Add 106 trajectories (phase 2, remaining runs; batch N03)

phase2/dexhand/assemble_hamburger/trajectories/dexhand-assemble-hamburger-i00-privileged--6givbpb.json

@@ -0,0 +1,250 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-assemble-hamburger-i00-privileged--6givbpb",
+ "taskKey": "dexhand/assemble_hamburger",
+ "dag_sha256": "8fcc440ea79579cd20bf3359bd20d84b2942ce3638aa32282b17b686831db048",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 1,
+   "hand_wuji": 0,
+   "infra_error": 0,
+   "n_actions": 884,
+   "sim_steps": 884,
+   "stages_completed": 1,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.546,
+   "peak_processes": 11,
+   "peak_disk_gb": 0.239,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 6,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "bottom_bun_on_plate": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 27,
+      "text_zh": "观察：inputs/video.json记为kind=none，且无状态日志；评分器reward只报告stages_completed=1，未报告底包的接盘、居中和正立联合判据。log step 27中的base=True与底包坐标来自agent本地测试，不作为评分轨迹的节点终值；依N9保留未知。"
+     }
+    ]
+   },
+   "patty_on_bottom_bun": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 27,
+      "text_zh": "观察：log step 27的agent脚本对肉饼执行接近、下降、闭指和抬升测试，输出884行动作；log step 28保存同一动作数组，评分器n_actions=sim_steps=884、deterministic=1。评分器未报告本层接触和姿态判据，且本次无视频、状态日志，不能从本地测试的肉饼坐标或总分判定该节点终值。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：log step 27记录单只右灵巧手的肉饼抓取测试，log step 28保存相应884行动作；本次没有评分轨迹视频或接触状态日志，无法确认实际接触端、接触方式及是否形成对向夹持，不能照搬脚本的抓取意图。"
+     }
+    ]
+   },
+   "cheese_on_patty": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 24,
+      "text_zh": "观察：log step 24曾在另一次2779步测试中操作芝士，该测试未作为最终动作交付；log step 27生成、log step 28保存的是884行动作。评分器未逐项报告芝士堆叠状态，本次无视频或状态日志，依N9保留终值未知。"
+     }
+    ]
+   },
+   "tomato_on_cheese": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 27,
+      "text_zh": "观察：log step 27的生成脚本列有番茄搬放计划，但测试在肉饼阶段结束，log step 28只保存884行动作；评分器未报告番茄堆叠状态，本次无视频或状态日志，不能用计划或测试输出赋予节点终值。"
+     }
+    ]
+   },
+   "lettuce_on_tomato": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 27,
+      "text_zh": "观察：log step 27的生成脚本列有生菜搬放计划，但测试在肉饼阶段结束，log step 28只保存884行动作；评分器未报告生菜堆叠状态，本次无视频或状态日志，依N9保留终值未知。"
+     }
+    ]
+   },
+   "top_bun_on_lettuce": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 27,
+      "text_zh": "观察：log step 27的生成脚本列有顶包搬放计划，但测试在肉饼阶段结束，log step 28只保存884行动作；评分器未报告顶包堆叠状态，本次无视频或状态日志，依N9保留终值未知。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "max_alignment_error_m",
+   "unit": "m",
+   "threshold": 0.012,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_tilt_rad",
+   "unit": "rad",
+   "threshold": 0.22,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_linear_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.025,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_angular_speed_rad_s",
+   "unit": "rad/s",
+   "threshold": 0.25,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "peak_bend_rad",
+   "unit": "rad",
+   "threshold": 0.65,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "patty_on_bottom_bun",
+   "t": null,
+   "log_step": 27,
+   "where": "操作中",
+   "evidence_zh": "推断：log step 27的生成脚本首先操作肉饼，下降和闭指后在抬升阶段以ingredient_dropped结束，生成884行动作；log step 28直接加载该数组保存，之后log step 29—32仅作诊断、没有再次保存。评分器n_actions=sim_steps=884、deterministic=1，与交付长度一致，故将断点定位于肉饼叠在底包的操作阶段；具体物理过程缺少评分视频或状态日志核对，t保持null。"
+  },
+  "unmet": {
+   "bottom_bun_on_plate": null,
+   "patty_on_bottom_bun": null,
+   "cheese_on_patty": null,
+   "tomato_on_cheese": null,
+   "lettuce_on_tomato": null,
+   "top_bun_on_lettuce": null
+  },
+  "phenomena": [],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "抓取和接触设计",
+    "alt_feasible": "未知",
+    "t": null,
+    "log_step": 30,
+    "source_zh": "日志",
+    "evidence_zh": "推断：会话证据支持接近与夹持的接触几何不适配。log step 27中，agent脚本在下降后即测得肉饼倾斜31.6度，闭指后43.6度，抬升时76.3度并报告ingredient_dropped；这份884行动作于log step 28保存。随后独立诊断log step 30直接枚举出右手食指中节、近节及拇指近节与肉饼的接触，肉饼倾斜33.5度，说明所用接近姿态已让非预期指节干扰肉饼，未建立可靠的搬运接触布局；这些是会话层面的根因证据，不作为评分轨迹的接触或节点值。没有验证可行的替代握法。"
+   }
+  ],
+  "gap_zh": "本次没有评分视频（inputs/video.json为kind=none，来源说明HF为404且无本地视频），也无状态日志。reward的stages_completed=1只是对齐层数，不能逐项决定六个包含层间接触、居中和正立条件的节点；log step 27的本地测试状态及log step 29—32的独立诊断均属于会话证据，依N9不设为节点终值或实际接触。动作数及保存链支持断点推断，但不足以确认评分尝试中曾形成承载抓握还是仅碰歪肉饼，故不标没保持住、没建立接触或碰撞现象，全部视频时刻及区间为null。五项原生指标未出现在评分器reward中，保留null；oracle仅为另一成功运行的参照，不能补足本次证据。"
+ },
+ "summary_zh": "观察：评分器判失败，记录884个动作、stages_completed=1、deterministic=1；这不是六个DAG节点的逐项验收结果，本次六个节点终值均未知。推断：log step 27—28的生成与保存链将断点定位于肉饼叠在底包的操作阶段，会话接触诊断支持抓取和接触设计问题，后续诊断未形成新交付。观察：本次无评分视频和状态日志，无法确认实际抓握、具体失败现象或视频时间；成功参照的开局及结束画面仅用于理解目标外观。",
+ "conf": 5,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "无视频、无状态日志，节点终值依N9保持未知合理；断点（肉饼叠底包，log step 27-28生成并保存884动作）与评分器n_actions=884一致，根因仅作会话证据且标注推断，空phenomena有gap_zh。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": "2026-10-30T22:19:26Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/dexhand-assemble-hamburger-i00-privileged--6givbpb/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-assemble-hamburger-i00-privileged--6givbpb/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "e65979f0a368a667e882cbf28fad1515559da24f15a29c07cb455affed36c44e",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/assemble_hamburger/trajectories/dexhand-assemble-hamburger-i00-privileged--kjbr2eq.json

@@ -0,0 +1,286 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-assemble-hamburger-i00-privileged--kjbr2eq",
+ "taskKey": "dexhand/assemble_hamburger",
+ "dag_sha256": "8fcc440ea79579cd20bf3359bd20d84b2942ce3638aa32282b17b686831db048",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 1,
+   "hand_wuji": 0,
+   "infra_error": 0,
+   "n_actions": 1090,
+   "sim_steps": 1028,
+   "stages_completed": 6,
+   "success": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.946,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.273,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "log",
+  "denominator": 6,
+  "base": null,
+  "peak": null,
+  "final": 6,
+  "nodes": {
+   "bottom_bun_on_plate": {
+    "method": "log",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 466,
+      "text_zh": "观察：agent在log step 465重新生成最终交付文件后，基准官方重放器在log step 466的verdict报告success=true、aligned_layers=6、base_on_plate=true；底包距盘心水平误差为0.005473080米，小于0.012米，全层最大倾斜为0.000157880弧度。log step 467再次官方重放成功，终态哈希一致；reward亦报告success=1、stages_completed=6、sim_steps=1028、deterministic=1。"
+     },
+     {
+      "t": null,
+      "log_step": 467,
+      "text_zh": "推断：最终交付文件的官方验收与评分记录一致，六层全部满足层序、相邻接触、居中和正立条件，故“底包留在餐盘上”终态为真。本次无视频及状态日志，无法确定本节点何时达成、过程中是否短暂失去，intervals保留null。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本次没有视频和状态日志；log step 466、467仅报告验收终态hand_contacts=0，未提供全过程接触记录，无法核实右灵巧手在本节点期间是否及如何接触底包，不把控制脚本的抓持计划当作实际接触。"
+     }
+    ]
+   },
+   "patty_on_bottom_bun": {
+    "method": "log",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 466,
+      "text_zh": "观察：agent在log step 465重新生成最终交付文件后，基准官方重放器在log step 466的verdict报告success=true、aligned_layers=6、ordered=true、adjacent_contacts=5（五个相邻层界面均接触）；肉饼距盘心水平误差为0.011437792米，小于0.012米，全层最大倾斜为0.000157880弧度。log step 467再次官方重放成功，终态哈希一致；reward亦报告success=1、stages_completed=6、sim_steps=1028、deterministic=1。"
+     },
+     {
+      "t": null,
+      "log_step": 467,
+      "text_zh": "推断：最终交付文件的官方验收与评分记录一致，六层全部满足层序、相邻接触、居中和正立条件，故“肉饼叠在底包上”终态为真。本次无视频及状态日志，无法确定本节点何时达成、过程中是否短暂失去，intervals保留null。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本次没有视频和状态日志；log step 466、467仅报告验收终态hand_contacts=0，未提供全过程接触记录，无法核实右灵巧手在本节点期间是否及如何接触肉饼，不把控制脚本的抓持计划当作实际接触。"
+     }
+    ]
+   },
+   "cheese_on_patty": {
+    "method": "log",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 466,
+      "text_zh": "观察：agent在log step 465重新生成最终交付文件后，基准官方重放器在log step 466的verdict报告success=true、aligned_layers=6、ordered=true、adjacent_contacts=5（五个相邻层界面均接触）；芝士距盘心水平误差为0.003434363米，小于0.012米，全层最大倾斜为0.000157880弧度。log step 467再次官方重放成功，终态哈希一致；reward亦报告success=1、stages_completed=6、sim_steps=1028、deterministic=1。"
+     },
+     {
+      "t": null,
+      "log_step": 467,
+      "text_zh": "推断：最终交付文件的官方验收与评分记录一致，六层全部满足层序、相邻接触、居中和正立条件，故“芝士叠在肉饼上”终态为真。本次无视频及状态日志，无法确定本节点何时达成、过程中是否短暂失去，intervals保留null。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本次��有视频和状态日志；log step 466、467仅报告验收终态hand_contacts=0，未提供全过程接触记录，无法核实右灵巧手在本节点期间是否及如何接触芝士，不把控制脚本的抓持计划当作实际接触。"
+     }
+    ]
+   },
+   "tomato_on_cheese": {
+    "method": "log",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 466,
+      "text_zh": "观察：agent在log step 465重新生成最终交付文件后，基准官方重放器在log step 466的verdict报告success=true、aligned_layers=6、ordered=true、adjacent_contacts=5（五个相邻层界面均接触）；番茄距盘心水平误差为0.010757302米，小于0.012米，全层最大倾斜为0.000157880弧度。log step 467再次官方重放成功，终态哈希一致；reward亦报告success=1、stages_completed=6、sim_steps=1028、deterministic=1。"
+     },
+     {
+      "t": null,
+      "log_step": 467,
+      "text_zh": "推断：最终交付文件的官方验收与评分记录一致，六层全部满足层序、相邻接触、居中和正立条件，故“番茄叠在芝士上”终态为真。本次无视频及状态日志，无法确定本节点何时达成、过程中是否短暂失去，intervals保留null。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本次没有视频和状态日志；log step 466、467仅报告验收终态hand_contacts=0，未提供全过程接触记录，无法核实右灵巧手在本节点期间是否及如何接触番茄，不把控制脚本的抓持计划当作实际接触。"
+     }
+    ]
+   },
+   "lettuce_on_tomato": {
+    "method": "log",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 466,
+      "text_zh": "观察：agent在log step 465重新生成最终交付文件后，基准官方重放器在log step 466的verdict报告success=true、aligned_layers=6、ordered=true、adjacent_contacts=5（五个相邻层界面均接触）；生菜距盘心水平误差为0.005856544米，小于0.012米，全层最大倾斜为0.000157880弧度。log step 467再次官方重放成功，终态哈希一致；reward亦报告success=1、stages_completed=6、sim_steps=1028、deterministic=1。"
+     },
+     {
+      "t": null,
+      "log_step": 467,
+      "text_zh": "推断：最终交付文件的官方验收与评分记录一致，六层全部满足层序、相邻接触、居中和正立条件，故“生菜叠在番茄上”终态为真。本次无视频及状态日志，无法确定本节点何时达成、过程中是否短暂失去，intervals保留null。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本次没有视频和状态日志；log step 466、467仅报告验收终态hand_contacts=0，未提供全过程接触记录，无法核实右灵巧手在本节点期间是否及如何接触生菜，不把控制脚本的抓持计划当作实际接触。"
+     }
+    ]
+   },
+   "top_bun_on_lettuce": {
+    "method": "log",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 466,
+      "text_zh": "观察：agent在log step 465重新生成最终交付文件后，基准官方重放器在log step 466的verdict报告success=true、aligned_layers=6、ordered=true、adjacent_contacts=5（五个相邻层界面均接触）；顶包距盘心水平误差为0.009074606米，小于0.012米，全层最大倾斜为0.000157880弧度。log step 467再次官方重放成功，终态哈希一致；reward亦报告success=1、stages_completed=6、sim_steps=1028、deterministic=1。"
+     },
+     {
+      "t": null,
+      "log_step": 467,
+      "text_zh": "推断：最终交付文件的官方验收与评分记录一致，六层全部满足层序、相邻接触、居中和正立条件，故“顶包叠在生菜上”终态为真。本次无视频及状态日志，无法确定本节点何时达成、过程中是否短暂失去，intervals保留null。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本次没有视频和状态日志；log step 466、467仅报告验收终态hand_contacts=0，未提供全过程接触记录，无法核实右灵巧手在本节点期间是否及如何接触顶包，不把控制脚本的抓持计划当作实际接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "max_alignment_error_m",
+   "unit": "m",
+   "threshold": 0.012,
+   "direction": "lower_better",
+   "value": 0.01143779190456854
+  },
+  {
+   "name": "max_tilt_rad",
+   "unit": "rad",
+   "threshold": 0.22,
+   "direction": "lower_better",
+   "value": 0.0001578804794327418
+  },
+  {
+   "name": "max_linear_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.025,
+   "direction": "lower_better",
+   "value": 0.00014243777977838363
+  },
+  {
+   "name": "max_angular_speed_rad_s",
+   "unit": "rad/s",
+   "threshold": 0.25,
+   "direction": "lower_better",
+   "value": 0.002850371997294105
+  },
+  {
+   "name": "peak_bend_rad",
+   "unit": "rad",
+   "threshold": 0.65,
+   "direction": "lower_better",
+   "value": 0.3536007999570926
+  }
+ ],
+ "failure": null,
+ "summary_zh": "观察：评分记录success=1、stages_completed=6；最终文件在log step 465重新生成后，基准官方重放器在log step 466、467均报告六层按配方接触、底包接盘、无手接触并稳定保持100控制步后成功，终态哈希一致，六个节点终态均为真。观察：本文件原生指标取自上述最终交付重放的verdict，未使用log step 460、461中早先文件的数值。观察：本次无视频及状态日志，无法标定节点达成或丢失的时间，也无法核实右灵巧手全过程的实际接触；参考成功视频属于另一次运行，不能补足本次证据。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "无视频、无状态日志；六个节点 final=true 与评分器 success=1、stages_completed=6 及 log step 466/467 官方重放 verdict（aligned_layers=6、各层误差<0.012 m、hand_contacts=0、released_hold_steps=100）一致，数值引用核对无误；intervals 为 null、contact_actual 为未知均符合 N9，未见需要修改之处。",
+    "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": "2026-10-30T22:20:30Z",
+  "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/dexhand-assemble-hamburger-i00-privileged--kjbr2eq/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-assemble-hamburger-i00-privileged--kjbr2eq/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "0aee3fa987b314094ca22e784d11539defa2e3a436cd687673161f3a4db6a298",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/assemble_hamburger/trajectories/dexhand-assemble-hamburger-i00-privileged--mj7gjea.json

@@ -0,0 +1,274 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-assemble-hamburger-i00-privileged--mj7gjea",
+ "taskKey": "dexhand/assemble_hamburger",
+ "dag_sha256": "8fcc440ea79579cd20bf3359bd20d84b2942ce3638aa32282b17b686831db048",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 1,
+   "hand_wuji": 0,
+   "infra_error": 0,
+   "n_actions": 212,
+   "sim_steps": 212,
+   "stages_completed": 1,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.823,
+   "peak_processes": 6,
+   "peak_disk_gb": 0.254,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 6,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "bottom_bun_on_plate": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 66,
+      "text_zh": "推断：inputs/run.json 的 reward.stages_completed=1 报告最终仅一层水平对齐；task/task_dag.json 说明该量不是按配方顺序完成的前缀计数。六个节点各自都要求其对应食材中心到盘心的水平距离不超过12毫米，因此最终成立节点数至多为1（分母6），但任一单独节点为真只需占用这一层，并不超过总量。该字段没有给出对齐食材身份，也未报告“底包留在盘上”所需的接触及正立状态，不能单独决定本节点终值。本次无视频、无状态日志；依docs/03_trajectory.md Step 2和DECISIONS.md N9，final保持null，intervals亦为null。 DAG记载的开局成立只能说明初始目标；S4允许该状态后来失去，不能据此证明终态仍成立。"
+     }
+    ]
+   },
+   "patty_on_bottom_bun": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 66,
+      "text_zh": "推断：inputs/run.json 的 reward.stages_completed=1 报告最终仅一层水平对齐；task/task_dag.json 说明该量不是按配方顺序完成的前缀计数。六个节点各自都要求其对应食材中心到盘心的水平距离不超过12毫米，因此最终成立节点数至多为1（分母6），但任一单独节点为真只需占用这一层，并不超过总量。该字段没有给出对齐食材身份，也未报告“肉饼叠在底包”所需的接触及正立状态，不能单独决定本节点终值。本次无视频、无状态日志；依docs/03_trajectory.md Step 2和DECISIONS.md N9，final保持null，intervals亦为null。"
+     },
+     {
+      "t": null,
+      "log_step": 66,
+      "text_zh": "观察：agent 的末版生成脚本仅打印肉饼处理结果，随后 Session.status() 返回 ingredient_dropped、success=False、steps=212、aligned=1、adj=1，并保存本次动作。肉饼坐标和29.81毫米误差来自脚本调用 features() 的自查；按N9只作会话证据，不用于设置节点终值。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：log step 66 的交付脚本命令单只右灵巧手执行拇指与食指闭合、搬运、张开和撤回以处理肉饼；本次无视频、无状态接触对，无法确认实际是否形成夹持、托举或推压接触。"
+     }
+    ]
+   },
+   "cheese_on_patty": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 66,
+      "text_zh": "推断：inputs/run.json 的 reward.stages_completed=1 报告最终仅一层水平对齐；task/task_dag.json 说明该量不是按配方顺序完成的前缀计数。六个节点各自都要求其对应食材中心到盘心的水平距离不超过12毫米，因此最终成立节点数至多为1（分母6），但任一单独节点为真只需占用这一层，并不超过总量。该字段没有给出对齐食材身份，也未报告“芝士叠在肉饼”所需的接触及正立状态，不能单独决定本节点终值。本次无视频、无状态日志；依docs/03_trajectory.md Step 2和DECISIONS.md N9，final保持null，intervals亦为null。"
+     }
+    ]
+   },
+   "tomato_on_cheese": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 66,
+      "text_zh": "推断：inputs/run.json 的 reward.stages_completed=1 报告最终仅一层水平对齐；task/task_dag.json 说明该量不是按配方顺序完成的前缀计数。六个节点各自都要求其对应食材中心到盘心的水平距离不超过12毫米，因此最终成立节点数至多为1（分母6），但任一单���节点为真只需占用这一层，并不超过总量。该字段没有给出对齐食材身份，也未报告“番茄叠在芝士”所需的接触及正立状态，不能单独决定本节点终值。本次无视频、无状态日志；依docs/03_trajectory.md Step 2和DECISIONS.md N9，final保持null，intervals亦为null。"
+     }
+    ]
+   },
+   "lettuce_on_tomato": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 66,
+      "text_zh": "推断：inputs/run.json 的 reward.stages_completed=1 报告最终仅一层水平对齐；task/task_dag.json 说明该量不是按配方顺序完成的前缀计数。六个节点各自都要求其对应食材中心到盘心的水平距离不超过12毫米，因此最终成立节点数至多为1（分母6），但任一单独节点为真只需占用这一层，并不超过总量。该字段没有给出对齐食材身份，也未报告“生菜叠在番茄”所需的接触及正立状态，不能单独决定本节点终值。本次无视频、无状态日志；依docs/03_trajectory.md Step 2和DECISIONS.md N9，final保持null，intervals亦为null。"
+     }
+    ]
+   },
+   "top_bun_on_lettuce": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 66,
+      "text_zh": "推断：inputs/run.json 的 reward.stages_completed=1 报告最终仅一层水平对齐；task/task_dag.json 说明该量不是按配方顺序完成的前缀计数。六个节点各自都要求其对应食材中心到盘心的水平距离不超过12毫米，因此最终成立节点数至多为1（分母6），但任一单独节点为真只需占用这一层，并不超过总量。该字段没有给出对齐食材身份，也未报告“顶包叠在生菜”所需的接触及正立状态，不能单独决定本节点终值。本次无视频、无状态日志；依docs/03_trajectory.md Step 2和DECISIONS.md N9，final保持null，intervals亦为null。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "max_alignment_error_m",
+   "unit": "m",
+   "threshold": 0.012,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_tilt_rad",
+   "unit": "rad",
+   "threshold": 0.22,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_linear_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.025,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_angular_speed_rad_s",
+   "unit": "rad/s",
+   "threshold": 0.25,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "peak_bend_rad",
+   "unit": "rad",
+   "threshold": 0.65,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "patty_on_bottom_bun",
+   "t": null,
+   "log_step": 66,
+   "where": "操作中",
+   "evidence_zh": "推断：log step 66 的末版脚本按肉饼、芝士、番茄、生菜、顶包循环，肉饼处理后即报告 ingredient_dropped 并因 s.done 退出，随后将动作保存为 trajectory.npz；其 Session.status() 的 steps=212 与评分器 n_actions=212、sim_steps=212 一致，评分器 deterministic=1、infra_error=0，支持该保存对应受评交付。故断点定位在肉饼叠到底包的操作中；未见后续食材处理或修复。无本次视频，无法给出视频时间，也不能确定掉落发生在张开还是撤手阶段。"
+  },
+  "unmet": {
+   "bottom_bun_on_plate": null,
+   "patty_on_bottom_bun": null,
+   "cheese_on_patty": null,
+   "tomato_on_cheese": null,
+   "lettuce_on_tomato": null,
+   "top_bun_on_lettuce": null
+  },
+  "phenomena": [],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "source_zh": "日志",
+    "evidence_zh": "推断：会话测试中，agent 脚本在 log step 42 读到肉饼到达释放高度时约为[-0.1597,0.0015,0.3533]米，释放并撤手后却读到[-0.0372,-0.0120,0.3172]米且 ingredient_dropped；log step 48、51、54 的不同张开和撤手动作也反复伴随肉饼偏移并以同一原因失败。末次保存的 log step 66 仍以 ingredient_dropped、success=False 结束，支持肉饼放置及撤手控制尚未实现稳定交接。上述自查只支持一般控制归因；无受评接触和运动观测，不能进一步断言抓取几何、轨迹跟踪或某一参数是确定根因。"
+   }
+  ],
+  "gap_zh": "本次没有视频（inputs/video.json: kind=none），也没有状态日志；inputs/oracle/replay.mp4属于另一条成功轨迹，不能替代本次观测。依据reward.stages_completed=1与六个节点各自的12毫米居中要求，可推断最终成立节点数至多为1（分母6）；旧标注没有写明这一有效上界，现已补入。但该字段是当前水平对齐层数，不是配方顺序的已完成前缀，也未给出食材身份、接触或正立终值，故不能将其等同于DAG完成1/6。对每个单独节点，其为真所需的一层对齐均未超过���量1，因此总量不能单独将任何一个节点判假。对I1具体赋值建议的分歧按DECISIONS.md N9的Now处理：无视频时仅采用报告节点的评分器或服务结果字段；log step 66的29.81毫米来自agent脚本调用features()的自查，不能代替节点终值。末版脚本未进入后续四层的操作也不等于评分器报告这四层的终态；底包开局成立则不能证明其在操作肉饼后未被带动。六节点final和failure.unmet保持null；无视频时所有intervals保持null。交付脚本和步数可定位处理肉饼时结束，但无法观察实际手物接触、完整落下过程及视频时刻；不能区分没保持住、位姿不准、没松开或放后不稳，故现象层为空。根因仅由会话重复测试支持到一般控制层级，未将测试中的物理事件当作受评轨迹现象；原生五项连续指标未在reward报告，均记null。"
+ },
+ "summary_zh": "评分器判定失败，受评动作共212步；末次保存脚本在处理肉饼叠到底包时以ingredient_dropped结束，后续四层未进入操作，会话多次测试支持一般控制问题。评分器stages_completed=1将最终成立节点数限定为至多1/6，但不报告对齐层身份或完整关系判据；无本次视频、状态日志或逐节点结果，六个节点终值仍未知。审查建议的底包为真及其余五节点为假尚不能由允许的证据确定，也未能逐一排除这些具体终值，故将conf从5降至4。",
+ "conf": 4,
+ "review": {
+  "status": "agreed",
+  "rounds": 2,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "fix",
+    "conf": 5,
+    "summary_zh": "断点（肉饼叠放）与根因层级基本站得住，受评动作数212与评分器n_actions=212吻合；但评分器字段 stages_completed=1 本身给出了节点终值的界，标签却把六个终值全部置 null、unmet 全空，削弱了有支持的标签。",
+    "n_issues": 1,
+    "n_major": 1
+   },
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 2,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "无本次视频与状态日志，标签仅用评分器字段（stages_completed=1、n_actions=212、deterministic=1）和 log step 66 的末版脚本输出。六节点 final/unmet 置 null 符合 N9，已独立核对断点、根因（控制）与空现象层加 gap_zh 的写法。未发现需修改的内容；第一轮提出的节点终值指定仍不能由评分器字段决定（stages_completed 是当下对齐层数，不指明食材身份）。",
+    "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": "2026-10-30T22:30:47Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/dexhand-assemble-hamburger-i00-privileged--mj7gjea/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-assemble-hamburger-i00-privileged--mj7gjea/04_review_r1"
+   },
+   {
+    "stage": "reconcile",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-assemble-hamburger-i00-privileged--mj7gjea/05_reconcile_r1"
+   },
+   {
+    "stage": "traj_review",
+    "round": 2,
+    "ok": true,
+    "space": "work/runs/dexhand-assemble-hamburger-i00-privileged--mj7gjea/04_review_r2"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "2ec7b94e52b655550612d3fdb422a111c7bc86b01f3aa8cfa8afa3afbb04e1c4",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/assemble_hamburger/trajectories/dexhand-assemble-hamburger-i00-privileged--mzjt2v3.json

@@ -0,0 +1,316 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-assemble-hamburger-i00-privileged--mzjt2v3",
+ "taskKey": "dexhand/assemble_hamburger",
+ "dag_sha256": "8fcc440ea79579cd20bf3359bd20d84b2942ce3638aa32282b17b686831db048",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 1,
+   "hand_wuji": 0,
+   "infra_error": 0,
+   "n_actions": 5651,
+   "sim_steps": 5651,
+   "stages_completed": 6,
+   "success": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.752,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.834,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "log",
+  "denominator": 6,
+  "base": null,
+  "peak": null,
+  "final": 6,
+  "nodes": {
+   "bottom_bun_on_plate": {
+    "method": "log",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的正式 reward 报告 success=1、stages_completed=6、deterministic=1、n_actions=5651、sim_steps=5651、infra_error=0；本次 video.kind=none，且无状态日志。"
+     },
+     {
+      "t": null,
+      "log_step": 97,
+      "text_zh": "观察：log step 96 对交付文件启动两次基准重放，log step 97 的重放器结果报告 success=true、stage=6、steps=5651；verdict 中底包水平误差为0.004921657米，aligned_layers=6、ordered=true、adjacent_contacts=5、base_on_plate=true、hand_contacts=0、released_hold_steps=100。"
+     },
+     {
+      "t": null,
+      "log_step": 97,
+      "text_zh": "推断：正式成功判定要求六层同时满足居中、正立、严格层序、相邻接触及底包接盘，故“底包留在盘上”终值为真；基准重放结果支持此判断。没有本次视频，无法给出该节点的达成时刻、途中失去或恢复区间，intervals 保留 null。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：没有本次视频或逐时接触状态日志；log step 97 只报告验收终态手接触为零，不能据此确定操作过程中右灵巧手是否及如何接触底包。底包初始已在盘上，不能把任务典型方案当作本次从未触碰的证据。"
+     }
+    ]
+   },
+   "patty_on_bottom_bun": {
+    "method": "log",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的正式 reward 报告 success=1、stages_completed=6、deterministic=1、n_actions=5651、sim_steps=5651、infra_error=0；本次 video.kind=none，且无状态日志。"
+     },
+     {
+      "t": null,
+      "log_step": 97,
+      "text_zh": "观察：log step 96 对交付文件启动两次基准重放，log step 97 的重放器结果报告 success=true、stage=6、steps=5651；verdict 中肉饼水平误差为0.008205203米，aligned_layers=6、ordered=true、adjacent_contacts=5、base_on_plate=true、hand_contacts=0、released_hold_steps=100。"
+     },
+     {
+      "t": null,
+      "log_step": 97,
+      "text_zh": "推断：正式成功判定要求六层同时满足居中、正立、严格层序、相邻接触及底包接盘，故“肉饼叠在底包”终值为真；基准重放结果支持此判断。没有本次视频，无法给出该节点的达成时刻、途中失去或恢复区间，intervals 保留 null。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：没有本次视频或逐时接触状态日志；log step 97 只报告验收终态手接触为零，不能据此确定操作过程中右灵巧手是否及如何接触肉饼。agent 的夹持脚本及调试接触输出不足以确认交付轨迹中的实际接触方式。"
+     }
+    ]
+   },
+   "cheese_on_patty": {
+    "method": "log",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的正式 reward 报告 success=1、stages_completed=6、deterministic=1、n_actions=5651、sim_steps=5651、infra_error=0；本次 video.kind=none，且无状态日志。"
+     },
+     {
+      "t": null,
+      "log_step": 97,
+      "text_zh": "观察：log step 96 对交付文件启动两次基准重放，log step 97 的重放器结果报告 success=true、stage=6、steps=5651；verdict 中芝士水平误差为0.004510767米，aligned_layers=6、ordered=true、adjacent_contacts=5、base_on_plate=true、hand_contacts=0、released_hold_steps=100。"
+     },
+     {
+      "t": null,
+      "log_step": 97,
+      "text_zh": "��断：正式成功判定要求六层同时满足居中、正立、严格层序、相邻接触及底包接盘，故“芝士叠在肉饼”终值为真；基准重放结果支持此判断。没有本次视频，无法给出该节点的达成时刻、途中失去或恢复区间，intervals 保留 null。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：没有本次视频或逐时接触状态日志；log step 97 只报告验收终态手接触为零，不能据此确定操作过程中右灵巧手是否及如何接触芝士。agent 的夹持脚本及调试接触输出不足以确认交付轨迹中的实际接触方式。"
+     }
+    ]
+   },
+   "tomato_on_cheese": {
+    "method": "log",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的正式 reward 报告 success=1、stages_completed=6、deterministic=1、n_actions=5651、sim_steps=5651、infra_error=0；本次 video.kind=none，且无状态日志。"
+     },
+     {
+      "t": null,
+      "log_step": 97,
+      "text_zh": "观察：log step 96 对交付文件启动两次基准重放，log step 97 的重放器结果报告 success=true、stage=6、steps=5651；verdict 中番茄水平误差为0.007576715米，aligned_layers=6、ordered=true、adjacent_contacts=5、base_on_plate=true、hand_contacts=0、released_hold_steps=100。"
+     },
+     {
+      "t": null,
+      "log_step": 97,
+      "text_zh": "推断：正式成功判定要求六层同时满足居中、正立、严格层序、相邻接触及底包接盘，故“番茄叠在芝士”终值为真；基准重放结果支持此判断。没有本次视频，无法给出该节点的达成时刻、途中失去或恢复区间，intervals 保留 null。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：没有本次视频或逐时接触状态日志；log step 97 只报告验收终态手接触为零，不能据此确定操作过程中右灵巧手是否及如何接触番茄。agent 的夹持脚本及调试接触输出不足以确认交付轨迹中的实际接触方式。"
+     }
+    ]
+   },
+   "lettuce_on_tomato": {
+    "method": "log",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的正式 reward 报告 success=1、stages_completed=6、deterministic=1、n_actions=5651、sim_steps=5651、infra_error=0；本次 video.kind=none，且无状态日志。"
+     },
+     {
+      "t": null,
+      "log_step": 97,
+      "text_zh": "观察：log step 96 对交付文件启动两次基准重放，log step 97 的重放器结果报告 success=true、stage=6、steps=5651；verdict 中生菜水平误差为0.007047983米，aligned_layers=6、ordered=true、adjacent_contacts=5、base_on_plate=true、hand_contacts=0、released_hold_steps=100。"
+     },
+     {
+      "t": null,
+      "log_step": 97,
+      "text_zh": "推断：正式成功判定要求六层同时满足居中、正立、严格层序、相邻接触及底包接盘，故“生菜叠在番茄”终值为真；基准重放结果支持此判断。没有本次视频，无法给出该节点的达成时刻、途中失去或恢复区间，intervals 保留 null。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：没有本次视频或逐时接触状态日志；log step 97 只报告验收终态手接触为零，不能据此确定操作过程中右灵巧手是否及如何接触生菜。agent 的夹持脚本及调试接触输出不足以确认交付轨迹中的实际接触方式。"
+     }
+    ]
+   },
+   "top_bun_on_lettuce": {
+    "method": "log",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的正式 reward 报告 success=1、stages_completed=6、deterministic=1、n_actions=5651、sim_steps=5651、infra_error=0；本次 video.kind=none，且无状态日志。"
+     },
+     {
+      "t": null,
+      "log_step": 97,
+      "text_zh": "观察：log step 96 对交付文件启动两次基准重放，log step 97 的重放器结果报告 success=true、stage=6、steps=5651；verdict 中顶包水平误差为0.005064915米，aligned_layers=6、ordered=true、adjacent_contacts=5、base_on_plate=true、hand_contacts=0、released_hold_steps=100。"
+     },
+     {
+      "t": null,
+      "log_step": 97,
+      "text_zh": "推断：正式成功判定要求六层同时满足居中、正立、严格层序、相邻接触及底包接盘，故“顶包叠在生菜”终值为真；基准重放结果支持此判断。没有本次视频，无法给出该节点的达成时刻、途中失去或恢复区间，intervals 保留 null。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：没有本次视频或逐时接触状态日志；log step 97 只报告验收终态手接触为零，不能据此确定操作过程中右灵巧手是否及如何接触顶包。agent 的夹持脚本及调试接触输出不足以确认交付轨迹中的实际接触方式。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "max_alignment_error_m",
+   "value": 0.008205202966923893,
+   "unit": "m",
+   "threshold": 0.012,
+   "direction": "lower_better"
+  },
+  {
+   "name": "max_tilt_rad",
+   "value": 0.00020608715125897996,
+   "unit": "rad",
+   "threshold": 0.22,
+   "direction": "lower_better"
+  },
+  {
+   "name": "max_linear_speed_m_s",
+   "value": 8.486304601974646e-05,
+   "unit": "m/s",
+   "threshold": 0.025,
+   "direction": "lower_better"
+  },
+  {
+   "name": "max_angular_speed_rad_s",
+   "value": 0.0027366430631092023,
+   "unit": "rad/s",
+   "threshold": 0.25,
+   "direction": "lower_better"
+  },
+  {
+   "name": "peak_bend_rad",
+   "value": 0.1471442733354059,
+   "unit": "rad",
+   "threshold": 0.65,
+   "direction": "lower_better"
+  }
+ ],
+ "failure": null,
+ "summary_zh": "观察：正式评分 success=1、stages_completed=6、deterministic=1，交付轨迹包含5651个动作；log step 96–97 的基准重放器也报告成功，verdict 确认六层居中、层序正确、五组相邻接触、底包接盘及无手接触连续保持100个控制步，原生指标取自该 verdict。推断：六个目标节点终值均为真，故无失败断点；log step 86–91 的早期调试偏差已由后续交付修复，不作为本次失败标签。观察：本次无视频且无状态日志，无法确定节点时间区间和操作中的实际接触方式，全部 intervals 为 null、contact_actual 为未知；提供的 oracle 视频属于另一成功运行，不能填补本次观测缺口。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "本次运行无视频、无状态日志；六个节点终值为真由正式 reward（success=1、stages_completed=6）和 log step 97 的重放器 verdict 支持，数值与原文一致；intervals 为 null、contact_actual 为未知、无失败断点均合理，未发现需要修改的错误。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T22:21:38Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/dexhand-assemble-hamburger-i00-privileged--mzjt2v3/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-assemble-hamburger-i00-privileged--mzjt2v3/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "c232f89ba494fa7d264f188bae5481cb6f33cf7be8d1737cbc4da0da674c9d6c",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/batter_whisk/trajectories/dexhand-batter-whisk-i00-privileged--lxshogx.json

@@ -0,0 +1,220 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-batter-whisk-i00-privileged--lxshogx",
+ "taskKey": "dexhand/batter_whisk",
+ "dag_sha256": "6e00e241e1946ef6b1177646e941187a35e0a039cdd0f28d568b3a649d569de9",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 2.564,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.254,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 4,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "whisk_picked_up": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 54,
+      "text_zh": "观察：agent 的会话脚本在 log step 54 重置并运行至本地 stage 2，保存内存快照；这不是已交付轨迹的节点结果。评分器 reward 为 missing_trajectory=1、n_actions=0、sim_steps=0；本次无评分视频或状态日志，也无该节点的评分结果，故 final 与 intervals 均未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付评分轨迹；会话测试中的机器人接触不作为本次评分尝试的实际接触。"
+     }
+    ]
+   },
+   "batter_whisked": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 56,
+      "text_zh": "观察：agent 的会话脚本在 log step 56 恢复内存快照后报告 stage 3、revolutions=5.020；这不是已交付轨迹的节点结果。评分器 reward 为 missing_trajectory=1、n_actions=0、sim_steps=0；本次无评分视频或状态日志，也无该节点的评分结果，故 final 与 intervals 均未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付评分轨迹；会话测试中的机器人接触不作为本次评分尝试的实际接触。"
+     }
+    ]
+   },
+   "whisk_drained": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 57,
+      "text_zh": "观察：agent 的会话脚本在 log step 57 恢复内存快照后报告 stage 4、drain_steps=25；这不是已交付轨迹的节点结果。评分器 reward 为 missing_trajectory=1、n_actions=0、sim_steps=0；本次无评分视频或状态日志，也无该节点的评分结果，故 final 与 intervals 均未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付评分轨迹；会话测试中的机器人接触不作为本次评分尝试的实际接触。"
+     }
+    ]
+   },
+   "whisk_on_counter": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 57,
+      "text_zh": "观察：agent 的会话脚本在 log step 57 报告本地测试 stage 5、hold_steps=50、success=True；这不是已交付轨迹的最终验收结果。评分器 reward 为 missing_trajectory=1、n_actions=0、sim_steps=0；本次无评分视频或状态日志，也无该节点的评分结果，故 final 与 intervals 均未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付评分轨迹；会话测试中的机器人接触不作为本次评分尝试的实际接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "revolutions",
+   "unit": "rev",
+   "threshold": 5,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "max_bowl_tilt_deg",
+   "unit": "deg",
+   "threshold": 10,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_bowl_slide_m",
+   "unit": "m",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 58,
+   "where": "首步前",
+   "evidence_zh": "观察：log step 58 的末条回复为供应商提示“This request was blocked by Gemini's filters”；评分器 reward 报告 missing_trajectory=1、n_actions=0、sim_steps=0。log step 57 的 success=True 来自 agent 恢复快照后的会话测试，未见交付保存或重放验证，无法定位评分轨迹中的物理断点。"
+  },
+  "unmet": {
+   "whisk_picked_up": null,
+   "batter_whisked": null,
+   "whisk_drained": null,
+   "whisk_on_counter": null
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：inputs/video.json 为 kind=none，未提供本次评分视频或状态日志；reward 未报告各节点结果及三个原生指标。log step 54、56、57 是 agent 重置或恢复快照后的本地测试，log step 58 随后出现 Gemini 过滤中断，评分器确认缺少交付轨迹。推断：这是交付前的外部因素中断，应重试并排除出机器人失败归因；保留目录中的 success=0，但不将会话成功或会话中的物理失败迁入评分节点，失败现象与根因均留空。"
+ },
+ "summary_zh": "观察：评分器报告缺少交付轨迹且动作数为零，本次无评分视频或状态日志，四个节点的评分终值均未知。观察：agent 在 log step 57 的恢复快照测试中报告五阶段成功，随后 log step 58 出现 Gemini 过滤中断，未见交付保存及重放验收。推断：本次属于交付前外部中断，不能归因于机器人操作失败，需重试并排除出失败分析。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "无视频/状态日志；评分器 missing_trajectory=1、n_actions=0、sim_steps=0，log step 58 为 Gemini 过滤中断，交付前被外部因素切断。标注将四节点 final 设为未知、断点 ref=unknown、现象与根因留空并写 gap_zh，符合规则，未把会话测试结果迁入评分节点。",
+    "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": "2026-10-30T22:17:10Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/dexhand-batter-whisk-i00-privileged--lxshogx/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-batter-whisk-i00-privileged--lxshogx/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "a676baec36ebe194382c07f67ca9726351f7245ad24badd59b64a2e45430ad47",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/batter_whisk/trajectories/dexhand-batter-whisk-i00-privileged--mzbxasw.json

@@ -0,0 +1,220 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-batter-whisk-i00-privileged--mzbxasw",
+ "taskKey": "dexhand/batter_whisk",
+ "dag_sha256": "6e00e241e1946ef6b1177646e941187a35e0a039cdd0f28d568b3a649d569de9",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 1.259,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.239,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 4,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "whisk_picked_up": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 20,
+      "text_zh": "观察：评分器 reward 报告 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“拿起打蛋器”的节点验收结果；log step 20 为响应超过32000输出 token 上限的 API 错误。无本次评分视频或状态日志，无法确定此节点终值；log step 11–12 的测试脚本及其输出仅属于会话探索。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告 missing_trajectory=1；没有交付轨迹，此节点无可标注的评分尝试接触。"
+     }
+    ]
+   },
+   "batter_whisked": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 20,
+      "text_zh": "观察：评分器 reward 报告 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“搅拌面糊”的节点验收结果；log step 20 为响应超过32000输出 token 上限的 API 错误。无本次评分视频或状态日志，无法确定此节点终值；log step 11–12 的测试脚本及其输出仅属于会话探索。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告 missing_trajectory=1；没有交付轨迹，此节点无可标注的评分尝试接触。"
+     }
+    ]
+   },
+   "whisk_drained": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 20,
+      "text_zh": "观察：评分器 reward 报告 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“打蛋器沥液”的节点验收结果；log step 20 为响应超过32000输出 token 上限的 API 错误。无本次评分视频或状态日志，无法确定此节点终值；log step 11–12 的测试脚本及其输出仅属于会话探索。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告 missing_trajectory=1；没有交付轨迹，此节点无可标注的评分尝试接触。"
+     }
+    ]
+   },
+   "whisk_on_counter": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 20,
+      "text_zh": "观察：评分器 reward 报告 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“打蛋器放台面”的节点验收结果；log step 20 为响应超过32000输出 token 上限的 API 错误。无本次评分视频或状态日志，无法确定此节点终值；log step 11–12 的测试脚本及其输出仅属于会话探索。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告 missing_trajectory=1；没有交付轨迹，此节点无可标注的评分尝试接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "revolutions",
+   "unit": "rev",
+   "threshold": 5,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "max_bowl_tilt_deg",
+   "unit": "deg",
+   "threshold": 10,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_bowl_slide_m",
+   "unit": "m",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 20,
+   "where": "首步前",
+   "evidence_zh": "观察：log step 11–12 仅创建并运行手部几何及闭合探索脚本；log step 13、15、17、19 均无工具调用或可见输出，log step 20 报告“API Error: Claude's response exceeded the 32000 output token maximum”。评分器 reward 为 missing_trajectory=1、n_actions=0、sim_steps=0；没有交付后的评分动作可定位到物理失败节点。"
+  },
+  "unmet": {
+   "whisk_picked_up": null,
+   "batter_whisked": null,
+   "whisk_drained": null,
+   "whisk_on_counter": null
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：inputs/video.json 为 kind=none，inputs/ 无本次 replay.mp4 或 states.jsonl，且 skeleton.state=null；四个节点均无评分器单项验收字段，故 intervals 和 final 保留 null。log step 20 明确记录模型响应超出输出上限的 API 中断，评分器确认未交付轨迹；按外因中断规则不标“没尝试”，现象和机器人根因均留空。log step 11–12 的闭手测试不是交付轨迹，不能据此标注抓握或节点完成；inputs/oracle/replay.mp4 属于另一成功运行，不能证明本次进展。"
+ },
+ "summary_zh": "观察：本次目录评分为 success=0，但评分器同时报告 missing_trajectory=1、n_actions=0、sim_steps=0，没有可供标注的交付轨迹；四个任务节点的终值均未知。观察：会话只记录了手部探索测试（log step 11–12），末尾因响应超过输出上限而发生 API 中断（log step 20），应按外因中断处理，断点为 unknown／首步前，现象与根因留空。观察：本次无视频、无状态日志或节点验收结果，不能把会话测试或参考成功视频作为本次进度。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "标注符合外因中断规则：评分器 missing_trajectory=1、n_actions=0，log step 20 为 API 输出上限错误，未交付轨迹；节点终值 unknown、断点 unknown/首步前、现象与根因留空并有 gap_zh，均合理。validator 通过。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T22:17:04Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/dexhand-batter-whisk-i00-privileged--mzbxasw/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-batter-whisk-i00-privileged--mzbxasw/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "a1158064686761b69554692c05437f3e69ca8134bbb78900788790353eced2d1",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/bicycle_pump_stroke/trajectories/dexhand-bicycle-pump-stroke-i00-privileged--a3preup.json

@@ -0,0 +1,768 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-bicycle-pump-stroke-i00-privileged--a3preup",
+ "taskKey": "dexhand/bicycle_pump_stroke",
+ "dag_sha256": "1e81e1d6887923dbaed96f9b3045c0eecce1c787fce830bd5914fcd0006f269f",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "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": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 30,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "head_connected": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“泵头接上气门”的节点结果；inputs/video.json 的 kind=none，亦无本次状态日志。依 N9，此节点的终值与达成时间无法判断，会话内的初态几何检查不作节点证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1；本次未交付受评动作轨迹，无实际接触可标注。"
+     }
+    ]
+   },
+   "pump_stroke_1": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第1次泵压”的节点结果；inputs/video.json 的 kind=none，亦无本次状态日志。依 N9，此节点的终值与达成时间无法判断，会话内的初态几何检查不作节点证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1；本次未交付受评动作轨迹，无实际接触可标注。"
+     }
+    ]
+   },
+   "pump_stroke_2": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第2次泵压”的节点结果；inputs/video.json 的 kind=none，亦无本次状态日志。依 N9，此节点的终值与达成时间无法判断，会话内的初态几何检查不作节点证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1；本次未交付受评动作轨迹，无实际接触可标注。"
+     }
+    ]
+   },
+   "pump_stroke_3": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第3次泵压”的节点结果；inputs/video.json 的 kind=none，亦无本次状态日志。依 N9，此节点的终值与达成时间无法判断，会话内的初态几何检查不作节点证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1；本次未交付受评动作轨迹，无实际接触可标注。"
+     }
+    ]
+   },
+   "pump_stroke_4": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第4次泵压”的节点结果；inputs/video.json 的 kind=none，亦无本次状态日志。依 N9，此节点的终值与达成时间无法判断，会话内的初态几何检查不作节点证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1；本次未交付受评动作轨迹，无实际接触可标注。"
+     }
+    ]
+   },
+   "pump_stroke_5": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第5次泵压”的节点结果；inputs/video.json 的 kind=none，亦无本次状态日志。依 N9，此节点的终值与达成时间无法判断，会话内的初态几何检查不作节点证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1；本次未交付受评动作轨迹，无实际接触可标注。"
+     }
+    ]
+   },
+   "pump_stroke_6": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第6次泵压”的节点结果；inputs/video.json 的 kind=none，亦无本次状态日志。依 N9，此节点的终值与达成时间无法判断，会话内的初态几何检查不作节点证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1；本次未交付受评动作轨迹，无实际接触可标注。"
+     }
+    ]
+   },
+   "pump_stroke_7": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第7次泵压”的节点结果；inputs/video.json 的 kind=none，亦无本次状态日志。依 N9，此节点的终值与达成时间无法判断，会话内的初态几何检查不作节点证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1；本次未交付受评动作轨迹，无实际接触可标注。"
+     }
+    ]
+   },
+   "pump_stroke_8": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第8次泵压”的节点结果；inputs/video.json 的 kind=none，亦无本次状态日志。依 N9，此节点的终值与达成时间无法判断，会话内的初态几何检查不作节点证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1；本次未交付受评动作轨迹，无实际接触可标注。"
+     }
+    ]
+   },
+   "pump_stroke_9": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第9次泵压”的节点结果；inputs/video.json 的 kind=none，亦无本次状态日志。依 N9，此节点的终值与达成时间无法判断，会话内的初态几何检查不作节点证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1；本次未交付受评动作轨迹，无实际接触可标注。"
+     }
+    ]
+   },
+   "pump_stroke_10": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第10次泵压”的节点结果；inputs/video.json 的 kind=none，亦无本次状态日志。依 N9，此节点的终值与达成时间无法判断，会话内的初态几何检查不作节点证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1；本次未交付受评动作轨迹，无实际接触可标注。"
+     }
+    ]
+   },
+   "pump_stroke_11": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第11次泵压”的节点结果；inputs/video.json 的 kind=none，亦无本次状态日志。依 N9，此节点的终值与达成时间无法判断，会话内的初态几何检查不作节点证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1；本次未交付受评动作轨迹，无实际接触可标注。"
+     }
+    ]
+   },
+   "pump_stroke_12": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第12次泵压”的节点结果；inputs/video.json 的 kind=none，亦无本次状态日志。依 N9，此节点的终值与达成时间无法判断，会话内的初态几何检查不作节点证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1；本次未交付受评动作轨迹，无实际接触可标注。"
+     }
+    ]
+   },
+   "pump_stroke_13": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第13次泵压”的节点结果；inputs/video.json 的 kind=none，亦无本次状态日志。依 N9，此节点的终值与达成时间无法判断，会话内的初态几何检查不作节点证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1；本次未交付受评动作轨迹，无实际接触可标注。"
+     }
+    ]
+   },
+   "pump_stroke_14": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第14次泵压”的节点结果；inputs/video.json 的 kind=none，亦无本次状态日志。依 N9，此节点的终值与达成时间无法判断，会话内的初态几何检查不作节点证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1；本次未交付受评动作轨迹，无实际接触可标注。"
+     }
+    ]
+   },
+   "pump_stroke_15": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第15次泵压”的节点结果；inputs/video.json 的 kind=none，亦无本次状态日志。依 N9，此节点的终值与达成时间无法判断，会话内的初态几何检查不作节点证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1；本次未交付受评动作轨迹，无实际接触可标注。"
+     }
+    ]
+   },
+   "pump_stroke_16": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第16次泵压”的节点结果；inputs/video.json 的 kind=none，亦无本次状态日志。依 N9，此节点的终值与达成时间无法判断，会话内的初态几何检查不作节点证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1；本次未交付受评动作轨迹，无实际接触可标注。"
+     }
+    ]
+   },
+   "pump_stroke_17": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第17次泵压”的节点结果；inputs/video.json 的 kind=none，亦无本次状态日志。依 N9，此节点的终值与达成时间无法判断，会话内的初态几何检查不作节点证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1；本次未交付受评动作轨迹，无实际接触可标注。"
+     }
+    ]
+   },
+   "pump_stroke_18": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第18次泵压”的节点结果；inputs/video.json 的 kind=none，亦无本次状态日志。依 N9，此节点的终值与达成时间无法判断，会话内的初态几何检查不作节点证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1；本次未交付受评动作轨迹，���实际接触可标注。"
+     }
+    ]
+   },
+   "pump_stroke_19": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第19次泵压”的节点结果；inputs/video.json 的 kind=none，亦无本次状态日志。依 N9，此节点的终值与达成时间无法判断，会话内的初态几何检查不作节点证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1；本次未交付受评动作轨迹，无实际接触可标注。"
+     }
+    ]
+   },
+   "pump_stroke_20": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第20次泵压”的节点结果；inputs/video.json 的 kind=none，亦无本次状态日志。依 N9，此节点的终值与达成时间无法判断，会话内的初态几何检查不作节点证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1；本次未交付受评动作轨迹，无实际接触可标注。"
+     }
+    ]
+   },
+   "pump_stroke_21": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第21次泵压”的节点结果；inputs/video.json 的 kind=none，亦无本次状态日志。依 N9，此节点的终值与达成时间无法判断，会话内的初态几何检查不作节点证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1；本次未交付受评动作轨迹，无实际接触可标注。"
+     }
+    ]
+   },
+   "pump_stroke_22": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第22次泵压”的节点结果；inputs/video.json 的 kind=none，亦无本次状态日志。依 N9，此节点的终值与达成时间无法判断，会话内的初态几何检查不作节点证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1；本次未交付受评动作轨迹，无实际接触可标注。"
+     }
+    ]
+   },
+   "pump_stroke_23": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第23次泵压”的节点结果；inputs/video.json 的 kind=none，亦无本次状态日志。依 N9，此节点的终值与达成时间无法判断，会话内的初态几何检查不作节点证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1；本次未交付受评动作轨迹，无实际接触可标注。"
+     }
+    ]
+   },
+   "pump_stroke_24": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第24次泵压”的节点结果；inputs/video.json 的 kind=none，亦无本次状态日志。依 N9，此节点的终值与达成时间无法判断，会话内的初态几何检查不作节点证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1；本次未交付受评动作轨迹，无实际接触可标注。"
+     }
+    ]
+   },
+   "pump_stroke_25": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第25次泵压”的节点结果；inputs/video.json 的 kind=none，亦无本次状态日志。依 N9，此节点的终值与达成时间无法判断，会话内的初态几何检查不作节点证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1；本次未交付受评动作轨迹，无实际接触可标注。"
+     }
+    ]
+   },
+   "pump_stroke_26": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第26次泵压”的节点结果；inputs/video.json 的 kind=none，亦无本次状态日志。依 N9，此节点的终值与达成时间无法判断，会话内的初态几何检查不作节点证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1；本次未交付受评动作轨迹，无实际接触可标注。"
+     }
+    ]
+   },
+   "pump_stroke_27": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第27次泵压”的节点结果；inputs/video.json 的 kind=none，亦无本次状态日志。依 N9，此节点的终值与达成时间无法判断，会话内的初态几何检查不作节点证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1；本次未交付受评动作轨迹，无实际接触可标注。"
+     }
+    ]
+   },
+   "pump_stroke_28": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第28次泵压”的节点结果；inputs/video.json 的 kind=none，亦无本次状态日志。依 N9，此节点的终值与达成时间无法判断，会话内的初态几何检查不作节点证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1；本次未交付受评动作轨迹，无实际接触可标注。"
+     }
+    ]
+   },
+   "head_disconnected": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“泵头脱离气门”的节点结果；inputs/video.json 的 kind=none，亦无本次状态日志。依 N9，此节点的终值与达成时间无法判断，会话内的初态几何检查不作节点证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1；本次未交付受评动作轨迹，无实际接触可标注。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "tire_gauge_pa",
+   "unit": "Pa",
+   "threshold": 210600,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "piston_down_depth_m",
+   "unit": "m",
+   "threshold": 0.18,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "piston_return_depth_m",
+   "unit": "m",
+   "threshold": 0.04,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "piston_release_depth_m",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "pump_release_tilt_rad",
+   "unit": "rad",
+   "threshold": 0.0873,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "pump_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "handle_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 18,
+   "where": "首步前",
+   "evidence_zh": "观察：评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0、success=0、infra_error=0。agent 在 log step 9–12 创建会话并检查复位几何，log step 11 的包围盒检查报 AttributeError 后在 log step 12 修正；最后一次工具调用 log step 17 仍仅检查辐条与桌面几何，log step 18 的 say、calls、out 均为空，无最终报告或轨迹保存记录。推断：受评轨迹未交付，失败位置为首步前；缺少可定位物理失败步骤的受评证据，ref 保留 unknown。"
+  },
+  "unmet": {
+   "head_connected": null,
+   "pump_stroke_1": null,
+   "pump_stroke_2": null,
+   "pump_stroke_3": null,
+   "pump_stroke_4": null,
+   "pump_stroke_5": null,
+   "pump_stroke_6": null,
+   "pump_stroke_7": null,
+   "pump_stroke_8": null,
+   "pump_stroke_9": null,
+   "pump_stroke_10": null,
+   "pump_stroke_11": null,
+   "pump_stroke_12": null,
+   "pump_stroke_13": null,
+   "pump_stroke_14": null,
+   "pump_stroke_15": null,
+   "pump_stroke_16": null,
+   "pump_stroke_17": null,
+   "pump_stroke_18": null,
+   "pump_stroke_19": null,
+   "pump_stroke_20": null,
+   "pump_stroke_21": null,
+   "pump_stroke_22": null,
+   "pump_stroke_23": null,
+   "pump_stroke_24": null,
+   "pump_stroke_25": null,
+   "pump_stroke_26": null,
+   "pump_stroke_27": null,
+   "pump_stroke_28": null,
+   "head_disconnected": null
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：本次无视频、无状态日志，评分器只报告缺轨迹和零动作，未报告接通标志、有效泵压次数、拆开标志或任何原生指标；inputs/oracle/ 的视频属于另一成功运行，不能证明本次进度。日志中部分工具输出与任务文本本身带省略标记，最后 log step 18 无最终报告；日志可见调用均为源码或复位几何检查，不能据此认定物理接触失败。推断：run.json 的 wall_s=3603 与 time_limit_s=3600 相符，支持预算结束前仍无交付，但不能把耗尽预算或缺交付直接归因为规划、控制或交付操作出错；log step 11 的检查脚本错误已在 log step 12 修正，不足以解释最终无交付。依 D23，现象和根因均留空；依 N9，全部节点终值保持未知，所有视频时间为 null。"
+ },
+ "summary_zh": "评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，本次未交付受评动作轨迹，30个语义节点均无可确认的终值。会话停留在源码和复位几何检查，最后一次工具调用为 log step 17，log step 18 无最终报告，故断点记 unknown／首步前。无本次视频、状态日志或逐节点评分，无法确认接通、泵压、拆开及物理失败原因，按 D23 留空现象和根因。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "本次无视频、无状态日志，评分器仅报 missing_trajectory=1、n_actions=0；run.json 日志显示 agent 只做源码与复位几何检查（log step 9–17），后几步为空输出（每步 32000 输出 token，无工具调用），从未提交动作。标注按 D23(a) 留空现象和根因并写 gap_zh，节点全 unknown，符合规则；未见需要修改之处。",
+    "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": "2026-10-30T22:18:31Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/dexhand-bicycle-pump-stroke-i00-privileged--a3preup/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-bicycle-pump-stroke-i00-privileged--a3preup/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "bcfc1b1949f45b8c8fde0af5c4ee949f7a87f061067ff519ef06151424a7903f",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/bicycle_pump_stroke/trajectories/dexhand-bicycle-pump-stroke-i00-privileged--qkoxr8f.json

@@ -0,0 +1,768 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-bicycle-pump-stroke-i00-privileged--qkoxr8f",
+ "taskKey": "dexhand/bicycle_pump_stroke",
+ "dag_sha256": "1e81e1d6887923dbaed96f9b3045c0eecce1c787fce830bd5914fcd0006f269f",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 2.361,
+   "peak_processes": 11,
+   "peak_disk_gb": 0.241,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 30,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "head_connected": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 98,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“泵头接上气门”的节点结果；agent 在 log step 98 明确自述未生成有效交付文件。本次无视频和状态日志，调试测试的计数或状态不能作为该节点评分终值，故保持未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付轨迹；会话调试中的右手接触不属于评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "pump_stroke_1": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 98,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第1次泵压”的节点结果；agent 在 log step 98 明确自述未生成有效交付文件。本次无视频和状态日志，调试测试的计数或状态不能作为该节点评分终值，故保持未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付轨迹；会话调试中的右手接触不属于评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "pump_stroke_2": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 98,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第2次泵压”的节点结果；agent 在 log step 98 明确自述未生成有效交付文件。本次无视频和状态日志，调试测试的计数或状态不能作为该节点评分终值，故保持未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付轨迹；会话调试中的右手接触不属于评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "pump_stroke_3": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 98,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第3次泵压”的节点结果；agent 在 log step 98 明确自述未生成有效交付文件。本次无视频和状态日志，调试测试的计数或状态不能作为该节点评分终值，故保持未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付轨迹；会话调试中的右手接触不属于评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "pump_stroke_4": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 98,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第4次泵压”的节点结果；agent 在 log step 98 明确自述未生成有效交付文件。本次无视频和状态日志，调试测试的计数或状态不能作为该节点评分终值，故保持未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付轨迹；会话调试中的右手接触不属于评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "pump_stroke_5": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 98,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第5次泵压”的节点结果；agent 在 log step 98 明确自述未生成有效交付文件。本次无视频和状态日志，调试测试的计数或状态不能作为该节点评分终值，故保持未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付轨迹；会话调试中的右手接触不属于评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "pump_stroke_6": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 98,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第6次泵压”的节点结果；agent 在 log step 98 明确自述未生成有效交付文件。本次无视频和状态日志，调试测试的计数或状态不能作为该节点评分终值，故保持未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付轨迹；会话调试中的右手接触不属于评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "pump_stroke_7": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 98,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第7次泵压”的节点结果；agent 在 log step 98 明确自述未生成有效交付文件。本次无视频和状态日志，调试测试的计数或状态不能作为该节点评分终值，故保持未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付轨迹；会话调试中的右手接触不属于评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "pump_stroke_8": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 98,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第8次泵压”的节点结果；agent 在 log step 98 明确自述未生成有效交付文件。本次无视频和状态日志，调试测试的计数或状态不能作为该节点评分终值，故保持未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付轨迹；会话调试中的右手接触不属于评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "pump_stroke_9": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 98,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第9次泵压”的节点结果；agent 在 log step 98 明确自述未生成有效交付文件。本次无视频和状态日志，调试测试的计数或状态不能作为该节点评分终值，故保持未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付轨迹；会话调试中的右手接触不属于评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "pump_stroke_10": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 98,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第10次泵压”的节点结果；agent 在 log step 98 明确自述未生成有效交付文件。本次无视频和状态日志，调试测试的计数或状态不能作为该节点评分终值，故保持未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付轨迹；会话调试中的右手接触不属于评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "pump_stroke_11": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 98,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第11次泵压”的节点结果；agent 在 log step 98 明确自述未生成有效交付文件。本次无视频和状态日志，调试测试的计数或状态不能作为该节点评分终值，故保持未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付轨迹；会话调试中的右手接触不属于评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "pump_stroke_12": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 98,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第12次泵压”的节点结果；agent 在 log step 98 明确自述未生成有效交付文件。本次无视频和状态日志，调试测试的计数或状态不能作为该节点评分终值，故保持未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付轨迹；会话调试中的右手接触不属于评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "pump_stroke_13": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 98,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第13次泵压”的节点结果；agent 在 log step 98 明确自述未生成有效交付文件。本次无视频和状态日志，调试测试的计数或状态不能作为该节点评分终值，故保持未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付轨迹；会话调试中的右手接触不属于评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "pump_stroke_14": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 98,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第14次泵压”的节点结果；agent 在 log step 98 明确自述未生成有效交付文件。本次无视频和状态日志，调试测试的计数或状态不能作为该节点评分终值，故保持未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付轨迹；会话调试中的右手接触不属于评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "pump_stroke_15": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 98,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第15次泵压”的节点结果；agent 在 log step 98 明确自述未生成有效交付文件。本次无视频和状态日志，调试测试的计数或状态不能作为该节点评分终值，故保持未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付轨迹；会话调试中的右手接触不属于评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "pump_stroke_16": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 98,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第16次泵压”的节点结果；agent 在 log step 98 明确自述未生成有效交付文件。本次无视频和状态日志，调试测试的计数或状态不能作为该节点评分终值，故保持未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付轨迹；会话调试中的右手接触不属于评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "pump_stroke_17": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 98,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第17次泵压”的节点结果；agent 在 log step 98 明确自述未生成有效交付文件。本次无视频和状态日志，调试测试的计数或状态不能作为该节点评分终值，故保持未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付轨迹；会话调试中的右手接触不属于评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "pump_stroke_18": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 98,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第18次泵压”的节点结果；agent 在 log step 98 明确自述未生成有效交付文件。本次无视频和状态日志，调试测试的计数或状态不能作为该节点评分终值，故保持未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付轨迹；会话调试中的右手接触不属于评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "pump_stroke_19": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 98,
+      "text_zh": "观察：inputs/run.json 的评���器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第19次泵压”的节点结果；agent 在 log step 98 明确自述未生成有效交付文件。本次无视频和状态日志，调试测试的计数或状态不能作为该节点评分终值，故保持未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付轨迹；会话调试中的右手接触不属于评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "pump_stroke_20": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 98,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第20次泵压”的节点结果；agent 在 log step 98 明确自述未生成有效交付文件。本次无视频和状态日志，调试测试的计数或状态不能作为该节点评分终值，故保持未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付轨迹；会话调试中的右手接触不属于评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "pump_stroke_21": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 98,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第21次泵压”的节点结果；agent 在 log step 98 明确自述未生成有效交付文件。本次无视频和状态日志，调试测试的计数或状态不能作为该节点评分终值，故保持未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付轨迹；会话调试中的右手接触不属于评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "pump_stroke_22": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 98,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第22次泵压”的节点结果；agent 在 log step 98 明确自述未生成有效交付文件。本次无视频和状态日志，调试测试的计数或状态不能作为该节点评分终值，故保持未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付轨迹；会话调试中的右手接触不属于评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "pump_stroke_23": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 98,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第23次泵压”的节点结果；agent 在 log step 98 明确自述未生成有效交付文件。本次无视频和状态日志，调试测试的计数或状态不能作为该节点评分终值，故保持未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付轨迹；会话调试中的右手接触不属于评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "pump_stroke_24": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 98,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第24次泵压”的节点结果；agent 在 log step 98 明确自述未生成有效交付文件。本次无视频和状态日志，调试测试的计数或状态不能作为该节点评分终值，故保持未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付轨迹；会话调试中的右手接触不属于评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "pump_stroke_25": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 98,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第25次泵压”的节点结果；agent 在 log step 98 明确自述未生成有效交付文件。本次无视频和状态日志，调试测试的计数或状态不能作为该节点评分终值，故保持未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付轨迹；会话调试中的右手接触不属于评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "pump_stroke_26": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 98,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第26次泵压”的节点结果；agent 在 log step 98 明确自述未生成有效交付文件。本次无视频和状态日志，调试测试的计数或状态不能作为该节点评分终值，故保持未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付轨迹；会话调试中的右手接触不属于评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "pump_stroke_27": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 98,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第27次泵压”的节点结果；agent 在 log step 98 明确自述未生成有效交付文件。本次无视频和状态日志，调试测试的计数或状态不能作为该节点评分终值，故保持未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付轨迹；会话调试中的右手接触不属于评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "pump_stroke_28": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 98,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第28次泵压”的节点结果；agent 在 log step 98 明确自述未生成有效交付文件。本次无视频和状态日志，调试测试的计数或状态不能作为该节点评分终值，故保持未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付轨迹；会话调试中的右手接触不属于评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "head_disconnected": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 98,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“泵头脱离气门”的节点结果；agent 在 log step 98 明确自述未生成有效交付文件。本次无视频和状态日志，调试测试的计数或状态不能作为该节点评分终值，故保持未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付轨迹；会话调试中的右手接触不属于评分轨迹的实际接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "tire_gauge_pa",
+   "unit": "Pa",
+   "threshold": 210600,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "piston_down_depth_m",
+   "unit": "m",
+   "threshold": 0.18,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "piston_return_depth_m",
+   "unit": "m",
+   "threshold": 0.04,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "piston_release_depth_m",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "pump_release_tilt_rad",
+   "unit": "rad",
+   "threshold": 0.0873,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "pump_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "handle_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 98,
+   "where": "首步前",
+   "evidence_zh": "观察：评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0、infra_error=0；log step 68、81 的生成器代码仅在测试成功后调用轨迹保存，log step 93—95 的测试分别以控制步预算耗尽、带锁退头、泵体倾倒结束，agent 在 log step 98 明确自述没有生成有效交付文件。因此没有评分轨迹可用于定位某个物理步骤的断点。"
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：本次 inputs/video.json 为 kind=none，inputs/ 中没有本次回放视频或状态日志，评分器也未给出各节点终值或 DAG 原生指标值。log step 95 的28次泵压、210640 Pa及泵体倾倒，和 log step 93—94 的开锁失败、带锁退头，均来自 agent 未交付的会话测试；不能据此标成本次评分轨迹的进度、接触或失败现象。agent 在 log step 96 自述“时间用完”，但 inputs/run.json 的 wall_s=2604、time_limit_s=3600，不能将其自述当作预算耗尽的事实或失败根因。依 DECISIONS.md D23，agent 结束会话且未交付轨迹时，现象和根因均留空；缺交付本身不足以推定某项能力根因。现有 oracle 视频属于另一成功运行，不能补足本次评分轨迹证据。",
+  "unmet": {
+   "head_connected": null,
+   "pump_stroke_1": null,
+   "pump_stroke_2": null,
+   "pump_stroke_3": null,
+   "pump_stroke_4": null,
+   "pump_stroke_5": null,
+   "pump_stroke_6": null,
+   "pump_stroke_7": null,
+   "pump_stroke_8": null,
+   "pump_stroke_9": null,
+   "pump_stroke_10": null,
+   "pump_stroke_11": null,
+   "pump_stroke_12": null,
+   "pump_stroke_13": null,
+   "pump_stroke_14": null,
+   "pump_stroke_15": null,
+   "pump_stroke_16": null,
+   "pump_stroke_17": null,
+   "pump_stroke_18": null,
+   "pump_stroke_19": null,
+   "pump_stroke_20": null,
+   "pump_stroke_21": null,
+   "pump_stroke_22": null,
+   "pump_stroke_23": null,
+   "pump_stroke_24": null,
+   "pump_stroke_25": null,
+   "pump_stroke_26": null,
+   "pump_stroke_27": null,
+   "pump_stroke_28": null,
+   "head_disconnected": null
+  }
+ },
+ "summary_zh": "评分器记录未交付轨迹（missing_trajectory=1，n_actions=0，sim_steps=0），本次30个节点均无可判定的终值，实际接触均为不适用。agent 的会话测试曾打印接通与至少28次泵压，但后续开锁退头及泵体稳定测试未成功，最终在 log step 98 承认未生成交付文件；这些调试结果不计入评分进度。由于本次无视频、无状态日志且无节点级评分结果，断点记 unknown／首步前，现象与根因依D23留空。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "本次无视频、无状态日志、无交付轨迹（missing_trajectory=1，n_actions=0）。标注全部节点 unknown、断点 unknown、现象与根因留空并写 gap_zh，符合 DECISIONS.md D23(a)；未把会话测试当作评分轨迹事实，也未把 agent 自述“时间用完”当作预算耗尽（wall_s=2604 < 3600）。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": "2026-10-30T22:17:23Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/dexhand-bicycle-pump-stroke-i00-privileged--qkoxr8f/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-bicycle-pump-stroke-i00-privileged--qkoxr8f/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "0eb66fd6b7c6ed8ce023753fcba214459ac63afc3f038729c19416cd7ef2a2bd",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/bicycle_pump_stroke/trajectories/dexhand-bicycle-pump-stroke-i00-standard--pg5jaq3.json

@@ -0,0 +1,598 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-bicycle-pump-stroke-i00-standard--pg5jaq3",
+ "taskKey": "dexhand/bicycle_pump_stroke",
+ "dag_sha256": "1e81e1d6887923dbaed96f9b3045c0eecce1c787fce830bd5914fcd0006f269f",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "calls": 140,
+   "control_steps_total": 2439,
+   "elapsed_s": 3650.598,
+   "finalized": 1,
+   "gave_up": 0,
+   "graded_by_live": 1,
+   "grader_error": 0,
+   "hand_wuji": 0,
+   "infra_error": 0,
+   "instance_mismatch": 0,
+   "live_success": 0,
+   "missing_log": 0,
+   "n_actions": 2090,
+   "no_verdict": 0,
+   "replays": 0,
+   "reset_budget": 20,
+   "resets": 1,
+   "resume_state_mismatch": 0,
+   "resumes": 0,
+   "sim_error": 0,
+   "sim_startups": 1,
+   "sim_steps": 2090,
+   "success": 0,
+   "video_rendered": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 0.415,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.855,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 30,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "head_connected": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 34,
+      "text_zh": "观察：服务在 log step 34 返回 episode=2、control_steps=0、resets_left=19；此前 log step 31 的 task_failure 所属回合已被复位，不是最终受评回合。"
+     },
+     {
+      "t": null,
+      "log_step": 128,
+      "text_zh": "观察：log step 125、127、128 的脚本仍在驱动右手中指合拢泵头拇指锁；log step 128 返回 control_steps=2090、done=False、reason=None，与 reward.n_actions=2090 对应。日志没有服务或评分器报告的 head_connected、泵头座深或锁角；无本次视频和状态日志，不能确定接通事件最终是否已锁存。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：log step 41、44、70、125、127、128 记录右手及小指、中指对泵头和拇指锁的接近、推入与合锁指令；本次无视频、状态接触对或服务接触标志，实际接触部位与方式未知。"
+     }
+    ]
+   },
+   "pump_stroke_1": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 128,
+      "text_zh": "观察：截至 log step 128，最终受评回合的日志未给出服务或评分器的有效泵压累计数、活塞深度或气压；reward 仅给出整体 success=0，未报告第1次泵压。无本次视频和状态日志，无法判定此节点终值，整体失败不等于该次未完成。"
+     }
+    ]
+   },
+   "pump_stroke_2": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 128,
+      "text_zh": "观察：截至 log step 128，最终受评回合的日志未给出服务或评分器的有效泵压累计数、活塞深度或气压；reward 仅给出整体 success=0，未报告第2次泵压。无本次视频和状态日志，无法判定此节点终值，整体失败不等于该次未完成。"
+     }
+    ]
+   },
+   "pump_stroke_3": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 128,
+      "text_zh": "观察：截至 log step 128，最终受评回合的日志未给出服务或评分器的有效泵压累计数、活塞深度或气压；reward 仅给出整体 success=0，未报告第3次泵压。无本次视频和状态日志，无法判定此节点终值，整体失败不等于该次未完成。"
+     }
+    ]
+   },
+   "pump_stroke_4": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 128,
+      "text_zh": "观察：截至 log step 128，最终受评回合的日志未给出服务或评分器的有效泵压累计数、活塞深度或气压；reward 仅给出整体 success=0，未报告第4次泵压。无本次视频和状态日志，无法判定此节点终值，整体失败不等于该次未完成。"
+     }
+    ]
+   },
+   "pump_stroke_5": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 128,
+      "text_zh": "观察：截至 log step 128，最终受评回合的日志未给出服务或评分器的有效泵压累计数、活塞深度或气压；reward 仅给出整体 success=0，未报告第5次泵压。无本次视频和状态日志，无法判定此节点终值，整体失败不等于该次未完成。"
+     }
+    ]
+   },
+   "pump_stroke_6": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 128,
+      "text_zh": "观察：截至 log step 128，最终受评回合的日志未给出服务或评分器的有效泵压累计数、活塞深度或气压；reward 仅给出整体 success=0，未报告第6次泵压。无本次视频和状态日志，无法判定此节点终值，整体失败不等于该次未完成。"
+     }
+    ]
+   },
+   "pump_stroke_7": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 128,
+      "text_zh": "观察：截至 log step 128，最终受评回合的日志未给出服务或评分器的有效泵压累计数、活塞深度或气压；reward 仅给出整体 success=0，未报告第7次泵压。无本次视频和状态日志，无法判定此节点终值，整体失败不等于该次未完成。"
+     }
+    ]
+   },
+   "pump_stroke_8": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 128,
+      "text_zh": "观察：截至 log step 128，最终受评回合的日志未给出服务或评分器的有效泵压累计数、活塞深度或气压；reward 仅给出整体 success=0，未报告第8次泵压。无本次视频和状态日志，无法判定此节点终值，整体失败不等于该次未完成。"
+     }
+    ]
+   },
+   "pump_stroke_9": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 128,
+      "text_zh": "观察：截至 log step 128，最终受评回合的日志未给出服务或评分器的有效泵压累计数、活塞深度或气压；reward 仅给出整体 success=0，未报告第9次泵压。无本次视频和状态日志，无法判定此节点终值，整体失败不等于该次未完成。"
+     }
+    ]
+   },
+   "pump_stroke_10": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 128,
+      "text_zh": "观察：截至 log step 128，最终受评回合的日志未给出服务或评分器的有效泵压累计数、活塞深度或气压；reward 仅给出整体 success=0，未报告第10次泵压。无本次视频和状态日志，无法判定此节点终值，整体失败不等于该次未完成。"
+     }
+    ]
+   },
+   "pump_stroke_11": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 128,
+      "text_zh": "观察：截至 log step 128，最终受评回合的日志未给出服务或评分器的有效泵压累计数、活塞深度或气压；reward 仅给出整体 success=0，未报告第11次泵压。无本次视频和状态日志，无法判定此节点终值，整体失败不等于该次未完成。"
+     }
+    ]
+   },
+   "pump_stroke_12": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 128,
+      "text_zh": "观察：截至 log step 128，最终受评回合的日志未给出服务或评分器的有效泵压累计数、活塞深度或气压；reward 仅给出整体 success=0，未报告第12次泵压。无本次视频和状态日志，无法判定此节点终值，整体失败不等于该次未完成。"
+     }
+    ]
+   },
+   "pump_stroke_13": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 128,
+      "text_zh": "观察：截至 log step 128，最终受评回合的日志未给出服务或评分器的有效泵压累计数、活塞深度或气压；reward 仅给出整体 success=0，未报告第13次泵压。无本次视频和状态日志，无法判定此节点终值，整体失败不等于该次未完成。"
+     }
+    ]
+   },
+   "pump_stroke_14": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 128,
+      "text_zh": "观察：截至 log step 128，最终受评回合的日志未给出服务或评分器的有效泵压累计数、活塞深度或气压；reward 仅给出整体 success=0，未报告第14次泵压。无本次视频和状态日志，无法判定此节点终值，整体失败不等于该次未完成。"
+     }
+    ]
+   },
+   "pump_stroke_15": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 128,
+      "text_zh": "观察：截至 log step 128，最终受评回合的日志未给出服务或评分器的有效泵压累计数、活塞深度或气压；reward 仅给出整体 success=0，未报告第15次泵压。无本次视频和状态日志，无法判定此节点终值，整体失败不等于该次未完成。"
+     }
+    ]
+   },
+   "pump_stroke_16": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 128,
+      "text_zh": "观察：截至 log step 128，最终受评回合的日志未给出服务或评分器的有效泵压累计数、活塞深度或气压；reward 仅给出整体 success=0，未报告第16次泵压。无本次视频和状态日志，无法判定此节点终值，整体失败不等于该次未完成。"
+     }
+    ]
+   },
+   "pump_stroke_17": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 128,
+      "text_zh": "观察：截至 log step 128，最终受评回合的日志未给出服务或评分器的有效泵压累计数、活塞深度或气压；reward 仅给出整体 success=0，未报告第17次泵压。无本次视频和状态日志，无法判定此节点终值，整体失败不等于该次未完成。"
+     }
+    ]
+   },
+   "pump_stroke_18": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 128,
+      "text_zh": "观察：截至 log step 128，最终受评回合的日志未给出服务或评分器的有效泵压累计数、活塞深度或气压；reward 仅给出整体 success=0，未报告第18次泵压。无本次视频和状态日志，无法判定此节点终值，整体失败不等于该次未完成。"
+     }
+    ]
+   },
+   "pump_stroke_19": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 128,
+      "text_zh": "观察：截至 log step 128，最终受评回合的日志未给出服务或评分器的有效泵压累计数、活塞深度或气压；reward 仅给出整体 success=0，未报告第19次泵压。无本次视频和状态日志，无法判定此节点终值，整体失败不等于该次未完成。"
+     }
+    ]
+   },
+   "pump_stroke_20": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 128,
+      "text_zh": "观察：截至 log step 128，最终受评回合的日志未给出服务或评分器的有效泵压累计数、活塞深度或气压；reward 仅给出整体 success=0，未报告第20次泵压。无本次视频和状态日志，无法判定此节点终值，整体失败不等于该次未完成。"
+     }
+    ]
+   },
+   "pump_stroke_21": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 128,
+      "text_zh": "观察：截至 log step 128，最终受评回合的日志未给出服务或评分器的有效泵压累计数、活塞深度或气压；reward 仅给出整体 success=0，未报告第21次泵压。无本次视频和状态日志，无法判定此节点终值，整体失败不等于该次未完成。"
+     }
+    ]
+   },
+   "pump_stroke_22": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 128,
+      "text_zh": "观察：截至 log step 128，最终受评回合的日志未给出服务或评分器的有效泵压累计数、活塞深度或气压；reward 仅给出整体 success=0，未报告第22次泵压。无本次视频和状态日志，无法判定此节点终值，整体失败不等于该次未完成。"
+     }
+    ]
+   },
+   "pump_stroke_23": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 128,
+      "text_zh": "观察：截至 log step 128，最终受评回合的日志未给出服务或评分器的有效泵压累计数、活塞深度或气压；reward 仅给出整体 success=0，未报告第23次泵压。无本次视频和状态日志，无法判定此节点终值，整体失败不等于该次未完成。"
+     }
+    ]
+   },
+   "pump_stroke_24": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 128,
+      "text_zh": "观察：截至 log step 128，最终受评回合的日志未给出服务或评分器的有效泵压累计数、活塞深度或气压；reward 仅给出整体 success=0，未报告第24次泵压。无本次视频和状态日志，无法判定此节点终值，整体失败不等于该次未完成。"
+     }
+    ]
+   },
+   "pump_stroke_25": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 128,
+      "text_zh": "观察：截至 log step 128，最终受评回合的日志未给出服务或评分器的有效泵压累计数、活塞深度或气压；reward 仅给出整体 success=0，未报告第25次泵压。无本次视频和状态日志，无法判定此节点终值，整体失败不等于该次未完成。"
+     }
+    ]
+   },
+   "pump_stroke_26": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 128,
+      "text_zh": "观察：截至 log step 128，最终受评回合的日志未给出服务或评分器的有效泵压累计数、活塞深度或气压；reward 仅给出整体 success=0，未报告第26次泵压。无本次视频和状态日志，无法判定此节点终值，整体失败不等于该次未完成。"
+     }
+    ]
+   },
+   "pump_stroke_27": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 128,
+      "text_zh": "观察：截至 log step 128，最终受评回合的日志未给出服务或评分器的有效泵压累计数、活塞深度或气压；reward 仅给出整体 success=0，未报告第27次泵压。无本次视频和状态日志，无法判定此节点终值，整体失败不等于该次未完成。"
+     }
+    ]
+   },
+   "pump_stroke_28": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 128,
+      "text_zh": "观察：截至 log step 128，最终受评回合的日志未给出服务或评分器的有效泵压累计数、活塞深度或气压；reward 仅给出整体 success=0，未报告第28次泵压。无本次视频和状态日志，无法判定此节点终值，整体失败不等于该次未完成。"
+     }
+    ]
+   },
+   "head_disconnected": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 128,
+      "text_zh": "观察：log step 125 至128 的指令仍为右手合锁，服务只返回控制步数和非终止状态；reward 没有 head_disconnected 或终态分阶段字段。无本次视频和状态日志，无法确认充气后的开锁退座事件；不能把初始未连接状态计作此步。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "tire_gauge_pa",
+   "unit": "Pa",
+   "threshold": 210600,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "piston_down_depth_m",
+   "unit": "m",
+   "threshold": 0.18,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "piston_return_depth_m",
+   "unit": "m",
+   "threshold": 0.04,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "piston_release_depth_m",
+   "unit": "m",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "pump_release_tilt_rad",
+   "unit": "rad",
+   "threshold": 0.0873,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "pump_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "handle_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "head_connected",
+   "t": null,
+   "log_step": 125,
+   "where": "操作中",
+   "evidence_zh": "推断：log step 34 复位后，机器人由 log step 41 开始接近泵头，log step 44 尝试推入，随后反复调整手指合锁；直到 log step 125、127、128 仍在执行合锁指令。log step 128 的 control_steps=2090 与 reward.n_actions=2090 一致，且未再复位，故将失败定位于最终回合持续操作的“泵头接上气门”；连接事件最终是否锁存仍未知，无视频时刻可定位。"
+  },
+  "unmet": {
+   "head_connected": null,
+   "pump_stroke_1": null,
+   "pump_stroke_2": null,
+   "pump_stroke_3": null,
+   "pump_stroke_4": null,
+   "pump_stroke_5": null,
+   "pump_stroke_6": null,
+   "pump_stroke_7": null,
+   "pump_stroke_8": null,
+   "pump_stroke_9": null,
+   "pump_stroke_10": null,
+   "pump_stroke_11": null,
+   "pump_stroke_12": null,
+   "pump_stroke_13": null,
+   "pump_stroke_14": null,
+   "pump_stroke_15": null,
+   "pump_stroke_16": null,
+   "pump_stroke_17": null,
+   "pump_stroke_18": null,
+   "pump_stroke_19": null,
+   "pump_stroke_20": null,
+   "pump_stroke_21": null,
+   "pump_stroke_22": null,
+   "pump_stroke_23": null,
+   "pump_stroke_24": null,
+   "pump_stroke_25": null,
+   "pump_stroke_26": null,
+   "pump_stroke_27": null,
+   "pump_stroke_28": null,
+   "head_disconnected": null
+  },
+  "phenomena": [],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "t": null,
+    "log_step": 125,
+    "evidence_zh": "推断：agent 的控制脚本在 log step 125 为合锁求得腕部目标 [0.383,-0.194,0.757] m，但根据服务关节位置经 FK 得到的实际腕部位置为 [0.354,-0.192,0.762] m；log step 128 再命令腕部沿 +x 移动0.02 m并降低0.01 m，返回腕部仅由约 [0.357,-0.191,0.763] 变为 [0.358,-0.191,0.759] m，中指末节的 x 坐标未向目标方向推进。结合该回合反复调整同一合锁动作的记录，支持局部操作的控制执行受限；缺少接触和锁角信息，无法区分接触布局、受阻与跟踪控制的具���原因。",
+    "source_zh": "日志",
+    "note_zh": "仅依据最终受评回合的指令及关节位置/FK诊断推断一般控制问题；不将 agent 的接通判断当作节点结果，也不据此断言具体碰撞或接触力异常。"
+   }
+  ],
+  "gap_zh": "本次 inputs/video.json 为 kind=none，未提供本次 replay.mp4 或状态日志；inputs/oracle/replay.mp4 是另一成功运行，不能证明本次进度或接触。日志中的图片查看记录仅有引用与偏移，没有可检查的本次图像；服务输出未报告连接标志、锁角、座深、泵压计数、气压和脱开标志，因此30个节点终值与全部视频时刻均保留未知。无法区分断点处究竟未碰到锁杆、接触后受阻还是锁角不足，现象层留空，控制根因仅为日志支持的推断。log step 70 的“手指骑在轮胎上”和 log step 125 的“泵头已入座、先前合锁方向错误”均为 agent 自述，未获画面或任务状态验证；最后一次反馈仍为 done=False、reason=None，最终仅有整体失败评分，不能补造物理终止原因。"
+ },
+ "summary_zh": "最终受评的是 log step 34 复位后的第二回合：机器人持续尝试泵头入座与合锁，log step 128 达到2090个控制步，评分器最终 success=0；可确认完成的节点为0个，30个节点终值均未知。断点定位于“泵头接上气门”的操作中，日志中合锁指令与实际腕部/FK位置的偏差支持一般控制问题，但不足以确定具体接触现象。没有本次视频、状态日志或节点级服务结果，不能由整体零分断言所有节点为假，也不能用成功参考运行补全本次进度。",
+ "conf": 5,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "本次无视频、无状态日志（video.kind=none），30个节点保持未知符合N9；断点定位于 head_connected（log step 34 复位后的最终回合持续合锁）有日志依据，现象层留空并有 gap_zh，控制根因仅为日志推断且措辞克制。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": "2026-10-30T22:20:51Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/dexhand-bicycle-pump-stroke-i00-standard--pg5jaq3/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-bicycle-pump-stroke-i00-standard--pg5jaq3/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "d1b7e97c30bdb1c84e46f1e68940dcd0391ce7be8830449552883023272e9e86",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/binder_page_load/trajectories/dexhand-binder-page-load-i00-privileged--jrxsjt5.json

@@ -0,0 +1,262 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-binder-page-load-i00-privileged--jrxsjt5",
+ "taskKey": "dexhand/binder_page_load",
+ "dag_sha256": "7b66f31f677549838769fa5d0443c5c82e455453dda042899107c711b92d17dd",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.725,
+   "peak_processes": 8,
+   "peak_disk_gb": 0.255,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 4,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "rings_opened": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 8,
+      "text_zh": "观察：log step 8 为 agent 新建 Session 的开环测试；评分器 reward 记录 missing_trajectory=1、n_actions=0、sim_steps=0，没有报告本次交付的开环结果。无本次视频及状态日志，测试结果不能作为受评节点值。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1、n_actions=0，未交付轨迹；会话测试中的接触不计入受评尝试。"
+     }
+    ]
+   },
+   "pages_threaded_1": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 61,
+      "text_zh": "观察：log step 61 为 agent 的单页穿环测试，输出 open_threaded=[0,3]；评分器 reward 记录 missing_trajectory=1、n_actions=0，没有交付回放的装页读数。无本次视频及状态日志，不能从会话测试认定第1张纸入环。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1、n_actions=0，未交付轨迹；会话测试中的接触不计入受评尝试。"
+     }
+    ]
+   },
+   "pages_threaded_2": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 89,
+      "text_zh": "观察：log step 84 展示测试脚本通过 Session.restore 恢复阶段快照，log step 89 的测试输出 linked=[3,3]、stage=4、hold_steps=50；评分器 reward 仍记录 missing_trajectory=1、n_actions=0。这些是会话测试读数，没有受评轨迹证明第2张纸入环。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1、n_actions=0，未交付轨迹；会话测试中的接触不计入受评尝试。"
+     }
+    ]
+   },
+   "rings_closed": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 89,
+      "text_zh": "观察：log step 89 的快照恢复测试输出 rings_closed=True，log step 7 另有测试初态 toggles=[0,0]；评分器未报告受评终态且 missing_trajectory=1。无本次视频及状态日志，不能把测试终态或任务默认初态记为受评终态。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1、n_actions=0，未交付轨迹；会话测试中的接触不计入受评尝试。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "min_toggle_open_angle",
+   "unit": "rad",
+   "threshold": 0.5,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "max_toggle_closed_angle",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "hole_capture_radius",
+   "unit": "m",
+   "threshold": 0.0035,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "binder_horizontal_shift",
+   "unit": "m",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "binder_tilt",
+   "unit": "rad",
+   "threshold": 0.35,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "wire_page_penetration",
+   "unit": "m",
+   "threshold": 0.0012,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "hand_page_penetration",
+   "unit": "m",
+   "threshold": 0.003,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "page_window_translation",
+   "unit": "m",
+   "threshold": 0.002,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "page_window_rotation",
+   "unit": "rad",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 90,
+   "where": "首步前",
+   "evidence_zh": "观察：log step 90 记录“This request was blocked by Gemini's filters”，此后无工具调用或交付记录；评分器 reward 为 missing_trajectory=1、n_actions=0、sim_steps=0。log step 80、84 保存及恢复的是阶段测试快照，log step 87、89 的成功读数属于该测试，未见交付动作文件保存及验证。"
+  },
+  "unmet": {
+   "rings_opened": null,
+   "pages_threaded_1": null,
+   "pages_threaded_2": null,
+   "rings_closed": null
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：本次 inputs/video.json 为 kind=none，inputs/ 中没有本次 replay.mp4 或 states.jsonl，评分器未提供四个节点的结果或原生指标；已有 oracle 视频属于另一成功运行，不能补足本次证据。log step 90 为模型提供方过滤器中断，评分器记录缺交付；log step 87、89 虽有测试成功读数，但测试脚本从阶段快照恢复，日志未记录交付文件保存及重放验证。推断：中断发生于交付及受评结果固定之前，按外因中断处理，不归入 agent 物理失败；四节点终值未知，全部时间区间为 null，实际接触为不适用，现象与根因留空，不以缺交付推断没尝试或交付操作出错。"
+ },
+ "summary_zh": "观察：评分器记录 success=0、missing_trajectory=1、n_actions=0、sim_steps=0，本次没有受评轨迹可确认四个节点的完成情况；log step 87、89 的阶段4成功和50控制步静置仅为恢复快照后的会话测试结果。观察：log step 90 出现 Gemini 提供方过滤器中断，日志未见交付文件保存及验证；推断：属于交付前的外因中断，断点记 unknown／首步前，现象和根因留空。观察：本次无视频、无状态日志、无逐节点评分字段，四节点终值均保留未知。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 9,
+    "summary_zh": "无视频、无状态日志；评分器记 missing_trajectory=1、n_actions=0，log step 90 为 Gemini 过滤器中断，全程无交付，会话测试读数未被当作受评事实。标签符合 03 第5条外因中断规则（ref unknown、首步前、现象与根因空、gap_zh 说明），四节点 unknown，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": "2026-10-30T22:17:33Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/dexhand-binder-page-load-i00-privileged--jrxsjt5/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-binder-page-load-i00-privileged--jrxsjt5/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "5172e778cdbf9aa7da78f34d8f52a69571809b55a1de83f1dd2586ebd2d03860",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/binder_page_load/trajectories/dexhand-binder-page-load-i00-privileged--kdzoggb.json

@@ -0,0 +1,262 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-binder-page-load-i00-privileged--kdzoggb",
+ "taskKey": "dexhand/binder_page_load",
+ "dag_sha256": "7b66f31f677549838769fa5d0443c5c82e455453dda042899107c711b92d17dd",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.857,
+   "peak_processes": 6,
+   "peak_disk_gb": 0.269,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 4,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "rings_opened": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 452,
+      "text_zh": "观察：评分器reward报告missing_trajectory=1、n_actions=0、sim_steps=0，未报告本节点结果；log step 452的stage=1及两板约0.602 rad来自agent自建Session测试，并非交付重放。本次无视频和状态日志，无法判定评分尝试是否完成开环。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告missing_trajectory=1、n_actions=0；未交付轨迹，本节点无可标注的评分尝试接触。"
+     }
+    ]
+   },
+   "pages_threaded_1": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 452,
+      "text_zh": "观察：评分器未报告装页节点结果且missing_trajectory=1；log step 452的open_threaded=[0,3]、pages_resting=[false,true]仅为agent测试脚本输出，不能作为评分尝试的节点值。本次无视频和状态日志，首张纸入环终值未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告missing_trajectory=1、n_actions=0；未交付轨迹，本节点无可标注的评分尝试接触。"
+     }
+    ]
+   },
+   "pages_threaded_2": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 492,
+      "text_zh": "观察：评分器未报告装页张数且missing_trajectory=1；log step 492的测试输出为open_threaded=[0,0]、failure=page_dropped，属于会话调试，不能据此判定评分尝试第二张纸入环为假。本次无视频和状态日志，节点终值未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告missing_trajectory=1、n_actions=0；未交付轨迹，本节点无可标注的评分尝试接触。"
+     }
+    ]
+   },
+   "rings_closed": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 465,
+      "text_zh": "观察：评分器未报告扣环终态且missing_trajectory=1；log step 465中测试脚本结束仍为stage=1、两板约0.602 rad，此为未交付的会话结果。本次无视频和状态日志，不能把测试终态或DAG描述的闭合初态作为评分尝试的最终节点值。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告missing_trajectory=1、n_actions=0；未交付轨迹，本节点无可标注的评分尝试接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "min_toggle_open_angle",
+   "unit": "rad",
+   "threshold": 0.5,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "max_toggle_closed_angle",
+   "unit": "rad",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "hole_capture_radius",
+   "unit": "m",
+   "threshold": 0.0035,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "binder_horizontal_shift",
+   "unit": "m",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "binder_tilt",
+   "unit": "rad",
+   "threshold": 0.35,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "wire_page_penetration",
+   "unit": "m",
+   "threshold": 0.0012,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "hand_page_penetration",
+   "unit": "m",
+   "threshold": 0.003,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "page_window_translation",
+   "unit": "m",
+   "threshold": 0.002,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "page_window_rotation",
+   "unit": "rad",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 499,
+   "where": "首步前",
+   "evidence_zh": "观察：inputs/run.json的评分器reward为missing_trajectory=1、n_actions=0、sim_steps=0、infra_error=0；log step 499仅启动agent的tune_p0_drag.py后台测试，日志至此结束，全部say为空，未见保存、提交或交付重放结果。推断：没有交付的评分轨迹，无法定位物理失败步骤，按未交付情形记为首步前。"
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：inputs/video.json与skeleton.json均记录video.kind=none，inputs/内无本次replay.mp4或states.jsonl；评分器也无逐节点结果，四个节点的intervals及final均保留null。log step 452、465是未交付的整段测试，log step 490、492、496的page_dropped是会话调试结果，不能转作评分轨迹的失败现象或节点值。run记录wall_s=3628、time_limit_s=3600，log step 499仍在运行测试，未记录最终保存动作或明确的外因中断；仅凭缺交付不能推断交付操作出错、控制或规划根因。按D23，未交付时现象与根因均留空。inputs/oracle/的成功录像属于另一运行，不能补足本次证据。",
+  "unmet": {
+   "rings_opened": null,
+   "pages_threaded_1": null,
+   "pages_threaded_2": null,
+   "rings_closed": null
+  }
+ },
+ "summary_zh": "评分器记录未交付轨迹（missing_trajectory=1、n_actions=0、sim_steps=0），四个节点均无可判定的最终结果。agent会话曾输出开环和单页入环的测试读数（log step 452、465），后续仍在调试纸张掉落（log step 492、496），没有证据将这些测试认定为交付尝试。断点记unknown／首步前；本次无视频、无状态日志且无逐节点评分，依D23不从会话测试补填失败现象或根因。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "本次无视频、无状态日志，评分器为missing_trajectory=1、n_actions=0；标注把四个节点留unknown、断点记unknown/首步前、现象与根因留空并写gap_zh，符合D23(a)，未把会话测试读数当作评分尝试事实。",
+    "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": "2026-10-30T22:17:22Z",
+  "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/dexhand-binder-page-load-i00-privileged--kdzoggb/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-binder-page-load-i00-privileged--kdzoggb/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "2b9fa345bdc315bfef2f67b40ff0127e6e07c0ecbedc7173fd1a491e8cd8daaa",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/book_page_turn/trajectories/dexhand-book-page-turn-i00-privileged--sarkret.json

@@ -0,0 +1,179 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-book-page-turn-i00-privileged--sarkret",
+ "taskKey": "dexhand/book_page_turn",
+ "dag_sha256": "b1614d794a716fb93ce7267d3c538874fe875ffd93b9add17265ecb5aed9cd91",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 2.053,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.242,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 1,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "target_page_turned": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器字段为 missing_trajectory=1、n_actions=0、sim_steps=0、success=0；未提供受评回合的翻页阶段、离手保持计数或节点终态，inputs/video.json 为 kind=none，且无状态日志，故无法判断“翻过指定页”的物理终态。"
+     },
+     {
+      "t": null,
+      "log_step": 21,
+      "text_zh": "观察：log step 20、21 是 agent 脚本的本地分段测试，分别以 --stop 130 和 --stop 300 运行，均未传入触发保存的 --save 参数；log step 23 又新建会话测量初始几何。上述会话读数不是已交付轨迹的节点证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告 missing_trajectory=1、n_actions=0，没有交付轨迹；不将本地测试中灵巧手与书页的动作或接触计划作为受评接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "target_angle_deg",
+   "unit": "deg",
+   "threshold": 170,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "target_speed_rad_s",
+   "unit": "rad/s",
+   "threshold": 0.2,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_other_leaf_change_deg",
+   "unit": "deg",
+   "threshold": 5,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "book_shift_m",
+   "unit": "m",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "book_rotation_deg",
+   "unit": "deg",
+   "threshold": 10,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": null,
+   "where": "首步前",
+   "evidence_zh": "观察：inputs/run.json 的评分器报告 missing_trajectory=1、n_actions=0、sim_steps=0，infra_error=0；日志末条 log step 23 仍为 agent 新建会话测量书页几何，没有交付或回放验证记录，无法定位受评轨迹内的物理失败步骤。"
+  },
+  "unmet": {
+   "target_page_turned": null
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：本次受评运行没有视频或状态日志，评分器明确报告缺少交付轨迹；成功参考视频 inputs/oracle/replay.mp4 属于另一运行，不能证明本次接触或节点终态。log step 17、19 的脚本错误之后仍有 log step 20、21 的本地测试，不能将这些会话事件当成受评失败现象，也没有足够证据确定导致最终缺交付的具体能力根因。推断：run.json 记录 wall_s=3603、time_limit_s=3600，与会话用尽预算相符，但预算本身不是现象或根因；依 DECISIONS.md D23，两层留空，不从缺交付反推没尝试或交付操作出错。"
+ },
+ "summary_zh": "评分器报告缺少交付轨迹、动作数为0，因此没有可确认完成的节点；“翻过指定页”的物理终态保持未知，断点记为 unknown／首步前。日志只记录了 agent 编写控制器、分段调试及末尾几何测量，不能据此标注受评接触或失败现象；无本次视频和状态日志，现象及根因均按 D23 留空。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "无视频、无状态日志，评分器 missing_trajectory=1、n_actions=0；标注为节点 unknown、两层留空加 gap_zh，符合 D23 与 N9，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": "2026-10-30T22:17:06Z",
+  "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/dexhand-book-page-turn-i00-privileged--sarkret/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-book-page-turn-i00-privileged--sarkret/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "18b64100d104cb44fb24c975965b40f4dfc109b071131a252427523043a953cc",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/boot_sole_brush/trajectories/dexhand-boot-sole-brush-i00-privileged--c3hgvqv.json

@@ -0,0 +1,307 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-boot-sole-brush-i00-privileged--c3hgvqv",
+ "taskKey": "dexhand/boot_sole_brush",
+ "dag_sha256": "cd4451f62f3532c4b6d69d8cb39e471594d0629b38c2d857aeb9b6ca1a36e3a1",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 3.004,
+   "peak_processes": 10,
+   "peak_disk_gb": 0.254,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 7,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "brush_in_hand": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 122,
+      "text_zh": "观察：评分器 reward 记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“拿起鞋刷”的节点结果；本次无评分视频、无状态日志。log step 122 是 agent 本地测试输出，不能作为已交付评分轨迹的节点值，故 final 保留未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "评分器记录 missing_trajectory=1，未交付评分轨迹；会话测试中的右手或鞋刷接触不计为本次评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "clods_in_tray_1": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 122,
+      "text_zh": "观察：评分器 reward 记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第1块泥入盘”的节点结果；本次无评分视频、无状态日志。log step 122 是 agent 本地测试输出，不能作为已交付评分轨迹的节点值，故 final 保留未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "评分器记录 missing_trajectory=1，未交付评分轨迹；会话测试中的右手或鞋刷接触不计为本次评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "clods_in_tray_2": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 122,
+      "text_zh": "观察：评分器 reward 记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第2块泥入盘”的节点结果；本次无评分视频、无状态日志。log step 122 是 agent 本地测试输出，不能作为已交付评分轨迹的节点值，故 final 保留未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "评分器记录 missing_trajectory=1，未交付评分轨迹；会话测试中的右手或鞋刷接触不计为本次评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "clods_in_tray_3": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 122,
+      "text_zh": "观察：评分器 reward 记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第3块泥入盘”的节点结果；本次无评分视频、无状态日志。log step 122 是 agent 本地测试输出，不能作为已交付评分轨迹的节点值，故 final 保留未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "评分器记录 missing_trajectory=1，未交付评分轨迹；会话测试中的右手或鞋刷接触不计为本次评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "clods_in_tray_4": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 122,
+      "text_zh": "观察：评分器 reward 记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第4块泥入盘”的节点结果；本次无评分视频、无状态日志。log step 122 是 agent 本地测试输出，不能作为已交付评分轨迹的节点值，故 final 保留未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "评分器记录 missing_trajectory=1，未交付评分轨迹；会话测试中的右手或鞋刷接触不计为本次评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "clods_in_tray_5": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 122,
+      "text_zh": "观察：评分器 reward 记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第5块泥入盘”的节点结果；本次无评分视频、无状态日志。log step 122 是 agent 本地测试输出，不能作为已交付评分轨迹的节点值，故 final 保留未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "评分器记录 missing_trajectory=1，未交付评分轨迹；会话测试中的右手或鞋刷接触不计为本次评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "clods_in_tray_6": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 122,
+      "text_zh": "观察：评分器 reward 记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第6块泥入盘”的节点结果；本次无评分视频、无状态日志。log step 122 是 agent 本地测试输出，不能作为已交付评分轨迹的节点值，故 final 保留未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "评分器记录 missing_trajectory=1，未交付评分轨迹；会话测试中的右手或鞋刷接触不计为本次评分轨迹的实际接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "clods_in_tray",
+   "unit": "count",
+   "threshold": 6,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "clods_brushed",
+   "unit": "count",
+   "threshold": 6,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "brush_contact_controls",
+   "unit": "control",
+   "threshold": 10,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "hold",
+   "unit": "control",
+   "threshold": 50,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "max_clod_speed",
+   "unit": "m/s",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "boot_shift_m",
+   "unit": "m",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "boot_tilt_rad",
+   "unit": "rad",
+   "threshold": 0.35,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 122,
+   "where": "首步前",
+   "evidence_zh": "观察：评分器 reward 为 success=0、missing_trajectory=1、n_actions=0、sim_steps=0、infra_error=0；log step 16 的 agent 测试服务器每次请求先恢复初始快照，log step 121 又恢复悬停快照，log step 122 最后一次本地测试因 hand_touched_debris 结束。完整会话未见保存或提交评分轨迹的调用，无法将某轮测试认定为评分尝试，也无法定位实际失败的物理节点。"
+  },
+  "unmet": {
+   "clods_in_tray_1": null,
+   "clods_in_tray_2": null,
+   "clods_in_tray_3": null,
+   "clods_in_tray_4": null,
+   "clods_in_tray_5": null,
+   "clods_in_tray_6": null
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：inputs/video.json 的 kind=none，inputs/ 只有本次 run.json、元数据和另一次成功运行的参考视频，没有本次评分视频或状态日志；评分器明确报告 missing_trajectory=1，未提供逐节点终态和 DAG 原生指标。log step 118 的两块泥入盘后掉刷、log step 122 的右手碰泥均属恢复快照后的会话测试，不能标为评分轨迹的进度、接触或失败现象。run.json 的 wall_s=3619、time_limit_s=3600，但没有最终交付报告或解释未交付的记录；按 D23，两层失败标签留空，不以缺交付或预算耗尽反推根因，亦无证据认定具体保存操作出错或外部中断。"
+ },
+ "summary_zh": "评分器判失败并报告未交付轨迹（missing_trajectory=1、n_actions=0、sim_steps=0），七个节点均无可判定的评分终态。agent 会话进行了多轮恢复快照后的拿刷、移刷和刷泥测试，最后 log step 122 因右手碰泥结束，但测试结果不能计入评分轨迹，断点记 unknown／首步前。缺少本次视频、状态日志和逐节点评分结果，实际接触记不适用，失败现象与根因留空。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "本次无评分视频、无状态日志，评分器报告 missing_trajectory=1、n_actions=0、sim_steps=0；七个节点 unknown、断点 unknown／首步前、两层留空并写 gap_zh，符合 D23(a)。会话测试（log step 118/122）未被当作评分轨迹事实，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": "2026-10-30T22:18:45Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/dexhand-boot-sole-brush-i00-privileged--c3hgvqv/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-boot-sole-brush-i00-privileged--c3hgvqv/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "258dd36f68cc5cd0422a895a5c6d57ceca0e08a29a49084c4d66460e5caa9449",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/boot_sole_brush/trajectories/dexhand-boot-sole-brush-i00-privileged--sb5aajd.json

@@ -0,0 +1,307 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-boot-sole-brush-i00-privileged--sb5aajd",
+ "taskKey": "dexhand/boot_sole_brush",
+ "dag_sha256": "cd4451f62f3532c4b6d69d8cb39e471594d0629b38c2d857aeb9b6ca1a36e3a1",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 3.305,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.266,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 7,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "brush_in_hand": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器字段为 missing_trajectory=1、n_actions=0、sim_steps=0；未报告“拿起鞋刷”的节点结果。本运行无视频、无状态日志，无法确定该节点终值；会话测试不作为评分轨迹进度。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告 missing_trajectory=1，未交付评分轨迹，故无可标注的轨迹级实际接触。"
+     }
+    ]
+   },
+   "clods_in_tray_1": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器字段为 missing_trajectory=1、n_actions=0、sim_steps=0；未报告“第1块泥入盘”的节点结果。本运行无视频、无状态日志，无法确定该节点终值；会话测试不作为评分轨迹进度。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告 missing_trajectory=1，未交付评分轨迹，故无可标注的轨迹级实际接触。"
+     }
+    ]
+   },
+   "clods_in_tray_2": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器字段为 missing_trajectory=1、n_actions=0、sim_steps=0；未报告“第2块泥入盘”的节点结果。本运行无视频、无状态日志，无法确定该节点终值；会话测试不作为评分轨迹进度。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告 missing_trajectory=1，未交付评分轨迹，故无可标注的轨迹级实际接触。"
+     }
+    ]
+   },
+   "clods_in_tray_3": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器字段为 missing_trajectory=1、n_actions=0、sim_steps=0；未报告“第3块泥入盘”的节点结果。本运行无视频、无状态日志，无法确定该节点终值；会话测试不作为评分轨迹进度。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告 missing_trajectory=1，未交付评分轨迹，故无可标注的轨迹级实际接触。"
+     }
+    ]
+   },
+   "clods_in_tray_4": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器字段为 missing_trajectory=1、n_actions=0、sim_steps=0；未报告“第4块泥入盘”的节点结果。本运行无视频、无状态日志，无法确定该节点终值；会话测试不作为评分轨迹进度。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告 missing_trajectory=1，未交付评分轨迹，故无可标注的轨迹级实际接触。"
+     }
+    ]
+   },
+   "clods_in_tray_5": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器字段为 missing_trajectory=1、n_actions=0、sim_steps=0；未报告“第5块泥入盘”的节点结果。本运行无视频、无状态日志，无法确定该节点终值；会话测试不作为评分轨迹进度。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告 missing_trajectory=1，未交付评分轨迹，故无可标注的轨迹级实际接触。"
+     }
+    ]
+   },
+   "clods_in_tray_6": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器字段为 missing_trajectory=1、n_actions=0、sim_steps=0；未报告“第6块泥入盘”的节点结果。本运行无视频、无状态日志，无法确定该节点终值；会话测试不作为评分轨迹进度。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告 missing_trajectory=1，未交付评分轨迹，故无可标注的轨迹级实际接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "clods_in_tray",
+   "unit": "count",
+   "threshold": 6,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "clods_brushed",
+   "unit": "count",
+   "threshold": 6,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "brush_contact_controls",
+   "unit": "control",
+   "threshold": 10,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "hold",
+   "unit": "control",
+   "threshold": 50,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "max_clod_speed",
+   "unit": "m/s",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "boot_shift_m",
+   "unit": "m",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "boot_tilt_rad",
+   "unit": "rad",
+   "threshold": 0.35,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 522,
+   "where": "首步前",
+   "evidence_zh": "观察：评分器报告 missing_trajectory=1、n_actions=0、sim_steps=0、infra_error=0。agent 在 log step 520 启动新的斜向扫刷会话测试，log step 521 仅返回后台任务 RUNNING，log step 522 最后自述“等待任务通知”；完整日志未见交付轨迹的保存及回放验证记录。推断：评分尝试未形成，无法将断点归到某一物理步骤，记为 unknown／首步前。"
+  },
+  "unmet": {
+   "clods_in_tray_1": null,
+   "clods_in_tray_2": null,
+   "clods_in_tray_3": null,
+   "clods_in_tray_4": null,
+   "clods_in_tray_5": null,
+   "clods_in_tray_6": null
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：本运行 inputs/video.json 的 kind=none，且 inputs/ 中没有本运行状态日志；仅有另一成功运行的 oracle 视频，不能证明本次结果。评分器未收到轨迹，也未报告拿刷、泥块入盘或 DAG 原生指标的节点级结果，因此全部节点终值与时间区间保留 null，实际接触记不适用。日志末尾仍为会话测试及等待，未给出最终物理结果或明确的未交付原因；wall_s=3620、time_limit_s=3600 只描述会话预算，不证明具体能力缺陷。按 D23，未交付情形的现象与根因均留空，不能把会话测试当作评分尝试，亦不能仅从交付缺失推定交付操作出错。"
+ },
+ "summary_zh": "评分器报告未交付轨迹（missing_trajectory=1），评分动作数及仿真步数均为0，七个节点没有可确认的终值。agent 持续进行抓刷与扫刷会话测试，最后在 log step 522 等待后台任务，断点记 unknown／首步前，未交付的具体原因证据不足。由于本运行无视频、无状态日志且评分器未报告节点级结果，按 D23 留空现象与根因，不将会话测试计入进度。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "无视频、无状态日志、评分器 missing_trajectory=1；全程只有会话测试，日志从未保存或提交交付轨迹。节点全为 unknown、现象与根因留空并附 gap_zh，符合 D23(a)；未把超时当现象或根因，无时钟转换，断点 unknown/首步前可接受。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T22:18:22Z",
+  "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/dexhand-boot-sole-brush-i00-privileged--sb5aajd/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-boot-sole-brush-i00-privileged--sb5aajd/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "071e45bcaf3ce1e78012d98258c016da4740cfcfe9d29dc90e2db9cfbe8b915b",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/drying_rack_unfold/trajectories/dexhand-drying-rack-unfold-i00-privileged--6x5hrab.json

@@ -0,0 +1,295 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-drying-rack-unfold-i00-privileged--6x5hrab",
+ "taskKey": "dexhand/drying_rack_unfold",
+ "dag_sha256": "044da74935ab7bbd4a0b895fad36baee09e495811264f09ed418f743b236ddb1",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.587,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.239,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 6,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "rack_unfolded": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器字段为 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“展开主架”的终态；inputs/video.json 为 kind=none，且无状态日志，无法判定本节点。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付受评分轨迹；会话调试中的右手接触不作为本次轨迹接触。"
+     }
+    ]
+   },
+   "legs_opened_1": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器字段为 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第1条腿展开”的终态；inputs/video.json 为 kind=none，且无状态日志，无法判定本节点。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付受评分轨迹；会话调试中的右手接触不作为本次轨迹接触。"
+     }
+    ]
+   },
+   "legs_opened_2": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器字段为 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第2条腿展开”的终态；inputs/video.json 为 kind=none，且无状态日志，无法判定本节点。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付受评分轨迹；会话调试中的右手接触不作为本次轨迹接触。"
+     }
+    ]
+   },
+   "legs_opened_3": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器字段为 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第3条腿展开”的终态；inputs/video.json 为 kind=none，且无状态日志，无法判定本节点。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付受评分轨迹；会话调试中的右手接触不作为本次轨迹接触。"
+     }
+    ]
+   },
+   "legs_opened_4": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器字段为 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第4条腿展开”的终态；inputs/video.json 为 kind=none，且无状态日志，无法判定本节点。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付受评分轨迹；会话调试中的右手接触不作为本次轨迹接触。"
+     }
+    ]
+   },
+   "rack_standing": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器字段为 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“晾衣架正放”的终态；inputs/video.json 为 kind=none，且无状态日志，无法判定本节点。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付受评分轨迹；会话调试中的右手接触不作为本次轨迹接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "fold_angle_rad",
+   "unit": "rad",
+   "threshold": 0.3141592653589793,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "leg_angles_rad[0]",
+   "unit": "rad",
+   "threshold": 1.5707963267948966,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "leg_angles_rad[1]",
+   "unit": "rad",
+   "threshold": 1.5707963267948966,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "leg_angles_rad[2]",
+   "unit": "rad",
+   "threshold": 1.5707963267948966,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "leg_angles_rad[3]",
+   "unit": "rad",
+   "threshold": 1.5707963267948966,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "tilt_deg",
+   "unit": "deg",
+   "threshold": 10.0,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "rack_linear_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "rack_angular_speed_rad_s",
+   "unit": "rad/s",
+   "threshold": 0.2,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 22,
+   "where": "首步前",
+   "evidence_zh": "观察：评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0、success=0、infra_error=0；agent 在 log step 13–19 运行抓取网格和诊断脚本，log step 20、22 均无调用、输出或报告，完整日志中未见保存或提交轨迹。推断：没有可定位物理失败步骤的受评分轨迹，断点记为 unknown／首步前。"
+  },
+  "unmet": {
+   "rack_unfolded": null,
+   "legs_opened_1": null,
+   "legs_opened_2": null,
+   "legs_opened_3": null,
+   "legs_opened_4": null,
+   "rack_standing": null
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：inputs/ 仅含运行记录与视频元信息等文件，没有视频或状态日志；评分器未报告六个节点的终态及八项原生物理指标。log step 13–19 的脚本错误、零接触、少量折角变化及 rack_dropped 等结果均来自独立会话测试，未保存为受评分轨迹；log step 20、22 无最终说明。运行 wall_s=3604、time_limit_s=3600，但没有记录能确定未交付的具体能力原因。依 DECISIONS.md D23，无交付时现象和根因均留空，不把会话测试的接触失败当作评分尝试的现象，也不从缺交付推断交付操作出错；依 N9，六个节点终值及所有时间区间保留 null。"
+ },
+ "summary_zh": "评分器判失败，记录 missing_trajectory=1、n_actions=0、sim_steps=0；没有受评分轨迹，六个节点终值均未知。agent 在 log step 13–19 进行主架端杆抓取与抬升测试，但完整日志未见保存或提交，断点为 unknown／首步前，按 DECISIONS.md D23 将现象与根因留空。没有视频、状态日志或节点级评分结果，会话测试不能代替受评分进度，八项原生指标也均未报告。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "无视频、无状态日志，评分器 missing_trajectory=1、n_actions=0；六节点 unknown、断点 unknown／首步前、现象与根因留空并写 gap_zh，符合 D23(a) 与 N9，会话测试未被当作受评分尝试。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": "2026-10-30T22:17:07Z",
+  "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/dexhand-drying-rack-unfold-i00-privileged--6x5hrab/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-drying-rack-unfold-i00-privileged--6x5hrab/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "34196a1c83747e34335d1129f88b6f435850486fa15e865247f87f87608ef0eb",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/drying_rack_unfold/trajectories/dexhand-drying-rack-unfold-i00-privileged--vhjnad5.json

@@ -0,0 +1,295 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-drying-rack-unfold-i00-privileged--vhjnad5",
+ "taskKey": "dexhand/drying_rack_unfold",
+ "dag_sha256": "044da74935ab7bbd4a0b895fad36baee09e495811264f09ed418f743b236ddb1",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 0.593,
+   "peak_processes": 6,
+   "peak_disk_gb": 0.262,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 6,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "rack_unfolded": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“展开主架”的正式评分状态；inputs/video.json 为 kind=none，且无状态日志。会话调试结果不作为本节点终值，故保留未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1，未交付供正式评分的动作轨迹；会话测试中右手与晾衣架的接触不计入轨迹级接触。"
+     }
+    ]
+   },
+   "legs_opened_1": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第1条腿展开”的正式评分状态；inputs/video.json 为 kind=none，且无状态日志。会话调试结果不作为本节点终值，故保留未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1，未交付供正式评分的动作轨迹；会话测试中右手与晾衣架的接触不计入轨迹级接触。"
+     }
+    ]
+   },
+   "legs_opened_2": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第2条腿展开”的正式评分状态；inputs/video.json 为 kind=none，且无状态日志。会话调试结果不作为本节点终值，故保留未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1，未交付供正式评分的动作轨迹；会话测试中右手与晾衣架的接触不计入轨迹级接触。"
+     }
+    ]
+   },
+   "legs_opened_3": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第3条腿展开”的正式评分状态；inputs/video.json 为 kind=none，且无状态日志。会话调试结果不作为本节点终值，故保留未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1，未交付供正式评分的动作轨迹；会话测试中右手与晾衣架的接触不计入轨迹级接触。"
+     }
+    ]
+   },
+   "legs_opened_4": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“第4条腿展开”的正式评分状态；inputs/video.json 为 kind=none，且无状态日志。会话调试结果不作为本节点终值，故保留未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1，未交付供正式评分的动作轨迹；会话测试中右手与晾衣架的接触不计入轨迹级接触。"
+     }
+    ]
+   },
+   "rack_standing": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“晾衣架正放”的正式评分状态；inputs/video.json 为 kind=none，且无状态日志。会话调试结果不作为本节点终值，故保留未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1，未交付供正式评分的动作轨迹；会话测试中右手与晾衣架的接触不计入轨迹级接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "fold_angle_rad",
+   "unit": "rad",
+   "threshold": 0.3141592653589793,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "leg_angles_rad[0]",
+   "unit": "rad",
+   "threshold": 1.5707963267948966,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "leg_angles_rad[1]",
+   "unit": "rad",
+   "threshold": 1.5707963267948966,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "leg_angles_rad[2]",
+   "unit": "rad",
+   "threshold": 1.5707963267948966,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "leg_angles_rad[3]",
+   "unit": "rad",
+   "threshold": 1.5707963267948966,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "tilt_deg",
+   "unit": "deg",
+   "threshold": 10.0,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "rack_linear_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "rack_angular_speed_rad_s",
+   "unit": "rad/s",
+   "threshold": 0.2,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 113,
+   "where": "首步前",
+   "evidence_zh": "观察：inputs/run.json 的评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0、infra_error=0；log step 84 的会话输出显示交付目录为空，最后的 log step 113 仍是三种腕部旋转模式的调试测试，均返回 rack_dropped。没有正式交付的评分轨迹，无法将断点归到某一物理步骤。"
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：inputs/video.json 标明无视频，inputs/ 中没有状态日志，评分器也未提供六个节点或原生指标的结果。log step 59 的展开测试及 log step 78、108、110、113 的翻转失败属于会话测试，不能认作交付轨迹的节点进度、接触或失败现象；run.json 的 wall_s=3602、time_limit_s=3600，日志最后仍在调试且无最终文字报告，未见外部过滤或服务中断的记录。推断：本次会话在预算结束前未形成轨迹交付；依 DECISIONS.md D23，无交付的 agent 侧失败将现象和根因均留空，不能仅凭交付缺失反推交付操作错误或把预算耗尽当作根因。",
+  "unmet": {
+   "rack_unfolded": null,
+   "legs_opened_1": null,
+   "legs_opened_2": null,
+   "legs_opened_3": null,
+   "legs_opened_4": null,
+   "rack_standing": null
+  }
+ },
+ "summary_zh": "观察：评分器判定失败，记录 missing_trajectory=1、n_actions=0、sim_steps=0；六个节点均无可判定的正式评分终值，确认完成数为0，未知数为6。观察：log step 59 的会话测试曾报告主架和四腿展开，但最后的 log step 113 仍在测试翻转，不能把这些测试计作交付进度。推断：断点记为 unknown／首步前；无视频、无状态日志且无交付轨迹，现象与根因按无交付规则留空。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "无交付（missing_trajectory=1、n_actions=0）、无视频、无状态日志；六节点 unknown、断点 unknown/首步前、现象与根因留空并写 gap_zh，符合 D23 与 03 的无交付规则；会话测试未被当作进度。",
+    "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": "2026-10-30T22:17:08Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/dexhand-drying-rack-unfold-i00-privileged--vhjnad5/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-drying-rack-unfold-i00-privileged--vhjnad5/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "9ca0cbfde23bf1714e7ac8302fd788b23d308a4c6c53948007d7a5f058118e1d",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/drying_rack_unfold/trajectories/dexhand-drying-rack-unfold-i00-privileged--ytjidgw.json

@@ -0,0 +1,295 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-drying-rack-unfold-i00-privileged--ytjidgw",
+ "taskKey": "dexhand/drying_rack_unfold",
+ "dag_sha256": "044da74935ab7bbd4a0b895fad36baee09e495811264f09ed418f743b236ddb1",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.761,
+   "peak_processes": 10,
+   "peak_disk_gb": 0.271,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 6,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "rack_unfolded": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 453,
+      "text_zh": "观察：评分器 reward 记录 missing_trajectory=1、n_actions=0、sim_steps=0，未提供“展开主架”的节点结果；log step 453 显示交付目录为空。输入无视频、无状态日志，无法判断本节点；log step 463、465 的本地测试快照不属于交付轨迹。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1，未交付受评轨迹；会话测试中的手与晾衣架接触不计入轨迹级接触。"
+     }
+    ]
+   },
+   "legs_opened_1": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 453,
+      "text_zh": "观察：评分器 reward 记录 missing_trajectory=1、n_actions=0、sim_steps=0，未提供“第1条腿展开”的节点结果；log step 453 显示交付目录为空。输入无视频、无状态日志，无法判断本节点；log step 463、465 的本地测试快照不属于交付轨迹。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1，未交付受评轨迹；会话测试中的手与晾衣架接触不计入轨迹级接触。"
+     }
+    ]
+   },
+   "legs_opened_2": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 453,
+      "text_zh": "观察：评分器 reward 记录 missing_trajectory=1、n_actions=0、sim_steps=0，未提供“第2条腿展开”的节点结果；log step 453 显示交付目录为空。输入无视频、无状态日志，无法判断本节点；log step 463、465 的本地测试快照不属于交付轨迹。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1，未交付受评轨迹；会话测试中的手与晾衣架接触不计入轨迹级接触。"
+     }
+    ]
+   },
+   "legs_opened_3": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 453,
+      "text_zh": "观察：评分器 reward 记录 missing_trajectory=1、n_actions=0、sim_steps=0，未提供“第3条腿展开”的节点结果；log step 453 显示交付目录为空。输入无视频、无状态日志，无法判断本节点；log step 463、465 的本地测试快照不属于交付轨迹。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1，未交付受评轨迹；会话测试中的手与晾衣架接触不计入轨迹级接触。"
+     }
+    ]
+   },
+   "legs_opened_4": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 453,
+      "text_zh": "观察：评分器 reward 记录 missing_trajectory=1、n_actions=0、sim_steps=0，未提供“第4条腿展开”的节点结果；log step 453 显示交付目录为空。输入无视频、无状态日志，无法判断本节点；log step 463、465 的本地测试快照不属于交付轨迹。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1，未交付受评轨迹；会话测试中的手与晾衣架接触不计入轨迹级接触。"
+     }
+    ]
+   },
+   "rack_standing": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 453,
+      "text_zh": "观察：评分器 reward 记录 missing_trajectory=1、n_actions=0、sim_steps=0，未提供“晾衣架正放”的节点结果；log step 453 显示交付目录为空。输入无视频、无状态日志，无法判断本节点；log step 463、465 的本地测试快照不属于交付轨迹。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1，未交付受评轨迹；会话测试中的手与晾衣架接触不计入轨迹级接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "fold_angle_rad",
+   "unit": "rad",
+   "threshold": 0.3141592653589793,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "leg_angles_rad[0]",
+   "unit": "rad",
+   "threshold": 1.5707963267948966,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "leg_angles_rad[1]",
+   "unit": "rad",
+   "threshold": 1.5707963267948966,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "leg_angles_rad[2]",
+   "unit": "rad",
+   "threshold": 1.5707963267948966,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "leg_angles_rad[3]",
+   "unit": "rad",
+   "threshold": 1.5707963267948966,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "tilt_deg",
+   "unit": "deg",
+   "threshold": 10.0,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "rack_linear_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "rack_angular_speed_rad_s",
+   "unit": "rad/s",
+   "threshold": 0.2,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 453,
+   "where": "首步前",
+   "evidence_zh": "观察：评分器 reward 为 missing_trajectory=1、n_actions=0、sim_steps=0、success=0。log step 32 仅保存并回放临时的5步保持动作测试；log step 453 显示交付目录为空，log step 463 保存的是测试快照，log step 465 仍在恢复该快照检查姿态，未见后续交付记录。没有受评轨迹可定位失败的物理步骤。"
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：inputs/ 只有 MANIFEST.json、meta.json、run.json、video.json；video.kind=none，无视频和状态日志，评分器未报告逐节点终值或原生物理指标。日志是多轮本地测试而非交付轨迹，不能据其角度、接触或失败输出标注受评节点和现象。run.wall_s=3612、time_limit_s=3600；最后一条 agent 自述仍在等待测试（log step 457），随后日志止于快照检查（log step 465），没有最终完成报告或足以证明缺交付具体原因的记录。按未交付的处理规则，现象和根因均留空，不从交付缺失反推控制、规划或交付操作错误。",
+  "unmet": {
+   "rack_unfolded": null,
+   "legs_opened_1": null,
+   "legs_opened_2": null,
+   "legs_opened_3": null,
+   "legs_opened_4": null,
+   "rack_standing": null
+  }
+ },
+ "summary_zh": "观察：评分器记录未交付轨迹（missing_trajectory=1、n_actions=0、sim_steps=0），判定 success=0；六个语义节点均无可核验的受评终值，全部保留未知。观察：log step 453 的交付目录为空，末尾 log step 463—465 仍是本地测试快照的保存与检查，故断点记为 unknown／首步前。观察：无视频、状态日志或逐节点评分结果，不能把会话测试的进展和接触记入受评轨迹，失败现象与根因留空。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "无视频、无状态日志、评分器 missing_trajectory=1；交付目录为空（log step 453），日志只有本地测试。六节点 unknown、断点 unknown/首步前、现象与根因留空并写 gap_zh，符合规则，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": "2026-10-30T22:17:29Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/dexhand-drying-rack-unfold-i00-privileged--ytjidgw/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-drying-rack-unfold-i00-privileged--ytjidgw/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "905db2802ea5eab6226b97f9c79c03fea077b2d35fde33372e7c53479db2b1ba",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/envelope_letter_insert/trajectories/dexhand-envelope-letter-insert-i00-privileged--egfrdbr.json

@@ -0,0 +1,193 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-envelope-letter-insert-i00-privileged--egfrdbr",
+ "taskKey": "dexhand/envelope_letter_insert",
+ "dag_sha256": "abdfffef3130431c83137bcf37b304b96623a00fa3c230811aac4c55022c6987",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 1.946,
+   "peak_processes": 8,
+   "peak_disk_gb": 0.253,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 1,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "letter_in_envelope": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 59,
+      "text_zh": "观察：评分器 reward 记录 missing_trajectory=1、n_actions=0、sim_steps=0；inputs/video.json 的 kind=none，且本次无状态日志。log step 59 为 Gemini 过滤器中断消息，没有可用于判断信纸插入终态的已交付轨迹。"
+     },
+     {
+      "t": null,
+      "log_step": 58,
+      "text_zh": "观察：log step 21 的 agent 测试服务在执行每个脚本前调用 s.reset()；log step 55、56、57、58 的本地测试打印 success=True，log step 58 打印 stage=3、released_hold_steps=50。推断：这些属于会话测试结果，评分器仍报告轨迹缺失，不能据此给受评节点赋真值或视频时间。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告 missing_trajectory=1、n_actions=0；未交付受评轨迹，会话测试中的右手接触不作为本节点的实际接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "outside_m",
+   "unit": "m",
+   "threshold": 0.00495,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "letter_linear_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "envelope_linear_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "letter_angular_speed_rad_s",
+   "unit": "rad/s",
+   "threshold": 0.2,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "envelope_angular_speed_rad_s",
+   "unit": "rad/s",
+   "threshold": 0.2,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "envelope_shift_m",
+   "unit": "m",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "envelope_tilt_rad",
+   "unit": "rad",
+   "threshold": 0.35,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 59,
+   "where": "首步前",
+   "evidence_zh": "观察：log step 59 的末条消息明确为“This request was blocked by Gemini's filters”，其后无工具调用；评分器报告 missing_trajectory=1、n_actions=0、sim_steps=0。推断：供应商过滤器在轨迹交付前中断会话，无受评物理步骤可定位，按外因中断记录 unknown／首步前。"
+  },
+  "unmet": {
+   "letter_in_envelope": null
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：本次没有受评视频或状态日志，评分器未收到轨迹，也未报告节点终态或 DAG 原生连续指标；log step 59 记录供应商过滤器中断。log step 55 至 log step 58 的成功状态来自 agent 的会话测试，不能替代已交付轨迹的证据。推断：依据 CASES.md FAIL-5，属于交付前外因中断，应重测并排除机器人失败分析，故现象与根因均为空，不标“没尝试”或“交付操作出错”；节点及全部原生指标保持未知。inputs/oracle/ 的成功参考属于另一运行，不能证明本次结果。"
+ },
+ "summary_zh": "观察：评分器记录 success=0、missing_trajectory=1、n_actions=0，唯一节点“信纸插入信封”终态未知；log step 55 至 log step 58 虽有本地测试成功读数，但无已交付轨迹支持节点完成。观察：log step 59 以 Gemini 过滤器拦截消息结束。推断：这是交付前外因中断，按 unknown／首步前记录并建议重测，现象和根因留空；本次无视频、无状态日志，无法确认受评物理进展及接触。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "无视频、无状态日志，评分器 missing_trajectory=1、n_actions=0；log step 59 为 Gemini 过滤器拦截且无已交付轨迹，按 FAIL-5 标 unknown/首步前、现象与根因为空并有 gap_zh，标注正确。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T22:19:27Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/dexhand-envelope-letter-insert-i00-privileged--egfrdbr/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-envelope-letter-insert-i00-privileged--egfrdbr/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "dc0cb8177090788a3597140055d80b0ef0a6cec5c1c424280d23430a0f0fbcc9",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/envelope_letter_insert/trajectories/dexhand-envelope-letter-insert-i00-privileged--hrc5yy6.json

@@ -0,0 +1,193 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-envelope-letter-insert-i00-privileged--hrc5yy6",
+ "taskKey": "dexhand/envelope_letter_insert",
+ "dag_sha256": "abdfffef3130431c83137bcf37b304b96623a00fa3c230811aac4c55022c6987",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 2.036,
+   "peak_processes": 8,
+   "peak_disk_gb": 0.239,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 1,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "letter_in_envelope": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器字段为 missing_trajectory=1、n_actions=0、sim_steps=0、success=0，未提供本节点的物理终态；本次无受评视频及状态日志，无法判定信纸插入并释放静置的节点值。"
+     },
+     {
+      "t": null,
+      "log_step": 29,
+      "text_zh": "观察：log step 29 是 agent 新建 Session 后运行控制器的接触诊断，输出右手与桌面及手指间接触；没有交付或受评回放记录，不能用该会话测试确定本节点的进度或实际接触。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1、n_actions=0，未交付轨迹；会话测试中右手的接触不计入受评轨迹。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "outside_m",
+   "unit": "m",
+   "threshold": 0.00495,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "letter_linear_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "envelope_linear_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "letter_angular_speed_rad_s",
+   "unit": "rad/s",
+   "threshold": 0.2,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "envelope_angular_speed_rad_s",
+   "unit": "rad/s",
+   "threshold": 0.2,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "envelope_shift_m",
+   "unit": "m",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "envelope_tilt_rad",
+   "unit": "rad",
+   "threshold": 0.35,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 29,
+   "where": "首步前",
+   "evidence_zh": "观察：评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0；最后一个日志条目 log step 29 仍为 agent 控制器的会话内接触诊断，没有交付记录或受评物理片段可定位失败步骤。"
+  },
+  "unmet": {
+   "letter_in_envelope": null
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：本次没有受评视频、状态日志或交付轨迹，评分器未报告信纸插入节点的终态及各项连续指标；inputs/oracle/replay.mp4 属于另一成功运行，不能证明本次进度。log steps 20–29 仅记录会话内测试与调试，不能作为受评失败现象；现有日志没有足够证据确定未交付的具体根因，依 DECISIONS.md D23 将现象与根因均留空。"
+ },
+ "summary_zh": "评分器记录未交付轨迹（missing_trajectory=1），受评动作数与仿真步数均为 0；唯一节点“信纸插入信封”的终态未知，不能由总成功标记为 0 推断其物理状态。log steps 20–29 显示 agent 仍在调试右臂及右手控制器，但这些是会话测试；断点保留 unknown／首步前，无受评视频或状态日志可支持物理失败现象与具体根因。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "本次无受评视频/状态日志，评分器 missing_trajectory=1、n_actions=0；agent 在 3600 s 预算内只做会话调试、未交付。标注按 D23(a) 留空现象与根因并写 gap_zh，节点 unknown、不把会话测试或超时当现象，符合规则，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": "2026-10-30T22:18:27Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/dexhand-envelope-letter-insert-i00-privileged--hrc5yy6/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-envelope-letter-insert-i00-privileged--hrc5yy6/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "d24709a2a6dd5e54be5847a2bd45cc6094aa22fa3c724f6655963689dc68f497",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/envelope_letter_insert/trajectories/dexhand-envelope-letter-insert-i00-privileged--t8z5pb5.json

@@ -0,0 +1,193 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-envelope-letter-insert-i00-privileged--t8z5pb5",
+ "taskKey": "dexhand/envelope_letter_insert",
+ "dag_sha256": "abdfffef3130431c83137bcf37b304b96623a00fa3c230811aac4c55022c6987",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 3.856,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.267,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 1,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "letter_in_envelope": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器字段为 missing_trajectory=1、n_actions=0、sim_steps=0、success=0，未提供信纸插入节点的阶段或终态读数；inputs/video.json 为 kind=none，且本次没有状态日志，无法判定节点终态。"
+     },
+     {
+      "t": null,
+      "log_step": 347,
+      "text_zh": "观察：log step 347 中 agent 启动的是新建 Session 的推送测试脚本，工具只返回后台任务已启动；log step 366 的最后一条自述仍表示等待几何诊断结果，log step 368 继续检查指尖最低点且无返回输出。这些会话测试不能作为已交付受评轨迹的插入证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告 missing_trajectory=1、n_actions=0、sim_steps=0，没有交付轨迹，故无可标注的受评右手与信纸或信封接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "outside_m",
+   "unit": "m",
+   "threshold": 0.00495,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "letter_linear_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "envelope_linear_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "letter_angular_speed_rad_s",
+   "unit": "rad/s",
+   "threshold": 0.2,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "envelope_angular_speed_rad_s",
+   "unit": "rad/s",
+   "threshold": 0.2,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "envelope_shift_m",
+   "unit": "m",
+   "threshold": 0.05,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "envelope_tilt_rad",
+   "unit": "rad",
+   "threshold": 0.35,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 368,
+   "where": "首步前",
+   "evidence_zh": "观察：评分器报告 missing_trajectory=1、n_actions=0、sim_steps=0；log step 368 的最后调用仍是 agent 的指尖几何诊断，没有返回结果或交付记录。没有受评轨迹，无法定位信纸插入过程中的物理断点，按首步前记录。"
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：本次没有视频（inputs/video.json: kind=none），inputs/ 中也没有状态日志；评分器没有节点阶段、接触或原生几何及速度指标。日志的最后自述（log step 366）仍在等待诊断，末次调用（log step 368）无返回；run.json 记录 wall_s=3606、time_limit_s=3600，但未明确记录终止机制。推断：会话持续测试直至预算附近而未交付；仅凭未交付不能确定控制、规划或保存操作的根因。依 DECISIONS.md D23，未交付时不把会话测试转换为受评失败现象，现象与根因均留空；依 N9，节点终态保留未知，所有时间区间为 null。成功参考视频属于另一次运行，不能补足本次执行证据。",
+  "unmet": {
+   "letter_in_envelope": null
+  }
+ },
+ "summary_zh": "观察：评分器报告 success=0、missing_trajectory=1，受评动作数和仿真步数均为 0；唯一的“信纸插入信封”节点没有可确认的完成记录，终态保留未知。观察：agent 在 log step 347 仍启动推送测试，log step 366 自述等待诊断，log step 368 继续检查指尖几何且无输出；未交付使断点只能记为 unknown／首步前。观察：本次无视频和状态日志，无法确定实际接触、物理失败现象或根因，两层均留空。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "无视频、无状态日志；评分器 missing_trajectory=1、n_actions=0，未交付。节点 unknown、breakpoint unknown/首步前、两层为空并带 gap_zh，符合 D23 与 03 的无交付规则；未把会话测试当作受评证据。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T22:17:24Z",
+  "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/dexhand-envelope-letter-insert-i00-privileged--t8z5pb5/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-envelope-letter-insert-i00-privileged--t8z5pb5/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "d2f8814a5ef7af1b79225578ba2832ba4957d0c7a5dc29ddb11709e319d563c5",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/essential_cube_reorient/trajectories/dexhand-essential-cube-reorient-i00-privileged--d4kpmt7.json

@@ -0,0 +1,266 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-essential-cube-reorient-i00-privileged--d4kpmt7",
+ "taskKey": "dexhand/essential_cube_reorient",
+ "dag_sha256": "6bc87a1901a74bc490999f5e2ecd2556658f7943851231829bff51c64a7e8707",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.635,
+   "peak_processes": 6,
+   "peak_disk_gb": 0.253,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 3,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "sphere_first_orientation": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 29,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告本节点的阶段结果；inputs/video.json 的 kind=none，且未提供状态日志。log step 29 的调试脚本输出 stg=1 属独立会话测试，不是已交付评分轨迹，不能据此确认首轴检查点；本节点终值未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告 missing_trajectory=1，没有交付评分轨迹；会话测试中的手指与球接触不计为本次轨迹的实际接触。"
+     }
+    ]
+   },
+   "sphere_second_orientation": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 40,
+      "text_zh": "观察：评分器仅报告 missing_trajectory=1、n_actions=0、sim_steps=0，未报告次轴检查点结果；本次没有视频或状态日志。log step 40 的调试脚本曾输出 stg=2，但该会话没有交付为评分轨迹，不能据此确认本节点；终值未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告 missing_trajectory=1，没有交付评分轨迹；会话测试中的手指与球接触不计为本次轨迹的实际接触。"
+     }
+    ]
+   },
+   "sphere_initial_orientation": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 67,
+      "text_zh": "观察：评分器报告 missing_trajectory=1、n_actions=0、sim_steps=0，未报告第三阶段或末段保持结果；本次没有视频或状态日志。log step 67 仍在独立测试反向转动，没有交付后的最终状态或验收结果，不能判断本节点终值。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告 missing_trajectory=1，没有交付评分轨迹；会话测试中的手指与球接触不计为本次轨迹的实际接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "cube_error_1",
+   "unit": "rad",
+   "threshold": 0.16,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "cube_error_2",
+   "unit": "rad",
+   "threshold": 0.16,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "cube_error_3",
+   "unit": "rad",
+   "threshold": 0.16,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "contacts",
+   "unit": "fingers",
+   "threshold": 2,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "speed_stage",
+   "unit": "m/s or rad/s",
+   "threshold": 0.25,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "speed_terminal",
+   "unit": "m/s or rad/s",
+   "threshold": 0.06,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "force",
+   "unit": "N",
+   "threshold": 80,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "speed_invariant",
+   "unit": "m/s or rad/s",
+   "threshold": 12,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "palm_drift",
+   "unit": "m",
+   "threshold": 0.003,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "wrist_rotation",
+   "unit": "rad",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "object_displacement",
+   "unit": "m",
+   "threshold": 0.07,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "object_height",
+   "unit": "m",
+   "threshold": [
+    0.18,
+    0.6
+   ],
+   "direction": "target",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": null,
+   "where": "首步前",
+   "evidence_zh": "观察：inputs/run.json 的评分器结果为 missing_trajectory=1、n_actions=0、sim_steps=0、success=0、infra_error=0；log step 67 结束于会话测试的数值输出，未见提交轨迹或成功验收记录。推断：没有可定位物理失败步骤的评分轨迹，按无交付处理，断点为 unknown，位置为首步前。"
+  },
+  "unmet": {
+   "sphere_first_orientation": null,
+   "sphere_second_orientation": null,
+   "sphere_initial_orientation": null
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "本次无视频（inputs/video.json: kind=none）、无状态日志，评分器明确报告未交付轨迹；三个节点均无对应的评分终值字段，因此 intervals 和 final 均为 null。inputs/oracle/replay.mp4 是另一次成功运行，不能证明本次进度或接触。log step 29、40 的阶段通过和 log step 41、63、66 的失去接触都是会话测试，不能移作评分尝试的现象。运行 wall_s=3603，预算 time_limit_s=3600，但日志末条为 log step 67（agent 时间 1701 s），未记载此后结束机制，也没有足够证据确定缺交付的具体原因或外部中断。按 DECISIONS.md D23 保留空现象、空根因，不能从缺交付反推交付操作出错或没尝试。"
+ },
+ "summary_zh": "评分器报告 missing_trajectory=1、n_actions=0、sim_steps=0，本次未交付可评分轨迹；三个节点均无可确认的终值，断点为 unknown／首步前。日志中的首轴、次轴阶段通过及反向转动测试仅为会话证据，不能计入本次评分进度或失败现象；没有视频、状态日志和节点级评分结果，现象与根因按 D23 留空。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "本次无视频、无状态日志，评分器报告 missing_trajectory=1、n_actions=0；标注把三个节点留 unknown、会话测试不计入进度、现象与根因留空并写 gap_zh，符合 D23(a)，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": "2026-10-30T22:17:19Z",
+  "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/dexhand-essential-cube-reorient-i00-privileged--d4kpmt7/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-essential-cube-reorient-i00-privileged--d4kpmt7/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "95fc1cc878024aa2201e961bf331b598e375618f8126b6f71a6ac53974ae737e",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/essential_cube_reorient/trajectories/dexhand-essential-cube-reorient-i00-standard--iowksk7.json

@@ -0,0 +1,281 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-essential-cube-reorient-i00-standard--iowksk7",
+ "taskKey": "dexhand/essential_cube_reorient",
+ "dag_sha256": "6bc87a1901a74bc490999f5e2ecd2556658f7943851231829bff51c64a7e8707",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "calls": 608,
+   "deterministic": 1,
+   "finalized": 1,
+   "hand_wuji": 0,
+   "infra_error": 0,
+   "live_matches_replay": 1,
+   "live_success": 0,
+   "n_actions": 6802,
+   "no_verdict": 0,
+   "replay_success": 0,
+   "reset_budget": 20,
+   "resets": 8,
+   "resume_state_mismatch": 0,
+   "resumes": 0,
+   "sim_error": 0,
+   "sim_startups": 1,
+   "sim_steps": 6802,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 0.767,
+   "peak_processes": 10,
+   "peak_disk_gb": 2.023,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "log",
+  "denominator": 3,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "sphere_first_orientation": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 92,
+      "text_zh": "观察：log step 91 的 agent 脚本按相机标记计算首轴误差并保持，log step 92 输出 HOLDERR=[-0.04199836,0.04019859,-0.09329607]，服务状态为 episode=9、control_steps=1695、done=False；该误差是 agent 自行计算的结果，不是评分器的 cube_error_1 或阶段验收。没有本次视频、状态日志或原生 stage_index、指尖接触及速度窗口记录，无法确认首检查点是否完成。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：log step 72 重置后，log step 79、87 的 agent 脚本向右手手指提交首轴滚动目标；本次无视频或接触日志，无法核实哪些指尖或掌面实际接触蓝色球及接触方式。"
+     }
+    ]
+   },
+   "sphere_second_orientation": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 112,
+      "text_zh": "观察：log step 92、99、110 的 agent 脚本运行第二目标控制，log step 100 的自算 TRACKERR 曾降至约0.86 rad，log step 112 的自算误差又增至约2.06 rad；这些均不是原生 cube_error_2 或 stage_index。没有本次视频、状态日志或服务逐阶段结果，既不能确认首检查点前置已通过，也不能据这些估计给第二检查点定真或定假。"
+     },
+     {
+      "t": null,
+      "log_step": 125,
+      "text_zh": "观察：log step 121 至125 的 agent 脚本继续提交拇指和食指旋转目标；log step 125 的服务状态为 episode=9、control_steps=6802、done=False、reason=None。agent 在 log step 126 自述第二检查点仍未收敛，但服务没有报告该节点的完成值。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：log step 101、107、115、121 的 agent 脚本先后改变右手拇指、食指和中指目标以尝试第二轴转动；无本次图像或原生接触对，实际指尖接触、掌面承托及是否发生滑动均无法核实。"
+     }
+    ]
+   },
+   "sphere_initial_orientation": {
+    "method": "log",
+    "intervals": null,
+    "final": false,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 125,
+      "text_zh": "观察：log step 125 的服务仍报告 episode=9、control_steps=6802、done=False；inputs/run.json 的评分字段 n_actions=6802、sim_steps=6802、finalized=1、live_matches_replay=1，且 live_success=0、replay_success=0、no_verdict=0。该控制步数与最后回合一致。"
+     },
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "推断：DAG 的末节点验收包含前两检查点、第三阶段、terminal_hold>=25 以及未失败，完整成立即触发本任务成功；评分器明确给出最终 live_success=0 且回放一致，因此该完整终态验收为假。此推断只决定末节点，不据失败总结果反推前两个历史检查点未完成。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "cube_error_1",
+   "unit": "rad",
+   "threshold": 0.16,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "cube_error_2",
+   "unit": "rad",
+   "threshold": 0.16,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "cube_error_3",
+   "unit": "rad",
+   "threshold": 0.16,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "contacts",
+   "unit": "fingers",
+   "threshold": 2,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "speed_stage",
+   "unit": "m/s or rad/s",
+   "threshold": 0.25,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "speed_terminal",
+   "unit": "m/s or rad/s",
+   "threshold": 0.06,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "force",
+   "unit": "N",
+   "threshold": 80,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "speed_invariant",
+   "unit": "m/s or rad/s",
+   "threshold": 12,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "palm_drift",
+   "unit": "m",
+   "threshold": 0.003,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "wrist_rotation",
+   "unit": "rad",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "object_displacement",
+   "unit": "m",
+   "threshold": 0.07,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "object_height",
+   "unit": "m",
+   "threshold": [
+    0.18,
+    0.6
+   ],
+   "direction": "target",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 126,
+   "where": "操作中",
+   "evidence_zh": "观察：log step 72 的 agent 脚本作最后一次重置，之后服务持续报告 episode=9；log step 125 的 control_steps=6802 与评分记录一致。log step 92 至93 中 agent 仅凭自算约0.11 rad 的首轴误差转向第二目标，随后持续重抓和旋转至 log step 125，log step 126 自述第二检查点仍未收敛。推断：最后回合确实已开始手内重定向操作，但缺少原生阶段确认，无法排除首检查点从未通过，因此不能把第二目标的反复调整直接认作最早未修复的失败节点。"
+  },
+  "unmet": {
+   "sphere_first_orientation": null,
+   "sphere_second_orientation": null,
+   "sphere_initial_orientation": false
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "本次 inputs/video.json 为 kind=none，且没有状态日志；inputs/oracle/replay.mp4 是另一次成功，仅供参考，不能替代本次证据。所有视频时刻与区间均留空。服务未发布 stage_index、stage_hold、terminal_hold、原生朝向误差或接触力/速度窗口，首两节点和最早未修复断点无法确认；末节点仅按其完整验收与评分器成功判据的对应关系定为假。log step 92、100、112、122至126 的朝向、位置、IK与接触点估计来自 agent 自己的脚本，不能代替实际物理观测；因此无法确定断点处是姿态偏差、指尖接触不足、滑移还是保持窗口未通过，现象层留空，也不足以区分控制、接触设计或感知误差等根因，根因层留空。log step 80 的脚本异常在 log step 81 已修复并继续，不当作最终失败原因；前八回合的 task_failure 不作为最后评分回合的失败现象。"
+ },
+ "summary_zh": "观察：最后第9回合的服务控制步数为6802，与评分记录一致；评分器确认实时与回放均失败且无基础设施错误。agent 在首轴相机估计接近目标后改做第二轴并多次调整，但首两检查点缺少原生验收证据，均为未知；包含完整有序阶段及末段保持的末节点由失败判定推断为假，最早未修复节点无法唯一定位。本次没有视频或状态日志，实际接触记未知，失败现象与根因因证据不足留空。",
+ "conf": 5,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "本次无视频、无状态日志、无原生 stage_index/cube_error；标注把前两节点记为 unknown、末节点按评分器 live_success=0 定为假、断点 unknown、现象与根因留空并写明 gap_zh，均有依据，未发现被证据否定或遗漏的结论。",
+    "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": "2026-10-30T22:23:31Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/dexhand-essential-cube-reorient-i00-standard--iowksk7/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-essential-cube-reorient-i00-standard--iowksk7/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "6f63e853acebb32dd8e9e6e29ec528dc45df76fb720b64a176109752cdbe2ee7",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/essential_pen_regrasp/trajectories/dexhand-essential-pen-regrasp-i00-privileged--7wen6cp.json

@@ -0,0 +1,244 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-essential-pen-regrasp-i00-privileged--7wen6cp",
+ "taskKey": "dexhand/essential_pen_regrasp",
+ "dag_sha256": "5d4eb54b0f602d691da20e76dd6190aca4275c6bc50b8abc6afeb7e5ecb805d2",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.824,
+   "peak_processes": 11,
+   "peak_disk_gb": 0.276,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 2,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "pen_advanced": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器 reward 记录 missing_trajectory=1、n_actions=0、sim_steps=0；未报告受评轨迹的推进阶段结果，本次无视频、无状态日志。"
+     },
+     {
+      "t": null,
+      "log_step": 557,
+      "text_zh": "观察：log step 557 中 agent 脚本新建 Session 执行推进测试，打印 stage_index=1，但只将控制量放入内存列表，未交付轨迹；该会话测试不能确定受评节点终值。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "评分器记录 missing_trajectory=1，未交付受评轨迹；会话推进测试的右手与笔接触不计入本轨迹。"
+     }
+    ]
+   },
+   "pen_tripod_grip": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器 reward 记录 missing_trajectory=1、success=0，未提供本节点的受评阶段或末态字段；无视频、无状态日志，无法判定三指换握终值。"
+     },
+     {
+      "t": null,
+      "log_step": 593,
+      "text_zh": "观察：log step 593 中 agent 脚本新建 Session 进行换握后保持测试，打印握带接触 th=1、idx=0、mid=1；这是会话测试输出，没有对应交付，不能作为受评节点值或失败现象。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "评分器记录 missing_trajectory=1，未交付受评轨迹；会话换握测试的右手与笔接触不计入本轨迹。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "pen_axial",
+   "value": null,
+   "unit": "m",
+   "threshold": [
+    0.024,
+    0.036
+   ],
+   "direction": "target"
+  },
+  {
+   "name": "pen_angle_advance",
+   "value": null,
+   "unit": "rad",
+   "threshold": [
+    0,
+    0.2
+   ],
+   "direction": "target"
+  },
+  {
+   "name": "pen_angle_tripod",
+   "value": null,
+   "unit": "rad",
+   "threshold": [
+    0.65,
+    0.92
+   ],
+   "direction": "target"
+  },
+  {
+   "name": "speed_terminal",
+   "value": null,
+   "unit": "m/s or rad/s",
+   "threshold": 0.06,
+   "direction": "lower_better"
+  },
+  {
+   "name": "force",
+   "value": null,
+   "unit": "N",
+   "threshold": 80,
+   "direction": "lower_better"
+  },
+  {
+   "name": "object_displacement",
+   "value": null,
+   "unit": "m",
+   "threshold": 0.07,
+   "direction": "lower_better"
+  },
+  {
+   "name": "object_height",
+   "value": null,
+   "unit": "m",
+   "threshold": [
+    0.18,
+    0.6
+   ],
+   "direction": "target"
+  },
+  {
+   "name": "palm_drift",
+   "value": null,
+   "unit": "m",
+   "threshold": 0.003,
+   "direction": "lower_better"
+  },
+  {
+   "name": "wrist_rotation",
+   "value": null,
+   "unit": "rad",
+   "threshold": 0.01,
+   "direction": "lower_better"
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 593,
+   "where": "首步前",
+   "evidence_zh": "观察：评分器 reward 为 missing_trajectory=1、n_actions=0、sim_steps=0、success=0；完整日志末条 log step 593 仍是 agent 新建 Session 后的握持测试，未记录轨迹交付。推断：无受评动作可定位物理失败节点，断点置 unknown、位置首步前。"
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：inputs/video.json 为 kind=none，inputs/ 中无本次 replay.mp4 或 states.jsonl，评分器也未给出逐节点结果；存在的 inputs/oracle/replay.mp4 属于另一成功运行，不能证明本次进度。log step 557、577、589、593 的阶段与接触记录均来自 agent 的独立会话测���，无交付轨迹可核对。运行 wall_s=3606、time_limit_s=3600，日志在测试中结束，但没有明确的终止事件或保存失败记录。推断：不能把会话推进成功或换握缺指接触作为受评进度、现象，也不能由缺交付或预算耗尽推断控制、认知或交付操作根因；按 D23 保留两层为空，按 N9 保留节点终值未知，全部视频时刻为空。",
+  "unmet": {
+   "pen_advanced": null,
+   "pen_tripod_grip": null
+  }
+ },
+ "summary_zh": "评分器记录 success=0、missing_trajectory=1，受评动作数与仿真步数均为 0；两个节点均无可确认的受评终值。会话中 agent 曾测试通过推进阶段（log step 557），末条仍在测试换握稳定性（log step 593），但没有交付轨迹，断点记 unknown／首步前。无本次视频、状态日志或逐节点评分字段，按 D23、N9 保留现象与根因为空，不将会话测试计入进度。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "本次无视频、无状态日志，评分器 missing_trajectory=1、n_actions=0；run.json 全部步骤中没有任何交付/提交调用。两节点 unknown、断点 unknown/首步前、现象与根因为空并附 gap_zh，符合 D23(a) 与 N9；会话测试未被当作受评进度，也没有时钟换算。",
+    "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": "2026-10-30T22:18:21Z",
+  "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/dexhand-essential-pen-regrasp-i00-privileged--7wen6cp/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-essential-pen-regrasp-i00-privileged--7wen6cp/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "0c9328f6fa4f6b650dcf2178fc26fed1664896b75fa89c3da34d892149048222",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/essential_pen_regrasp/trajectories/dexhand-essential-pen-regrasp-i00-privileged--e2krr4p.json

@@ -0,0 +1,234 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-essential-pen-regrasp-i00-privileged--e2krr4p",
+ "taskKey": "dexhand/essential_pen_regrasp",
+ "dag_sha256": "5d4eb54b0f602d691da20e76dd6190aca4275c6bc50b8abc6afeb7e5ecb805d2",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.612,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.256,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 2,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "pen_advanced": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 44,
+      "text_zh": "观察：inputs/run.json 的评分器字段为 missing_trajectory=1、n_actions=0、sim_steps=0，未报告受评推进阶段；inputs/video.json 为 kind=none，亦无状态日志。log step 42、44 的 stage=1 来自 agent 的独立参数测试，未交付，不能用于本节点终值。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "评分器记录 missing_trajectory=1，未交付受评轨迹；会话测试中的手指与笔接触不计入本次受评接触。"
+     }
+    ]
+   },
+   "pen_tripod_grip": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 67,
+      "text_zh": "观察：inputs/run.json 的评分器字段为 missing_trajectory=1、n_actions=0、sim_steps=0，未报告受评转向三指握持阶段或末段保持计数；本次无视频和状态日志。log step 53 的三指握带接触及 log step 67 的 stage=1 均属于会话测试，不是交付结果。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "评分器记录 missing_trajectory=1，未交付受评轨迹；会话测试中的手指与笔接触不计入本次受评接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "pen_axial",
+   "unit": "m",
+   "threshold": [
+    0.024,
+    0.036
+   ],
+   "direction": "target",
+   "value": null
+  },
+  {
+   "name": "pen_angle_advance",
+   "unit": "rad",
+   "threshold": [
+    0,
+    0.2
+   ],
+   "direction": "target",
+   "value": null
+  },
+  {
+   "name": "pen_angle_tripod",
+   "unit": "rad",
+   "threshold": [
+    0.65,
+    0.92
+   ],
+   "direction": "target",
+   "value": null
+  },
+  {
+   "name": "speed_terminal",
+   "unit": "m/s or rad/s",
+   "threshold": 0.06,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "force",
+   "unit": "N",
+   "threshold": 80,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "object_displacement",
+   "unit": "m",
+   "threshold": 0.07,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "object_height",
+   "unit": "m",
+   "threshold": [
+    0.18,
+    0.6
+   ],
+   "direction": "target",
+   "value": null
+  },
+  {
+   "name": "palm_drift",
+   "unit": "m",
+   "threshold": 0.003,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "wrist_rotation",
+   "unit": "rad",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 83,
+   "where": "首步前",
+   "evidence_zh": "观察：评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0、success=0。最后的 log step 83 中，agent 自述仍在等待 test_lead.py，工具仅返回后台任务 Status: DONE，没有轨迹保存或提交结果。推断：没有交付可定位物理失败步骤的受评轨迹，断点保留 unknown，位置为首步前。"
+  },
+  "unmet": {
+   "pen_advanced": null,
+   "pen_tripod_grip": null
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：本次无受评视频、无状态日志，评分器明确报告缺少轨迹，且没有两个节点的阶段判定或原生指标。log step 42、44、46、53、67 记录的是 agent 会话中的测试，不能当作受评进度、实际接触或失败现象；成功 oracle 是另一运行，也不能填补本次证据。log step 83 后没有进一步执行或交付记录；run.json 的 wall_s=3603、time_limit_s=3600 说明会话用时接近预算，但未记录为何没有交付，不能仅据此指定控制、规划或交付操作根因。按 D23，未交付的现象与根因均留空。"
+ },
+ "summary_zh": "观察：评分器判定失败，missing_trajectory=1、n_actions=0、sim_steps=0；本次没有可确认完成的受评节点，沿轴推进与转向三指握持两个节点终值均为未知。观察：log step 44 的测试达到推进阶段，log step 53 的另一测试出现三指握带接触，但均没有交付记录，不能计入本次进度。推断：断点只能标为 unknown／首步前；缺少受评视频和状态日志，且日志未说明交付缺失的具体原因，故现象、根因留空。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "本次无受评视频、无状态日志，评分器 missing_trajectory=1、n_actions=0、sim_steps=0；两节点终值 unknown、会话测试不计入受评进度、现象与根因留空并写 gap_zh（D23 (a)），均与证据一致，未发现需改动之处。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T22:18:27Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/dexhand-essential-pen-regrasp-i00-privileged--e2krr4p/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-essential-pen-regrasp-i00-privileged--e2krr4p/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "d6b3ea706a217f63d3ddf1a595f64f9b54267e42be6537599ae9b4c51d069d86",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/essential_pen_regrasp/trajectories/dexhand-essential-pen-regrasp-i00-privileged--kgc9cxz.json

@@ -0,0 +1,234 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-essential-pen-regrasp-i00-privileged--kgc9cxz",
+ "taskKey": "dexhand/essential_pen_regrasp",
+ "dag_sha256": "5d4eb54b0f602d691da20e76dd6190aca4275c6bc50b8abc6afeb7e5ecb805d2",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.766,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.241,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 2,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "pen_advanced": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 39,
+      "text_zh": "观察：inputs/run.json 的评分器字段为 missing_trajectory=1、n_actions=0、sim_steps=0，未报告推进阶段是否完成；inputs/video.json 为 kind=none，且无本次运行状态日志。log step 39 是 agent 的控制器会话测试，不能据此设定评分轨迹的推进节点终值。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1，未收到可评分动作轨迹；会话测试中的右手与笔接触不计为本次评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "pen_tripod_grip": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 39,
+      "text_zh": "观察：inputs/run.json 的评分器字段为 missing_trajectory=1、n_actions=0、sim_steps=0，未报告转向三指握阶段或末段保持结果；无本次运行视频和状态日志。log step 39 的 stage=0、success=False 属于会话测试，不能代替未交付的评分轨迹节点终值。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1，未收到可评分动作轨迹；会话测试中的右手与笔接触不计为本次评分轨迹的实际接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "pen_axial",
+   "unit": "m",
+   "threshold": [
+    0.024,
+    0.036
+   ],
+   "direction": "target",
+   "value": null
+  },
+  {
+   "name": "pen_angle_advance",
+   "unit": "rad",
+   "threshold": [
+    0,
+    0.2
+   ],
+   "direction": "target",
+   "value": null
+  },
+  {
+   "name": "pen_angle_tripod",
+   "unit": "rad",
+   "threshold": [
+    0.65,
+    0.92
+   ],
+   "direction": "target",
+   "value": null
+  },
+  {
+   "name": "speed_terminal",
+   "unit": "m/s or rad/s",
+   "threshold": 0.06,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "force",
+   "unit": "N",
+   "threshold": 80,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "object_displacement",
+   "unit": "m",
+   "threshold": 0.07,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "object_height",
+   "unit": "m",
+   "threshold": [
+    0.18,
+    0.6
+   ],
+   "direction": "target",
+   "value": null
+  },
+  {
+   "name": "palm_drift",
+   "unit": "m",
+   "threshold": 0.003,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "wrist_rotation",
+   "unit": "rad",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 39,
+   "where": "首步前",
+   "evidence_zh": "观察：评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0、infra_error=0。日志止于 log step 39 的 g5 控制器会话测试；log steps 33–37 中动作数组保存路径均在会话工作目录，未见调用 save_trajectory 将动作交付到指定输出目录。推断：未形成可供定位物理断点的评分轨迹，按无交付规则记 unknown／首步前，不把测试中的失握或速度越界当作评分尝试的失败现象。"
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：本次运行没有视频、状态日志或可评分交付轨迹；评分器只报告缺轨迹及零动作，未提供任何节点阶段终值或原生物理指标。日志最后的 log step 39 仍是会话测试，没有最终文字报告；run.json 记录 wall_s=3606，任务预算为 agent 墙钟 3600 s。推断：记录符合预算结束前未交付的情况，但没有独立终止事件记录；不将预���耗尽本身标为现象或根因。依据 DECISIONS.md D23，现象及根因均留空；会话调试中的接触丢失、速度越界和脚本异常不证明评分轨迹的物理断点或缺交付的具体根因。两个节点终值保持未知，所有视频时间为空。",
+  "unmet": {
+   "pen_advanced": null,
+   "pen_tripod_grip": null
+  }
+ },
+ "summary_zh": "评分器记录 success=0、missing_trajectory=1、n_actions=0、sim_steps=0，未收到可评分轨迹；笔沿轴前移和笔转向三指握两个节点均无可确认的终值。日志结尾仍在进行控制器会话测试，断点按无交付规则记 unknown／首步前，实际接触记不适用，现象和根因留空。缺少本次运行视频、状态日志和节点级评分结果，不能将会话试验的动作或失败移作评分轨迹证据。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "本次运行无视频、无状态日志，评分器记录 missing_trajectory=1、n_actions=0；run.json 日志（steps 2–39，约 3544 s）只有会话内测试，且没有 save_trajectory 交付。标注按 D23(a)：节点终值 unknown、现象和根因留空并附 gap_zh，预算耗尽不作现象或根因，未把会话测试当作评分轨迹，我认为合理。",
+    "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": "2026-10-30T22:18:25Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/dexhand-essential-pen-regrasp-i00-privileged--kgc9cxz/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-essential-pen-regrasp-i00-privileged--kgc9cxz/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "fb4f02c72783a360fd21f89929798ef30c5f6631b7f2408bdb3cdb2a4a380594",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/essential_pen_regrasp/trajectories/dexhand-essential-pen-regrasp-i00-standard--kspkphx.json

@@ -0,0 +1,256 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-essential-pen-regrasp-i00-standard--kspkphx",
+ "taskKey": "dexhand/essential_pen_regrasp",
+ "dag_sha256": "5d4eb54b0f602d691da20e76dd6190aca4275c6bc50b8abc6afeb7e5ecb805d2",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "calls": 1473,
+   "deterministic": 1,
+   "finalized": 1,
+   "hand_wuji": 0,
+   "infra_error": 0,
+   "live_matches_replay": 1,
+   "live_success": 0,
+   "n_actions": 3215,
+   "no_verdict": 0,
+   "replay_success": 0,
+   "reset_budget": 20,
+   "resets": 1,
+   "resume_state_mismatch": 0,
+   "resumes": 0,
+   "sim_error": 0,
+   "sim_startups": 1,
+   "sim_steps": 3215,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 19.38,
+   "peak_processes": 8,
+   "peak_disk_gb": 0.245,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 2,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "pen_advanced": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 37,
+      "text_zh": "观察：log step 36 的服务状态为 episode 1、done=True、reason=task_failure；log step 37 复位后服务返回 episode 2，log step 105 仍为 episode 2。reward.resets=1、resumes=0、live_matches_replay=1，故以复位后的最后一回合作为受评尝试，不将第一回合的失败直接标在本回合。"
+     },
+     {
+      "t": null,
+      "log_step": 134,
+      "text_zh": "观察：log step 54 的脚本注明第一阶段推进；log step 106、111、114 继续尝试以拇指推进笔，log step 116、119–137 改用食指屈曲和侧摆，注释仍要求沿笔轴向远端滑动。log step 134 服务仅返回 done=False、reason=None；笔帽位移 delta 来自 agent 自写 exp.measure，评分记录未报告 pen_axial、pen_angle 或 stage_index。本回合无视频、无状态日志，不能据这些估计确认推进阶段是否通过，final 留未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：log step 38–40、54–58、106–137 的脚本控制右手拇指、中指及食指尝试推进笔；本回合没有可检查的图像或接触记录，无法确认哪根手指实际碰到笔、是否夹住或完成手内滑移。"
+     }
+    ]
+   },
+   "pen_tripod_grip": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 115,
+      "text_zh": "观察：log step 111、114 的脚本试图抬高笔尾并推进，log step 115、116 及后续食指动作试图改变笔的倾角；日志未给出服务端的三指握带接触、离掌、转角或保持计数。inputs/run.json 的 reward 只报告整体 live_success=0、replay_success=0，没有逐节点验收值；不能由整体失败确定本节点的终值。"
+     },
+     {
+      "t": null,
+      "log_step": 137,
+      "text_zh": "观察：log step 134 最后一次有输出的动作仍报告服务 done=False、reason=None，log step 137 最后一次动作调用没有输出，且没有最终完成报告。本回合无视频、无状态日志，无法确认转向、三指握带夹持及稳定保持是否成立，final 留未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：log step 111–137 记录右手拇指与食指改变笔姿的动作意图；缺少本回合图像及实际接触对，无法确认是否已开展三指握带换握，也无法判定各指与笔或掌面的实际接触方式。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "pen_axial",
+   "unit": "m",
+   "threshold": [
+    0.024,
+    0.036
+   ],
+   "direction": "target",
+   "value": null
+  },
+  {
+   "name": "pen_angle_advance",
+   "unit": "rad",
+   "threshold": [
+    0,
+    0.2
+   ],
+   "direction": "target",
+   "value": null
+  },
+  {
+   "name": "pen_angle_tripod",
+   "unit": "rad",
+   "threshold": [
+    0.65,
+    0.92
+   ],
+   "direction": "target",
+   "value": null
+  },
+  {
+   "name": "speed_terminal",
+   "unit": "m/s or rad/s",
+   "threshold": 0.06,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "force",
+   "unit": "N",
+   "threshold": 80,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "object_displacement",
+   "unit": "m",
+   "threshold": 0.07,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "object_height",
+   "unit": "m",
+   "threshold": [
+    0.18,
+    0.6
+   ],
+   "direction": "target",
+   "value": null
+  },
+  {
+   "name": "palm_drift",
+   "unit": "m",
+   "threshold": 0.003,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "wrist_rotation",
+   "unit": "rad",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "pen_advanced",
+   "t": null,
+   "log_step": 134,
+   "where": "操作中",
+   "evidence_zh": "推断：log step 37 唯一复位后的 episode 2 是最后受评回合；log step 54 明确尝试第一阶段推进，log step 106、111、114 继续推进笔，log step 116、119–137 的食指动作仍注明向远端滑动，直到最后调用均未记录切换至三指握带终态验收。依据这些动作目标，将断点推定在笔沿轴前移的操作中；这只定位仍在尝试的步骤，不证明推进阶段的验收值。log step 134 服务仍为 done=False、reason=None，reward 最终 success=0；本回合无视频，无法定位物理失败时刻。"
+  },
+  "unmet": {
+   "pen_advanced": null,
+   "pen_tripod_grip": null
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "本回合 inputs/video.json 明确 kind=none，未提供本回合 replay.mp4 或 states.jsonl；inputs/run.json 中看图调用仅留下图片标记与偏移信息，不能检查实际画面。评分记录没有阶段索引、保持计数、逐节点结果及原生位姿/接触指标，两个节点的 final 和全部视频区间均为 null，实际接触为未知。断点仅由复位后脚本持续推进笔的动作目标推定，不能排除推进曾通过而后续换握未完成。agent 自写的笔帽深度估计与关节 effort 不能替代笔的真实位姿、接触力或验收结果；agent 在 log step 54 自述测量不可靠。log step 37 对第一回合失败的力峰值猜测没有服务接触力证据，而且该回合随后复位，不作为本回合现象或根因。最后调用 log step 137 没有返回，run.wall_s=3607、time_limit_s=3600 与预算结束时停止一致，但时间用完不作失败现象或根因；缺少受评物理失败和因果证据，现象与根因均留空。inputs/oracle/replay.mp4 存在，已查看成功参考的概览接触表；它不能补作本回合证据。"
+ },
+ "summary_zh": "评分记录确认最后受评回合失败（success=0、live_matches_replay=1），但未报告逐阶段结果，两个节点终值均未知。唯一复位后的脚本持续尝试用拇指及食指推进并调整笔姿，故将断点推定为 pen_advanced 的操作中，无法确认具体物理失败现象或根因。本回合没有视频和状态日志，所有视频区间与未报告的九项原生指标均留空；成功 oracle 只用于了解目标外观。",
+ "conf": 4,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "本回合无视频、无状态日志，两节点 final 与区间均为未知，符合 N9；以复位后的 episode 2 作为受评回合有 log step 36/37 支持；现象与根因留空并写了 gap_zh，未把超时当原因。断点为推断并已标明，可接受。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T22:20:32Z",
+  "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/dexhand-essential-pen-regrasp-i00-standard--kspkphx/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-essential-pen-regrasp-i00-standard--kspkphx/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "c2948791bae9ec156fac441502f65ba70b7d505feefe72d8be420f5623ddff51",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/essential_rotary_gait/trajectories/dexhand-essential-rotary-gait-i00-privileged--5sospex.json

@@ -0,0 +1,215 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-essential-rotary-gait-i00-privileged--5sospex",
+ "taskKey": "dexhand/essential_rotary_gait",
+ "dag_sha256": "dcf31bae70eb0330bb07ebf73158e14c1b5b7a89486e26ab83954530a3b6331d",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.804,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.838,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 2,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "knob_forward_720": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 132,
+      "text_zh": "观察：评分器 reward 记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告正转阶段验收结果；inputs/video.json 为 kind=none，且无状态日志。log step 132 是 agent 的开发搜索输出（候选 turn 约 11.214 rad），没有成为交付轨迹的证据，不能据此判定本节点最终值。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "评分器记录未交付轨迹；开发测试中的右手与旋钮接触不作为受评轨迹接触。"
+     }
+    ]
+   },
+   "knob_reversed_360_held": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 132,
+      "text_zh": "观察：评分器 reward 记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告反转阶段或 terminal_hold 验收结果；本次无视频、无状态日志。log step 132 仍为 agent 的正向候选搜索，没有受评反转停稳结果可供判定。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "评分器记录未交付轨迹；开发测试中的右手与旋钮接触不作为受评轨迹接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "turn:two_forward_turns",
+   "unit": "rad",
+   "threshold": 12.566370614359172,
+   "direction": "target",
+   "value": null
+  },
+  {
+   "name": "turn:one_reverse_turn",
+   "unit": "rad",
+   "threshold": 6.283185307179586,
+   "direction": "target",
+   "value": null
+  },
+  {
+   "name": "speed:stage",
+   "unit": "rad/s",
+   "threshold": 0.25,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "speed:terminal",
+   "unit": "rad/s",
+   "threshold": 0.06,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "speed:invariant",
+   "unit": "rad/s",
+   "threshold": 12,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "force",
+   "unit": "N",
+   "threshold": 80,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "palm_drift",
+   "unit": "m",
+   "threshold": 0.003,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "wrist_rotation",
+   "unit": "rad",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 132,
+   "where": "首步前",
+   "evidence_zh": "观察：评分器 reward 为 missing_trajectory=1、n_actions=0、sim_steps=0、infra_error=0。log step 131 仅拼接开发用动作数组，log step 132 继续搜索下一组手指目标并输出候选结果；全日志未见保存受评轨迹的调用，也没有最终交付报告。推断：没有形成可定位物理失败步骤的受评轨迹，因此断点未知、位置为首步前。"
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：inputs/video.json 明确本次无视频，inputs/ 中也无状态日志；存在的 inputs/oracle/replay.mp4 属于另一成功运行，不能证明本次动作或进度。评分器未收到轨迹且未给出两个节点的阶段验收或原生指标；log step 132 的转角、接触与候选有效性均来自开发测试，不能赋给受评节点。运行 wall_s=3607、预算 time_limit_s=3600，日志止于开发搜索，没有最终汇报或明确结束原因记录。推断：记录与预算结束前未交付相符，但不能从缺交付推定具体控制、规划或保存错误；依 D23 两个失败解释层均留空，依 N9 两节点 final 和所有视频区间为 null。",
+  "unmet": {
+   "knob_forward_720": null,
+   "knob_reversed_360_held": null
+  }
+ },
+ "summary_zh": "评分器记录未收到轨迹文件、受评动作数为 0，正转两圈与反转停稳两个节点的最终值均无法判定。log step 132 仍在开发搜索，未形成可供定位物理失败的受评轨迹，故断点未知、位置为首步前。此次无视频或状态日志，开发测试读数及另一成功运行的参考录像不能计为本次节点进度；现象和根因依缺交付规则留空。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "评分器记录 missing_trajectory=1、n_actions=0，无视频无状态日志；两个节点 final 为 null、断点 unknown/首步前、现象与根因留空并附 gap_zh，符合 D23(a) 与 N9。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": "2026-10-30T22:18:11Z",
+  "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/dexhand-essential-rotary-gait-i00-privileged--5sospex/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-essential-rotary-gait-i00-privileged--5sospex/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "c5cafb6adaccdbcca820105db1905b7edf390aad9525de0868edf26943ad5c67",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/essential_rotary_gait/trajectories/dexhand-essential-rotary-gait-i00-privileged--p5srguc.json

@@ -0,0 +1,215 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-essential-rotary-gait-i00-privileged--p5srguc",
+ "taskKey": "dexhand/essential_rotary_gait",
+ "dag_sha256": "dcf31bae70eb0330bb07ebf73158e14c1b5b7a89486e26ab83954530a3b6331d",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.571,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.243,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 2,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "knob_forward_720": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 52,
+      "text_zh": "观察：inputs/run.json 的评分器字段为 missing_trajectory=1、n_actions=0、sim_steps=0，未报告正转阶段验收结果；inputs/video.json 为 kind=none，且无状态日志。log step 52 的 stage 0、269 个控制步和接触中断属于开发测试，并非交付回放，不能据此填写本节点终值（N9）。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告 missing_trajectory=1，未交付可供评分的轨迹；会话测试中的右手与旋钮接触不计为本节点的实际接触。"
+     }
+    ]
+   },
+   "knob_reversed_360_held": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 48,
+      "text_zh": "观察：评分器只报告未交付轨迹及 success=0，没有反转阶段或 terminal_hold 结果；本次无视频、无状态日志。log step 48 的生成器仅保存开发动作数组，log step 52 仍为正转阶段的开发测试，未见正式交付，故无法判断本节点终值（N9）。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告 missing_trajectory=1，未交付可供评分的轨迹；会话测试中的右手与旋钮接触不计为本节点的实际接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "turn:two_forward_turns",
+   "unit": "rad",
+   "threshold": 12.566370614359172,
+   "direction": "target",
+   "value": null
+  },
+  {
+   "name": "turn:one_reverse_turn",
+   "unit": "rad",
+   "threshold": 6.283185307179586,
+   "direction": "target",
+   "value": null
+  },
+  {
+   "name": "speed:stage",
+   "unit": "rad/s",
+   "threshold": 0.25,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "speed:terminal",
+   "unit": "rad/s",
+   "threshold": 0.06,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "speed:invariant",
+   "unit": "rad/s",
+   "threshold": 12,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "force",
+   "unit": "N",
+   "threshold": 80,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "palm_drift",
+   "unit": "m",
+   "threshold": 0.003,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "wrist_rotation",
+   "unit": "rad",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 52,
+   "where": "首步前",
+   "evidence_zh": "观察：评分器报告 missing_trajectory=1、n_actions=0、sim_steps=0、infra_error=0。log step 48 的生成器末尾仅将动作保存为开发目录中的 rows.npy；日志止于 log step 52 的开发测试输出，未出现正式轨迹保存或重放验收。推断：没有可定位物理失败步骤的评分轨迹，断点记 unknown，位置为首步前。"
+  },
+  "unmet": {
+   "knob_forward_720": null,
+   "knob_reversed_360_held": null
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：本次无视频（inputs/video.json: kind=none）、无状态日志，评分器报告未交付轨迹；reward 未提供两个物理节点及任何原生指标的终值。日志中的转角、接力和接触中断均来自开发测试，不能作为评分尝试的进度或失败现象；成功参照也不能证明本次运行发生了什么。推断：依 DECISIONS.md D23，未交付且无已证实外因时，现象与根因均留空；不能仅凭缺交付推定交付操作错误或控制根因。"
+ },
+ "summary_zh": "评分器报告未交付轨迹（missing_trajectory=1，n_actions=0，sim_steps=0），两个物理节点均无法判定，未确认任何节点达成。日志止于 log step 52 的开发测试；log step 48 的生成器仅保存开发动作数组，不能将测试进展或接触中断当作评分尝试。无本次视频及状态日志，按 D23 将断点记为 unknown／首步前，现象与根因留空，原生指标均未报告。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "评分器 missing_trajectory=1、n_actions=0、sim_steps=0，无视频无状态日志；两节点 unknown、断点 unknown/首步前、现象与根因留空并写 gap_zh，符合 03 与 D23(a)。日志最后一步（step 52）仍是开发测试，未见交付，标注未把会话测试当作评分尝试。",
+    "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": "2026-10-30T22:17:12Z",
+  "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/dexhand-essential-rotary-gait-i00-privileged--p5srguc/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-essential-rotary-gait-i00-privileged--p5srguc/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "50f6adad1aebe33584353b7ab61c542fd9c8e57e7180774ba048f8a9f98af53d",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/essential_rotary_gait/trajectories/dexhand-essential-rotary-gait-i00-privileged--uodurxc.json

@@ -0,0 +1,215 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-essential-rotary-gait-i00-privileged--uodurxc",
+ "taskKey": "dexhand/essential_rotary_gait",
+ "dag_sha256": "dcf31bae70eb0330bb07ebf73158e14c1b5b7a89486e26ab83954530a3b6331d",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.64,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.256,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 2,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "knob_forward_720": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 74,
+      "text_zh": "观察：inputs/run.json 的评分器 reward 报告 missing_trajectory=1、n_actions=0、sim_steps=0，未提供正转阶段验收字段；inputs/video.json 为 kind=none，且无状态日志。log step 73 仅写入生成脚本，log step 74 启动后只返回后台任务通知，log step 75 为 agent 自述正在生成和验证，没有生成结果；会话测试不能证明受评正转节点取值。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告 missing_trajectory=1，未交付受评轨迹；会话测试中右手与旋钮的接触不计入受评接触。"
+     }
+    ]
+   },
+   "knob_reversed_360_held": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 74,
+      "text_zh": "观察：inputs/run.json 的评分器 reward 报告 missing_trajectory=1、n_actions=0、sim_steps=0，未报告反转阶段或终末保持结果；inputs/video.json 为 kind=none，且无状态日志。log step 73 的生成脚本仅在自身 success 为真时保存轨迹，log step 74–75 没有返回最终状态或保存确认；会话中的反转测试不是交付轨迹证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告 missing_trajectory=1，未交付受评轨迹；会话测试中右手与旋钮的接触不计入受评接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "turn:two_forward_turns",
+   "unit": "rad",
+   "threshold": 12.566370614359172,
+   "direction": "target",
+   "value": null
+  },
+  {
+   "name": "turn:one_reverse_turn",
+   "unit": "rad",
+   "threshold": 6.283185307179586,
+   "direction": "target",
+   "value": null
+  },
+  {
+   "name": "speed:stage",
+   "unit": "rad/s",
+   "threshold": 0.25,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "speed:terminal",
+   "unit": "rad/s",
+   "threshold": 0.06,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "speed:invariant",
+   "unit": "rad/s",
+   "threshold": 12,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "force",
+   "unit": "N",
+   "threshold": 80,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "palm_drift",
+   "unit": "m",
+   "threshold": 0.003,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "wrist_rotation",
+   "unit": "rad",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 74,
+   "where": "首步前",
+   "evidence_zh": "观察：评分器 reward 报告 missing_trajectory=1、n_actions=0、sim_steps=0、infra_error=0；log step 73 写入生成脚本，log step 74 启动该脚本后仅返回后台运行通知，log step 75 的最后自述是正在生成和验证轨迹，之后无结果记录。没有交付的受评动作，无法定位某个物理步骤的失败。"
+  },
+  "unmet": {
+   "knob_forward_720": null,
+   "knob_reversed_360_held": null
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "本次运行无视频、无状态日志，评分器没有收到轨迹，也未报告两个物理节点的阶段验收或任何原生物理指标；因此所有 intervals 与 final 均为 null，接触为不适用。log step 74 启动的生成进程没有后续输出，无法确认其终止位置或未保存原因；运行 wall_s=3603、预算 time_limit_s=3600 只说明会话预算情况，不能作为失败现象或根因。log step 45、53、57、72 等调试回合的接触中断不能���为受评轨迹现象；按 DECISIONS.md D23 / CASES.md OPEN-16，未交付时两层留空，不从缺交付推断控制或交付操作错误。inputs/oracle/replay.mp4 是另一成功运行，不能替代本次受评证据。"
+ },
+ "summary_zh": "评分器以 missing_trajectory=1 判定失败，受评动作数与仿真步数均为 0；两项旋钮节点均无可确认的终值，不能将会话调试进度记为受评进度。log step 73–75 只记录编写并启动生成脚本及 agent 自述正在生成验证，没有交付保存或验证成功的结果，故断点为 unknown／首步前。无本次视频、状态日志或节点验收字段，失败现象与根因按 D23 留空。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "无视频/状态日志，评分器 missing_trajectory=1、n_actions=0；两节点 unknown、断点 unknown/首步前、现象与根因留空并有 gap_zh，符合 D23/OPEN-16，未把预算耗尽当作原因，也未把会话测试当作受评事实。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": "2026-10-30T22:18:12Z",
+  "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/dexhand-essential-rotary-gait-i00-privileged--uodurxc/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-essential-rotary-gait-i00-privileged--uodurxc/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "e67bfe7244d70a736839d80c05be3a0a845695fbb576de8aaead25abb2fa6b59",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/essential_two_object/trajectories/dexhand-essential-two-object-i00-privileged--5y5ktk6.json

@@ -0,0 +1,280 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-essential-two-object-i00-privileged--5y5ktk6",
+ "taskKey": "dexhand/essential_two_object",
+ "dag_sha256": "531229b39d851ab1d907e1264ab1da4b275b30aec7af561984af7640acd7c097",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.968,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.261,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 3,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "a_quarter_turned": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 262,
+      "text_zh": "观察：评分器 reward.missing_trajectory=1、n_actions=0、sim_steps=0；无本次视频或状态日志。log step 262 的 stage_index=1 属于 agent 的会话测试，未见该测试被保存并交付为评分轨迹，不能据此确定本节点终值。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告 missing_trajectory=1，未交付评分轨迹，无该轨迹的实际接触可标注。"
+     }
+    ]
+   },
+   "a_returned": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 262,
+      "text_zh": "观察：log step 262 的会话测试输出 stage_index=2，但评分器报告 missing_trajectory=1、n_actions=0、sim_steps=0，未提供评分轨迹的阶段结果；无本次视频或状态日志，本节点终值未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告 missing_trajectory=1，未交付评分轨迹，无该轨迹的实际接触可标注。"
+     }
+    ]
+   },
+   "a_in_tray": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 415,
+      "text_zh": "观察：log step 415 是 agent 新建 Session 后运行的落盘静置测试，输出 stage_index=3、terminal_hold=0，未见保存或提交；评分器报告 missing_trajectory=1、n_actions=0、sim_steps=0。该会话测试不是评分轨迹，不能用来确定本节点终值。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告 missing_trajectory=1，未交付评分轨迹，无该轨迹的实际接触可标注。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "a_quarter_turn_error",
+   "unit": "rad",
+   "threshold": 0.16,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "a_orientation_error",
+   "unit": "rad",
+   "threshold": 0.16,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "a_position_error",
+   "unit": "m",
+   "threshold": 0.035,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "a_deposit_distance",
+   "unit": "m",
+   "threshold": 0.04,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "b_position_error",
+   "unit": "m",
+   "threshold": 0.015,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "b_orientation_error",
+   "unit": "rad",
+   "threshold": 0.2,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "force",
+   "unit": "N",
+   "threshold": 80,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "palm_drift",
+   "unit": "m",
+   "threshold": 0.003,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "wrist_rotation",
+   "unit": "rad",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "a_linear_speed",
+   "unit": "m/s",
+   "threshold": 2.5,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "a_angular_speed",
+   "unit": "rad/s",
+   "threshold": 80,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "a_height",
+   "unit": "m",
+   "threshold": [
+    0.135,
+    0.6
+   ],
+   "direction": "target",
+   "value": null
+  },
+  {
+   "name": "speed_rotate_return",
+   "unit": "m/s or rad/s",
+   "threshold": 0.25,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "speed_terminal",
+   "unit": "m/s or rad/s",
+   "threshold": 0.06,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 415,
+   "where": "首步前",
+   "evidence_zh": "观察：评分器 reward.missing_trajectory=1、n_actions=0、sim_steps=0、infra_error=0；完整日志末尾 log step 415 仍为 agent 的独立落盘静置测试，没有评分轨迹交付记录，无法定位某个物理步骤的失败。"
+  },
+  "unmet": {
+   "a_quarter_turned": null,
+   "a_returned": null,
+   "a_in_tray": null
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：本次无视频、无状态日志，评分器只报告轨迹缺失和零执行动作，未报告各节点或原生指标。log step 262、415 的阶段及接触数值来自 agent 会话测试，不属于已交付的评分轨迹；inputs/oracle/replay.mp4 属于另一次成功运行，也不能证明本次进度。推断：无法确定本次三个节点的终值、实际物理失败现象或根因；按无交付规则保留空现象和空根因，不以会话测试替代评分尝试，也不从轨迹缺失反推交付操作出错。"
+ },
+ "summary_zh": "观察：评分器报告 success=0、missing_trajectory=1、n_actions=0、sim_steps=0，本次没有交付评分轨迹，三个节点终值均未知。观察：log step 262 的转动与回正、log step 415 的落盘静置均为 agent 会话测试，末次测试的 terminal_hold 始终为0，不能计入本次评分进度。推断：断点只能记为 unknown／首步前；因无本次视频、状态日志或交付轨迹，失败现象和根因留空。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "评分器 missing_trajectory=1、n_actions=0、sim_steps=0，无视频/状态日志；agent 在 3600 s 上限（wall 3608 s）内始终只做会话测试，未交付。标注为 unknown／首步前、两层为空并带 gap_zh、节点均 unknown，符合 03 无交付规则（D23）。",
+    "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": "2026-10-30T22:19:49Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/dexhand-essential-two-object-i00-privileged--5y5ktk6/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-essential-two-object-i00-privileged--5y5ktk6/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "94814266b11bb2c165e24e3f205cd8c7f82de6058981f0ab6d2ee4952b7d17f3",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/essential_two_object/trajectories/dexhand-essential-two-object-i00-privileged--9sacyfj.json

@@ -0,0 +1,280 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-essential-two-object-i00-privileged--9sacyfj",
+ "taskKey": "dexhand/essential_two_object",
+ "dag_sha256": "531229b39d851ab1d907e1264ab1da4b275b30aec7af561984af7640acd7c097",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.976,
+   "peak_processes": 10,
+   "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": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 3,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "a_quarter_turned": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 53,
+      "text_zh": "观察：评分器 reward.missing_trajectory=1、n_actions=0、sim_steps=0，未提供该节点的受评阶段结果；inputs/video.json 的 kind=none，且无状态日志。log step 53 的候选测试输出 stage_index=1 属于会话开发记录，不能作为受评轨迹的节点终值，故保留未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1、n_actions=0；没有交付的受评轨迹，无法标注该步的实际接触，会话测试不计入。"
+     }
+    ]
+   },
+   "a_returned": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 68,
+      "text_zh": "观察：log step 68 的候选测试输出 stage_index=2，但评分器记录 missing_trajectory=1、n_actions=0，没有交付轨迹或该节点的受评结果；无本次视频及状态日志，不能把会话测试的回正进展计入受评轨迹，终值未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1、n_actions=0；没有交付的受评轨迹，无法标注该步的实际接触，会话测试不计入。"
+     }
+    ]
+   },
+   "a_in_tray": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 115,
+      "text_zh": "观察：log step 115 的候选测试输出 stage_index=3、terminal_hold=0、success=False，随后测试脚本出现 NameError；这些是会话测试结果。评分器记录 missing_trajectory=1、n_actions=0，未报告受评落盘或静置终值，且无本次视频及状态日志，故节点未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1、n_actions=0；没有交付的受评轨迹，无法标注该步的实际接触，会话测试不计入。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "a_quarter_turn_error",
+   "unit": "rad",
+   "threshold": 0.16,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "a_orientation_error",
+   "unit": "rad",
+   "threshold": 0.16,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "a_position_error",
+   "unit": "m",
+   "threshold": 0.035,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "a_deposit_distance",
+   "unit": "m",
+   "threshold": 0.04,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "b_position_error",
+   "unit": "m",
+   "threshold": 0.015,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "b_orientation_error",
+   "unit": "rad",
+   "threshold": 0.2,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "force",
+   "unit": "N",
+   "threshold": 80,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "palm_drift",
+   "unit": "m",
+   "threshold": 0.003,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "wrist_rotation",
+   "unit": "rad",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "a_linear_speed",
+   "unit": "m/s",
+   "threshold": 2.5,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "a_angular_speed",
+   "unit": "rad/s",
+   "threshold": 80,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "a_height",
+   "unit": "m",
+   "threshold": [
+    0.135,
+    0.6
+   ],
+   "direction": "target",
+   "value": null
+  },
+  {
+   "name": "speed_rotate_return",
+   "unit": "m/s or rad/s",
+   "threshold": 0.25,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "speed_terminal",
+   "unit": "m/s or rad/s",
+   "threshold": 0.06,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 130,
+   "where": "首步前",
+   "evidence_zh": "观察：评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0、success=0、infra_error=0；log step 127、129 的脚本仅在测试 success 为真时保存最终轨迹，log step 130 仍在查询候选测试日志及交付目录，输出为空，没有保存成功或回放验证的记录。没有交付的受评轨迹，不能以会话中正在调试的释放步骤指定物理断点。"
+  },
+  "unmet": {
+   "a_quarter_turned": null,
+   "a_returned": null,
+   "a_in_tray": null
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "本次无视频（inputs/video.json: kind=none），无状态日志，评分器未提供各节点终值或原生物理指标，且 missing_trajectory=1。log step 53、68、115 的阶段记录与 log step 123 关于B缓慢转动的自述均属于候选测试，不能当作受评轨迹的物理现象；log step 127—130 没有返回保存或验证成功的结果。运行记录 wall_s=3606、time_limit_s=3600，与预算结束相符，但不能据此归因为控制、规划或交付操作出错，也没有证据确认外部中断。按 DECISIONS.md D23，无交付且未证实外因时保留空现象和根因，不标没尝试；缺少受评轨迹也无法定位具体失败节点或视频时刻。"
+ },
+ "summary_zh": "评分器判定失败且缺少交付轨迹（missing_trajectory=1、n_actions=0），三个节点的受评终值均未知，实际接触均为不适用。会话候选测试曾输出 stage_index=3、terminal_hold=0（log step 115），最后仍在调试释放与稳持（log step 123—130），这些不能算作受评进度。无本次视频及状态日志，断点记 unknown／首步前，并按 DECISIONS.md D23 将现象和根因留空。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "无视频、无状态日志、评分器 missing_trajectory=1、n_actions=0，无交付轨迹；三节点终值 unknown、接触不适用、断点 unknown/首步前、现象与根因留空并写 gap_zh，符合 DECISIONS.md D23(a)。会话测试的 stage_index 未被当作受评进度，也未把预算耗尽当作原因。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": "2026-10-30T22:21:29Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/dexhand-essential-two-object-i00-privileged--9sacyfj/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-essential-two-object-i00-privileged--9sacyfj/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "9f128130b52c26f50fe660939962e4d01a94decdff47ec1904bd47d1d2c0fa73",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/essential_two_object/trajectories/dexhand-essential-two-object-i00-privileged--cdpetpm.json

@@ -0,0 +1,280 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-essential-two-object-i00-privileged--cdpetpm",
+ "taskKey": "dexhand/essential_two_object",
+ "dag_sha256": "531229b39d851ab1d907e1264ab1da4b275b30aec7af561984af7640acd7c097",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.721,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.253,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 3,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "a_quarter_turned": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 37,
+      "text_zh": "观察：log step 37中，agent的会话测试脚本输出stage_index=1；评分器reward却记录missing_trajectory=1、n_actions=0、sim_steps=0，未提供此节点的受评结果。无本次受评视频或状态日志，不能把测试中的转动进度记为交付节点值。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "评分器记录missing_trajectory=1，无交付轨迹；会话测试中的右手与核桃接触不计入受评轨迹接触。"
+     }
+    ]
+   },
+   "a_returned": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 67,
+      "text_zh": "观察：log step 67中，agent的会话测试脚本输出stage=2；本次评分器只记录轨迹缺失和零受评动作，没有回正节点的受评结果。无受评视频或状态日志，测试输出不能决定本节点终值。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "评分器记录missing_trajectory=1，无交付轨迹；会话测试中的右手与核桃接触不计入受评轨迹接触。"
+     }
+    ]
+   },
+   "a_in_tray": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 71,
+      "text_zh": "观察：log step 71中，agent的释放测试脚本输出success=True、stage_index=3、terminal_hold=25；log step 70显示该脚本从复制的第二阶段状态测试释放，未保存交付动作文件。评分器reward记录missing_trajectory=1且n_actions=0；该会话测试不是受评落盘结果，本节点终值未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "评分器记录missing_trajectory=1，无交付轨迹；会话测试中的右手与核桃接触不计入受评轨迹接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "a_quarter_turn_error",
+   "unit": "rad",
+   "threshold": 0.16,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "a_orientation_error",
+   "unit": "rad",
+   "threshold": 0.16,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "a_position_error",
+   "unit": "m",
+   "threshold": 0.035,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "a_deposit_distance",
+   "unit": "m",
+   "threshold": 0.04,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "b_position_error",
+   "unit": "m",
+   "threshold": 0.015,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "b_orientation_error",
+   "unit": "rad",
+   "threshold": 0.2,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "force",
+   "unit": "N",
+   "threshold": 80,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "palm_drift",
+   "unit": "m",
+   "threshold": 0.003,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "wrist_rotation",
+   "unit": "rad",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "a_linear_speed",
+   "unit": "m/s",
+   "threshold": 2.5,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "a_angular_speed",
+   "unit": "rad/s",
+   "threshold": 80,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "a_height",
+   "unit": "m",
+   "threshold": [
+    0.135,
+    0.6
+   ],
+   "direction": "target",
+   "value": null
+  },
+  {
+   "name": "speed_rotate_return",
+   "unit": "m/s or rad/s",
+   "threshold": 0.25,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "speed_terminal",
+   "unit": "m/s or rad/s",
+   "threshold": 0.06,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 72,
+   "where": "首步前",
+   "evidence_zh": "观察：log step 72以“This request was blocked by Gemini's filters”结束，之后没有交付调用；评分器reward记录missing_trajectory=1、n_actions=0、sim_steps=0。log step 71的success=True来自agent的会话测试，未保存并验证交付轨迹，因此没有可定位的受评物理失败步骤。"
+  },
+  "unmet": {
+   "a_quarter_turned": null,
+   "a_returned": null,
+   "a_in_tray": null
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：inputs/video.json明确kind=none，本次没有受评视频或状态日志；inputs/oracle/replay.mp4属于另一成功运行，不能证明本次节点进度。log step 72记录Gemini供应商过滤拦截，评分器记录missing_trajectory=1，未交付受评轨迹；log step 37、67、71均为会话测试输出，不能替代受评节点结果。推断：按外部因素中断处理，应重试并排除出机器人失败分析；断点为unknown／首步前，现象和根因留空，不标没尝试，也不从缺交付推断交付操作出错。所有节点的终值及视频区间均未知，原生指标未在评分器字段报告。"
+ },
+ "summary_zh": "本次未交付受评动作轨迹，评分器记录missing_trajectory=1、n_actions=0，三个节点均无可确认的受评终值。log step 71的会话脚本虽输出成功，但未形成已保存并验证的交付，随后log step 72出现Gemini过滤拦截，按外部中断处理，断点unknown／首步前，现象与根因留空。没有本次受评视频或状态日志，成功参考视频属于另一运行，无法补足节点与接触证据。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "同意：本次无受评视频/状态日志，评分器 missing_trajectory=1、n_actions=0；log step 70-71 的脚本只做会话内验证，未保存交付动作，step 72 为 Gemini 过滤拦截，按外部因素中断处理（空现象/根因+gap_zh、断点 unknown/首步前）合理，节点均为 unknown 且 contact_actual 为不适用，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": "2026-10-30T22:18:42Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/dexhand-essential-two-object-i00-privileged--cdpetpm/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-essential-two-object-i00-privileged--cdpetpm/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "87234f222b8722383f64db6b9d6cd619917939ceacaec6e79c37cfafafe3489e",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/fan_guard_install/trajectories/dexhand-fan-guard-install-i00-privileged--e7ctvui.json

@@ -0,0 +1,296 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-fan-guard-install-i00-privileged--e7ctvui",
+ "taskKey": "dexhand/fan_guard_install",
+ "dag_sha256": "780b495033f927e2f99d6c0c605a490e62e3855b405d828e8318daf24d90f374",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.754,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.261,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 5,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "guard_seated": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 345,
+      "text_zh": "观察：评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0；本次无视频和状态日志，评分器未报告“前护罩就位”的终态。log step 345 仅启动另一次基于缓存前缀的新建 Session 调试，未给出交付结果，不能据此判断该节点。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1，本次没有交付轨迹，故无受评轨迹接触可标注；会话调试接触不计入。"
+     }
+    ]
+   },
+   "clip_45_closed": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 345,
+      "text_zh": "观察：评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0；本次无视频和状态日志，评分器未报告“合上45度扣”的终态。log step 345 仅启动另一次基于缓存前缀的新建 Session 调试，未给出交付结果，不能据此判断该节点。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1，本次没有交付轨迹，故无受评轨迹接触可标注；会话调试接触不计入。"
+     }
+    ]
+   },
+   "clip_135_closed": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 345,
+      "text_zh": "观察：评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0；本次无视频和状态日志，评分器未报告“合上135度扣”的终态。log step 345 仅启动另一次基于缓存前缀的新建 Session 调试，未给出交付结果，不能据此判断该节点。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1，本次没有交付轨迹，故无受评轨迹接触可标注；会话调试接触不计入。"
+     }
+    ]
+   },
+   "clip_225_closed": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 345,
+      "text_zh": "观察：评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0；本次无视频和状态日志，评分器未报告“合上225度扣”的终态。log step 345 仅启动另一次基于缓存前缀的新建 Session 调试，未给出交付结果，不能据此判断该节点。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1，本次没有交付轨迹，故无受评轨迹接触可标注；会话调试接触不计入。"
+     }
+    ]
+   },
+   "clip_315_closed": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 345,
+      "text_zh": "观察：评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0；本次无视频和状态日志，评分器未报告“合上315度扣”的终态。log step 345 仅启动另一次基于缓存前缀的新建 Session 调试，未给出交付结果，不能据此判断该节点。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1，本次没有交付轨迹，故无受评轨迹接触可标注；会话调试接触不计入。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "concentric_m",
+   "unit": "m",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "abs_seat_axial_m",
+   "unit": "m",
+   "threshold": 0.0025,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "parallel_rad",
+   "unit": "rad",
+   "threshold": 0.0872664626,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "clip_45_angle_rad",
+   "unit": "rad",
+   "threshold": 0.12,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "clip_135_angle_rad",
+   "unit": "rad",
+   "threshold": 0.12,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "clip_225_angle_rad",
+   "unit": "rad",
+   "threshold": 0.12,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "clip_315_angle_rad",
+   "unit": "rad",
+   "threshold": 0.12,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "guard_linear_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "guard_angular_speed_rad_s",
+   "unit": "rad/s",
+   "threshold": 0.2,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "fan_displacement_m",
+   "unit": "m",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "fan_tilt_rad",
+   "unit": "rad",
+   "threshold": 0.1,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 345,
+   "where": "首步前",
+   "evidence_zh": "观察：评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0、success=0、infra_error=0；log step 343 中 agent 仍自述正在评估腕部撤回与放松抓握，最后的 log step 345 只返回调试后台任务已启动，没有交付或验证结果。推断：无可定位的受评物理步骤，按未交付轨迹记 unknown／首步前。"
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：inputs/video.json 的 kind=none，且无本次状态日志；评分器仅报告缺交付及零动作，未报告任何节点或原生几何指标。log step 262 的 Stage=1、Clips locked=2，以及 log step 323、341 的松手或撤回失败读数，均来自 agent 自建 Session 的调试测试，不能视为受评终态或失败现象；log step 345 启动的后台测试没有返回最终结果。推断：按 DECISIONS.md D23 与 N9，五节点终值保留未知，缺交付不足以确定物理断点或根因，现象及根因均留空；成功参考视频属于另一运行，不能补足本次证据。",
+  "unmet": {
+   "guard_seated": null,
+   "clip_45_closed": null,
+   "clip_135_closed": null,
+   "clip_225_closed": null,
+   "clip_315_closed": null
+  }
+ },
+ "summary_zh": "评分器判定未交付轨迹（missing_trajectory=1，n_actions=0，sim_steps=0），本次五个节点均无可确认的受评终态。日志会话曾测试护罩就位及合扣，至 log step 343、345 仍在调试松手、腕部撤回和就位位置；这些测试不计入轨迹进度，断点为 unknown／首步前。无本次视频、状态日志或逐节点评分结果，依 D23、N9 保留节点未知并将失败现象、根因留空。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "本次无视频、无状态日志，评分器仅报告 missing_trajectory=1、n_actions=0、sim_steps=0；标注文件按 D23(a)/N9 将五节点留未知，断点 unknown/首步前，现象与根因留空并写 gap_zh，且不把会话调试读数当作受评事实。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": "2026-10-30T22:29:58Z",
+  "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/dexhand-fan-guard-install-i00-privileged--e7ctvui/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-fan-guard-install-i00-privileged--e7ctvui/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "83b44501780aae9ed56694166fbec18a715ff7ef8f35ff9fcb45e0caba631107",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/fan_guard_install/trajectories/dexhand-fan-guard-install-i00-privileged--ftzpnvq.json

@@ -0,0 +1,296 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-fan-guard-install-i00-privileged--ftzpnvq",
+ "taskKey": "dexhand/fan_guard_install",
+ "dag_sha256": "780b495033f927e2f99d6c0c605a490e62e3855b405d828e8318daf24d90f374",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 0.487,
+   "peak_processes": 8,
+   "peak_disk_gb": 0.254,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 5,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "guard_seated": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 139,
+      "text_zh": "观察：inputs/video.json标记本次无视频，且无状态日志；评分器reward仅报告missing_trajectory=1、n_actions=0、sim_steps=0和success=0，未报告“前护罩就位”的受评终值。log step 139仍是agent自建Session的抓取测试输出，不能作为受评节点值；按N9保留未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "评分器记录missing_trajectory=1，未交付受评轨迹；会话测试中的手指接触不计入本节点。"
+     }
+    ]
+   },
+   "clip_45_closed": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 139,
+      "text_zh": "观察：inputs/video.json标记本次无视频，且无状态日志；评分器reward仅报告missing_trajectory=1、n_actions=0、sim_steps=0和success=0，未报告“合上45度扣”的受评终值。log step 139仍是agent自建Session的抓取测试输出，不能作为受评节点值；按N9保留未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "评分器记录missing_trajectory=1，未交付受评轨迹；会话测试中的手指接触不计入本节点。"
+     }
+    ]
+   },
+   "clip_135_closed": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 139,
+      "text_zh": "观察：inputs/video.json标记本次无视频，且无状态日志；评分器reward仅报告missing_trajectory=1、n_actions=0、sim_steps=0和success=0，未报告“合上135度扣”的受评终值。log step 139仍是agent自建Session的抓取测试输出，不能作为受评节点值；按N9保留未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "评分器记录missing_trajectory=1，未交付受评轨迹；会话测试中的手指接触不计入本节点。"
+     }
+    ]
+   },
+   "clip_225_closed": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 139,
+      "text_zh": "观察：inputs/video.json标记本次无视频，且无状态日志；评分器reward仅报告missing_trajectory=1、n_actions=0、sim_steps=0和success=0，未报告“合上225度扣”的受评终值。log step 139仍是agent自建Session的抓取测试输出，不能作为受评节点值；按N9保留未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "评分器记录missing_trajectory=1，未交付受评轨迹；会话测试中的手指接触不计入本节点。"
+     }
+    ]
+   },
+   "clip_315_closed": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 139,
+      "text_zh": "观察：inputs/video.json标记本次无视频，且无状态日志；评分器reward仅报告missing_trajectory=1、n_actions=0、sim_steps=0和success=0，未报告“合上315度扣”的受评终值。log step 139仍是agent自建Session的抓取测试输出，不能作为受评节点值；按N9保留未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "评分器记录missing_trajectory=1，未交付受评轨迹；会话测试中的手指接触不计入本节点。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "concentric_m",
+   "unit": "m",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "abs_seat_axial_m",
+   "unit": "m",
+   "threshold": 0.0025,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "parallel_rad",
+   "unit": "rad",
+   "threshold": 0.0872664626,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "clip_45_angle_rad",
+   "unit": "rad",
+   "threshold": 0.12,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "clip_135_angle_rad",
+   "unit": "rad",
+   "threshold": 0.12,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "clip_225_angle_rad",
+   "unit": "rad",
+   "threshold": 0.12,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "clip_315_angle_rad",
+   "unit": "rad",
+   "threshold": 0.12,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "guard_linear_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "guard_angular_speed_rad_s",
+   "unit": "rad/s",
+   "threshold": 0.2,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "fan_displacement_m",
+   "unit": "m",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "fan_tilt_rad",
+   "unit": "rad",
+   "threshold": 0.1,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 139,
+   "where": "首步前",
+   "evidence_zh": "观察：评分器reward记录missing_trajectory=1、n_actions=0、sim_steps=0、infra_error=0；log step 139最后执行的仍是agent抓取测试，输出After upright后fail: guard_dropped，日志没有后续保存交付或最终报告。推断：没有可定位物理断点的受评轨迹，断点记unknown、首步前；测试掉落不当作受评失败现象。"
+  },
+  "unmet": {
+   "guard_seated": null,
+   "clip_45_closed": null,
+   "clip_135_closed": null,
+   "clip_225_closed": null,
+   "clip_315_closed": null
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "本次无视频、无状态日志且未交付受评轨迹，评分器也没有逐节点终值，五个节点的intervals与final均为null。inputs/oracle/replay.mp4属于另一成功运行，不能补作本次证据。日志中的下扣闭合、搬运与guard_dropped均为agent会话测试（如log step 133、139），不能标为本次受评轨迹的进度或现象。run.json记录会话wall_s=3602、time_limit_s=3600，末尾仍在调试且无最终报告，会话耗时接近预算；日志未明确给出终止事件或可归因的交付操作错误。依D23，对agent侧未交付保留现象和根因两层为空，不从缺交付或会话测试反推受评根因。"
+ },
+ "summary_zh": "评分器记录未交付轨迹（missing_trajectory=1、n_actions=0、sim_steps=0），本次成功值为0；五个节点均无可确认的受评终值，不能把会话测试中合下扣或搬运护罩的结果计入进度。断点为unknown／首步前；会话在预算附近结束时仍在调试抓取，未见保存交付或最终报告。本次没有视频和状态日志，按N9及D23保留节点未知、接触不适用，现象与根因均为空。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "无视频、无状态日志，评分器仅报告missing_trajectory=1、n_actions=0；五节点unknown、断点unknown/首步前、现象与根因为空并附gap_zh，符合N9与D23(a)，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": "2026-10-30T22:34:17Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work_rerun1/runs/dexhand-fan-guard-install-i00-privileged--ftzpnvq/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work_rerun1/runs/dexhand-fan-guard-install-i00-privileged--ftzpnvq/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "4be4b23257f871406912bec1d414fbfe1bcf2f8726cab8fd37c71ae5b4da9aa7",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/fan_guard_install/trajectories/dexhand-fan-guard-install-i00-privileged--wkigebp.json

@@ -0,0 +1,296 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-fan-guard-install-i00-privileged--wkigebp",
+ "taskKey": "dexhand/fan_guard_install",
+ "dag_sha256": "780b495033f927e2f99d6c0c605a490e62e3855b405d828e8318daf24d90f374",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.787,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.241,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 5,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "guard_seated": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 92,
+      "text_zh": "观察：log step 92 的输出目录检查没有列出交付文件；inputs/run.json 的评分器字段为 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“前护罩就位”的节点结果；本次没有视频或状态日志，无法判定该节点终值。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付受评轨迹；会话调试中的手部接触不计入本次受评轨迹。"
+     }
+    ]
+   },
+   "clip_45_closed": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 92,
+      "text_zh": "观察：log step 92 的输出目录检查没有列出交付文件；inputs/run.json 的评分器字段为 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“合上45度扣”的节点结果；本次没有视频或状态日志，无法判定该节点终值。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付受评轨迹；会话调试中的手部接触不计入本次受评轨迹。"
+     }
+    ]
+   },
+   "clip_135_closed": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 92,
+      "text_zh": "观察：log step 92 的输出目录检查没有列出交付文件；inputs/run.json 的评分器字段为 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“合上135度扣”的节点结果；本次没有视频或状态日志，无法判定该节点终值。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付受评轨迹；会话调试中的手部接触不计入本次受评轨迹。"
+     }
+    ]
+   },
+   "clip_225_closed": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 92,
+      "text_zh": "观察：log step 92 的输出目录检查没有列出交付文件；inputs/run.json 的评分器字段为 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“合上225度扣”的节点结果；本次没有视频或状态日志，无法判定该节点终值。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付受评轨迹；会话调试中的手部接触不计入本次受评轨迹。"
+     }
+    ]
+   },
+   "clip_315_closed": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 92,
+      "text_zh": "观察：log step 92 的输出目录检查没有列出交付文件；inputs/run.json 的评分器字段为 missing_trajectory=1、n_actions=0、sim_steps=0，未报告“合上315度扣”的节点结果；本次没有视频或状态日志，无法判定该节点终值。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "未交付受评轨迹；会话调试中的手部接触不计入本次受评轨迹。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "concentric_m",
+   "unit": "m",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "abs_seat_axial_m",
+   "unit": "m",
+   "threshold": 0.0025,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "parallel_rad",
+   "unit": "rad",
+   "threshold": 0.0872664626,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "clip_45_angle_rad",
+   "unit": "rad",
+   "threshold": 0.12,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "clip_135_angle_rad",
+   "unit": "rad",
+   "threshold": 0.12,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "clip_225_angle_rad",
+   "unit": "rad",
+   "threshold": 0.12,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "clip_315_angle_rad",
+   "unit": "rad",
+   "threshold": 0.12,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "guard_linear_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "guard_angular_speed_rad_s",
+   "unit": "rad/s",
+   "threshold": 0.2,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "fan_displacement_m",
+   "unit": "m",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "fan_tilt_rad",
+   "unit": "rad",
+   "threshold": 0.1,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 92,
+   "where": "首步前",
+   "evidence_zh": "观察：log step 92 的最终目录检查未列出任何交付文件；log step 93 中 agent 明确自述未生成轨迹。评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0、infra_error=0，因而没有可定位物理步骤的受评轨迹，断点记 unknown、首步前。"
+  },
+  "unmet": {
+   "guard_seated": null,
+   "clip_45_closed": null,
+   "clip_135_closed": null,
+   "clip_225_closed": null,
+   "clip_315_closed": null
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：本次 inputs/video.json 为 kind=none，输入中没有本次回放或状态日志；评分器只报告轨迹缺失，未报告五个节点终值或原生几何、速度指标。log step 79 的下扣闭合结果及 log step 84、87、91 的抓罩测试均来自 agent 会话调试，未保存为交付轨迹，不能作为本次节点、实际接触或物理失败现象的依据。log step 90 的预算读数仍为 time_left_s=1874.9，log step 93 随后主动结束并承认无交付；没有外部中断证据，不采信其“时间用完”的自述为归因。按 DECISIONS.md D23，agent 侧未交付且无受评物理尝试时，现象与根因留空；不由交付缺失反推交付操作出错，也不由其抓罩诊断推断任务不可行。参考成功视频属于另一运行，不能补足本次证据。"
+ },
+ "summary_zh": "本次没有交付受评轨迹：评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，log step 92 检查输出目录为空，五个节点终值均未知。agent 在 log step 79 的会话测试中闭合了两个下扣，随后继续测试抓罩，但 log step 93 承认未生成轨迹，故断点记 unknown／首步前。没有本次视频、状态日志或逐节点评分结果，按 D23 不将会话测试记为受评进展，现象和根因均留空。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "本次无视频、无状态日志、无交付轨迹（missing_trajectory=1，log step 92 输出目录为空）；五节点 unknown、断点 unknown/首步前、现象与根因留空并附 gap_zh，符合 D23(a)，未把会话测试当作受评事实，也未把 agent 的“时间用完”当作原因。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T22:27:38Z",
+  "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/dexhand-fan-guard-install-i00-privileged--wkigebp/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-fan-guard-install-i00-privileged--wkigebp/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "2612d9a4d17ad575f88e1c5bd77a243dcf0f00d6fe7d583777aabab18822c363",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/fan_guard_install/trajectories/dexhand-fan-guard-install-i00-privileged--z4k5tse.json

@@ -0,0 +1,318 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-fan-guard-install-i00-privileged--z4k5tse",
+ "taskKey": "dexhand/fan_guard_install",
+ "dag_sha256": "780b495033f927e2f99d6c0c605a490e62e3855b405d828e8318daf24d90f374",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.236,
+   "peak_processes": 6,
+   "peak_disk_gb": 0.239,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 5,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "guard_seated": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 13,
+      "text_zh": "观察：本次无视频、无状态日志；评分器记录missing_trajectory=1、n_actions=0、sim_steps=0，未报告“前护罩就位”的节点结果。log step 13记录响应超过32000输出token上限的API错误，无法据此判定该节点的物理终态。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录missing_trajectory=1，未交付可评分的机器人轨迹，因此轨迹级接触不适用。"
+     }
+    ]
+   },
+   "clip_45_closed": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 13,
+      "text_zh": "观察：本次无视频、无状态日志；评分器记录missing_trajectory=1、n_actions=0、sim_steps=0，未报告“合上45度扣”的节点结果。log step 13记录响应超过32000输出token上限的API错误，无法据此判定该节点的物理终态。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录missing_trajectory=1，未交付可评分的机器人轨迹，因此轨迹级接触不适用。"
+     }
+    ]
+   },
+   "clip_135_closed": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 13,
+      "text_zh": "观察：本次无视频、无状态日志；评分器记录missing_trajectory=1、n_actions=0、sim_steps=0，未报告“合上135度扣”的节点结果。log step 13记录响应超过32000输出token上限的API错误，无法据此判定该节点的物理终态。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录missing_trajectory=1，未交付可评分的机器人轨迹，因此轨迹级接触不适用。"
+     }
+    ]
+   },
+   "clip_225_closed": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 13,
+      "text_zh": "观察：本次无视频、无状态日志；评分器记录missing_trajectory=1、n_actions=0、sim_steps=0，未报告“合上225度扣”的节点结果。log step 13记录响应超过32000输出token上限的API错误，无法据此判定该节点的物理终态。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录missing_trajectory=1，未交付可评分的机器人轨迹，因此轨迹级接触不适用。"
+     }
+    ]
+   },
+   "clip_315_closed": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 13,
+      "text_zh": "观察：本次无视频、无状态日志；评分器记录missing_trajectory=1、n_actions=0、sim_steps=0，未报告“合上315度扣”的节点结果。log step 13记录响应超过32000输出token上限的API错误，无法据此判定该节点的物理终态。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录missing_trajectory=1，未交付可评分的机器人轨迹，因此轨迹级接触不适用。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "concentric_m",
+   "unit": "m",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "abs_seat_axial_m",
+   "unit": "m",
+   "threshold": 0.0025,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "parallel_rad",
+   "unit": "rad",
+   "threshold": 0.0872664626,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "clip_45_angle_rad",
+   "unit": "rad",
+   "threshold": 0.12,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "clip_135_angle_rad",
+   "unit": "rad",
+   "threshold": 0.12,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "clip_225_angle_rad",
+   "unit": "rad",
+   "threshold": 0.12,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "clip_315_angle_rad",
+   "unit": "rad",
+   "threshold": 0.12,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "guard_linear_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "guard_angular_speed_rad_s",
+   "unit": "rad/s",
+   "threshold": 0.2,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "fan_displacement_m",
+   "unit": "m",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "fan_tilt_rad",
+   "unit": "rad",
+   "threshold": 0.1,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 13,
+   "where": "首步前",
+   "evidence_zh": "观察：log steps 2–5仅记录agent读取任务及模拟器源码，log steps 6、8、10、12没有工具调用或输出；log step 13以“API Error: Claude's response exceeded the 32000 output token maximum”结束。评分器记录missing_trajectory=1、n_actions=0、sim_steps=0，未产生已交付的评分尝试，无法定位到某个物理步骤。"
+  },
+  "unmet": {
+   "guard_seated": null,
+   "clip_45_closed": null,
+   "clip_135_closed": null,
+   "clip_225_closed": null,
+   "clip_315_closed": null
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：inputs/video.json的kind为none，本次无视频、无状态日志；log steps 6、8、10、12的tok.out各为32000，calls均为空且say、out.text均为空，log step 13以响应超过32000输出token上限的API错误结束。评分器记录missing_trajectory=1、n_actions=0、sim_steps=0、infra_error=0，未报告五个节点或原生物理指标的结果。推断：连续耗尽输出上限后仍无交付，按docs/03_trajectory.md Step 3第5项及DECISIONS.md D23当前处理归为agent侧无交付，不作为外部因素；保留断点unknown／首步前，两层留空，不标“没尝试”，也不从缺交付反推具体能力根因。inputs/oracle/replay.mp4属于另一成功运行，不能作为本次动作或终态证据。"
+ },
+ "summary_zh": "本次没有交付可评分的机器人轨迹，评分器记录missing_trajectory=1、n_actions=0、sim_steps=0；五个节点均无可确定的终态，保留未知。log steps 6、8、10、12各记录32000个输出token却无工具调用或可见文本，log step 13以输出上限错误结束，据此按agent侧无交付处理，非外部因素，保留success=0及断点unknown／首步前。本次无视频、无状态日志或逐节点评分结果，无法判定物理失败现象或具体根因，按D23当前处理将两层留空。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 2,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "fix",
+    "conf": 6,
+    "summary_zh": "节点全unknown、断点unknown/首步前、两层空，结构正确（无视频无状态日志）；但gap_zh与summary把这次中断定性为“外部中断/不是机器人操作失败”，证据不支持：log steps 6/8/10/12四次连续各输出32000 token却无任何工具调用或文本，是agent自身的超长输出（API错误只是它触发的输出上限），按D23应为agent侧无交付，而非外部因素。",
+    "n_issues": 2,
+    "n_major": 1
+   },
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 2,
+    "verdict": "agree",
+    "conf": 9,
+    "summary_zh": "无视频无状态日志；节点全unknown、断点unknown/首步前、两层留空并有gap_zh，符合D23(a)；已按上轮意见改为agent侧无交付，与run.json一致。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T22:30:34Z",
+  "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/dexhand-fan-guard-install-i00-privileged--z4k5tse/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-fan-guard-install-i00-privileged--z4k5tse/04_review_r1"
+   },
+   {
+    "stage": "reconcile",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-fan-guard-install-i00-privileged--z4k5tse/05_reconcile_r1"
+   },
+   {
+    "stage": "traj_review",
+    "round": 2,
+    "ok": true,
+    "space": "work/runs/dexhand-fan-guard-install-i00-privileged--z4k5tse/04_review_r2"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "5f08a2feaf4ae69b4be59867cb0d510efc8233d80074d6d42cea89655d82b500",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/fan_guard_install/trajectories/dexhand-fan-guard-install-i00-standard--d3rrwwm.json

@@ -0,0 +1,278 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-fan-guard-install-i00-standard--d3rrwwm",
+ "taskKey": "dexhand/fan_guard_install",
+ "dag_sha256": "780b495033f927e2f99d6c0c605a490e62e3855b405d828e8318daf24d90f374",
+ "outcome": {
+  "success": 0,
+  "reward": null,
+  "source": "run.json state",
+  "raw": {}
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 5,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "guard_seated": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 43,
+      "text_zh": "观察：log step 43 的机器人服务返回 episode=1、control_steps=25、done=False、reason=None，但未报告“前护罩就位”的判据或节点结果；run.json 的 reward={}、success=null。本次无视频及状态日志，无法判定该节点的终值或达到时间。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本次没有视频、状态日志或任务接触读数，无法确认右手与前护罩之间是否及如何接触；不以任务DAG的典型接触方案代填。"
+     }
+    ]
+   },
+   "clip_45_closed": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 43,
+      "text_zh": "观察：log step 43 的机器人服务返回 episode=1、control_steps=25、done=False、reason=None，但未报告“合上45度扣”的判据或节点结果；run.json 的 reward={}、success=null。本次无视频及状态日志，无法判定该节点的终值或达到时间。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本次没有视频、状态日志或任务接触读数，无法确认右手与45度夹扣之间是否及如何接触；不以任务DAG的典型接触方案代填。"
+     }
+    ]
+   },
+   "clip_135_closed": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 43,
+      "text_zh": "观察：log step 43 的机器人服务返回 episode=1、control_steps=25、done=False、reason=None，但未报告“合上135度扣”的判据或节点结果；run.json 的 reward={}、success=null。本次无视频及状态日志，无法判定该节点的终值或达到时间。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本次没有视频、状态日志或任务接触读数，无法确认右手与135度夹扣之间是否及如何接触；不以任务DAG的典型接触方案代填。"
+     }
+    ]
+   },
+   "clip_225_closed": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 43,
+      "text_zh": "观察：log step 43 的机器人服务返回 episode=1、control_steps=25、done=False、reason=None，但未报告“合上225度扣”的判据或节点结果；run.json 的 reward={}、success=null。本次无视频及状态日志，无法判定该节点的终值或达到时间。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本次没有视频、状态日志或任务接触读数，无法确认右手与225度夹扣之间是否及如何接触；不以任务DAG的典型接触方案代填。"
+     }
+    ]
+   },
+   "clip_315_closed": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 43,
+      "text_zh": "观察：log step 43 的机器人服务返回 episode=1、control_steps=25、done=False、reason=None，但未报告“合上315度扣”的判据或节点结果；run.json 的 reward={}、success=null。本次无视频及状态日志，无法判定该节点的终值或达到时间。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：本次没有视频、状态日志或任务接触读数，无法确认右手与315度夹扣之间是否及如何接触；不以任务DAG的典型接触方案代填。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "concentric_m",
+   "unit": "m",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "abs_seat_axial_m",
+   "unit": "m",
+   "threshold": 0.0025,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "parallel_rad",
+   "unit": "rad",
+   "threshold": 0.0872664626,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "clip_45_angle_rad",
+   "unit": "rad",
+   "threshold": 0.12,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "clip_135_angle_rad",
+   "unit": "rad",
+   "threshold": 0.12,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "clip_225_angle_rad",
+   "unit": "rad",
+   "threshold": 0.12,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "clip_315_angle_rad",
+   "unit": "rad",
+   "threshold": 0.12,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "guard_linear_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "guard_angular_speed_rad_s",
+   "unit": "rad/s",
+   "threshold": 0.2,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "fan_displacement_m",
+   "unit": "m",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "fan_tilt_rad",
+   "unit": "rad",
+   "threshold": 0.1,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 40,
+   "where": "首步前",
+   "evidence_zh": "推断：完整日志中唯一返回执行结果的运动为 log step 40：agent 脚本将腕部z目标从0.95 m提高至0.97 m，机器人服务返回执行25个控制步；log step 43仍报告episode=1、done=False。其余动作是观察、正运动学计算和环境诊断，没有向护罩或夹扣提交操作轨迹的记录，因此只能定位到任务步骤开始前，不能唯一指定某个DAG节点；无视频，不填写视频时刻。"
+  },
+  "unmet": {
+   "guard_seated": null,
+   "clip_45_closed": null,
+   "clip_135_closed": null,
+   "clip_225_closed": null,
+   "clip_315_closed": null
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：inputs/video.json 的kind=none，inputs/中没有本次replay.mp4或states.jsonl；inputs/oracle/replay.mp4属于另一个成功运行，不能证明本次进展。run.json记录state=failed、success=null、reward={}，没有任务节点终值或评分器判负原因；log step 40仅记录25个控制步的腕部抬升测试，log step 43的服务状态仍为done=False。log step 51、54、58只返回后台任务启动信息，最后log step 59是agent自述仍在检查手指配置，未给出完成报告或中断原因。无本次可视证据，无法确认实际接触与物理失败现象；亦无充分证据区分控制困难、决策问题或外部中断，故两层留空，不从failed或总墙钟时长推断根因。"
+ },
+ "summary_zh": "运行记录为failed，但success=null、reward={}，五个任务节点均无可判定的终值，不能将未知计作已完成或未完成。log step 40只确认机器人执行了25个控制步的腕部抬升测试，log step 43仍为未终结的episode；没有开始护罩装配或合扣操作的日志证据，断点记首步前、具体节点未知。本次无视频和状态日志，且缺少最终服务裁决及明确中断原因，实际接触、失败现象和根因均无法确认。",
+ "conf": 5,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "本次无视频、无状态日志，run.json 为 state=failed、success=null、reward={}；五个节点均为 unknown 且 contact_actual 为未知，符合 N9。日志只有只读检查和一次 25 控制步的腕部抬升测试，没有向护罩或夹扣提交操作，断点记首步前（ref unknown）、两层留空并写 gap_zh，符合 D23(a)。引用的 log step 40/43 存在，未把 agent 或墙钟时间换算成视频时间。未发现需要修改的问题。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T22:23:52Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/dexhand-fan-guard-install-i00-standard--d3rrwwm/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-fan-guard-install-i00-standard--d3rrwwm/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "3e1793a8e6e506b7ee6c961d61ba33cd6697d03de9f50a80499289b6b2acc65b",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/jar_lid_open/trajectories/dexhand-jar-lid-open-i00-privileged--n4suxtn.json

@@ -0,0 +1,222 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-jar-lid-open-i00-privileged--n4suxtn",
+ "taskKey": "dexhand/jar_lid_open",
+ "dag_sha256": "b2060ab237a25379dd35ea85a182cf91836d35380185b84c082a6e96311fd522",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 2.399,
+   "peak_processes": 11,
+   "peak_disk_gb": 0.242,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 2,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "lid_unscrewed": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 50,
+      "text_zh": "观察：评分器 reward 报告 missing_trajectory=1、n_actions=0、sim_steps=0，未报告 stage 或本节点结果；inputs/video.json 为 kind=none，且无状态日志。log step 50 是 agent 本地脚本测试，输出 stage 0、turn=0.623 rad、success=False，不能作为已交付评分轨迹的节点值。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告 missing_trajectory=1，没有交付轨迹；会话中右手的测试接触不计为本次评分轨迹的实际接触。"
+     }
+    ]
+   },
+   "lid_on_worktop": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 42,
+      "text_zh": "观察：评分器 reward 报告 missing_trajectory=1，未提供抬离、放置、hold_steps 或本节点终值；本次无视频及状态日志。log step 42 的脚本仅在 s.success 为真时保存轨迹，后续测试没有成功保存记录；不能从总分为零推定罐盖放置节点为假。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告 missing_trajectory=1，没有可标注放置接触的交付轨迹。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "max_seated_turn_rad",
+   "unit": "rad",
+   "threshold": 1.049342105263158,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "max_clearance_m",
+   "unit": "m",
+   "threshold": 0.03,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "lid_distance_m",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "lid_tilt_rad",
+   "unit": "rad",
+   "threshold": 0.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "lid_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "lid_spin_rad_s",
+   "unit": "rad/s",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "jar_tilt_rad",
+   "unit": "rad",
+   "threshold": 0.1,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "jar_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "jar_slide_m",
+   "unit": "m",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 50,
+   "where": "首步前",
+   "evidence_zh": "观察：评分器报告 missing_trajectory=1、n_actions=0、sim_steps=0、infra_error=0。log step 40、42 的保存语句均由 if s.success 保护；最后的 log step 50 仍为 agent 的本地测试，输出 success=False，之后无交付确认或最终汇报。推断：没有形成可评分的物理轨迹，不能把会话中的拧盖测试选作评分轨迹的断点，故定位为 unknown／首步前。"
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：inputs/ 只有本次 run.json、video.json 等记录及另一次成功运行的 oracle 视频，没有本次 replay.mp4 或 states.jsonl；评分器确认轨迹缺失，且没有逐节点或原生物理指标结果。run.json 的 wall_s=3605、time_limit_s=3600，最后的 log step 50 仍是未成功的本地脚本测试。推断：会话在预算附近结束且未交付；时间耗尽本身不作现象或根因。按 DECISIONS.md D23，未交付时不把会话测试中的接触、罐身倾斜或逆解问题移作评分轨迹现象，也不据缺交付反推根因；两层均留空。按 N9，两节��� final 均为 null，全部 intervals 为 null；九项原生指标未由评分器报告，均留 null。成功 oracle 属于另一运行，不能补足本次证据。",
+  "unmet": {
+   "lid_unscrewed": null,
+   "lid_on_worktop": null
+  }
+ },
+ "summary_zh": "评分器未收到交付轨迹（missing_trajectory=1、n_actions=0），本次无视频及状态日志，两节点的达成情况均未知。agent 的保存代码只在本地测试成功后写出轨迹（log step 40、42），最后一次测试仍未成功（log step 50），会话用时接近预算上限；这些会话测试不能代替评分轨迹。断点记为 unknown／首步前，按 D23 留空现象及根因；评分器也未报告九项原生物理指标。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "评分器 missing_trajectory=1、n_actions=0，无视频无状态日志；两节点 unknown、断点 unknown／首步前、现象与根因留空并写 gap_zh，符合 D23(a) 与 N9。log step 40、42 的保存语句确由 if s.success 保护，step 50 为未成功的本地测试，引用属实。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": "2026-10-30T22:19:57Z",
+  "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/dexhand-jar-lid-open-i00-privileged--n4suxtn/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-jar-lid-open-i00-privileged--n4suxtn/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "7f7b924621f200d72ad9de18b1cb9a259c4cc4d1e1719ea786b7522e17e740a3",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/jar_lid_open/trajectories/dexhand-jar-lid-open-i00-privileged--pntki5r.json

@@ -0,0 +1,222 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-jar-lid-open-i00-privileged--pntki5r",
+ "taskKey": "dexhand/jar_lid_open",
+ "dag_sha256": "b2060ab237a25379dd35ea85a182cf91836d35380185b84c082a6e96311fd522",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 2.124,
+   "peak_processes": 7,
+   "peak_disk_gb": 0.251,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 2,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "lid_unscrewed": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 57,
+      "text_zh": "观察：评分器 reward 记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告评分轨迹的 stage 或拧盖结果；inputs/video.json 的 kind=none，亦无状态日志。log step 57 的 stage=1 来自 agent 的会话测试，不能作为交付轨迹完成拧盖的证据，故终值未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器 reward.missing_trajectory=1；未交付评分轨迹，轨迹级接触不适用，会话测试中的右手与罐盖接触不计入。"
+     }
+    ]
+   },
+   "lid_on_worktop": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 60,
+      "text_zh": "观察：评分器 reward 记录 missing_trajectory=1，未报告评分轨迹的放盖状态或 hold_steps；本次没有评分视频及状态日志。log step 60 的两组 success=True、stage=3、hold_steps=50 是从 snap_lifted 恢复后进行的会话测试；log step 63 仍在测试放盖，log step 64 随后显示过滤器中断，不能据此给评分节点赋真值。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器 reward.missing_trajectory=1；未交付评分轨迹，轨迹级接触不适用，会话测试中的右手与罐盖接触不计入。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "max_seated_turn_rad",
+   "unit": "rad",
+   "threshold": 1.049342105263158,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "max_clearance_m",
+   "unit": "m",
+   "threshold": 0.03,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "lid_distance_m",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "lid_tilt_rad",
+   "unit": "rad",
+   "threshold": 0.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "lid_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "lid_spin_rad_s",
+   "unit": "rad/s",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "jar_tilt_rad",
+   "unit": "rad",
+   "threshold": 0.1,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "jar_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "jar_slide_m",
+   "unit": "m",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 64,
+   "where": "首步前",
+   "evidence_zh": "观察：log step 64 的最终消息为“This request was blocked by Gemini's filters”，无后续调用或交付；评分器 reward 为 missing_trajectory=1、n_actions=0、sim_steps=0。推断：供应商过滤器在交付前中断会话，无评分轨迹可定位物理失败步骤；首步前仅指评分尝试，不代表 agent 未进行会话测试。"
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：本次 inputs/video.json 为 kind=none，无评分视频和状态日志，评分器仅报告缺少轨迹，未给出两个节点的终态及原生指标。log step 57–63 是包含快照恢复的会话测试，其中 log step 60 虽出现成功结果，但并非已保存且验证的交付；log step 64 显示 Gemini 过滤器中断。推断：按外部中断规则，本次应重试并排除出 agent 失败分析；保留目录 success=0，节点终值均未知，断点 unknown／首步前，现象与根因均留空，不将会话失败、缺交付或中断标为没尝试、控制或交付错误。",
+  "unmet": {
+   "lid_unscrewed": null,
+   "lid_on_worktop": null
+  }
+ },
+ "summary_zh": "观察：目录 success=0，评分器报告缺少轨迹且 n_actions=0，无法确认拧开罐盖或罐盖放到桌面两个评分节点，终值均保留未知。观察：log step 60 的会话测试曾返回成功，但 log step 64 最终显示 Gemini 过滤器拦截，未产生评分交付。推断：这是交付前的外部中断，应重试并排除出 agent 失败分析；没有本次评分视频或状态日志，无法定位物理断点或填写现象、根因。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 9,
+    "summary_zh": "无视频、无状态日志；评分器 missing_trajectory=1、n_actions=0；log step 64 为 Gemini 过滤器拦截，未交付。标注（节点 unknown、断点 unknown/首步前、现象与根因为空并附 gap_zh）符合外部中断规则，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": "2026-10-30T22:19:18Z",
+  "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/dexhand-jar-lid-open-i00-privileged--pntki5r/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-jar-lid-open-i00-privileged--pntki5r/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "bc9033268668712ba4a4bde58f63a2d537f26fe24a9bcbd7499a1a348fc045a5",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/jar_lid_open/trajectories/dexhand-jar-lid-open-i00-privileged--ryz2dvf.json

@@ -0,0 +1,232 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-jar-lid-open-i00-privileged--ryz2dvf",
+ "taskKey": "dexhand/jar_lid_open",
+ "dag_sha256": "b2060ab237a25379dd35ea85a182cf91836d35380185b84c082a6e96311fd522",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 2.784,
+   "peak_processes": 11,
+   "peak_disk_gb": 0.26,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 2,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "lid_unscrewed": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器 reward 记录 missing_trajectory=1、n_actions=0、sim_steps=0、success=0，没有报告受评轨迹的 stage 或拧盖节点结果；inputs/video.json 为 kind=none，且没有状态日志，无法判定该节点终值。"
+     },
+     {
+      "t": null,
+      "log_step": 459,
+      "text_zh": "观察：log step 459 的输出来自 agent 的 test_grasp_and_lift.py 会话测试，含测试控制步 430 的“Stage=1, Turn=1.054”；该测试并非已交付轨迹的重放，不能据此认定受评节点完成，也不能转换成视频时刻。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "评分器记录 missing_trajectory=1，未交付受评轨迹；会话调试中的右手接触不作为本次轨迹的实际接触。"
+     }
+    ]
+   },
+   "lid_on_worktop": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：评分器 reward 只有缺交付及整体失败记录，没有报告放盖节点、stage、hold_steps 或终态验收字段；本次没有视频或状态日志，无法判定盖是否曾抬离并在桌面松手保持。"
+     },
+     {
+      "t": null,
+      "log_step": 549,
+      "text_zh": "观察：log step 549 是 agent 启动 test_symmetric_lift.py 会话抬盖测试，log step 550 显示该后台任务仍在运行，log step 551 仅自述等待；没有该测试结果的交付或重放验证记录。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "评分器记录 missing_trajectory=1，未交付受评轨迹；会话调试中的右手接触不作为本次轨迹的实际接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "max_seated_turn_rad",
+   "unit": "rad",
+   "threshold": 1.049342105263158,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "max_clearance_m",
+   "unit": "m",
+   "threshold": 0.03,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "lid_distance_m",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "lid_tilt_rad",
+   "unit": "rad",
+   "threshold": 0.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "lid_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "lid_spin_rad_s",
+   "unit": "rad/s",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "jar_tilt_rad",
+   "unit": "rad",
+   "threshold": 0.1,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "jar_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "jar_slide_m",
+   "unit": "m",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 551,
+   "where": "首步前",
+   "evidence_zh": "观察：评分器 reward 的 missing_trajectory=1、n_actions=0、sim_steps=0，说明没有交付受评动作序列。log step 549–551 仍是 agent 启动并等待会话抬盖测试，最后自述“Waiting for task-876 to complete.”；没有受评物理过程可定位失败步骤或视频时刻。"
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：inputs/ 中有完整 inputs/run.json、inputs/video.json 和另一成功运行的 oracle 视频，但没有本次 replay 视频或状态日志；评分器记录缺交付，未报告任���节点终值或原生物理指标。会话测试的 stage、接触和脚本异常不能替代受评轨迹，故两个节点 final 与 intervals 均为 null，接触为不适用，现象及根因留空。运行记录 wall_s=3669、time_limit_s=3600，日志末尾仍在等待测试；这些记录不足以证明某项物理失败或特定保存操作错误，也没有依据把失败归为基础设施故障。",
+  "unmet": {
+   "lid_unscrewed": null,
+   "lid_on_worktop": null
+  }
+ },
+ "summary_zh": "评分器记录未交付轨迹（missing_trajectory=1），受评动作数与仿真步数均为 0，拧开罐盖和罐盖放到桌面两个节点均无法确认终值。日志中的 Stage 1 是会话测试结果，末尾 log step 549–551 仍在启动并等待抬盖测试，不能视作已交付进度；断点记为 unknown／首步前。没有本次视频或状态日志，无法确定受评物理失败现象与根因，均留空。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "本次无 replay 视频、无状态日志，评分器记录 missing_trajectory=1、n_actions=0；标注把节点 final 置 unknown、断点 unknown/首步前、现象与根因留空并写 gap_zh，符合 03 规则（未交付、不用会话测试作事实、不标没尝试、不把超时当原因）。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": "2026-10-30T22:19:55Z",
+  "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/dexhand-jar-lid-open-i00-privileged--ryz2dvf/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-jar-lid-open-i00-privileged--ryz2dvf/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "cd54db59e301afa7bcff4e43b8d583b0854ebe2d981e3592d1d6860176bae24b",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/pancake_serve/trajectories/dexhand-pancake-serve-i00-privileged--gm4qrmm.json

@@ -0,0 +1,241 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-pancake-serve-i00-privileged--gm4qrmm",
+ "taskKey": "dexhand/pancake_serve",
+ "dag_sha256": "88a62bb839c4f72417c49b00bde10e86ccf1f60461bb3d1dbe0d25557e9b5275",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 3.414,
+   "peak_processes": 9,
+   "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": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 3,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "turner_in_hand": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 165,
+      "text_zh": "观察：inputs/run.json 的评分器字段 missing_trajectory=1、n_actions=0、sim_steps=0，未报告拿起锅铲的节点结果，且本次无视频或状态日志。log step 165 的握铲、提铲输出属于 agent 的会话测试，不能确定受评节点终值。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1；没有交付轨迹，无法记录该步骤的受评接触。"
+     }
+    ]
+   },
+   "pancake_on_blade": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 453,
+      "text_zh": "观察：评分器仅报告 missing_trajectory=1、success=0，未报告 on_blade 或阶段结果；本次无视频或状态日志。log step 447、451 是 agent 的托饼参数搜索及检查脚本，log step 453 仍在等待测试，不能作为受评轨迹托饼成功或失败的节点证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1；没有交付轨迹，无法记录该步骤的受评接触。"
+     }
+    ]
+   },
+   "pancake_on_plate": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 453,
+      "text_zh": "观察：评分器未报告 served、hold 或装盘阶段结果，missing_trajectory=1 且本次无视频或状态日志；总 success=0 不单独决定装盘节点终值。日志末尾 log step 451—453 仍为托饼测试，没有可核实的交付装盘结果。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1；没有交付轨迹，无法记录该步骤的受评接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "pancake_lift_m",
+   "unit": "m",
+   "threshold": 0.015,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "pancake_tilt_deg",
+   "unit": "deg",
+   "threshold": 30,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "pancake_plate_dist_m",
+   "unit": "m",
+   "threshold": 0.058,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "served_pancake_tilt_deg",
+   "unit": "deg",
+   "threshold": 20,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "pancake_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "pancake_ang_speed_rad_s",
+   "unit": "rad/s",
+   "threshold": 0.3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "blade_clearance_m",
+   "unit": "m",
+   "threshold": 0.08,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "pan_shift_m",
+   "unit": "m",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "plate_shift_m",
+   "unit": "m",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 453,
+   "where": "首步前",
+   "evidence_zh": "观察：评分器报告 missing_trajectory=1、n_actions=0、sim_steps=0、infra_error=0；log step 451—452 中 agent 编写并启动托饼检查脚本，log step 453 最后自述仍在等待 task-796 完成，未见轨迹导出及新进程回放验证记录。没有交付的受评动作，无法定位某个物理步骤的失败。"
+  },
+  "unmet": {
+   "pancake_on_blade": null,
+   "pancake_on_plate": null
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：本次 inputs/video.json 的 kind=none，inputs/ 无本次回放和状态日志；评分器明确报告缺少交付轨迹，亦未给出逐节点终值及原生指标。完整会话包含多次 reset、调试和 snapshot/restore（如 log step 447、451），这些测试不构成已交付受评轨迹；oracle 视频仅属于另一成功运行，不能补作本次证据。按 DECISIONS.md N9，各节点 intervals 与 final 均为 null；按 D23，缺交付时现象和根因均留空，不能把会话测试的接触失败或缺交付本身推为物理现象或交付操作根因。日志未明确记载最终终止事件，不能仅凭 wall_s=3622 与 time_limit_s=3600 断定具体终止机制。"
+ },
+ "summary_zh": "本次评分失败，评分器明确记录缺少交付轨迹、受评动作数和仿真步数均为0，三个节点都没有可确认的终值。agent 在会话中反复调试握铲与托饼，最后仍等待托饼检查测试（log step 453）；断点只能记为 unknown／首步前，缺少证据确定物理失败现象及根因。本次无视频、无状态日志，按 N9、D23 保留节点未知及空的失败原因层，不能将会话测试或另一运行的成功视频当作本次进展。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 7,
+    "summary_zh": "本次无视频、无状态日志，评分器给出 missing_trajectory=1、n_actions=0。节点全 unknown、断点 unknown/首步前、现象与根因留空并写 gap_zh，符合 N9 和 D23(a)。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": "2026-10-30T22:28:53Z",
+  "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/dexhand-pancake-serve-i00-privileged--gm4qrmm/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-pancake-serve-i00-privileged--gm4qrmm/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "e12735c660f8e06abba317ece87e4e3869f3560c04fe74dec8ac4408f589fb79",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/picture_hang/trajectories/dexhand-picture-hang-i00-privileged--lcjsttr.json

@@ -0,0 +1,241 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-picture-hang-i00-privileged--lcjsttr",
+ "taskKey": "dexhand/picture_hang",
+ "dag_sha256": "aa6a1b7c76b911e4015893803bb156ab69766ab91b47214535622219f48b7b8b",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "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": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 3,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "picture_lifted": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 18,
+      "text_zh": "观察：inputs/run.json 的评分器字段为 missing_trajectory=1、n_actions=0、sim_steps=0，未报告受评相框离台结果；本次无视频及状态日志。log step 18 的 stage=1 来自 agent 的夹持抬升测试，不能作为未交付轨迹的节点终值。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告 missing_trajectory=1，未交付受评轨迹；会话测试中的机器人双手接触不计为本轨迹的实际接触。"
+     }
+    ]
+   },
+   "left_ring_on_hook": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 19,
+      "text_zh": "观察：评分器报告 missing_trajectory=1，未报告受评左环穿挂或承重结果；本次无视频及状态日志。log step 19 的 threaded 与 load 读数来自 agent 的草稿测试，不是已交付轨迹的评分结果，左环节点终值无法判定。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告 missing_trajectory=1，未交付受评轨迹；会话测试中的机器人双手接触不计为本轨迹的实际接触。"
+     }
+    ]
+   },
+   "right_ring_on_hook": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 19,
+      "text_zh": "观察：评分器报告 missing_trajectory=1，未报告受评右环穿挂或承重结果；本次无视频及状态日志。log step 19 的 threaded 与 load 读数来自 agent 的草稿测试，不是已交付轨迹的评分结果，右环节点终值无法判定。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告 missing_trajectory=1，未交付受评轨迹；会话测试中的机器人双手接触不计为本轨迹的实际接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "lift_clearance_m",
+   "unit": "m",
+   "threshold": 0.02,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "hook_load_share_l",
+   "unit": "ratio",
+   "threshold": 0.2,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "hook_load_share_r",
+   "unit": "ratio",
+   "threshold": 0.2,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "total_hook_load_share",
+   "unit": "ratio",
+   "threshold": 0.85,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "abs_roll_deg",
+   "unit": "deg",
+   "threshold": 3,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "tilt_rad",
+   "unit": "rad",
+   "threshold": 0.1,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "angular_speed_rad_s",
+   "unit": "rad/s",
+   "threshold": 0.15,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "hold_steps",
+   "unit": "control_step",
+   "threshold": 50,
+   "direction": "higher_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 20,
+   "where": "首步前",
+   "evidence_zh": "观察：评分器报告 missing_trajectory=1、n_actions=0、sim_steps=0、infra_error=0；log step 19 中 agent 将草稿动作保存到会话工作区的 draft_actions.npy，log step 20 仍在重置并调试该草稿，未见交付成功记录。推断：没有可定位物理失败步骤的受评轨迹，断点记为 unknown、位置为首步前；不���会话测试失败作为受评断点。"
+  },
+  "unmet": {
+   "left_ring_on_hook": null,
+   "right_ring_on_hook": null
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "本次 inputs/video.json 为 kind=none，且无状态日志；评分器仅报告缺交付及零动作，没有逐节点结果，故各节点 final 与 intervals 均为 null。inputs/oracle/replay.mp4 是另一成功运行，只能作任务参考，不能补充本次执行证据。log step 18–20 是会话测试，部分代码和输出还被原始记录截断；没有最终完成报告，也没有足以解释缺交付的具体保存错误或控制故障证据。运行 wall_s=3603、time_limit_s=3600 提示预算耗尽，但预算本身不是失败现象或根因；依 D23 两层均留空，不从缺交付反推没尝试或交付操作出错。"
+ },
+ "summary_zh": "评分器报告未交付轨迹（missing_trajectory=1，n_actions=0，sim_steps=0），相框离台和左右环挂钩三个节点均无可判定的受评终值。log step 18–20 中 agent 进行了抬框、草稿生成和回放调试，但这些会话测试不能计入受评进度；断点记为首步前的 unknown。由于本次无视频、无状态日志，且缺交付的具体原因证据不足，失败现象与根因均留空。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "无视频、无状态日志；评分器 missing_trajectory=1、n_actions=0，未交付受评轨迹。三个节点 unknown、断点 unknown/首步前、现象与根因留空并写 gap_zh，符合 D23(a)；log step 18-20 的会话读数未被当作受评事实，预算耗尽也未被当作现象或根因。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T22:20:27Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/dexhand-picture-hang-i00-privileged--lcjsttr/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-picture-hang-i00-privileged--lcjsttr/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "f8992e72ea110ca6178ccae8b24f725a627b42911409563513aab75d748f4b07",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/ram_module_install/trajectories/dexhand-ram-module-install-i00-privileged--lswxfgb.json

@@ -0,0 +1,222 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-ram-module-install-i00-privileged--lswxfgb",
+ "taskKey": "dexhand/ram_module_install",
+ "dag_sha256": "49c2ef7342ff96d98bcaa9c9bbf565c347b910e312bf0fe85bfe24ce2574b91f",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.251,
+   "peak_processes": 6,
+   "peak_disk_gb": 0.239,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 2,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "module_seated": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 26,
+      "text_zh": "观察：评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，未报告本节点评分状态；本次没有视频或状态日志。log step 24–26 是 agent 的指尖运动学调试，并非交付轨迹，不能证明内存插入，故终值未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1，未交付可供评分的动作轨迹，实际接触标注不适用。"
+     }
+    ]
+   },
+   "latch_closed": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 32,
+      "text_zh": "观察：评分器记录 missing_trajectory=1，未报告 latched、seated 或锁扣终态；本次没有视频或状态日志。log step 27–28 仍是 agent 查阅手模型几何，后续 log step 29、32 无调用、输出或最终陈述，不能确定本节点终值。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1，未交付可供评分的动作轨迹，实际接触标注不适用。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "min_end_depths_m",
+   "unit": "m",
+   "threshold": 0.0015,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "max_end_depths_m",
+   "unit": "m",
+   "threshold": 0.0026,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "lateral_m",
+   "unit": "m",
+   "threshold": 0.0005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "abs_along_slot_offset_m",
+   "unit": "m",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "upright_err_deg",
+   "unit": "deg",
+   "threshold": 3.0,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "keyed_err_deg",
+   "unit": "deg",
+   "threshold": 3.0,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "latch_rad",
+   "unit": "rad",
+   "threshold": 0.08,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "module_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "module_ang_speed_rad_s",
+   "unit": "rad/s",
+   "threshold": 0.2,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": null,
+   "where": "首步前",
+   "evidence_zh": "观察：评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0、infra_error=0；完整日志未显示轨迹保存或提交，最后 log step 32 无调用、输出或最终陈述。推断：没有交付评分回合，断点位于首步前，不能从会话调试选择某个物理失败节点。"
+  },
+  "unmet": {
+   "module_seated": null,
+   "latch_closed": null
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：本次无视频、无状态日志，评分器未报告两节点终值或任何原生几何指标；inputs/oracle/replay.mp4 属于另一成功运行，不能证明本次进展。log step 18 的几何检查报错及 log step 25 的数组维度报错均发生于会话调试，分别在 log step 19–20、26 修正，不能作为未交付评分回合的失败现象或根因。观察：run.wall_s=3603 s、time_limit_s=3600 s，末尾无停止原因陈述；推断：会话耗尽墙钟预算时仍无交付，但没有证据定位具体能力缺陷，按 D23 将现象和根因均留空，不以缺交付或超时反推原因。"
+ },
+ "summary_zh": "观察：评分器记录未交付轨迹（missing_trajectory=1），评分动作数及仿真步数均为0，插入内存与合上锁扣两节点均无已确认完成，终值保留未知。推断：本次在墙钟预算耗尽时仍未形成评分轨迹，断点记 unknown／首步前；日志中的几何检查、保持动作测速与指尖运动学调试不代表交付回合进展。观察：没有本次视频、状态日志或逐节点服务结果，无法确认物理失败现象及其根因，故两层留空。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "本次无视频、无状态日志，评分器 missing_trajectory=1、n_actions=0；文件按 D23(a) 将两节点终值记为未知、现象与根因留空并写 gap_zh，断点 unknown／首步前，会话调试未被当作交付回合事实，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": "2026-10-30T22:24:34Z",
+  "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/dexhand-ram-module-install-i00-privileged--lswxfgb/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-ram-module-install-i00-privileged--lswxfgb/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "d1835529618f815cbff2d968965ab3d643b4541539ef04c879016cd20d8db11a",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/ram_module_install/trajectories/dexhand-ram-module-install-i00-privileged--ptnjpwp.json

@@ -0,0 +1,222 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-ram-module-install-i00-privileged--ptnjpwp",
+ "taskKey": "dexhand/ram_module_install",
+ "dag_sha256": "49c2ef7342ff96d98bcaa9c9bbf565c347b910e312bf0fe85bfe24ce2574b91f",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.671,
+   "peak_processes": 6,
+   "peak_disk_gb": 0.252,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 2,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "module_seated": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 76,
+      "text_zh": "观察：log step 76 的输出来自 agent 新建 Session 的本地测试，报告 seated=True、latched=True 或 stage=3、succ=True；日志没有将该测试保存为交付轨迹的操作。评分器 reward 记录 missing_trajectory=1、n_actions=0、sim_steps=0，且 inputs/video.json 的 kind=none；无受评回放或状态日志可确认“插入内存”节点的最终值，本地测试结果不能替代交付结果。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1、n_actions=0；没有交付轨迹，故不将本地测试中的右手与内存条接触计入受评轨迹。"
+     }
+    ]
+   },
+   "latch_closed": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 77,
+      "text_zh": "观察：log step 77 的输出来自 agent 新建 Session 的本地测试，报告 seated=True、latched=True 或 stage=3、succ=True；日志没有将该测试保存为交付轨迹的操作。评分器 reward 记录 missing_trajectory=1、n_actions=0、sim_steps=0，且 inputs/video.json 的 kind=none；无受评回放或状态日志可确认“合上锁扣”节点的最终值，本地测试结果不能替代交付结果。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器记录 missing_trajectory=1、n_actions=0；没有交付轨迹，故不将本地测试中的右手与内存条接触计入受评轨迹。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "min_end_depths_m",
+   "unit": "m",
+   "threshold": 0.0015,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "max_end_depths_m",
+   "unit": "m",
+   "threshold": 0.0026,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "lateral_m",
+   "unit": "m",
+   "threshold": 0.0005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "abs_along_slot_offset_m",
+   "unit": "m",
+   "threshold": 0.001,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "upright_err_deg",
+   "unit": "deg",
+   "threshold": 3.0,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "keyed_err_deg",
+   "unit": "deg",
+   "threshold": 3.0,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "latch_rad",
+   "unit": "rad",
+   "threshold": 0.08,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "module_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.01,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "module_ang_speed_rad_s",
+   "unit": "rad/s",
+   "threshold": 0.2,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 79,
+   "where": "首步前",
+   "evidence_zh": "观察：log step 79 的最终消息为“This request was blocked by Gemini's filters”，随后没有工具调用或交付记录；评分器 reward 记录 missing_trajectory=1、n_actions=0、sim_steps=0。本次未形成已交付的受评轨迹，无法定位物理步骤断点。"
+  },
+  "unmet": {
+   "module_seated": null,
+   "latch_closed": null
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：本次没有受评视频（inputs/video.json：kind=none），也没有状态日志；评分器未报告两个节点的最终状态或任何 DAG 原生指标。log step 76、77 的成功输出仅属未保存交付的本地测试，log step 78 又进行了另一测试，log step 79 随后记录 Gemini 过滤器阻断。推断：这是交��前被模型提供方过滤器中断的外因运行，按 CASES.md FAIL-5 留待重测，不纳入 agent 失败归因；现象、根因均留空，不标“没尝试”，也不凭缺少交付推断交付操作出错。成功参考录像不能证明本次轨迹的节点或接触。"
+ },
+ "summary_zh": "观察：评分器记录 success=0、missing_trajectory=1、n_actions=0，没有交付的受评轨迹，两个节点的最终值均无法确认；log step 76、77 的本地测试虽报告成功，但未保存交付。观察：log step 79 记录 Gemini 过滤器阻断；推断：本次属于交付前的外因中断，应重测并排除在 agent 失败归因之外，断点为 unknown／首步前，现象和根因留空。观察：本次没有受评视频或状态日志，全部节点时间与原生指标均不可确定。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 9,
+    "summary_zh": "标注正确：无交付轨迹（missing_trajectory=1, n_actions=0），无视频；log step 79 为 Gemini 过滤器阻断，属外因，断点 unknown/首步前，现象与根因留空且 gap_zh 说明，符合 CASES.md FAIL-5；本地测试未当作受评结果。",
+    "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": "2026-10-30T22:22:33Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/dexhand-ram-module-install-i00-privileged--ptnjpwp/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-ram-module-install-i00-privileged--ptnjpwp/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "189c98486714cbb9ee9cc5f11af364bc7c6a6dc1a1818446df760db4618ce600",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/remote_battery_swap/trajectories/dexhand-remote-battery-swap-i00-privileged--8zcfmjs.json

@@ -0,0 +1,297 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-remote-battery-swap-i00-privileged--8zcfmjs",
+ "taskKey": "dexhand/remote_battery_swap",
+ "dag_sha256": "448d38d5df99705f24cec0642f2338ce7100127136b831930f8d1fee7f80f537",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0,
+   "peak_memory_gb": 3.629,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.268,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 4,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "battery_cover_opened": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 598,
+      "text_zh": "观察：评分器 reward 报告 missing_trajectory=1、n_actions=0、sim_steps=0，未报告正式回放的开盖阶段；inputs/video.json 为 kind=none，亦无状态日志。log step 598 的 Stage 1 latched: True 来自 agent 运行 test_2f_pinch.py 的会话测试，不能作为交付回合的开盖结果。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器 reward 报告 missing_trajectory=1、n_actions=0；无交付轨迹可标注实际接触，会话测试接触不计入。"
+     }
+    ]
+   },
+   "old_cell_in_tray": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 598,
+      "text_zh": "观察：评分器 reward 报告 missing_trajectory=1，未报告旧电池入托盘的正式结果；本次无视频或状态日志。log step 598 的旧电池坐标、接触列表和 remote_tipped 终止信息属于 agent 的抓取测试，不能确定正式节点末值。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器 reward 报告 missing_trajectory=1、n_actions=0；无交付轨迹可标注实际接触，会话测试接触不计入。"
+     }
+    ]
+   },
+   "new_cell_seated": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 606,
+      "text_zh": "观察：评分器 reward 仅报告缺少轨迹和零动作，没有新电池入槽的阶段或末态字段；本次无视频和状态日志。log step 606 仍是 agent 对捏取姿态指尖向量的计算，输出为空，不能据此确定新电池入槽。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器 reward 报告 missing_trajectory=1、n_actions=0；无交付轨迹可标注实际接触，会话测试接触不计入。"
+     }
+    ]
+   },
+   "battery_cover_closed": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：评分器 reward 未报告电池盖的正式末态，且 missing_trajectory=1；本次无视频或状态日志。任务 DAG 描述的初始闭盖状态并非本次正式末态观测，不能替代末值证据。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器 reward 报告 missing_trajectory=1、n_actions=0；无交付轨迹可标注实际接触，会话测试接触不计入。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "cover_slide_m_open",
+   "unit": "m",
+   "threshold": 0.06,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "cover_slide_m_closed",
+   "unit": "m",
+   "threshold": 0.002,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "tab_depth_m_closed",
+   "unit": "m",
+   "threshold": 0.0015,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "old_cell_tray_local_x_abs",
+   "unit": "m",
+   "threshold": 0.037,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "old_cell_tray_local_y_abs",
+   "unit": "m",
+   "threshold": 0.032,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "old_cell_tray_local_z",
+   "unit": "m",
+   "threshold": 0.014,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "new_cell_slot_error_x_abs",
+   "unit": "m",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "new_cell_slot_error_y_abs",
+   "unit": "m",
+   "threshold": 0.003,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "new_cell_slot_error_z_abs",
+   "unit": "m",
+   "threshold": 0.002,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "new_cell_axis_angle_deg",
+   "unit": "deg",
+   "threshold": 10,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_spin_rad_s",
+   "unit": "rad/s",
+   "threshold": 0.5,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "remote_displacement_m",
+   "unit": "m",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "remote_tilt_deg",
+   "unit": "deg",
+   "threshold": 15,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 606,
+   "where": "首步前",
+   "evidence_zh": "观察：评分器 reward 的 missing_trajectory=1、n_actions=0、sim_steps=0、infra_error=0；log step 604 中 agent 最后自述仍在等待指尖坐标检查，log step 606 又发起捏取姿态向量计算且无返回输出，未见保存并验证交付轨迹的记录。推断：没有可定位物理失败步骤的正式轨迹，按未交付情形标为 unknown／首步前，不把会话抓取测试的失败当作正式断点。"
+  },
+  "unmet": {
+   "battery_cover_opened": null,
+   "old_cell_in_tray": null,
+   "new_cell_seated": null,
+   "battery_cover_closed": null
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：本次 inputs/video.json 为 kind=none（HF 视频 404 且无本地视频），没有状态日志；评分器仅给出 missing_trajectory=1 与零动作，未给四个节点的正式阶段、末态或原生几何指标。inputs/run.json 的 wall_s=3609、time_limit_s=3600，日志结束于 log step 606 的调试调用，未记录明确退出原因；log step 598 的开盖成功、旧电池接触及 remote_tipped 均属于未交付的会话测试。推断：会话持续至预算附近，但不能据此给出物理失败现象或能力根因；依 DECISIONS.md D23，未交付且无可识别正式回合时两层留空，也不从缺交付反推交付操作出错。成功参考视频属于另一运行，不能补足本次证据。"
+ },
+ "summary_zh": "评分器报告未找到交付轨迹、正式动作数和仿真步数均为零，四个节点的末值均未知，无法确认本次正式进度。日志中 agent 曾测试开盖及抓取旧电池，但最后仍在调试，未见保存并验证交付的记录，因此断点为未知／首步前。由于本次没有视频、状态日志或节点级评分结果，接触均不适用，失败现象和根因按未交付规则留空。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "无视频、无状态日志；评分器 missing_trajectory=1、n_actions=0，四个节点 unknown、接触不适用、两层留空并写 gap_zh，符合 D23(a)；会话测试未被当作正式回合事实。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": "2026-10-30T22:29:33Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/dexhand-remote-battery-swap-i00-privileged--8zcfmjs/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-remote-battery-swap-i00-privileged--8zcfmjs/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "83712d4bbcc7e8d5d98ea6aa4d64412a4856daf72bc262a91a29d29d31a09a7b",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/remote_battery_swap/trajectories/dexhand-remote-battery-swap-i00-privileged--vc3unz7.json

@@ -0,0 +1,297 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-remote-battery-swap-i00-privileged--vc3unz7",
+ "taskKey": "dexhand/remote_battery_swap",
+ "dag_sha256": "448d38d5df99705f24cec0642f2338ce7100127136b831930f8d1fee7f80f537",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 2.255,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.266,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 4,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "battery_cover_opened": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 92,
+      "text_zh": "观察：inputs/run.json 的 reward 为 missing_trajectory=1、n_actions=0、sim_steps=0；inputs/video.json 为 kind=none，且无状态日志。评分记录没有报告受评轨迹的开盖节点值；log step 91 的开盖测试以 remote_tipped 结束，log step 92 仍是姿态运动学调试，均非交付回放。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告 missing_trajectory=1，没有交付轨迹；会话测试中的机器人接触不计为本节点的轨迹级接触。"
+     }
+    ]
+   },
+   "old_cell_in_tray": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的 reward 为 missing_trajectory=1、n_actions=0、sim_steps=0；inputs/video.json 为 kind=none，且无状态日志。评分记录没有报告灰色旧电池入托盘的节点值；不能从整体 success=0 或未交付轨迹推定该物理状态为假。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告 missing_trajectory=1，没有交付轨迹；会话测试中的机器人接触不计为本节点的轨迹级接触。"
+     }
+    ]
+   },
+   "new_cell_seated": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的 reward 为 missing_trajectory=1、n_actions=0、sim_steps=0；inputs/video.json 为 kind=none，且无状态日志。评分记录没有报告绿色新电池入槽的节点值；会话中的开盖测试不能证明受评回放中电池的位置、极性或接触。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告 missing_trajectory=1，没有交付轨迹；会话测试中的机器人接触不计为本节点的轨迹级接触。"
+     }
+    ]
+   },
+   "battery_cover_closed": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的 reward 为 missing_trajectory=1、n_actions=0、sim_steps=0；inputs/video.json 为 kind=none，且无状态日志。评分记录没有报告受评轨迹的盖合锁定状态；log step 10、11 中盖已关闭的读数属于调试会话初态，不能替代不存在的交付回放终态。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器报告 missing_trajectory=1，没有交付轨迹；会话测试中的机器人接触不计为本节点的轨迹级接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "cover_slide_m_open",
+   "unit": "m",
+   "threshold": 0.06,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "cover_slide_m_closed",
+   "unit": "m",
+   "threshold": 0.002,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "tab_depth_m_closed",
+   "unit": "m",
+   "threshold": 0.0015,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "old_cell_tray_local_x_abs",
+   "unit": "m",
+   "threshold": 0.037,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "old_cell_tray_local_y_abs",
+   "unit": "m",
+   "threshold": 0.032,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "old_cell_tray_local_z",
+   "unit": "m",
+   "threshold": 0.014,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "new_cell_slot_error_x_abs",
+   "unit": "m",
+   "threshold": 0.005,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "new_cell_slot_error_y_abs",
+   "unit": "m",
+   "threshold": 0.003,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "new_cell_slot_error_z_abs",
+   "unit": "m",
+   "threshold": 0.002,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "new_cell_axis_angle_deg",
+   "unit": "deg",
+   "threshold": 10,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_spin_rad_s",
+   "unit": "rad/s",
+   "threshold": 0.5,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "remote_displacement_m",
+   "unit": "m",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "remote_tilt_deg",
+   "unit": "deg",
+   "threshold": 15,
+   "direction": "lower_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 92,
+   "where": "首步前",
+   "evidence_zh": "观察：评分器 reward 报告 missing_trajectory=1、n_actions=0、sim_steps=0、success=0、infra_error=0；log step 92 是 agent 对开盖手指姿态的运动学验证，之后没有最终汇报或保存轨迹的记录。没有交付可供定位物理失败步骤，故断点为 unknown，位置为首步前。"
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：本次没有受评视频、状态日志或交付轨迹，评分器也未报告四个节点的终值。log step 91 的 remote_tipped 和 log step 92 的姿态验证只属于作者调试，不能作为受评轨迹的失败现象或节点值；成功参考视频 inputs/oracle/replay.mp4 属于另一运行，不能补足本次证据。inputs/run.json 记录 wall_s=3605、time_limit_s=3600，日志结束时仍在调试，但没有可确认具体保存错误或物理断点的交付记录。依 DECISIONS.md D23，无交付且无已证实外因的运行保留失败判定，现象和根因均留空；不从缺交付推导没尝试、交付操作出错或控制根因。",
+  "unmet": {
+   "battery_cover_opened": null,
+   "old_cell_in_tray": null,
+   "new_cell_seated": null,
+   "battery_cover_closed": null
+  }
+ },
+ "summary_zh": "评分器报告未交付轨迹（missing_trajectory=1、n_actions=0、sim_steps=0），本次未通过；四个节点均无受评终态证据，保留未知。日志末尾仍在调试开盖（log step 91、92），不能将调试中的开盖失败当作受评轨迹的断点，因此标为 unknown／首步前。没有本次视频或状态日志，按无交付规则留空失败现象与根因，全部原生指标因评分记录未报告而为 null。",
+ "conf": 8,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "本次无视频、无状态日志、无交付轨迹（评分器 missing_trajectory=1、n_actions=0）。四个节点保持 unknown/null，contact_actual 用 不适用，断点 unknown/首步前，现象与根因为空并带 gap_zh，符合 DECISIONS.md D23 与 N9；会话调试读数未被当作受评事实。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": "2026-10-30T22:29:05Z",
+  "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/dexhand-remote-battery-swap-i00-privileged--vc3unz7/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-remote-battery-swap-i00-privileged--vc3unz7/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "174491f241e4d42f492afb2f6f479eed7eb31a05bae832d405f0943d9b0285a1",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/stool_brace_insert/trajectories/dexhand-stool-brace-insert-i00-privileged--9eo8gcy.json

@@ -0,0 +1,271 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-stool-brace-insert-i00-privileged--9eo8gcy",
+ "taskKey": "dexhand/stool_brace_insert",
+ "dag_sha256": "de74f08a423d134ae656560a17c6b1fb178bebc88b5333b7e92ab23845028b3f",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 0,
+   "infra_error": 0,
+   "missing_trajectory": 1,
+   "n_actions": 0,
+   "sim_steps": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 0.63,
+   "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
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 2,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "brace_in_lower_frame": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 34,
+      "text_zh": "观察：log step 34 的 test_brace.py 会话测试输出 stage=2、seated=True；评分器 reward 却记录 missing_trajectory=1、n_actions=0、sim_steps=0，且未报告本节点结果。本次无受评视频或状态日志，会话测试不作为节点完成证据，final 与 intervals 均为 null。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器 reward.missing_trajectory=1、n_actions=0；未交付受评轨迹，故本节点的实际接触不适用。"
+     }
+    ]
+   },
+   "upper_frame_on_dowels": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 98,
+      "text_zh": "观察：log step 98 仅启动 test_stage34.py 后台测试，log step 99 的 agent 自述仍在等待完成，随后 log step 101 出现 Gemini 过滤器拦截。评分器 reward.missing_trajectory=1、n_actions=0，未报告四个上榫插接结果；无受评视频或状态日志，final 与 intervals 均为 null。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "不适用",
+      "how": "不适用",
+      "note_zh": "观察：评分器 reward.missing_trajectory=1、n_actions=0；未交付受评轨迹，故本节点的实际接触不适用。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "seat_depth_min",
+   "unit": "m",
+   "threshold": 0.012,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "seat_depth_max",
+   "unit": "m",
+   "threshold": 0.019,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "seat_lateral",
+   "unit": "m",
+   "threshold": 0.0016,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "seat_axis_angle",
+   "unit": "rad",
+   "threshold": 0.105,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_linear_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.02,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "max_angular_speed_rad_s",
+   "unit": "rad/s",
+   "threshold": 0.2,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "lower_frame_shift_m",
+   "unit": "m",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "lower_frame_tilt",
+   "unit": "rad",
+   "threshold": 0.1745,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "preseated_depth_min",
+   "unit": "m",
+   "threshold": 0.008,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "preseated_tilt",
+   "unit": "rad",
+   "threshold": 0.1745,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "post_lift_brace_height_above_table",
+   "unit": "m",
+   "threshold": 0.09,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "post_seating_brace_tilt",
+   "unit": "rad",
+   "threshold": 0.349,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "post_lift_upper_frame_tilt",
+   "unit": "rad",
+   "threshold": 0.349,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "unsupported_downward_speed",
+   "unit": "m/s",
+   "threshold": 0.2,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "lift_center_rise",
+   "unit": "m",
+   "threshold": 0.05,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "lift_lowest_corner_clearance",
+   "unit": "m",
+   "threshold": 0.02,
+   "direction": "higher_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": 101,
+   "where": "首步前",
+   "evidence_zh": "观察：log step 98 ���动会话测试，log step 99 的 agent 自述仍在等待，log step 101 的最终消息为“This request was blocked by Gemini's filters”；评分器 reward.missing_trajectory=1、n_actions=0、sim_steps=0。未交付受评轨迹，无法定位物理步骤断点。"
+  },
+  "unmet": {
+   "brace_in_lower_frame": null,
+   "upper_frame_on_dowels": null
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "观察：本次 inputs/video.json 的 kind=none，未提供本次受评视频或状态日志；inputs/oracle/replay.mp4 是另一成功运行的参考视频，不能证明本次进度。log step 101 显示模型提供方 Gemini 过滤器拦截，评分器记录 missing_trajectory=1、n_actions=0、sim_steps=0，交付前受外因中断；按外因规则不标机器人失败现象或根因，也不标“没尝试”。会话中 log step 34 的横撑插好测试及 log step 57 起恢复快照的对侧框测试均非交付轨迹，不能确定两个节点的受评终值；评分器未报告各项原生指标。"
+ },
+ "summary_zh": "评分器记录 success=0、missing_trajectory=1、n_actions=0，未交付受评轨迹，横撑插入下框与装上对侧框两个节点的受评进度均未知。log step 98–99 仍在启动并等待会话装配测试，log step 101 随即显示 Gemini 过滤器拦截，属于交付前外因中断，断点记 unknown／首步前，失败现象与根因留空。本次无受评视频或状态日志，会话测试与其他运行的成功参考视频均不能代替本次节点证据。",
+ "conf": 9,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 9,
+    "summary_zh": "本次无受评视频，评分器 missing_trajectory=1、n_actions=0，log step 101 为 Gemini 过滤器拦截，交付前外因中断。标注（节点 unknown、断点 unknown/首步前、两层为空并有 gap_zh、不标没尝试）符合 03 外因规则。",
+    "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": "2026-10-30T22:27:53Z",
+  "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/dexhand-stool-brace-insert-i00-privileged--9eo8gcy/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-stool-brace-insert-i00-privileged--9eo8gcy/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "35c68aba66f821ac1d058e3d6adc9cc36aa254f38b50b03ab5609b4b6e72f168",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/sweep_dustpan/trajectories/dexhand-sweep-dustpan-i00-privileged--npghzqz.json

@@ -0,0 +1,338 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-sweep-dustpan-i00-privileged--npghzqz",
+ "taskKey": "dexhand/sweep_dustpan",
+ "dag_sha256": "4d82017ee30ab51b0c47be7c7dca8760f7e8fdbb7e1109052499805be552add8",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 1,
+   "hand_wuji": 0,
+   "infra_error": 0,
+   "n_actions": 17400,
+   "sim_steps": 17386,
+   "stages_completed": 0,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 2.082,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.269,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "log",
+  "denominator": 7,
+  "base": null,
+  "peak": null,
+  "final": 0,
+  "nodes": {
+   "orange_package_in_pan": {
+    "method": "log",
+    "intervals": null,
+    "final": false,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 44,
+      "text_zh": "观察：log step 44 中，基准 Session.status() 在仿真 8.693 s 返回 done=True、failed=True、in_pan=0；Session.features() 的 debris_0.inside_pan=False（橙包装），未满足完整入箕判据。这些是基准检查器字段，不是 agent 自定义几何检查。"
+     },
+     {
+      "t": null,
+      "log_step": 44,
+      "text_zh": "推断：log step 44 的基准 Session 返回终止步17386、保存动作17400行，与 reward.sim_steps=17386、n_actions=17400、deterministic=1一致；log step 47 无返回或保存记录，因此将 log step 44 识别为本次评分交付。据该次基准终止结果判定本节点终值为假；没有本次视频，不能确定历史区间。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "机器人脚本用共同推扫动作处理六件碎屑（log step 43、44）；缺少本次视频和状态接触对，不能核实右手扫帚及左手簸箕对橙包装的实际作用方式。"
+     }
+    ]
+   },
+   "red_cap_in_pan": {
+    "method": "log",
+    "intervals": null,
+    "final": false,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 44,
+      "text_zh": "观察：log step 44 中，基准 Session.status() 在仿真 8.693 s 返回 done=True、failed=True、in_pan=0；Session.features() 的 debris_1.inside_pan=False（红瓶盖），未满足完整入箕判据。这些是基准检查器字段，不是 agent 自定义几何检查。"
+     },
+     {
+      "t": null,
+      "log_step": 44,
+      "text_zh": "推断：log step 44 的基准 Session 返回终止步17386、保存动作17400行，与 reward.sim_steps=17386、n_actions=17400、deterministic=1一致；log step 47 无返回或保存记录，因此将 log step 44 识别为本次评分交付。据该次基准终止结果判定本节点终值为假；没有本次视频，不能确定历史区间。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "机器人脚本用共同推扫动作处理六件碎屑（log step 43、44）；缺少本次视频和状态接触对，不能核实右手扫帚及左手簸箕对红瓶盖的实际作用方式。"
+     }
+    ]
+   },
+   "wood_chip_in_pan": {
+    "method": "log",
+    "intervals": null,
+    "final": false,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 44,
+      "text_zh": "观察：log step 44 中，基准 Session.status() 在仿真 8.693 s 返回 done=True、failed=True、in_pan=0；Session.features() 的 debris_2.inside_pan=False（木屑），未满足完整入箕判据。这些是基准检查器字段，不是 agent 自定义几何检查。"
+     },
+     {
+      "t": null,
+      "log_step": 44,
+      "text_zh": "推断：log step 44 的基准 Session 返回终止步17386、保存动作17400行，与 reward.sim_steps=17386、n_actions=17400、deterministic=1一致；log step 47 无返回或保存记录，因此将 log step 44 识别为本次评分交付。据该次基准终止结果判定本节点终值为假；没有本次视频，不能确定历史区间。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "机器人脚本用共同推扫动作处理六件碎屑（log step 43、44）；缺少本次视频和状态接触对，不能核实右手扫帚及左手簸箕对木屑的实际作用方式。"
+     }
+    ]
+   },
+   "blue_package_in_pan": {
+    "method": "log",
+    "intervals": null,
+    "final": false,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 44,
+      "text_zh": "观察：log step 44 中，基准 Session.status() 在仿真 8.693 s 返回 done=True、failed=True、in_pan=0；Session.features() 的 debris_3.inside_pan=False（蓝包装），未满足完整入箕判据。这些是基准检查器字段，不是 agent 自定义几何检查。"
+     },
+     {
+      "t": null,
+      "log_step": 44,
+      "text_zh": "推断：log step 44 的基准 Session 返回终止步17386、保存动作17400行，与 reward.sim_steps=17386、n_actions=17400、deterministic=1一致；log step 47 无返回或保存记录，因此将 log step 44 识别为本次评分交付。据该次基准终止结果判定本节点终值为假；没有本次视频，不能确定历史区间。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "机器人脚本用共同推扫动作处理六件碎屑（log step 43、44）；缺少本次视频和状态接触对，不能核实右手扫帚及左手簸箕对蓝包装的实际作用方式。"
+     }
+    ]
+   },
+   "paper_ball_in_pan": {
+    "method": "log",
+    "intervals": null,
+    "final": false,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 44,
+      "text_zh": "观察：log step 44 中，基准 Session.status() 在仿真 8.693 s 返回 done=True、failed=True、in_pan=0；Session.features() 的 debris_4.inside_pan=False（纸球），未满足完整入箕判据。这些是基准检查器字段，不是 agent 自定义几何检查。"
+     },
+     {
+      "t": null,
+      "log_step": 44,
+      "text_zh": "推断：log step 44 的基准 Session 返回终止步17386、保存动作17400行，与 reward.sim_steps=17386、n_actions=17400、deterministic=1一致；log step 47 无返回或保存记录，因此将 log step 44 识别为本次评分交付。据该次基准终止结果判定本节点终值为假；没有本次视频，不能确定历史区间。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "机器人脚本用共同推扫动作处理六件碎屑（log step 43、44）；缺少本次视频和状态接触对，不能核实右手扫帚及左手簸箕对纸球的实际作用方式。"
+     }
+    ]
+   },
+   "yellow_cap_in_pan": {
+    "method": "log",
+    "intervals": null,
+    "final": false,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 44,
+      "text_zh": "观察：log step 44 中，基准 Session.status() 在仿真 8.693 s 返回 done=True、failed=True、in_pan=0；Session.features() 的 debris_5.inside_pan=False（黄瓶盖），未满足完整入箕判据。这些是基准检查器字段，不是 agent 自定义几何检查。"
+     },
+     {
+      "t": null,
+      "log_step": 44,
+      "text_zh": "推断：log step 44 的基准 Session 返回终止步17386、保存动作17400行，与 reward.sim_steps=17386、n_actions=17400、deterministic=1一致；log step 47 无返回或保存记录，因此将 log step 44 识别为本次评分交付。据该次基准终止结果判定本节点终值为假；没有本次视频，不能确定历史区间。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "机器人脚本用共同推扫动作处理六件碎屑（log step 43、44）；缺少本次视频和状态接触对，不能核实右手扫帚及左手簸箕对黄瓶盖的实际作用方式。"
+     }
+    ]
+   },
+   "pan_raised": {
+    "method": "log",
+    "intervals": null,
+    "final": false,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 44,
+      "text_zh": "观察：log step 44 的基准 Session.features() 终态字段 pan_lift_m=0.022606951667184447，pan_position 的世界 z=-0.07739304833281556 m，低于节点要求的抬高0.05 m；脚本在终止后打印这些基准字段并保存17400行动作。"
+     },
+     {
+      "t": null,
+      "log_step": 44,
+      "text_zh": "推断：log step 44 的基准 Session 返回终止步17386、保存动作17400行，与 reward.sim_steps=17386、n_actions=17400、deterministic=1一致；log step 47 无返回或保存记录，因此将 log step 44 识别为本次评分交付。据对应交付的基准终态判定抬箕节点为假，不从其他会话测试取值。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "log step 43、44 的脚本包含左臂抬动及前移簸箕；缺少本次视频与状态接触对，不能核实左手对簸箕柄的实际抓握接触。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "pan_lift_m",
+   "unit": "m",
+   "threshold": 0.05,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "linear_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "angular_speed_rad_s",
+   "unit": "rad/s",
+   "threshold": 0.8,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "terminal_run",
+   "unit": "physics_steps",
+   "threshold": 2000,
+   "direction": "higher_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "red_cap_in_pan",
+   "t": null,
+   "log_step": 44,
+   "where": "操作中",
+   "evidence_zh": "推断：log step 43 的脚本将仿真 4.0–9.0 s 设为扫帚前推动作；对应保存交付的 log step 44 中，基准刷毛历史在仿真 5.6 s 首先成为010000，确认先处理红瓶盖，随后共同推扫其他碎屑，但红瓶盖终态 inside_pan=False、in_pan=0。在仿真 8.693 s，基准以 tool_pressed_by_other_robot_link 永久判负，未完成或修复该扫入步骤；以最早有基准接触记录且未完成的红瓶盖节点锚定共同推扫断点。该次保存17400行、终止17386步均与评分器记录一致；没有视频时刻，t留空。"
+  },
+  "phenomena": [
+   {
+    "cat": "约束",
+    "item": "违反约束",
+    "t": null,
+    "log_step": 44,
+    "evidence_zh": "观察：log step 44 的基准 Session.status() 在仿真 8.693 s（17386物理步）返回 failed=True、reason=tool_pressed_by_other_robot_link；任务文本将该条件定义为工具受到所属手以外的机器人部位超过0.01 N的压碰。该段终止和保存行数与本次 reward 的 sim_steps、n_actions一致。基准记录直接支持禁用接触违规，但未报告具体工具、机器人部位或阻挡形态。"
+   }
+  ],
+  "root_causes": [
+   {
+    "cat": "控制",
+    "item": "控制",
+    "source_zh": "日志",
+    "evidence_zh": "推断：log step 43、44 的机器人脚本目标仍是用扫帚将碎屑推入簸箕，执行该动作时却由基准返回 tool_pressed_by_other_robot_link 而终止；会话测试 log step 36、37 也返回相同判负，改为先前移簸箕的保存版本 log step 44 仍未消除这类受禁接触。由此将根因保留在控制层级；现有证据不足以区分接触设计、轨迹或逆解、反馈控制，也不足以指定某个参数错误。"
+   }
+  ],
+  "gap_zh": "本次 inputs/video.json 为 kind=none，inputs/ 无本次 replay.mp4 或 states.jsonl；所有视频区间及视频锚点均留空，contact_actual 均为未知。节点终值来自与评分器动作行数及终止步数吻合的 log step 44 基准 Session.status()/features() 字段；此交付对应关系属于推断，不能独立复核整段运动。日志已有输出截断，且 log step 47 未返回，无法核实具体违规工具、机器人部位、接触力大小及精细控制原因；成功参考视频不作为本次轨迹证据。reward 未独立报告 DAG 的四个原生指标，native_metrics.value 留空。",
+  "unmet": {
+   "orange_package_in_pan": false,
+   "red_cap_in_pan": false,
+   "wood_chip_in_pan": false,
+   "blue_package_in_pan": false,
+   "paper_ball_in_pan": false,
+   "yellow_cap_in_pan": false,
+   "pan_raised": false
+  }
+ },
+ "summary_zh": "推断：log step 44 保存的17400行轨迹及仿真终止步17386与评分器记录一致；该次基准终态显示六件碎屑均未完整入箕，簸箕仅抬高约22.6毫米，七个节点终值均为假，会话其他测试的进展不计入本次交付。共同推扫中最早有基准刷毛接触记录的红瓶盖步骤未完成，基准在仿真 8.693 s 以工具受非所属手机器人部位压碰而判负，根因仅支持到控制层级。本次无视频及状态日志，无法给出视频区间或实际接触细节。",
+ "conf": 6,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "本次无视频/状态日志；七个节点终值为假、交付识别（log step 44：17400行、17386步，与reward一致）、违反约束现象均有基准Session字段支撑，断点与根因保守且标明为推断。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": "2026-10-30T22:30:10Z",
+  "review_prompt": "prompts/04_trajectory_review.md",
+  "review_prompt_sha256": "9afbc4ac19b1e1cf03732c8961ec46a618d682ee229343f66e3316794b14f9ba",
+  "reconcile_prompt": "prompts/05_reconcile.md",
+  "reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
+  "calls": [
+   {
+    "stage": "traj",
+    "round": 0,
+    "ok": true,
+    "space": "work/runs/dexhand-sweep-dustpan-i00-privileged--npghzqz/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-sweep-dustpan-i00-privileged--npghzqz/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "300b9cddbf57059c914f55d8e065a2e204e764437457bab08764ded6eb76edc4",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/sweep_dustpan/trajectories/dexhand-sweep-dustpan-i00-privileged--uvbbhft.json

@@ -0,0 +1,288 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-sweep-dustpan-i00-privileged--uvbbhft",
+ "taskKey": "dexhand/sweep_dustpan",
+ "dag_sha256": "4d82017ee30ab51b0c47be7c7dca8760f7e8fdbb7e1109052499805be552add8",
+ "outcome": {
+  "success": 0,
+  "reward": 0.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 1,
+   "hand_wuji": 0,
+   "infra_error": 0,
+   "n_actions": 13500,
+   "sim_steps": 13500,
+   "stages_completed": 1,
+   "success": 0,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 2.306,
+   "peak_processes": 9,
+   "peak_disk_gb": 0.256,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "none",
+  "denominator": 7,
+  "base": null,
+  "peak": null,
+  "final": null,
+  "nodes": {
+   "orange_package_in_pan": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 9,
+      "text_zh": "观察：inputs/video.json 标明本次无视频，亦无状态日志；评分器只报告 stages_completed=1、success=0，没有逐件入箕字段。log step 9 的接口代码定义 stage=in_pan+int(success)，log step 11 的重放代码输出终态 stage。推断：评分记录支持一件碎屑入箕的总体进展，但不能确定是否为橙包装，故本节点终值未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：log step 227 的交付生成脚本安排右臂持扫帚扫向碎屑、左臂随后抬簸箕；log step 236 确认交付含13500行控制。推断：此节点在整批扫入的尝试范围内，但无本次画面或接触日志，无法确认扫帚刷毛或簸箕与该碎屑实际如何接触。"
+     }
+    ]
+   },
+   "red_cap_in_pan": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 9,
+      "text_zh": "观察：inputs/video.json 标明本次无视频，亦无状态日志；评分器只报告 stages_completed=1、success=0，没有逐件入箕字段。log step 9 的接口代码定义 stage=in_pan+int(success)，log step 11 的重放代码输出终态 stage。推断：评分记录支持一件碎屑入箕的总体进展，但不能确定是否为红瓶盖，故本节点终值未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：log step 227 的交付生成脚本安排右臂持扫帚扫向碎屑、左臂随后抬簸箕；log step 236 确认交付含13500行控制。推断：此节点在整批扫入的尝试范围内，但无本次画面或接触日志，无法确认扫帚刷毛或簸箕与该碎屑实际如何接触。"
+     }
+    ]
+   },
+   "wood_chip_in_pan": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 9,
+      "text_zh": "观察：inputs/video.json 标明本次无视频，亦无状态日志；评分器只报告 stages_completed=1、success=0，没有逐件入箕字段。log step 9 的接口代码定义 stage=in_pan+int(success)，log step 11 的重放代码输出终态 stage。推断：评分记录支持一件碎屑入箕的总体进展，但不能确定是否为木屑，故本节点终值未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：log step 227 的交付生成脚本安排右臂持扫帚扫向碎屑、左臂随后抬簸箕；log step 236 确认交付含13500行控制。推断：此节点在整批扫入的尝试范围内，但无本次画面或接触日志，无法确认扫帚刷毛或簸箕与该碎屑实际如何接触。"
+     }
+    ]
+   },
+   "blue_package_in_pan": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 9,
+      "text_zh": "观察：inputs/video.json 标明本次无视频，亦无状态日志；评分器只报告 stages_completed=1、success=0，没有逐件入箕字段。log step 9 的接口代码定义 stage=in_pan+int(success)，log step 11 的重放代码输出终态 stage。推断：评分记录支持一件碎屑入箕的总体进展，但不能确定是否为蓝包装，故本节点终值未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：log step 227 的交付生成脚本安排右臂持扫帚扫向碎屑、左臂随后抬簸箕；log step 236 确认交付含13500行控制。推断：此节点在整批扫入的尝试范围内，但无本次画面或接触日志，无法确认扫帚���毛或簸箕与该碎屑实际如何接触。"
+     }
+    ]
+   },
+   "paper_ball_in_pan": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 9,
+      "text_zh": "观察：inputs/video.json 标明本次无视频，亦无状态日志；评分器只报告 stages_completed=1、success=0，没有逐件入箕字段。log step 9 的接口代码定义 stage=in_pan+int(success)，log step 11 的重放代码输出终态 stage。推断：评分记录支持一件碎屑入箕的总体进展，但不能确定是否为纸球，故本节点终值未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：log step 227 的交付生成脚本安排右臂持扫帚扫向碎屑、左臂随后抬簸箕；log step 236 确认交付含13500行控制。推断：此节点在整批扫入的尝试范围内，但无本次画面或接触日志，无法确认扫帚刷毛或簸箕与该碎屑实际如何接触。"
+     }
+    ]
+   },
+   "yellow_cap_in_pan": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 9,
+      "text_zh": "观察：inputs/video.json 标明本次无视频，亦无状态日志；评分器只报告 stages_completed=1、success=0，没有逐件入箕字段。log step 9 的接口代码定义 stage=in_pan+int(success)，log step 11 的重放代码输出终态 stage。推断：评分记录支持一件碎屑入箕的总体进展，但不能确定是否为黄瓶盖，故本节点终值未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：log step 227 的交付生成脚本安排右臂持扫帚扫向碎屑、左臂随后抬簸箕；log step 236 确认交付含13500行控制。推断：此节点在整批扫入的尝试范围内，但无本次画面或接触日志，无法确认扫帚刷毛或簸箕与该碎屑实际如何接触。"
+     }
+    ]
+   },
+   "pan_raised": {
+    "method": "unknown",
+    "intervals": null,
+    "final": null,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": 227,
+      "text_zh": "观察：log step 227 的交付生成脚本安排左臂抬高目标0.12 m；log step 236 确认交付含13500行控制，评分器也报告 n_actions=sim_steps=13500。评分器没有 pan_lift_m，log step 231、237 的重放调用只返回后台运行提示，没有终态高度或服务判定。推断：抬箕控制已被安排，但控制目标不等于实际簸箕高度，抬起节点终值未知。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "观察：log step 227 的交付生成脚本安排左臂抬簸箕并同时撤起右臂扫帚。推断：抬箕被尝试，但没有本次画面或手柄接触记录，无法核实左手与簸箕柄的实际接触方式。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "pan_lift_m",
+   "unit": "m",
+   "threshold": 0.05,
+   "direction": "higher_better",
+   "value": null
+  },
+  {
+   "name": "linear_speed_m_s",
+   "unit": "m/s",
+   "threshold": 0.03,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "angular_speed_rad_s",
+   "unit": "rad/s",
+   "threshold": 0.8,
+   "direction": "lower_better",
+   "value": null
+  },
+  {
+   "name": "terminal_run",
+   "unit": "physics_steps",
+   "threshold": 2000,
+   "direction": "higher_better",
+   "value": null
+  }
+ ],
+ "failure": {
+  "breakpoint": {
+   "ref": "unknown",
+   "t": null,
+   "log_step": null,
+   "where": "未知",
+   "evidence_zh": "观察：评分器记录 success=0、deterministic=1、infra_error=0、stages_completed=1、sim_steps=13500；log step 227 的最终脚本包含整批扫入、抬箕和保持，log step 235、236 确认轨迹文件存在且含13500行控制。log step 231、237 没有保留重放结果明细。推断：已有被评分的交付，不能按未交付或首步前处理；缺少逐件状态及失败原因，无法确定最早未修复的是哪件碎屑入箕、抬箕还是共同终态验收。"
+  },
+  "unmet": {
+   "orange_package_in_pan": null,
+   "red_cap_in_pan": null,
+   "wood_chip_in_pan": null,
+   "blue_package_in_pan": null,
+   "paper_ball_in_pan": null,
+   "yellow_cap_in_pan": null,
+   "pan_raised": null
+  },
+  "phenomena": [],
+  "root_causes": [],
+  "gap_zh": "本次无视频（inputs/video.json 为 none），无状态日志。评分器的 stages_completed=1 只有总量，未报告物件身份、簸箕高度、碎屑速度、terminal_run 或失败原因；这些量的原生指标均保留 null。log step 231、237 只有重放启动提示，log step 241 关于验证和哈希一致的说法是 agent 自述，不能补成重放判定或物理状态。会话中的碰撞、扫入和抬箕测试没有可绑定本次评分轨迹的结果，不能据测试名称判定现象或根因；因此两层留空���已查看成功参考视频的粗采样，但它属于另一运行，不能证明本次节点或接触。"
+ },
+ "summary_zh": "评分器判定失败，记录13500个仿真步、stages_completed=1且重放确定性成立；结合 log step 9、11 的接口与重放代码，可推断有一件碎屑入箕的总体进展，但无法识别物件，七个具体节点的终值均未知。最终交付脚本包含扫入、抬箕和保持，现有证据无法定位最早未修复的失败步骤，也不足以归因。本次没有视频或状态日志，成功参考视频仅用于理解任务视角，不作为本次状态证据。",
+ "conf": 3,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 6,
+    "summary_zh": "本次无视频、无状态日志；评分器只给出 stages_completed=1（log step 9 的 status() 定义 stage=in_pan+int(success)，success=0 即终态恰有一件碎屑入箕），无法识别物件，七个节点 unknown、断点 unknown、现象与根因留空并写明 gap_zh，均符合 03 的 N9 规则；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": "2026-10-30T22:29:57Z",
+  "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/dexhand-sweep-dustpan-i00-privileged--uvbbhft/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-sweep-dustpan-i00-privileged--uvbbhft/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "fc8c982275d41068328a7824abcc50a60efb41c3f51e88d9943d3635e267208e",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}

phase2/dexhand/sweep_dustpan/trajectories/dexhand-sweep-dustpan-i00-privileged--vup9yjl.json

@@ -0,0 +1,303 @@
+{
+ "schema": "efe-trajectory-labels/1",
+ "taxonomy": "v2",
+ "runId": "dexhand-sweep-dustpan-i00-privileged--vup9yjl",
+ "taskKey": "dexhand/sweep_dustpan",
+ "dag_sha256": "4d82017ee30ab51b0c47be7c7dca8760f7e8fdbb7e1109052499805be552add8",
+ "outcome": {
+  "success": 1,
+  "reward": 1.0,
+  "source": "catalog",
+  "raw": {
+   "deterministic": 1,
+   "hand_wuji": 0,
+   "infra_error": 0,
+   "n_actions": 15060,
+   "sim_steps": 15060,
+   "stages_completed": 7,
+   "success": 1,
+   "resource_enforced": 0,
+   "limits_violation_observed": 0,
+   "resource_monitor_missing": 0,
+   "peak_gpu_memory_gb": 0.0,
+   "peak_memory_gb": 2.48,
+   "peak_processes": 7,
+   "peak_disk_gb": 0.249,
+   "resource_emergency_kills": 0,
+   "resource_oom_kills": 0,
+   "resource_host_oom_kills": 0,
+   "resource_service_oom_kills": 0
+  }
+ },
+ "video": {
+  "kind": "none",
+  "url": null,
+  "duration_s": null,
+  "fps": null,
+  "frames": null,
+  "sha256": null
+ },
+ "state": null,
+ "progress": {
+  "method": "log",
+  "denominator": 7,
+  "base": null,
+  "peak": null,
+  "final": 7,
+  "nodes": {
+   "orange_package_in_pan": {
+    "method": "log",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器 reward.success=1、stages_completed=7、deterministic=1、n_actions=sim_steps=15060，infra_error=0；本次无视频且无状态日志。"
+     },
+     {
+      "t": null,
+      "log_step": 62,
+      "text_zh": "推断：log step 55 保存的最终轨迹哈希为 d73a8cda1588c6f1b917b0ae09103db2a01706631bbf811a99809b8c555295b5；log step 62 中基准 replay.py 对该文件的两次独立重放均报告 success=true、stage=7、steps=15060、bad_trajectory=0，且 state_sha 相同，与最终评分一致。按 DAG 的联合成功判据，橙包装入簸箕在验收时成立；无本次视频，无法确定首次达成、途中失去或恢复的时间，intervals 留 null。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "橙包装入簸箕已由成功验收确认，但本次无视频或状态接触对，无法直接核实作用端及接触方式；不以生成脚本、agent 自述或另一运行的参考视频代替实际接触证据。"
+     }
+    ]
+   },
+   "red_cap_in_pan": {
+    "method": "log",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器 reward.success=1、stages_completed=7、deterministic=1、n_actions=sim_steps=15060，infra_error=0；本次无视频且无状态日志。"
+     },
+     {
+      "t": null,
+      "log_step": 62,
+      "text_zh": "推断：log step 55 保存的最终轨迹哈希为 d73a8cda1588c6f1b917b0ae09103db2a01706631bbf811a99809b8c555295b5；log step 62 中基准 replay.py 对该文件的两次独立重放均报告 success=true、stage=7、steps=15060、bad_trajectory=0，且 state_sha 相同，与最终评分一致。按 DAG 的联合成功判据，红瓶盖入簸箕在验收时成立；无本次视频，无法确定首次达成、途中失去或恢复的时间，intervals 留 null。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "红瓶盖入簸箕已由成功验收确认，但本次无视频或状态接触对，无法直接核实作用端及接触方式；不以生成脚本、agent 自述或另一运行的参考视频代替实际接触证据。"
+     }
+    ]
+   },
+   "wood_chip_in_pan": {
+    "method": "log",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器 reward.success=1、stages_completed=7、deterministic=1、n_actions=sim_steps=15060，infra_error=0；本次无视频且无状态日志。"
+     },
+     {
+      "t": null,
+      "log_step": 62,
+      "text_zh": "推断：log step 55 保存的最终轨迹哈希为 d73a8cda1588c6f1b917b0ae09103db2a01706631bbf811a99809b8c555295b5；log step 62 中基准 replay.py 对该文件的两次独立重放均报告 success=true、stage=7、steps=15060、bad_trajectory=0，且 state_sha 相同，与最终评分一致。按 DAG 的联合成功判据，木屑入簸箕在验收时成立；无本次视频，无法确定首次达成、途中失去或恢复的时间，intervals 留 null。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "木屑入簸箕已由成功验收确认，但本次无视频或状态接触对，无法直接核实作用端及接触方式；不以生成脚本、agent 自述或另一运行的参考视频代替实际接触证据。"
+     }
+    ]
+   },
+   "blue_package_in_pan": {
+    "method": "log",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器 reward.success=1、stages_completed=7、deterministic=1、n_actions=sim_steps=15060，infra_error=0；本次无视频且无状态日志。"
+     },
+     {
+      "t": null,
+      "log_step": 62,
+      "text_zh": "推断：log step 55 保存的最终轨迹哈希为 d73a8cda1588c6f1b917b0ae09103db2a01706631bbf811a99809b8c555295b5；log step 62 中基准 replay.py 对该文件的两次独立重放均报告 success=true、stage=7、steps=15060、bad_trajectory=0，且 state_sha 相同，与最终评分一致。按 DAG 的联合成功判据，蓝包装入簸箕在验收时成立；无本次视频，无法确定首次达成、途中失去或恢复的时间，intervals 留 null。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "蓝包装入簸箕已由成功验收确认，但本次无视频或状态接触对，无法直接核实作用端及接触方式；不以生成脚本、agent 自述或另一运行的参考视频代替实际接触证据。"
+     }
+    ]
+   },
+   "paper_ball_in_pan": {
+    "method": "log",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器 reward.success=1、stages_completed=7、deterministic=1、n_actions=sim_steps=15060，infra_error=0；本次无视频且无状态日志。"
+     },
+     {
+      "t": null,
+      "log_step": 62,
+      "text_zh": "推断：log step 55 保存的最终轨迹哈希为 d73a8cda1588c6f1b917b0ae09103db2a01706631bbf811a99809b8c555295b5；log step 62 中基准 replay.py 对该文件的两次独立重放均报告 success=true、stage=7、steps=15060、bad_trajectory=0，且 state_sha 相同，与最终评分一致。按 DAG 的联合成功判据，纸球入簸箕在验收时成立；无本次视频，无法确定首次达成、途中失去或恢复的时间，intervals 留 null。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "纸球入簸箕已由成功验收确认，但本次无视频或状态接触对，无法直接核实作用端及接触方式；不以生成脚本、agent 自述或另一运行的参考视频代替实际接触证据。"
+     }
+    ]
+   },
+   "yellow_cap_in_pan": {
+    "method": "log",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器 reward.success=1、stages_completed=7、deterministic=1、n_actions=sim_steps=15060，infra_error=0；本次无视频且无状态日志。"
+     },
+     {
+      "t": null,
+      "log_step": 62,
+      "text_zh": "推断：log step 55 保存的最终轨迹哈希为 d73a8cda1588c6f1b917b0ae09103db2a01706631bbf811a99809b8c555295b5；log step 62 中基准 replay.py 对该文件的两次独立重放均报告 success=true、stage=7、steps=15060、bad_trajectory=0，且 state_sha 相同，与最终评分一致。按 DAG 的联合成功判据，黄瓶盖入簸箕在验收时成立；无本次视频，无法确定首次达成、途中失去或恢复的时间，intervals 留 null。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "黄瓶盖入簸箕已由成功验收确认，但本次无视频或状态接触对，无法直接核实作用端及接触方式；不以生成脚本、agent 自述或另一运行的参考视频代替实际接触证据。"
+     }
+    ]
+   },
+   "pan_raised": {
+    "method": "log",
+    "intervals": null,
+    "final": true,
+    "evidence": [
+     {
+      "t": null,
+      "log_step": null,
+      "text_zh": "观察：inputs/run.json 的评分器 reward.success=1、stages_completed=7、deterministic=1、n_actions=sim_steps=15060，infra_error=0；本次无视频且无状态日志。"
+     },
+     {
+      "t": null,
+      "log_step": 62,
+      "text_zh": "推断：log step 55 保存的最终轨迹哈希为 d73a8cda1588c6f1b917b0ae09103db2a01706631bbf811a99809b8c555295b5；log step 62 中基准 replay.py 对该文件的两次独立重放均报告 success=true、stage=7、steps=15060、bad_trajectory=0，且 state_sha 相同，与最终评分一致。按 DAG 的联合成功判据，抬起簸箕在验收时成立；无本次视频，无法确定首次达成、途中失去或恢复的时间，intervals 留 null。"
+     }
+    ],
+    "contact_actual": [
+     {
+      "who": "未知",
+      "how": "未知",
+      "note_zh": "抬起簸箕已由成功验收确认，但本次无视频或状态接触对，无法直接核实作用端及接触方式；不以生成脚本、agent 自述或另一运行的参考视频代替实际接触证据。"
+     }
+    ]
+   }
+  },
+  "constraints": {}
+ },
+ "native_metrics": [
+  {
+   "name": "pan_lift_m",
+   "value": null,
+   "unit": "m",
+   "threshold": 0.05,
+   "direction": "higher_better"
+  },
+  {
+   "name": "linear_speed_m_s",
+   "value": null,
+   "unit": "m/s",
+   "threshold": 0.03,
+   "direction": "lower_better"
+  },
+  {
+   "name": "angular_speed_rad_s",
+   "value": null,
+   "unit": "rad/s",
+   "threshold": 0.8,
+   "direction": "lower_better"
+  },
+  {
+   "name": "terminal_run",
+   "value": null,
+   "unit": "physics_steps",
+   "threshold": 2000,
+   "direction": "higher_better"
+  }
+ ],
+ "failure": null,
+ "summary_zh": "观察：评分器报告 success=1、stages_completed=7、deterministic=1；最终15060步交付轨迹在 log step 62 的两次独立重放中均成功且终态哈希相同，据联合验收可确定六件碎屑入簸箕及抬箕共7/7节点终值为真，无失败断点。推断：该成功说明共同装载、刷毛接触历史与抬箕保持条件通过验收，但本次无视频、无状态日志，所有节点时间区间留 null、实际接触标未知；已查看的 oracle 画面属于另一成功运行，不能填补本次证据。观察：本次 reward 未报告四项原生指标数值，均留 null；log step 49 的详细诊断属于较早14454步版本，log step 65 的抬高数值属于 agent 自述，均不转作最终评分指标。",
+ "conf": 7,
+ "review": {
+  "status": "agreed",
+  "rounds": 1,
+  "reviewers": [
+   {
+    "engine": "claude",
+    "model": "claude-sonnet-5-5",
+    "round": 1,
+    "verdict": "agree",
+    "conf": 8,
+    "summary_zh": "本次无视频、无状态日志；评分器 success=1、stages_completed=7、n_actions=15060，七个目标节点终值为真、failure 为 null 均有分级结果支撑；log step 55/62 的哈希与重放结果与标注一致；时间区间留 null、contact_actual 为未知均合理。",
+    "n_issues": 0,
+    "n_major": 0
+   }
+  ],
+  "open_zh": []
+ },
+ "provenance": {
+  "prompt": "prompts/03_trajectory.md",
+  "prompt_sha256": "2385b35930667330fc4ea181a15ea0818b49cfeae0df1700b8b08c988e72ad8e",
+  "engine": "codex",
+  "model": "gpt-6.1-sol",
+  "effort": "high",
+  "created": "2026-10-30T22:28:06Z",
+  "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/dexhand-sweep-dustpan-i00-privileged--vup9yjl/03_label"
+   },
+   {
+    "stage": "traj_review",
+    "round": 1,
+    "ok": true,
+    "space": "work/runs/dexhand-sweep-dustpan-i00-privileged--vup9yjl/04_review_r1"
+   }
+  ],
+  "review_engine": "claude",
+  "review_model": "claude-sonnet-5-5",
+  "inputs": {
+   "run_json_sha256": "74bdd9f3d0e393bcbd1e7955c4b3ca9c4e936aa1a496025b98efef5e98ddd6a2",
+   "video_sha256": null,
+   "states_sha256": null,
+   "video_choice": "HF replay.mp4 status 404; no local video"
+  }
+ }
+}