值，没有 distance_m；结合已见脱手与倒地，达标落地节点未完成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "观察：视频 0.00–1.48 s，机器人右手持握铁饼，铁饼随右手转体摆臂运动；视频 1.52–1.56 s 脱手后进入无接触飞行。初始手指细节较小，但持握随动及评分器确认的脱手阶段共同支持抓握。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–1.48 s，机器人双脚在投掷圈内承重，腿和躯干支撑转体投掷；视频 1.60–1.96 s 身体倾倒并失去正常足部支撑。"
+     }
+    ]
+   },
+   "rear_exit_stood": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 1.96,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–1.96 s，机器人只完成圈内转体投掷，随后倒向画面左侧，末帧身体接近地面、单腿翘起；没有有效落地后向后半圈迈出并连续站稳的过程。"
+     },
+     {
+      "t": null,
+      "log_step": 953,
+      "text_zh": "观察：log step 953 的最终生成脚本包含蓄力、加速投掷及收臂阻尼恢复，随后保存轨迹，没有后圈离圈动作；评分器仅到达一个原生阶段且 fell=1。推断：本节点的有效落地前置未满足，亦未进入合法离圈站稳步骤。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "distance_m",
+   "value": null,
+   "unit": "m",
+   "threshold": 15.0,
+   "direction": "higher_better"
+  },
+  {
+   "name": "retouch_dv_m_s",
+   "value": null,
+   "unit": "m/s",
+   "threshold": 0.5,
+   "direction": "lower_better"
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "discus_landed_qualified",
+   "t": 1.6,
+   "log_step": null,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 1.52–1.56 s 铁饼刚脱离右手，机器人投掷随动后明显倾斜；视频 1.60–1.96 s 躯干持续下降直至倒地，铁饼在视频 1.80 s 尚在空中。评分器 reward 的 fell=1、episode_over=1，原生三阶段只完成一个。推断：机器人在完成达标落地前、仍进行投掷及收势时失去平衡，本次投掷未修复，故断点为铁饼落区达标。"
+  },
+  "unmet": {
+   "discus_landed_qualified": false,
+   "rear_exit_stood": false
+  },
+  "phenomena": [
+   {
+    "cat": "全身",
+    "item": "平衡",
+    "t": 1.6,
+    "log_step": null,
+    "evidence_zh": "观察：视频 1.60–1.76 s，机器人躯干从斜立转为接近水平，支撑脚未能接住身体；视频 1.80–1.96 s 躯干继续跌落、另一条腿翘起，末帧已倒地。评分器 reward 同时记录 fell=1。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "log_step": 953,
+    "evidence_zh": "推断：log step 953 的交付生成代码以 lowcmd 执行快速腰部转动与肩腕加速，投后仅收右臂、保持腿部关节目标并对腰部加阻尼；视频 1.52–1.96 s 显示该动作未能在脱手后恢复身体支撑而持续倒地，评分器记录 deterministic=1、fell=1。证据支持投掷与收势的全身稳定控制失败，但不足以单独归因于接触布局、反馈律或某个增益。",
+    "source_zh": "日志 / 视频 / 评分器字段"
+   }
+  ],
+  "gap_zh": "没有状态日志，视频也不能分辨逐指接触、足部受力及关节跟踪误差，因而控制根因不细分。铁饼在视频结束前没有可见首次落地，评分器未报告 distance_m 或 retouch_dv_m_s，两个原生指标均留 null；log step 965 中 agent 自述的落点 13.05 m 不能作为本次评分轨迹的落地事实，视频与评分器显示其先因跌倒终止。"
+ },
+ "summary_zh": "机器人完成右手持饼转体并脱手，但视频 1.60–1.96 s 显示其在铁饼仍飞行时失衡倒地；两个 DAG 节点均未完成，原生脱手阶段的 1/3 不等于一个 DAG 节点完成。断点为铁饼落区达标的操作中，现象为全身平衡失败，最终生成脚本与画面支持控制根因。无状态日志或首次落地记录，不能核验落距、接触冲量或进一步区分失稳的控制机制。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "标签与视频一致：圈内转体投掷、约1.5 s脱手、随后倒地，铁饼在视频结束前未见落地；两个节点未完成、断点在落区达标、现象平衡、根因停在控制均有证据支持。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:10:36Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-discus-i00-privileged--pxbblpf/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-discus-i00-privileged--pxbblpf/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "9b925ff099229d377044799d86b65c4a0872a19c4e00097c1b013459b30bcb94",
+   "video_sha256": "1765980400ad76d749738e6ddc5f139c2487b4d40d82d595003f02b333b9de73",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/discus/trajectories/embodied-olympics-fun-discus-i00-privileged--qz6geqr.json

@@ -0,0 +1,222 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-discus-i00-privileged--qz6geqr",
+ "taskKey": "embodied-olympics-fun/discus",
+ "dag_sha256": "ffb63c003bdbc581b50872ffe890e6ce579309876e24b407597d79050cb48e95",
+ "outcome": {
+  "success": 0,
+  "reward": 0.3333,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "n_actions": 285,
+   "sim_steps": 285,
+   "horizon": 6000,
+   "episode_over": 1,
+   "progress": 0.3333,
+   "stages_reached": 1,
+   "stages_total": 3,
+   "fell": 0,
+   "replay_s": 5.87,
+   "bad_trajectory": 0,
+   "metric_distance_m": 14.075,
+   "metric_release_height_m": 1.753,
+   "metric_release_speed_m_s": 11.97,
+   "metric_release_spin_rad_s": 17.9,
+   "metric_release_tilt_deg": 54.5,
+   "metric_release_time_s": 1.244,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 131,
+   "agent_resets": 14677,
+   "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": 2.188,
+   "peak_processes": 20,
+   "peak_disk_gb": 0.243,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-discus-i00-privileged--qz6geqr/replay.mp4",
+  "duration_s": 2.88,
+  "fps": 25.0,
+  "frames": 72,
+  "sha256": "5e8a1937a49c6001be2bc6ca4ce975b6694b6dd51d820f3bf8720b51e22b5900"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "discus_landed_qualified": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.88,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00 s 橙红色铁饼位于机器人右手，视频 0.60–1.20 s 随右手转体、摆臂；视频 1.24–1.32 s 铁饼离手向场地飞出，随后机器人收臂并留在圈内。"
+     },
+     {
+      "t": 2.8,
+      "log_step": 157,
+      "text_zh": "观察：视频 2.48–2.84 s 铁饼向远处草地下降，末帧可见小橙点落在场地；最终交付的重放输出（log step 157）报告 failed=\"the throw is short of the qualifying mark\"、stages=[\"release\"]，交付文件为 285 行。评分器 reward.metric_distance_m=14.075，低于节点要求的 15.0 米；远处落点的精确距离取评分器记录，未由画面测量。"
+     },
+     {
+      "t": 2.84,
+      "log_step": 158,
+      "text_zh": "推断：本次投掷虽已脱手并落地，但首次落点距离不足，故该节点从未成立。agent 在 log step 158 自述的 14.15 米沿用了会话优化结果（log step 156 为 14.153 米），本标签采用正式评分器的 14.075 米；不把调试试投当成本次评分结果。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "推断：视频 0.00–1.20 s 右侧灵巧手持住铁饼边缘，铁饼从水平转向近竖直并随手摆动，之后脱手飞行；结合任务初始手指扣住饼缘的说明，判为右手抓握，未见左手参与持饼。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–2.84 s 机器人双脚在投掷圈内支撑身体，躯干转动配合右臂投掷，投后仍直立在圈内。"
+     }
+    ]
+   },
+   "rear_exit_stood": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.84,
+      "log_step": 157,
+      "text_zh": "观察：视频 1.56–2.84 s 机器人收臂后仍在灰色投掷圈内，未向后圈外迈出。成功参考视频约 4.80–6.80 s 可见机器人向圈后绿色地台后退并停住，本次没有这一动作；log step 157 的最终重放只达到 release 阶段，未出现 landing 或 exit。"
+     },
+     {
+      "t": 2.84,
+      "log_step": 141,
+      "text_zh": "推断：前置的合格落地未成立，且本次视频没有后圈离圈动作，故后圈离圈站稳节点未达成，也未开始尝试。log step 141 的生成脚本修改明确在落地判负后保存投掷轨迹并返回，未执行离圈分支；log step 157 确认最终保存的是这类投掷判负结果。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "distance_m",
+   "unit": "m",
+   "threshold": 15.0,
+   "direction": "higher_better",
+   "value": 14.075
+  },
+  {
+   "name": "retouch_dv_m_s",
+   "unit": "m/s",
+   "threshold": 0.5,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "discus_landed_qualified",
+   "t": 2.8,
+   "log_step": 157,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 1.24–1.32 s 机器人将铁饼掷出，视频 2.76–2.84 s 铁饼降到远处草地，机器人仍在圈内直立。log step 157 的最终交付重放明确报告投掷距离不足，评分器首次落点距离为 14.075 米，未达到 15.0 米；这是正在执行的投掷步骤失败，后圈离圈尚未开始。"
+  },
+  "unmet": {
+   "discus_landed_qualified": false,
+   "rear_exit_stood": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 2.8,
+    "log_step": 157,
+    "evidence_zh": "观察：视频 1.28–2.84 s 可见铁饼离手飞行后降向场地，机器人没有跌倒；log step 157 的交付重放报告 \"the throw is short of the qualifying mark\"，正式评分器 metric_distance_m=14.075。推断：投掷的首次落点落在所需距离之前，差 0.925 米，属于投掷落点偏差。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "source_zh": "日志 / 视频 / 评分器字段",
+    "evidence_zh": "推断：agent 的摆臂与释放控制未形成能达到距离门槛的投掷。会话优化（log step 156）围绕摆臂、手腕及释放参数搜索，最佳候选仍被裁判判为距离不足；最终保存并重放的 285 行轨迹（log step 157）同样距离不足，视频 1.24–2.84 s 与投出后短距离落地的过程一致，评分器记录脱手速度 11.97 米/秒、首次落点 14.075 米。目标与阶段顺序正确，证据只支持控制层归因，不能确定是哪一项参数或接触细节造成。"
+   }
+  ],
+  "gap_zh": "无状态日志；斜视视频中远处铁饼只有少量像素，首次落地距离依正式评分器和最终重放判词确定，无法独立精测首次触地时刻或接触冲量，retouch_dv_m_s 未报告。agent 在 log step 158 自述摩擦滑移限制速度，但没有本次投掷的手指接触力、滑移及指令跟踪记录，不能据此确认抓握设计、参数调试等更细根因，更不能认定所有合法方案不可行。"
+ },
+ "summary_zh": "本次评分轨迹完成了持饼转体、摆臂和脱手，但首次落点仅为 14.075 米，未达到 15.0 米门槛，两个语义节点最终均未完成（0/2）。断点为铁饼落区达标，现象为投掷落点过近，根因仅归到控制；机器人未跌倒，后圈离圈分支尚未执行。无状态日志，精确落点依评分器，agent 所称 14.15 米及摩擦速度上限不作为本次精确结果或已证实根因。",
+ "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（投掷步骤，操作中）、位姿不准+控制根因与评分器 14.075 m 一致，未发现需改动的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:12:30Z",
+  "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-discus-i00-privileged--qz6geqr/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-discus-i00-privileged--qz6geqr/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "514d1513fc840007d3099e7b8833dc1fc7f94190a19df1f41ae6b8c0f09c2f3a",
+   "video_sha256": "5e8a1937a49c6001be2bc6ca4ce975b6694b6dd51d820f3bf8720b51e22b5900",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/discus/trajectories/embodied-olympics-fun-discus-i00-privileged--ugrcl8n.json

@@ -0,0 +1,251 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-discus-i00-privileged--ugrcl8n",
+ "taskKey": "embodied-olympics-fun/discus",
+ "dag_sha256": "ffb63c003bdbc581b50872ffe890e6ce579309876e24b407597d79050cb48e95",
+ "outcome": {
+  "success": 0,
+  "reward": 0.3333,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "n_actions": 175,
+   "sim_steps": 175,
+   "horizon": 6000,
+   "episode_over": 1,
+   "progress": 0.3333,
+   "stages_reached": 1,
+   "stages_total": 3,
+   "fell": 0,
+   "replay_s": 4.52,
+   "bad_trajectory": 0,
+   "metric_distance_m": 2.251,
+   "metric_release_height_m": 2.26,
+   "metric_release_speed_m_s": 6.08,
+   "metric_release_spin_rad_s": 2.3,
+   "metric_release_tilt_deg": 52.7,
+   "metric_release_time_s": 0.189,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 172,
+   "agent_resets": 1042,
+   "agent_replays": 2,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.828,
+   "peak_processes": 13,
+   "peak_disk_gb": 0.239,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-discus-i00-privileged--ugrcl8n/replay.mp4",
+  "duration_s": 1.76,
+  "fps": 25.0,
+  "frames": 44,
+  "sha256": "352d6524f390a768ecf17d23b8f7330fd6ba76fdddcff938575f22a1315aa098"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "discus_landed_qualified": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.28,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–0.20 s，机器人右手带着橙红色铁饼向上摆动；视频 0.24–0.56 s，铁饼被抛至机器人头顶上方并接近画面上边缘，随后向画面左侧移动，目标扇形位于画面右上方。"
+     },
+     {
+      "t": 1.72,
+      "log_step": 149,
+      "text_zh": "观察：视频 1.36–1.72 s，橙红色铁饼从画面左上部向左边缘草地方向下降，未见它进入右上方目标扇形；圈左侧的深色斑点在视频 1.00、1.40、1.72 s 持续向左移动，并非保持不动的落地物体。推断：深色斑点是飞行中铁饼的投影，不能据此认定铁饼较早落地。观察：log step 149 的官方交付文件重放结果为 failed=\"the discus landed outside the sector\"、stages=[\"release\"]、distance_m=2.251；log step 154 再次生成并重放最终交付文件，得到相同结果，且 run.json 的评分器字段 metric_distance_m=2.251、deterministic=1。"
+     },
+     {
+      "t": 1.72,
+      "log_step": 154,
+      "text_zh": "推断：最终交付文件的官方重放与本段视频的反向飞行相符，首次落地既在扇形外，距离 2.251 米也低于 15 米，因此本节点全程未达成；不以原生脱手阶段代替本 DAG 的落区达标节点。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "推断：视频 0.00–0.20 s 铁饼随右手摆动，随后脱离；结合 inputs/run.json 的 task_text 所述初始手指卷住饼沿，以及 log step 149 的原生 release 阶段，右侧五指手初始抓握铁饼再投出。斜视画面不能分辨每根手指的接触。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–0.52 s，机器人双脚在灰色投掷圈内接地，腿和躯干支撑右臂摆动投饼。"
+     }
+    ]
+   },
+   "rear_exit_stood": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 1.72,
+      "log_step": 147,
+      "text_zh": "观察：视频 0.60–1.72 s，机器人有脚步与身体位移，仍在圈内并逐渐向画面右侧明显前倾，未出现从后侧绿色地台离圈并站稳的画面；log step 147 的最终生成脚本在投掷后切入负向行走指令，log step 154 的最终官方重放仍仅报告 release 阶段。"
+     },
+     {
+      "t": 1.72,
+      "log_step": 154,
+      "text_zh": "推断：有效落地这一前置节点始终未达，且视频至 1.72 s 末帧没有后圈离圈站稳结果，因此本 final 节点未达成。成功参考视频约 4.8–6.4 s 可见机器人到达圈后绿色地台，而本次视频没有该结果；参考只用于识别视图中的离圈表现。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.60–1.72 s，机器人脚部在投掷圈内支撑并迈动，身体发生位移。推断：结合 log step 147 的负向行走指令，这是交付动作中离圈阶段的尝试，但没有完成后半圈离圈或站稳。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "distance_m",
+   "unit": "m",
+   "threshold": 15.0,
+   "direction": "higher_better",
+   "value": 2.251
+  },
+  {
+   "name": "retouch_dv_m_s",
+   "unit": "m/s",
+   "threshold": 0.5,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "discus_landed_qualified",
+   "t": 0.32,
+   "log_step": 149,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 0.28–0.52 s，机器人右手投出的铁饼向画面左上方飞去，目标扇形在画面右上方；此后没有补救投掷。log step 149 的交付文件官方重放及 log step 154 的最终复验均报告扇形外落地、distance_m=2.251、仅 release 阶段；失败发生在投掷落区这一步。"
+  },
+  "unmet": {
+   "discus_landed_qualified": false,
+   "rear_exit_stood": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 0.32,
+    "log_step": 149,
+    "evidence_zh": "观察：视频 0.24–0.56 s，铁饼脱离机器人右手后被抛至头顶上方，随后向画面左侧移动；视频 1.36–1.72 s，橙红色铁饼在左缘附近下降，背离画面右上方目标扇形。log step 149 官方重放判定首次落点在扇形外，距离仅 2.251 米，未投到规定落区。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "evidence_zh": "推断：log step 147 的最终生成脚本从初始持饼姿态对右肩俯仰关节施加恒定 -130 N·m 力矩并撤去该关节位置/速度增益，按固定时序打开右手；log step 154 确认该脚本生成的最终文件重复得到扇形外落点和 2.251 米距离，与视频 0.28–0.52 s 的偏向飞行一致。机器人已执行投掷，但这套摆臂与松手控制没有形成朝目标扇形的有效投射；证据不足以进一步区分接触设计、轨迹或反馈子类。",
+    "source_zh": "日志"
+   }
+  ],
+  "gap_zh": "没有状态日志。视频只有斜视场地视角，手指接触细节无法逐根辨认，首次触地点位于画面左边缘附近，精确距离与界外判定取自最终交付文件的官方重放和评分器字段；未报告 retouch_dv_m_s，不能从外观估算。缺少交付投掷的逐步接触力和关节跟踪数据，根因仅确定到控制大类。"
+ },
+ "summary_zh": "机器人从右手初始持饼姿态摆臂投出铁饼，但视频 0.28–0.52 s 可见铁饼背离目标扇形飞去；最终官方重放确认扇形外落地、距离 2.251 米，两项 DAG 节点均未达成。断点是铁饼落区达标，现象为位姿不准，证据支持摆臂和松手控制未形成合格投射；后续脚步也未完成后圈离圈站稳。无状态日志，精确落点和距离依赖官方重放，控制子类及细粒度手指接触尚不能确定。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 2,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "fix",
+    "conf": 7,
+    "summary_zh": "节点值（两节点均未达成）、断点（discus_landed_qualified，操作中）、现象与根因大类与视频和官方重放（扇形外、2.251 m、仅 release 阶段）相符；唯一需改的是证据中对铁饼飞行/落点画面的描述与帧不符。",
+    "n_issues": 1,
+    "n_major": 0
+   },
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 2,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "两节点均未达成、断点（discus_landed_qualified，操作中）、位姿不准现象与控制大类根因，和视频及官方重放（扇形外、distance_m=2.251、仅 release 阶段）一致；log step 147/149/154 存在且引用属实（147 的 -130 N·m 肩俯仰力矩脚本、149/154 的重放结果）。第 1 轮争议的铁饼后段描述已改为区分观察与推断，不再有需改动的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:14:58Z",
+  "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-discus-i00-privileged--ugrcl8n/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-discus-i00-privileged--ugrcl8n/04_review_r1"
+   },
+   {
+    "stage": "reconcile",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-discus-i00-privileged--ugrcl8n/05_reconcile_r1"
+   },
+   {
+    "stage": "traj_review",
+    "round": 2,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-discus-i00-privileged--ugrcl8n/04_review_r2"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "b40af8ae00b30e92c1a030bf529032743bedd851d661b8507c5f603a3ab73498",
+   "video_sha256": "352d6524f390a768ecf17d23b8f7330fd6ba76fdddcff938575f22a1315aa098",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/discus/trajectories/embodied-olympics-fun-discus-i00-privileged--uhxr9x9.json

@@ -0,0 +1,219 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-discus-i00-privileged--uhxr9x9",
+ "taskKey": "embodied-olympics-fun/discus",
+ "dag_sha256": "ffb63c003bdbc581b50872ffe890e6ce579309876e24b407597d79050cb48e95",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "success": 1,
+   "replay_success": 1,
+   "deterministic": 1,
+   "n_actions": 683,
+   "sim_steps": 683,
+   "horizon": 6000,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 3,
+   "stages_total": 3,
+   "fell": 0,
+   "replay_s": 7.44,
+   "bad_trajectory": 0,
+   "metric_distance_m": 16.267,
+   "metric_release_height_m": 2.087,
+   "metric_release_speed_m_s": 12.68,
+   "metric_release_spin_rad_s": 24.4,
+   "metric_release_tilt_deg": 71.7,
+   "metric_release_time_s": 1.118,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 124,
+   "agent_resets": 9839,
+   "agent_replays": 37,
+   "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": 2.443,
+   "peak_processes": 20,
+   "peak_disk_gb": 0.899,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-discus-i00-privileged--uhxr9x9/replay.mp4",
+  "duration_s": 6.84,
+  "fps": 25.0,
+  "frames": 171,
+  "sha256": "8bd1349417b6d1cda7f0d3be0a0ccc1fe73c8a4160f66bf316b88318e805bf53"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 2,
+  "final": 2,
+  "nodes": {
+   "discus_landed_qualified": {
+    "method": "visual",
+    "intervals": [
+     [
+      3.12,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 1.2,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–1.12 s 橙红色铁饼随机器人右手摆动，视频 1.16 s 铁饼仍靠近右手，视频 1.20 s 已与手分离并飞向扇形方向；机器人双脚在圈内支撑投掷。视频 2.96–3.24 s 的画面只覆盖近处场地，远处首次落点不可见。"
+     },
+     {
+      "t": 3.12,
+      "log_step": 200,
+      "text_zh": "观察：最终生成器返回的原生裁判状态（log step 200）记录 landing 阶段在仿真 3.113 s 达成、distance_m=16.267、failed=None，随后 success=True；log step 201 验证再生成动作与已验证重放逐行相同。评分器字段 replay_success=1、deterministic=1、metric_distance_m=16.267，与该最终交付一致。"
+     },
+     {
+      "t": 3.12,
+      "log_step": 200,
+      "text_zh": "推断：log step 200 的交付轨迹为 683 控制步、仿真 6.83 s，即每步 0.01 s，与视频 6.84 s 的长度吻合；其 release 阶段仿真 1.196 s 对应视频 1.20 s 可见离手，exit 阶段仿真 4.956 s 对应视频 4.96–5.00 s 可见后侧脚越沿落地，支持仿真时钟与视频时钟近似同起点同速。故将该次有效首次落地锚定于视频约 3.12 s；达标依据是最终交付裁判记录及评分器成功结果，未用滑滚位置估计落点。"
+     },
+     {
+      "t": 1.0,
+      "log_step": null,
+      "text_zh": "推断：任务文本给定初始右手手指卷住饼沿，视频 0.00、0.60、1.00、1.08 s 中铁饼随右手转动并保持在手端，支持右手抓握后投出；画面分辨率不足以逐指辨认所有接触点。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "右侧灵巧手持住橙红色铁饼并随摆臂投出；依据初始持握条件及视频 0.00–1.20 s 的随手运动与离手画面，具体手指接触点不能逐一分辨。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "机器人双脚接触圈内地面，腿和躯干承重并配合右臂转体投掷。"
+     }
+    ]
+   },
+   "rear_exit_stood": {
+    "method": "visual",
+    "intervals": [
+     [
+      6.83,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 4.96,
+      "log_step": 200,
+      "text_zh": "观察：视频 3.40–4.80 s 机器人向扇形反方向后退；视频 4.96–5.00 s 可见后侧脚越过灰色圈的后沿，随后继续后退。最终交付裁判记录（log step 200）在仿真 4.956 s 锁存 exit 阶段，与上述画面相符。"
+     },
+     {
+      "t": 6.8,
+      "log_step": 200,
+      "text_zh": "观察：视频 5.80–6.80 s 的逐帧画面显示机器人保持直立，双脚在后侧绿色地台小幅调整，没有倒地或手膝撑地；末帧视频 6.80 s 仍站立。log step 200 的最终交付原生裁判状态在仿真 6.83 s 返回 success=True、done=True，release、landing、exit 阶段齐全，��分器也报告 success=1、replay_success=1、fell=0。"
+     },
+     {
+      "t": 6.83,
+      "log_step": 200,
+      "text_zh": "推断：沿用已由视频 1.20 s 离手及视频 4.96–5.00 s 后圈触地核对的控制步时钟映射，最终成功事件仿真 6.83 s 对应视频约 6.83 s，故此时计入后圈离圈并站稳节点；不在仅锁存 exit 的视频约 4.96 s 提前计入。连续站稳窗口内的精确躯干方向余弦及骨盆速度由最终原生成功结果确认，录像不能精确量出。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "机器人左右脚交替接触圈内地面及后侧圈外绿色地台，腿和躯干支撑后退离圈，再保持直立低速站立；视频 4.96–6.80 s 可见。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "distance_m",
+   "unit": "m",
+   "threshold": 15.0,
+   "direction": "higher_better",
+   "value": 16.267
+  },
+  {
+   "name": "retouch_dv_m_s",
+   "unit": "m/s",
+   "threshold": 0.5,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "本次交付完成两个节点：右手投出的铁饼在视频约 3.12 s 首次有效落地，裁判距离为 16.267 米；机器人随后从后圈离圈，至视频约 6.83 s 满足站稳验收，评分与确定性重放均成功。节点时间结合视频与最终交付的控制步时钟核对，未采用会话中其他试投的结果。无状态日志，首次落点出画，逐指接触、精确站稳阈值及脱手后擦碰冲量不能由录像独立量出，其中 retouch_dv_m_s 未报告，保留为 null。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "成功运行，两节点均达成；log step 193/200 的原生裁判事件（release 1.196、landing 3.113、exit 4.956、success 6.83 仿真秒）与视频 1.2 s 离手、4.96 s 后脚越圈相符，控制步时钟映射有可见事件核对，时间、final、contact_actual 与无 failure 均无误。",
+    "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": "2026-10-31T00:13:53Z",
+  "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-discus-i00-privileged--uhxr9x9/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-discus-i00-privileged--uhxr9x9/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "861b2e6a8d4a231cc761463c42b707f310e60eb1a5edf2d59f426753de04ca0f",
+   "video_sha256": "8bd1349417b6d1cda7f0d3be0a0ccc1fe73c8a4160f66bf316b88318e805bf53",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/discus/trajectories/embodied-olympics-fun-discus-i00-standard--4gapg9s.json

@@ -0,0 +1,236 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-discus-i00-standard--4gapg9s",
+ "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": 497,
+   "sim_steps": 2647,
+   "calls": 140,
+   "rows": 2429,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 2913.1,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 0.932,
+   "peak_processes": 11,
+   "peak_disk_gb": 1.445,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-discus-i00-standard--4gapg9s/replay.mp4",
+  "duration_s": 5.0,
+  "fps": 25.0,
+  "frames": 125,
+  "sha256": "20483243b68a3dd0d0c325ebbc481408d61a627132a24c6f2de875c28df9e3b7"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "discus_landed_qualified": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–1.60 s，机器人在圈内站立，右手持橙红色铁饼；视频 2.00–4.12 s，机器人转动躯干、将持饼右臂后摆再前挥，双脚持续支撑身体。"
+     },
+     {
+      "t": 4.24,
+      "log_step": 184,
+      "text_zh": "观察：视频 4.20–4.24 s，铁饼从右手附近分离；视频 4.36–4.96 s，铁饼向画面左侧、白色扇形之外飞去和下降。log step 184 的第6次试验返回裁判字段 OUTCOME: the discus landed outside the sector、failed、stages_reached=1/3（服务的原生投掷阶段计数，并非本 DAG 的两节点计数），没有有效落地。"
+     },
+     {
+      "t": 4.96,
+      "log_step": 189,
+      "text_zh": "观察：视频末帧 4.96 s，机器人仍直立在投掷圈内；log step 189 的服务状态为 episode=6、control_steps=497、done=true、reason=failed。log step 187 的后续重置被拒绝，故第6次试验仍是被评分的一局，前几次调试试验不计入节点进度。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "右侧灵巧手弯指扣住铁饼边缘、掌部承托，持饼后摆并挥臂投出；视频可见铁饼随手运动，log step 184 的 holding_right 在摆臂多数采样为1，支持实际持握。脱手后铁饼自由飞行。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "机器人双脚在投掷圈内承重，腿和躯干支撑转体、摆臂及投后恢复姿态。"
+     }
+    ]
+   },
+   "rear_exit_stood": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 4.96,
+      "log_step": 184,
+      "text_zh": "观察：视频 0.00–4.96 s，机器人只在圈内持饼、摆臂、投出并恢复姿态，没有向后跨出圈沿后站稳的过程，视频 4.96 s 双脚仍在灰色圈内。log step 184 裁判报告铁饼落在扇形外，前置有效落地未达成；log step 172 的脚本只有在落地后服务仍未结束时才执行向后离圈，故本局未执行该段。参考成功录像的后退离圈在其视频约 4.50–6.80 s 可见，本次无该行为。"
+     }
+    ]
+   }
+  },
+  "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": 4.24,
+   "log_step": 184,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 4.24 s，机器人正在挥动右臂投饼，铁饼开始离手；视频 4.36–4.96 s，铁饼飞到白色扇形侧外。log step 184 的第6次试验由裁判以 the discus landed outside the sector 判负，投掷落区达标这一步未完成，尚未进入后圈离圈。"
+  },
+  "unmet": {
+   "discus_landed_qualified": false,
+   "rear_exit_stood": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 4.24,
+    "log_step": 184,
+    "evidence_zh": "观察：视频 4.24–4.96 s，铁饼离开机器人右手后向白线扇形的侧外飞去，没有沿扇形内方向投出；log step 184 的本局裁判反馈为 the discus landed outside the sector。投掷的出手方向及落区不符合目标。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "source_zh": "日志",
+    "evidence_zh": "推断：本局摆臂与释放控制未使铁饼沿目标扇形投出，具体控制子类尚不能确定。视频 4.24–4.96 s 可见铁饼离手后向扇形侧外飞去，log step 184 的服务裁判反馈为 the discus landed outside the sector。log step 176 的脚本使用 agent 运动学模型预测的方位、仰角和速度触发张手；log step 184 设置 min_speed=8.0 m/s、force_rel_row=48，行35–37的部分预测方位进入目标±11度时，预测速度约6.66–7.15 m/s，未满足速度门槛；行48进入强制释放条件时预测方位已为106.2/116.6度。这些数值并非原生铁饼释放指标，且 holding_right 在行47为0、行48又为1，不能据此确认实际脱手由门槛延迟造成。log step 187 中 agent 自述其本地模型估算达15米需12–14 m/s，该估算亦未独立验证；即使按其模型，窗口内约7 m/s也未显示足以达标。缺少放宽门槛改善实际投掷的验证，无法在参数调试、接触设计等具体原因间定论，故仅标控制。"
+   }
+  ],
+  "gap_zh": "无状态日志；单个场地斜视视频不足以精确确定首次地面接触点、distance_m、短时复触及冲量，最终落区不合格由本局服务裁判反馈补足。run.json 的评分字段未报告 distance_m 和 retouch_dv_m_s，均记为null。手指接触细节分辨率有限，持握结合随手运动、初始持饼布局和本局 holding_right 读数判断；无法仅凭画面区分最终脱手中的滑移与主动张手各自的贡献，因此不另标抓取失效根因。 释放判据使用 agent 模型预测，缺少原生释放速度、方向及改动门槛后的对照验证，不能确认参数调试子类，也不能将 agent 自述的握力限制当作已证实的接触设计根因。"
+ },
+ "summary_zh": "第6次实时试验中，机器人在圈内右手持饼、后摆并投出，铁饼在视频约4.24 s离手后飞向扇形侧外，裁判以落区越界判负；两个节点均未完成，后圈离圈未启动。摆臂与释放控制未实现目标方向，但模型预测与实际脱手的关系、调整门槛的效果均未验证，根因仅标控制，因具体原因仍有推断而将置信度降为7。无状态日志，精确首次落点和复触冲量未报告，最终落区判断结合本局裁判反馈。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 2,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "fix",
+    "conf": 6,
+    "summary_zh": "节点值、断点、时钟、现象均与视频和 log step 184 一致，未发现错误。但根因“参数调试”只靠 agent 自己的运动学预测，且按该预测即使在瞄准窗口内释放，速度也仅约 7 m/s，远低于达标所需，故“只要门槛合适就能成功”的因果没有证据，参数调试（须已证实参数不当）未证实。",
+    "n_issues": 1,
+    "n_major": 0
+   },
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 2,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "节点（均未达成）、断点（落区步骤，操作中）、现象（位姿不准）与降为控制的根因均与视频及 log step 184 裁判反馈一致，第1轮意见已落实，未发现新错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:17:54Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-discus-i00-standard--4gapg9s/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-discus-i00-standard--4gapg9s/04_review_r1"
+   },
+   {
+    "stage": "reconcile",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-discus-i00-standard--4gapg9s/05_reconcile_r1"
+   },
+   {
+    "stage": "traj_review",
+    "round": 2,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-discus-i00-standard--4gapg9s/04_review_r2"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "2d62123b04dc869e200a12f77be58d95b7a95fd26b804ba9459bb1124da0b647",
+   "video_sha256": "20483243b68a3dd0d0c325ebbc481408d61a627132a24c6f2de875c28df9e3b7",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/discus/trajectories/embodied-olympics-fun-discus-i00-standard--cttuyqf.json

@@ -0,0 +1,210 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-discus-i00-standard--cttuyqf",
+ "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": 88,
+   "sim_steps": 405,
+   "calls": 45,
+   "rows": 405,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 6220.2,
+   "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": 4.772,
+   "peak_processes": 13,
+   "peak_disk_gb": 4.643,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-discus-i00-standard--cttuyqf/replay.mp4",
+  "duration_s": 0.92,
+  "fps": 25.0,
+  "frames": 23,
+  "sha256": "7b53d97630abfcec815fc1ddb43cc3a0588e7c26ec02f981b0a65a1463dd061e"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "discus_landed_qualified": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.44,
+      "log_step": 133,
+      "text_zh": "观察：已检查本次视频全部23帧（视频0.00–0.88 s）；机器人在圈内直立，右手附近的橙红色铁饼随右臂做小幅预摆，始终未见铁饼脱手飞向扇形或触地。"
+     },
+     {
+      "t": 0.88,
+      "log_step": 133,
+      "text_zh": "观察：视频0.80–0.88 s铁饼仍紧邻右手，机器人没有完成投出；log step 133明确重置至episode 6，实时服务在最后一段动作返回done=True、reason=failed，原生裁判结果为“the discus was touched by something other than the throwing hand”，stages_reached为0。这与本次短录像的未投出终态一致；此前其他episode的落地记录不计入本次进度。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "观察：视频0.00–0.88 s右手托持并以卷曲手指扣住橙红色铁饼边缘，随臂预摆；log step 133的实时持握读数hold始终为1.0，支持这是持握而非仅平掌承托。裁判另确认存在非投掷手部位触饼，但具体接触连杆在画面中无法辨明。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频0.00–0.88 s机器人的双脚在投掷圈内接地，腿和躯干支撑持饼预摆，全程保持直立。"
+     }
+    ]
+   },
+   "rear_exit_stood": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.88,
+      "log_step": 133,
+      "text_zh": "观察：视频0.00–0.88 s机器人始终位于灰色投掷圈内，没有向后跨出圈沿；视频0.88 s仍持饼站在圈内。log step 133的最终episode在持饼预摆时被裁判终止，未执行后退离圈；有效落地这一前置也未达到。"
+     }
+    ]
+   }
+  },
+  "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": 0.84,
+   "log_step": 133,
+   "where": "操作中",
+   "evidence_zh": "观察：视频约0.80–0.88 s机器人正在圈内持饼预摆，铁饼仍在右手附近，尚未进入飞行；log step 133的最终episode 6被原生裁判以“the discus was touched by something other than the throwing hand”终止，阶段数为0。视频0.84 s作为这次预摆失败的近似导航锚点，非精确接触时刻。"
+  },
+  "unmet": {
+   "discus_landed_qualified": false,
+   "rear_exit_stood": false
+  },
+  "phenomena": [
+   {
+    "cat": "约束",
+    "item": "违反约束",
+    "t": 0.84,
+    "log_step": 133,
+    "evidence_zh": "观察：log step 133中最终episode 6的实时服务输出原生裁判原因“the discus was touched by something other than the throwing hand”，即触发非右手触饼规则；视频0.80–0.88 s机器人仍在持饼预摆且直立，铁饼未飞出。这里没有足以证明碰撞阻挡、掉饼或失衡的画面，故按已确认的禁触规则违规标注。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "抓取和接触设计",
+    "alt_feasible": "未知",
+    "source_zh": "日志 / 视频",
+    "evidence_zh": "推断：log step 129的动作参数在持握阶段加入右腕偏航预摆rwy_w=0.9；log step 131、133只改肩肘预摆而沿用此腕部设置，log step 133最终episode仍在持握读数1.0时触发非投掷手触饼。结合视频0.44–0.88 s铁饼随右手转动、未脱手且机器人直立，失败指向预摆时铁饼与本体的接触几何和间隙设计未能维持合法接触，而非落点瞄准或平衡。agent在log step 134将其解释为铁饼后缘扫到腕部连杆；具体连杆及单一关节的因果尚未核实，替代预摆方案也未在本次最终episode中验证。"
+   }
+  ],
+  "gap_zh": "没有状态日志或原生接触对记录，录像仅为640×360场地斜视，放大后仍不能分辨触饼的具体非投掷手连杆及精确接触时刻。裁判对禁触事件的报告明确，但agent在log step 134提出的“腕偏航使铁饼边缘碰腕部连杆”仍是其诊断，不能据此判定具体腕部构件超出合法右手范围；根因只归到预摆接触几何。评分器未报告distance_m或retouch_dv_m_s，两个原生指标均留null。"
+ },
+ "summary_zh": "本次评分采用log step 133重置后的episode 6；视频0.00–0.88 s机器人仅在圈内持饼预摆，铁饼未投出，两个目标节点最终均未达到（0/2）。投掷步骤在视频约0.84 s附近因原生裁判确认的非右手触饼而终止，证据支持预摆接触几何设计失当。缺少接触对和近景，无法核实agent所称的具体腕部连杆及精确碰触时刻。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "视频仅23帧（0.92 s），机器人全程在圈内持饼预摆，铁饼未投出，两节点均未达成；断点、约束违反现象与log step 133（id）的裁判结果“touched by something other than the throwing hand”、stages_reached 0一致，根因已标明是推断，未发现需要修改的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:17:06Z",
+  "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-discus-i00-standard--cttuyqf/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-discus-i00-standard--cttuyqf/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "24965e2f0753a886a85df11e14569950b292aac63832b9d5448bb98ab6a765f4",
+   "video_sha256": "7b53d97630abfcec815fc1ddb43cc3a0588e7c26ec02f981b0a65a1463dd061e",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/discus/trajectories/embodied-olympics-fun-discus-i00-standard--kjvqvxd.json

@@ -0,0 +1,243 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-discus-i00-standard--kjvqvxd",
+ "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": 123,
+   "sim_steps": 1097,
+   "calls": 88,
+   "rows": 77,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 2334.0,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 1,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.0,
+   "peak_processes": 0.0,
+   "peak_disk_gb": 0.0,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-discus-i00-standard--kjvqvxd/replay.mp4",
+  "duration_s": 1.24,
+  "fps": 25.0,
+  "frames": 31,
+  "sha256": "4df9d68448d519123d8c5b52c85212576bb40feb414c6a42462cd4e722763389"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "discus_landed_qualified": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": 87,
+      "text_zh": "观察：视频 0.00 s 橙红色铁饼在机器人右手中；视频 0.00–0.76 s 机器人把持饼右臂向下、向后预摆。log step 87 明确重置为 episode 6，该次试验是评分所用的最后一次实时试验。"
+     },
+     {
+      "t": 0.84,
+      "log_step": 87,
+      "text_zh": "观察：逐帧检查视频 0.64–1.20 s，铁饼在视频约 0.84 s 已与右手分离，随后在机器人身旁下落，没有飞向远处扇形；视频 1.20 s 铁饼已降到圆圈后侧近地面。log step 87 的该次试验反馈在仿真 0.65 s 为 hold=1，在仿真 0.80 s 为 hold=0；控制脚本仍在预摆，尚未进入计划的挥臂及开手循环。"
+     },
+     {
+      "t": 1.2,
+      "log_step": 88,
+      "text_zh": "观察：log step 88 的机器人服务报告 episode=6、control_steps=123、done=True、outcome=\"the discus landed outside the sector\"、stages_reached=1/3。视频末段铁饼在机器人近旁落下，与该次试验的界外落地判负相符；不是有效落区达标。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "观察：视频 0.00 s 右手手指围住橙红色铁饼边缘，机器人以右手持饼预摆；随后铁饼从右手滑出。log step 87 的持握反馈先为1，后为0，支持最初确有持握。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–1.20 s 机器人双脚在灰色投掷圈内支撑身体，躯干随右臂预摆小幅转动，未见跌倒。"
+     }
+    ]
+   },
+   "rear_exit_stood": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 1.2,
+      "log_step": 88,
+      "text_zh": "观察：视频 0.00–1.20 s 机器人双脚始终在投掷圈内，没有向后半圈外迈出；末帧仍站在圈内，铁饼在身旁近地面。log step 88 的该次试验已因界外落地结束。"
+     },
+     {
+      "t": 1.2,
+      "log_step": 86,
+      "text_zh": "推断：有效落地前置节点未达成，后圈离圈站稳不能成立。log step 86 的控制脚本把后退离圈安排在挥臂及落地等待之后；log step 87 显示脚本在预摆丢饼后尝试返回初始姿态，服务已结束该试验，离圈分支未执行。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "distance_m",
+   "value": null,
+   "unit": "m",
+   "threshold": 15.0,
+   "direction": "higher_better"
+  },
+  {
+   "name": "retouch_dv_m_s",
+   "value": null,
+   "unit": "m/s",
+   "threshold": 0.5,
+   "direction": "lower_better"
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "discus_landed_qualified",
+   "t": 0.8,
+   "log_step": 87,
+   "where": "操作中",
+   "evidence_zh": "观察：视频约 0.80–0.88 s，机器人预摆右臂时橙红色铁饼脱离右手，并在视频 0.96–1.20 s 向圈后侧近地面下落；未完成计划的前向投掷。log step 87 显示预摆后的 holding_right 已丢失，脚本进入返回初始姿态分支；log step 88 的服务最终判定铁饼落在扇形外，此步未修复。"
+  },
+  "unmet": {
+   "discus_landed_qualified": false,
+   "rear_exit_stood": false
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没保持住",
+    "t": 0.8,
+    "log_step": 87,
+    "evidence_zh": "观察：视频 0.00 s 铁饼随右手一起运动，视频约 0.84 s 铁饼与右手出现分离，视频 0.96–1.20 s 右手留在较高处而铁饼持续下落。log step 87 的该次试验反馈从仿真 0.65 s 的 hold=1 变为仿真 0.80 s 的 hold=0；丢饼发生在预摆阶段，早于脚本计划的开手释放。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "抓取和接触设计",
+    "alt_feasible": "未知",
+    "evidence_zh": "推断：预摆时的手掌朝向与手指承载布局未能维持已有持握。视频 0.00–0.76 s 铁饼从近水平随右手转为明显倾斜，随后在视频约 0.84 s 滑离；log step 86 的脚本在预摆阶段保持握紧命令，仅在后续挥臂循环才计划开手，而 log step 87 在预摆结束前已报告 hold=0。log step 85 的局部模型检查仅以掌面朝上分量判断姿态，log step 89 中 agent 自述该模型预测可持握但实际丢饼；该模型的具体倾角和承载裕量没有独立状态记录验证，不能据此断言手指力不足或某个数值参数错误。",
+    "source_zh": "日志 / 视频",
+    "note_zh": "该预摆姿态下的可行替代握法或承载布局未在本次试验中验证；成功参考展示另一条成功投掷轨迹，但未提供其逐指接触及受力。"
+   }
+  ],
+  "gap_zh": "无状态日志及逐指接触、接触力记录，不能精确定位最后一次持握或区分具体滑移受力机制；脱手视频锚点约有0.1秒不确定性。评分器未报告 distance_m 与 retouch_dv_m_s，故两项原生指标为 null；界外落地结论由该次试验的服务裁判结果和近处下落画面共同支持，不从画面估算距离。log step 87 的 EpisodeOver 异常发生在服务已经判负之后，不作为导致丢饼的根因。"
+ },
+ "summary_zh": "评分所用的是最后重置后的第6次试验：机器人右手持饼预摆时丢失持握，铁饼在身旁落下，服务判为扇形外落地，两个节点均未达成（0/2）。断点为“铁饼落区达标”的操作中，现象为没保持住，预摆姿态下的抓取和接触设计未能保持铁饼；机器人未开始后圈离圈。已逐帧检查完整评分视频并对照成功参考；没有状态及逐指受力日志，精确脱手时刻和具体力学机制仍有限制。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 2,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "fix",
+    "conf": 7,
+    "summary_zh": "节点未达成、断点节点、没保持住现象均成立；但脱手时刻错：逐帧显示铁饼在视频约0.16–0.32 s已离开右手并下落，落至地面约0.5–0.7 s，标签的0.84 s（及“0.96–1.20 s 才下落”）晚了约0.5 s。",
+    "n_issues": 1,
+    "n_major": 1
+   },
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 2,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "两节点未达成、断点在 discus_landed_qualified 操作中、现象没保持住均有画面与服务读数支持；第1轮提出的脱手时刻问题经逐帧复核不成立（视频约0.8 s前铁饼仍紧贴右手/大腿旁，0.88 s起明显分离），标签的0.8 s锚点合理。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:16:09Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-discus-i00-standard--kjvqvxd/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-discus-i00-standard--kjvqvxd/04_review_r1"
+   },
+   {
+    "stage": "reconcile",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-discus-i00-standard--kjvqvxd/05_reconcile_r1"
+   },
+   {
+    "stage": "traj_review",
+    "round": 2,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-discus-i00-standard--kjvqvxd/04_review_r2"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "2c09b109ada426aa39ea68ef52eb97ddb279954d7f6adf90ea8e4d10386d16d1",
+   "video_sha256": "4df9d68448d519123d8c5b52c85212576bb40feb414c6a42462cd4e722763389",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/discus/trajectories/embodied-olympics-fun-discus-i00-standard--nhu8nbc.json

@@ -0,0 +1,211 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-discus-i00-standard--nhu8nbc",
+ "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": 116,
+   "sim_steps": 644,
+   "calls": 17,
+   "rows": 644,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 3492.4,
+   "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": 1.114,
+   "peak_processes": 12,
+   "peak_disk_gb": 1.929,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-discus-i00-standard--nhu8nbc/replay.mp4",
+  "duration_s": 1.2,
+  "fps": 25.0,
+  "frames": 30,
+  "sha256": "84147243445e6f19867f3dd7ecd75e4ff5ea48c671a55fe2d3e216fa24a9d1c7"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "discus_landed_qualified": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–0.64 s，机器人右手持橙红色铁饼随右臂后摆，双脚在投掷圈内支撑；视频 0.72–0.80 s 铁饼与右手分离，随后向画面左下方圈旁下落，没有向白线之间的远处扇形飞行。"
+     },
+     {
+      "t": 1.16,
+      "log_step": 148,
+      "text_zh": "观察：视频 1.00–1.16 s，铁饼在圈旁绿色地面上方继续下降；log step 148 的最后一次 reset 后，服务在第二段动作返回 done=True、reason=failed、outcome=\"the discus landed outside the sector\"。log step 150 服务状态为 episode=6、control_steps=116、resets_left=0，与评分字段 n_actions=116 及此短视频一致，故只标最后一次实时试验，界外首次落地未达标。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "观察：视频 0.00–0.64 s，右侧五指手持饼随臂摆动，随后铁饼从手中脱离；log step 148 最后试验的第一段结束时 holding_right=0，log step 150 对该段本体传感记录的检查显示此前有持握。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–1.16 s，机器人双脚在圈内承重，腿和躯干支撑右臂摆动与身体转向。"
+     }
+    ]
+   },
+   "rear_exit_stood": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 1.16,
+      "log_step": 148,
+      "text_zh": "观察：已检查视频 0.00–1.16 s 的全部帧，机器人始终在灰色投掷圈内，末帧仍在摆臂转身，未向后侧绿色地台迈出；对照成功参照视频 5.00–6.50 s 可见的后退离圈过程，本次没有该变化。log step 148 的服务已因铁饼落在扇形外终止，脚本在执行后退段前返回；有效落地前置和后圈离圈站稳均未达到。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "distance_m",
+   "value": null,
+   "unit": "m",
+   "threshold": 15.0,
+   "direction": "higher_better"
+  },
+  {
+   "name": "retouch_dv_m_s",
+   "value": null,
+   "unit": "m/s",
+   "threshold": 0.5,
+   "direction": "lower_better"
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "discus_landed_qualified",
+   "t": 0.76,
+   "log_step": 148,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 0.64 s 铁饼仍紧邻右手，视频 0.72–0.80 s 在摆臂过程中从右手下方分离并向圈旁下落，未恢复持饼或投出有效远投；log step 148 服务最终判定 \"the discus landed outside the sector\"。失败发生在铁饼落区达标步骤的试摆持饼过程中，导航锚点取视频 0.76 s。"
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没保持住",
+    "t": 0.76,
+    "log_step": 148,
+    "evidence_zh": "观察：视频 0.00–0.64 s 铁饼随机器人右手运动，视频 0.72–0.80 s 铁饼从右手下方脱离，视频 0.84–1.16 s 继续向圈旁地面下降。log step 148 将第一段动作称为用于校准的 swing A，期望留饼供第二段投掷，却在该段结束读到 holding_right=0；log step 150 对最后试验的记录检查显示第一段有持握、第二段全部无持握，支持先握住后丢失。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "t": 0.76,
+    "log_step": 148,
+    "source_zh": "日志",
+    "evidence_zh": "推断：持饼与腕臂摆动的协调未维持住接触。log step 142 的控制库设计是在预摆期间固定腕姿、摆动起始快速转腕形成承载姿态；log step 148 最后试验仍调用该库，以 30% 速率执行校准摆动，第一段结束已无持握，脚本随后执行第二段空手投掷。视频 0.72–0.80 s 的离手下落与这一控制失效相符；不能仅据 agent 在 log step 151、155 的事后力学估算细分为某个参数或握持几何错误。"
+   }
+  ],
+  "gap_zh": "本次没有状态日志，单一斜视画面的手部像素不足以辨清逐指接触、滑脱受力和精确脱手时刻；控制根因保留在上层，不能独立验证 agent 在 log step 151、155 所称预摆加速度超出承载能力的具体解释。run.json 的评分字段未报告 distance_m 或 retouch_dv_m_s，均保留 null；界外落地由最后实时试验的服务裁判结果及圈旁下落画面共同确认。",
+  "unmet": {
+   "discus_landed_qualified": false,
+   "rear_exit_stood": false
+  }
+ },
+ "summary_zh": "最后一次实时试验中，机器人右手原先持饼，但在视频约 0.72–0.80 s 的试摆过程中失去持握，铁饼落向圈旁，服务裁判判定落在扇形外，铁饼落区达标与后圈离圈站稳均未完成（0/2）。断点为铁饼落区达标，现象为没保持住，根因为持饼与腕臂摆动协调的控制失效。没有状态日志及逐指接触力记录，无法核实 agent 的具体加速度解释，两个原生指标也未报告。",
+ "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 1.2 s 与最后一次实时试验（control_steps=116）对应；0.0–0.6 s 铁饼在右手中，0.72–0.80 s 起脱手下落到圈旁，两节点均未达成（0/2）；断点、没保持住、控制根因均有 proprio 读数（hold-sum 22/31 行，末行 holding_right=0）和画面支持，未发现需改动之处。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:13:28Z",
+  "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-discus-i00-standard--nhu8nbc/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-discus-i00-standard--nhu8nbc/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "046affd68fe813bac40e9d4bc12efffb5ad5def08b355ff72b0aed5a2e83e3ce",
+   "video_sha256": "84147243445e6f19867f3dd7ecd75e4ff5ea48c671a55fe2d3e216fa24a9d1c7",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/discus/trajectories/embodied-olympics-fun-discus-i00-standard--pemlwm5.json

@@ -0,0 +1,201 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-discus-i00-standard--pemlwm5",
+ "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": 0,
+   "sim_steps": 0,
+   "calls": 2,
+   "rows": 0,
+   "resets": 0,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 3087.2,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 4,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 1.007,
+   "peak_processes": 7,
+   "peak_disk_gb": 0.36,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-discus-i00-standard--pemlwm5/replay.mp4",
+  "duration_s": 0.04,
+  "fps": 25.0,
+  "frames": 1,
+  "sha256": "5d9c1503eb2984af754c58117202641d12a476911fe2c89c8ad2ffe0fad0599d"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "discus_landed_qualified": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": 86,
+      "text_zh": "观察：本次视频仅有视频 0.00 s 一帧，机器人站在圈内，橙红色铁饼仍位于右手处，未见脱手、飞行或首次落地；log step 86 的实时机器人服务状态为 rows=0、control_steps=0、done=false，评分记录亦为 n_actions=0、sim_steps=0。"
+     },
+     {
+      "t": null,
+      "log_step": 104,
+      "text_zh": "观察：log step 104 只报告 agent 本地建立的 MuJoCo 投掷模型加载成功；这不是实时评分试验的投掷结果，不能用于认定落区达标。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分记录 n_actions=0、rows=0；会话被过滤器中断前未交付实时控制轨迹，初始右手持饼不作为已执行投掷的接触。"
+     }
+    ]
+   },
+   "rear_exit_stood": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": 86,
+      "text_zh": "观察：视频 0.00 s 的唯一帧中机器人双脚仍在灰色投掷圈内；log step 86 的实时服务状态为 control_steps=0，未发生后半圈离圈。"
+     },
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "推断：本次没有投出铁饼及有效落地，后圈离圈站稳的前置事件也未发生；圈内初始直立不满足该终态节点。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分记录 n_actions=0、sim_steps=0；无交付的实时离圈动作轨迹。"
+     }
+    ]
+   }
+  },
+  "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": "unknown",
+   "t": null,
+   "log_step": 105,
+   "where": "首步前",
+   "evidence_zh": "观察：log step 105 的末条消息为“This request was blocked by Gemini's filters”；此前 log step 104 仅成功加载 agent 的本地模型，log step 86 的实时服务仍为 control_steps=0、done=false，最终评分记录 n_actions=0、sim_steps=0。提供方过滤器在实时动作交付前中断会话，未出现可定位到某个任务步骤的物理失败。"
+  },
+  "unmet": {
+   "discus_landed_qualified": false,
+   "rear_exit_stood": false
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：log step 105 明确记录提供方过滤器拦截；本次只有视频 0.00 s 的初始帧（文件时长 0.04 s），没有后续动作或状态日志，distance_m 与 retouch_dv_m_s 均未报告。推断：这是评分结果固定前的外因中断（CASES.md FAIL-5），应重测并排除出 agent 失败分析；保留目录 success=0，但不标“没尝试”或推断控制、规划根因。agent 的本地 MuJoCo 调试不代表评分试验；中断所在的 log step 105 无对应视频时刻，故 breakpoint.t=null。"
+ },
+ "summary_zh": "本次评分视频只有视频 0.00 s 的圈���持饼初始帧，实时控制动作及仿真步数均为 0，铁饼落区达标和后圈离圈站稳均未完成，最终为 0/2。log step 105 记录 Gemini 过滤器拦截，发生在动作交付前，按外因中断处理：保留目录 success=0，断点为 unknown／首步前，失败现象与根因留空，不计为 agent 失败。缺少后续运动证据，日志中的本地模型试验及 oracle 的成功动作均不计入本次进度。",
+ "conf": 10,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 9,
+    "summary_zh": "评分视频仅一帧初始帧，n_actions=0；log step 105 为 Gemini 过滤器拦截，动作交付前被外因中断。标注（节点均未达成、断点 unknown/首步前、现象与根因为空并附 gap_zh）符合 FAIL-5，未发现需修改之处。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:12:09Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-discus-i00-standard--pemlwm5/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-discus-i00-standard--pemlwm5/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "e4f975400df3926400e42d0d4ad11cc4afb68d73dc0a25530187b6831e718345",
+   "video_sha256": "5d9c1503eb2984af754c58117202641d12a476911fe2c89c8ad2ffe0fad0599d",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/discus/trajectories/embodied-olympics-fun-discus-i00-standard--soz6cpv.json

@@ -0,0 +1,214 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-discus-i00-standard--soz6cpv",
+ "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": 1613,
+   "sim_steps": 3565,
+   "calls": 1162,
+   "rows": 2472,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 6757.0,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 1,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.0,
+   "peak_processes": 0.0,
+   "peak_disk_gb": 0.0,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-discus-i00-standard--soz6cpv/replay.mp4",
+  "duration_s": 16.16,
+  "fps": 25.0,
+  "frames": 404,
+  "sha256": "e8d5f3d00ccc980325c4b62df00e3eaf08c37fa6b8cade71a59c748c3c35e27c"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "discus_landed_qualified": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 8.8,
+      "log_step": 186,
+      "text_zh": "观察：视频 7.8–9.4 s，机器人右手持续携带橙红色铁饼转动到身体侧后方，双脚在圈内支撑，正在准备投掷。"
+     },
+     {
+      "t": 15.68,
+      "log_step": 242,
+      "text_zh": "推断：这是最后一次实时试验 episode 6（log step 163 开始，log step 254 结束），前五次试验不计入本视频。log step 221 给出控制周期 0.01 s，log step 242 的 control_steps=1568 对应视频约 15.68 s；已与视频 15.64–15.84 s 的摆臂、铁饼离手下落核对，末次 1613 控制步也与视频约 16.16 s 时长相符。"
+     },
+     {
+      "t": 16.12,
+      "log_step": 254,
+      "text_zh": "观察：视频 15.80–16.12 s，铁饼离开右手向近处绿色地台下落，末帧位于灰色投掷圈的画面左下侧，没有朝前方白线扇形远投。log step 254 中实时机器人服务报告 episode=6、done=true、outcome=\"the discus landed outside the sector\"、stages_reached=1/3；有效首次落地节点未达到。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "右手手指与掌部持住橙红色铁饼，随右臂转动、摆臂；视频 7.8–9.4 s 铁饼随手移动，log step 240、242 的持握读数先为 1，随后变为 0，视频 15.80–16.12 s 可见铁饼脱离并下落。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "机器人双脚接触投掷圈内地面，腿与躯干支撑持饼、转体和摆臂；视频 15.00–16.12 s 仍直立站在圈内。"
+     }
+    ]
+   },
+   "rear_exit_stood": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 16.12,
+      "log_step": 254,
+      "text_zh": "观察：视频全程粗看及视频 15.00–16.12 s 逐帧检查，机器人始终在灰色投掷圈内，没有向后侧绿色地台迈出的离圈动作；末帧仍直立留在圈内。log step 254 的服务以扇形外落地终止试验，前置有效落地未达成，故后圈离圈站稳节点未达到。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "distance_m",
+   "value": null,
+   "unit": "m",
+   "threshold": 15.0,
+   "direction": "higher_better"
+  },
+  {
+   "name": "retouch_dv_m_s",
+   "value": null,
+   "unit": "m/s",
+   "threshold": 0.5,
+   "direction": "lower_better"
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "discus_landed_qualified",
+   "t": 15.7,
+   "log_step": 242,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 15.64–15.84 s，机器人右臂开始加速摆动，铁饼随即脱离并向下掉落，视频 16.12 s 已落到圈旁近处；机器人没有修复这次投掷。log step 242 的最后一次实时试验持握读数从 1 降至 0，log step 254 的服务随后报告铁饼落在扇形外，试验失败。"
+  },
+  "unmet": {
+   "discus_landed_qualified": false,
+   "rear_exit_stood": false
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没保持住",
+    "t": 15.7,
+    "log_step": 242,
+    "evidence_zh": "观察：视频 15.60 s 铁饼仍在右手处，视频 15.80–15.92 s 铁饼离手下落而手臂留在上方，视频 16.12 s 铁饼到达近处地台。log step 242 中实时持握读数在摆臂相对时间 0.16 s 仍为 1、0.18 s 变为 0，log step 245、249、252 的后续读数一直为 0；原有持握失去后未恢复。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "source_zh": "日志",
+    "evidence_zh": "推断：log step 231 的实际投掷脚本在摆臂相对时间达到参数 x[8]-0.02 后才命令张手；log step 238 选定 x[8]=0.37468773，故计划张手约在摆臂相对时间 0.355 s。log step 242 的实时读数却在摆臂相对时间 0.18 s 已失去持握，早于计划张手；随后调整手腕、重新闭手仍未恢复（log step 244–252），视频 15.80–16.12 s 也显示铁饼直接下落。这支持摆臂过程中未能维持持饼接触的控制问题，不能据此确定具体控制子类。"
+   }
+  ],
+  "gap_zh": "没有状态日志；视频只有场地斜视画面，手指接触与接触力无法精确辨认，无法进一步区分握持几何、负载参数或运动跟踪的作用。首次落点在扇形外由末次实时服务判定与画面共同支持，但 reward 未报告 distance_m 或 retouch_dv_m_s，故两项原生指标填 null。"
+ },
+ "summary_zh": "本次评分对应最后一次实时试验 episode 6：机器人持饼转体并摆臂，但铁饼在计划张手前失去持握，视频约 15.8–16.12 s 掉至圈旁；实时服务判定首次落地在扇形外。铁饼落区达标与后圈离圈站稳均未达到，最终进度为 0/2，断点是投掷中的持握丢失，根因仅归入控制。没有状态日志及接触力记录，无法进一步细分控制原因或量出首次落点距离。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "标注与视频和日志一致：末次实时试验（episode 6）中铁饼在摆臂中提前脱手并掉在圈旁，两个节点未达到，断点、「没保持住」和控制根因均有证据支持。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:03:32Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-discus-i00-standard--soz6cpv/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-discus-i00-standard--soz6cpv/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "dc3d7670cc96936370babe035e5e11a3e5a86f5c8380500daa1f81281dab713e",
+   "video_sha256": "e8d5f3d00ccc980325c4b62df00e3eaf08c37fa6b8cade71a59c748c3c35e27c",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/discus/trajectories/embodied-olympics-fun-discus-i00-standard--vfwgwja.json

@@ -0,0 +1,209 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-discus-i00-standard--vfwgwja",
+ "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": 267,
+   "sim_steps": 1708,
+   "calls": 16,
+   "rows": 1658,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 580.7,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.188,
+   "peak_processes": 6,
+   "peak_disk_gb": 0.283,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-discus-i00-standard--vfwgwja/replay.mp4",
+  "duration_s": 2.68,
+  "fps": 25.0,
+  "frames": 67,
+  "sha256": "da140beca494a0c264d3f3104dbd86fd32466bb7c028a011040025c4a1280b62"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "discus_landed_qualified": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": 114,
+      "text_zh": "观察：log step 114 将服务重置至第6次试验，log step 115 从该试验的 control_steps=0 开始执行；视频0.00–1.84 s 中机器人在圈内持橙红色铁饼转腰蓄力。此前各次试验不是本段评分录像。"
+     },
+     {
+      "t": 2.08,
+      "log_step": 115,
+      "text_zh": "观察：视频2.08 s 铁饼已与机器人的右手分离，向扇形方向飞行；视频2.24–2.64 s 机器人持续向后倾倒，视频2.64 s 铁饼仍高于地面，与其地面阴影分离，尚未首次触地。log step 115 的机器人服务结果为 outcome=fell、stages_reached=1 of 3，未完成落地阶段。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "推断：右手手掌与弯曲手指持住橙红色铁饼，随转腰和摆臂携带，随后脱手。依据视频0.00–1.84 s 铁饼随右手移动、视频2.08 s 已分离，以及log step 10 的固定初态说明（掌托、指尖扣住饼缘）；不将左臂摆动当作左手触饼。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频0.00–2.08 s 机器人双脚在投掷圈内接地，腿和躯干支撑持饼、转腰及投掷；视频2.24–2.64 s 支撑失稳，机器人向后倒下。"
+     }
+    ]
+   },
+   "rear_exit_stood": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.64,
+      "log_step": 115,
+      "text_zh": "观察：视频全段0.00–2.64 s 未出现有效落地后向后圈迈出并站稳的动作，末帧机器人已后仰倒下。log step 115 显示第6次试验在首个投掷动作块内即因 fell 终止；log step 113 的后退离圈动作块位于该终止返回之后，没有执行。成功参考视频3.60–6.80 s 可见机器人向圆圈后侧行进并保持直立，本段录像未出现对应动作。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "distance_m",
+   "value": null,
+   "unit": "m",
+   "threshold": 15.0,
+   "direction": "higher_better"
+  },
+  {
+   "name": "retouch_dv_m_s",
+   "value": null,
+   "unit": "m/s",
+   "threshold": 0.5,
+   "direction": "lower_better"
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "discus_landed_qualified",
+   "t": 2.3,
+   "log_step": 115,
+   "where": "操作中",
+   "evidence_zh": "观察：视频2.24–2.64 s，机器人在投出铁饼后的动作中向后倾倒，骨盆降低、躯干接近地面；铁饼仍在飞行，落地目标尚未达到。log step 115 的第6次实时试验在首个投掷动作块的267个控制步后报告 fell，仅到脱手阶段，未进入后退离圈动作块。推断：断点属于铁饼落区达标步骤，而非尚未开始的离圈步骤。"
+  },
+  "unmet": {
+   "discus_landed_qualified": false,
+   "rear_exit_stood": false
+  },
+  "phenomena": [
+   {
+    "cat": "全身",
+    "item": "平衡",
+    "t": 2.3,
+    "log_step": 115,
+    "evidence_zh": "观察：视频2.08 s 机器人投出铁饼后已后仰；视频2.24–2.64 s 后仰角继续扩大、身体降低并向圈后侧倒下，未恢复站立。log step 115 的机器人服务明确报告 outcome=fell 并结束本次试验。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "source_zh": "日志 / 视频",
+    "evidence_zh": "推断：投掷及其后的全身稳定控制未能维持支撑。log step 113 的实际执行脚本在stand模式下以固定关键姿态生成转腰摆臂，并用左臂反向摆动补偿；log step 115 执行此版本后报告fell。视频2.08–2.64 s 中铁饼已投出而机器人持续向后倾倒，说明该次实际运动未保持全身稳定；不足以进一步确定是轨迹设计还是底层反馈问题。"
+   }
+  ],
+  "gap_zh": "没有状态日志，无法核验接触对、投掷期间的力矩及支撑裕量，控制根因不细分。录像在首次落地前终止，评分字段未报告distance_m或retouch_dv_m_s，不能外推最终落点、距离或擦碰冲量；log step 120 对发射速度及反作用力的解释仅为agent自述，不作为实测结论。"
+ },
+ "summary_zh": "评分轨迹是log step 114重置后的第6次试验：机器人在圈内右手持饼蓄力，并于视频约2.08 s前投出；铁饼尚未首次落地，两个节点均未完成（0/2）。机器人在视频约2.3 s持续向后失衡，服务以fell终止试验，断在铁饼落区达标的操作中，根因归为控制。无状态日志和原生落点指标，无法细分失稳机制或判断终止后的飞行结果。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "标注与视频和日志一致：评分段为第6次试验，视频2.08 s铁饼脱手，之后机器人后倒，铁饼未落地；两节点均未达成，断点、平衡现象与控制根因（已声明无法细分）均有证据支持。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:10:05Z",
+  "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-discus-i00-standard--vfwgwja/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-discus-i00-standard--vfwgwja/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "f02ef0d5457248308f2a0e18a6c5297b09126a4bb08dea39d9e4fefae8c5ffac",
+   "video_sha256": "da140beca494a0c264d3f3104dbd86fd32466bb7c028a011040025c4a1280b62",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/discus/trajectories/embodied-olympics-fun-discus-i00-standard--zfzysxx.json

@@ -0,0 +1,219 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-discus-i00-standard--zfzysxx",
+ "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": 290,
+   "sim_steps": 1427,
+   "calls": 45,
+   "rows": 1218,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 625.1,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 3,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.329,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.835,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-discus-i00-standard--zfzysxx/replay.mp4",
+  "duration_s": 2.92,
+  "fps": 25.0,
+  "frames": 73,
+  "sha256": "b3ee806306a28174285c87a75371616906f7952bb9163404a6a87912f6948499"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 2,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "discus_landed_qualified": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–2.08 s，橙红色铁饼随机器人右手移动；视频 2.12–2.16 s，铁饼与右手分离并向圈前方飞出。"
+     },
+     {
+      "t": 2.88,
+      "log_step": null,
+      "text_zh": "观察：视频 2.60–2.88 s，铁饼下降到灰色投掷圈紧邻的绿色地台，视频 2.88 s 铁饼与其地面阴影汇合；落点在通向远处的两条白色扇形线所围区域之外，未飞入远处落区。"
+     },
+     {
+      "t": 2.88,
+      "log_step": 69,
+      "text_zh": "推断：视频 2.88 s 所见圈旁首次落地不可能达到距圈内沿至少 15 米的条件，因此本节点未达到；agent 在 log step 64、69 也自述裁判判为“the discus landed outside the sector”，该自述仅作辅助，未当作独立裁判读数。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "手指/夹爪",
+      "how": "抓握",
+      "note_zh": "推断：视频 0.00–2.08 s 铁饼贴随右手运动，结合任务初态说明手指卷住铁饼边缘，右手持饼摆臂；视频 2.12–2.16 s 松开后铁饼独立飞行。画面分辨率不足以分辨各手指接触点。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–2.88 s 机器人双脚在投掷圈内接地，腿和躯干支撑转体、摆臂与收臂，未跌倒。"
+     }
+    ]
+   },
+   "rear_exit_stood": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 2.88,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–2.88 s 机器人始终在灰色圈内，投掷后原地收臂，未向后半圈圈沿或圈外迈出；视频末尾仍直立于圈内。"
+     },
+     {
+      "t": 2.88,
+      "log_step": null,
+      "text_zh": "推断：视频 2.88 s 首次落地不合格，且全过程未出现离圈动作，未满足有效落地后的后圈离圈及站稳验收。成功参考视频约 4.8–6.6 s 可见机器人向后跨出灰色圈的动作，本次视频没有这一变化。"
+     }
+    ]
+   }
+  },
+  "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": 2.88,
+   "log_step": null,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 2.12–2.16 s 机器人右手将铁饼释放，视频 2.60–2.88 s 铁饼下降并在投掷圈旁绿色地台首次触地，没有进入远处扇形落区；机器人仍在圈内，投掷达标这一步未完成。"
+  },
+  "unmet": {
+   "discus_landed_qualified": false,
+   "rear_exit_stood": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 2.88,
+    "log_step": null,
+    "evidence_zh": "观察：视频 2.12–2.88 s 铁饼脱手后仅作短距离飞行，首次落在圈旁绿色地台，偏离两条白色边线之间的目标落区；这是已实施投掷的落点偏差。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "evidence_zh": "推断：log step 55 的最后一次重置后，agent 脚本依次下发转腰蓄力、右肩预加速、摆臂并开手命令，log step 56–57 下发收臂命令；视频 1.00–2.16 s 可见对应的转体、右手摆动与脱手，但视频 2.88 s 铁饼仅落到圈旁。脚本实施了任务所需的投掷方向与阶段，实际投掷运动未产生到达远处落区的飞行，支持归于控制；缺少释放速度和关节跟踪读数，不能确定具体控制子类。",
+    "source_zh": "日志 / 视频"
+   }
+  ],
+  "gap_zh": "没有状态日志；inputs/run.json 中最后一次试验的服务调用输出文本为空，缺少独立的裁判阶段、distance_m、retouch_dv_m_s、释放速度及接触力记录。斜视视频足以判定圈旁落地与未离圈，但无法精确量出首次落点距离、各手指接触及释放动力学，因此原生指标填 null，控制根因不细分。"
+ },
+ "summary_zh": "评分采用最后一次实时试验（log step 55 重置后），此前试验不计入本次进度；机器人完成右手摆臂脱手，但视频 2.88 s 铁饼首次落在圈旁绿色地台，铁饼落区达标和后圈离圈站稳均未完成，最终为 0/2。断点是铁饼落区达标，表现为近距离偏区落地，结合投掷命令与画面推断为控制问题。缺少最终试验的裁判数值、释放速度及接触日志，不能进一步确定控制子类。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "视频 2.88 s 铁饼落在圈旁绿色地台、未进扇形，两节点均未达成，断点、现象（位姿不准）与停在“控制”的根因均有帧和 log 支持；未发现需要修改之处。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:11:18Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-discus-i00-standard--zfzysxx/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-discus-i00-standard--zfzysxx/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "e4abc3e62b7181a179e6a8263d8b3d106e4b9c052ea4b7ff669fa17f2cd9a625",
+   "video_sha256": "b3ee806306a28174285c87a75371616906f7952bb9163404a6a87912f6948499",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/free_throw/trajectories/embodied-olympics-fun-free-throw-i00-privileged--2tvslbf.json

@@ -0,0 +1,173 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-free-throw-i00-privileged--2tvslbf",
+ "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,
+   "n_actions": 113,
+   "sim_steps": 113,
+   "horizon": 1200,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 2,
+   "stages_total": 2,
+   "fell": 0,
+   "replay_s": 3.25,
+   "bad_trajectory": 0,
+   "metric_ball_height_m": 2.772,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 102,
+   "agent_resets": 17278,
+   "agent_replays": 24,
+   "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": 1.903,
+   "peak_processes": 18,
+   "peak_disk_gb": 0.851,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-free-throw-i00-privileged--2tvslbf/replay.mp4",
+  "duration_s": 1.16,
+  "fps": 25.0,
+  "frames": 29,
+  "sha256": "60ac47056617b4a739b27b1741b7e4bea9ab9cf109687875714524abfd68e45c"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "free_throw_scored": {
+    "method": "visual",
+    "intervals": [
+     [
+      1.12,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–0.16 s，机器人双脚着地站立，双手在篮球两侧共同抬球并向前上方送出；视频 0.20 s 球已与双手明显分离。"
+     },
+     {
+      "t": 1.12,
+      "log_step": null,
+      "text_zh": "观察：已逐帧查看视频 0.00–1.12 s 的全部 29 帧。篮球离手后沿弧线飞向篮圈，视频 0.80–1.00 s 从圈上方向下接近，视频 1.04–1.12 s 穿圈并降至篮网区域；末帧机器人仍站立。将达到圈下验收高度的时间标为视频约 1.12 s，受帧率和远景分辨率限制。"
+     },
+     {
+      "t": 1.12,
+      "log_step": 428,
+      "text_zh": "观察：log step 428 对最终 113 行交付文件重新生成并作两次独立重放，均返回 success=true、failed=null、阶段 release 与 through the ring、相同 state_sha；重放事件记载仿真 0.17 s 离手及仿真 1.121 s 穿圈成功。inputs/run.json 的评分器字段 success=1、replay_success=1、deterministic=1、stages_reached=2，与视频末段穿圈一致；节点时刻取自视频帧，未把 agent 时间换算为视频时间。"
+     },
+     {
+      "t": 0.12,
+      "log_step": 381,
+      "text_zh": "推断：视频 0.00–0.16 s 可见双手从球两侧共同夹持并送球；开局张手夹持由 task_text 的起始条件补充，log step 381 的最终生成脚本在送球时给左右手闭合指令 0.3673 和 0.5。远景帧不能逐指确认后续是否形成包络抓握，因此只记录可确认的双手夹持，不将闭合指令本身当作抓握成立的证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 0.00–0.16 s 两只灵巧手从篮球两侧共同夹持、抬高并向前上方送球，视频 0.20 s 篮球已自由飞行。候选：夹持；开局张手相向夹紧，单手包络抓握未能从远景核实，托/舀也不能完整表达两侧相向夹持。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–1.12 s 机器人双脚接触球场地面并支撑躯干，配合双臂送球，出手后保持站立。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "ball_height_m",
+   "value": 2.772,
+   "unit": "m",
+   "threshold": null,
+   "direction": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人站立将开局持有的篮球投向篮筐，视频约 1.12 s 篮球从上方穿圈并下降，罚球命中节点完成，终态为 1/1。最终交付文件的两次重放均成功且终态哈希相同（log step 428），与评分器成功判定及视频一致。无状态日志，远景录像不能独立核验逐指接触、毫秒级离手和精确穿圈交点；完成判定结合评分器结果，时间取视频近似值。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "罚球命中节点的时间（视频约 1.12 s）、final=true、接触（双臂协作/其他 + 腿/躯干支撑）与成功运行的空 failure 均与视频帧和评分器字段一致，未发现需修改之处。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:08:01Z",
+  "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-free-throw-i00-privileged--2tvslbf/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-free-throw-i00-privileged--2tvslbf/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "99117ee99cad79623e6712148f18767e7ce12f6f70d51ff9b9e66c30d1c4de95",
+   "video_sha256": "60ac47056617b4a739b27b1741b7e4bea9ab9cf109687875714524abfd68e45c",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/free_throw/trajectories/embodied-olympics-fun-free-throw-i00-privileged--5oaws6e.json

@@ -0,0 +1,168 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-free-throw-i00-privileged--5oaws6e",
+ "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,
+   "n_actions": 179,
+   "sim_steps": 179,
+   "horizon": 1200,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 2,
+   "stages_total": 2,
+   "fell": 0,
+   "replay_s": 4.23,
+   "bad_trajectory": 0,
+   "metric_ball_height_m": 2.754,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 152,
+   "agent_resets": 7827,
+   "agent_replays": 21,
+   "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": 3.205,
+   "peak_processes": 18,
+   "peak_disk_gb": 0.24,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-free-throw-i00-privileged--5oaws6e/replay.mp4",
+  "duration_s": 1.8,
+  "fps": 25.0,
+  "frames": 45,
+  "sha256": "01e9aad5210c4440ace780fe091a1e42a99a45c058ae63ec137f2af23ffa7905"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "free_throw_scored": {
+    "method": "visual",
+    "intervals": [
+     [
+      1.79,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.4,
+      "log_step": 142,
+      "text_zh": "观察：视频 0.00–0.32 s，机器人双手在胸前篮球两侧相向接触，手指张开；视频 0.36–0.40 s 双臂共同向前上方送球，视频 0.44 s 球与手之间已出现间隙并向篮筐飞行。机器人双脚在地面支撑，直至视频 1.76 s 均保持站立。log step 142 的交付说明同样记载张手、双掌相向夹持后送球。"
+     },
+     {
+      "t": 1.76,
+      "log_step": null,
+      "text_zh": "观察：已逐帧检查视频 0.00–1.76 s；篮球离手后上升再下降，视频 1.52–1.68 s 从篮圈上方接近，视频 1.72 s 球在圈口上沿，视频 1.76 s 已进入圈口及网部并受到遮挡。成功参考在其视频 1.00–1.12 s 呈现相同的下降入圈过程；参考仅用于辨认画面。"
+     },
+     {
+      "t": 1.79,
+      "log_step": 158,
+      "text_zh": "推断：罚球命中在视频末端约 1.79 s 完成并锁存。log step 158 对最终交付文件的两次新进程重放均显示 success=True、179 个控制步、release 和 through the ring 两阶段完成且终态哈希一致；本次评分器 reward 同样为 success=1、replay_success=1、stages_reached=2/2。log step 141 的脚本按每控制步 0.01 s 计时，离手为仿真 0.43 s，与视频 0.40–0.44 s 的可见离手核对一致；据此把最终控制步 179×0.01=仿真 1.79 s 对应到约视频 1.79 s（视频总长 1.80 s）。最后渲染帧为视频 1.76 s，精确下方验收平面的通过由评分器成功结果补足，不从 agent 工作时间换算。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 0.00–0.40 s，机器人两只张开的灵巧手在篮球两侧相向夹持并共同送球；候选名称“夹持”。两手未合指包住球，不适用抓握；单纯托/舀不能完整描述相向夹紧。视频 0.44 s 后球离手自由飞行。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–1.76 s，机器人双脚接触罚球区地面，支撑站立躯干并配合双臂投球。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "ball_height_m",
+   "value": 2.754,
+   "unit": "m",
+   "threshold": null,
+   "direction": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人双手相向夹持开局篮球并向前上方送出，约视频 0.44 s 可见离手，末段篮球从上方进入篮圈，约视频 1.79 s 完成罚球命中，最终进度为 1/1。评分器成功、两阶段完成及最终交付文件的确定性重放结果与画面一致，没有失败断点。没有状态日志，25 fps 录像末帧在视频 1.76 s 且球受篮圈及网遮挡，精确下方平面验收和毫秒级接触不能直接逐帧核验，完成值结合评分器判定，时间结合已核对的控制步映射估计。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "标签���画面和评分器一致：球离手后飞行并在约视频 1.7 s 进入篮圈，评分器 success=1、阶段 2/2，节点 reached、无失败断点；接触（双臂协作/其他 + 腿/躯干支撑）合理，时间换算有控制步核对。",
+    "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": "2026-10-31T00:02:40Z",
+  "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-free-throw-i00-privileged--5oaws6e/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-free-throw-i00-privileged--5oaws6e/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "13b56e903391668b9c33b30eefe9778be6b3152cbf2999911ce90f505cd122ac",
+   "video_sha256": "01e9aad5210c4440ace780fe091a1e42a99a45c058ae63ec137f2af23ffa7905",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/free_throw/trajectories/embodied-olympics-fun-free-throw-i00-privileged--6uq7st2.json

@@ -0,0 +1,168 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-free-throw-i00-privileged--6uq7st2",
+ "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,
+   "n_actions": 180,
+   "sim_steps": 180,
+   "horizon": 1200,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 2,
+   "stages_total": 2,
+   "fell": 0,
+   "replay_s": 3.65,
+   "bad_trajectory": 0,
+   "metric_ball_height_m": 2.755,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 92,
+   "agent_resets": 105909,
+   "agent_replays": 112,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 3.518,
+   "peak_processes": 30,
+   "peak_disk_gb": 0.812,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-free-throw-i00-privileged--6uq7st2/replay.mp4",
+  "duration_s": 1.84,
+  "fps": 25.0,
+  "frames": 46,
+  "sha256": "638afecd39efcc63504726fc70345e03c46740750a6c0a2988699f771299cc34"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "free_throw_scored": {
+    "method": "visual",
+    "intervals": [
+     [
+      1.8,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.12,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–0.12 s，机器人两只张开的手从篮球两侧共同夹持并向上送球；视频 0.16 s 球已与双手分离，双脚仍落地支撑。"
+     },
+     {
+      "t": 1.8,
+      "log_step": null,
+      "text_zh": "观察：逐帧查看完整录像；视频 1.60–1.72 s 橙球从篮板上方朝篮圈下降，视频 1.76 s 球到达橙色圈口，视频 1.80 s 球已在圈下篮网内。参考成功录像的末段也呈现球从上方入圈后在网内下落的可见形态；本次达成时间按本次末帧约定为视频 1.80 s，精度约一个视频帧。"
+     },
+     {
+      "t": 1.8,
+      "log_step": 219,
+      "text_zh": "观察：inputs/run.json 的评分器字段 success=1、replay_success=1、deterministic=1、stages_reached=stages_total=2、fell=0；log step 219 的最终封装脚本输出报告交付文件为 180 行、两次新进程重放成功、生成器逐数组完全一致，并记录仿真离手 0.156 s、仿真命中 1.793 s。节点时间取上述已检查的视频帧，不用 agent 工作时间换算。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 0.00–0.12 s 左右手从球两侧相向夹持并共同上送，未见任一手合拢包住篮球；候选“夹持送球”，抓握不适用，单纯托/舀不能完整描述两侧相向夹持；视频 0.16 s 后球离开双手自由飞行。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–1.80 s 机器人双脚接触罚球区地面，支撑躯干完成送球及随挥，录像中保持站立。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "ball_height_m",
+   "value": 2.755,
+   "unit": "m",
+   "threshold": null,
+   "direction": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "观察：机器人双手夹持上送篮球，球约在视频 0.16 s 离手，沿高弧线飞行后于视频 1.76–1.80 s 从上方穿过篮圈并降入网内，罚球命中节点最终完成 1/1；评分器 success=1、两个原生阶段均完成且重放确定。观察：没有状态日志，25 fps 场地录像不能独立核验毫秒级接触或精确穿圈交点，相关合法性由评分器成功结果佐证，节点时间取视频约 1.80 s。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "成功运行，单节点 free_throw_scored 在视频约 1.8 s 达成，帧上球入网，评分器 success=1，接触标注（双臂协作+其他、腿/躯干支撑）与视频一致，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": "2026-10-30T23:59:22Z",
+  "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-free-throw-i00-privileged--6uq7st2/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-free-throw-i00-privileged--6uq7st2/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "0f49ed7a4316b7e1cc2b7c94f8244e7d3c1fc721a42aeb86e0c9af7c530269f0",
+   "video_sha256": "638afecd39efcc63504726fc70345e03c46740750a6c0a2988699f771299cc34",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/free_throw/trajectories/embodied-olympics-fun-free-throw-i00-privileged--gtwhdfb.json

@@ -0,0 +1,178 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-free-throw-i00-privileged--gtwhdfb",
+ "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,
+   "n_actions": 307,
+   "sim_steps": 307,
+   "horizon": 1200,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 2,
+   "stages_total": 2,
+   "fell": 0,
+   "replay_s": 4.25,
+   "bad_trajectory": 0,
+   "metric_ball_height_m": 2.803,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 170,
+   "agent_resets": 8528,
+   "agent_replays": 2,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 4.492,
+   "peak_processes": 28,
+   "peak_disk_gb": 0.241,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-free-throw-i00-privileged--gtwhdfb/replay.mp4",
+  "duration_s": 3.08,
+  "fps": 25.0,
+  "frames": 77,
+  "sha256": "8e425eb5f53bc820ec2a6d09a94dd1607aa10a68d5bc4917cbf21240c1eed102"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "free_throw_scored": {
+    "method": "visual",
+    "intervals": [
+     [
+      3.07,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00 s，机器人站在罚球线后，橙色篮球位于两只张开的手之间；视频 0.64–1.52 s，机器人屈膝下蹲，篮球随双手下降，仍在两手之间。"
+     },
+     {
+      "t": 1.72,
+      "log_step": null,
+      "text_zh": "观察：视频 1.56–1.68 s，机器人伸腿并将双臂从下方向前上方摆起，两只手共同送出篮球；视频 1.72 s，球已与双手明显分离并继续向篮筐上方飞行。"
+     },
+     {
+      "t": 3.04,
+      "log_step": null,
+      "text_zh": "观察：逐帧检查视频 2.00–3.04 s，篮球先上升再下降，视频 2.88–3.00 s 从篮圈上方接近圈口，视频 3.04 s 球位于圈口及篮网顶部区域；机器人此时明显向前倾斜，录像在其完全倒地前结束。"
+     },
+     {
+      "t": 3.07,
+      "log_step": 56,
+      "text_zh": "推断：log step 56 是交付 trajectory.npz 的官方重放，记录仿真 1.652 s 离手、仿真 3.069 s through the ring 和 success，307 个控制步结束；log step 2 给出 control_dt=0.01 s，故终点映射为视频约 307×0.01=3.07 s，与视频总长 3.08 s 一致，并已用视频 1.64–1.72 s 的离手动作核对。将命中事件定位在视频约 3.07 s，精度受 25 fps 末帧采样限制。"
+     },
+     {
+      "t": 3.07,
+      "log_step": 91,
+      "text_zh": "观察：inputs/run.json 的评分器字段 success=1、replay_success=1、deterministic=1、stages_reached=2/2、fell=0；log step 91 显示两次交付文件重放均成功且终态哈希相同，log step 74 确认再生成的 307 行 actions 与此前交付相同，log step 96 确认最终再生成文件的哈希不变。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 0.00–1.64 s，机器人左右手分别位于篮球两侧，双臂共同夹持、降低并向上送球；可见手保持张开，没有单手合拢包住球。候选“夹持送球”：两手相向承载并送出，不符合单手包络的抓握，托/舀也不能完整描述两侧夹持；视频 1.72 s 后球与双手分离，自由飞向篮筐。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–1.60 s，机器人双脚接触场地地面，双腿屈膝下蹲后伸展，为双臂投篮提供支撑；出手后机器人身体上升再落回脚部支撑，视频 2.72–3.04 s 逐渐向前倾斜。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "ball_height_m",
+   "value": 2.803,
+   "unit": "m",
+   "threshold": null,
+   "direction": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "观察：机器人下蹲后以双手夹持送球，视频约 1.7 s 出手，篮球在视频末段从上方进入篮圈；评分器确认罚球命中，唯一节点完成，最终 1/1。推断：结合交付文件官方重放的控制步和离手动作核对，命中定位为视频约 3.07 s；末段前倾未被评分器判为命中前跌倒。证据缺口：没有状态日志，25 fps 固定远景不能独立核验毫秒级接触间断和精确穿圈交点，末帧之后的验收细节由官方重放及评分器成功字段支持。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "唯一节点 free_throw_scored 达成（评分器 success=1、stages 2/2），时间由重放事件 t=3.069 s 与 307 步映射得出，视频抽帧显示约 1.7 s 出手、末段球在篮筐上方，与标注一致；双臂协作/其他与腿支撑接触合理，无失败层。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T23:59:43Z",
+  "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-free-throw-i00-privileged--gtwhdfb/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-free-throw-i00-privileged--gtwhdfb/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "9ee6e278be5d928097bbe21a546107b1c5cc1ce36a03d6c67749a56bcac2cd5d",
+   "video_sha256": "8e425eb5f53bc820ec2a6d09a94dd1607aa10a68d5bc4917cbf21240c1eed102",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/free_throw/trajectories/embodied-olympics-fun-free-throw-i00-privileged--kpkajvd.json

@@ -0,0 +1,194 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-free-throw-i00-privileged--kpkajvd",
+ "taskKey": "embodied-olympics-fun/free_throw",
+ "dag_sha256": "c2d4b07a9ca0ff585b81cafb3eecf9f2484164969c066ba3cd5ef111abe016a5",
+ "outcome": {
+  "success": 0,
+  "reward": 0.5,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "n_actions": 135,
+   "sim_steps": 135,
+   "horizon": 1200,
+   "episode_over": 1,
+   "progress": 0.5,
+   "stages_reached": 1,
+   "stages_total": 2,
+   "fell": 0,
+   "replay_s": 2.53,
+   "bad_trajectory": 0,
+   "metric_ball_height_m": 0.097,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 203,
+   "agent_resets": 2545,
+   "agent_replays": 3,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.892,
+   "peak_processes": 10,
+   "peak_disk_gb": 0.264,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-free-throw-i00-privileged--kpkajvd/replay.mp4",
+  "duration_s": 1.36,
+  "fps": 25.0,
+  "frames": 34,
+  "sha256": "4da866449f81e746c799290319eefad2dad0fadbe78081d1092928bb97a59f34"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "free_throw_scored": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.24,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–0.20 s，机器人双手将篮球从身前向上送出；视频 0.24–0.28 s，篮球与双手分离并向篮筐方向飞行。"
+     },
+     {
+      "t": 0.64,
+      "log_step": null,
+      "text_zh": "观察：逐帧检查视频 0.28–1.32 s，篮球飞行弧线低于篮圈，随后下降到篮下附近地面；没有从上向下穿圈。成功参考视频 0.80–1.12 s 则可见篮球从篮圈上方降入圈内。"
+     },
+     {
+      "t": 1.32,
+      "log_step": 453,
+      "text_zh": "观察：视频 1.20–1.32 s，篮球接近并落到地面，机器人仍以双脚支撑；log step 453 对最终生成文件的重放报告 success=false、stages=[release]、failed=the ball touched the floor，与画面一致。原生离手阶段不满足本 DAG 的罚球命中节点。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 0.00–0.20 s，机器人两只张开的手在篮球两侧共同夹持、上送篮球，随后分开让球自由飞行。候选：夹持；未见任一手合拢包住篮球，抓握不适用，单纯托/舀不能完整描述两侧相向夹持。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–1.32 s，机器人双脚接触罚球区地面支撑躯干，腿部伸展配合双臂送球，末段躯干前倾但未见倒地。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "ball_height_m",
+   "value": 0.097,
+   "unit": "m",
+   "threshold": null,
+   "direction": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "free_throw_scored",
+   "t": 0.64,
+   "log_step": 453,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 0.24–0.68 s，机器人投出的篮球沿低于篮圈的弧线前进，继而下降；视频 1.20–1.32 s，球落到地面而未穿圈。log step 453 最终文件重放只记录 release 阶段并因篮球触地结束，故失败步骤为罚球命中。"
+  },
+  "unmet": {
+   "free_throw_scored": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 0.64,
+    "log_step": 453,
+    "evidence_zh": "观察：视频 0.28–1.32 s，篮球离开机器人双手后未达到篮圈高度便开始下降，最终落到篮下附近地面，没有从上方进入篮圈；log step 453 的最终重放也报告只有离手阶段、篮球触地失败。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "t": 0.24,
+    "log_step": 349,
+    "source_zh": "日志 / 视频",
+    "evidence_zh": "推断：log step 349 的最终生成脚本以预设关节目标、前馈力矩和固定动作步数协调腿臂伸展，再外展双臂放球；log step 453 重新生成并重放该文件仍因篮球触地失败，且视频 0.24–1.32 s 显示出手弧线过低。结合脚本明确朝篮筐投球的目标，支持送球控制未产生所需飞行轨迹；不能据此确定是接触几何、关节跟踪还是具体参数造成。"
+   }
+  ],
+  "gap_zh": "没有状态日志；固定远景录像不足以��辨逐指接触、接触力及精确离手速度，因而控制根因保留大类，未细分为抓取和接触设计、轨迹或逆解或参数调试。log step 453 的离手与触地事件带仿真时间，本标签的视频锚点均来自已检查画面，未将 agent 工作时间换算成视频时间。"
+ },
+ "summary_zh": "机器人从双手持球开始，双脚支撑并协调腿臂伸展投出篮球，但球的弧线低于篮圈，最终落地，罚球命中节点始终未达成（0/1）。最终交付文件的重放只达到原生离手阶段，失败定位于罚球命中操作中的位姿不准，根因支持控制大类。缺少逐帧状态及接触力数据，无法进一步确定出手控制的具体失效机制。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "标签与视频和 log 一致：球离手后弧线低于篮圈并落地，罚球命中节点未达成，断点、位姿不准、控制根因均符合 03 文档对投篮类的规则，未发现需改动之处。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:05:57Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-free-throw-i00-privileged--kpkajvd/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-free-throw-i00-privileged--kpkajvd/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "8174ea18d710fa5260f01979a58426176378a2d10ce2841596bbb64d6fa95ba6",
+   "video_sha256": "4da866449f81e746c799290319eefad2dad0fadbe78081d1092928bb97a59f34",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/free_throw/trajectories/embodied-olympics-fun-free-throw-i00-privileged--o3budzo.json

@@ -0,0 +1,178 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-free-throw-i00-privileged--o3budzo",
+ "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,
+   "n_actions": 137,
+   "sim_steps": 137,
+   "horizon": 1200,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 2,
+   "stages_total": 2,
+   "fell": 0,
+   "replay_s": 3.91,
+   "bad_trajectory": 0,
+   "metric_ball_height_m": 2.765,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 100,
+   "agent_resets": 5864,
+   "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": 3.005,
+   "peak_processes": 19,
+   "peak_disk_gb": 0.982,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-free-throw-i00-privileged--o3budzo/replay.mp4",
+  "duration_s": 1.4,
+  "fps": 25.0,
+  "frames": 35,
+  "sha256": "f1223491fcff8c354c7751e58d80fe5c604e4562a8d638c313c97db19a6b24e6"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "free_throw_scored": {
+    "method": "visual",
+    "intervals": [
+     [
+      1.36,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.12,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–0.12 s，机器人两只张开的手在篮球两侧夹持并共同向前上方送球；未见单手合指包住篮球，双脚接触地面支撑身体。"
+     },
+     {
+      "t": 0.2,
+      "log_step": null,
+      "text_zh": "观察：视频 0.16 s 篮球位于上举双手前上方，视频 0.20 s 已与两手明显分离，随后沿弧线飞向篮筐；机器人双脚仍在原地支撑。"
+     },
+     {
+      "t": 1.36,
+      "log_step": null,
+      "text_zh": "观察：已检查视频全部 35 帧；视频 1.08–1.28 s 篮球从篮圈上方持续下降，视频 1.32 s 到达圈口，视频 1.36 s 可见篮球进入圈下篮网。节点到达取视频约 1.36 s，精度受 25 fps 及远景分辨率限制。"
+     },
+     {
+      "t": null,
+      "log_step": 210,
+      "text_zh": "观察：log step 210 最终交付文件的重放结果为 success=true、stages=[release, through the ring]、executed_rows=137、ball_height_m=2.765；log step 219 对该文件的两次独立重放均成功且终态哈希相同。inputs/run.json 的评分器字段 success=1、replay_success=1、deterministic=1、stages_reached=2，与视频末段穿圈一致。"
+     },
+     {
+      "t": 1.36,
+      "log_step": 210,
+      "text_zh": "推断：结合视频末段从上向下穿圈及评分器成功结果，罚球命中验收已完成，事件锁存至回合结束；无状态日志，录像无法独立量出球心精确交点、圈下一个球直径阈值或毫秒级接触间断。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 0.00–0.16 s，机器人左、右张开的灵巧手从篮球两侧相向夹持并共同向上送球，随后篮球离手。候选：夹持；未见任一手合指包络篮球，故不标抓握，单纯托/舀不足以描述开局两侧夹紧。离手后篮球自由飞行。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–1.36 s，机器人双脚持续接触罚球区地面，原地支撑躯干完成送球和随挥。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "ball_height_m",
+   "value": 2.765,
+   "unit": "m",
+   "threshold": null,
+   "direction": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "观察：机器人双手夹持送出篮球，球沿弧线从上方进入篮圈，视频约 1.36 s 进入圈下篮网，罚球命中节点完成（1/1）；评分器成功且最终交付文件两次重放成功，故无失败断点。推断：视频与裁判结果共同支持成功；没有状态日志，精确穿圈交点、圈下球径阈值及毫秒级接触规则依赖裁判验收，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": 7,
+    "summary_zh": "成功运行，评分器 success=1；视频中球离手后飞向篮筐，节点时间约 1.36 s 合理，接触���注（双臂协作+其他、腿/躯干支撑）与帧一致，无失败断点。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:00:55Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-free-throw-i00-privileged--o3budzo/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-free-throw-i00-privileged--o3budzo/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "3030a4c604397762745543dd1ea6b4d5a94c4bf88bfc4ee494af5f0d97da731c",
+   "video_sha256": "f1223491fcff8c354c7751e58d80fe5c604e4562a8d638c313c97db19a6b24e6",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/free_throw/trajectories/embodied-olympics-fun-free-throw-i00-privileged--yraw2rj.json

@@ -0,0 +1,168 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-free-throw-i00-privileged--yraw2rj",
+ "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,
+   "n_actions": 259,
+   "sim_steps": 221,
+   "horizon": 1200,
+   "episode_over": 1,
+   "progress": 1.0,
+   "stages_reached": 2,
+   "stages_total": 2,
+   "fell": 0,
+   "replay_s": 4.67,
+   "bad_trajectory": 0,
+   "metric_ball_height_m": 2.797,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 430,
+   "agent_resets": 3083,
+   "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": 3.164,
+   "peak_processes": 22,
+   "peak_disk_gb": 0.242,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-free-throw-i00-privileged--yraw2rj/replay.mp4",
+  "duration_s": 2.24,
+  "fps": 25.0,
+  "frames": 56,
+  "sha256": "3150ef218adebd465ac2add3f640db83fd6e1a47cae7984ded163bdcf10288f5"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 1,
+  "final": 1,
+  "nodes": {
+   "free_throw_scored": {
+    "method": "visual",
+    "intervals": [
+     [
+      2.2,
+      null
+     ]
+    ],
+    "final": true,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–1.12 s，机器人双脚站在罚球线后，橙色篮球位于两只张开的灵巧手之间；视频 1.16–1.24 s，双手共同向上送球，视频 1.28 s 篮球与双手已明显分开并向篮筐飞行。"
+     },
+     {
+      "t": 2.2,
+      "log_step": null,
+      "text_zh": "观察：逐帧检查视频 1.28–2.20 s，篮球先上升再下降；视频 2.08–2.12 s 球仍在篮圈上方，视频 2.16 s 球正在圈口，视频 2.20 s 球已进入网中并降至篮圈下方，机器人仍由双脚支撑。节点达到时间取视频约 2.20 s。"
+     },
+     {
+      "t": 2.2,
+      "log_step": 160,
+      "text_zh": "观察：log step 159 显示修正后的生成文件与已验证轨迹数组相同；log step 160 对该最终交付文件的官方 CLI 重放返回 success=true、failed=null、stages=[release, through the ring]、steps=221、time_s=2.21（仿真秒）、ball_height_m=2.797，与 log step 156 重放的 state_sha 一致；inputs/run.json 的最终评分亦为 success=1、replay_success=1、deterministic=1、stages_reached=2/2。该重放结果与视频末段穿圈一致，确认已满足圈下继续下降及合法投篮的验收。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 0.00–1.12 s，机器人左右张开的灵巧手从篮球两侧相向夹持；视频 1.16–1.24 s 双手共同向上送球，视频 1.28 s 球已脱离双手。候选：夹持；两手没有各自合拢包住球，抓握不适用，单纯托/舀不能完整描述开局相向夹球（D12）。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–2.20 s，机器人左右脚接触球场地面，在罚球线后支撑身体完成送球并在篮球飞行期间保持站立。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "ball_height_m",
+   "value": 2.797,
+   "unit": "m",
+   "threshold": null,
+   "direction": null
+  }
+ ],
+ "failure": null,
+ "summary_zh": "机器人完成罚球命中节点，最终进度为 1/1：双手夹球后共同上送，篮球飞向篮筐并于视频约 2.20 s 从上方穿圈、继续降入网中。最终评分成功，log step 160 对最终文件的官方重放记录 release 与 through the ring 两阶段均完成，支持视频所见；会话中的其他试投不计入本次轨迹。无状态日志且视频仅 25 fps，无法独立核验毫秒级接触和精确穿圈交点，这些验收细节依据最终裁判结果确认。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "视频帧与标签一致：约1.2 s双手上送离手，2.0–2.2 s球入篮圈/网；评分success=1、stages 2/2支持节点达成，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": "2026-10-31T00:06:26Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-free-throw-i00-privileged--yraw2rj/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-free-throw-i00-privileged--yraw2rj/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "141a4078b785dccfac3d91d3dacd6789624b20bc9bbffed73eff6834b845715a",
+   "video_sha256": "3150ef218adebd465ac2add3f640db83fd6e1a47cae7984ded163bdcf10288f5",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/free_throw/trajectories/embodied-olympics-fun-free-throw-i00-standard--28vucfr.json

@@ -0,0 +1,169 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-free-throw-i00-standard--28vucfr",
+ "taskKey": "embodied-olympics-fun/free_throw",
+ "dag_sha256": "c2d4b07a9ca0ff585b81cafb3eecf9f2484164969c066ba3cd5ef111abe016a5",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "live_success": 0,
+   "graded_by_live": 1,
+   "replays": 0,
+   "live_rule": 1,
+   "live_verdict_error": 0,
+   "finalized": 1,
+   "missing_log": 0,
+   "infra_error": 0,
+   "grader_error": 0,
+   "sim_error": 0,
+   "sim_crashed_on_call": 0,
+   "instance_mismatch": 0,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "calls": 1,
+   "rows": 0,
+   "resets": 0,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 2699.2,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 4,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 1,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.0,
+   "peak_processes": 0.0,
+   "peak_disk_gb": 0.0,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-free-throw-i00-standard--28vucfr/replay.mp4",
+  "duration_s": 0.04,
+  "fps": 25.0,
+  "frames": 1,
+  "sha256": "f5ea9092e2daeb267e46d92b7987d621f520594a6f3c3d8c9f692d8920fb8cd6"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "free_throw_scored": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": 4,
+      "text_zh": "观察：本次视频唯一一帧（视频 0.00 s，片长 0.04 s）显示机器人站在罚球区，橙色篮球仍位于胸前双手之间，未见离手、飞向篮圈或穿圈；log step 4 的现场服务观测为 seconds=0.0、control_steps_watch=0。"
+     },
+     {
+      "t": 0.0,
+      "log_step": 79,
+      "text_zh": "观察：评分字段 n_actions=0、sim_steps=0、rows=0、live_success=0；log step 79 记录 Gemini 过滤器阻断。此前 log step 63、74、77、78 的投球数值来自 agent 自建 MuJoCo 场景测试，不是本次现场评分回合，不能计入节点进度。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "提供方过滤器中断前没有交付或执行现场动作；录像仅为开局持球静帧，没有可标注的投篮操作接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "ball_height_m",
+   "value": null,
+   "unit": "m",
+   "threshold": null,
+   "direction": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 79,
+   "where": "首步前",
+   "evidence_zh": "观察：log step 79 以“This request was blocked by Gemini's filters”结束，评分记录 n_actions=0、sim_steps=0、rows=0；视频 0.00 s 仅有开局持球画面。推断：现场投篮尚未启动即被提供方过滤器中断，没有可定位的物理失败步骤，按外因中断规则将断点记为 unknown。"
+  },
+  "unmet": {
+   "free_throw_scored": false
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：inputs/replay.mp4 仅含视频 0.00 s 的一帧，inputs/ 中没有状态日志；评分记录为零动作、零仿真步，log step 79 记录提供方过滤器阻断。推断：这是结果尚未由实际投篮确定前的外因中断，依 CASES.md FAIL-5 不列为机器人能力失败，不标“没尝试”或根因；会话内自建仿真测试不能替代评分轨迹。原生 ball_height_m 未在评分字段中报告，保留 null。"
+ },
+ "summary_zh": "本次评分视频只有开局持球的一帧，现场动作数及仿真步数均为零，“罚球命中”未达成，节点终值为 0/1。log step 79 记录 Gemini 提供方过滤器中断，故按外因中断处理，断点为 unknown／首步前，现象与根因留空；会话内自建 MuJoCo 测试不属于评分回合。没有实际投篮过程或状态日志可用于进一步分析，应重测并排除于能力失败统计。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "标注与证据一致：全程无任何提交（77 个步骤 sim.submit 均为 0），n_actions=0，log step 79 为 Gemini 过滤器阻断，视频仅一帧开局画面；按 FAIL-5 外因中断，断点 unknown/首步前，现象与根因为空且有 gap_zh，节点 false 合理。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:03:02Z",
+  "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-free-throw-i00-standard--28vucfr/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-free-throw-i00-standard--28vucfr/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "a28bfaa0e80f4cf6d3d609d77ca3b3a1b6aaf9994c661efaf0f33e4abcdb8ddd",
+   "video_sha256": "f5ea9092e2daeb267e46d92b7987d621f520594a6f3c3d8c9f692d8920fb8cd6",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/free_throw/trajectories/embodied-olympics-fun-free-throw-i00-standard--dkxv6yq.json

@@ -0,0 +1,196 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-free-throw-i00-standard--dkxv6yq",
+ "taskKey": "embodied-olympics-fun/free_throw",
+ "dag_sha256": "c2d4b07a9ca0ff585b81cafb3eecf9f2484164969c066ba3cd5ef111abe016a5",
+ "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": 303,
+   "sim_steps": 961,
+   "calls": 37,
+   "rows": 920,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 3971.7,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.33,
+   "peak_processes": 9,
+   "peak_disk_gb": 1.093,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-free-throw-i00-standard--dkxv6yq/replay.mp4",
+  "duration_s": 3.04,
+  "fps": 25.0,
+  "frames": 76,
+  "sha256": "39499c6194e1806ffeedb43a1752afec9b90f402918530ed3d0bfadd733e03c1"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "free_throw_scored": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": 169,
+      "text_zh": "观察：视频 0.00 s 篮球位于机器人胸前双手之间；视频 1.00–2.40 s 双臂将球抬至面前。log step 169 的控制脚本重置到 episode 6，log step 170 的服务回执确认该回合开始并执行持球抬升；评分字段 graded_by_live=1，采用最后现场回合。"
+     },
+     {
+      "t": 2.64,
+      "log_step": 176,
+      "text_zh": "观察：视频 2.40–2.64 s 机器人双手转动，篮球逐渐落到掌下；视频 2.68–3.00 s 球与双手分离并落向机器人脚前，全程未飞向篮圈，也未穿圈。agent 自述“球在最后转掌时滑脱”（log step 176），与上述画面一致。"
+     },
+     {
+      "t": 3.0,
+      "log_step": 177,
+      "text_zh": "观察：视频 3.00 s 篮球已在机器人脚前近地面处，机器人仍站立；log step 177 的机器人服务裁判回执为 episode=6、done=true、outcome=\"the ball touched the floor\"、stages_reached=1/2，评分 success=0。原生离手阶段不等于 DAG 的罚球命中，故该节点始终未达到。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 0.00 s 双手从篮球两侧相向夹持，随后共同抬球，视频 2.40–2.64 s 转掌时球下滑。候选：夹持；开局不是单手合拢包球的抓握，后续远景不能精确辨认各手指的包络及接触变化。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–3.00 s 机器人双脚接触球场地面，支撑躯干及双臂持球、抬球和转掌，末帧仍站立。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "ball_height_m",
+   "value": null,
+   "unit": "m",
+   "threshold": null,
+   "direction": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "free_throw_scored",
+   "t": 2.6,
+   "log_step": 173,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 2.40–2.64 s 机器人正在转动双手准备送球，球从手间下滑；视频 2.68–3.00 s 球继续落向脚前，未形成朝篮圈的投射。log step 171、173 提交的是转掌控制段，log step 176 改为站立等待；log step 177 的裁判以篮球触地结束最后现场回合，罚球命中未补成。"
+  },
+  "unmet": {
+   "free_throw_scored": false
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没保持住",
+    "t": 2.6,
+    "log_step": 176,
+    "evidence_zh": "观察：视频 1.00–2.40 s 篮球随双手抬升并停留在面前，视频 2.60–2.72 s 球落至手下并脱离，随后继续下落；这是已有持球在转掌过程中丢失。agent 自述“球在最后转掌时滑脱”（log step 176），裁判在 log step 177 报告篮球触地。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "t": 2.6,
+    "log_step": 171,
+    "evidence_zh": "推断：log step 171 的控制脚本要求双手围绕目标球心高度 1.5 m 调整掌向，并持续发送双手闭合命令；log step 173 执行其后半段，视频 2.40–2.72 s 却显示球从两手下方掉出，未被维持到计划的送球阶段。结合 agent 在 log step 176 确认转掌滑脱，归为持球到转掌过渡的控制失败；没有证据进一步区分接触几何、力控制或轨迹跟踪。",
+    "source_zh": "日志 / 视频 / agent 自述"
+   }
+  ],
+  "gap_zh": "没有状态日志；远景录像不能分辨后续每根手指是否形成包络抓握，也不能核验毫秒级接触先后，实际接触记录仅明确可见的双手夹持和双脚支撑。日志没有报告最后回合各接触点的力及足以归因的跟踪误差，故控制根因不细分；原生 ball_height_m 未在评分字段中报告，保留 null。"
+ },
+ "summary_zh": "最后现场回合中，机器人双手持球并抬至面前，但视频约 2.6 s 转掌时球滑出双手，随后落向脚前；裁判报告篮球触地，罚球命中节点最终为 0/1。现象为没保持住，根因归为控制；无状态日志且远景无法分辨手指接触细节，不能进一步确定接触设计或跟踪问题。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "视频 2.64 s 前球仍在双手间，约 2.68–3.00 s 球滑落至脚前，节点未达成、断点与「没保持住」现象均有画面和裁判回执支持；根因归控制并声明无法细分，措辞克制，未发现需改内容的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:09:35Z",
+  "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-free-throw-i00-standard--dkxv6yq/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-free-throw-i00-standard--dkxv6yq/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "c52f8829d4db0156cdd80328ebc46755beb878b344363b413a67fa370ecc74e6",
+   "video_sha256": "39499c6194e1806ffeedb43a1752afec9b90f402918530ed3d0bfadd733e03c1",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/free_throw/trajectories/embodied-olympics-fun-free-throw-i00-standard--e78dxun.json

@@ -0,0 +1,189 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-free-throw-i00-standard--e78dxun",
+ "taskKey": "embodied-olympics-fun/free_throw",
+ "dag_sha256": "c2d4b07a9ca0ff585b81cafb3eecf9f2484164969c066ba3cd5ef111abe016a5",
+ "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": 162,
+   "sim_steps": 1189,
+   "calls": 20,
+   "rows": 939,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 1963.7,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 0.381,
+   "peak_processes": 10,
+   "peak_disk_gb": 1.806,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-free-throw-i00-standard--e78dxun/replay.mp4",
+  "duration_s": 1.64,
+  "fps": 25.0,
+  "frames": 41,
+  "sha256": "92262515387df4083ce5afb5e44eb34af47216c12e9b0415820cb47e9a4b2b08"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "free_throw_scored": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–1.20 s，机器人双脚站立，篮球停在胸腹前的左右手之间；视频 1.28–1.60 s 篮球脱离双手向脚前下落，全片未见球飞至篮圈或从上穿圈。"
+     },
+     {
+      "t": 1.6,
+      "log_step": 141,
+      "text_zh": "观察：log step 134 的服务状态为 episode=6、control_steps=162、done=true、reason=failed；log step 141 读取该最后回合的服务元数据，outcome 为“the ball touched the floor”、stages_reached=1，与视频末段失球下落一致。该原生阶段数不是 DAG 节点完成数，罚球命中仍为未达。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 0.00–1.20 s，机器人左右手从篮球两侧相向夹持，随后球从两手下方掉出。候选：夹持；开局手掌相向承球，未见单手以相对手指包络夹住球，抓握不适用，托/舀不能完整描述两侧夹持。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–1.60 s，机器人双脚持续接触罚球区地面，支撑持球及屈臂动作，未见跌倒。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "ball_height_m",
+   "value": null,
+   "unit": "m",
+   "threshold": null,
+   "direction": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "free_throw_scored",
+   "t": 1.28,
+   "log_step": 141,
+   "where": "操作中",
+   "evidence_zh": "观察：视频约 1.28 s，机器人仍在胸腹前屈臂准备投篮时，篮球开始从双手下方滑出；视频 1.40–1.60 s 球降至腿旁及脚前，未向篮圈飞行。log step 141 的最后回合服务元数据报告篮球触地失败，之后未修复。"
+  },
+  "unmet": {
+   "free_throw_scored": false
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没保持住",
+    "t": 1.28,
+    "log_step": 141,
+    "evidence_zh": "观察：视频 0.00–1.20 s 篮球保持在机器人两手之间，视频 1.28–1.36 s 从手下方露出并持续下落，视频 1.40–1.60 s 两手空着、篮球降到腿旁及脚前；没有完成向上送球出手。log step 141 对应最后回合的服务输出为“the ball touched the floor”。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "交付",
+    "item": "交付操作出错",
+    "source_zh": "日志",
+    "evidence_zh": "推断：最终现场回合误选了旧动作策略。log step 132 新增 run_final() 及其快速上挥、分掌出手方案，log step 134 执行 python gen.py final 却产生旧分支的 t1 数据；log step 141 确认该回合实际仅运行 grip、load 段并以篮球触地结束。log step 110 所写入口读取 what 后无条件调用 run_throw(p, tag)，log step 143 的带 assert 替换成功记录再次核实旧入口未按 final 分派；修复发生在 episode=6 已结束之后。此策略选择错误使评分回合重复旧屈臂蓄力动作，与视频约 1.28 s 失球一致；新方案是否能命中未获验证。"
+   }
+  ],
+  "gap_zh": "没有状态日志；固定远景无法精确区分每根手指的接触���失球时的接触力，故不进一步归因为握力、摩擦或关节跟踪误差。原生指标 ball_height_m 未见评分器字段报告，保留 null。"
+ },
+ "summary_zh": "评分对象是最后一次现场回合（episode=6），唯一节点“罚球命中”未达成，最终进度为 0/1；视频约 1.28 s 篮球从双手下滑，至视频 1.60 s 已落向脚前，机器人仍站立。日志核实最终调用未分派到新投篮函数，而重复旧屈臂蓄力方案，修复发生在回合结束后；失球现象为没保持住。没有状态日志及接触力记录，不能进一步判定具体失球力学原因。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "节点未达、断点、没保持住现象和交付操作出错根因均与视频及 log step 134/141/143 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": "2026-10-31T00:05:05Z",
+  "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-free-throw-i00-standard--e78dxun/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-free-throw-i00-standard--e78dxun/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "d1fb789af5a218a5182169c899ea65be54192a484a47e99164540666ac06b938",
+   "video_sha256": "92262515387df4083ce5afb5e44eb34af47216c12e9b0415820cb47e9a4b2b08",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/free_throw/trajectories/embodied-olympics-fun-free-throw-i00-standard--syygljz.json

@@ -0,0 +1,191 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-free-throw-i00-standard--syygljz",
+ "taskKey": "embodied-olympics-fun/free_throw",
+ "dag_sha256": "c2d4b07a9ca0ff585b81cafb3eecf9f2484164969c066ba3cd5ef111abe016a5",
+ "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": 109,
+   "sim_steps": 200,
+   "calls": 10,
+   "rows": 200,
+   "resets": 1,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 253.8,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 2.486,
+   "peak_processes": 11,
+   "peak_disk_gb": 0.921,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-free-throw-i00-standard--syygljz/replay.mp4",
+  "duration_s": 1.12,
+  "fps": 25.0,
+  "frames": 28,
+  "sha256": "e37408f21138821ef9e442ef69e6142e88e7034b0646b96da86f191d060864a4"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "free_throw_scored": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.4,
+      "log_step": 26,
+      "text_zh": "观察：视频 0.00–0.32 s 机器人双手在胸前夹持并抬起篮球；视频 0.40 s 球已与双手分离，视频 0.68–0.80 s 双臂继续抬伸时球已在手前自由飞行。log step 26 记录唯一一次重置后第二回合，四段动作执行后裁判返回 the ball touched the floor，stages_reached=1；这是最终受评现场回合。"
+     },
+     {
+      "t": 1.08,
+      "log_step": 26,
+      "text_zh": "观察：逐帧检查视频 0.00–1.08 s，篮球仅在机器人前方形成低而短的弧线，视频 0.56–0.64 s 附近达到最高处，随后下降，视频 1.08 s 已到篮筐前方地面附近；全程未到篮圈处，更未从上向下穿圈。成功参考视频 0.88–1.12 s 可见球在篮板及篮圈处下降，本回合没有该现象；裁判触地失败与本回合画面一致。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 0.00–0.32 s 左右手同时在篮球两侧相向夹持并共同抬球，视频 0.36–0.40 s 球逐渐脱离手部；开局手指未包住球，不能称抓握。候选：夹持；相向夹紧不属于现有推/扫、压/擦或托/舀取值。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–1.08 s 机器人双脚接触球场地面并支撑躯干，抬球及后续空手推送时保持站姿。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "ball_height_m",
+   "value": null,
+   "unit": "m",
+   "threshold": null,
+   "direction": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "free_throw_scored",
+   "t": 0.4,
+   "log_step": 26,
+   "where": "操作中",
+   "evidence_zh": "推断：视频 0.36–0.40 s 球在机器人抬臂时提前脱手，视频 0.68–0.80 s 后续推送已为空手动作，篮球始终远低于并未到达篮圈，最终落向地面。log step 22 的脚本将抬球列为推送之前的独立阶段，log step 26 确认此回合以触地失败结束，仅达到离手阶段，罚球命中未完成且之后没有新的现场回合修复。"
+  },
+  "unmet": {
+   "free_throw_scored": false
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没保持住",
+    "t": 0.4,
+    "log_step": 26,
+    "evidence_zh": "推断：视频 0.00–0.32 s 机器人双手夹持球并抬起；视频 0.40–0.64 s 球已离手而双臂停留在抬球位置，视频 0.68–0.80 s 才继续作空手推送。结合 log step 22 的先抬球、再推送及开手程序，球未保持到计划推送阶段便已脱手；随后视频 0.84–1.08 s 球在篮筐前方下落，log step 26 裁判报告篮球触地。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "t": 0.4,
+    "log_step": 24,
+    "source_zh": "日志",
+    "evidence_zh": "推断：log step 22 的脚本在各阶段调用 ramp([], target, T)，而 log step 24 修补为 rows 为空时令 cur=target，因此抬球阶段的全部目标行直接等于抬球终点，没有从上一姿态渐变；log step 26 随后实际执行此脚本。视频 0.24–0.40 s 双臂上抬���程中球已脱离，后续推送空挥，说明此控制序列未协调持球、加速及释放。缺少接触力及关节跟踪记录，子类不进一步细分。"
+   }
+  ],
+  "gap_zh": "没有状态日志或逐帧接触力，无法确定最后一次手球接触的毫秒时刻，也不能进一步区分接触几何、夹持力与关节跟踪的贡献；脱手时间取视频 0.36–0.40 s 的近似锚点。输入评分器未报告 ball_height_m，故原生指标留 null。log step 48 之后的本地模型搜索、log step 169 的说明文件及 log step 190 的末次搜索自述均不是新增现场回合，不能替代最终受评视频。"
+ },
+ "summary_zh": "最终受评的是 log step 26 重置后的第二次现场投篮，罚球命中节点始终未达成，最终为 0/1。机器人抬球时约视频 0.36–0.40 s 提前失去手球接触，随后空手推送，球短距离飞行后落地；日志中的轨迹生成方式支持控制序列协调不足的归因。没有状态日志及接触力记录，精确脱手时刻和控制子类仍有缺口，后续本地搜索未在现场执行。",
+ "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（罚球命中·操作中，球在抬臂时脱手）、没保持住现象与控制根因均与视频及 log step 22/24/26 一致；未发现需要修改的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:05:46Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-free-throw-i00-standard--syygljz/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-free-throw-i00-standard--syygljz/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "1de06028c0bd5761535f26f88a6cddb545a42e19ed25922d70112105c18d1bfd",
+   "video_sha256": "e37408f21138821ef9e442ef69e6142e88e7034b0646b96da86f191d060864a4",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/free_throw/trajectories/embodied-olympics-fun-free-throw-i00-standard--waqmafq.json

@@ -0,0 +1,195 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-free-throw-i00-standard--waqmafq",
+ "taskKey": "embodied-olympics-fun/free_throw",
+ "dag_sha256": "c2d4b07a9ca0ff585b81cafb3eecf9f2484164969c066ba3cd5ef111abe016a5",
+ "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": 313,
+   "sim_steps": 1795,
+   "calls": 18,
+   "rows": 1615,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 2578.7,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 3.939,
+   "peak_processes": 13,
+   "peak_disk_gb": 0.38,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-free-throw-i00-standard--waqmafq/replay.mp4",
+  "duration_s": 3.16,
+  "fps": 25.0,
+  "frames": 79,
+  "sha256": "892dc231939729f6fb43c15e84354574e0a9c42126e1f7f65c0271fac65a621b"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "free_throw_scored": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00 s，机器人站在罚球区，左右张开的手共同夹持胸前的橙色篮球；视频 0.80–2.72 s，双手把球缓慢降到腰前。"
+     },
+     {
+      "t": 2.8,
+      "log_step": null,
+      "text_zh": "观察：视频 2.76–2.84 s，机器人双手迅速上抬，篮球却从两手之间向下脱离，未随手上升；视频 2.88–3.12 s，球继续落到机器人脚前，始终远离篮圈，全片没有从上向下穿圈。"
+     },
+     {
+      "t": 3.12,
+      "log_step": 119,
+      "text_zh": "观察：log step 119 的服务返回标明 episode=6、resets_left=0、done=True、outcome=\"the ball touched the floor\"、stages_reached=1；这是最后一次现场回合，与视频 3.12 s 球落到脚前地面的画面一致。run.json 的 graded_by_live=1、live_success=0，罚球命中未达成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 0.00–2.72 s，机器人左右张开的灵巧手从两侧共同夹持篮球并下移，未见任一只手合拢包住球；视频 2.76–2.84 s，双手上抬而球脱离下落。候选：夹持；相向夹紧不属于单手包络抓握，托/舀不能完整描述两侧共同夹持。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–3.12 s，机器人双脚接触球场地面，双腿支撑躯干完成持球、下移及双臂上抬过程，未见倒伏。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "ball_height_m",
+   "value": null,
+   "unit": "m",
+   "threshold": null,
+   "direction": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "free_throw_scored",
+   "t": 2.76,
+   "log_step": 119,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 2.72 s 篮球仍在机器人腰前两手之间；视频 2.76–2.84 s，两臂上抬而球向下脱离，随后一直落至视频 3.12 s 的脚前地面，投篮未获修复。log step 119 的最后一局服务结果为篮球触地失败，因此断点是正在进行的罚球命中步骤。"
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没保持住",
+    "t": 2.76,
+    "log_step": 119,
+    "evidence_zh": "观察：视频 2.64–2.72 s，篮球仍随双手保持在腰前；视频 2.76–2.84 s，双手快速上抬后球留在下方，视频 2.88–3.12 s 几乎向下落至脚前，没有形成飞向篮圈的上抛。log step 119 的服务裁判返回 \"the ball touched the floor\"。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "抓取和接触设计",
+    "alt_feasible": "未知",
+    "source_zh": "日志",
+    "evidence_zh": "推断：log step 117 的生成脚本用双臂向内偏移与夹紧力矩保持球，log step 118 为最后一局生成命令时 grip=0.0，依靠张开双手的接触承载快速送球；视频 2.76–2.84 s 显示双手上抬而球向下脱离，说明该接触布局未把手部运动有效传递给球。agent 在 log step 121 自述最后一次 V 形掌姿在投篮启动时失球，且本地模型未能验证真实手球��触；这与画面一致，支持接触设计不适配，但不足以量化摩擦或夹紧力。日志未验证能修复本控制器的具体替代握法。"
+   }
+  ],
+  "gap_zh": "本任务没有状态日志，场地远景和放大裁剪只能辨认双手共同持球及随后失球，无法区分各掌、拇指的精确接触、接触力或毫秒级离手先后；抓取和接触设计根因由命令脚本、视频与 agent 自述综合推断。评分器未报告 ball_height_m，故原生指标值留空。",
+  "unmet": {
+   "free_throw_scored": false
+  }
+ },
+ "summary_zh": "评分采用最后一次现场回合：机器人双手持球并下移后尝试上抛，但视频约 2.76 s 双手抬起时未保持住球，视频 3.12 s 球落到脚前，罚球命中最终为 0/1。控制脚本与 agent 自述支持张开双手的夹持接触设计未能承载送球动作；前五次测试不计入本次进度和现象。没有状态日志或接触力记录，无法精确核实滑脱机制，球心高度指标也未报告。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "标注与视频和日志一致：最后一次现场回合中，视频约 2.76 s 双手上抬而球向下脱离，3.12 s 落到脚前，罚球命中未达成；断点、现象（没保持住）与根因（抓取和接触设计，推断）有据，未发现需要修改的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:03:22Z",
+  "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-free-throw-i00-standard--waqmafq/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-free-throw-i00-standard--waqmafq/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "bc1e196d697527c33aae6c4de116584b8fcf849bbcb28d646dfdb33b99af20f1",
+   "video_sha256": "892dc231939729f6fb43c15e84354574e0a9c42126e1f7f65c0271fac65a621b",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/free_throw/trajectories/embodied-olympics-fun-free-throw-i00-standard--xxktbfx.json

@@ -0,0 +1,189 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-free-throw-i00-standard--xxktbfx",
+ "taskKey": "embodied-olympics-fun/free_throw",
+ "dag_sha256": "c2d4b07a9ca0ff585b81cafb3eecf9f2484164969c066ba3cd5ef111abe016a5",
+ "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": 160,
+   "sim_steps": 659,
+   "calls": 23,
+   "rows": 645,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 553.9,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.273,
+   "peak_processes": 8,
+   "peak_disk_gb": 0.836,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-free-throw-i00-standard--xxktbfx/replay.mp4",
+  "duration_s": 1.64,
+  "fps": 25.0,
+  "frames": 41,
+  "sha256": "be5b38741438f37a857cc6b8ff4b52153f5b5aff7a19ea8a061205ffb868a338"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "free_throw_scored": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.16,
+      "log_step": 71,
+      "text_zh": "观察：视频 0.00–0.12 s，机器人双手夹持胸前篮球并一起上抬；视频 0.16 s，球已与双手分离，随后主要向上飞行。log step 71 启动最后一局 shotfinal，log step 72 的服务返回 episode=6、control_steps=160、stages_reached=1/2、reason=failed；本视频为该最后一局，而非前五局测试的拼接。"
+     },
+     {
+      "t": 1.6,
+      "log_step": 72,
+      "text_zh": "观察：已逐帧检查视频 0.00–1.60 s；球约在视频 0.64 s 到达最高处，仍远在篮圈前方，随后在视频 1.20–1.60 s 下落，末帧落至机器人与篮架之间的地面，始终未到达篮圈。成功参考视频 1.04–1.12 s 可见球从圈上下降穿圈；本局没有这一过程，且评分字段 live_success=0。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 0.00–0.12 s，机器人左右张开的手从篮球两侧夹持并同时向上送球，未见任一只手合拢包住篮球；候选“夹持”，故不用抓握。视频 0.16 s 后球离手自由飞行。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–1.60 s，机器人双脚着地支撑躯干完成上肢投篮，未见跌倒。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "ball_height_m",
+   "value": null,
+   "unit": "m",
+   "threshold": null,
+   "direction": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "free_throw_scored",
+   "t": 0.16,
+   "log_step": 71,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 0.16 s，机器人已把球向上送离双手；视频 0.64–1.60 s，球在篮圈前方达到顶点并下落至地面，全程未穿圈。log step 72 返回最后一局失败且仅达到两个原生阶段中的一个。断点为正在执行但未完成的罚球命中步骤，出手时刻为粗略导航锚点。"
+  },
+  "unmet": {
+   "free_throw_scored": false
+  },
+  "phenomena": [
+   {
+    "cat": "运动",
+    "item": "位姿不准",
+    "t": 0.64,
+    "log_step": null,
+    "evidence_zh": "观察：视频 0.16–0.64 s，篮球离开机器人双手后以上升为主，朝篮圈方向只前进较小距离；视频 1.20–1.60 s，球在篮圈前方下落至地面，未接近圈内穿越路径。投篮轨迹明显过短。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "evidence_zh": "推断：log step 76 所打印的最终 shotfinal 控制代码按预设关节目标和速度同步抬肩、屈伸肘，再延迟改变肩部偏航，并保持双手张开；视频 0.00–0.16 s 显示机器人执行上送并离手，但视频 0.64–1.60 s 的球路前向位移不足而落在篮圈之前。这支持投篮控制未产生合适的出手运动；现有证据不足以细分为接触设计、轨迹跟踪或某个参数错误。",
+    "source_zh": "日志 / 视频"
+   }
+  ],
+  "gap_zh": "没有状态日志；录像仅有场地视角，缺少出手瞬间的球速、手球接触力和实际关节跟踪数据，无法细分控制根因或确定毫秒级离手��间。评分记录未报告 ball_height_m，原生指标值保留 null。"
+ },
+ "summary_zh": "评分视频对应最后一次现场罚球：机器人双手上送并出球，但篮球前向位移不足，随后在篮圈前方落地，罚球命中节点始终未达成，最终为 0/1。断点在投篮操作中，现象为位姿不准，结合最终控制脚本将根因归为控制。没有状态日志或离手速度、接触力数据，无法进一步区分具体控制子类。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "节点未达成、断点、接触与现象（位姿不准）均与视频和服务返回（最后一局 failed，stages 1/2，control_steps 160，视频 1.64 s 对应）一致；未发现需要修改的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:13:35Z",
+  "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-free-throw-i00-standard--xxktbfx/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-free-throw-i00-standard--xxktbfx/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "ba0469307fb52db53a6d12242a823bf1b9ad88c45cbdc3cd8b0b299b3405958b",
+   "video_sha256": "be5b38741438f37a857cc6b8ff4b52153f5b5aff7a19ea8a061205ffb868a338",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/free_throw/trajectories/embodied-olympics-fun-free-throw-i00-standard--y78ah7z.json

@@ -0,0 +1,189 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-free-throw-i00-standard--y78ah7z",
+ "taskKey": "embodied-olympics-fun/free_throw",
+ "dag_sha256": "c2d4b07a9ca0ff585b81cafb3eecf9f2484164969c066ba3cd5ef111abe016a5",
+ "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": 55,
+   "sim_steps": 315,
+   "calls": 12,
+   "rows": 315,
+   "resets": 5,
+   "reset_budget": 5,
+   "sim_startups": 1,
+   "elapsed_s": 4918.9,
+   "video_rendered": 1,
+   "first_success": 0,
+   "first_reason_code": 2,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.206,
+   "peak_processes": 12,
+   "peak_disk_gb": 0.778,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-free-throw-i00-standard--y78ah7z/replay.mp4",
+  "duration_s": 0.56,
+  "fps": 25.0,
+  "frames": 14,
+  "sha256": "9478a299da0082c94a231f3cc61dbb7f89a8898f8e2b8c6d5dd4d069a53576f0"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 1,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "free_throw_scored": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 0.0,
+      "log_step": null,
+      "text_zh": "观察：视频 0.00–0.36 s，机器人站在罚球区，左右张开的手共同夹持身前的橙色篮球；视频 0.40–0.52 s，双手和篮球向腰部下降，球没有飞向篮板或穿过篮圈。已检查全部14帧，包括末帧视频 0.52 s。"
+     },
+     {
+      "t": 0.52,
+      "log_step": 99,
+      "text_zh": "观察：log step 99 中 agent 重置后执行第六次现场回合，机器人服务返回 control_steps=55、done=True、reason=failed，原生裁判结果为“the ball was touched again after the release”，阶段1/2；log step 100 确认 episode=6、resets_left=0。此为最后受评回合；与视频中球仍在机器人身前低处、未入篮一致，罚球命中未达成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "双臂协作",
+      "how": "其他",
+      "note_zh": "观察：视频 0.00–0.36 s，左右张开的灵巧手从篮球两侧相向夹持，共同持球；视频 0.40–0.52 s 双手和球向下运动。候选：夹持；未见单手合拢包络篮球，抓握不适用，侧向相向夹紧也不能仅用托/舀描述。log step 99 的脚本以双手 grip=0 执行，服务记录离手后再触球，但画面不能细分再接触的具体手指。"
+     },
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 0.00–0.52 s，机器人双脚接触罚球区地面，双腿支撑躯干并配合双手持球及投篮准备动作，未见跌倒。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "ball_height_m",
+   "value": null,
+   "unit": "m",
+   "threshold": null,
+   "direction": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "free_throw_scored",
+   "t": 0.48,
+   "log_step": 99,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 0.40–0.52 s，机器人正在执行双手投篮动作，双手及篮球反而向腰部下降，球未飞向篮圈；log step 99 中最后现场回合的服务结果为离手后再次触球、failed、阶段1/2，log step 100 确认回合已结束。因此失败步骤为罚球命中，视频 0.48 s 仅作失败附近的导航锚点。"
+  },
+  "unmet": {
+   "free_throw_scored": false
+  },
+  "phenomena": [
+   {
+    "cat": "接触",
+    "item": "没保持住",
+    "t": 0.48,
+    "log_step": 99,
+    "evidence_zh": "推断：视频 0.00–0.36 s 机器人双手原本持住球，视频 0.40–0.52 s 手和球向下而未形成向篮圈的飞行；log step 99 的脚本打印计划离手为仿真 0.59 s，但机器人服务已在55个控制步内以“离手后再次触球”终止。这说明送球过程中持球接触提前中断并再次建立，未保持至计划出手；这里记录一次持球失稳，不将再触球判罚重复标为第二现象。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "evidence_zh": "推断：log steps 97–99 中 agent 用自建离线机械臂模型搜索双手推投参数，再把预计算轨迹交给机器人服务执行；离线预测球向前上方出手，但视频 0.40–0.52 s 的实际双手及球向下，且 log step 99 服务记录提前离手后再触球。送球控制未维持球与双手的连续接触。log step 103 将其解释为肩部滞后于肘部，该机制仅为 agent 自述，现有证据不足以细分控制子类。",
+    "source_zh": "日志 / 视频"
+   }
+  ],
+  "gap_zh": "没有状态日志；25 fps 的场地固定视角不能分辨毫秒级离手、再接触的先后及具体接触指节，依最后现场回合的原生服务结果确认该判罚。log step 103 所述肩关节力矩滞后和手掌下降幅度没有本回合可独立核验的关节或力矩记录，因此根因只标控制。评分器 reward 未报告 ball_height_m，原生指标值保留 null。"
+ },
+ "summary_zh": "受评的是最后一次、即第六次现场回合，罚球命中未达成，最终进度0/1。视频 0.40–0.52 s 可见机器人双手与球向下运动，log step 99 的服务记录离手后再次触球并判负，说明送球过程中未保持住篮球；根因归为控制。没有状态日志，具体接触时刻与肩肘滞后机制不能独立核实。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "受评回合为第六次现场回合，视频仅14帧(0.56 s)，所见双手持球下降、球未飞向篮圈，与 log step 99 服务结果（离手后再触球、failed、阶段1/2）一致；节点未达成、断点、接触标注和单一现象/根因均有证据支撑，未发现需修改之处。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:14:21Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-free-throw-i00-standard--y78ah7z/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-free-throw-i00-standard--y78ah7z/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "e792176461d1be57911e511254628a84127fb4def1b6403615d6a99e99b877b0",
+   "video_sha256": "9478a299da0082c94a231f3cc61dbb7f89a8898f8e2b8c6d5dd4d069a53576f0",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}

phase2/embodied-olympics-fun/hurdles_freestanding/trajectories/embodied-olympics-fun-hurdles-freestanding-i00-privileged--4uxwkep.json

@@ -0,0 +1,258 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "embodied-olympics-fun-hurdles-freestanding-i00-privileged--4uxwkep",
+ "taskKey": "embodied-olympics-fun/hurdles_freestanding",
+ "dag_sha256": "631f06ede76d388c24511754cca206280db214cee4838b8a909aa9aebd7ae56e",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "success": 0,
+   "replay_success": 0,
+   "deterministic": 1,
+   "n_actions": 2872,
+   "sim_steps": 2872,
+   "horizon": 14000,
+   "episode_over": 1,
+   "progress": 0.0,
+   "stages_reached": 0,
+   "stages_total": 5,
+   "fell": 0,
+   "replay_s": 26.8,
+   "bad_trajectory": 0,
+   "metric_distance_m": 13.6,
+   "metric_hurdles_cleared": 0.0,
+   "grader_error": 0,
+   "video_rendered": 1,
+   "agent_sessions": 122,
+   "agent_resets": 197,
+   "agent_replays": 0,
+   "agent_cost_self_reported": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 1.087,
+   "peak_processes": 11,
+   "peak_disk_gb": 0.247,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0,
+   "session_log_tampered": 0,
+   "session_log_from_monitor": 0
+  }
+ },
+ "video": {
+  "kind": "hf",
+  "url": "https://huggingface.co/datasets/MLL-Lab/embodied-exam-media/resolve/main/runs/embodied-olympics-fun-hurdles-freestanding-i00-privileged--4uxwkep/replay.mp4",
+  "duration_s": 28.76,
+  "fps": 25.0,
+  "frames": 719,
+  "sha256": "6be123d35ef4fff66c4240c1e46b6d7be6977efb97466d651c0b270341ae9b87"
+ },
+ "state": null,
+ "progress": {
+  "method": "visual",
+  "denominator": 5,
+  "base": 0,
+  "peak": 0,
+  "final": 0,
+  "nodes": {
+   "hurdle_1_cleared": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 27.8,
+      "log_step": null,
+      "text_zh": "观察：视频 0–27.8 s，机器人沿跑道向第一栏行走；视频 27.8–28.72 s 的逐帧放大图中，机器人仍在第一栏近端贴地迈步，双腿未腾空越过横杆，视频结束前未完成第一栏。"
+     },
+     {
+      "t": 28.72,
+      "log_step": 198,
+      "text_zh": "观察：log step 198 运行最终生成脚本并重新保存轨迹，输出 2872 行动作及裁判状态 failed=went under the top of the first hurdle，最终骨盆坐标为 [13.597,0.088,1.007] 米；评分器 n_actions=2872、stages_reached=0、metric_hurdles_cleared=0，与视频中止于第一栏前的情况一致。该日志对应最终交付，不采用此前其他测试的跨栏高度。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "腿/躯干",
+      "how": "支撑",
+      "note_zh": "观察：视频 2–28.72 s，机器人双脚交替支撑红色跑道、向第一栏迈步，末段脚部仍贴近跑道；未见完成腾空越栏。全景分辨率不足以确认脚与栏杆有无瞬时接触，故不额外标碰栏或避开。"
+     }
+    ]
+   },
+   "hurdle_2_cleared": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 28.72,
+      "log_step": null,
+      "text_zh": "观察：视频 0–28.72 s，机器人只从起点行进到第一栏近端，未到达或尝试第2栏；视频 28.72 s 的全景中，第2栏仍在机器人前方远处。评分器 stages_reached=0、metric_hurdles_cleared=0。"
+     }
+    ]
+   },
+   "hurdle_3_cleared": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 28.72,
+      "log_step": null,
+      "text_zh": "观察：视频 0–28.72 s，机器人只从起点行进到第一栏近端，未到达或尝试第3栏；视频 28.72 s 的全景中，第3栏仍在机器人前方远处。评分器 stages_reached=0、metric_hurdles_cleared=0。"
+     }
+    ]
+   },
+   "hurdle_4_cleared": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 28.72,
+      "log_step": null,
+      "text_zh": "观察：视频 0–28.72 s，机器人只从起点行进到第一栏近端，未到达或尝试第4栏；视频 28.72 s 的全景中，第4栏仍在机器人前方远处。评分器 stages_reached=0、metric_hurdles_cleared=0。"
+     }
+    ]
+   },
+   "torso_at_finish": {
+    "method": "visual",
+    "intervals": [],
+    "final": false,
+    "evidence": [
+     {
+      "t": 28.72,
+      "log_step": null,
+      "text_zh": "观察：视频 28.72 s，机器人停留在第一栏处，终点白色立柱仍在画面右端远处，躯干未抵达终点；全程也未完成四栏。评分器 metric_distance_m=13.6、stages_reached=0。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "time_s",
+   "unit": "s",
+   "threshold": 140,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "distance_m",
+   "unit": "m",
+   "threshold": null,
+   "direction": "higher_better",
+   "value": 13.6
+  },
+  {
+   "name": "hurdles_cleared",
+   "unit": "count",
+   "threshold": 4,
+   "direction": "higher_better",
+   "value": 0.0
+  },
+  {
+   "name": "hurdles_down",
+   "unit": "count",
+   "threshold": null,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "hurdle_1_cleared",
+   "t": 28.6,
+   "log_step": 198,
+   "where": "操作中",
+   "evidence_zh": "观察：视频 27.8–28.72 s，机器人在第一栏近端继续贴地迈步，未出现三点腾空越栏，随后视频结束；log step 198 最终保存的 2872 行轨迹收到裁判 went under the top of the first hurdle，评分器确认跨栏数为 0。推断：第一栏接近和起跳衔接未成功，第一栏是首次未修复的失败步骤，视频约 28.6 s 为末段贴地接近的导航锚点。"
+  },
+  "unmet": {
+   "hurdle_1_cleared": false,
+   "hurdle_2_cleared": false,
+   "hurdle_3_cleared": false,
+   "hurdle_4_cleared": false,
+   "torso_at_finish": false
+  },
+  "phenomena": [
+   {
+    "cat": "全身",
+    "item": "步态",
+    "t": 28.6,
+    "log_step": 198,
+    "evidence_zh": "观察：视频 27.8–28.72 s，机器人双脚仍在跑道高度附近交替迈动，身体未完成跨栏腾空，也未翻倒；log step 198 最终保存轨迹的裁判状态为低于第一栏顶部穿过，评分器 fell=0。推断：迈步接近第一栏时未及时形成满足越栏高度的运动，属于步态失败。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "t": 28.6,
+    "log_step": 180,
+    "source_zh": "日志 / 视频",
+    "evidence_zh": "观察：log step 180 的最终生成脚本先以 walk 接近第一栏、再执行 45 个停止行走指令步，之后才安排下蹲、蹬伸和收腿；log step 198 重新执行该脚本保存的最终轨迹在第一栏被裁判判为低过栏顶，最终骨盆高度为 1.007 米。视频 27.8–28.72 s 显示末段仍是贴地迈步，没有脚本预期的完整起跳过程。推断：该控制序列未能在脚部进入第一栏判定平面前衔接出有效的跨栏动作；现有记录不足以细分为制动、起跳触发或关节跟踪问题，根因保留为控制。"
+   }
+  ],
+  "gap_zh": "输入仅有低分辨率场地全景视频，无状态日志、第一栏侧视视频或成功参考视频；放大后仍无法精确确认触发低过栏顶的左右脚身份、穿面高度及有无瞬时碰栏。最终保存日志能核对判负类型，但没有该轨迹逐控制步的姿态及执行分支记录，无法判断失败发生在停止行走的哪一步或刚开始下蹲时；会话后续测试及 agent 自述的收腿饱和、后仰诊断不作为本次评分轨迹的已证实细分根因。"
+ },
+ "summary_zh": "机器人沿跑道走到第一栏，但评分轨迹末段仍在贴地迈步，最终未完成任何跨栏或冲线节点（0/5）。断点为越过第1栏：最终保存日志报告低于栏顶穿过，视频未见有效腾空越栏，判为步态失败及控制衔接不足。全景分辨率低且无状态日志，无法确定触发判负的脚及更细的控制原因，其他会话测试的腾空高度不计入本次进度。",
+ "conf": 6,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "第1栏未越过、其余节点未达成与视频（机器人停在第一栏近端）及评分器（stages_reached=0, 13.6 m）一致；断点、步态现象与控制根因有证据且已承认细分不足。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-31T00:04:24Z",
+  "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-hurdles-freestanding-i00-privileged--4uxwkep/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/embodied-olympics-fun-hurdles-freestanding-i00-privileged--4uxwkep/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "0297baf70184344c4f8681cdce9a66713e381d7f29ed6da03ca2bd82835cac09",
+   "video_sha256": "6be123d35ef4fff66c4240c1e46b6d7be6977efb97466d651c0b270341ae9b87",
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4"
+  }
+ }
+}