Add GPT-6 Astra 1× + ICL native evaluation logs (#2)

- Add GPT-6 Astra 1× + ICL native evaluation logs (7562e4f8d4e3cec1f3681922c68a8b72c735d205)
- Record historical missing-log and rerun audit hashes (cf97cb8a32fe699dc29c9aa6a822e58aae1500dc)


Co-authored-by: Quanzhou Li <licandow@users.noreply.huggingface.co>

gpt6_astra_1x_icl/README.md

@@ -0,0 +1,40 @@
+# GPT-6 Astra — 1× + ICL
+
+This entry reports the **Full / unseen** Bench2Dex channel for GPT-6 Astra with
+frozen per-task in-context preparation. The policy uses RGB camera observations,
+proprioception, and a task description; the model runs with max reasoning effort.
+Task-related preparation is frozen before test episodes, and each test episode
+uses an isolated model session. The evaluation uses the upstream stable-success
+and latched-stage metrics at 60 Hz physics and 20 Hz control.
+
+[Technical report](https://www.sensenova.ai/has-gpt-6-astra-achieved-embodied-intelligence)
+· [SenseNova-RoboRSI project](https://github.com/OpenSenseNova/SenseNova-RoboRSI)
+
+## Result and coverage
+
+- **Stable success rate:** 68/260 = **26.15%**.
+- **Latched stage completion rate:** **48.77%**, averaged equally over 26 tasks.
+- **Coverage:** 26 tasks × 10 Full-channel episodes = 260 episodes. The reference
+  protocol uses 50 episodes per task and channel.
+- **Efficiency:** not reported because the retained aggregate evidence does not
+  establish the successful-step/expert-reference inputs for every selected episode.
+- **Other channels:** None, Equi., and Inv. were not evaluated in this cohort;
+  the four-channel mean is therefore not reported.
+
+## Evaluation protocol
+
+The selected indices 5–14 form a 260-episode Full-channel cohort. Seventeen damaged outcomes use same-index certified replacements. Task 61 episode 8 was additionally rerun at its original index and seed because its original native evaluator log was lost. The rerun's first valid outcome was selected regardless of success; the original recorded success and 100% LSCR remain historical evidence, not the selected score. The cohort is an amended reporting selection.
+
+The official 1× task budget is `ceil(1.5 × expert_time_step / 3)` policy steps,
+with three physics steps per action row. The JSON gives each task's actual 1×
+budget, the 1× allowance, seeds, and episode indices.
+The two entries must be interpreted as separate evaluation settings.
+
+The 1× source cohort includes provenance-selected damage replacements; a full
+archive certification was not completed. Historical Task 22/26 extensions and
+the partly unresolved Task 51 preparation provenance are disclosed limitations.
+
+The linked Hugging Face dataset entry includes the selected native Bench2Dex
+evaluator rows, sanitized single-episode source logs and summaries under
+`source_runs/`, and a SHA-256 provenance manifest. The per-task summaries are
+recomputed with the upstream Bench2Dex summary writer from the selected rows.

gpt6_astra_1x_icl/leaderboard_gpt6_astra_1x_icl.json

@@ -0,0 +1,1855 @@
+{
+  "model": {
+    "id": "gpt6_astra_1x_icl",
+    "label": "GPT-6 Astra",
+    "variantLabel": "1× + ICL",
+    "family": "GPT-6 Astra",
+    "modelVersion": "gpt-6-astra",
+    "checkpoint": null,
+    "inputs": "RGB vision + proprioception + task description",
+    "protocolId": "standard",
+    "displayOrder": 100,
+    "trainingSetting": "Frozen per-task in-context preparation; no weight updates",
+    "resultVersion": "Amended 260, 1× budget",
+    "updatedAt": null,
+    "evaluationSetting": "task_icl_v3",
+    "sourceLabel": "Has GPT-6-Astra Achieved Embodied Intelligence?",
+    "sourceUrl": "https://www.sensenova.ai/has-gpt-6-astra-achieved-embodied-intelligence",
+    "resultDetailsUrl": null,
+    "contributor": "SenseNova-RoboRSI",
+    "contributorUrl": "https://github.com/OpenSenseNova/SenseNova-RoboRSI",
+    "type": "Agent",
+    "modelUrl": "https://www.sensenova.ai/has-gpt-6-astra-achieved-embodied-intelligence",
+    "logoPath": "leaderboard/gpt6_astra_1x_icl/logo.png",
+    "logoAlt": "SenseNova logo",
+    "weightsUrl": null,
+    "codeUrl": "https://github.com/OpenSenseNova/SenseNova-RoboRSI",
+    "externalReport": true,
+    "episodesPerChannel": 10,
+    "stepBudgetMultiplier": 1
+  },
+  "overall": {
+    "taskCount": 26,
+    "scope": "task-macro",
+    "status": "not-reported",
+    "sr": null,
+    "lscr": null,
+    "efficiency": null,
+    "efficiencyStatus": "not-reported",
+    "efficiencyTaskCount": null,
+    "channelSr": {
+      "None": null,
+      "Equi.": null,
+      "Inv.": null,
+      "Full": 26.153846153846153
+    },
+    "channelLscr": {
+      "None": null,
+      "Equi.": null,
+      "Inv.": null,
+      "Full": 48.76923076923077
+    },
+    "channelTaskCounts": {
+      "None": 0,
+      "Equi.": 0,
+      "Inv.": 0,
+      "Full": 26
+    }
+  },
+  "per_task": {
+    "12": {
+      "channels": {
+        "None": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Equi.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Inv.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Full": {
+          "status": "reported",
+          "episodes": 10,
+          "successes": 4,
+          "lscr": 85.0,
+          "source": {
+            "runId": "task_icl_v3-amended260-index05-14-task61_08-rerun",
+            "perEpisodeFile": "per_task_eval/12_screwdriver_box_and_hammer/Full/per_episode.jsonl",
+            "summaryFile": "per_task_eval/12_screwdriver_box_and_hammer/Full/summary.json"
+          }
+        }
+      },
+      "mean": {
+        "status": "not-reported",
+        "sr": null,
+        "srScope": "four-channel-mean",
+        "srEpisodes": null,
+        "lscr": null,
+        "lscrScope": "four-channel-mean",
+        "lscrEpisodes": null,
+        "efficiency": null,
+        "efficiencyStatus": "not-reported"
+      }
+    },
+    "21": {
+      "channels": {
+        "None": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Equi.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Inv.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Full": {
+          "status": "reported",
+          "episodes": 10,
+          "successes": 2,
+          "lscr": 23.333333333333332,
+          "source": {
+            "runId": "task_icl_v3-amended260-index05-14-task61_08-rerun",
+            "perEpisodeFile": "per_task_eval/21_condiment_box_loading/Full/per_episode.jsonl",
+            "summaryFile": "per_task_eval/21_condiment_box_loading/Full/summary.json"
+          }
+        }
+      },
+      "mean": {
+        "status": "not-reported",
+        "sr": null,
+        "srScope": "four-channel-mean",
+        "srEpisodes": null,
+        "lscr": null,
+        "lscrScope": "four-channel-mean",
+        "lscrEpisodes": null,
+        "efficiency": null,
+        "efficiencyStatus": "not-reported"
+      }
+    },
+    "22": {
+      "channels": {
+        "None": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Equi.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Inv.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Full": {
+          "status": "reported",
+          "episodes": 10,
+          "successes": 5,
+          "lscr": 72.5,
+          "source": {
+            "runId": "task_icl_v3-amended260-index05-14-task61_08-rerun",
+            "perEpisodeFile": "per_task_eval/22_tool_box_loading/Full/per_episode.jsonl",
+            "summaryFile": "per_task_eval/22_tool_box_loading/Full/summary.json"
+          }
+        }
+      },
+      "mean": {
+        "status": "not-reported",
+        "sr": null,
+        "srScope": "four-channel-mean",
+        "srEpisodes": null,
+        "lscr": null,
+        "lscrScope": "four-channel-mean",
+        "lscrEpisodes": null,
+        "efficiency": null,
+        "efficiencyStatus": "not-reported"
+      }
+    },
+    "24": {
+      "channels": {
+        "None": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Equi.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Inv.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Full": {
+          "status": "reported",
+          "episodes": 10,
+          "successes": 0,
+          "lscr": 25.0,
+          "source": {
+            "runId": "task_icl_v3-amended260-index05-14-task61_08-rerun",
+            "perEpisodeFile": "per_task_eval/24_stationery_category_sorting/Full/per_episode.jsonl",
+            "summaryFile": "per_task_eval/24_stationery_category_sorting/Full/summary.json"
+          }
+        }
+      },
+      "mean": {
+        "status": "not-reported",
+        "sr": null,
+        "srScope": "four-channel-mean",
+        "srEpisodes": null,
+        "lscr": null,
+        "lscrScope": "four-channel-mean",
+        "lscrEpisodes": null,
+        "efficiency": null,
+        "efficiencyStatus": "not-reported"
+      }
+    },
+    "26": {
+      "channels": {
+        "None": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Equi.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Inv.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Full": {
+          "status": "reported",
+          "episodes": 10,
+          "successes": 0,
+          "lscr": 57.49999999999999,
+          "source": {
+            "runId": "task_icl_v3-amended260-index05-14-task61_08-rerun",
+            "perEpisodeFile": "per_task_eval/26_canned_food_tray_line_arrangement/Full/per_episode.jsonl",
+            "summaryFile": "per_task_eval/26_canned_food_tray_line_arrangement/Full/summary.json"
+          }
+        }
+      },
+      "mean": {
+        "status": "not-reported",
+        "sr": null,
+        "srScope": "four-channel-mean",
+        "srEpisodes": null,
+        "lscr": null,
+        "lscrScope": "four-channel-mean",
+        "lscrEpisodes": null,
+        "efficiency": null,
+        "efficiencyStatus": "not-reported"
+      }
+    },
+    "27": {
+      "channels": {
+        "None": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Equi.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Inv.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Full": {
+          "status": "reported",
+          "episodes": 10,
+          "successes": 10,
+          "lscr": 100.0,
+          "source": {
+            "runId": "task_icl_v3-amended260-index05-14-task61_08-rerun",
+            "perEpisodeFile": "per_task_eval/27_ball_box_loading/Full/per_episode.jsonl",
+            "summaryFile": "per_task_eval/27_ball_box_loading/Full/summary.json"
+          }
+        }
+      },
+      "mean": {
+        "status": "not-reported",
+        "sr": null,
+        "srScope": "four-channel-mean",
+        "srEpisodes": null,
+        "lscr": null,
+        "lscrScope": "four-channel-mean",
+        "lscrEpisodes": null,
+        "efficiency": null,
+        "efficiencyStatus": "not-reported"
+      }
+    },
+    "32": {
+      "channels": {
+        "None": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Equi.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Inv.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Full": {
+          "status": "reported",
+          "episodes": 10,
+          "successes": 0,
+          "lscr": 7.5,
+          "source": {
+            "runId": "task_icl_v3-amended260-index05-14-task61_08-rerun",
+            "perEpisodeFile": "per_task_eval/32_baking_tray_prep_with_tools/Full/per_episode.jsonl",
+            "summaryFile": "per_task_eval/32_baking_tray_prep_with_tools/Full/summary.json"
+          }
+        }
+      },
+      "mean": {
+        "status": "not-reported",
+        "sr": null,
+        "srScope": "four-channel-mean",
+        "srEpisodes": null,
+        "lscr": null,
+        "lscrScope": "four-channel-mean",
+        "lscrEpisodes": null,
+        "efficiency": null,
+        "efficiencyStatus": "not-reported"
+      }
+    },
+    "34": {
+      "channels": {
+        "None": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Equi.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Inv.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Full": {
+          "status": "reported",
+          "episodes": 10,
+          "successes": 3,
+          "lscr": 68.0,
+          "source": {
+            "runId": "task_icl_v3-amended260-index05-14-task61_08-rerun",
+            "perEpisodeFile": "per_task_eval/34_fridge_wine_interhand_pour/Full/per_episode.jsonl",
+            "summaryFile": "per_task_eval/34_fridge_wine_interhand_pour/Full/summary.json"
+          }
+        }
+      },
+      "mean": {
+        "status": "not-reported",
+        "sr": null,
+        "srScope": "four-channel-mean",
+        "srEpisodes": null,
+        "lscr": null,
+        "lscrScope": "four-channel-mean",
+        "lscrEpisodes": null,
+        "efficiency": null,
+        "efficiencyStatus": "not-reported"
+      }
+    },
+    "42": {
+      "channels": {
+        "None": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Equi.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Inv.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Full": {
+          "status": "reported",
+          "episodes": 10,
+          "successes": 5,
+          "lscr": 70.0,
+          "source": {
+            "runId": "task_icl_v3-amended260-index05-14-task61_08-rerun",
+            "perEpisodeFile": "per_task_eval/42_trash_disposal/Full/per_episode.jsonl",
+            "summaryFile": "per_task_eval/42_trash_disposal/Full/summary.json"
+          }
+        }
+      },
+      "mean": {
+        "status": "not-reported",
+        "sr": null,
+        "srScope": "four-channel-mean",
+        "srEpisodes": null,
+        "lscr": null,
+        "lscrScope": "four-channel-mean",
+        "lscrEpisodes": null,
+        "efficiency": null,
+        "efficiencyStatus": "not-reported"
+      }
+    },
+    "43": {
+      "channels": {
+        "None": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Equi.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Inv.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Full": {
+          "status": "reported",
+          "episodes": 10,
+          "successes": 3,
+          "lscr": 55.00000000000001,
+          "source": {
+            "runId": "task_icl_v3-amended260-index05-14-task61_08-rerun",
+            "perEpisodeFile": "per_task_eval/43_fridge_fruit_shelf_sorting/Full/per_episode.jsonl",
+            "summaryFile": "per_task_eval/43_fridge_fruit_shelf_sorting/Full/summary.json"
+          }
+        }
+      },
+      "mean": {
+        "status": "not-reported",
+        "sr": null,
+        "srScope": "four-channel-mean",
+        "srEpisodes": null,
+        "lscr": null,
+        "lscrScope": "four-channel-mean",
+        "lscrEpisodes": null,
+        "efficiency": null,
+        "efficiencyStatus": "not-reported"
+      }
+    },
+    "44": {
+      "channels": {
+        "None": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Equi.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Inv.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Full": {
+          "status": "reported",
+          "episodes": 10,
+          "successes": 4,
+          "lscr": 62.5,
+          "source": {
+            "runId": "task_icl_v3-amended260-index05-14-task61_08-rerun",
+            "perEpisodeFile": "per_task_eval/44_microwave_bowl_loading/Full/per_episode.jsonl",
+            "summaryFile": "per_task_eval/44_microwave_bowl_loading/Full/summary.json"
+          }
+        }
+      },
+      "mean": {
+        "status": "not-reported",
+        "sr": null,
+        "srScope": "four-channel-mean",
+        "srEpisodes": null,
+        "lscr": null,
+        "lscrScope": "four-channel-mean",
+        "lscrEpisodes": null,
+        "efficiency": null,
+        "efficiencyStatus": "not-reported"
+      }
+    },
+    "51": {
+      "channels": {
+        "None": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Equi.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Inv.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Full": {
+          "status": "reported",
+          "episodes": 10,
+          "successes": 6,
+          "lscr": 83.33333333333334,
+          "source": {
+            "runId": "task_icl_v3-amended260-index05-14-task61_08-rerun",
+            "perEpisodeFile": "per_task_eval/51_toilet_lid_cleaner_pour/Full/per_episode.jsonl",
+            "summaryFile": "per_task_eval/51_toilet_lid_cleaner_pour/Full/summary.json"
+          }
+        }
+      },
+      "mean": {
+        "status": "not-reported",
+        "sr": null,
+        "srScope": "four-channel-mean",
+        "srEpisodes": null,
+        "lscr": null,
+        "lscrScope": "four-channel-mean",
+        "lscrEpisodes": null,
+        "efficiency": null,
+        "efficiencyStatus": "not-reported"
+      }
+    },
+    "60": {
+      "channels": {
+        "None": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Equi.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Inv.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Full": {
+          "status": "reported",
+          "episodes": 10,
+          "successes": 0,
+          "lscr": 0.0,
+          "source": {
+            "runId": "task_icl_v3-amended260-index05-14-task61_08-rerun",
+            "perEpisodeFile": "per_task_eval/60_breadbasket_fast_food_loading/Full/per_episode.jsonl",
+            "summaryFile": "per_task_eval/60_breadbasket_fast_food_loading/Full/summary.json"
+          }
+        }
+      },
+      "mean": {
+        "status": "not-reported",
+        "sr": null,
+        "srScope": "four-channel-mean",
+        "srEpisodes": null,
+        "lscr": null,
+        "lscrScope": "four-channel-mean",
+        "lscrEpisodes": null,
+        "efficiency": null,
+        "efficiencyStatus": "not-reported"
+      }
+    },
+    "61": {
+      "channels": {
+        "None": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Equi.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Inv.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Full": {
+          "status": "reported",
+          "episodes": 10,
+          "successes": 8,
+          "lscr": 83.33333333333334,
+          "source": {
+            "runId": "task_icl_v3-amended260-index05-14-task61_08-rerun",
+            "perEpisodeFile": "per_task_eval/61_medicine_shoebox_pack/Full/per_episode.jsonl",
+            "summaryFile": "per_task_eval/61_medicine_shoebox_pack/Full/summary.json"
+          }
+        }
+      },
+      "mean": {
+        "status": "not-reported",
+        "sr": null,
+        "srScope": "four-channel-mean",
+        "srEpisodes": null,
+        "lscr": null,
+        "lscrScope": "four-channel-mean",
+        "lscrEpisodes": null,
+        "efficiency": null,
+        "efficiencyStatus": "not-reported"
+      }
+    },
+    "62": {
+      "channels": {
+        "None": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Equi.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Inv.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Full": {
+          "status": "reported",
+          "episodes": 10,
+          "successes": 1,
+          "lscr": 50.0,
+          "source": {
+            "runId": "task_icl_v3-amended260-index05-14-task61_08-rerun",
+            "perEpisodeFile": "per_task_eval/62_shoebox_accessory_pack/Full/per_episode.jsonl",
+            "summaryFile": "per_task_eval/62_shoebox_accessory_pack/Full/summary.json"
+          }
+        }
+      },
+      "mean": {
+        "status": "not-reported",
+        "sr": null,
+        "srScope": "four-channel-mean",
+        "srEpisodes": null,
+        "lscr": null,
+        "lscrScope": "four-channel-mean",
+        "lscrEpisodes": null,
+        "efficiency": null,
+        "efficiencyStatus": "not-reported"
+      }
+    },
+    "64": {
+      "channels": {
+        "None": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Equi.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Inv.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Full": {
+          "status": "reported",
+          "episodes": 10,
+          "successes": 6,
+          "lscr": 72.5,
+          "source": {
+            "runId": "task_icl_v3-amended260-index05-14-task61_08-rerun",
+            "perEpisodeFile": "per_task_eval/64_sports_ball_cup_sort/Full/per_episode.jsonl",
+            "summaryFile": "per_task_eval/64_sports_ball_cup_sort/Full/summary.json"
+          }
+        }
+      },
+      "mean": {
+        "status": "not-reported",
+        "sr": null,
+        "srScope": "four-channel-mean",
+        "srEpisodes": null,
+        "lscr": null,
+        "lscrScope": "four-channel-mean",
+        "lscrEpisodes": null,
+        "efficiency": null,
+        "efficiencyStatus": "not-reported"
+      }
+    },
+    "67": {
+      "channels": {
+        "None": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Equi.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Inv.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Full": {
+          "status": "reported",
+          "episodes": 10,
+          "successes": 1,
+          "lscr": 40.0,
+          "source": {
+            "runId": "task_icl_v3-amended260-index05-14-task61_08-rerun",
+            "perEpisodeFile": "per_task_eval/67_faucet_cup_water_fill/Full/per_episode.jsonl",
+            "summaryFile": "per_task_eval/67_faucet_cup_water_fill/Full/summary.json"
+          }
+        }
+      },
+      "mean": {
+        "status": "not-reported",
+        "sr": null,
+        "srScope": "four-channel-mean",
+        "srEpisodes": null,
+        "lscr": null,
+        "lscrScope": "four-channel-mean",
+        "lscrEpisodes": null,
+        "efficiency": null,
+        "efficiencyStatus": "not-reported"
+      }
+    },
+    "73": {
+      "channels": {
+        "None": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Equi.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Inv.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Full": {
+          "status": "reported",
+          "episodes": 10,
+          "successes": 0,
+          "lscr": 20.0,
+          "source": {
+            "runId": "task_icl_v3-amended260-index05-14-task61_08-rerun",
+            "perEpisodeFile": "per_task_eval/73_jigsaw_puzzle_assembly/Full/per_episode.jsonl",
+            "summaryFile": "per_task_eval/73_jigsaw_puzzle_assembly/Full/summary.json"
+          }
+        }
+      },
+      "mean": {
+        "status": "not-reported",
+        "sr": null,
+        "srScope": "four-channel-mean",
+        "srEpisodes": null,
+        "lscr": null,
+        "lscrScope": "four-channel-mean",
+        "lscrEpisodes": null,
+        "efficiency": null,
+        "efficiencyStatus": "not-reported"
+      }
+    },
+    "76": {
+      "channels": {
+        "None": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Equi.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Inv.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Full": {
+          "status": "reported",
+          "episodes": 10,
+          "successes": 0,
+          "lscr": 42.5,
+          "source": {
+            "runId": "task_icl_v3-amended260-index05-14-task61_08-rerun",
+            "perEpisodeFile": "per_task_eval/76_soup_serving/Full/per_episode.jsonl",
+            "summaryFile": "per_task_eval/76_soup_serving/Full/summary.json"
+          }
+        }
+      },
+      "mean": {
+        "status": "not-reported",
+        "sr": null,
+        "srScope": "four-channel-mean",
+        "srEpisodes": null,
+        "lscr": null,
+        "lscrScope": "four-channel-mean",
+        "lscrEpisodes": null,
+        "efficiency": null,
+        "efficiencyStatus": "not-reported"
+      }
+    },
+    "79": {
+      "channels": {
+        "None": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Equi.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Inv.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Full": {
+          "status": "reported",
+          "episodes": 10,
+          "successes": 7,
+          "lscr": 85.0,
+          "source": {
+            "runId": "task_icl_v3-amended260-index05-14-task61_08-rerun",
+            "perEpisodeFile": "per_task_eval/79_bimanual_piano_melody/Full/per_episode.jsonl",
+            "summaryFile": "per_task_eval/79_bimanual_piano_melody/Full/summary.json"
+          }
+        }
+      },
+      "mean": {
+        "status": "not-reported",
+        "sr": null,
+        "srScope": "four-channel-mean",
+        "srEpisodes": null,
+        "lscr": null,
+        "lscrScope": "four-channel-mean",
+        "lscrEpisodes": null,
+        "efficiency": null,
+        "efficiencyStatus": "not-reported"
+      }
+    },
+    "80": {
+      "channels": {
+        "None": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Equi.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Inv.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Full": {
+          "status": "reported",
+          "episodes": 10,
+          "successes": 0,
+          "lscr": 0.0,
+          "source": {
+            "runId": "task_icl_v3-amended260-index05-14-task61_08-rerun",
+            "perEpisodeFile": "per_task_eval/80_gaming_desk_setup/Full/per_episode.jsonl",
+            "summaryFile": "per_task_eval/80_gaming_desk_setup/Full/summary.json"
+          }
+        }
+      },
+      "mean": {
+        "status": "not-reported",
+        "sr": null,
+        "srScope": "four-channel-mean",
+        "srEpisodes": null,
+        "lscr": null,
+        "lscrScope": "four-channel-mean",
+        "lscrEpisodes": null,
+        "efficiency": null,
+        "efficiencyStatus": "not-reported"
+      }
+    },
+    "03": {
+      "channels": {
+        "None": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Equi.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Inv.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Full": {
+          "status": "reported",
+          "episodes": 10,
+          "successes": 0,
+          "lscr": 40.0,
+          "source": {
+            "runId": "task_icl_v3-amended260-index05-14-task61_08-rerun",
+            "perEpisodeFile": "per_task_eval/03_wine_glass_plate_balance/Full/per_episode.jsonl",
+            "summaryFile": "per_task_eval/03_wine_glass_plate_balance/Full/summary.json"
+          }
+        }
+      },
+      "mean": {
+        "status": "not-reported",
+        "sr": null,
+        "srScope": "four-channel-mean",
+        "srEpisodes": null,
+        "lscr": null,
+        "lscrScope": "four-channel-mean",
+        "lscrEpisodes": null,
+        "efficiency": null,
+        "efficiencyStatus": "not-reported"
+      }
+    },
+    "06": {
+      "channels": {
+        "None": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Equi.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Inv.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Full": {
+          "status": "reported",
+          "episodes": 10,
+          "successes": 1,
+          "lscr": 52.5,
+          "source": {
+            "runId": "task_icl_v3-amended260-index05-14-task61_08-rerun",
+            "perEpisodeFile": "per_task_eval/06_fruit_bowl_loading/Full/per_episode.jsonl",
+            "summaryFile": "per_task_eval/06_fruit_bowl_loading/Full/summary.json"
+          }
+        }
+      },
+      "mean": {
+        "status": "not-reported",
+        "sr": null,
+        "srScope": "four-channel-mean",
+        "srEpisodes": null,
+        "lscr": null,
+        "lscrScope": "four-channel-mean",
+        "lscrEpisodes": null,
+        "efficiency": null,
+        "efficiencyStatus": "not-reported"
+      }
+    },
+    "07": {
+      "channels": {
+        "None": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Equi.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Inv.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Full": {
+          "status": "reported",
+          "episodes": 10,
+          "successes": 0,
+          "lscr": 57.49999999999999,
+          "source": {
+            "runId": "task_icl_v3-amended260-index05-14-task61_08-rerun",
+            "perEpisodeFile": "per_task_eval/07_citrus_plate_loading/Full/per_episode.jsonl",
+            "summaryFile": "per_task_eval/07_citrus_plate_loading/Full/summary.json"
+          }
+        }
+      },
+      "mean": {
+        "status": "not-reported",
+        "sr": null,
+        "srScope": "four-channel-mean",
+        "srEpisodes": null,
+        "lscr": null,
+        "lscrScope": "four-channel-mean",
+        "lscrEpisodes": null,
+        "efficiency": null,
+        "efficiencyStatus": "not-reported"
+      }
+    },
+    "08": {
+      "channels": {
+        "None": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Equi.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Inv.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Full": {
+          "status": "reported",
+          "episodes": 10,
+          "successes": 0,
+          "lscr": 15.0,
+          "source": {
+            "runId": "task_icl_v3-amended260-index05-14-task61_08-rerun",
+            "perEpisodeFile": "per_task_eval/08_frypan_stand_pour/Full/per_episode.jsonl",
+            "summaryFile": "per_task_eval/08_frypan_stand_pour/Full/summary.json"
+          }
+        }
+      },
+      "mean": {
+        "status": "not-reported",
+        "sr": null,
+        "srScope": "four-channel-mean",
+        "srEpisodes": null,
+        "lscr": null,
+        "lscrScope": "four-channel-mean",
+        "lscrEpisodes": null,
+        "efficiency": null,
+        "efficiencyStatus": "not-reported"
+      }
+    },
+    "09": {
+      "channels": {
+        "None": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Equi.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Inv.": {
+          "status": "not-reported",
+          "episodes": null,
+          "successes": null,
+          "lscr": null,
+          "source": null
+        },
+        "Full": {
+          "status": "reported",
+          "episodes": 10,
+          "successes": 2,
+          "lscr": 0.0,
+          "source": {
+            "runId": "task_icl_v3-amended260-index05-14-task61_08-rerun",
+            "perEpisodeFile": "per_task_eval/09_cleaner_moisturizer_box_loading/Full/per_episode.jsonl",
+            "summaryFile": "per_task_eval/09_cleaner_moisturizer_box_loading/Full/summary.json"
+          }
+        }
+      },
+      "mean": {
+        "status": "not-reported",
+        "sr": null,
+        "srScope": "four-channel-mean",
+        "srEpisodes": null,
+        "lscr": null,
+        "lscrScope": "four-channel-mean",
+        "lscrEpisodes": null,
+        "efficiency": null,
+        "efficiencyStatus": "not-reported"
+      }
+    }
+  },
+  "evaluation": {
+    "schemaVersion": "1.0",
+    "isTemplate": false,
+    "protocol": {
+      "benchmarkCommit": "f96a8b2b4eb475483af66e9e03916b35bc43f1be",
+      "taskCount": 26,
+      "channels": [
+        "Full"
+      ],
+      "episodesPerChannel": 10,
+      "seeds": {
+        "episodeIndices": [
+          5,
+          6,
+          7,
+          8,
+          9,
+          10,
+          11,
+          12,
+          13,
+          14
+        ],
+        "byTask": {
+          "03": [
+            200300004,
+            200300005,
+            200300006,
+            200300007,
+            200300008,
+            200300009,
+            200300010,
+            200300011,
+            200300012,
+            200300013
+          ],
+          "06": [
+            200600004,
+            200600005,
+            200600006,
+            200600007,
+            200600008,
+            200600009,
+            200600010,
+            200600011,
+            200600012,
+            200600013
+          ],
+          "07": [
+            200700004,
+            200700005,
+            200700006,
+            200700007,
+            200700008,
+            200700009,
+            200700010,
+            200700011,
+            200700012,
+            200700013
+          ],
+          "08": [
+            200800004,
+            200800005,
+            200800006,
+            200800007,
+            200800008,
+            200800009,
+            200800010,
+            200800011,
+            200800012,
+            200800013
+          ],
+          "09": [
+            200900004,
+            200900005,
+            200900006,
+            200900007,
+            200900008,
+            200900009,
+            200900010,
+            200900011,
+            200900012,
+            200900013
+          ],
+          "12": [
+            201200004,
+            201200005,
+            201200006,
+            201200007,
+            201200008,
+            201200009,
+            201200010,
+            201200011,
+            201200012,
+            201200013
+          ],
+          "21": [
+            202100004,
+            202100005,
+            202100006,
+            202100007,
+            202100008,
+            202100009,
+            202100010,
+            202100011,
+            202100012,
+            202100013
+          ],
+          "22": [
+            202200004,
+            202200005,
+            202200006,
+            202200007,
+            202200008,
+            202200009,
+            202200010,
+            202200011,
+            202200012,
+            202200013
+          ],
+          "24": [
+            202400004,
+            202400005,
+            202400006,
+            202400007,
+            202400008,
+            202400009,
+            202400010,
+            202400011,
+            202400012,
+            202400013
+          ],
+          "26": [
+            202600004,
+            202600005,
+            202600006,
+            202600007,
+            202600008,
+            202600009,
+            202600010,
+            202600011,
+            202600012,
+            202600013
+          ],
+          "27": [
+            202700004,
+            202700005,
+            202700006,
+            202700007,
+            202700008,
+            202700009,
+            202700010,
+            202700011,
+            202700012,
+            202700013
+          ],
+          "32": [
+            203200004,
+            203200005,
+            203200006,
+            203200007,
+            203200008,
+            203200009,
+            203200010,
+            203200011,
+            203200012,
+            203200013
+          ],
+          "34": [
+            203400004,
+            203400005,
+            203400006,
+            203400007,
+            203400008,
+            203400009,
+            203400010,
+            203400011,
+            203400012,
+            203400013
+          ],
+          "42": [
+            204200004,
+            204200005,
+            204200006,
+            204200007,
+            204200008,
+            204200009,
+            204200010,
+            204200011,
+            204200012,
+            204200013
+          ],
+          "43": [
+            204300004,
+            204300005,
+            204300006,
+            204300007,
+            204300008,
+            204300009,
+            204300010,
+            204300011,
+            204300012,
+            204300013
+          ],
+          "44": [
+            204400004,
+            204400005,
+            204400006,
+            204400007,
+            204400008,
+            204400009,
+            204400010,
+            204400011,
+            204400012,
+            204400013
+          ],
+          "51": [
+            205100004,
+            205100005,
+            205100006,
+            205100007,
+            205100008,
+            205100009,
+            205100010,
+            205100011,
+            205100012,
+            205100013
+          ],
+          "60": [
+            206000004,
+            206000005,
+            206000006,
+            206000007,
+            206000008,
+            206000009,
+            206000010,
+            206000011,
+            206000012,
+            206000013
+          ],
+          "61": [
+            206100004,
+            206100005,
+            206100006,
+            206100007,
+            206100008,
+            206100009,
+            206100010,
+            206100011,
+            206100012,
+            206100013
+          ],
+          "62": [
+            206200004,
+            206200005,
+            206200006,
+            206200007,
+            206200008,
+            206200009,
+            206200010,
+            206200011,
+            206200012,
+            206200013
+          ],
+          "64": [
+            206400004,
+            206400005,
+            206400006,
+            206400007,
+            206400008,
+            206400009,
+            206400010,
+            206400011,
+            206400012,
+            206400013
+          ],
+          "67": [
+            206700004,
+            206700005,
+            206700006,
+            206700007,
+            206700008,
+            206700009,
+            206700010,
+            206700011,
+            206700012,
+            206700013
+          ],
+          "73": [
+            207300004,
+            207300005,
+            207300006,
+            207300007,
+            207300008,
+            207300009,
+            207300010,
+            207300011,
+            207300012,
+            207300013
+          ],
+          "76": [
+            207600004,
+            207600005,
+            207600006,
+            207600007,
+            207600008,
+            207600009,
+            207600010,
+            207600011,
+            207600012,
+            207600013
+          ],
+          "79": [
+            207900004,
+            207900005,
+            207900006,
+            207900007,
+            207900008,
+            207900009,
+            207900010,
+            207900011,
+            207900012,
+            207900013
+          ],
+          "80": [
+            208000004,
+            208000005,
+            208000006,
+            208000007,
+            208000008,
+            208000009,
+            208000010,
+            208000011,
+            208000012,
+            208000013
+          ]
+        }
+      },
+      "stepUnit": "physics_steps",
+      "physicsHz": 60,
+      "controlHz": 20,
+      "executionStepBudget": {
+        "rule": "ceil(1.5 × expert_time_step / 3) policy steps",
+        "perTaskPolicySteps": {
+          "03": 1082,
+          "06": 640,
+          "07": 602,
+          "08": 1144,
+          "09": 642,
+          "12": 1149,
+          "21": 827,
+          "22": 681,
+          "24": 1542,
+          "26": 871,
+          "27": 1190,
+          "32": 852,
+          "34": 1080,
+          "42": 668,
+          "43": 1375,
+          "44": 1061,
+          "51": 1023,
+          "60": 995,
+          "61": 1034,
+          "62": 1077,
+          "64": 665,
+          "67": 964,
+          "73": 1213,
+          "76": 986,
+          "79": 681,
+          "80": 593
+        },
+        "perTaskPhysicsSteps": {
+          "03": 3246,
+          "06": 1920,
+          "07": 1806,
+          "08": 3432,
+          "09": 1926,
+          "12": 3447,
+          "21": 2481,
+          "22": 2043,
+          "24": 4626,
+          "26": 2613,
+          "27": 3570,
+          "32": 2556,
+          "34": 3240,
+          "42": 2004,
+          "43": 4125,
+          "44": 3183,
+          "51": 3069,
+          "60": 2985,
+          "61": 3102,
+          "62": 3231,
+          "64": 1995,
+          "67": 2892,
+          "73": 3639,
+          "76": 2958,
+          "79": 2043,
+          "80": 1779
+        }
+      },
+      "expertReference": "Bench2Dex scene expert_time_step at benchmarkCommit",
+      "protocolChanges": [
+        "Full/unseen only; 10 episodes per task rather than the 50-episode reference.",
+        "17 same-index damage-recovery replacements were selected by provenance, not outcome.",
+        "Task 61 episode 8 was rerun at its original index/seed because its original native evaluator log was lost; the new valid outcome replaces that reporting slot."
+      ]
+    },
+    "coverage": {
+      "evaluatedTaskCount": 26,
+      "evaluatedEpisodeCount": 260,
+      "completedTaskCount": {
+        "None": 0,
+        "Equi.": 0,
+        "Inv.": 0,
+        "Full": 26,
+        "mean": 0
+      },
+      "expectedFullEpisodeCount": 1300,
+      "expectedFourChannelEpisodeCount": 5200,
+      "missingTaskChannels": {
+        "03": [
+          "None",
+          "Equi.",
+          "Inv."
+        ],
+        "06": [
+          "None",
+          "Equi.",
+          "Inv."
+        ],
+        "07": [
+          "None",
+          "Equi.",
+          "Inv."
+        ],
+        "08": [
+          "None",
+          "Equi.",
+          "Inv."
+        ],
+        "09": [
+          "None",
+          "Equi.",
+          "Inv."
+        ],
+        "12": [
+          "None",
+          "Equi.",
+          "Inv."
+        ],
+        "21": [
+          "None",
+          "Equi.",
+          "Inv."
+        ],
+        "22": [
+          "None",
+          "Equi.",
+          "Inv."
+        ],
+        "24": [
+          "None",
+          "Equi.",
+          "Inv."
+        ],
+        "26": [
+          "None",
+          "Equi.",
+          "Inv."
+        ],
+        "27": [
+          "None",
+          "Equi.",
+          "Inv."
+        ],
+        "32": [
+          "None",
+          "Equi.",
+          "Inv."
+        ],
+        "34": [
+          "None",
+          "Equi.",
+          "Inv."
+        ],
+        "42": [
+          "None",
+          "Equi.",
+          "Inv."
+        ],
+        "43": [
+          "None",
+          "Equi.",
+          "Inv."
+        ],
+        "44": [
+          "None",
+          "Equi.",
+          "Inv."
+        ],
+        "51": [
+          "None",
+          "Equi.",
+          "Inv."
+        ],
+        "60": [
+          "None",
+          "Equi.",
+          "Inv."
+        ],
+        "61": [
+          "None",
+          "Equi.",
+          "Inv."
+        ],
+        "62": [
+          "None",
+          "Equi.",
+          "Inv."
+        ],
+        "64": [
+          "None",
+          "Equi.",
+          "Inv."
+        ],
+        "67": [
+          "None",
+          "Equi.",
+          "Inv."
+        ],
+        "73": [
+          "None",
+          "Equi.",
+          "Inv."
+        ],
+        "76": [
+          "None",
+          "Equi.",
+          "Inv."
+        ],
+        "79": [
+          "None",
+          "Equi.",
+          "Inv."
+        ],
+        "80": [
+          "None",
+          "Equi.",
+          "Inv."
+        ]
+      }
+    },
+    "aggregation": {
+      "units": {
+        "sr": "percent",
+        "lscr": "percent",
+        "efficiency": "ratio",
+        "steps": "physics_steps"
+      },
+      "sr": "100 * successes / episodes per task/channel; equal channel mean per task; equal task mean over all 26 settings.",
+      "lscr": "100 * mean episode latched_stage_completion_rate, including failures; equal channel mean per task; equal task mean over all 26 settings.",
+      "efficiency": "expert_time_step / mean successful steps_to_stable_success. Pool successful episodes across all four channels before calculating task mean efficiency. Overall values average task efficiencies over tasks with success; report valid-task counts.",
+      "missing": "null means unreported (NR). Efficiency without successes is null with no-success status. Valid-task counts are null when unreported, and 0 for an evaluated scope without successes.",
+      "scope": "Overall channel aggregates require all 26 settings. Four-channel means require all four channels. Never substitute Full-only scores for means or include duplicate reruns."
+    },
+    "logs": {
+      "repository": "Bench2Dex/Leaderboard",
+      "policyPath": "gpt6_astra_1x_icl",
+      "revision": null,
+      "status": "not-archived"
+    }
+  }
+}

gpt6_astra_1x_icl/per_task_eval/03_wine_glass_plate_balance/Full/summary.json

@@ -0,0 +1,212 @@
+{
+  "metric_schema_version": "core-v1.8-policy-step-query-split",
+  "metadata": {
+    "policy_name": "gpt-6-astra",
+    "base_seed": 200300000,
+    "seed_policy": "episode_seed = base_seed + episode_index - 1",
+    "randomization_recorded": true,
+    "n_episodes": 10,
+    "successes": 0,
+    "error_count": 0,
+    "evaluation_protocol": "reach_and_stop",
+    "early_stop": true,
+    "success_definition": "terminal.raw_condition holds for dwell_time_s",
+    "measurement_window": "episode_start_to_stable_success_or_failure_budget",
+    "max_episode_steps": 3246,
+    "dwell_time_s": 0.5,
+    "field_semantics": {
+      "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
+      "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
+      "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
+      "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
+      "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
+      "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
+      "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
+      "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
+      "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
+      "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
+      "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
+      "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
+      "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
+      "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
+      "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
+      "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
+      "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
+      "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
+      "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
+      "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
+      "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
+      "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
+      "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
+      "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
+      "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
+    }
+  },
+  "core_metrics": {
+    "completion": {
+      "success_rate": 0.0,
+      "success_rate_ci": {
+        "mean": 0.0,
+        "ci_low": 0.0,
+        "ci_high": 0.0
+      }
+    },
+    "progress": {
+      "latched_stage_completion_rate": 0.39999999999999997
+    },
+    "efficiency": {
+      "avg_time_to_success_s": null
+    },
+    "safety": {
+      "safe_success_rate": 0.0,
+      "hard_violation_rate": 0.9,
+      "drop_rate": 0.9,
+      "high_speed_violation_rate": 0.1,
+      "safety_violation_step_rate": 0.04460874922982132
+    },
+    "robustness": {}
+  },
+  "diagnostic_metrics": {
+    "completion": {
+      "ever_instant_success_rate": 0.0,
+      "at_end_success_rate": 0.0,
+      "task_family_success_rate": {
+        "table_setting": 0.0
+      }
+    },
+    "progress": {
+      "current_stage_completion_rate": 0.16666666666666666,
+      "current_chain_depth_progress_score": 0.16666666666666666,
+      "latched_chain_depth_progress_score": 0.39999999999999997,
+      "avg_current_chain_depth": 0.5,
+      "avg_latched_chain_depth": 1.2
+    },
+    "efficiency": {
+      "task_efficiency": null,
+      "expert_time_s": 36.06666666666666,
+      "expert_time_step": 2164,
+      "avg_steps_to_success": null,
+      "avg_policy_steps_to_success": null,
+      "avg_policy_queries_to_success": null
+    },
+    "safety": {
+      "safety_violation_events_per_step": 0.0004313000616142945,
+      "violation_counts": {
+        "drop_obj_194_wineglass_1": 1,
+        "drop_to_table_obj_194_wineglass_1": 9,
+        "drop_to_table_obj_194_wineglass_2": 1,
+        "drop_to_table_obj_194_wineglass_3": 2,
+        "high_speed_obj_194_wineglass_1": 1
+      },
+      "robot_constraint_violation_rate": 0.7,
+      "finger_joint_limit_saturation_rate": 0.0,
+      "joint_limit_active_step_rate": 0.0012630930375847197,
+      "finger_joint_limit_saturation_step_rate": 0.0,
+      "max_joint_limit_excess_rad": 0.06860432028770447,
+      "robot_constraint_diagnostics": {
+        "joint_limit_soft_counts": {
+          "middle_q2": 29,
+          "multi_pinky_q1": 3,
+          "multi_ring_q1": 1,
+          "multi_ring_q2": 8
+        },
+        "joint_limit_hard_counts": {
+          "middle_q2": 1,
+          "multi_pinky_q1": 2,
+          "multi_ring_q1": 3,
+          "multi_ring_q2": 2,
+          "ring_q1": 1
+        },
+        "joint_limit_hard_episode_events": 9,
+        "finger_joint_limit_saturation_counts": {},
+        "joint_limit_max_excess_by_joint": {
+          "index_q1": 0.0030562784522771835,
+          "index_q2": 2.8614797198400765e-09,
+          "middle_q1": 0.0012579697649925947,
+          "middle_q2": 0.05049429088830948,
+          "multi_index_q1": 0.00476967403665185,
+          "multi_index_q2": 2.294862300189493e-09,
+          "multi_middle_q1": 0.002055603312328458,
+          "multi_middle_q2": 0.001359704416245222,
+          "multi_pinky_q1": 0.026384813711047173,
+          "multi_pinky_q2": 0.0026463610120117664,
+          "multi_ring_q1": 0.0110697066411376,
+          "multi_ring_q2": 0.06860432028770447,
+          "multi_thumb_q1": 0.0005444049602374434,
+          "multi_thumb_q2": 0.00013726763427257538,
+          "pinky_q1": 0.001833398942835629,
+          "pinky_q2": 0.0018782949773594737,
+          "ring_q1": 0.006766872014850378,
+          "ring_q2": 0.0024923672899603844,
+          "thumb_q1": 0.0006504592020064592,
+          "thumb_q2": 0.0006376263336278498
+        },
+        "coverage": {
+          "hard_limit_source": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          },
+          "joint_role_manifest": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          }
+        },
+        "hard_limit_sources": {
+          "runtime.joint_pos_limits": 10
+        },
+        "soft_limit_sources": {
+          "runtime.soft_joint_pos_limits": 10
+        },
+        "unknown_runtime_joint_names": [],
+        "baseline_checks": {
+          "post_home": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          },
+          "post_reset": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          }
+        }
+      }
+    },
+    "robot_motion": {
+      "joint_acceleration_rms": 9.036153273750015,
+      "joint_effort_rms": 91.83046455785966,
+      "joint_jerk_rms": 915.2846308193771,
+      "joint_velocity_rms": 0.480255237631612
+    },
+    "grasp": {
+      "kinematic_grasp_stability_index": 0.9986104463301466,
+      "mean_kinematic_grasp_stability": 0.5942982607822139,
+      "per_object_best_kinematic_grasp_stability": {
+        "obj_194_wineglass_1": 0.8956681838712092,
+        "obj_194_wineglass_2": 0.995692048610797,
+        "obj_194_wineglass_3": 0.9360362115771004
+      },
+      "grasp_gsi_diagnostics": {
+        "mean_anti_slip_score": 0.8491921681385672,
+        "mean_duration_score": 0.46705483672212,
+        "mean_hold_score": 0.3785936537276675,
+        "mean_kinematic_stability_score": 0.23044707313794188,
+        "mean_legacy_gsi": 0.4845957828923964,
+        "mean_lift_score": 0.44093242965701374,
+        "mean_motion_stability_score": 0.3529374016477624
+      }
+    },
+    "tool": {
+      "tool_selection_accuracy": null,
+      "tool_switch_success_rate": null,
+      "tool_switch_total": 0,
+      "tool_switch_successful": 0
+    },
+    "runtime": {
+      "avg_steps": 3246,
+      "error_count": 0
+    }
+  }
+}

gpt6_astra_1x_icl/per_task_eval/06_fruit_bowl_loading/Full/summary.json

@@ -0,0 +1,275 @@
+{
+  "metric_schema_version": "core-v1.8-policy-step-query-split",
+  "metadata": {
+    "policy_name": "gpt-6-astra",
+    "base_seed": 200600000,
+    "seed_policy": "episode_seed = base_seed + episode_index - 1",
+    "randomization_recorded": true,
+    "n_episodes": 10,
+    "successes": 1,
+    "error_count": 1,
+    "evaluation_protocol": "reach_and_stop",
+    "early_stop": true,
+    "success_definition": "terminal.raw_condition holds for dwell_time_s",
+    "measurement_window": "episode_start_to_stable_success_or_failure_budget",
+    "max_episode_steps": 1920,
+    "dwell_time_s": 0.5,
+    "field_semantics": {
+      "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
+      "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
+      "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
+      "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
+      "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
+      "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
+      "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
+      "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
+      "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
+      "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
+      "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
+      "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
+      "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
+      "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
+      "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
+      "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
+      "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
+      "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
+      "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
+      "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
+      "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
+      "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
+      "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
+      "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
+      "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
+    }
+  },
+  "core_metrics": {
+    "completion": {
+      "success_rate": 0.1,
+      "success_rate_ci": {
+        "mean": 0.1,
+        "ci_low": 0.0,
+        "ci_high": 0.3
+      }
+    },
+    "progress": {
+      "latched_stage_completion_rate": 0.525
+    },
+    "efficiency": {
+      "avg_time_to_success_s": 31.016666666666186
+    },
+    "safety": {
+      "safe_success_rate": 0.0,
+      "hard_violation_rate": 0.9,
+      "drop_rate": 0.9,
+      "high_speed_violation_rate": 0.1,
+      "safety_violation_step_rate": 0.0042205222896333425
+    },
+    "robustness": {}
+  },
+  "diagnostic_metrics": {
+    "completion": {
+      "ever_instant_success_rate": 0.1,
+      "at_end_success_rate": 0.1,
+      "task_family_success_rate": {
+        "rigid_pick_place": 0.1
+      }
+    },
+    "progress": {
+      "current_stage_completion_rate": 0.325,
+      "current_chain_depth_progress_score": 0.36666666666666664,
+      "latched_chain_depth_progress_score": 0.5666666666666667,
+      "avg_current_chain_depth": 1.1,
+      "avg_latched_chain_depth": 1.7
+    },
+    "efficiency": {
+      "task_efficiency": 0.6878022568511659,
+      "expert_time_s": 21.333333333333332,
+      "expert_time_step": 1280,
+      "avg_steps_to_success": 1861,
+      "avg_policy_steps_to_success": 621,
+      "avg_policy_queries_to_success": 25
+    },
+    "safety": {
+      "safety_violation_events_per_step": 0.0006330783434450013,
+      "violation_counts": {
+        "drop_obj_013_apple_3": 1,
+        "drop_to_table_obj_013_apple_3": 2,
+        "drop_to_table_obj_013_apple_4": 8,
+        "high_speed_obj_013_apple_3": 1
+      },
+      "robot_constraint_violation_rate": 1.0,
+      "finger_joint_limit_saturation_rate": 1.0,
+      "joint_limit_active_step_rate": 0.17246109206014243,
+      "finger_joint_limit_saturation_step_rate": 0.1280400949617515,
+      "max_joint_limit_excess_rad": 3.382310390472412,
+      "robot_constraint_diagnostics": {
+        "joint_limit_soft_counts": {
+          "left_index_1_joint": 193,
+          "left_index_2_joint": 9,
+          "left_little_1_joint": 382,
+          "left_little_2_joint": 84,
+          "left_middle_1_joint": 303,
+          "left_middle_2_joint": 5,
+          "left_ring_1_joint": 308,
+          "left_ring_2_joint": 29,
+          "left_thumb_1_joint": 59,
+          "left_thumb_2_joint": 216,
+          "left_thumb_3_joint": 190,
+          "left_thumb_4_joint": 70,
+          "right_index_1_joint": 680,
+          "right_index_2_joint": 25,
+          "right_little_1_joint": 129,
+          "right_little_2_joint": 57,
+          "right_middle_1_joint": 798,
+          "right_middle_2_joint": 35,
+          "right_ring_1_joint": 382,
+          "right_ring_2_joint": 56,
+          "right_thumb_1_joint": 10,
+          "right_thumb_2_joint": 1635,
+          "right_thumb_3_joint": 36,
+          "right_thumb_4_joint": 1
+        },
+        "joint_limit_hard_counts": {
+          "left_index_1_joint": 7,
+          "left_index_2_joint": 2,
+          "left_little_1_joint": 22,
+          "left_little_2_joint": 5,
+          "left_middle_1_joint": 16,
+          "left_middle_2_joint": 3,
+          "left_ring_1_joint": 21,
+          "left_ring_2_joint": 6,
+          "left_thumb_1_joint": 3,
+          "left_thumb_2_joint": 15,
+          "left_thumb_3_joint": 5,
+          "left_thumb_4_joint": 5,
+          "right_index_1_joint": 22,
+          "right_index_2_joint": 2,
+          "right_little_1_joint": 8,
+          "right_little_2_joint": 2,
+          "right_middle_1_joint": 25,
+          "right_middle_2_joint": 3,
+          "right_ring_1_joint": 15,
+          "right_ring_2_joint": 2,
+          "right_thumb_1_joint": 3,
+          "right_thumb_2_joint": 25,
+          "right_thumb_3_joint": 2,
+          "right_thumb_4_joint": 2
+        },
+        "joint_limit_hard_episode_events": 79,
+        "finger_joint_limit_saturation_counts": {
+          "left_index_1_joint": 140,
+          "left_index_2_joint": 3,
+          "left_little_1_joint": 226,
+          "left_little_2_joint": 71,
+          "left_middle_1_joint": 170,
+          "left_ring_1_joint": 162,
+          "left_thumb_1_joint": 39,
+          "left_thumb_2_joint": 93,
+          "left_thumb_3_joint": 184,
+          "left_thumb_4_joint": 56,
+          "right_index_1_joint": 409,
+          "right_index_2_joint": 3,
+          "right_little_1_joint": 82,
+          "right_little_2_joint": 53,
+          "right_middle_1_joint": 617,
+          "right_middle_2_joint": 32,
+          "right_ring_1_joint": 241,
+          "right_ring_2_joint": 48,
+          "right_thumb_2_joint": 1231,
+          "right_thumb_3_joint": 34
+        },
+        "joint_limit_max_excess_by_joint": {
+          "left_index_1_joint": 0.7339382171630859,
+          "left_index_2_joint": 0.3402869701385498,
+          "left_little_1_joint": 0.9096411466598511,
+          "left_little_2_joint": 1.1495881080627441,
+          "left_middle_1_joint": 0.9641392827033997,
+          "left_middle_2_joint": 0.2039506435394287,
+          "left_ring_1_joint": 1.0324233770370483,
+          "left_ring_2_joint": 0.24595904350280762,
+          "left_thumb_1_joint": 0.7501708269119263,
+          "left_thumb_2_joint": 0.987554132938385,
+          "left_thumb_3_joint": 3.382310390472412,
+          "left_thumb_4_joint": 1.9342007637023926,
+          "right_index_1_joint": 0.8946888446807861,
+          "right_index_2_joint": 0.273584246635437,
+          "right_little_1_joint": 0.932692289352417,
+          "right_little_2_joint": 0.7767856121063232,
+          "right_middle_1_joint": 0.9934417605400085,
+          "right_middle_2_joint": 0.6010532379150391,
+          "right_ring_1_joint": 0.9061310887336731,
+          "right_ring_2_joint": 0.626173734664917,
+          "right_thumb_1_joint": 0.0660252571105957,
+          "right_thumb_2_joint": 0.6172922253608704,
+          "right_thumb_3_joint": 1.7565410137176514,
+          "right_thumb_4_joint": 0.025886058807373047
+        },
+        "coverage": {
+          "hard_limit_source": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          },
+          "joint_role_manifest": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          }
+        },
+        "hard_limit_sources": {
+          "runtime.joint_pos_limits": 10
+        },
+        "soft_limit_sources": {
+          "runtime.soft_joint_pos_limits": 10
+        },
+        "unknown_runtime_joint_names": [],
+        "baseline_checks": {
+          "post_home": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          },
+          "post_reset": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          }
+        }
+      }
+    },
+    "robot_motion": {
+      "joint_acceleration_rms": 54.53032325899591,
+      "joint_effort_rms": 49.91163010593858,
+      "joint_jerk_rms": 5538.471765760078,
+      "joint_velocity_rms": 1.1476248155165663
+    },
+    "grasp": {
+      "kinematic_grasp_stability_index": 0.6750588898147976,
+      "mean_kinematic_grasp_stability": 0.17570324777356666,
+      "per_object_best_kinematic_grasp_stability": {
+        "obj_011_banana_2": 0.21171605483204742,
+        "obj_013_apple_3": 0.6421852607741502,
+        "obj_013_apple_4": 0.2131514025765569
+      },
+      "grasp_gsi_diagnostics": {
+        "mean_anti_slip_score": 0.9908254908253259,
+        "mean_duration_score": 0.03099287995244664,
+        "mean_hold_score": 0.030345580896439787,
+        "mean_kinematic_stability_score": 0.01797651457830877,
+        "mean_legacy_gsi": 0.22654757608459167,
+        "mean_lift_score": 0.08310973547380095,
+        "mean_motion_stability_score": 0.027594197289731306
+      }
+    },
+    "tool": {
+      "tool_selection_accuracy": null,
+      "tool_switch_success_rate": null,
+      "tool_switch_total": 0,
+      "tool_switch_successful": 0
+    },
+    "runtime": {
+      "avg_steps": 1895.5,
+      "error_count": 1
+    }
+  }
+}

gpt6_astra_1x_icl/per_task_eval/07_citrus_plate_loading/Full/summary.json

@@ -0,0 +1,278 @@
+{
+  "metric_schema_version": "core-v1.8-policy-step-query-split",
+  "metadata": {
+    "policy_name": "gpt-6-astra",
+    "base_seed": 200700000,
+    "seed_policy": "episode_seed = base_seed + episode_index - 1",
+    "randomization_recorded": true,
+    "n_episodes": 10,
+    "successes": 0,
+    "error_count": 1,
+    "evaluation_protocol": "reach_and_stop",
+    "early_stop": true,
+    "success_definition": "terminal.raw_condition holds for dwell_time_s",
+    "measurement_window": "episode_start_to_stable_success_or_failure_budget",
+    "max_episode_steps": 1806,
+    "dwell_time_s": 0.5,
+    "field_semantics": {
+      "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
+      "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
+      "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
+      "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
+      "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
+      "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
+      "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
+      "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
+      "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
+      "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
+      "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
+      "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
+      "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
+      "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
+      "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
+      "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
+      "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
+      "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
+      "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
+      "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
+      "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
+      "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
+      "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
+      "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
+      "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
+    }
+  },
+  "core_metrics": {
+    "completion": {
+      "success_rate": 0.0,
+      "success_rate_ci": {
+        "mean": 0.0,
+        "ci_low": 0.0,
+        "ci_high": 0.0
+      }
+    },
+    "progress": {
+      "latched_stage_completion_rate": 0.575
+    },
+    "efficiency": {
+      "avg_time_to_success_s": null
+    },
+    "safety": {
+      "safe_success_rate": 0.0,
+      "hard_violation_rate": 0.9,
+      "drop_rate": 0.9,
+      "high_speed_violation_rate": 0.3,
+      "safety_violation_step_rate": 0.05840922890103218
+    },
+    "robustness": {}
+  },
+  "diagnostic_metrics": {
+    "completion": {
+      "ever_instant_success_rate": 0.4,
+      "at_end_success_rate": 0.2,
+      "task_family_success_rate": {
+        "rigid_pick_place": 0.0
+      }
+    },
+    "progress": {
+      "current_stage_completion_rate": 0.35,
+      "current_chain_depth_progress_score": 0.4,
+      "latched_chain_depth_progress_score": 0.7,
+      "avg_current_chain_depth": 0.8,
+      "avg_latched_chain_depth": 1.4
+    },
+    "efficiency": {
+      "task_efficiency": null,
+      "expert_time_s": 20.05,
+      "expert_time_step": 1203,
+      "avg_steps_to_success": null,
+      "avg_policy_steps_to_success": null,
+      "avg_policy_queries_to_success": null
+    },
+    "safety": {
+      "safety_violation_events_per_step": 0.001639344262295082,
+      "violation_counts": {
+        "drop_obj_014_lemon_2": 2,
+        "drop_obj_014_lemon_3": 1,
+        "drop_obj_017_orange_4": 2,
+        "drop_to_table_obj_014_lemon_2": 3,
+        "drop_to_table_obj_014_lemon_3": 1,
+        "drop_to_table_obj_017_orange_4": 9,
+        "drop_to_table_obj_017_orange_5": 3,
+        "high_speed_obj_014_lemon_2": 2,
+        "high_speed_obj_014_lemon_3": 1,
+        "high_speed_obj_017_orange_4": 2,
+        "high_speed_obj_029_plate_1": 1
+      },
+      "robot_constraint_violation_rate": 0.9,
+      "finger_joint_limit_saturation_rate": 0.6,
+      "joint_limit_active_step_rate": 0.053005464480874315,
+      "finger_joint_limit_saturation_step_rate": 0.009653916211293261,
+      "max_joint_limit_excess_rad": 1.5978164672851562,
+      "robot_constraint_diagnostics": {
+        "joint_limit_soft_counts": {
+          "left_index_dip_joint": 11,
+          "left_index_mcp_joint": 4,
+          "left_index_pip_joint": 16,
+          "left_middle_dip_joint": 35,
+          "left_middle_mcp_joint": 29,
+          "left_middle_pip_joint": 73,
+          "left_middle_pitch_joint": 17,
+          "left_pinky_mcp_joint": 92,
+          "left_pinky_pip_joint": 37,
+          "left_ring_dip_joint": 23,
+          "left_ring_mcp_joint": 46,
+          "left_ring_pip_joint": 52,
+          "left_thumb_dip_joint": 21,
+          "left_thumb_mcp_joint": 358,
+          "left_thumb_pip_joint": 1,
+          "left_thumb_yaw_joint": 52,
+          "right_index_pip_joint": 2,
+          "right_middle_pip_joint": 7,
+          "right_ring_pip_joint": 3,
+          "right_thumb_mcp_joint": 250
+        },
+        "joint_limit_hard_counts": {
+          "left_index_dip_joint": 1,
+          "left_index_mcp_joint": 1,
+          "left_index_pip_joint": 32,
+          "left_middle_dip_joint": 2,
+          "left_middle_mcp_joint": 2,
+          "left_middle_pip_joint": 133,
+          "left_middle_pitch_joint": 1,
+          "left_pinky_dip_joint": 1,
+          "left_pinky_mcp_joint": 9,
+          "left_pinky_pip_joint": 3,
+          "left_ring_dip_joint": 2,
+          "left_ring_mcp_joint": 4,
+          "left_ring_pip_joint": 10,
+          "left_thumb_dip_joint": 2,
+          "left_thumb_mcp_joint": 30,
+          "left_thumb_pip_joint": 1,
+          "left_thumb_yaw_joint": 2,
+          "right_index_pip_joint": 30,
+          "right_middle_pip_joint": 72,
+          "right_ring_pip_joint": 2,
+          "right_thumb_mcp_joint": 27
+        },
+        "joint_limit_hard_episode_events": 220,
+        "finger_joint_limit_saturation_counts": {
+          "left_index_dip_joint": 6,
+          "left_middle_dip_joint": 29,
+          "left_middle_mcp_joint": 15,
+          "left_pinky_mcp_joint": 6,
+          "left_pinky_pip_joint": 26,
+          "left_ring_dip_joint": 11,
+          "left_ring_mcp_joint": 23,
+          "left_ring_pip_joint": 25,
+          "left_thumb_dip_joint": 12,
+          "left_thumb_mcp_joint": 44,
+          "left_thumb_yaw_joint": 44,
+          "right_thumb_mcp_joint": 10
+        },
+        "joint_limit_max_excess_by_joint": {
+          "left_index_dip_joint": 0.40528249740600586,
+          "left_index_mcp_joint": 0.06639307737350464,
+          "left_index_pip_joint": 0.14446650445461273,
+          "left_index_pitch_joint": 0.0033638954628258944,
+          "left_middle_dip_joint": 1.5978164672851562,
+          "left_middle_mcp_joint": 0.6392132043838501,
+          "left_middle_pip_joint": 0.2057327926158905,
+          "left_middle_pitch_joint": 0.03984842076897621,
+          "left_pinky_dip_joint": 0.006871405057609081,
+          "left_pinky_mcp_joint": 0.28638166189193726,
+          "left_pinky_pip_joint": 1.280145525932312,
+          "left_ring_dip_joint": 0.534247875213623,
+          "left_ring_mcp_joint": 0.7535475492477417,
+          "left_ring_pip_joint": 0.8452719449996948,
+          "left_thumb_dip_joint": 0.7821993231773376,
+          "left_thumb_mcp_joint": 0.6103029251098633,
+          "left_thumb_pip_joint": 0.017655611038208008,
+          "left_thumb_yaw_joint": 0.7936220169067383,
+          "right_index_dip_joint": 5.028687519370578e-05,
+          "right_index_pip_joint": 0.052236296236515045,
+          "right_index_pitch_joint": 0.00014959948020987213,
+          "right_middle_dip_joint": 0.0003015365800820291,
+          "right_middle_pip_joint": 0.19331547617912292,
+          "right_middle_pitch_joint": 0.0002718656905926764,
+          "right_pinky_dip_joint": 0.00038979147211648524,
+          "right_pinky_mcp_joint": 0.0006187328253872693,
+          "right_pinky_pip_joint": 0.0008713925490155816,
+          "right_ring_dip_joint": 0.00021358171943575144,
+          "right_ring_mcp_joint": 0.0001470772986067459,
+          "right_ring_pip_joint": 0.03229715675115585,
+          "right_thumb_dip_joint": 2.9477143471012823e-05,
+          "right_thumb_mcp_joint": 0.29763340950012207,
+          "right_thumb_pip_joint": 4.7241366701200604e-05,
+          "right_thumb_yaw_joint": 1.998916657197647e-11
+        },
+        "coverage": {
+          "hard_limit_source": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          },
+          "joint_role_manifest": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          }
+        },
+        "hard_limit_sources": {
+          "runtime.joint_pos_limits": 10
+        },
+        "soft_limit_sources": {
+          "runtime.soft_joint_pos_limits": 10
+        },
+        "unknown_runtime_joint_names": [],
+        "baseline_checks": {
+          "post_home": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          },
+          "post_reset": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          }
+        }
+      }
+    },
+    "robot_motion": {
+      "joint_acceleration_rms": 24.961714088569405,
+      "joint_effort_rms": 32.44377230761677,
+      "joint_jerk_rms": 2483.7157956358647,
+      "joint_velocity_rms": 0.5874940997851043
+    },
+    "grasp": {
+      "kinematic_grasp_stability_index": 0.8622755407368967,
+      "mean_kinematic_grasp_stability": 0.41917867169956197,
+      "per_object_best_kinematic_grasp_stability": {
+        "obj_014_lemon_2": 0.7741835969714579,
+        "obj_014_lemon_3": 0.7023812118533483,
+        "obj_017_orange_4": 0.4459987070772581,
+        "obj_017_orange_5": 0.3496105877623669
+      },
+      "grasp_gsi_diagnostics": {
+        "mean_anti_slip_score": 0.7558600959763804,
+        "mean_duration_score": 0.33843992248061877,
+        "mean_hold_score": 0.20455287929125243,
+        "mean_kinematic_stability_score": 0.05351032437012617,
+        "mean_legacy_gsi": 0.3464732089348715,
+        "mean_lift_score": 0.2882613510520487,
+        "mean_motion_stability_score": 0.20219920816885556
+      }
+    },
+    "tool": {
+      "tool_selection_accuracy": null,
+      "tool_switch_success_rate": null,
+      "tool_switch_total": 0,
+      "tool_switch_successful": 0
+    },
+    "runtime": {
+      "avg_steps": 1647,
+      "error_count": 1
+    }
+  }
+}

gpt6_astra_1x_icl/per_task_eval/08_frypan_stand_pour/Full/summary.json

@@ -0,0 +1,307 @@
+{
+  "metric_schema_version": "core-v1.8-policy-step-query-split",
+  "metadata": {
+    "policy_name": "gpt-6-astra",
+    "base_seed": 200800000,
+    "seed_policy": "episode_seed = base_seed + episode_index - 1",
+    "randomization_recorded": true,
+    "n_episodes": 10,
+    "successes": 0,
+    "error_count": 1,
+    "evaluation_protocol": "reach_and_stop",
+    "early_stop": true,
+    "success_definition": "terminal.raw_condition holds for dwell_time_s",
+    "measurement_window": "episode_start_to_stable_success_or_failure_budget",
+    "max_episode_steps": 3432,
+    "dwell_time_s": 0.5,
+    "field_semantics": {
+      "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
+      "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
+      "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
+      "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
+      "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
+      "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
+      "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
+      "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
+      "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
+      "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
+      "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
+      "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
+      "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
+      "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
+      "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
+      "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
+      "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
+      "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
+      "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
+      "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
+      "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
+      "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
+      "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
+      "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
+      "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
+    }
+  },
+  "core_metrics": {
+    "completion": {
+      "success_rate": 0.0,
+      "success_rate_ci": {
+        "mean": 0.0,
+        "ci_low": 0.0,
+        "ci_high": 0.0
+      }
+    },
+    "progress": {
+      "latched_stage_completion_rate": 0.15
+    },
+    "efficiency": {
+      "avg_time_to_success_s": null
+    },
+    "safety": {
+      "safe_success_rate": 0.0,
+      "hard_violation_rate": 0.3,
+      "drop_rate": 0.0,
+      "high_speed_violation_rate": 0.3,
+      "safety_violation_step_rate": 0.00032802528776763884
+    },
+    "robustness": {}
+  },
+  "diagnostic_metrics": {
+    "completion": {
+      "ever_instant_success_rate": 0.0,
+      "at_end_success_rate": 0.0,
+      "task_family_success_rate": {
+        "pour_tilt": 0.0
+      }
+    },
+    "progress": {
+      "current_stage_completion_rate": 0.0,
+      "current_chain_depth_progress_score": 0.0,
+      "latched_chain_depth_progress_score": 0.3,
+      "avg_current_chain_depth": 0,
+      "avg_latched_chain_depth": 0.3
+    },
+    "efficiency": {
+      "task_efficiency": null,
+      "expert_time_s": 38.13333333333333,
+      "expert_time_step": 2288,
+      "avg_steps_to_success": null,
+      "avg_policy_steps_to_success": null,
+      "avg_policy_queries_to_success": null
+    },
+    "safety": {
+      "safety_violation_events_per_step": 8.946144211844694e-05,
+      "violation_counts": {
+        "high_speed_obj_172_soy_sauce_1": 1,
+        "high_speed_obj_182_bread_2": 2
+      },
+      "robot_constraint_violation_rate": 1.0,
+      "finger_joint_limit_saturation_rate": 1.0,
+      "joint_limit_active_step_rate": 0.10458042583646449,
+      "finger_joint_limit_saturation_step_rate": 0.0571062205522753,
+      "max_joint_limit_excess_rad": 2.122424602508545,
+      "robot_constraint_diagnostics": {
+        "joint_limit_soft_counts": {
+          "left_hand_Finger_Spread": 285,
+          "left_hand_Index_Finger_Distal": 422,
+          "left_hand_Index_Finger_Proximal": 629,
+          "left_hand_Middle_Finger_Distal": 208,
+          "left_hand_Middle_Finger_Proximal": 474,
+          "left_hand_Pinky": 261,
+          "left_hand_Ring_Finger": 276,
+          "left_hand_Thumb_Flexion": 1534,
+          "left_hand_Thumb_Opposition": 238,
+          "left_hand_index_spread": 265,
+          "left_hand_j12": 214,
+          "left_hand_j13": 210,
+          "left_hand_j14": 14,
+          "left_hand_j15": 3,
+          "left_hand_j16": 6,
+          "left_hand_j17": 41,
+          "left_hand_j3": 574,
+          "left_hand_j4": 2,
+          "left_hand_j5": 402,
+          "left_hand_ring_spread": 388,
+          "right_hand_Index_Finger_Distal": 62,
+          "right_hand_Index_Finger_Proximal": 64,
+          "right_hand_Middle_Finger_Distal": 105,
+          "right_hand_Middle_Finger_Proximal": 299,
+          "right_hand_Pinky": 76,
+          "right_hand_Ring_Finger": 181,
+          "right_hand_Thumb_Flexion": 105,
+          "right_hand_j12": 15,
+          "right_hand_j13": 92,
+          "right_hand_j15": 101,
+          "right_hand_j3": 159
+        },
+        "joint_limit_hard_counts": {
+          "left_hand_Finger_Spread": 19,
+          "left_hand_Index_Finger_Distal": 13,
+          "left_hand_Index_Finger_Proximal": 30,
+          "left_hand_Middle_Finger_Distal": 11,
+          "left_hand_Middle_Finger_Proximal": 22,
+          "left_hand_Pinky": 10,
+          "left_hand_Ring_Finger": 16,
+          "left_hand_Thumb_Flexion": 35,
+          "left_hand_Thumb_Opposition": 18,
+          "left_hand_index_spread": 47,
+          "left_hand_j12": 10,
+          "left_hand_j13": 12,
+          "left_hand_j14": 1,
+          "left_hand_j15": 1,
+          "left_hand_j16": 3,
+          "left_hand_j17": 2,
+          "left_hand_j3": 21,
+          "left_hand_j4": 3,
+          "left_hand_j5": 62,
+          "left_hand_ring_spread": 12,
+          "right_hand_Index_Finger_Distal": 25,
+          "right_hand_Index_Finger_Proximal": 18,
+          "right_hand_Middle_Finger_Distal": 27,
+          "right_hand_Middle_Finger_Proximal": 48,
+          "right_hand_Pinky": 22,
+          "right_hand_Ring_Finger": 25,
+          "right_hand_Thumb_Flexion": 18,
+          "right_hand_j12": 16,
+          "right_hand_j13": 26,
+          "right_hand_j14": 2,
+          "right_hand_j15": 3,
+          "right_hand_j3": 19,
+          "right_hand_j5": 1
+        },
+        "joint_limit_hard_episode_events": 180,
+        "finger_joint_limit_saturation_counts": {
+          "left_hand_Finger_Spread": 198,
+          "left_hand_Index_Finger_Distal": 316,
+          "left_hand_Index_Finger_Proximal": 442,
+          "left_hand_Middle_Finger_Distal": 156,
+          "left_hand_Middle_Finger_Proximal": 314,
+          "left_hand_Pinky": 150,
+          "left_hand_Ring_Finger": 88,
+          "left_hand_Thumb_Flexion": 647,
+          "left_hand_Thumb_Opposition": 147,
+          "left_hand_index_spread": 86,
+          "left_hand_j12": 159,
+          "left_hand_j13": 134,
+          "left_hand_j14": 8,
+          "left_hand_j17": 22,
+          "left_hand_j3": 363,
+          "left_hand_j5": 138,
+          "left_hand_ring_spread": 300,
+          "right_hand_Middle_Finger_Distal": 62,
+          "right_hand_Middle_Finger_Proximal": 56,
+          "right_hand_Ring_Finger": 51,
+          "right_hand_Thumb_Flexion": 17,
+          "right_hand_j15": 75,
+          "right_hand_j3": 44
+        },
+        "joint_limit_max_excess_by_joint": {
+          "left_hand_Finger_Spread": 0.8473438620567322,
+          "left_hand_Index_Finger_Distal": 1.539358377456665,
+          "left_hand_Index_Finger_Proximal": 1.3137761354446411,
+          "left_hand_Middle_Finger_Distal": 1.0814626216888428,
+          "left_hand_Middle_Finger_Proximal": 1.446799397468567,
+          "left_hand_Pinky": 0.8448187708854675,
+          "left_hand_Ring_Finger": 0.5920495986938477,
+          "left_hand_Thumb_Flexion": 2.122424602508545,
+          "left_hand_Thumb_Opposition": 0.7682660818099976,
+          "left_hand_index_spread": 0.7858690023422241,
+          "left_hand_j12": 1.524500846862793,
+          "left_hand_j13": 1.586573600769043,
+          "left_hand_j14": 0.7159442901611328,
+          "left_hand_j15": 0.2375551015138626,
+          "left_hand_j16": 0.014156644232571125,
+          "left_hand_j17": 0.6165305376052856,
+          "left_hand_j3": 1.3230681419372559,
+          "left_hand_j4": 0.03332408145070076,
+          "left_hand_j5": 0.9578635692596436,
+          "left_hand_ring_spread": 1.350319266319275,
+          "right_hand_Finger_Spread": 0.002514712745323777,
+          "right_hand_Index_Finger_Distal": 0.22093699872493744,
+          "right_hand_Index_Finger_Proximal": 0.10431328415870667,
+          "right_hand_Middle_Finger_Distal": 0.6995282173156738,
+          "right_hand_Middle_Finger_Proximal": 0.4700015187263489,
+          "right_hand_Pinky": 0.11842221766710281,
+          "right_hand_Ring_Finger": 0.4743174910545349,
+          "right_hand_Thumb_Flexion": 0.41061514616012573,
+          "right_hand_Thumb_Opposition": 0.000317111931508407,
+          "right_hand_index_spread": 0.0016010317485779524,
+          "right_hand_j12": 0.041416678577661514,
+          "right_hand_j13": 0.14545877277851105,
+          "right_hand_j14": 0.009076248854398727,
+          "right_hand_j15": 0.4486370384693146,
+          "right_hand_j16": 0.00037610915023833513,
+          "right_hand_j17": 0.0011840634979307652,
+          "right_hand_j3": 0.9112940430641174,
+          "right_hand_j4": 3.488648508209735e-05,
+          "right_hand_j5": 0.006437512114644051,
+          "right_hand_ring_spread": 0.0024567407090216875
+        },
+        "coverage": {
+          "hard_limit_source": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          },
+          "joint_role_manifest": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          }
+        },
+        "hard_limit_sources": {
+          "runtime.joint_pos_limits": 10
+        },
+        "soft_limit_sources": {
+          "runtime.soft_joint_pos_limits": 10
+        },
+        "unknown_runtime_joint_names": [],
+        "baseline_checks": {
+          "post_home": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          },
+          "post_reset": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          }
+        }
+      }
+    },
+    "robot_motion": {
+      "joint_acceleration_rms": 17.57303284739676,
+      "joint_effort_rms": 30.554318080512665,
+      "joint_jerk_rms": 1740.172251966746,
+      "joint_velocity_rms": 0.41925120570845065
+    },
+    "grasp": {
+      "kinematic_grasp_stability_index": 0.5465015212467665,
+      "mean_kinematic_grasp_stability": 0.4917737156866544,
+      "per_object_best_kinematic_grasp_stability": {
+        "obj_146_olive_oil_1": 0.2118868034065897,
+        "obj_172_soy_sauce_1": 0.5428749003249655
+      },
+      "grasp_gsi_diagnostics": {
+        "mean_anti_slip_score": 0.9057135985707381,
+        "mean_duration_score": 0.1296814891457749,
+        "mean_hold_score": 0.11354288238216828,
+        "mean_kinematic_stability_score": 0.03007668700669158,
+        "mean_legacy_gsi": 0.2725563161287371,
+        "mean_lift_score": 0.13398981970410542,
+        "mean_motion_stability_score": 0.10237528498868878
+      }
+    },
+    "tool": {
+      "tool_selection_accuracy": 0.7,
+      "tool_switch_success_rate": 0.0,
+      "tool_switch_total": 18,
+      "tool_switch_successful": 0
+    },
+    "runtime": {
+      "avg_steps": 3353.4,
+      "error_count": 1
+    }
+  }
+}

gpt6_astra_1x_icl/per_task_eval/09_cleaner_moisturizer_box_loading/Full/summary.json

@@ -0,0 +1,237 @@
+{
+  "metric_schema_version": "core-v1.8-policy-step-query-split",
+  "metadata": {
+    "policy_name": "gpt-6-astra",
+    "base_seed": 200900000,
+    "seed_policy": "episode_seed = base_seed + episode_index - 1",
+    "randomization_recorded": true,
+    "n_episodes": 10,
+    "successes": 2,
+    "error_count": 0,
+    "evaluation_protocol": "reach_and_stop",
+    "early_stop": true,
+    "success_definition": "terminal.raw_condition holds for dwell_time_s",
+    "measurement_window": "episode_start_to_stable_success_or_failure_budget",
+    "max_episode_steps": 1926,
+    "dwell_time_s": 0.5,
+    "field_semantics": {
+      "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
+      "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
+      "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
+      "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
+      "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
+      "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
+      "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
+      "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
+      "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
+      "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
+      "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
+      "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
+      "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
+      "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
+      "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
+      "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
+      "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
+      "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
+      "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
+      "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
+      "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
+      "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
+      "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
+      "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
+      "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
+    }
+  },
+  "core_metrics": {
+    "completion": {
+      "success_rate": 0.2,
+      "success_rate_ci": {
+        "mean": 0.2,
+        "ci_low": 0.0,
+        "ci_high": 0.5
+      }
+    },
+    "progress": {
+      "latched_stage_completion_rate": 0.0
+    },
+    "efficiency": {
+      "avg_time_to_success_s": 29.258333333332953
+    },
+    "safety": {
+      "safe_success_rate": 0.2,
+      "hard_violation_rate": 0.3,
+      "drop_rate": 0.3,
+      "high_speed_violation_rate": 0.1,
+      "safety_violation_step_rate": 0.0005285691632750145
+    },
+    "robustness": {}
+  },
+  "diagnostic_metrics": {
+    "completion": {
+      "ever_instant_success_rate": 0.2,
+      "at_end_success_rate": 0.2,
+      "task_family_success_rate": {
+        "rigid_pick_place": 0.2
+      }
+    },
+    "progress": {
+      "current_stage_completion_rate": 0.0,
+      "current_chain_depth_progress_score": 0.0,
+      "latched_chain_depth_progress_score": 0.0,
+      "avg_current_chain_depth": 0,
+      "avg_latched_chain_depth": 0
+    },
+    "efficiency": {
+      "task_efficiency": 0.7309270866394292,
+      "expert_time_s": 21.383333333333333,
+      "expert_time_step": 1283,
+      "avg_steps_to_success": 1755.5,
+      "avg_policy_steps_to_success": 585.5,
+      "avg_policy_queries_to_success": 22
+    },
+    "safety": {
+      "safety_violation_events_per_step": 0.00021142766531000582,
+      "violation_counts": {
+        "drop_to_table_obj_200_cleaner_2": 3,
+        "high_speed_obj_154_wooden_box_1": 1
+      },
+      "robot_constraint_violation_rate": 1.0,
+      "finger_joint_limit_saturation_rate": 0.9,
+      "joint_limit_active_step_rate": 0.08853533484856493,
+      "finger_joint_limit_saturation_step_rate": 0.04635551561921877,
+      "max_joint_limit_excess_rad": 1.7215934991836548,
+      "robot_constraint_diagnostics": {
+        "joint_limit_soft_counts": {
+          "leap_l_10": 363,
+          "leap_l_11": 13,
+          "leap_l_12": 8,
+          "leap_l_13": 8,
+          "leap_l_2": 425,
+          "leap_l_5": 71,
+          "leap_l_6": 58,
+          "leap_l_9": 59,
+          "multi_leap_r_10": 655,
+          "multi_leap_r_11": 19,
+          "multi_leap_r_2": 311,
+          "multi_leap_r_5": 9,
+          "multi_leap_r_6": 259,
+          "multi_leap_r_9": 33
+        },
+        "joint_limit_hard_counts": {
+          "leap_l_10": 18,
+          "leap_l_11": 12,
+          "leap_l_12": 1,
+          "leap_l_13": 9,
+          "leap_l_2": 98,
+          "leap_l_3": 2,
+          "leap_l_5": 3,
+          "leap_l_6": 7,
+          "leap_l_9": 4,
+          "multi_leap_r_10": 23,
+          "multi_leap_r_11": 1,
+          "multi_leap_r_2": 18,
+          "multi_leap_r_5": 2,
+          "multi_leap_r_6": 7,
+          "multi_leap_r_9": 9
+        },
+        "joint_limit_hard_episode_events": 135,
+        "finger_joint_limit_saturation_counts": {
+          "leap_l_10": 208,
+          "leap_l_12": 2,
+          "leap_l_2": 149,
+          "leap_l_5": 2,
+          "leap_l_6": 13,
+          "leap_l_9": 3,
+          "multi_leap_r_10": 435,
+          "multi_leap_r_2": 91,
+          "multi_leap_r_6": 157
+        },
+        "joint_limit_max_excess_by_joint": {
+          "joint7": 0.002178192138671875,
+          "leap_l_10": 1.7215934991836548,
+          "leap_l_11": 0.04692891240119934,
+          "leap_l_12": 0.2714962959289551,
+          "leap_l_13": 0.016356706619262695,
+          "leap_l_15": 0.0019141435623168945,
+          "leap_l_2": 1.59174382686615,
+          "leap_l_3": 0.007471174001693726,
+          "leap_l_5": 0.2800409197807312,
+          "leap_l_6": 0.32280123233795166,
+          "leap_l_9": 0.2903289794921875,
+          "multi_leap_r_1": 0.004976987838745117,
+          "multi_leap_r_10": 1.1505019664764404,
+          "multi_leap_r_11": 0.1929008960723877,
+          "multi_leap_r_2": 1.406436800956726,
+          "multi_leap_r_3": 0.00039336085319519043,
+          "multi_leap_r_5": 0.06958889961242676,
+          "multi_leap_r_6": 0.7499661445617676,
+          "multi_leap_r_9": 0.14962145686149597
+        },
+        "coverage": {
+          "hard_limit_source": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          },
+          "joint_role_manifest": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          }
+        },
+        "hard_limit_sources": {
+          "runtime.joint_pos_limits": 10
+        },
+        "soft_limit_sources": {
+          "runtime.soft_joint_pos_limits": 10
+        },
+        "unknown_runtime_joint_names": [],
+        "baseline_checks": {
+          "post_home": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          },
+          "post_reset": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          }
+        }
+      }
+    },
+    "robot_motion": {
+      "joint_acceleration_rms": 23.810451645998292,
+      "joint_effort_rms": 62.29403347460127,
+      "joint_jerk_rms": 2366.79009938777,
+      "joint_velocity_rms": 0.7054899200340649
+    },
+    "grasp": {
+      "kinematic_grasp_stability_index": 0.8618147740430608,
+      "mean_kinematic_grasp_stability": 0.6386741189676691,
+      "per_object_best_kinematic_grasp_stability": {
+        "obj_200_cleaner_2": 0.8438837776737047,
+        "obj_209_soap_3": 0.2996346957430446
+      },
+      "grasp_gsi_diagnostics": {
+        "mean_anti_slip_score": 0.7500015372419915,
+        "mean_duration_score": 0.34143576379041307,
+        "mean_hold_score": 0.2573252645977591,
+        "mean_kinematic_stability_score": 0.07110006258989891,
+        "mean_legacy_gsi": 0.36872778290899044,
+        "mean_lift_score": 0.29982892424063234,
+        "mean_motion_stability_score": 0.23394034772178426
+      }
+    },
+    "tool": {
+      "tool_selection_accuracy": null,
+      "tool_switch_success_rate": null,
+      "tool_switch_total": 0,
+      "tool_switch_successful": 0
+    },
+    "runtime": {
+      "avg_steps": 1891.9,
+      "error_count": 0
+    }
+  }
+}

gpt6_astra_1x_icl/per_task_eval/12_screwdriver_box_and_hammer/Full/summary.json

@@ -0,0 +1,247 @@
+{
+  "metric_schema_version": "core-v1.8-policy-step-query-split",
+  "metadata": {
+    "policy_name": "gpt-6-astra",
+    "base_seed": 201200000,
+    "seed_policy": "episode_seed = base_seed + episode_index - 1",
+    "randomization_recorded": true,
+    "n_episodes": 10,
+    "successes": 4,
+    "error_count": 1,
+    "evaluation_protocol": "reach_and_stop",
+    "early_stop": true,
+    "success_definition": "terminal.raw_condition holds for dwell_time_s",
+    "measurement_window": "episode_start_to_stable_success_or_failure_budget",
+    "max_episode_steps": 3447,
+    "dwell_time_s": 0.5,
+    "field_semantics": {
+      "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
+      "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
+      "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
+      "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
+      "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
+      "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
+      "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
+      "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
+      "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
+      "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
+      "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
+      "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
+      "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
+      "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
+      "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
+      "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
+      "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
+      "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
+      "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
+      "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
+      "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
+      "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
+      "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
+      "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
+      "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
+    }
+  },
+  "core_metrics": {
+    "completion": {
+      "success_rate": 0.4,
+      "success_rate_ci": {
+        "mean": 0.4,
+        "ci_low": 0.1,
+        "ci_high": 0.7
+      }
+    },
+    "progress": {
+      "latched_stage_completion_rate": 0.85
+    },
+    "efficiency": {
+      "avg_time_to_success_s": 52.94166666666494
+    },
+    "safety": {
+      "safe_success_rate": 0.4,
+      "hard_violation_rate": 0.2,
+      "drop_rate": 0.2,
+      "high_speed_violation_rate": 0.1,
+      "safety_violation_step_rate": 0.013190515542559279
+    },
+    "robustness": {}
+  },
+  "diagnostic_metrics": {
+    "completion": {
+      "ever_instant_success_rate": 0.8,
+      "at_end_success_rate": 0.5,
+      "task_family_success_rate": {
+        "tool_precision_place": 0.4
+      }
+    },
+    "progress": {
+      "current_stage_completion_rate": 0.8,
+      "current_chain_depth_progress_score": 0.8,
+      "latched_chain_depth_progress_score": 0.85,
+      "avg_current_chain_depth": 3.2,
+      "avg_latched_chain_depth": 3.4
+    },
+    "efficiency": {
+      "task_efficiency": 0.7246924442871527,
+      "expert_time_s": 38.28333333333333,
+      "expert_time_step": 2297,
+      "avg_steps_to_success": 3176.5,
+      "avg_policy_steps_to_success": 1059.25,
+      "avg_policy_queries_to_success": 19.25
+    },
+    "safety": {
+      "safety_violation_events_per_step": 0.00015892187400673829,
+      "violation_counts": {
+        "drop_obj_044_flat_screwdriver_2": 1,
+        "drop_to_table_obj_044_flat_screwdriver_2": 1,
+        "drop_to_table_obj_048_hammer_3": 1,
+        "high_speed_obj_044_flat_screwdriver_2": 1,
+        "high_speed_obj_154_wooden_box_1": 1
+      },
+      "robot_constraint_violation_rate": 1.0,
+      "finger_joint_limit_saturation_rate": 0.9,
+      "joint_limit_active_step_rate": 0.11515478990528256,
+      "finger_joint_limit_saturation_step_rate": 0.07513826203038586,
+      "max_joint_limit_excess_rad": 5.223601818084717,
+      "robot_constraint_diagnostics": {
+        "joint_limit_soft_counts": {
+          "left_index_1_joint": 274,
+          "left_little_1_joint": 317,
+          "left_middle_1_joint": 246,
+          "left_ring_1_joint": 306,
+          "left_thumb_2_joint": 870,
+          "left_thumb_3_joint": 20,
+          "left_thumb_4_joint": 5,
+          "right_index_1_joint": 936,
+          "right_little_1_joint": 77,
+          "right_middle_1_joint": 494,
+          "right_ring_1_joint": 482,
+          "right_thumb_2_joint": 471,
+          "right_thumb_3_joint": 119,
+          "right_thumb_4_joint": 10
+        },
+        "joint_limit_hard_counts": {
+          "left_index_1_joint": 19,
+          "left_little_1_joint": 8,
+          "left_middle_1_joint": 9,
+          "left_ring_1_joint": 15,
+          "left_thumb_2_joint": 25,
+          "left_thumb_3_joint": 1,
+          "left_thumb_4_joint": 5,
+          "right_index_1_joint": 32,
+          "right_little_1_joint": 4,
+          "right_middle_1_joint": 15,
+          "right_ring_1_joint": 9,
+          "right_thumb_2_joint": 7,
+          "right_thumb_3_joint": 6,
+          "right_thumb_4_joint": 4
+        },
+        "joint_limit_hard_episode_events": 91,
+        "finger_joint_limit_saturation_counts": {
+          "left_index_1_joint": 62,
+          "left_little_1_joint": 223,
+          "left_middle_1_joint": 102,
+          "left_ring_1_joint": 109,
+          "left_thumb_2_joint": 531,
+          "right_index_1_joint": 625,
+          "right_little_1_joint": 68,
+          "right_middle_1_joint": 238,
+          "right_ring_1_joint": 341,
+          "right_thumb_2_joint": 329,
+          "right_thumb_3_joint": 38
+        },
+        "joint_limit_max_excess_by_joint": {
+          "left_index_1_joint": 0.27620208263397217,
+          "left_index_2_joint": 1.139794562732277e-06,
+          "left_little_1_joint": 5.223601818084717,
+          "left_little_2_joint": 0.0018775563221424818,
+          "left_middle_1_joint": 0.32757097482681274,
+          "left_middle_2_joint": 1.1005318810930476e-05,
+          "left_ring_1_joint": 0.7387237548828125,
+          "left_ring_2_joint": 0.004270508419722319,
+          "left_thumb_1_joint": 2.2825703126727603e-05,
+          "left_thumb_2_joint": 0.4792882204055786,
+          "left_thumb_3_joint": 0.09909138828516006,
+          "left_thumb_4_joint": 0.03685218095779419,
+          "right_index_1_joint": 0.9538151621818542,
+          "right_index_2_joint": 7.616848506586393e-06,
+          "right_little_1_joint": 1.3297919034957886,
+          "right_little_2_joint": 3.0642575438832864e-05,
+          "right_middle_1_joint": 0.30515244603157043,
+          "right_middle_2_joint": 5.51865559828002e-05,
+          "right_ring_1_joint": 0.2909194231033325,
+          "right_ring_2_joint": 5.6689455959713086e-05,
+          "right_thumb_1_joint": 0.0003198256599716842,
+          "right_thumb_2_joint": 0.3879449963569641,
+          "right_thumb_3_joint": 0.40510305762290955,
+          "right_thumb_4_joint": 0.05127860605716705
+        },
+        "coverage": {
+          "hard_limit_source": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          },
+          "joint_role_manifest": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          }
+        },
+        "hard_limit_sources": {
+          "runtime.joint_pos_limits": 10
+        },
+        "soft_limit_sources": {
+          "runtime.soft_joint_pos_limits": 10
+        },
+        "unknown_runtime_joint_names": [],
+        "baseline_checks": {
+          "post_home": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          },
+          "post_reset": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          }
+        }
+      }
+    },
+    "robot_motion": {
+      "joint_acceleration_rms": 57.26923484351931,
+      "joint_effort_rms": 45.39985160217687,
+      "joint_jerk_rms": 5931.196605298128,
+      "joint_velocity_rms": 1.196371621151782
+    },
+    "grasp": {
+      "kinematic_grasp_stability_index": 0.9977184232259461,
+      "mean_kinematic_grasp_stability": 0.45023017764548745,
+      "per_object_best_kinematic_grasp_stability": {
+        "obj_043_phillips_screwdriver_1": 0.6986435163758,
+        "obj_044_flat_screwdriver_2": 0.9947615671029998,
+        "obj_048_hammer_3": 0.4458844659441989
+      },
+      "grasp_gsi_diagnostics": {
+        "mean_anti_slip_score": 0.7957915891988776,
+        "mean_duration_score": 0.3958068690583115,
+        "mean_hold_score": 0.3459987049394094,
+        "mean_kinematic_stability_score": 0.15518423436059983,
+        "mean_legacy_gsi": 0.45035378089465894,
+        "mean_lift_score": 0.4747552273531635,
+        "mean_motion_stability_score": 0.33414277492197964
+      }
+    },
+    "tool": {
+      "tool_selection_accuracy": 0.0,
+      "tool_switch_success_rate": 0.0,
+      "tool_switch_total": 40,
+      "tool_switch_successful": 0
+    },
+    "runtime": {
+      "avg_steps": 3146.2,
+      "error_count": 1
+    }
+  }
+}

gpt6_astra_1x_icl/per_task_eval/21_condiment_box_loading/Full/summary.json

@@ -0,0 +1,250 @@
+{
+  "metric_schema_version": "core-v1.8-policy-step-query-split",
+  "metadata": {
+    "policy_name": "gpt-6-astra",
+    "base_seed": 202100000,
+    "seed_policy": "episode_seed = base_seed + episode_index - 1",
+    "randomization_recorded": true,
+    "n_episodes": 10,
+    "successes": 2,
+    "error_count": 1,
+    "evaluation_protocol": "reach_and_stop",
+    "early_stop": true,
+    "success_definition": "terminal.raw_condition holds for dwell_time_s",
+    "measurement_window": "episode_start_to_stable_success_or_failure_budget",
+    "max_episode_steps": 2481,
+    "dwell_time_s": 0.5,
+    "field_semantics": {
+      "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
+      "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
+      "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
+      "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
+      "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
+      "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
+      "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
+      "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
+      "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
+      "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
+      "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
+      "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
+      "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
+      "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
+      "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
+      "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
+      "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
+      "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
+      "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
+      "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
+      "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
+      "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
+      "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
+      "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
+      "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
+    }
+  },
+  "core_metrics": {
+    "completion": {
+      "success_rate": 0.2,
+      "success_rate_ci": {
+        "mean": 0.2,
+        "ci_low": 0.0,
+        "ci_high": 0.5
+      }
+    },
+    "progress": {
+      "latched_stage_completion_rate": 0.23333333333333334
+    },
+    "efficiency": {
+      "avg_time_to_success_s": 40.49166666666565
+    },
+    "safety": {
+      "safe_success_rate": 0.2,
+      "hard_violation_rate": 0.5,
+      "drop_rate": 0.5,
+      "high_speed_violation_rate": 0.0,
+      "safety_violation_step_rate": 0.0002968365702654567
+    },
+    "robustness": {}
+  },
+  "diagnostic_metrics": {
+    "completion": {
+      "ever_instant_success_rate": 0.2,
+      "at_end_success_rate": 0.2,
+      "task_family_success_rate": {
+        "rigid_pick_place": 0.2
+      }
+    },
+    "progress": {
+      "current_stage_completion_rate": 0.23333333333333334,
+      "current_chain_depth_progress_score": 0.25,
+      "latched_chain_depth_progress_score": 0.25,
+      "avg_current_chain_depth": 0.5,
+      "avg_latched_chain_depth": 0.5
+    },
+    "efficiency": {
+      "task_efficiency": 0.6810269994526049,
+      "expert_time_s": 27.566666666666666,
+      "expert_time_step": 1654,
+      "avg_steps_to_success": 2429.5,
+      "avg_policy_steps_to_success": 810,
+      "avg_policy_queries_to_success": 22.5
+    },
+    "safety": {
+      "safety_violation_events_per_step": 0.0002968365702654567,
+      "violation_counts": {
+        "drop_to_table_obj_171_msg_4": 5,
+        "drop_to_table_obj_172_soy_sauce_2": 2
+      },
+      "robot_constraint_violation_rate": 0.9,
+      "finger_joint_limit_saturation_rate": 0.7,
+      "joint_limit_active_step_rate": 0.03290645407514206,
+      "finger_joint_limit_saturation_step_rate": 0.020693749469934696,
+      "max_joint_limit_excess_rad": 2.4376306533813477,
+      "robot_constraint_diagnostics": {
+        "joint_limit_soft_counts": {
+          "left_finger2_joint4": 47,
+          "left_finger3_joint4": 55,
+          "left_finger4_joint2": 9,
+          "left_finger4_joint4": 55,
+          "left_finger5_joint4": 52,
+          "right_finger1_joint1": 216,
+          "right_finger1_joint2": 244,
+          "right_finger1_joint3": 222,
+          "right_finger2_joint1": 107,
+          "right_finger3_joint1": 53,
+          "right_finger3_joint4": 72,
+          "right_finger4_joint1": 13,
+          "right_finger4_joint2": 20,
+          "right_finger4_joint4": 46,
+          "right_finger5_joint1": 14,
+          "right_finger5_joint2": 27,
+          "right_finger5_joint4": 28
+        },
+        "joint_limit_hard_counts": {
+          "left_finger2_joint4": 1,
+          "left_finger3_joint4": 1,
+          "left_finger4_joint2": 1,
+          "left_finger4_joint4": 1,
+          "left_finger5_joint4": 1,
+          "right_finger1_joint1": 5,
+          "right_finger1_joint2": 50,
+          "right_finger1_joint3": 11,
+          "right_finger2_joint1": 7,
+          "right_finger3_joint1": 2,
+          "right_finger3_joint4": 2,
+          "right_finger4_joint1": 3,
+          "right_finger4_joint2": 6,
+          "right_finger4_joint4": 3,
+          "right_finger5_joint1": 1,
+          "right_finger5_joint2": 5,
+          "right_finger5_joint4": 1
+        },
+        "joint_limit_hard_episode_events": 64,
+        "finger_joint_limit_saturation_counts": {
+          "left_finger2_joint4": 45,
+          "left_finger3_joint4": 53,
+          "left_finger4_joint4": 53,
+          "left_finger5_joint4": 48,
+          "right_finger1_joint1": 174,
+          "right_finger1_joint2": 137,
+          "right_finger1_joint3": 78,
+          "right_finger2_joint1": 47,
+          "right_finger3_joint1": 12,
+          "right_finger3_joint4": 65,
+          "right_finger4_joint4": 36,
+          "right_finger5_joint2": 10,
+          "right_finger5_joint4": 24
+        },
+        "joint_limit_max_excess_by_joint": {
+          "left_finger1_joint2": 0.00391620397567749,
+          "left_finger2_joint1": 0.0003159046173095703,
+          "left_finger2_joint2": 0.00010144710540771484,
+          "left_finger2_joint4": 1.8637301921844482,
+          "left_finger3_joint4": 2.3013296127319336,
+          "left_finger4_joint2": 0.11612790822982788,
+          "left_finger4_joint4": 2.4376306533813477,
+          "left_finger5_joint2": 0.0035106539726257324,
+          "left_finger5_joint4": 2.2746286392211914,
+          "right_finger1_joint1": 1.5812128782272339,
+          "right_finger1_joint2": 1.083169937133789,
+          "right_finger1_joint3": 0.7230508923530579,
+          "right_finger2_joint1": 0.3803626298904419,
+          "right_finger2_joint2": 0.0009592175483703613,
+          "right_finger3_joint1": 0.1703876256942749,
+          "right_finger3_joint4": 1.728808045387268,
+          "right_finger4_joint1": 0.018409013748168945,
+          "right_finger4_joint2": 0.20231959223747253,
+          "right_finger4_joint4": 1.2384870052337646,
+          "right_finger5_joint1": 0.23881781101226807,
+          "right_finger5_joint2": 0.40025654435157776,
+          "right_finger5_joint4": 0.6091189384460449
+        },
+        "coverage": {
+          "hard_limit_source": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          },
+          "joint_role_manifest": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          }
+        },
+        "hard_limit_sources": {
+          "runtime.joint_pos_limits": 10
+        },
+        "soft_limit_sources": {
+          "runtime.soft_joint_pos_limits": 10
+        },
+        "unknown_runtime_joint_names": [],
+        "baseline_checks": {
+          "post_home": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          },
+          "post_reset": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          }
+        }
+      }
+    },
+    "robot_motion": {
+      "joint_acceleration_rms": 21.106462569404204,
+      "joint_effort_rms": 34.15535031136554,
+      "joint_jerk_rms": 2137.8700609128655,
+      "joint_velocity_rms": 0.48582973757766323
+    },
+    "grasp": {
+      "kinematic_grasp_stability_index": 0.875453319366878,
+      "mean_kinematic_grasp_stability": 0.40674579740861994,
+      "per_object_best_kinematic_grasp_stability": {
+        "obj_171_msg_4": 0.8669026389450034,
+        "obj_172_soy_sauce_2": 0.3484896544064934,
+        "obj_173_vinegar_3": 0.3677465198821677
+      },
+      "grasp_gsi_diagnostics": {
+        "mean_anti_slip_score": 0.7685570102324817,
+        "mean_duration_score": 0.3333672806617162,
+        "mean_hold_score": 0.24902185413207156,
+        "mean_kinematic_stability_score": 0.06303769278693115,
+        "mean_legacy_gsi": 0.36836577022870376,
+        "mean_lift_score": 0.32760037535194875,
+        "mean_motion_stability_score": 0.23132505296156392
+      }
+    },
+    "tool": {
+      "tool_selection_accuracy": null,
+      "tool_switch_success_rate": null,
+      "tool_switch_total": 0,
+      "tool_switch_successful": 0
+    },
+    "runtime": {
+      "avg_steps": 2358.2,
+      "error_count": 1
+    }
+  }
+}

gpt6_astra_1x_icl/per_task_eval/22_tool_box_loading/Full/summary.json

@@ -0,0 +1,210 @@
+{
+  "metric_schema_version": "core-v1.8-policy-step-query-split",
+  "metadata": {
+    "policy_name": "gpt-6-astra",
+    "base_seed": 202200000,
+    "seed_policy": "episode_seed = base_seed + episode_index - 1",
+    "randomization_recorded": true,
+    "n_episodes": 10,
+    "successes": 5,
+    "error_count": 0,
+    "evaluation_protocol": "reach_and_stop",
+    "early_stop": true,
+    "success_definition": "terminal.raw_condition holds for dwell_time_s",
+    "measurement_window": "episode_start_to_stable_success_or_failure_budget",
+    "max_episode_steps": 2043,
+    "dwell_time_s": 0.5,
+    "field_semantics": {
+      "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
+      "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
+      "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
+      "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
+      "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
+      "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
+      "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
+      "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
+      "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
+      "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
+      "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
+      "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
+      "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
+      "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
+      "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
+      "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
+      "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
+      "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
+      "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
+      "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
+      "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
+      "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
+      "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
+      "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
+      "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
+    }
+  },
+  "core_metrics": {
+    "completion": {
+      "success_rate": 0.5,
+      "success_rate_ci": {
+        "mean": 0.5,
+        "ci_low": 0.2,
+        "ci_high": 0.8
+      }
+    },
+    "progress": {
+      "latched_stage_completion_rate": 0.725
+    },
+    "efficiency": {
+      "avg_time_to_success_s": 32.91999999999941
+    },
+    "safety": {
+      "safe_success_rate": 0.5,
+      "hard_violation_rate": 0.3,
+      "drop_rate": 0.3,
+      "high_speed_violation_rate": 0.1,
+      "safety_violation_step_rate": 0.0004479617739286248
+    },
+    "robustness": {}
+  },
+  "diagnostic_metrics": {
+    "completion": {
+      "ever_instant_success_rate": 0.7,
+      "at_end_success_rate": 0.6,
+      "task_family_success_rate": {
+        "rigid_assembly": 0.5
+      }
+    },
+    "progress": {
+      "current_stage_completion_rate": 0.7,
+      "current_chain_depth_progress_score": 0.75,
+      "latched_chain_depth_progress_score": 0.75,
+      "avg_current_chain_depth": 1.5,
+      "avg_latched_chain_depth": 1.5
+    },
+    "efficiency": {
+      "task_efficiency": 0.6893364199929786,
+      "expert_time_s": 22.683333333333334,
+      "expert_time_step": 1361,
+      "avg_steps_to_success": 1975.2,
+      "avg_policy_steps_to_success": 658.8,
+      "avg_policy_queries_to_success": 24.8
+    },
+    "safety": {
+      "safety_violation_events_per_step": 0.00029864118261908315,
+      "violation_counts": {
+        "drop_to_table_obj_042_adjustable_wrench_4": 1,
+        "drop_to_table_obj_044_flat_screwdriver_3": 1,
+        "drop_to_table_obj_147_drill_5": 3,
+        "high_speed_obj_147_drill_5": 1
+      },
+      "robot_constraint_violation_rate": 0.7,
+      "finger_joint_limit_saturation_rate": 0.0,
+      "joint_limit_active_step_rate": 0.004081429495794137,
+      "finger_joint_limit_saturation_step_rate": 0.0,
+      "max_joint_limit_excess_rad": 0.11344873905181885,
+      "robot_constraint_diagnostics": {
+        "joint_limit_soft_counts": {
+          "left_thumb_abd": 1,
+          "left_thumb_pip": 5,
+          "right_middle_mcp": 2,
+          "right_ring_mcp": 1,
+          "right_thumb_abd": 1,
+          "right_thumb_mcp": 73
+        },
+        "joint_limit_hard_counts": {
+          "left_middle_mcp": 4,
+          "left_thumb_abd": 12,
+          "left_thumb_pip": 3,
+          "right_middle_mcp": 4,
+          "right_ring_mcp": 6,
+          "right_thumb_abd": 2,
+          "right_thumb_mcp": 5,
+          "right_wrist": 1
+        },
+        "joint_limit_hard_episode_events": 35,
+        "finger_joint_limit_saturation_counts": {},
+        "joint_limit_max_excess_by_joint": {
+          "left_middle_mcp": 0.006831526756286621,
+          "left_ring_mcp": 0.003512263298034668,
+          "left_thumb_abd": 0.012391950003802776,
+          "left_thumb_pip": 0.06341838836669922,
+          "left_wrist": 0.0015559792518615723,
+          "multi_panda_joint1": 0.0007355213165283203,
+          "multi_panda_joint6": 0.0014863014221191406,
+          "panda_joint6": 0.00025081634521484375,
+          "right_middle_mcp": 0.01965653896331787,
+          "right_ring_mcp": 0.021506071090698242,
+          "right_thumb_abd": 0.01598169095814228,
+          "right_thumb_mcp": 0.11344873905181885,
+          "right_wrist": 0.005901217460632324
+        },
+        "coverage": {
+          "hard_limit_source": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          },
+          "joint_role_manifest": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          }
+        },
+        "hard_limit_sources": {
+          "runtime.joint_pos_limits": 10
+        },
+        "soft_limit_sources": {
+          "runtime.soft_joint_pos_limits": 10
+        },
+        "unknown_runtime_joint_names": [],
+        "baseline_checks": {
+          "post_home": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          },
+          "post_reset": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          }
+        }
+      }
+    },
+    "robot_motion": {
+      "joint_acceleration_rms": 22.367337886593,
+      "joint_effort_rms": 38.81418593137456,
+      "joint_jerk_rms": 2296.63084483884,
+      "joint_velocity_rms": 0.4387224129909315
+    },
+    "grasp": {
+      "kinematic_grasp_stability_index": 0.9368400181971847,
+      "mean_kinematic_grasp_stability": 0.40969288911124657,
+      "per_object_best_kinematic_grasp_stability": {
+        "obj_042_adjustable_wrench_4": 0.4613274562594772,
+        "obj_043_phillips_screwdriver_2": 0.519796901095459,
+        "obj_044_flat_screwdriver_3": 0.8068015206125777,
+        "obj_147_drill_5": 0.8072000871020392
+      },
+      "grasp_gsi_diagnostics": {
+        "mean_anti_slip_score": 0.7625753955019094,
+        "mean_duration_score": 0.35017685997261283,
+        "mean_hold_score": 0.2890241535183454,
+        "mean_kinematic_stability_score": 0.08842905670716783,
+        "mean_legacy_gsi": 0.3922435900200157,
+        "mean_lift_score": 0.3596693608186037,
+        "mean_motion_stability_score": 0.26173983225426856
+      }
+    },
+    "tool": {
+      "tool_selection_accuracy": null,
+      "tool_switch_success_rate": null,
+      "tool_switch_total": 0,
+      "tool_switch_successful": 0
+    },
+    "runtime": {
+      "avg_steps": 2009.1,
+      "error_count": 0
+    }
+  }
+}

gpt6_astra_1x_icl/per_task_eval/24_stationery_category_sorting/Full/summary.json

@@ -0,0 +1,266 @@
+{
+  "metric_schema_version": "core-v1.8-policy-step-query-split",
+  "metadata": {
+    "policy_name": "gpt-6-astra",
+    "base_seed": 202400000,
+    "seed_policy": "episode_seed = base_seed + episode_index - 1",
+    "randomization_recorded": true,
+    "n_episodes": 10,
+    "successes": 0,
+    "error_count": 0,
+    "evaluation_protocol": "reach_and_stop",
+    "early_stop": true,
+    "success_definition": "terminal.raw_condition holds for dwell_time_s",
+    "measurement_window": "episode_start_to_stable_success_or_failure_budget",
+    "max_episode_steps": 4626,
+    "dwell_time_s": 0.5,
+    "field_semantics": {
+      "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
+      "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
+      "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
+      "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
+      "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
+      "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
+      "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
+      "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
+      "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
+      "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
+      "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
+      "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
+      "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
+      "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
+      "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
+      "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
+      "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
+      "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
+      "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
+      "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
+      "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
+      "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
+      "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
+      "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
+      "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
+    }
+  },
+  "core_metrics": {
+    "completion": {
+      "success_rate": 0.0,
+      "success_rate_ci": {
+        "mean": 0.0,
+        "ci_low": 0.0,
+        "ci_high": 0.0
+      }
+    },
+    "progress": {
+      "latched_stage_completion_rate": 0.25
+    },
+    "efficiency": {
+      "avg_time_to_success_s": null
+    },
+    "safety": {
+      "safe_success_rate": 0.0,
+      "hard_violation_rate": 1.0,
+      "drop_rate": 1.0,
+      "high_speed_violation_rate": 0.0,
+      "safety_violation_step_rate": 0.00032425421530479895
+    },
+    "robustness": {}
+  },
+  "diagnostic_metrics": {
+    "completion": {
+      "ever_instant_success_rate": 0.0,
+      "at_end_success_rate": 0.0,
+      "task_family_success_rate": {
+        "rigid_pick_place": 0.0
+      }
+    },
+    "progress": {
+      "current_stage_completion_rate": 0.15,
+      "current_chain_depth_progress_score": 0.3,
+      "latched_chain_depth_progress_score": 0.35,
+      "avg_current_chain_depth": 0.6,
+      "avg_latched_chain_depth": 0.7
+    },
+    "efficiency": {
+      "task_efficiency": null,
+      "expert_time_s": 51.38333333333333,
+      "expert_time_step": 3083,
+      "avg_steps_to_success": null,
+      "avg_policy_steps_to_success": null,
+      "avg_policy_queries_to_success": null
+    },
+    "safety": {
+      "safety_violation_events_per_step": 0.00032425421530479895,
+      "violation_counts": {
+        "drop_to_table_obj_166_markpen_4": 7,
+        "drop_to_table_obj_168_battery_6": 2,
+        "drop_to_table_obj_199_glue_5": 1,
+        "drop_to_table_obj_221_pen_3": 5
+      },
+      "robot_constraint_violation_rate": 1.0,
+      "finger_joint_limit_saturation_rate": 0.9,
+      "joint_limit_active_step_rate": 0.06288370082144401,
+      "finger_joint_limit_saturation_step_rate": 0.03938607868568958,
+      "max_joint_limit_excess_rad": 2.296665906906128,
+      "robot_constraint_diagnostics": {
+        "joint_limit_soft_counts": {
+          "left_index_distal_joint": 9,
+          "left_index_proximal_joint": 55,
+          "left_middle_distal_joint": 3,
+          "left_middle_proximal_joint": 16,
+          "left_pinky_distal_joint": 84,
+          "left_pinky_proximal_joint": 100,
+          "left_ring_distal_joint": 217,
+          "left_ring_proximal_joint": 45,
+          "left_thumb_distal_joint": 78,
+          "left_thumb_metacarpal_joint": 8,
+          "left_thumb_proximal_joint": 1,
+          "right_index_distal_joint": 111,
+          "right_index_proximal_joint": 344,
+          "right_middle_distal_joint": 226,
+          "right_middle_proximal_joint": 629,
+          "right_pinky_distal_joint": 453,
+          "right_pinky_proximal_joint": 428,
+          "right_ring_distal_joint": 816,
+          "right_ring_proximal_joint": 686,
+          "right_thumb_distal_joint": 86,
+          "right_thumb_metacarpal_joint": 252,
+          "right_thumb_proximal_joint": 116
+        },
+        "joint_limit_hard_counts": {
+          "left_index_distal_joint": 2,
+          "left_index_proximal_joint": 16,
+          "left_middle_distal_joint": 6,
+          "left_middle_proximal_joint": 17,
+          "left_pinky_distal_joint": 11,
+          "left_pinky_proximal_joint": 42,
+          "left_ring_distal_joint": 15,
+          "left_ring_proximal_joint": 22,
+          "left_thumb_distal_joint": 6,
+          "left_thumb_metacarpal_joint": 2,
+          "left_thumb_proximal_joint": 5,
+          "right_index_distal_joint": 12,
+          "right_index_proximal_joint": 27,
+          "right_middle_distal_joint": 22,
+          "right_middle_proximal_joint": 39,
+          "right_pinky_distal_joint": 38,
+          "right_pinky_proximal_joint": 69,
+          "right_ring_distal_joint": 52,
+          "right_ring_proximal_joint": 59,
+          "right_thumb_distal_joint": 13,
+          "right_thumb_metacarpal_joint": 3,
+          "right_thumb_proximal_joint": 13
+        },
+        "joint_limit_hard_episode_events": 242,
+        "finger_joint_limit_saturation_counts": {
+          "left_index_proximal_joint": 10,
+          "left_pinky_distal_joint": 39,
+          "left_pinky_proximal_joint": 37,
+          "left_ring_distal_joint": 99,
+          "left_thumb_distal_joint": 40,
+          "right_index_distal_joint": 38,
+          "right_index_proximal_joint": 275,
+          "right_middle_distal_joint": 103,
+          "right_middle_proximal_joint": 445,
+          "right_pinky_distal_joint": 191,
+          "right_pinky_proximal_joint": 271,
+          "right_ring_distal_joint": 442,
+          "right_ring_proximal_joint": 290,
+          "right_thumb_distal_joint": 26,
+          "right_thumb_metacarpal_joint": 225,
+          "right_thumb_proximal_joint": 15
+        },
+        "joint_limit_max_excess_by_joint": {
+          "left_index_distal_joint": 0.017278848215937614,
+          "left_index_proximal_joint": 0.11591135710477829,
+          "left_middle_distal_joint": 0.011272430419921875,
+          "left_middle_proximal_joint": 0.028246808797121048,
+          "left_pinky_distal_joint": 0.318048357963562,
+          "left_pinky_proximal_joint": 0.1895831823348999,
+          "left_ring_distal_joint": 0.29407820105552673,
+          "left_ring_proximal_joint": 0.043559636920690536,
+          "left_thumb_distal_joint": 0.3178062438964844,
+          "left_thumb_metacarpal_joint": 0.07565286010503769,
+          "left_thumb_proximal_joint": 0.013899445533752441,
+          "right_index_distal_joint": 0.4855976104736328,
+          "right_index_proximal_joint": 2.296665906906128,
+          "right_middle_distal_joint": 0.4053751528263092,
+          "right_middle_proximal_joint": 0.8717591762542725,
+          "right_pinky_distal_joint": 0.2755752503871918,
+          "right_pinky_proximal_joint": 1.1545370817184448,
+          "right_ring_distal_joint": 0.7506979703903198,
+          "right_ring_proximal_joint": 1.6100006103515625,
+          "right_thumb_distal_joint": 0.523934006690979,
+          "right_thumb_metacarpal_joint": 1.271401286125183,
+          "right_thumb_proximal_joint": 0.149954691529274
+        },
+        "coverage": {
+          "hard_limit_source": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          },
+          "joint_role_manifest": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          }
+        },
+        "hard_limit_sources": {
+          "runtime.joint_pos_limits": 10
+        },
+        "soft_limit_sources": {
+          "runtime.soft_joint_pos_limits": 10
+        },
+        "unknown_runtime_joint_names": [],
+        "baseline_checks": {
+          "post_home": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          },
+          "post_reset": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          }
+        }
+      }
+    },
+    "robot_motion": {
+      "joint_acceleration_rms": 17.79519686065434,
+      "joint_effort_rms": 44.77046983146167,
+      "joint_jerk_rms": 1787.0659899737534,
+      "joint_velocity_rms": 0.4567406371406277
+    },
+    "grasp": {
+      "kinematic_grasp_stability_index": 0.9885670708624079,
+      "mean_kinematic_grasp_stability": 0.5321034546628869,
+      "per_object_best_kinematic_grasp_stability": {
+        "obj_166_markpen_4": 0.6682970848742789,
+        "obj_168_battery_6": 0.7739798940890414,
+        "obj_199_glue_5": 0.7085178799598442,
+        "obj_221_pen_3": 0.7578735524374395
+      },
+      "grasp_gsi_diagnostics": {
+        "mean_anti_slip_score": 0.8266284767257528,
+        "mean_duration_score": 0.31778516356823755,
+        "mean_hold_score": 0.21126486165153333,
+        "mean_kinematic_stability_score": 0.09919096057946239,
+        "mean_legacy_gsi": 0.35022192409635716,
+        "mean_lift_score": 0.27171962818849976,
+        "mean_motion_stability_score": 0.18910388950269644
+      }
+    },
+    "tool": {
+      "tool_selection_accuracy": null,
+      "tool_switch_success_rate": null,
+      "tool_switch_total": 0,
+      "tool_switch_successful": 0
+    },
+    "runtime": {
+      "avg_steps": 4626,
+      "error_count": 0
+    }
+  }
+}

gpt6_astra_1x_icl/per_task_eval/26_canned_food_tray_line_arrangement/Full/summary.json

@@ -0,0 +1,278 @@
+{
+  "metric_schema_version": "core-v1.8-policy-step-query-split",
+  "metadata": {
+    "policy_name": "gpt-6-astra",
+    "base_seed": 202600000,
+    "seed_policy": "episode_seed = base_seed + episode_index - 1",
+    "randomization_recorded": true,
+    "n_episodes": 10,
+    "successes": 0,
+    "error_count": 0,
+    "evaluation_protocol": "reach_and_stop",
+    "early_stop": true,
+    "success_definition": "terminal.raw_condition holds for dwell_time_s",
+    "measurement_window": "episode_start_to_stable_success_or_failure_budget",
+    "max_episode_steps": 2613,
+    "dwell_time_s": 0.5,
+    "field_semantics": {
+      "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
+      "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
+      "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
+      "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
+      "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
+      "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
+      "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
+      "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
+      "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
+      "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
+      "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
+      "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
+      "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
+      "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
+      "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
+      "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
+      "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
+      "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
+      "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
+      "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
+      "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
+      "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
+      "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
+      "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
+      "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
+    }
+  },
+  "core_metrics": {
+    "completion": {
+      "success_rate": 0.0,
+      "success_rate_ci": {
+        "mean": 0.0,
+        "ci_low": 0.0,
+        "ci_high": 0.0
+      }
+    },
+    "progress": {
+      "latched_stage_completion_rate": 0.575
+    },
+    "efficiency": {
+      "avg_time_to_success_s": null
+    },
+    "safety": {
+      "safe_success_rate": 0.0,
+      "hard_violation_rate": 0.1,
+      "drop_rate": 0.1,
+      "high_speed_violation_rate": 0.0,
+      "safety_violation_step_rate": 3.827018752391887e-05
+    },
+    "robustness": {}
+  },
+  "diagnostic_metrics": {
+    "completion": {
+      "ever_instant_success_rate": 0.0,
+      "at_end_success_rate": 0.0,
+      "task_family_success_rate": {
+        "rigid_pick_place": 0.0
+      }
+    },
+    "progress": {
+      "current_stage_completion_rate": 0.4,
+      "current_chain_depth_progress_score": 0.6,
+      "latched_chain_depth_progress_score": 0.8,
+      "avg_current_chain_depth": 1.2,
+      "avg_latched_chain_depth": 1.6
+    },
+    "efficiency": {
+      "task_efficiency": null,
+      "expert_time_s": 29.03333333333333,
+      "expert_time_step": 1742,
+      "avg_steps_to_success": null,
+      "avg_policy_steps_to_success": null,
+      "avg_policy_queries_to_success": null
+    },
+    "safety": {
+      "safety_violation_events_per_step": 3.827018752391887e-05,
+      "violation_counts": {
+        "drop_to_table_obj_002_master_chef_can_2": 1
+      },
+      "robot_constraint_violation_rate": 0.7,
+      "finger_joint_limit_saturation_rate": 0.4,
+      "joint_limit_active_step_rate": 0.017106773823191732,
+      "finger_joint_limit_saturation_step_rate": 0.005625717566016073,
+      "max_joint_limit_excess_rad": 1.7231225967407227,
+      "robot_constraint_diagnostics": {
+        "joint_limit_soft_counts": {
+          "left_index_MCP_FE": 1,
+          "left_pinky_CMC": 55,
+          "left_pinky_MCP_FE": 19,
+          "multi_right_index_DIP": 63,
+          "multi_right_index_MCP_AA": 52,
+          "multi_right_index_MCP_FE": 69,
+          "multi_right_index_PIP": 49,
+          "multi_right_middle_DIP": 31,
+          "multi_right_middle_MCP_AA": 56,
+          "multi_right_middle_MCP_FE": 66,
+          "multi_right_middle_PIP": 28,
+          "multi_right_pinky_CMC": 53,
+          "multi_right_pinky_DIP": 19,
+          "multi_right_pinky_MCP_AA": 8,
+          "multi_right_pinky_MCP_FE": 5,
+          "multi_right_ring_MCP_AA": 23,
+          "multi_right_ring_MCP_FE": 53,
+          "multi_right_ring_PIP": 41,
+          "multi_right_thumb_CMC_AA": 276,
+          "multi_right_thumb_IP": 49,
+          "multi_right_thumb_MCP_AA": 50,
+          "multi_right_thumb_MCP_FE": 19
+        },
+        "joint_limit_hard_counts": {
+          "left_index_MCP_FE": 2,
+          "left_pinky_CMC": 4,
+          "left_pinky_MCP_FE": 1,
+          "multi_right_index_DIP": 2,
+          "multi_right_index_MCP_AA": 2,
+          "multi_right_index_MCP_FE": 2,
+          "multi_right_index_PIP": 2,
+          "multi_right_middle_DIP": 1,
+          "multi_right_middle_MCP_AA": 2,
+          "multi_right_middle_MCP_FE": 2,
+          "multi_right_middle_PIP": 3,
+          "multi_right_pinky_CMC": 5,
+          "multi_right_pinky_DIP": 1,
+          "multi_right_pinky_MCP_AA": 1,
+          "multi_right_pinky_MCP_FE": 2,
+          "multi_right_ring_MCP_AA": 1,
+          "multi_right_ring_MCP_FE": 2,
+          "multi_right_ring_PIP": 2,
+          "multi_right_thumb_CMC_AA": 2,
+          "multi_right_thumb_IP": 2,
+          "multi_right_thumb_MCP_AA": 1,
+          "multi_right_thumb_MCP_FE": 1
+        },
+        "joint_limit_hard_episode_events": 13,
+        "finger_joint_limit_saturation_counts": {
+          "left_pinky_CMC": 44,
+          "left_pinky_MCP_FE": 11,
+          "multi_right_index_DIP": 58,
+          "multi_right_index_MCP_AA": 28,
+          "multi_right_index_MCP_FE": 53,
+          "multi_right_index_PIP": 10,
+          "multi_right_middle_DIP": 28,
+          "multi_right_middle_MCP_AA": 41,
+          "multi_right_middle_MCP_FE": 54,
+          "multi_right_middle_PIP": 6,
+          "multi_right_pinky_CMC": 17,
+          "multi_right_pinky_DIP": 15,
+          "multi_right_ring_MCP_AA": 18,
+          "multi_right_ring_MCP_FE": 41,
+          "multi_right_ring_PIP": 12,
+          "multi_right_thumb_CMC_AA": 17,
+          "multi_right_thumb_IP": 40,
+          "multi_right_thumb_MCP_AA": 47,
+          "multi_right_thumb_MCP_FE": 3
+        },
+        "joint_limit_max_excess_by_joint": {
+          "A5": 1.3828277587890625e-05,
+          "left_index_MCP_FE": 0.010401219129562378,
+          "left_index_PIP": 0.00014018648653291166,
+          "left_middle_MCP_FE": 0.0015854686498641968,
+          "left_middle_PIP": 5.984597737551667e-05,
+          "left_pinky_CMC": 1.6369332075119019,
+          "left_pinky_DIP": 4.6774202928645536e-05,
+          "left_pinky_MCP_AA": 0.000651627779006958,
+          "left_pinky_MCP_FE": 0.2908959984779358,
+          "left_pinky_PIP": 0.00024309821310453117,
+          "left_ring_DIP": 5.443812187877484e-05,
+          "left_ring_PIP": 0.00024855288211256266,
+          "left_thumb_CMC_AA": 0.0033279061317443848,
+          "multi_right_index_DIP": 1.7231225967407227,
+          "multi_right_index_MCP_AA": 0.5683627128601074,
+          "multi_right_index_MCP_FE": 1.1254463195800781,
+          "multi_right_index_PIP": 0.5688197612762451,
+          "multi_right_middle_DIP": 0.428147554397583,
+          "multi_right_middle_MCP_AA": 0.5996490716934204,
+          "multi_right_middle_MCP_FE": 1.4571937322616577,
+          "multi_right_middle_PIP": 0.39574775099754333,
+          "multi_right_pinky_CMC": 0.3349069654941559,
+          "multi_right_pinky_DIP": 0.6046924591064453,
+          "multi_right_pinky_MCP_AA": 0.16932150721549988,
+          "multi_right_pinky_MCP_FE": 0.03357513248920441,
+          "multi_right_pinky_PIP": 9.58083196112014e-11,
+          "multi_right_ring_DIP": 4.77389135167261e-11,
+          "multi_right_ring_MCP_AA": 0.46329352259635925,
+          "multi_right_ring_MCP_FE": 0.9943783283233643,
+          "multi_right_ring_PIP": 0.22688937187194824,
+          "multi_right_thumb_CMC_AA": 0.22878697514533997,
+          "multi_right_thumb_IP": 0.5456024408340454,
+          "multi_right_thumb_MCP_AA": 1.2203001976013184,
+          "multi_right_thumb_MCP_FE": 0.2699267268180847
+        },
+        "coverage": {
+          "hard_limit_source": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          },
+          "joint_role_manifest": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          }
+        },
+        "hard_limit_sources": {
+          "runtime.joint_pos_limits": 10
+        },
+        "soft_limit_sources": {
+          "runtime.soft_joint_pos_limits": 10
+        },
+        "unknown_runtime_joint_names": [],
+        "baseline_checks": {
+          "post_home": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          },
+          "post_reset": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          }
+        }
+      }
+    },
+    "robot_motion": {
+      "joint_acceleration_rms": 21.13053212454924,
+      "joint_effort_rms": 39.6013760297938,
+      "joint_jerk_rms": 2153.4881366924997,
+      "joint_velocity_rms": 0.8151944282105695
+    },
+    "grasp": {
+      "kinematic_grasp_stability_index": 0.9930390387800315,
+      "mean_kinematic_grasp_stability": 0.5591430792387045,
+      "per_object_best_kinematic_grasp_stability": {
+        "obj_002_master_chef_can_2": 0.8581517543203825,
+        "obj_010_potted_meat_can_5": 0.8543908082115873,
+        "obj_151_milk_box_4": 0.9615970431595371,
+        "obj_171_msg_3": 0.9200750754081339
+      },
+      "grasp_gsi_diagnostics": {
+        "mean_anti_slip_score": 0.8663700727133495,
+        "mean_duration_score": 0.4545120551090722,
+        "mean_hold_score": 0.24647292384232733,
+        "mean_kinematic_stability_score": 0.1433447166025016,
+        "mean_legacy_gsi": 0.4040219572268817,
+        "mean_lift_score": 0.28454841178721774,
+        "mean_motion_stability_score": 0.22292427740378531
+      }
+    },
+    "tool": {
+      "tool_selection_accuracy": null,
+      "tool_switch_success_rate": null,
+      "tool_switch_total": 0,
+      "tool_switch_successful": 0
+    },
+    "runtime": {
+      "avg_steps": 2613,
+      "error_count": 0
+    }
+  }
+}

gpt6_astra_1x_icl/per_task_eval/27_ball_box_loading/Full/summary.json

@@ -0,0 +1,209 @@
+{
+  "metric_schema_version": "core-v1.8-policy-step-query-split",
+  "metadata": {
+    "policy_name": "gpt-6-astra",
+    "base_seed": 202700000,
+    "seed_policy": "episode_seed = base_seed + episode_index - 1",
+    "randomization_recorded": true,
+    "n_episodes": 10,
+    "successes": 10,
+    "error_count": 0,
+    "evaluation_protocol": "reach_and_stop",
+    "early_stop": true,
+    "success_definition": "terminal.raw_condition holds for dwell_time_s",
+    "measurement_window": "episode_start_to_stable_success_or_failure_budget",
+    "max_episode_steps": 3570,
+    "dwell_time_s": 0.5,
+    "field_semantics": {
+      "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
+      "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
+      "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
+      "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
+      "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
+      "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
+      "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
+      "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
+      "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
+      "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
+      "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
+      "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
+      "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
+      "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
+      "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
+      "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
+      "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
+      "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
+      "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
+      "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
+      "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
+      "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
+      "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
+      "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
+      "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
+    }
+  },
+  "core_metrics": {
+    "completion": {
+      "success_rate": 1.0,
+      "success_rate_ci": {
+        "mean": 1.0,
+        "ci_low": 1.0,
+        "ci_high": 1.0
+      }
+    },
+    "progress": {
+      "latched_stage_completion_rate": 1.0
+    },
+    "efficiency": {
+      "avg_time_to_success_s": 45.18999999999872
+    },
+    "safety": {
+      "safe_success_rate": 0.3,
+      "hard_violation_rate": 0.7,
+      "drop_rate": 0.1,
+      "high_speed_violation_rate": 0.7,
+      "safety_violation_step_rate": 0.0004425757911042266
+    },
+    "robustness": {}
+  },
+  "diagnostic_metrics": {
+    "completion": {
+      "ever_instant_success_rate": 1.0,
+      "at_end_success_rate": 1.0,
+      "task_family_success_rate": {
+        "rigid_pick_place": 1.0
+      }
+    },
+    "progress": {
+      "current_stage_completion_rate": 1.0,
+      "current_chain_depth_progress_score": 1.0,
+      "latched_chain_depth_progress_score": 1.0,
+      "avg_current_chain_depth": 2,
+      "avg_latched_chain_depth": 2
+    },
+    "efficiency": {
+      "task_efficiency": 0.8815334164071097,
+      "expert_time_s": 39.666666666666664,
+      "expert_time_step": 2380,
+      "avg_steps_to_success": 2711.4,
+      "avg_policy_steps_to_success": 904.2,
+      "avg_policy_queries_to_success": 19.8
+    },
+    "safety": {
+      "safety_violation_events_per_step": 0.00036881315925352216,
+      "violation_counts": {
+        "drop_to_table_obj_058_golf_ball_4": 1,
+        "high_speed_obj_058_golf_ball_4": 3,
+        "high_speed_obj_144_table_tennis_5": 6
+      },
+      "robot_constraint_violation_rate": 0.7,
+      "finger_joint_limit_saturation_rate": 0.6,
+      "joint_limit_active_step_rate": 0.02268200929409161,
+      "finger_joint_limit_saturation_step_rate": 0.016227779007154976,
+      "max_joint_limit_excess_rad": 0.5655606985092163,
+      "robot_constraint_diagnostics": {
+        "joint_limit_soft_counts": {
+          "left_finger5_joint3": 23,
+          "right_finger1_joint3": 230,
+          "right_finger2_joint1": 47,
+          "right_finger3_joint1": 252,
+          "right_finger4_joint1": 252
+        },
+        "joint_limit_hard_counts": {
+          "left_finger5_joint3": 1,
+          "right_finger1_joint2": 14,
+          "right_finger1_joint3": 2,
+          "right_finger2_joint1": 1,
+          "right_finger3_joint1": 4,
+          "right_finger4_joint1": 4
+        },
+        "joint_limit_hard_episode_events": 20,
+        "finger_joint_limit_saturation_counts": {
+          "left_finger5_joint3": 13,
+          "right_finger1_joint3": 171,
+          "right_finger2_joint1": 26,
+          "right_finger3_joint1": 168,
+          "right_finger4_joint1": 161
+        },
+        "joint_limit_max_excess_by_joint": {
+          "left_finger1_joint2": 0.004154786467552185,
+          "left_finger2_joint1": 0.0003505498170852661,
+          "left_finger2_joint2": 0.0003935396671295166,
+          "left_finger5_joint3": 0.2794632911682129,
+          "right_finger1_joint2": 0.005902811884880066,
+          "right_finger1_joint3": 0.31119072437286377,
+          "right_finger2_joint1": 0.3970223665237427,
+          "right_finger2_joint2": 0.0016947686672210693,
+          "right_finger3_joint1": 0.5655606985092163,
+          "right_finger4_joint1": 0.5568047761917114
+        },
+        "coverage": {
+          "hard_limit_source": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          },
+          "joint_role_manifest": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          }
+        },
+        "hard_limit_sources": {
+          "runtime.joint_pos_limits": 10
+        },
+        "soft_limit_sources": {
+          "runtime.soft_joint_pos_limits": 10
+        },
+        "unknown_runtime_joint_names": [],
+        "baseline_checks": {
+          "post_home": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          },
+          "post_reset": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          }
+        }
+      }
+    },
+    "robot_motion": {
+      "joint_acceleration_rms": 9.256581322588325,
+      "joint_effort_rms": 20.439332879679043,
+      "joint_jerk_rms": 945.797398486884,
+      "joint_velocity_rms": 0.32309312904499365
+    },
+    "grasp": {
+      "kinematic_grasp_stability_index": 0.9688283562851823,
+      "mean_kinematic_grasp_stability": 0.46235651122128657,
+      "per_object_best_kinematic_grasp_stability": {
+        "obj_053_mini_soccer_ball_2": 0.8429700090279171,
+        "obj_056_tennis_ball_3": 0.7501144634352336,
+        "obj_058_golf_ball_4": 0.6582300475117814,
+        "obj_144_table_tennis_5": 0.5581942281640864
+      },
+      "grasp_gsi_diagnostics": {
+        "mean_anti_slip_score": 0.6459908498738125,
+        "mean_duration_score": 0.46322743964187374,
+        "mean_hold_score": 0.33652721456083634,
+        "mean_kinematic_stability_score": 0.07720277747525262,
+        "mean_legacy_gsi": 0.41907977307571204,
+        "mean_lift_score": 0.38939890230149105,
+        "mean_motion_stability_score": 0.31800925605330843
+      }
+    },
+    "tool": {
+      "tool_selection_accuracy": null,
+      "tool_switch_success_rate": null,
+      "tool_switch_total": 0,
+      "tool_switch_successful": 0
+    },
+    "runtime": {
+      "avg_steps": 2711.4,
+      "error_count": 0
+    }
+  }
+}

gpt6_astra_1x_icl/per_task_eval/32_baking_tray_prep_with_tools/Full/summary.json

@@ -0,0 +1,313 @@
+{
+  "metric_schema_version": "core-v1.8-policy-step-query-split",
+  "metadata": {
+    "policy_name": "gpt-6-astra",
+    "base_seed": 203200000,
+    "seed_policy": "episode_seed = base_seed + episode_index - 1",
+    "randomization_recorded": true,
+    "n_episodes": 10,
+    "successes": 0,
+    "error_count": 0,
+    "evaluation_protocol": "reach_and_stop",
+    "early_stop": true,
+    "success_definition": "terminal.raw_condition holds for dwell_time_s",
+    "measurement_window": "episode_start_to_stable_success_or_failure_budget",
+    "max_episode_steps": 2556,
+    "dwell_time_s": 0.5,
+    "field_semantics": {
+      "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
+      "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
+      "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
+      "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
+      "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
+      "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
+      "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
+      "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
+      "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
+      "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
+      "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
+      "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
+      "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
+      "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
+      "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
+      "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
+      "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
+      "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
+      "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
+      "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
+      "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
+      "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
+      "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
+      "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
+      "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
+    }
+  },
+  "core_metrics": {
+    "completion": {
+      "success_rate": 0.0,
+      "success_rate_ci": {
+        "mean": 0.0,
+        "ci_low": 0.0,
+        "ci_high": 0.0
+      }
+    },
+    "progress": {
+      "latched_stage_completion_rate": 0.075
+    },
+    "efficiency": {
+      "avg_time_to_success_s": null
+    },
+    "safety": {
+      "safe_success_rate": 0.0,
+      "hard_violation_rate": 0.2,
+      "drop_rate": 0.2,
+      "high_speed_violation_rate": 0.0,
+      "safety_violation_step_rate": 0.00011737089201877934
+    },
+    "robustness": {}
+  },
+  "diagnostic_metrics": {
+    "completion": {
+      "ever_instant_success_rate": 0.0,
+      "at_end_success_rate": 0.0,
+      "task_family_success_rate": {
+        "rigid_pick_place": 0.0
+      }
+    },
+    "progress": {
+      "current_stage_completion_rate": 0.05,
+      "current_chain_depth_progress_score": 0.1,
+      "latched_chain_depth_progress_score": 0.15,
+      "avg_current_chain_depth": 0.2,
+      "avg_latched_chain_depth": 0.3
+    },
+    "efficiency": {
+      "task_efficiency": null,
+      "expert_time_s": 28.383333333333333,
+      "expert_time_step": 1703,
+      "avg_steps_to_success": null,
+      "avg_policy_steps_to_success": null,
+      "avg_policy_queries_to_success": null
+    },
+    "safety": {
+      "safety_violation_events_per_step": 0.00011737089201877934,
+      "violation_counts": {
+        "drop_to_table_obj_009_gelatin_box_3": 1,
+        "drop_to_table_obj_033_spatula_4": 1,
+        "drop_to_table_obj_189_brush_5": 1
+      },
+      "robot_constraint_violation_rate": 1.0,
+      "finger_joint_limit_saturation_rate": 1.0,
+      "joint_limit_active_step_rate": 0.06083724569640062,
+      "finger_joint_limit_saturation_step_rate": 0.037363067292644755,
+      "max_joint_limit_excess_rad": 2.900862455368042,
+      "robot_constraint_diagnostics": {
+        "joint_limit_soft_counts": {
+          "left_index_MCP_AA": 13,
+          "left_index_MCP_FE": 7,
+          "left_middle_DIP": 121,
+          "left_middle_PIP": 36,
+          "left_pinky_CMC": 336,
+          "left_pinky_DIP": 146,
+          "left_pinky_MCP_FE": 47,
+          "left_ring_DIP": 84,
+          "left_ring_PIP": 5,
+          "left_thumb_CMC_AA": 57,
+          "left_thumb_MCP_AA": 20,
+          "multi_right_index_DIP": 67,
+          "multi_right_index_MCP_FE": 222,
+          "multi_right_index_PIP": 50,
+          "multi_right_middle_DIP": 158,
+          "multi_right_middle_MCP_AA": 2,
+          "multi_right_middle_MCP_FE": 63,
+          "multi_right_middle_PIP": 128,
+          "multi_right_pinky_CMC": 254,
+          "multi_right_pinky_DIP": 1,
+          "multi_right_pinky_MCP_AA": 118,
+          "multi_right_pinky_MCP_FE": 42,
+          "multi_right_pinky_PIP": 100,
+          "multi_right_ring_DIP": 90,
+          "multi_right_ring_MCP_AA": 56,
+          "multi_right_ring_MCP_FE": 57,
+          "multi_right_ring_PIP": 131,
+          "multi_right_thumb_CMC_AA": 107,
+          "multi_right_thumb_CMC_FE": 68,
+          "multi_right_thumb_IP": 45,
+          "multi_right_thumb_MCP_AA": 39,
+          "multi_right_thumb_MCP_FE": 94
+        },
+        "joint_limit_hard_counts": {
+          "left_index_MCP_AA": 1,
+          "left_index_MCP_FE": 1,
+          "left_middle_DIP": 8,
+          "left_middle_PIP": 2,
+          "left_pinky_CMC": 6,
+          "left_pinky_DIP": 7,
+          "left_pinky_MCP_FE": 1,
+          "left_ring_DIP": 4,
+          "left_ring_PIP": 1,
+          "left_thumb_CMC_AA": 5,
+          "left_thumb_MCP_AA": 1,
+          "multi_right_index_DIP": 3,
+          "multi_right_index_MCP_FE": 7,
+          "multi_right_index_PIP": 4,
+          "multi_right_middle_DIP": 5,
+          "multi_right_middle_MCP_AA": 1,
+          "multi_right_middle_MCP_FE": 4,
+          "multi_right_middle_PIP": 4,
+          "multi_right_pinky_CMC": 12,
+          "multi_right_pinky_DIP": 4,
+          "multi_right_pinky_MCP_AA": 3,
+          "multi_right_pinky_MCP_FE": 9,
+          "multi_right_pinky_PIP": 8,
+          "multi_right_ring_DIP": 5,
+          "multi_right_ring_MCP_AA": 1,
+          "multi_right_ring_MCP_FE": 3,
+          "multi_right_ring_PIP": 4,
+          "multi_right_thumb_CMC_AA": 22,
+          "multi_right_thumb_CMC_FE": 3,
+          "multi_right_thumb_IP": 2,
+          "multi_right_thumb_MCP_AA": 1,
+          "multi_right_thumb_MCP_FE": 19
+        },
+        "joint_limit_hard_episode_events": 70,
+        "finger_joint_limit_saturation_counts": {
+          "left_middle_DIP": 64,
+          "left_middle_PIP": 23,
+          "left_pinky_CMC": 242,
+          "left_pinky_DIP": 68,
+          "left_ring_DIP": 33,
+          "left_thumb_CMC_AA": 13,
+          "left_thumb_MCP_AA": 3,
+          "multi_right_index_DIP": 61,
+          "multi_right_index_MCP_FE": 80,
+          "multi_right_index_PIP": 29,
+          "multi_right_middle_DIP": 115,
+          "multi_right_middle_MCP_FE": 25,
+          "multi_right_middle_PIP": 94,
+          "multi_right_pinky_CMC": 150,
+          "multi_right_pinky_MCP_AA": 65,
+          "multi_right_pinky_PIP": 65,
+          "multi_right_ring_DIP": 62,
+          "multi_right_ring_MCP_AA": 51,
+          "multi_right_ring_MCP_FE": 11,
+          "multi_right_ring_PIP": 77,
+          "multi_right_thumb_CMC_AA": 56,
+          "multi_right_thumb_CMC_FE": 58,
+          "multi_right_thumb_IP": 25,
+          "multi_right_thumb_MCP_AA": 18,
+          "multi_right_thumb_MCP_FE": 50
+        },
+        "joint_limit_max_excess_by_joint": {
+          "A2": 9.959936141967773e-05,
+          "left_index_DIP": 0.0004909038543701172,
+          "left_index_MCP_AA": 0.08945390582084656,
+          "left_index_MCP_FE": 0.04542422294616699,
+          "left_index_PIP": 0.0021076472476124763,
+          "left_middle_DIP": 0.6832666993141174,
+          "left_middle_PIP": 0.44560933113098145,
+          "left_pinky_CMC": 2.900862455368042,
+          "left_pinky_DIP": 0.4595421552658081,
+          "left_pinky_MCP_AA": 0.003502368927001953,
+          "left_pinky_MCP_FE": 0.08163511753082275,
+          "left_pinky_PIP": 0.0019508898258209229,
+          "left_ring_DIP": 0.5615358948707581,
+          "left_ring_MCP_AA": 0.0045351386070251465,
+          "left_ring_PIP": 0.057178858667612076,
+          "left_thumb_CMC_AA": 0.14517754316329956,
+          "left_thumb_CMC_FE": 0.0029706209897994995,
+          "left_thumb_IP": 0.00015700132644269615,
+          "left_thumb_MCP_AA": 0.26343855261802673,
+          "multi_right_index_DIP": 1.5347728729248047,
+          "multi_right_index_MCP_AA": 0.00014585256576538086,
+          "multi_right_index_MCP_FE": 0.23827871680259705,
+          "multi_right_index_PIP": 0.44316303730010986,
+          "multi_right_middle_DIP": 1.5245301723480225,
+          "multi_right_middle_MCP_AA": 0.03112727403640747,
+          "multi_right_middle_MCP_FE": 0.1781976819038391,
+          "multi_right_middle_PIP": 0.9427446126937866,
+          "multi_right_pinky_CMC": 0.854444682598114,
+          "multi_right_pinky_DIP": 0.010975774377584457,
+          "multi_right_pinky_MCP_AA": 0.18510952591896057,
+          "multi_right_pinky_MCP_FE": 0.14120236039161682,
+          "multi_right_pinky_PIP": 1.0234969854354858,
+          "multi_right_ring_DIP": 0.5525888204574585,
+          "multi_right_ring_MCP_AA": 0.6991322040557861,
+          "multi_right_ring_MCP_FE": 0.15312358736991882,
+          "multi_right_ring_PIP": 0.5413898825645447,
+          "multi_right_thumb_CMC_AA": 0.9160000085830688,
+          "multi_right_thumb_CMC_FE": 0.9682929515838623,
+          "multi_right_thumb_IP": 0.4199718236923218,
+          "multi_right_thumb_MCP_AA": 0.3752760589122772,
+          "multi_right_thumb_MCP_FE": 0.5875681638717651
+        },
+        "coverage": {
+          "hard_limit_source": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          },
+          "joint_role_manifest": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          }
+        },
+        "hard_limit_sources": {
+          "runtime.joint_pos_limits": 10
+        },
+        "soft_limit_sources": {
+          "runtime.soft_joint_pos_limits": 10
+        },
+        "unknown_runtime_joint_names": [],
+        "baseline_checks": {
+          "post_home": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          },
+          "post_reset": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          }
+        }
+      }
+    },
+    "robot_motion": {
+      "joint_acceleration_rms": 41.85202679742137,
+      "joint_effort_rms": 43.6663283644778,
+      "joint_jerk_rms": 4176.809250717952,
+      "joint_velocity_rms": 0.7754996672267569
+    },
+    "grasp": {
+      "kinematic_grasp_stability_index": 0.6549562059102851,
+      "mean_kinematic_grasp_stability": 0.3007984859248549,
+      "per_object_best_kinematic_grasp_stability": {
+        "obj_008_pudding_box_2": 0.0,
+        "obj_009_gelatin_box_3": 0.3662892199049133,
+        "obj_033_spatula_4": 0.3254284429457169,
+        "obj_189_brush_5": 0.39449756256106144
+      },
+      "grasp_gsi_diagnostics": {
+        "mean_anti_slip_score": 0.9704225352112645,
+        "mean_duration_score": 0.06400723787167452,
+        "mean_hold_score": 0.04528022300469473,
+        "mean_kinematic_stability_score": 0.02021204991771197,
+        "mean_legacy_gsi": 0.23476010926168536,
+        "mean_lift_score": 0.0626858372456964,
+        "mean_motion_stability_score": 0.04377626796963671
+      }
+    },
+    "tool": {
+      "tool_selection_accuracy": null,
+      "tool_switch_success_rate": null,
+      "tool_switch_total": 0,
+      "tool_switch_successful": 0
+    },
+    "runtime": {
+      "avg_steps": 2556,
+      "error_count": 0
+    }
+  }
+}

gpt6_astra_1x_icl/per_task_eval/34_fridge_wine_interhand_pour/Full/summary.json

@@ -0,0 +1,315 @@
+{
+  "metric_schema_version": "core-v1.8-policy-step-query-split",
+  "metadata": {
+    "policy_name": "gpt-6-astra",
+    "base_seed": 203400000,
+    "seed_policy": "episode_seed = base_seed + episode_index - 1",
+    "randomization_recorded": true,
+    "n_episodes": 10,
+    "successes": 3,
+    "error_count": 0,
+    "evaluation_protocol": "reach_and_stop",
+    "early_stop": true,
+    "success_definition": "terminal.raw_condition holds for dwell_time_s",
+    "measurement_window": "episode_start_to_stable_success_or_failure_budget",
+    "max_episode_steps": 3240,
+    "dwell_time_s": 0.5,
+    "field_semantics": {
+      "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
+      "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
+      "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
+      "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
+      "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
+      "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
+      "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
+      "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
+      "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
+      "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
+      "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
+      "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
+      "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
+      "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
+      "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
+      "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
+      "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
+      "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
+      "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
+      "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
+      "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
+      "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
+      "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
+      "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
+      "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
+    }
+  },
+  "core_metrics": {
+    "completion": {
+      "success_rate": 0.3,
+      "success_rate_ci": {
+        "mean": 0.3,
+        "ci_low": 0.0,
+        "ci_high": 0.6
+      }
+    },
+    "progress": {
+      "latched_stage_completion_rate": 0.6799999999999999
+    },
+    "efficiency": {
+      "avg_time_to_success_s": 47.79999999999857
+    },
+    "safety": {
+      "safe_success_rate": 0.2,
+      "hard_violation_rate": 0.7,
+      "drop_rate": 0.7,
+      "high_speed_violation_rate": 0.0,
+      "safety_violation_step_rate": 0.00022375655287047692
+    },
+    "robustness": {}
+  },
+  "diagnostic_metrics": {
+    "completion": {
+      "ever_instant_success_rate": 0.3,
+      "at_end_success_rate": 0.3,
+      "task_family_success_rate": {
+        "articulation_pour": 0.3
+      }
+    },
+    "progress": {
+      "current_stage_completion_rate": 0.42,
+      "current_chain_depth_progress_score": 0.62,
+      "latched_chain_depth_progress_score": 0.6799999999999999,
+      "avg_current_chain_depth": 3.1,
+      "avg_latched_chain_depth": 3.4
+    },
+    "efficiency": {
+      "task_efficiency": 0.7564199283806476,
+      "expert_time_s": 36.0,
+      "expert_time_step": 2160,
+      "avg_steps_to_success": 2868,
+      "avg_policy_steps_to_success": 956.3333333333334,
+      "avg_policy_queries_to_success": 27.666666666666668
+    },
+    "safety": {
+      "safety_violation_events_per_step": 0.00022375655287047692,
+      "violation_counts": {
+        "drop_to_table_obj_224_wine_bottle_1": 7
+      },
+      "robot_constraint_violation_rate": 1.0,
+      "finger_joint_limit_saturation_rate": 1.0,
+      "joint_limit_active_step_rate": 0.3917018284106892,
+      "finger_joint_limit_saturation_step_rate": 0.3418360823424115,
+      "max_joint_limit_excess_rad": 2.037945032119751,
+      "robot_constraint_diagnostics": {
+        "joint_limit_soft_counts": {
+          "left_index_dip_joint": 61,
+          "left_index_mcp_joint": 12,
+          "left_index_pip_joint": 697,
+          "left_index_pitch_joint": 494,
+          "left_middle_dip_joint": 40,
+          "left_middle_mcp_joint": 61,
+          "left_middle_pip_joint": 1173,
+          "left_middle_pitch_joint": 344,
+          "left_pinky_mcp_joint": 213,
+          "left_pinky_pip_joint": 140,
+          "left_ring_dip_joint": 27,
+          "left_ring_mcp_joint": 277,
+          "left_ring_pip_joint": 896,
+          "left_thumb_dip_joint": 56,
+          "left_thumb_mcp_joint": 3259,
+          "left_thumb_pip_joint": 125,
+          "left_thumb_yaw_joint": 32,
+          "right_index_dip_joint": 272,
+          "right_index_mcp_joint": 251,
+          "right_index_pip_joint": 645,
+          "right_index_pitch_joint": 379,
+          "right_middle_dip_joint": 327,
+          "right_middle_mcp_joint": 436,
+          "right_middle_pip_joint": 634,
+          "right_middle_pitch_joint": 300,
+          "right_pinky_dip_joint": 6,
+          "right_pinky_mcp_joint": 378,
+          "right_pinky_pip_joint": 356,
+          "right_ring_dip_joint": 393,
+          "right_ring_mcp_joint": 383,
+          "right_ring_pip_joint": 732,
+          "right_thumb_dip_joint": 141,
+          "right_thumb_mcp_joint": 6681,
+          "right_thumb_pip_joint": 94,
+          "right_thumb_yaw_joint": 92
+        },
+        "joint_limit_hard_counts": {
+          "left_index_dip_joint": 4,
+          "left_index_mcp_joint": 2,
+          "left_index_pip_joint": 89,
+          "left_index_pitch_joint": 17,
+          "left_middle_dip_joint": 3,
+          "left_middle_mcp_joint": 5,
+          "left_middle_pip_joint": 62,
+          "left_middle_pitch_joint": 14,
+          "left_pinky_mcp_joint": 22,
+          "left_pinky_pip_joint": 10,
+          "left_ring_dip_joint": 4,
+          "left_ring_mcp_joint": 24,
+          "left_ring_pip_joint": 57,
+          "left_thumb_dip_joint": 1,
+          "left_thumb_mcp_joint": 37,
+          "left_thumb_pip_joint": 4,
+          "left_thumb_yaw_joint": 2,
+          "right_index_dip_joint": 16,
+          "right_index_mcp_joint": 2,
+          "right_index_pip_joint": 8,
+          "right_index_pitch_joint": 12,
+          "right_middle_dip_joint": 24,
+          "right_middle_mcp_joint": 4,
+          "right_middle_pip_joint": 8,
+          "right_middle_pitch_joint": 8,
+          "right_pinky_dip_joint": 1,
+          "right_pinky_mcp_joint": 15,
+          "right_pinky_pip_joint": 5,
+          "right_ring_dip_joint": 15,
+          "right_ring_mcp_joint": 16,
+          "right_ring_pip_joint": 6,
+          "right_thumb_dip_joint": 8,
+          "right_thumb_mcp_joint": 15,
+          "right_thumb_pip_joint": 2,
+          "right_thumb_yaw_joint": 2
+        },
+        "joint_limit_hard_episode_events": 167,
+        "finger_joint_limit_saturation_counts": {
+          "left_index_dip_joint": 35,
+          "left_index_pip_joint": 348,
+          "left_index_pitch_joint": 336,
+          "left_middle_dip_joint": 7,
+          "left_middle_pip_joint": 662,
+          "left_middle_pitch_joint": 254,
+          "left_pinky_mcp_joint": 102,
+          "left_pinky_pip_joint": 106,
+          "left_ring_mcp_joint": 171,
+          "left_ring_pip_joint": 528,
+          "left_thumb_dip_joint": 52,
+          "left_thumb_mcp_joint": 2601,
+          "left_thumb_pip_joint": 115,
+          "left_thumb_yaw_joint": 26,
+          "right_index_dip_joint": 141,
+          "right_index_mcp_joint": 211,
+          "right_index_pip_joint": 540,
+          "right_index_pitch_joint": 315,
+          "right_middle_dip_joint": 191,
+          "right_middle_mcp_joint": 357,
+          "right_middle_pip_joint": 542,
+          "right_middle_pitch_joint": 218,
+          "right_pinky_mcp_joint": 253,
+          "right_pinky_pip_joint": 309,
+          "right_ring_dip_joint": 310,
+          "right_ring_mcp_joint": 269,
+          "right_ring_pip_joint": 663,
+          "right_thumb_dip_joint": 111,
+          "right_thumb_mcp_joint": 6338,
+          "right_thumb_pip_joint": 74,
+          "right_thumb_yaw_joint": 68
+        },
+        "joint_limit_max_excess_by_joint": {
+          "left_index_dip_joint": 0.2799852788448334,
+          "left_index_mcp_joint": 0.1833624243736267,
+          "left_index_pip_joint": 1.2716604471206665,
+          "left_index_pitch_joint": 1.605282187461853,
+          "left_middle_dip_joint": 0.16115638613700867,
+          "left_middle_mcp_joint": 0.2435968816280365,
+          "left_middle_pip_joint": 1.4053504467010498,
+          "left_middle_pitch_joint": 1.5653107166290283,
+          "left_pinky_dip_joint": 0.0007799672312103212,
+          "left_pinky_mcp_joint": 1.7634187936782837,
+          "left_pinky_pip_joint": 1.73162043094635,
+          "left_ring_dip_joint": 0.15498420596122742,
+          "left_ring_mcp_joint": 1.5648345947265625,
+          "left_ring_pip_joint": 1.2001138925552368,
+          "left_thumb_dip_joint": 1.2471123933792114,
+          "left_thumb_mcp_joint": 1.5431932210922241,
+          "left_thumb_pip_joint": 0.9053640961647034,
+          "left_thumb_yaw_joint": 0.5279140472412109,
+          "right_index_dip_joint": 1.0168943405151367,
+          "right_index_mcp_joint": 0.565585732460022,
+          "right_index_pip_joint": 1.4217863082885742,
+          "right_index_pitch_joint": 0.7631471753120422,
+          "right_middle_dip_joint": 0.8343515396118164,
+          "right_middle_mcp_joint": 0.5970035791397095,
+          "right_middle_pip_joint": 1.8758935928344727,
+          "right_middle_pitch_joint": 0.835150420665741,
+          "right_pinky_dip_joint": 0.22969281673431396,
+          "right_pinky_mcp_joint": 0.9765199422836304,
+          "right_pinky_pip_joint": 1.4550007581710815,
+          "right_ring_dip_joint": 1.306758165359497,
+          "right_ring_mcp_joint": 0.5623847246170044,
+          "right_ring_pip_joint": 2.037945032119751,
+          "right_thumb_dip_joint": 1.4873428344726562,
+          "right_thumb_mcp_joint": 0.6458135843276978,
+          "right_thumb_pip_joint": 0.2756539583206177,
+          "right_thumb_yaw_joint": 0.6658437252044678
+        },
+        "coverage": {
+          "hard_limit_source": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          },
+          "joint_role_manifest": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          }
+        },
+        "hard_limit_sources": {
+          "runtime.joint_pos_limits": 10
+        },
+        "soft_limit_sources": {
+          "runtime.soft_joint_pos_limits": 10
+        },
+        "unknown_runtime_joint_names": [],
+        "baseline_checks": {
+          "post_home": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          },
+          "post_reset": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          }
+        }
+      }
+    },
+    "robot_motion": {
+      "joint_acceleration_rms": 47.787378654259044,
+      "joint_effort_rms": 38.57953411576559,
+      "joint_jerk_rms": 4881.414279506254,
+      "joint_velocity_rms": 0.9058797532993996
+    },
+    "grasp": {
+      "kinematic_grasp_stability_index": 0.8886124110017146,
+      "mean_kinematic_grasp_stability": 0.4925575938786643,
+      "per_object_best_kinematic_grasp_stability": {
+        "obj_224_wine_bottle_1": 0.8886124110017146
+      },
+      "grasp_gsi_diagnostics": {
+        "mean_anti_slip_score": 0.48982951166362576,
+        "mean_duration_score": 0.7648623145760967,
+        "mean_hold_score": 0.408584463530777,
+        "mean_kinematic_stability_score": 0.14470559194835245,
+        "mean_legacy_gsi": 0.48046335493133674,
+        "mean_lift_score": 0.5189258376899576,
+        "mean_motion_stability_score": 0.3453879590047513
+      }
+    },
+    "tool": {
+      "tool_selection_accuracy": null,
+      "tool_switch_success_rate": null,
+      "tool_switch_total": 0,
+      "tool_switch_successful": 0
+    },
+    "runtime": {
+      "avg_steps": 3128.4,
+      "error_count": 0
+    }
+  }
+}

gpt6_astra_1x_icl/per_task_eval/42_trash_disposal/Full/summary.json

@@ -0,0 +1,267 @@
+{
+  "metric_schema_version": "core-v1.8-policy-step-query-split",
+  "metadata": {
+    "policy_name": "gpt-6-astra",
+    "base_seed": 204200000,
+    "seed_policy": "episode_seed = base_seed + episode_index - 1",
+    "randomization_recorded": true,
+    "n_episodes": 10,
+    "successes": 5,
+    "error_count": 0,
+    "evaluation_protocol": "reach_and_stop",
+    "early_stop": true,
+    "success_definition": "terminal.raw_condition holds for dwell_time_s",
+    "measurement_window": "episode_start_to_stable_success_or_failure_budget",
+    "max_episode_steps": 2004,
+    "dwell_time_s": 0.5,
+    "field_semantics": {
+      "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
+      "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
+      "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
+      "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
+      "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
+      "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
+      "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
+      "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
+      "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
+      "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
+      "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
+      "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
+      "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
+      "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
+      "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
+      "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
+      "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
+      "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
+      "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
+      "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
+      "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
+      "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
+      "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
+      "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
+      "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
+    }
+  },
+  "core_metrics": {
+    "completion": {
+      "success_rate": 0.5,
+      "success_rate_ci": {
+        "mean": 0.5,
+        "ci_low": 0.2,
+        "ci_high": 0.8
+      }
+    },
+    "progress": {
+      "latched_stage_completion_rate": 0.7
+    },
+    "efficiency": {
+      "avg_time_to_success_s": 24.73333333333321
+    },
+    "safety": {
+      "safe_success_rate": 0.1,
+      "hard_violation_rate": 0.9,
+      "drop_rate": 0.8,
+      "high_speed_violation_rate": 0.8,
+      "safety_violation_step_rate": 0.22832568807339448
+    },
+    "robustness": {}
+  },
+  "diagnostic_metrics": {
+    "completion": {
+      "ever_instant_success_rate": 0.7,
+      "at_end_success_rate": 0.5,
+      "task_family_success_rate": {
+        "trash_sorting_disposal": 0.5
+      }
+    },
+    "progress": {
+      "current_stage_completion_rate": 0.6333333333333333,
+      "current_chain_depth_progress_score": 0.7,
+      "latched_chain_depth_progress_score": 0.7,
+      "avg_current_chain_depth": 2.1,
+      "avg_latched_chain_depth": 2.1
+    },
+    "efficiency": {
+      "task_efficiency": 0.9178965738222508,
+      "expert_time_s": 22.266666666666666,
+      "expert_time_step": 1336,
+      "avg_steps_to_success": 1484,
+      "avg_policy_steps_to_success": 495,
+      "avg_policy_queries_to_success": 16.8
+    },
+    "safety": {
+      "safety_violation_events_per_step": 0.0017775229357798166,
+      "violation_counts": {
+        "drop_obj_080_bottle_3": 5,
+        "drop_to_table_obj_080_bottle_3": 1,
+        "drop_to_table_obj_341_banana_peel_1": 7,
+        "drop_to_table_obj_349_paper_low_2": 4,
+        "high_speed_obj_080_bottle_3": 2,
+        "high_speed_obj_341_banana_peel_1": 2,
+        "high_speed_obj_349_paper_low_2": 10
+      },
+      "robot_constraint_violation_rate": 1.0,
+      "finger_joint_limit_saturation_rate": 1.0,
+      "joint_limit_active_step_rate": 0.2680619266055046,
+      "finger_joint_limit_saturation_step_rate": 0.1694380733944954,
+      "max_joint_limit_excess_rad": 1.6444499492645264,
+      "robot_constraint_diagnostics": {
+        "joint_limit_soft_counts": {
+          "left_index_1_joint": 254,
+          "left_index_2_joint": 1,
+          "left_little_1_joint": 186,
+          "left_little_2_joint": 28,
+          "left_middle_1_joint": 217,
+          "left_middle_2_joint": 10,
+          "left_ring_1_joint": 218,
+          "left_ring_2_joint": 82,
+          "left_thumb_1_joint": 2,
+          "left_thumb_2_joint": 1915,
+          "right_index_1_joint": 1308,
+          "right_little_1_joint": 328,
+          "right_little_2_joint": 80,
+          "right_middle_1_joint": 1150,
+          "right_middle_2_joint": 10,
+          "right_ring_1_joint": 857,
+          "right_ring_2_joint": 69,
+          "right_thumb_1_joint": 38,
+          "right_thumb_2_joint": 367,
+          "right_thumb_3_joint": 71,
+          "right_thumb_4_joint": 28
+        },
+        "joint_limit_hard_counts": {
+          "left_index_1_joint": 25,
+          "left_index_2_joint": 1,
+          "left_little_1_joint": 8,
+          "left_little_2_joint": 2,
+          "left_middle_1_joint": 10,
+          "left_middle_2_joint": 3,
+          "left_ring_1_joint": 14,
+          "left_ring_2_joint": 10,
+          "left_thumb_1_joint": 3,
+          "left_thumb_2_joint": 40,
+          "right_index_1_joint": 37,
+          "right_little_1_joint": 12,
+          "right_little_2_joint": 2,
+          "right_middle_1_joint": 45,
+          "right_middle_2_joint": 4,
+          "right_ring_1_joint": 49,
+          "right_ring_2_joint": 6,
+          "right_thumb_1_joint": 1,
+          "right_thumb_2_joint": 9,
+          "right_thumb_3_joint": 3,
+          "right_thumb_4_joint": 10
+        },
+        "joint_limit_hard_episode_events": 142,
+        "finger_joint_limit_saturation_counts": {
+          "left_index_1_joint": 123,
+          "left_little_1_joint": 130,
+          "left_middle_1_joint": 110,
+          "left_ring_1_joint": 145,
+          "left_ring_2_joint": 55,
+          "left_thumb_2_joint": 1353,
+          "right_index_1_joint": 812,
+          "right_little_1_joint": 225,
+          "right_little_2_joint": 65,
+          "right_middle_1_joint": 676,
+          "right_ring_1_joint": 534,
+          "right_ring_2_joint": 39,
+          "right_thumb_1_joint": 32,
+          "right_thumb_2_joint": 47,
+          "right_thumb_3_joint": 47
+        },
+        "joint_limit_max_excess_by_joint": {
+          "left_index_1_joint": 0.9848292469978333,
+          "left_index_2_joint": 0.02854819782078266,
+          "left_little_1_joint": 1.0754948854446411,
+          "left_little_2_joint": 0.11992128938436508,
+          "left_middle_1_joint": 1.0950721502304077,
+          "left_middle_2_joint": 0.040635135024785995,
+          "left_ring_1_joint": 1.0245829820632935,
+          "left_ring_2_joint": 0.5659329891204834,
+          "left_thumb_1_joint": 0.02052757702767849,
+          "left_thumb_2_joint": 0.8127384185791016,
+          "left_thumb_3_joint": 0.004792214836925268,
+          "left_thumb_4_joint": 0.0034056901931762695,
+          "right_index_1_joint": 1.2606998682022095,
+          "right_index_2_joint": 6.519936141557992e-05,
+          "right_little_1_joint": 1.6444499492645264,
+          "right_little_2_joint": 0.8218870162963867,
+          "right_middle_1_joint": 1.1572142839431763,
+          "right_middle_2_joint": 0.017551619559526443,
+          "right_ring_1_joint": 0.8695412874221802,
+          "right_ring_2_joint": 0.3739415407180786,
+          "right_thumb_1_joint": 0.34854838252067566,
+          "right_thumb_2_joint": 1.0910316705703735,
+          "right_thumb_3_joint": 0.7988921999931335,
+          "right_thumb_4_joint": 0.04604159668087959
+        },
+        "coverage": {
+          "hard_limit_source": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          },
+          "joint_role_manifest": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          }
+        },
+        "hard_limit_sources": {
+          "runtime.joint_pos_limits": 10
+        },
+        "soft_limit_sources": {
+          "runtime.soft_joint_pos_limits": 10
+        },
+        "unknown_runtime_joint_names": [],
+        "baseline_checks": {
+          "post_home": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          },
+          "post_reset": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          }
+        }
+      }
+    },
+    "robot_motion": {
+      "joint_acceleration_rms": 54.30332184461524,
+      "joint_effort_rms": 56.989484342180546,
+      "joint_jerk_rms": 5524.436829787346,
+      "joint_velocity_rms": 1.1177463837560624
+    },
+    "grasp": {
+      "kinematic_grasp_stability_index": 0.9756752204324011,
+      "mean_kinematic_grasp_stability": 0.5476756719477086,
+      "per_object_best_kinematic_grasp_stability": {
+        "obj_080_bottle_3": 0.5705879291621311,
+        "obj_341_banana_peel_1": 0.3622530188542202,
+        "obj_349_paper_low_2": 0.9696077133505674
+      },
+      "grasp_gsi_diagnostics": {
+        "mean_anti_slip_score": 0.9019469384984251,
+        "mean_duration_score": 0.2985466809434927,
+        "mean_hold_score": 0.2966979557272879,
+        "mean_kinematic_stability_score": 0.1956420106718901,
+        "mean_legacy_gsi": 0.41795077791011626,
+        "mean_lift_score": 0.39630676539031257,
+        "mean_motion_stability_score": 0.27402355721377775
+      }
+    },
+    "tool": {
+      "tool_selection_accuracy": null,
+      "tool_switch_success_rate": null,
+      "tool_switch_total": 0,
+      "tool_switch_successful": 0
+    },
+    "runtime": {
+      "avg_steps": 1744,
+      "error_count": 0
+    }
+  }
+}

gpt6_astra_1x_icl/per_task_eval/43_fridge_fruit_shelf_sorting/Full/summary.json

@@ -0,0 +1,341 @@
+{
+  "metric_schema_version": "core-v1.8-policy-step-query-split",
+  "metadata": {
+    "policy_name": "gpt-6-astra",
+    "base_seed": 204300000,
+    "seed_policy": "episode_seed = base_seed + episode_index - 1",
+    "randomization_recorded": true,
+    "n_episodes": 10,
+    "successes": 3,
+    "error_count": 1,
+    "evaluation_protocol": "reach_and_stop",
+    "early_stop": true,
+    "success_definition": "terminal.raw_condition holds for dwell_time_s",
+    "measurement_window": "episode_start_to_stable_success_or_failure_budget",
+    "max_episode_steps": 4125,
+    "dwell_time_s": 0.5,
+    "field_semantics": {
+      "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
+      "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
+      "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
+      "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
+      "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
+      "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
+      "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
+      "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
+      "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
+      "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
+      "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
+      "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
+      "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
+      "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
+      "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
+      "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
+      "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
+      "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
+      "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
+      "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
+      "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
+      "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
+      "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
+      "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
+      "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
+    }
+  },
+  "core_metrics": {
+    "completion": {
+      "success_rate": 0.3,
+      "success_rate_ci": {
+        "mean": 0.3,
+        "ci_low": 0.0,
+        "ci_high": 0.6
+      }
+    },
+    "progress": {
+      "latched_stage_completion_rate": 0.55
+    },
+    "efficiency": {
+      "avg_time_to_success_s": 63.69444444444211
+    },
+    "safety": {
+      "safe_success_rate": 0.1,
+      "hard_violation_rate": 0.9,
+      "drop_rate": 0.9,
+      "high_speed_violation_rate": 0.7,
+      "safety_violation_step_rate": 0.0013635194237200925
+    },
+    "robustness": {}
+  },
+  "diagnostic_metrics": {
+    "completion": {
+      "ever_instant_success_rate": 0.3,
+      "at_end_success_rate": 0.3,
+      "task_family_success_rate": {
+        "articulation_multi": 0.3
+      }
+    },
+    "progress": {
+      "current_stage_completion_rate": 0.35,
+      "current_chain_depth_progress_score": 0.5333333333333333,
+      "latched_chain_depth_progress_score": 0.6333333333333333,
+      "avg_current_chain_depth": 1.6,
+      "avg_latched_chain_depth": 1.9
+    },
+    "efficiency": {
+      "task_efficiency": 0.7231427052588435,
+      "expert_time_s": 45.833333333333336,
+      "expert_time_step": 2750,
+      "avg_steps_to_success": 3821.6666666666665,
+      "avg_policy_steps_to_success": 1274.3333333333333,
+      "avg_policy_queries_to_success": 33.333333333333336
+    },
+    "safety": {
+      "safety_violation_events_per_step": 0.0007203498842294829,
+      "violation_counts": {
+        "drop_to_table_obj_011_banana_2": 9,
+        "drop_to_table_obj_013_apple_1": 6,
+        "high_speed_obj_013_apple_1": 13
+      },
+      "robot_constraint_violation_rate": 1.0,
+      "finger_joint_limit_saturation_rate": 1.0,
+      "joint_limit_active_step_rate": 0.1631849755595575,
+      "finger_joint_limit_saturation_step_rate": 0.09773604322099305,
+      "max_joint_limit_excess_rad": 1.9151273965835571,
+      "robot_constraint_diagnostics": {
+        "joint_limit_soft_counts": {
+          "FFJ2": 110,
+          "FFJ3": 155,
+          "LFJ2": 146,
+          "LFJ3": 163,
+          "LFJ4": 211,
+          "LFJ5": 515,
+          "MFJ1": 24,
+          "MFJ2": 270,
+          "MFJ3": 210,
+          "MFJ4": 32,
+          "RFJ1": 23,
+          "RFJ2": 583,
+          "RFJ3": 218,
+          "RFJ4": 30,
+          "THJ1": 44,
+          "THJ3": 28,
+          "THJ4": 195,
+          "THJ5": 535,
+          "WRJ1": 209,
+          "WRJ2": 296,
+          "l_FFJ2": 486,
+          "l_FFJ3": 1310,
+          "l_LFJ2": 398,
+          "l_LFJ3": 709,
+          "l_LFJ4": 85,
+          "l_LFJ5": 1343,
+          "l_MFJ1": 1,
+          "l_MFJ2": 1429,
+          "l_MFJ3": 812,
+          "l_MFJ4": 80,
+          "l_RFJ1": 1,
+          "l_RFJ2": 1563,
+          "l_RFJ3": 250,
+          "l_RFJ4": 67,
+          "l_THJ1": 14,
+          "l_THJ2": 90,
+          "l_THJ3": 35,
+          "l_THJ4": 480,
+          "l_THJ5": 467,
+          "l_WRJ1": 84,
+          "l_WRJ2": 388
+        },
+        "joint_limit_hard_counts": {
+          "FFJ2": 17,
+          "FFJ3": 4,
+          "FFJ4": 3,
+          "LFJ1": 2,
+          "LFJ2": 14,
+          "LFJ3": 13,
+          "LFJ4": 22,
+          "LFJ5": 47,
+          "MFJ1": 7,
+          "MFJ2": 16,
+          "MFJ3": 12,
+          "MFJ4": 8,
+          "RFJ1": 9,
+          "RFJ2": 28,
+          "RFJ3": 10,
+          "RFJ4": 10,
+          "THJ1": 1,
+          "THJ3": 1,
+          "THJ4": 4,
+          "THJ5": 67,
+          "WRJ1": 3,
+          "WRJ2": 6,
+          "l_FFJ2": 131,
+          "l_FFJ3": 43,
+          "l_LFJ2": 31,
+          "l_LFJ3": 31,
+          "l_LFJ4": 15,
+          "l_LFJ5": 204,
+          "l_MFJ1": 2,
+          "l_MFJ2": 86,
+          "l_MFJ3": 61,
+          "l_MFJ4": 7,
+          "l_RFJ1": 1,
+          "l_RFJ2": 95,
+          "l_RFJ3": 33,
+          "l_RFJ4": 11,
+          "l_THJ1": 1,
+          "l_THJ2": 2,
+          "l_THJ3": 1,
+          "l_THJ4": 188,
+          "l_THJ5": 38,
+          "l_WRJ1": 6,
+          "l_WRJ2": 12
+        },
+        "joint_limit_hard_episode_events": 468,
+        "finger_joint_limit_saturation_counts": {
+          "FFJ2": 13,
+          "FFJ3": 82,
+          "LFJ2": 105,
+          "LFJ3": 43,
+          "LFJ4": 102,
+          "LFJ5": 350,
+          "MFJ2": 127,
+          "MFJ3": 113,
+          "RFJ2": 354,
+          "RFJ3": 123,
+          "THJ1": 42,
+          "THJ3": 24,
+          "THJ4": 91,
+          "THJ5": 322,
+          "l_FFJ2": 272,
+          "l_FFJ3": 395,
+          "l_LFJ2": 168,
+          "l_LFJ3": 403,
+          "l_LFJ4": 70,
+          "l_LFJ5": 533,
+          "l_MFJ2": 597,
+          "l_MFJ3": 282,
+          "l_MFJ4": 40,
+          "l_RFJ2": 925,
+          "l_RFJ3": 47,
+          "l_RFJ4": 32,
+          "l_THJ2": 81,
+          "l_THJ3": 31,
+          "l_THJ4": 157,
+          "l_THJ5": 340
+        },
+        "joint_limit_max_excess_by_joint": {
+          "FFJ1": 0.004584662150591612,
+          "FFJ2": 0.5536534786224365,
+          "FFJ3": 0.339205265045166,
+          "FFJ4": 0.005583912134170532,
+          "LFJ1": 0.006359064485877752,
+          "LFJ2": 0.6927093267440796,
+          "LFJ3": 0.42599403858184814,
+          "LFJ4": 0.6373152732849121,
+          "LFJ5": 1.2247565984725952,
+          "MFJ1": 0.24392759799957275,
+          "MFJ2": 0.7013113498687744,
+          "MFJ3": 0.2703166604042053,
+          "MFJ4": 0.1731272041797638,
+          "RFJ1": 0.13311851024627686,
+          "RFJ2": 0.9399458169937134,
+          "RFJ3": 0.38790106773376465,
+          "RFJ4": 0.12483513355255127,
+          "THJ1": 1.548326015472412,
+          "THJ3": 0.6530141234397888,
+          "THJ4": 0.7885133028030396,
+          "THJ5": 1.0876133441925049,
+          "WRJ1": 1.4698245525360107,
+          "WRJ2": 0.6440258622169495,
+          "l_FFJ1": 0.0005501087289303541,
+          "l_FFJ2": 1.1656535863876343,
+          "l_FFJ3": 1.08669114112854,
+          "l_LFJ1": 2.3546695047116373e-06,
+          "l_LFJ2": 0.4393867254257202,
+          "l_LFJ3": 1.440045952796936,
+          "l_LFJ4": 0.6306201219558716,
+          "l_LFJ5": 1.5674127340316772,
+          "l_MFJ1": 0.01010024081915617,
+          "l_MFJ2": 1.9151273965835571,
+          "l_MFJ3": 1.060526728630066,
+          "l_MFJ4": 0.7785158157348633,
+          "l_RFJ1": 0.01199818029999733,
+          "l_RFJ2": 1.4461613893508911,
+          "l_RFJ3": 0.6890462040901184,
+          "l_RFJ4": 0.4180395305156708,
+          "l_THJ1": 0.1456013321876526,
+          "l_THJ2": 1.0489020347595215,
+          "l_THJ3": 1.1338406801223755,
+          "l_THJ4": 0.6453428268432617,
+          "l_THJ5": 1.0347974300384521,
+          "l_WRJ1": 0.3298899829387665,
+          "l_WRJ2": 0.4654872417449951
+        },
+        "coverage": {
+          "hard_limit_source": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          },
+          "joint_role_manifest": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          }
+        },
+        "hard_limit_sources": {
+          "runtime.joint_pos_limits": 10
+        },
+        "soft_limit_sources": {
+          "runtime.soft_joint_pos_limits": 10
+        },
+        "unknown_runtime_joint_names": [],
+        "baseline_checks": {
+          "post_home": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          },
+          "post_reset": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          }
+        }
+      }
+    },
+    "robot_motion": {
+      "joint_acceleration_rms": 21.6003658770578,
+      "joint_effort_rms": 52.903143749283146,
+      "joint_jerk_rms": 2143.0483888617514,
+      "joint_velocity_rms": 0.5002505032033382
+    },
+    "grasp": {
+      "kinematic_grasp_stability_index": 0.9489701401741613,
+      "mean_kinematic_grasp_stability": 0.4333703160657618,
+      "per_object_best_kinematic_grasp_stability": {
+        "obj_011_banana_2": 0.8037451133980107,
+        "obj_013_apple_1": 0.9116650967013875
+      },
+      "grasp_gsi_diagnostics": {
+        "mean_anti_slip_score": 0.7634205380824515,
+        "mean_duration_score": 0.6876188743290731,
+        "mean_hold_score": 0.31131175593053584,
+        "mean_kinematic_stability_score": 0.16574074223314023,
+        "mean_legacy_gsi": 0.47877921251336375,
+        "mean_lift_score": 0.402605518136838,
+        "mean_motion_stability_score": 0.3023697254228749
+      }
+    },
+    "tool": {
+      "tool_selection_accuracy": null,
+      "tool_switch_success_rate": null,
+      "tool_switch_total": 0,
+      "tool_switch_successful": 0
+    },
+    "runtime": {
+      "avg_steps": 3887,
+      "error_count": 1
+    }
+  }
+}

gpt6_astra_1x_icl/per_task_eval/44_microwave_bowl_loading/Full/summary.json

@@ -0,0 +1,294 @@
+{
+  "metric_schema_version": "core-v1.8-policy-step-query-split",
+  "metadata": {
+    "policy_name": "gpt-6-astra",
+    "base_seed": 204400000,
+    "seed_policy": "episode_seed = base_seed + episode_index - 1",
+    "randomization_recorded": true,
+    "n_episodes": 10,
+    "successes": 4,
+    "error_count": 0,
+    "evaluation_protocol": "reach_and_stop",
+    "early_stop": true,
+    "success_definition": "terminal.raw_condition holds for dwell_time_s",
+    "measurement_window": "episode_start_to_stable_success_or_failure_budget",
+    "max_episode_steps": 3183,
+    "dwell_time_s": 0.5,
+    "field_semantics": {
+      "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
+      "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
+      "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
+      "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
+      "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
+      "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
+      "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
+      "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
+      "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
+      "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
+      "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
+      "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
+      "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
+      "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
+      "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
+      "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
+      "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
+      "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
+      "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
+      "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
+      "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
+      "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
+      "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
+      "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
+      "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
+    }
+  },
+  "core_metrics": {
+    "completion": {
+      "success_rate": 0.4,
+      "success_rate_ci": {
+        "mean": 0.4,
+        "ci_low": 0.1,
+        "ci_high": 0.7
+      }
+    },
+    "progress": {
+      "latched_stage_completion_rate": 0.625
+    },
+    "efficiency": {
+      "avg_time_to_success_s": 42.0874999999989
+    },
+    "safety": {
+      "safe_success_rate": 0.4,
+      "hard_violation_rate": 0.6,
+      "drop_rate": 0.6,
+      "high_speed_violation_rate": 0.1,
+      "safety_violation_step_rate": 0.0004109729785266619
+    },
+    "robustness": {}
+  },
+  "diagnostic_metrics": {
+    "completion": {
+      "ever_instant_success_rate": 0.4,
+      "at_end_success_rate": 0.4,
+      "task_family_success_rate": {
+        "articulation_door_place": 0.4
+      }
+    },
+    "progress": {
+      "current_stage_completion_rate": 0.425,
+      "current_chain_depth_progress_score": 0.525,
+      "latched_chain_depth_progress_score": 0.625,
+      "avg_current_chain_depth": 2.1,
+      "avg_latched_chain_depth": 2.5
+    },
+    "efficiency": {
+      "task_efficiency": 0.8481173070861898,
+      "expert_time_s": 35.36666666666667,
+      "expert_time_step": 2122,
+      "avg_steps_to_success": 2525.25,
+      "avg_policy_steps_to_success": 842.25,
+      "avg_policy_queries_to_success": 22.5
+    },
+    "safety": {
+      "safety_violation_events_per_step": 0.0003082297338949964,
+      "violation_counts": {
+        "drop_to_table_obj_024_bowl_3": 2,
+        "drop_to_table_obj_163_baguette_1": 6,
+        "high_speed_obj_163_baguette_1": 1
+      },
+      "robot_constraint_violation_rate": 1.0,
+      "finger_joint_limit_saturation_rate": 1.0,
+      "joint_limit_active_step_rate": 0.15298469125654987,
+      "finger_joint_limit_saturation_step_rate": 0.0633925819377376,
+      "max_joint_limit_excess_rad": 3.174018144607544,
+      "robot_constraint_diagnostics": {
+        "joint_limit_soft_counts": {
+          "left_hand_Finger_Spread": 19,
+          "left_hand_Index_Finger_Distal": 61,
+          "left_hand_Index_Finger_Proximal": 475,
+          "left_hand_Middle_Finger_Distal": 296,
+          "left_hand_Middle_Finger_Proximal": 933,
+          "left_hand_Pinky": 132,
+          "left_hand_Ring_Finger": 1062,
+          "left_hand_Thumb_Flexion": 1116,
+          "left_hand_Thumb_Opposition": 1,
+          "left_hand_j12": 86,
+          "left_hand_j13": 184,
+          "left_hand_j3": 626,
+          "left_hand_j5": 568,
+          "left_hand_ring_spread": 17,
+          "right_hand_Index_Finger_Distal": 590,
+          "right_hand_Index_Finger_Proximal": 12,
+          "right_hand_Middle_Finger_Distal": 23,
+          "right_hand_Middle_Finger_Proximal": 50,
+          "right_hand_Pinky": 1,
+          "right_hand_Ring_Finger": 2,
+          "right_hand_Thumb_Flexion": 281,
+          "right_hand_index_spread": 9,
+          "right_hand_j12": 13,
+          "right_hand_j13": 3,
+          "right_hand_j3": 260,
+          "right_hand_j5": 1,
+          "right_hand_ring_spread": 10
+        },
+        "joint_limit_hard_counts": {
+          "left_hand_Finger_Spread": 41,
+          "left_hand_Index_Finger_Distal": 2,
+          "left_hand_Index_Finger_Proximal": 15,
+          "left_hand_Middle_Finger_Distal": 7,
+          "left_hand_Middle_Finger_Proximal": 29,
+          "left_hand_Pinky": 9,
+          "left_hand_Ring_Finger": 25,
+          "left_hand_Thumb_Flexion": 30,
+          "left_hand_Thumb_Opposition": 2,
+          "left_hand_index_spread": 2,
+          "left_hand_j12": 3,
+          "left_hand_j13": 7,
+          "left_hand_j3": 14,
+          "left_hand_j5": 83,
+          "left_hand_ring_spread": 45,
+          "right_hand_Index_Finger_Distal": 21,
+          "right_hand_Index_Finger_Proximal": 23,
+          "right_hand_Middle_Finger_Distal": 21,
+          "right_hand_Middle_Finger_Proximal": 31,
+          "right_hand_Pinky": 9,
+          "right_hand_Ring_Finger": 17,
+          "right_hand_Thumb_Flexion": 51,
+          "right_hand_index_spread": 1,
+          "right_hand_j12": 6,
+          "right_hand_j13": 9,
+          "right_hand_j15": 1,
+          "right_hand_j3": 63,
+          "right_hand_j5": 1,
+          "right_hand_ring_spread": 1
+        },
+        "joint_limit_hard_episode_events": 224,
+        "finger_joint_limit_saturation_counts": {
+          "left_hand_Index_Finger_Distal": 53,
+          "left_hand_Index_Finger_Proximal": 200,
+          "left_hand_Middle_Finger_Distal": 183,
+          "left_hand_Middle_Finger_Proximal": 483,
+          "left_hand_Pinky": 40,
+          "left_hand_Ring_Finger": 510,
+          "left_hand_Thumb_Flexion": 261,
+          "left_hand_j12": 14,
+          "left_hand_j13": 91,
+          "left_hand_j3": 323,
+          "left_hand_j5": 362,
+          "right_hand_Index_Finger_Distal": 14,
+          "right_hand_Middle_Finger_Distal": 17,
+          "right_hand_Middle_Finger_Proximal": 15,
+          "right_hand_Thumb_Flexion": 91,
+          "right_hand_index_spread": 2,
+          "right_hand_j12": 4,
+          "right_hand_j3": 98
+        },
+        "joint_limit_max_excess_by_joint": {
+          "left_hand_Finger_Spread": 0.037375323474407196,
+          "left_hand_Index_Finger_Distal": 0.8622639179229736,
+          "left_hand_Index_Finger_Proximal": 1.5071290731430054,
+          "left_hand_Middle_Finger_Distal": 0.39120006561279297,
+          "left_hand_Middle_Finger_Proximal": 1.09234619140625,
+          "left_hand_Pinky": 0.3397405743598938,
+          "left_hand_Ring_Finger": 0.5145901441574097,
+          "left_hand_Thumb_Flexion": 0.747669517993927,
+          "left_hand_Thumb_Opposition": 0.010924256406724453,
+          "left_hand_index_spread": 0.006348968017846346,
+          "left_hand_j12": 0.1434175968170166,
+          "left_hand_j13": 0.7270541191101074,
+          "left_hand_j14": 3.901387025939673e-12,
+          "left_hand_j15": 1.140336089444638e-12,
+          "left_hand_j16": 8.541977888909535e-12,
+          "left_hand_j17": 7.827029822882192e-12,
+          "left_hand_j3": 0.3234946131706238,
+          "left_hand_j4": 1.0819940863410205e-12,
+          "left_hand_j5": 3.174018144607544,
+          "left_hand_ring_spread": 0.12098025530576706,
+          "right_hand_Finger_Spread": 0.0024728020653128624,
+          "right_hand_Index_Finger_Distal": 0.5602082014083862,
+          "right_hand_Index_Finger_Proximal": 0.050545454025268555,
+          "right_hand_Middle_Finger_Distal": 0.5714244842529297,
+          "right_hand_Middle_Finger_Proximal": 0.5611324310302734,
+          "right_hand_Pinky": 0.010315727442502975,
+          "right_hand_Ring_Finger": 0.015359331853687763,
+          "right_hand_Thumb_Flexion": 0.4125230014324188,
+          "right_hand_index_spread": 0.31571322679519653,
+          "right_hand_j12": 0.29430997371673584,
+          "right_hand_j13": 0.01638295128941536,
+          "right_hand_j14": 0.0032558327075093985,
+          "right_hand_j15": 0.006125228479504585,
+          "right_hand_j16": 8.439960536010815e-11,
+          "right_hand_j17": 3.214399579842819e-11,
+          "right_hand_j3": 0.6654275059700012,
+          "right_hand_j4": 0.003800993785262108,
+          "right_hand_j5": 0.02329612523317337,
+          "right_hand_ring_spread": 0.08119967579841614
+        },
+        "coverage": {
+          "hard_limit_source": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          },
+          "joint_role_manifest": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          }
+        },
+        "hard_limit_sources": {
+          "runtime.joint_pos_limits": 10
+        },
+        "soft_limit_sources": {
+          "runtime.soft_joint_pos_limits": 10
+        },
+        "unknown_runtime_joint_names": [],
+        "baseline_checks": {
+          "post_home": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          },
+          "post_reset": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          }
+        }
+      }
+    },
+    "robot_motion": {
+      "joint_acceleration_rms": 15.823111828504633,
+      "joint_effort_rms": 28.26060521301849,
+      "joint_jerk_rms": 1579.5593131692594,
+      "joint_velocity_rms": 0.3536156621531375
+    },
+    "grasp": {
+      "kinematic_grasp_stability_index": 0.96396805981083,
+      "mean_kinematic_grasp_stability": 0.5654986083088973,
+      "per_object_best_kinematic_grasp_stability": {
+        "obj_024_bowl_3": 0.96396805981083,
+        "obj_163_baguette_1": 0.5525038150096471
+      },
+      "grasp_gsi_diagnostics": {
+        "mean_anti_slip_score": 0.7031216264935697,
+        "mean_duration_score": 0.4613686127754932,
+        "mean_hold_score": 0.21695957292544113,
+        "mean_kinematic_stability_score": 0.07164192046786215,
+        "mean_legacy_gsi": 0.35909566133479026,
+        "mean_lift_score": 0.2872173414761005,
+        "mean_motion_stability_score": 0.19473740983633647
+      }
+    },
+    "tool": {
+      "tool_selection_accuracy": null,
+      "tool_switch_success_rate": null,
+      "tool_switch_total": 0,
+      "tool_switch_successful": 0
+    },
+    "runtime": {
+      "avg_steps": 2919.9,
+      "error_count": 0
+    }
+  }
+}

gpt6_astra_1x_icl/per_task_eval/51_toilet_lid_cleaner_pour/Full/summary.json

@@ -0,0 +1,251 @@
+{
+  "metric_schema_version": "core-v1.8-policy-step-query-split",
+  "metadata": {
+    "policy_name": "gpt-6-astra",
+    "base_seed": 205100000,
+    "seed_policy": "episode_seed = base_seed + episode_index - 1",
+    "randomization_recorded": true,
+    "n_episodes": 10,
+    "successes": 6,
+    "error_count": 1,
+    "evaluation_protocol": "reach_and_stop",
+    "early_stop": true,
+    "success_definition": "terminal.raw_condition holds for dwell_time_s",
+    "measurement_window": "episode_start_to_stable_success_or_failure_budget",
+    "max_episode_steps": 3069,
+    "dwell_time_s": 0.5,
+    "field_semantics": {
+      "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
+      "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
+      "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
+      "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
+      "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
+      "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
+      "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
+      "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
+      "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
+      "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
+      "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
+      "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
+      "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
+      "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
+      "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
+      "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
+      "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
+      "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
+      "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
+      "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
+      "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
+      "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
+      "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
+      "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
+      "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
+    }
+  },
+  "core_metrics": {
+    "completion": {
+      "success_rate": 0.6,
+      "success_rate_ci": {
+        "mean": 0.6,
+        "ci_low": 0.3,
+        "ci_high": 0.9
+      }
+    },
+    "progress": {
+      "latched_stage_completion_rate": 0.8333333333333334
+    },
+    "efficiency": {
+      "avg_time_to_success_s": 37.361111111110276
+    },
+    "safety": {
+      "safe_success_rate": 0.4,
+      "hard_violation_rate": 0.3,
+      "drop_rate": 0.2,
+      "high_speed_violation_rate": 0.2,
+      "safety_violation_step_rate": 0.00017086715079026057
+    },
+    "robustness": {}
+  },
+  "diagnostic_metrics": {
+    "completion": {
+      "ever_instant_success_rate": 0.7,
+      "at_end_success_rate": 0.7,
+      "task_family_success_rate": {
+        "articulation_multi": 0.6
+      }
+    },
+    "progress": {
+      "current_stage_completion_rate": 0.5,
+      "current_chain_depth_progress_score": 0.7666666666666666,
+      "latched_chain_depth_progress_score": 0.8333333333333334,
+      "avg_current_chain_depth": 2.3,
+      "avg_latched_chain_depth": 2.5
+    },
+    "efficiency": {
+      "task_efficiency": 0.9439523668688949,
+      "expert_time_s": 34.1,
+      "expert_time_step": 2046,
+      "avg_steps_to_success": 2241.6666666666665,
+      "avg_policy_steps_to_success": 747.6666666666666,
+      "avg_policy_queries_to_success": 25.666666666666668
+    },
+    "safety": {
+      "safety_violation_events_per_step": 0.00017086715079026057,
+      "violation_counts": {
+        "drop_to_table_obj_021_bleach_cleanser_2": 2,
+        "high_speed_obj_021_bleach_cleanser_2": 2
+      },
+      "robot_constraint_violation_rate": 1.0,
+      "finger_joint_limit_saturation_rate": 1.0,
+      "joint_limit_active_step_rate": 0.18752669799231098,
+      "finger_joint_limit_saturation_step_rate": 0.11828278513455788,
+      "max_joint_limit_excess_rad": 2.1122384071350098,
+      "robot_constraint_diagnostics": {
+        "joint_limit_soft_counts": {
+          "joint7": 174,
+          "leap_l_1": 142,
+          "leap_l_10": 1303,
+          "leap_l_11": 27,
+          "leap_l_12": 226,
+          "leap_l_13": 102,
+          "leap_l_14": 24,
+          "leap_l_15": 1,
+          "leap_l_2": 948,
+          "leap_l_5": 28,
+          "leap_l_6": 80,
+          "leap_l_9": 72,
+          "multi_leap_r_1": 3,
+          "multi_leap_r_10": 35,
+          "multi_leap_r_11": 43,
+          "multi_leap_r_2": 7,
+          "multi_leap_r_5": 1341,
+          "multi_leap_r_6": 124,
+          "multi_leap_r_9": 2083
+        },
+        "joint_limit_hard_counts": {
+          "joint7": 3,
+          "leap_l_1": 6,
+          "leap_l_10": 146,
+          "leap_l_11": 32,
+          "leap_l_12": 48,
+          "leap_l_13": 9,
+          "leap_l_14": 1,
+          "leap_l_15": 2,
+          "leap_l_2": 179,
+          "leap_l_3": 1,
+          "leap_l_5": 1,
+          "leap_l_6": 3,
+          "leap_l_9": 1,
+          "multi_leap_r_1": 3,
+          "multi_leap_r_10": 5,
+          "multi_leap_r_11": 1,
+          "multi_leap_r_2": 5,
+          "multi_leap_r_5": 2,
+          "multi_leap_r_6": 2,
+          "multi_leap_r_9": 19
+        },
+        "joint_limit_hard_episode_events": 277,
+        "finger_joint_limit_saturation_counts": {
+          "leap_l_1": 45,
+          "leap_l_10": 457,
+          "leap_l_12": 129,
+          "leap_l_13": 52,
+          "leap_l_14": 19,
+          "leap_l_2": 376,
+          "leap_l_6": 45,
+          "leap_l_9": 23,
+          "multi_leap_r_11": 29,
+          "multi_leap_r_5": 1279,
+          "multi_leap_r_6": 108,
+          "multi_leap_r_9": 1874
+        },
+        "joint_limit_max_excess_by_joint": {
+          "joint7": 0.5528488159179688,
+          "leap_l_1": 0.3120039701461792,
+          "leap_l_10": 2.1122384071350098,
+          "leap_l_11": 0.21100544929504395,
+          "leap_l_12": 0.7742586135864258,
+          "leap_l_13": 0.5815486907958984,
+          "leap_l_14": 0.5133529901504517,
+          "leap_l_15": 0.032002806663513184,
+          "leap_l_2": 1.4947503805160522,
+          "leap_l_3": 0.0061879754066467285,
+          "leap_l_5": 0.10505503416061401,
+          "leap_l_6": 0.37968534231185913,
+          "leap_l_9": 0.2436368465423584,
+          "multi_leap_r_1": 0.015098214149475098,
+          "multi_leap_r_10": 0.2028316855430603,
+          "multi_leap_r_11": 0.3777793049812317,
+          "multi_leap_r_2": 0.07765775918960571,
+          "multi_leap_r_3": 0.003427267074584961,
+          "multi_leap_r_5": 1.4568877220153809,
+          "multi_leap_r_6": 1.092919945716858,
+          "multi_leap_r_9": 1.5552098751068115
+        },
+        "coverage": {
+          "hard_limit_source": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          },
+          "joint_role_manifest": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          }
+        },
+        "hard_limit_sources": {
+          "runtime.joint_pos_limits": 10
+        },
+        "soft_limit_sources": {
+          "runtime.soft_joint_pos_limits": 10
+        },
+        "unknown_runtime_joint_names": [],
+        "baseline_checks": {
+          "post_home": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          },
+          "post_reset": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          }
+        }
+      }
+    },
+    "robot_motion": {
+      "joint_acceleration_rms": 28.7850080674395,
+      "joint_effort_rms": 70.2196282205132,
+      "joint_jerk_rms": 2909.0275627193028,
+      "joint_velocity_rms": 0.8121275827835532
+    },
+    "grasp": {
+      "kinematic_grasp_stability_index": 0.8292787128415665,
+      "mean_kinematic_grasp_stability": 0.47345078266978513,
+      "per_object_best_kinematic_grasp_stability": {
+        "obj_021_bleach_cleanser_2": 0.8292787128415665
+      },
+      "grasp_gsi_diagnostics": {
+        "mean_anti_slip_score": 0.7595510349149899,
+        "mean_duration_score": 0.46520827010389804,
+        "mean_hold_score": 0.3410071222080645,
+        "mean_kinematic_stability_score": 0.15772402105129632,
+        "mean_legacy_gsi": 0.4430089669697321,
+        "mean_lift_score": 0.41464080583433066,
+        "mean_motion_stability_score": 0.3092388872798334
+      }
+    },
+    "tool": {
+      "tool_selection_accuracy": 0.9,
+      "tool_switch_success_rate": 0.0,
+      "tool_switch_total": 26,
+      "tool_switch_successful": 0
+    },
+    "runtime": {
+      "avg_steps": 2341,
+      "error_count": 1
+    }
+  }
+}

gpt6_astra_1x_icl/per_task_eval/60_breadbasket_fast_food_loading/Full/summary.json

@@ -0,0 +1,314 @@
+{
+  "metric_schema_version": "core-v1.8-policy-step-query-split",
+  "metadata": {
+    "policy_name": "gpt-6-astra",
+    "base_seed": 206000000,
+    "seed_policy": "episode_seed = base_seed + episode_index - 1",
+    "randomization_recorded": true,
+    "n_episodes": 10,
+    "successes": 0,
+    "error_count": 0,
+    "evaluation_protocol": "reach_and_stop",
+    "early_stop": true,
+    "success_definition": "terminal.raw_condition holds for dwell_time_s",
+    "measurement_window": "episode_start_to_stable_success_or_failure_budget",
+    "max_episode_steps": 2985,
+    "dwell_time_s": 0.5,
+    "field_semantics": {
+      "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
+      "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
+      "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
+      "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
+      "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
+      "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
+      "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
+      "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
+      "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
+      "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
+      "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
+      "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
+      "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
+      "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
+      "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
+      "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
+      "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
+      "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
+      "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
+      "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
+      "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
+      "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
+      "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
+      "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
+      "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
+    }
+  },
+  "core_metrics": {
+    "completion": {
+      "success_rate": 0.0,
+      "success_rate_ci": {
+        "mean": 0.0,
+        "ci_low": 0.0,
+        "ci_high": 0.0
+      }
+    },
+    "progress": {
+      "latched_stage_completion_rate": 0.0
+    },
+    "efficiency": {
+      "avg_time_to_success_s": null
+    },
+    "safety": {
+      "safe_success_rate": 0.0,
+      "hard_violation_rate": 0.8,
+      "drop_rate": 0.8,
+      "high_speed_violation_rate": 0.3,
+      "safety_violation_step_rate": 0.03296482412060302
+    },
+    "robustness": {}
+  },
+  "diagnostic_metrics": {
+    "completion": {
+      "ever_instant_success_rate": 0.0,
+      "at_end_success_rate": 0.0,
+      "task_family_success_rate": {
+        "rigid_pick_place": 0.0
+      }
+    },
+    "progress": {
+      "current_stage_completion_rate": 0.0,
+      "current_chain_depth_progress_score": 0.0,
+      "latched_chain_depth_progress_score": 0.0,
+      "avg_current_chain_depth": 0,
+      "avg_latched_chain_depth": 0
+    },
+    "efficiency": {
+      "task_efficiency": null,
+      "expert_time_s": 33.15,
+      "expert_time_step": 1989,
+      "avg_steps_to_success": null,
+      "avg_policy_steps_to_success": null,
+      "avg_policy_queries_to_success": null
+    },
+    "safety": {
+      "safety_violation_events_per_step": 0.0008040201005025126,
+      "violation_counts": {
+        "drop_obj_163_baguette_5": 2,
+        "drop_obj_182_bread_4": 1,
+        "drop_to_table_obj_128_hamburg_3": 2,
+        "drop_to_table_obj_163_baguette_5": 5,
+        "drop_to_table_obj_182_bread_4": 7,
+        "high_speed_distractor_01_clutter_006_mustard_bottle": 1,
+        "high_speed_distractor_01_clutter_019_pitcher_base": 1,
+        "high_speed_obj_163_baguette_5": 2,
+        "high_speed_obj_182_bread_4": 1,
+        "high_speed_obj_183_breadbasket_1": 2
+      },
+      "robot_constraint_violation_rate": 1.0,
+      "finger_joint_limit_saturation_rate": 1.0,
+      "joint_limit_active_step_rate": 0.17688442211055277,
+      "finger_joint_limit_saturation_step_rate": 0.12254606365159129,
+      "max_joint_limit_excess_rad": 3.084315538406372,
+      "robot_constraint_diagnostics": {
+        "joint_limit_soft_counts": {
+          "left_index_dip_joint": 12,
+          "left_index_mcp_joint": 32,
+          "left_index_pip_joint": 24,
+          "left_index_pitch_joint": 26,
+          "left_middle_dip_joint": 8,
+          "left_middle_mcp_joint": 9,
+          "left_middle_pip_joint": 41,
+          "left_middle_pitch_joint": 2,
+          "left_pinky_mcp_joint": 89,
+          "left_pinky_pip_joint": 30,
+          "left_ring_dip_joint": 3,
+          "left_ring_mcp_joint": 1,
+          "left_ring_pip_joint": 78,
+          "left_thumb_mcp_joint": 174,
+          "left_thumb_pip_joint": 5,
+          "right_index_dip_joint": 13,
+          "right_index_mcp_joint": 20,
+          "right_index_pip_joint": 210,
+          "right_index_pitch_joint": 1316,
+          "right_middle_dip_joint": 58,
+          "right_middle_mcp_joint": 4,
+          "right_middle_pip_joint": 701,
+          "right_middle_pitch_joint": 974,
+          "right_pinky_dip_joint": 8,
+          "right_pinky_mcp_joint": 2924,
+          "right_pinky_pip_joint": 511,
+          "right_ring_dip_joint": 165,
+          "right_ring_mcp_joint": 1679,
+          "right_ring_pip_joint": 1189,
+          "right_thumb_dip_joint": 130,
+          "right_thumb_mcp_joint": 794,
+          "right_thumb_pip_joint": 98
+        },
+        "joint_limit_hard_counts": {
+          "left_index_dip_joint": 1,
+          "left_index_mcp_joint": 1,
+          "left_index_pip_joint": 5,
+          "left_index_pitch_joint": 3,
+          "left_middle_dip_joint": 2,
+          "left_middle_mcp_joint": 1,
+          "left_middle_pip_joint": 2,
+          "left_middle_pitch_joint": 1,
+          "left_pinky_mcp_joint": 5,
+          "left_pinky_pip_joint": 3,
+          "left_ring_dip_joint": 2,
+          "left_ring_mcp_joint": 1,
+          "left_ring_pip_joint": 6,
+          "left_thumb_mcp_joint": 11,
+          "left_thumb_pip_joint": 1,
+          "right_index_dip_joint": 1,
+          "right_index_mcp_joint": 2,
+          "right_index_pip_joint": 58,
+          "right_index_pitch_joint": 30,
+          "right_middle_dip_joint": 8,
+          "right_middle_mcp_joint": 1,
+          "right_middle_pip_joint": 58,
+          "right_middle_pitch_joint": 35,
+          "right_pinky_dip_joint": 4,
+          "right_pinky_mcp_joint": 50,
+          "right_pinky_pip_joint": 40,
+          "right_ring_dip_joint": 6,
+          "right_ring_mcp_joint": 36,
+          "right_ring_pip_joint": 151,
+          "right_thumb_dip_joint": 5,
+          "right_thumb_mcp_joint": 24,
+          "right_thumb_pip_joint": 3,
+          "right_thumb_yaw_joint": 2
+        },
+        "joint_limit_hard_episode_events": 170,
+        "finger_joint_limit_saturation_counts": {
+          "left_index_dip_joint": 3,
+          "left_index_mcp_joint": 27,
+          "left_index_pip_joint": 15,
+          "left_index_pitch_joint": 2,
+          "left_middle_pip_joint": 31,
+          "left_pinky_mcp_joint": 50,
+          "left_pinky_pip_joint": 5,
+          "left_ring_pip_joint": 37,
+          "left_thumb_mcp_joint": 68,
+          "right_index_dip_joint": 7,
+          "right_index_pip_joint": 80,
+          "right_index_pitch_joint": 909,
+          "right_middle_dip_joint": 31,
+          "right_middle_pip_joint": 314,
+          "right_middle_pitch_joint": 631,
+          "right_pinky_mcp_joint": 2226,
+          "right_pinky_pip_joint": 284,
+          "right_ring_dip_joint": 120,
+          "right_ring_mcp_joint": 1252,
+          "right_ring_pip_joint": 453,
+          "right_thumb_dip_joint": 105,
+          "right_thumb_mcp_joint": 476,
+          "right_thumb_pip_joint": 73
+        },
+        "joint_limit_max_excess_by_joint": {
+          "left_index_dip_joint": 0.3007436990737915,
+          "left_index_mcp_joint": 1.2541817426681519,
+          "left_index_pip_joint": 0.5510872006416321,
+          "left_index_pitch_joint": 0.14136695861816406,
+          "left_middle_dip_joint": 0.061401475220918655,
+          "left_middle_mcp_joint": 0.1364496946334839,
+          "left_middle_pip_joint": 0.9280629754066467,
+          "left_middle_pitch_joint": 0.01650257222354412,
+          "left_pinky_dip_joint": 1.1730775440810248e-05,
+          "left_pinky_mcp_joint": 0.43559882044792175,
+          "left_pinky_pip_joint": 0.43403732776641846,
+          "left_ring_dip_joint": 0.04406795650720596,
+          "left_ring_mcp_joint": 0.011595302261412144,
+          "left_ring_pip_joint": 0.9072043895721436,
+          "left_thumb_dip_joint": 0.0003824085579253733,
+          "left_thumb_mcp_joint": 0.2585655152797699,
+          "left_thumb_pip_joint": 0.18143272399902344,
+          "left_thumb_yaw_joint": 0.00297143729403615,
+          "right_index_dip_joint": 0.5543527603149414,
+          "right_index_mcp_joint": 0.22716423869132996,
+          "right_index_pip_joint": 1.683140516281128,
+          "right_index_pitch_joint": 1.3602131605148315,
+          "right_middle_dip_joint": 0.8403565883636475,
+          "right_middle_mcp_joint": 0.0468006432056427,
+          "right_middle_pip_joint": 3.084315538406372,
+          "right_middle_pitch_joint": 1.2197375297546387,
+          "right_pinky_dip_joint": 0.08165574073791504,
+          "right_pinky_mcp_joint": 2.348736047744751,
+          "right_pinky_pip_joint": 2.373507499694824,
+          "right_ring_dip_joint": 1.5018227100372314,
+          "right_ring_mcp_joint": 2.1126577854156494,
+          "right_ring_pip_joint": 2.5233230590820312,
+          "right_thumb_dip_joint": 1.1205986738204956,
+          "right_thumb_mcp_joint": 1.225667953491211,
+          "right_thumb_pip_joint": 1.7652851343154907,
+          "right_thumb_yaw_joint": 0.0067640733905136585
+        },
+        "coverage": {
+          "hard_limit_source": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          },
+          "joint_role_manifest": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          }
+        },
+        "hard_limit_sources": {
+          "runtime.joint_pos_limits": 10
+        },
+        "soft_limit_sources": {
+          "runtime.soft_joint_pos_limits": 10
+        },
+        "unknown_runtime_joint_names": [],
+        "baseline_checks": {
+          "post_home": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          },
+          "post_reset": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          }
+        }
+      }
+    },
+    "robot_motion": {
+      "joint_acceleration_rms": 42.03115120574183,
+      "joint_effort_rms": 40.095110257162816,
+      "joint_jerk_rms": 4255.74555520429,
+      "joint_velocity_rms": 0.8570957403328379
+    },
+    "grasp": {
+      "kinematic_grasp_stability_index": 0.9053341556108909,
+      "mean_kinematic_grasp_stability": 0.45365309487160427,
+      "per_object_best_kinematic_grasp_stability": {
+        "obj_127_french_fries_2": 0.0005072203712319478,
+        "obj_128_hamburg_3": 0.4519183539365257,
+        "obj_163_baguette_5": 0.8854239570936067,
+        "obj_182_bread_4": 0.727510702783414
+      },
+      "grasp_gsi_diagnostics": {
+        "mean_anti_slip_score": 0.7374288107202726,
+        "mean_duration_score": 0.3526976549413741,
+        "mean_hold_score": 0.21971691792295156,
+        "mean_kinematic_stability_score": 0.06098248369518157,
+        "mean_legacy_gsi": 0.34845968288010737,
+        "mean_lift_score": 0.2932328308207705,
+        "mean_motion_stability_score": 0.20608393419595297
+      }
+    },
+    "tool": {
+      "tool_selection_accuracy": null,
+      "tool_switch_success_rate": null,
+      "tool_switch_total": 0,
+      "tool_switch_successful": 0
+    },
+    "runtime": {
+      "avg_steps": 2985,
+      "error_count": 0
+    }
+  }
+}

gpt6_astra_1x_icl/per_task_eval/61_medicine_shoebox_pack/Full/summary.json

@@ -0,0 +1,217 @@
+{
+  "metric_schema_version": "core-v1.8-policy-step-query-split",
+  "metadata": {
+    "policy_name": "gpt-6-astra",
+    "base_seed": 206100000,
+    "seed_policy": "episode_seed = base_seed + episode_index - 1",
+    "randomization_recorded": true,
+    "n_episodes": 10,
+    "successes": 8,
+    "error_count": 1,
+    "evaluation_protocol": "reach_and_stop",
+    "early_stop": true,
+    "success_definition": "terminal.raw_condition holds for dwell_time_s",
+    "measurement_window": "episode_start_to_stable_success_or_failure_budget",
+    "max_episode_steps": 3102,
+    "dwell_time_s": 0.5,
+    "field_semantics": {
+      "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
+      "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
+      "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
+      "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
+      "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
+      "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
+      "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
+      "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
+      "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
+      "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
+      "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
+      "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
+      "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
+      "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
+      "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
+      "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
+      "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
+      "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
+      "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
+      "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
+      "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
+      "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
+      "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
+      "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
+      "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
+    }
+  },
+  "core_metrics": {
+    "completion": {
+      "success_rate": 0.8,
+      "success_rate_ci": {
+        "mean": 0.8,
+        "ci_low": 0.5,
+        "ci_high": 1.0
+      }
+    },
+    "progress": {
+      "latched_stage_completion_rate": 0.8333333333333334
+    },
+    "efficiency": {
+      "avg_time_to_success_s": 47.587499999998585
+    },
+    "safety": {
+      "safe_success_rate": 0.7,
+      "hard_violation_rate": 0.2,
+      "drop_rate": 0.2,
+      "high_speed_violation_rate": 0.0,
+      "safety_violation_step_rate": 7.488953793155097e-05
+    },
+    "robustness": {}
+  },
+  "diagnostic_metrics": {
+    "completion": {
+      "ever_instant_success_rate": 0.8,
+      "at_end_success_rate": 0.8,
+      "task_family_success_rate": {
+        "rigid_pick_place": 0.8
+      }
+    },
+    "progress": {
+      "current_stage_completion_rate": 0.8333333333333334,
+      "current_chain_depth_progress_score": 0.8333333333333334,
+      "latched_chain_depth_progress_score": 0.8333333333333334,
+      "avg_current_chain_depth": 2.5,
+      "avg_latched_chain_depth": 2.5
+    },
+    "efficiency": {
+      "task_efficiency": 0.7252476778155035,
+      "expert_time_s": 34.45,
+      "expert_time_step": 2067,
+      "avg_steps_to_success": 2855.25,
+      "avg_policy_steps_to_success": 952.125,
+      "avg_policy_queries_to_success": 27
+    },
+    "safety": {
+      "safety_violation_events_per_step": 7.488953793155097e-05,
+      "violation_counts": {
+        "drop_to_table_obj_186_pillbottle_2": 2
+      },
+      "robot_constraint_violation_rate": 0.6,
+      "finger_joint_limit_saturation_rate": 0.2,
+      "joint_limit_active_step_rate": 0.010971317306972216,
+      "finger_joint_limit_saturation_step_rate": 0.0063656107241818315,
+      "max_joint_limit_excess_rad": 0.2870100140571594,
+      "robot_constraint_diagnostics": {
+        "joint_limit_soft_counts": {
+          "left_index_mcp": 2,
+          "left_thumb_mcp": 227,
+          "left_thumb_pip": 3,
+          "panda_joint1": 12,
+          "right_index_mcp": 9,
+          "right_thumb_mcp": 39,
+          "right_thumb_pip": 2
+        },
+        "joint_limit_hard_counts": {
+          "left_index_mcp": 8,
+          "left_thumb_abd": 1,
+          "left_thumb_mcp": 4,
+          "left_thumb_pip": 5,
+          "panda_joint1": 1,
+          "right_index_mcp": 2,
+          "right_thumb_abd": 1,
+          "right_thumb_mcp": 3,
+          "right_thumb_pip": 4
+        },
+        "joint_limit_hard_episode_events": 26,
+        "finger_joint_limit_saturation_counts": {
+          "left_thumb_mcp": 170
+        },
+        "joint_limit_max_excess_by_joint": {
+          "left_index_mcp": 0.013265013694763184,
+          "left_middle_mcp": 0.0038971900939941406,
+          "left_ring_mcp": 0.004880070686340332,
+          "left_thumb_abd": 0.006793883629143238,
+          "left_thumb_mcp": 0.2870100140571594,
+          "left_thumb_pip": 0.03088366985321045,
+          "left_wrist": 0.0008555054664611816,
+          "multi_panda_joint1": 0.0007903575897216797,
+          "multi_panda_joint2": 8.082389831542969e-05,
+          "multi_panda_joint6": 0.0005898475646972656,
+          "panda_joint1": 0.03150486946105957,
+          "panda_joint4": 3.7670135498046875e-05,
+          "panda_joint6": 0.00022268295288085938,
+          "right_index_mcp": 0.044898033142089844,
+          "right_middle_mcp": 0.0024698972702026367,
+          "right_ring_mcp": 0.0024514198303222656,
+          "right_thumb_abd": 0.008111393079161644,
+          "right_thumb_mcp": 0.04530054330825806,
+          "right_thumb_pip": 0.01618802547454834,
+          "right_wrist": 0.0023905038833618164
+        },
+        "coverage": {
+          "hard_limit_source": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          },
+          "joint_role_manifest": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          }
+        },
+        "hard_limit_sources": {
+          "runtime.joint_pos_limits": 10
+        },
+        "soft_limit_sources": {
+          "runtime.soft_joint_pos_limits": 10
+        },
+        "unknown_runtime_joint_names": [],
+        "baseline_checks": {
+          "post_home": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          },
+          "post_reset": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          }
+        }
+      }
+    },
+    "robot_motion": {
+      "joint_acceleration_rms": 13.352779522053197,
+      "joint_effort_rms": 30.984043879517564,
+      "joint_jerk_rms": 1354.7758293435568,
+      "joint_velocity_rms": 0.3676393883821293
+    },
+    "grasp": {
+      "kinematic_grasp_stability_index": 0.8784805893632315,
+      "mean_kinematic_grasp_stability": 0.4564553016247831,
+      "per_object_best_kinematic_grasp_stability": {
+        "obj_186_pillbottle_2": 0.7485942393665792,
+        "obj_211_hydrating_oil_4": 0.574519497127236,
+        "obj_218_tooth_paste_3": 0.722047709770264
+      },
+      "grasp_gsi_diagnostics": {
+        "mean_anti_slip_score": 0.9197153007036462,
+        "mean_duration_score": 0.21457679051247536,
+        "mean_hold_score": 0.20949915391856863,
+        "mean_kinematic_stability_score": 0.12378331291105998,
+        "mean_legacy_gsi": 0.355359782415228,
+        "mean_lift_score": 0.2858099626193437,
+        "mean_motion_stability_score": 0.20054731827339617
+      }
+    },
+    "tool": {
+      "tool_selection_accuracy": null,
+      "tool_switch_success_rate": null,
+      "tool_switch_total": 0,
+      "tool_switch_successful": 0
+    },
+    "runtime": {
+      "avg_steps": 2670.6,
+      "error_count": 1
+    }
+  }
+}

gpt6_astra_1x_icl/per_task_eval/62_shoebox_accessory_pack/Full/summary.json

@@ -0,0 +1,237 @@
+{
+  "metric_schema_version": "core-v1.8-policy-step-query-split",
+  "metadata": {
+    "policy_name": "gpt-6-astra",
+    "base_seed": 206200000,
+    "seed_policy": "episode_seed = base_seed + episode_index - 1",
+    "randomization_recorded": true,
+    "n_episodes": 10,
+    "successes": 1,
+    "error_count": 1,
+    "evaluation_protocol": "reach_and_stop",
+    "early_stop": true,
+    "success_definition": "terminal.raw_condition holds for dwell_time_s",
+    "measurement_window": "episode_start_to_stable_success_or_failure_budget",
+    "max_episode_steps": 3231,
+    "dwell_time_s": 0.5,
+    "field_semantics": {
+      "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
+      "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
+      "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
+      "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
+      "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
+      "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
+      "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
+      "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
+      "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
+      "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
+      "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
+      "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
+      "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
+      "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
+      "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
+      "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
+      "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
+      "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
+      "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
+      "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
+      "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
+      "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
+      "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
+      "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
+      "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
+    }
+  },
+  "core_metrics": {
+    "completion": {
+      "success_rate": 0.1,
+      "success_rate_ci": {
+        "mean": 0.1,
+        "ci_low": 0.0,
+        "ci_high": 0.3
+      }
+    },
+    "progress": {
+      "latched_stage_completion_rate": 0.5
+    },
+    "efficiency": {
+      "avg_time_to_success_s": 50.233333333331764
+    },
+    "safety": {
+      "safe_success_rate": 0.1,
+      "hard_violation_rate": 0.6,
+      "drop_rate": 0.6,
+      "high_speed_violation_rate": 0.2,
+      "safety_violation_step_rate": 0.0008907657286133747
+    },
+    "robustness": {}
+  },
+  "diagnostic_metrics": {
+    "completion": {
+      "ever_instant_success_rate": 0.1,
+      "at_end_success_rate": 0.1,
+      "task_family_success_rate": {
+        "rigid_pick_place": 0.1
+      }
+    },
+    "progress": {
+      "current_stage_completion_rate": 0.36666666666666664,
+      "current_chain_depth_progress_score": 0.36666666666666664,
+      "latched_chain_depth_progress_score": 0.5,
+      "avg_current_chain_depth": 1.1,
+      "avg_latched_chain_depth": 1.5
+    },
+    "efficiency": {
+      "task_efficiency": 0.7146648971466713,
+      "expert_time_s": 35.9,
+      "expert_time_step": 2154,
+      "avg_steps_to_success": 3014,
+      "avg_policy_steps_to_success": 1005,
+      "avg_policy_queries_to_success": 22
+    },
+    "safety": {
+      "safety_violation_events_per_step": 0.00029692190953779156,
+      "violation_counts": {
+        "drop_to_table_obj_143_pet_collar_3": 2,
+        "drop_to_table_obj_153_shoe_2": 1,
+        "drop_to_table_obj_203_seal_4": 4,
+        "high_speed_obj_129_shoe_box_1": 2
+      },
+      "robot_constraint_violation_rate": 1.0,
+      "finger_joint_limit_saturation_rate": 1.0,
+      "joint_limit_active_step_rate": 0.08284121276104385,
+      "finger_joint_limit_saturation_step_rate": 0.041107188809343144,
+      "max_joint_limit_excess_rad": 1.3980273008346558,
+      "robot_constraint_diagnostics": {
+        "joint_limit_soft_counts": {
+          "leap_l_10": 983,
+          "leap_l_12": 1,
+          "leap_l_15": 34,
+          "leap_l_2": 88,
+          "leap_l_5": 8,
+          "leap_l_6": 63,
+          "multi_leap_r_1": 37,
+          "multi_leap_r_10": 55,
+          "multi_leap_r_2": 1003,
+          "multi_leap_r_5": 309,
+          "multi_leap_r_6": 40,
+          "multi_leap_r_9": 125
+        },
+        "joint_limit_hard_counts": {
+          "leap_l_1": 1,
+          "leap_l_10": 112,
+          "leap_l_12": 1,
+          "leap_l_13": 1,
+          "leap_l_15": 1,
+          "leap_l_2": 22,
+          "leap_l_5": 3,
+          "leap_l_6": 2,
+          "multi_leap_r_1": 11,
+          "multi_leap_r_10": 5,
+          "multi_leap_r_2": 121,
+          "multi_leap_r_5": 27,
+          "multi_leap_r_6": 2,
+          "multi_leap_r_9": 15
+        },
+        "joint_limit_hard_episode_events": 286,
+        "finger_joint_limit_saturation_counts": {
+          "leap_l_10": 448,
+          "leap_l_6": 33,
+          "multi_leap_r_1": 23,
+          "multi_leap_r_10": 6,
+          "multi_leap_r_2": 646,
+          "multi_leap_r_5": 96,
+          "multi_leap_r_6": 16,
+          "multi_leap_r_9": 65
+        },
+        "joint_limit_max_excess_by_joint": {
+          "leap_l_1": 0.006041496992111206,
+          "leap_l_10": 0.9413332939147949,
+          "leap_l_11": 0.0007315278053283691,
+          "leap_l_12": 0.030606746673583984,
+          "leap_l_13": 0.006847858428955078,
+          "leap_l_15": 0.07977128028869629,
+          "leap_l_2": 0.14914929866790771,
+          "leap_l_3": 0.0003237128257751465,
+          "leap_l_5": 0.01879701018333435,
+          "leap_l_6": 0.2482045292854309,
+          "multi_joint2": 8.726119995117188e-05,
+          "multi_joint5": 0.001739501953125,
+          "multi_leap_r_1": 0.7594497799873352,
+          "multi_leap_r_10": 0.3137786388397217,
+          "multi_leap_r_14": 0.0032968521118164062,
+          "multi_leap_r_2": 1.3980273008346558,
+          "multi_leap_r_3": 0.0005801916122436523,
+          "multi_leap_r_5": 0.5519753694534302,
+          "multi_leap_r_6": 0.6966503858566284,
+          "multi_leap_r_9": 0.4708576798439026
+        },
+        "coverage": {
+          "hard_limit_source": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          },
+          "joint_role_manifest": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          }
+        },
+        "hard_limit_sources": {
+          "runtime.joint_pos_limits": 10
+        },
+        "soft_limit_sources": {
+          "runtime.soft_joint_pos_limits": 10
+        },
+        "unknown_runtime_joint_names": [],
+        "baseline_checks": {
+          "post_home": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          },
+          "post_reset": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          }
+        }
+      }
+    },
+    "robot_motion": {
+      "joint_acceleration_rms": 13.450612894861683,
+      "joint_effort_rms": 52.50730358736812,
+      "joint_jerk_rms": 1317.9484685470466,
+      "joint_velocity_rms": 0.5117744167445009
+    },
+    "grasp": {
+      "kinematic_grasp_stability_index": 0.8428762547557698,
+      "mean_kinematic_grasp_stability": 0.42899256368516175,
+      "per_object_best_kinematic_grasp_stability": {
+        "obj_143_pet_collar_3": 0.3203506381371494,
+        "obj_153_shoe_2": 0.534497211861279,
+        "obj_203_seal_4": 0.734455183686091
+      },
+      "grasp_gsi_diagnostics": {
+        "mean_anti_slip_score": 0.8422642140748882,
+        "mean_duration_score": 0.34615112041520985,
+        "mean_hold_score": 0.08341946009800057,
+        "mean_kinematic_stability_score": 0.04143370823434997,
+        "mean_legacy_gsi": 0.28596229946024676,
+        "mean_lift_score": 0.11013936975425967,
+        "mean_motion_stability_score": 0.07251795644534301
+      }
+    },
+    "tool": {
+      "tool_selection_accuracy": null,
+      "tool_switch_success_rate": null,
+      "tool_switch_total": 0,
+      "tool_switch_successful": 0
+    },
+    "runtime": {
+      "avg_steps": 3031.1,
+      "error_count": 1
+    }
+  }
+}

gpt6_astra_1x_icl/per_task_eval/64_sports_ball_cup_sort/Full/summary.json

@@ -0,0 +1,226 @@
+{
+  "metric_schema_version": "core-v1.8-policy-step-query-split",
+  "metadata": {
+    "policy_name": "gpt-6-astra",
+    "base_seed": 206400000,
+    "seed_policy": "episode_seed = base_seed + episode_index - 1",
+    "randomization_recorded": true,
+    "n_episodes": 10,
+    "successes": 6,
+    "error_count": 0,
+    "evaluation_protocol": "reach_and_stop",
+    "early_stop": true,
+    "success_definition": "terminal.raw_condition holds for dwell_time_s",
+    "measurement_window": "episode_start_to_stable_success_or_failure_budget",
+    "max_episode_steps": 1995,
+    "dwell_time_s": 0.5,
+    "field_semantics": {
+      "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
+      "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
+      "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
+      "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
+      "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
+      "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
+      "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
+      "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
+      "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
+      "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
+      "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
+      "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
+      "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
+      "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
+      "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
+      "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
+      "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
+      "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
+      "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
+      "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
+      "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
+      "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
+      "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
+      "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
+      "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
+    }
+  },
+  "core_metrics": {
+    "completion": {
+      "success_rate": 0.6,
+      "success_rate_ci": {
+        "mean": 0.6,
+        "ci_low": 0.3,
+        "ci_high": 0.9
+      }
+    },
+    "progress": {
+      "latched_stage_completion_rate": 0.725
+    },
+    "efficiency": {
+      "avg_time_to_success_s": 32.62777777777721
+    },
+    "safety": {
+      "safe_success_rate": 0.6,
+      "hard_violation_rate": 0.3,
+      "drop_rate": 0.3,
+      "high_speed_violation_rate": 0.1,
+      "safety_violation_step_rate": 0.00035486160397445
+    },
+    "robustness": {}
+  },
+  "diagnostic_metrics": {
+    "completion": {
+      "ever_instant_success_rate": 0.7,
+      "at_end_success_rate": 0.6,
+      "task_family_success_rate": {
+        "rigid_pick_place": 0.6
+      }
+    },
+    "progress": {
+      "current_stage_completion_rate": 0.675,
+      "current_chain_depth_progress_score": 0.725,
+      "latched_chain_depth_progress_score": 0.725,
+      "avg_current_chain_depth": 2.9,
+      "avg_latched_chain_depth": 2.9
+    },
+    "efficiency": {
+      "task_efficiency": 0.6789434557829453,
+      "expert_time_s": 22.15,
+      "expert_time_step": 1329,
+      "avg_steps_to_success": 1957.6666666666667,
+      "avg_policy_steps_to_success": 652.8333333333334,
+      "avg_policy_queries_to_success": 30.5
+    },
+    "safety": {
+      "safety_violation_events_per_step": 0.00035486160397445,
+      "violation_counts": {
+        "drop_to_table_obj_055_baseball_4": 3,
+        "drop_to_table_obj_057_racquetball_5": 2,
+        "drop_to_table_obj_058_golf_ball_6": 1,
+        "high_speed_obj_055_baseball_4": 1
+      },
+      "robot_constraint_violation_rate": 0.9,
+      "finger_joint_limit_saturation_rate": 0.4,
+      "joint_limit_active_step_rate": 0.027324343506032647,
+      "finger_joint_limit_saturation_step_rate": 0.01333265740646862,
+      "max_joint_limit_excess_rad": 0.8996438980102539,
+      "robot_constraint_diagnostics": {
+        "joint_limit_soft_counts": {
+          "joint_10_0": 265,
+          "multi_joint_2_0": 106,
+          "multi_joint_6_0": 15,
+          "multi_joint_7_0": 45,
+          "multi_joint_9_0": 87,
+          "multi_panda_joint5": 18,
+          "panda_joint7": 35
+        },
+        "joint_limit_hard_counts": {
+          "joint_10_0": 22,
+          "joint_12_0": 1,
+          "joint_9_0": 1,
+          "multi_joint_1_0": 1,
+          "multi_joint_2_0": 73,
+          "multi_joint_5_0": 4,
+          "multi_joint_6_0": 1,
+          "multi_joint_7_0": 2,
+          "multi_joint_9_0": 3,
+          "multi_panda_joint5": 1,
+          "panda_joint7": 2
+        },
+        "joint_limit_hard_episode_events": 108,
+        "finger_joint_limit_saturation_counts": {
+          "joint_10_0": 184,
+          "multi_joint_2_0": 24,
+          "multi_joint_7_0": 11,
+          "multi_joint_9_0": 44
+        },
+        "joint_limit_max_excess_by_joint": {
+          "joint_10_0": 0.66518634557724,
+          "joint_12_0": 0.0050629377365112305,
+          "joint_13_0": 3.2261013984680176e-06,
+          "joint_1_0": 0.00023233890533447266,
+          "joint_2_0": 0.0004210472106933594,
+          "joint_5_0": 0.00020384788513183594,
+          "joint_6_0": 0.0002518892288208008,
+          "joint_9_0": 0.006655454635620117,
+          "multi_joint_10_0": 0.001398324966430664,
+          "multi_joint_1_0": 0.007284760475158691,
+          "multi_joint_2_0": 0.8996438980102539,
+          "multi_joint_5_0": 0.006478667259216309,
+          "multi_joint_6_0": 0.04621243476867676,
+          "multi_joint_7_0": 0.24186664819717407,
+          "multi_joint_9_0": 0.22497618198394775,
+          "multi_panda_joint1": 7.200241088867188e-05,
+          "multi_panda_joint5": 0.06462550163269043,
+          "multi_panda_joint7": 0.0021522045135498047,
+          "panda_joint5": 0.0001461505889892578,
+          "panda_joint7": 0.08028483390808105
+        },
+        "coverage": {
+          "hard_limit_source": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          },
+          "joint_role_manifest": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          }
+        },
+        "hard_limit_sources": {
+          "runtime.joint_pos_limits": 10
+        },
+        "soft_limit_sources": {
+          "runtime.soft_joint_pos_limits": 10
+        },
+        "unknown_runtime_joint_names": [],
+        "baseline_checks": {
+          "post_home": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          },
+          "post_reset": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          }
+        }
+      }
+    },
+    "robot_motion": {
+      "joint_acceleration_rms": 9.207911887023146,
+      "joint_effort_rms": 29.024835436637847,
+      "joint_jerk_rms": 927.7374989751764,
+      "joint_velocity_rms": 0.2729033813019237
+    },
+    "grasp": {
+      "kinematic_grasp_stability_index": 0.9867421040812419,
+      "mean_kinematic_grasp_stability": 0.40980121245964235,
+      "per_object_best_kinematic_grasp_stability": {
+        "obj_054_softball_3": 0.6421239324384325,
+        "obj_055_baseball_4": 0.9296671411791132,
+        "obj_057_racquetball_5": 0.8341595343153546,
+        "obj_058_golf_ball_6": 0.737731666754543
+      },
+      "grasp_gsi_diagnostics": {
+        "mean_anti_slip_score": 0.8336579913254804,
+        "mean_duration_score": 0.31413740111441035,
+        "mean_hold_score": 0.24307938818638286,
+        "mean_kinematic_stability_score": 0.09424760398579528,
+        "mean_legacy_gsi": 0.3746012329278188,
+        "mean_lift_score": 0.3328560603057352,
+        "mean_motion_stability_score": 0.227613904503579
+      }
+    },
+    "tool": {
+      "tool_selection_accuracy": null,
+      "tool_switch_success_rate": null,
+      "tool_switch_total": 0,
+      "tool_switch_successful": 0
+    },
+    "runtime": {
+      "avg_steps": 1972.6,
+      "error_count": 0
+    }
+  }
+}

gpt6_astra_1x_icl/per_task_eval/67_faucet_cup_water_fill/Full/summary.json

@@ -0,0 +1,255 @@
+{
+  "metric_schema_version": "core-v1.8-policy-step-query-split",
+  "metadata": {
+    "policy_name": "gpt-6-astra",
+    "base_seed": 206700000,
+    "seed_policy": "episode_seed = base_seed + episode_index - 1",
+    "randomization_recorded": true,
+    "n_episodes": 10,
+    "successes": 1,
+    "error_count": 0,
+    "evaluation_protocol": "reach_and_stop",
+    "early_stop": true,
+    "success_definition": "terminal.raw_condition holds for dwell_time_s",
+    "measurement_window": "episode_start_to_stable_success_or_failure_budget",
+    "max_episode_steps": 2892,
+    "dwell_time_s": 3.0,
+    "field_semantics": {
+      "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
+      "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
+      "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
+      "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
+      "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
+      "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
+      "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
+      "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
+      "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
+      "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
+      "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
+      "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
+      "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
+      "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
+      "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
+      "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
+      "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
+      "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
+      "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
+      "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
+      "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
+      "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
+      "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
+      "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
+      "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
+    }
+  },
+  "core_metrics": {
+    "completion": {
+      "success_rate": 0.1,
+      "success_rate_ci": {
+        "mean": 0.1,
+        "ci_low": 0.0,
+        "ci_high": 0.3
+      }
+    },
+    "progress": {
+      "latched_stage_completion_rate": 0.4
+    },
+    "efficiency": {
+      "avg_time_to_success_s": 42.46666666666554
+    },
+    "safety": {
+      "safe_success_rate": 0.1,
+      "hard_violation_rate": 0.7,
+      "drop_rate": 0.7,
+      "high_speed_violation_rate": 0.0,
+      "safety_violation_step_rate": 0.00024496080627099665
+    },
+    "robustness": {}
+  },
+  "diagnostic_metrics": {
+    "completion": {
+      "ever_instant_success_rate": 0.1,
+      "at_end_success_rate": 0.1,
+      "task_family_success_rate": {
+        "articulation_multi": 0.1
+      }
+    },
+    "progress": {
+      "current_stage_completion_rate": 0.15,
+      "current_chain_depth_progress_score": 0.25,
+      "latched_chain_depth_progress_score": 0.4,
+      "avg_current_chain_depth": 1,
+      "avg_latched_chain_depth": 1.6
+    },
+    "efficiency": {
+      "task_efficiency": 0.7566718995290624,
+      "expert_time_s": 32.13333333333333,
+      "expert_time_step": 1928,
+      "avg_steps_to_success": 2548,
+      "avg_policy_steps_to_success": 850,
+      "avg_policy_queries_to_success": 28
+    },
+    "safety": {
+      "safety_violation_events_per_step": 0.00024496080627099665,
+      "violation_counts": {
+        "drop_to_table_obj_031_spoon_1": 7
+      },
+      "robot_constraint_violation_rate": 1.0,
+      "finger_joint_limit_saturation_rate": 0.6,
+      "joint_limit_active_step_rate": 0.021871500559910415,
+      "finger_joint_limit_saturation_step_rate": 0.00566909294512878,
+      "max_joint_limit_excess_rad": 1.2255189418792725,
+      "robot_constraint_diagnostics": {
+        "joint_limit_soft_counts": {
+          "l_joint4": 69,
+          "left_index_proximal_joint": 101,
+          "left_middle_distal_joint": 34,
+          "left_middle_proximal_joint": 62,
+          "left_pinky_distal_joint": 38,
+          "left_pinky_proximal_joint": 10,
+          "left_ring_distal_joint": 11,
+          "left_ring_proximal_joint": 141,
+          "left_thumb_distal_joint": 57,
+          "left_thumb_metacarpal_joint": 11,
+          "left_thumb_proximal_joint": 32,
+          "right_index_proximal_joint": 9,
+          "right_middle_distal_joint": 45,
+          "right_middle_proximal_joint": 73,
+          "right_pinky_proximal_joint": 108,
+          "right_ring_distal_joint": 12,
+          "right_ring_proximal_joint": 68,
+          "right_thumb_metacarpal_joint": 93,
+          "right_thumb_proximal_joint": 72
+        },
+        "joint_limit_hard_counts": {
+          "l_joint4": 1,
+          "left_index_proximal_joint": 31,
+          "left_middle_distal_joint": 4,
+          "left_middle_proximal_joint": 32,
+          "left_pinky_distal_joint": 3,
+          "left_pinky_proximal_joint": 23,
+          "left_ring_distal_joint": 8,
+          "left_ring_proximal_joint": 29,
+          "left_thumb_distal_joint": 6,
+          "left_thumb_metacarpal_joint": 3,
+          "left_thumb_proximal_joint": 6,
+          "right_index_proximal_joint": 11,
+          "right_middle_distal_joint": 1,
+          "right_middle_proximal_joint": 10,
+          "right_pinky_distal_joint": 2,
+          "right_pinky_proximal_joint": 26,
+          "right_ring_distal_joint": 2,
+          "right_ring_proximal_joint": 12,
+          "right_thumb_distal_joint": 6,
+          "right_thumb_metacarpal_joint": 2,
+          "right_thumb_proximal_joint": 2
+        },
+        "joint_limit_hard_episode_events": 116,
+        "finger_joint_limit_saturation_counts": {
+          "left_index_proximal_joint": 21,
+          "left_middle_proximal_joint": 8,
+          "left_pinky_distal_joint": 7,
+          "left_ring_proximal_joint": 1,
+          "left_thumb_distal_joint": 43,
+          "left_thumb_proximal_joint": 11,
+          "right_middle_distal_joint": 37,
+          "right_middle_proximal_joint": 60,
+          "right_pinky_proximal_joint": 58,
+          "right_ring_proximal_joint": 62,
+          "right_thumb_metacarpal_joint": 73,
+          "right_thumb_proximal_joint": 66
+        },
+        "joint_limit_max_excess_by_joint": {
+          "joint3": 0.0006670951843261719,
+          "l_joint4": 0.6398665904998779,
+          "left_index_distal_joint": 7.247085598294234e-09,
+          "left_index_proximal_joint": 0.21345362067222595,
+          "left_middle_distal_joint": 0.0739472284913063,
+          "left_middle_proximal_joint": 0.49430060386657715,
+          "left_pinky_distal_joint": 0.18549209833145142,
+          "left_pinky_proximal_joint": 0.052287131547927856,
+          "left_ring_distal_joint": 0.08776852488517761,
+          "left_ring_proximal_joint": 0.27737730741500854,
+          "left_thumb_distal_joint": 0.5415257215499878,
+          "left_thumb_metacarpal_joint": 0.14287704229354858,
+          "left_thumb_proximal_joint": 0.19011513888835907,
+          "right_index_distal_joint": 3.43435807570458e-10,
+          "right_index_proximal_joint": 0.044638969004154205,
+          "right_middle_distal_joint": 0.5703709125518799,
+          "right_middle_proximal_joint": 1.2255189418792725,
+          "right_pinky_distal_joint": 0.006779804825782776,
+          "right_pinky_proximal_joint": 0.4822751581668854,
+          "right_ring_distal_joint": 0.07349101454019547,
+          "right_ring_proximal_joint": 0.729503333568573,
+          "right_thumb_distal_joint": 0.006306767463684082,
+          "right_thumb_metacarpal_joint": 1.004553198814392,
+          "right_thumb_proximal_joint": 0.3518408536911011
+        },
+        "coverage": {
+          "hard_limit_source": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          },
+          "joint_role_manifest": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          }
+        },
+        "hard_limit_sources": {
+          "runtime.joint_pos_limits": 10
+        },
+        "soft_limit_sources": {
+          "runtime.soft_joint_pos_limits": 10
+        },
+        "unknown_runtime_joint_names": [],
+        "baseline_checks": {
+          "post_home": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          },
+          "post_reset": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          }
+        }
+      }
+    },
+    "robot_motion": {
+      "joint_acceleration_rms": 17.557697685911634,
+      "joint_effort_rms": 34.88281651243846,
+      "joint_jerk_rms": 1792.867224072184,
+      "joint_velocity_rms": 0.42139582796471897
+    },
+    "grasp": {
+      "kinematic_grasp_stability_index": 0.7970236393824165,
+      "mean_kinematic_grasp_stability": 0.2994397114519785,
+      "per_object_best_kinematic_grasp_stability": {
+        "obj_025_mug_1": 0.5855821488893681,
+        "obj_031_spoon_1": 0.7227022694059623
+      },
+      "grasp_gsi_diagnostics": {
+        "mean_anti_slip_score": 0.8081044589343345,
+        "mean_duration_score": 0.38472007081372744,
+        "mean_hold_score": 0.2551040411376831,
+        "mean_kinematic_stability_score": 0.1264663544545423,
+        "mean_legacy_gsi": 0.39177606879777344,
+        "mean_lift_score": 0.37053483219013744,
+        "mean_motion_stability_score": 0.23423136351870943
+      }
+    },
+    "tool": {
+      "tool_selection_accuracy": null,
+      "tool_switch_success_rate": null,
+      "tool_switch_total": 0,
+      "tool_switch_successful": 0
+    },
+    "runtime": {
+      "avg_steps": 2857.6,
+      "error_count": 0
+    }
+  }
+}

gpt6_astra_1x_icl/per_task_eval/73_jigsaw_puzzle_assembly/Full/summary.json

@@ -0,0 +1,329 @@
+{
+  "metric_schema_version": "core-v1.8-policy-step-query-split",
+  "metadata": {
+    "policy_name": "gpt-6-astra",
+    "base_seed": 207300000,
+    "seed_policy": "episode_seed = base_seed + episode_index - 1",
+    "randomization_recorded": true,
+    "n_episodes": 10,
+    "successes": 0,
+    "error_count": 0,
+    "evaluation_protocol": "reach_and_stop",
+    "early_stop": true,
+    "success_definition": "terminal.raw_condition holds for dwell_time_s",
+    "measurement_window": "episode_start_to_stable_success_or_failure_budget",
+    "max_episode_steps": 3639,
+    "dwell_time_s": 0.5,
+    "field_semantics": {
+      "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
+      "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
+      "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
+      "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
+      "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
+      "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
+      "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
+      "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
+      "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
+      "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
+      "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
+      "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
+      "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
+      "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
+      "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
+      "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
+      "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
+      "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
+      "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
+      "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
+      "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
+      "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
+      "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
+      "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
+      "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
+    }
+  },
+  "core_metrics": {
+    "completion": {
+      "success_rate": 0.0,
+      "success_rate_ci": {
+        "mean": 0.0,
+        "ci_low": 0.0,
+        "ci_high": 0.0
+      }
+    },
+    "progress": {
+      "latched_stage_completion_rate": 0.2
+    },
+    "efficiency": {
+      "avg_time_to_success_s": null
+    },
+    "safety": {
+      "safe_success_rate": 0.0,
+      "hard_violation_rate": 0.3,
+      "drop_rate": 0.3,
+      "high_speed_violation_rate": 0.0,
+      "safety_violation_step_rate": 0.00010992030777686178
+    },
+    "robustness": {}
+  },
+  "diagnostic_metrics": {
+    "completion": {
+      "ever_instant_success_rate": 0.0,
+      "at_end_success_rate": 0.0,
+      "task_family_success_rate": {
+        "tetris_block_manipulation": 0.0
+      }
+    },
+    "progress": {
+      "current_stage_completion_rate": 0.025,
+      "current_chain_depth_progress_score": 0.1,
+      "latched_chain_depth_progress_score": 0.7,
+      "avg_current_chain_depth": 0.1,
+      "avg_latched_chain_depth": 0.7
+    },
+    "efficiency": {
+      "task_efficiency": null,
+      "expert_time_s": 40.43333333333333,
+      "expert_time_step": 2426,
+      "avg_steps_to_success": null,
+      "avg_policy_steps_to_success": null,
+      "avg_policy_queries_to_success": null
+    },
+    "safety": {
+      "safety_violation_events_per_step": 0.0001374003847210772,
+      "violation_counts": {
+        "drop_to_table_obj_326_piece_2": 3,
+        "drop_to_table_obj_326_piece_3": 1,
+        "drop_to_table_obj_326_piece_5": 1
+      },
+      "robot_constraint_violation_rate": 1.0,
+      "finger_joint_limit_saturation_rate": 0.9,
+      "joint_limit_active_step_rate": 0.05798296235229459,
+      "finger_joint_limit_saturation_step_rate": 0.027892278098378676,
+      "max_joint_limit_excess_rad": 1.9653520584106445,
+      "robot_constraint_diagnostics": {
+        "joint_limit_soft_counts": {
+          "left_index_DIP": 278,
+          "left_index_MCP_AA": 24,
+          "left_index_MCP_FE": 147,
+          "left_index_PIP": 126,
+          "left_middle_DIP": 13,
+          "left_middle_MCP_AA": 17,
+          "left_middle_MCP_FE": 122,
+          "left_middle_PIP": 54,
+          "left_pinky_CMC": 67,
+          "left_pinky_DIP": 13,
+          "left_pinky_MCP_FE": 103,
+          "left_pinky_PIP": 21,
+          "left_ring_DIP": 2,
+          "left_ring_MCP_AA": 20,
+          "left_ring_MCP_FE": 55,
+          "left_ring_PIP": 55,
+          "left_thumb_IP": 116,
+          "left_thumb_MCP_AA": 2,
+          "multi_A7": 172,
+          "multi_right_index_DIP": 275,
+          "multi_right_index_MCP_AA": 86,
+          "multi_right_index_MCP_FE": 225,
+          "multi_right_index_PIP": 98,
+          "multi_right_middle_DIP": 119,
+          "multi_right_middle_MCP_FE": 194,
+          "multi_right_middle_PIP": 181,
+          "multi_right_pinky_CMC": 296,
+          "multi_right_pinky_DIP": 21,
+          "multi_right_pinky_MCP_FE": 187,
+          "multi_right_pinky_PIP": 104,
+          "multi_right_ring_DIP": 189,
+          "multi_right_ring_MCP_FE": 268,
+          "multi_right_ring_PIP": 231,
+          "multi_right_thumb_CMC_AA": 36,
+          "multi_right_thumb_IP": 42,
+          "multi_right_thumb_MCP_AA": 25,
+          "multi_right_thumb_MCP_FE": 33
+        },
+        "joint_limit_hard_counts": {
+          "left_index_DIP": 15,
+          "left_index_MCP_AA": 2,
+          "left_index_MCP_FE": 10,
+          "left_index_PIP": 5,
+          "left_middle_DIP": 3,
+          "left_middle_MCP_AA": 1,
+          "left_middle_MCP_FE": 5,
+          "left_middle_PIP": 3,
+          "left_pinky_CMC": 11,
+          "left_pinky_DIP": 2,
+          "left_pinky_MCP_FE": 6,
+          "left_pinky_PIP": 2,
+          "left_ring_DIP": 2,
+          "left_ring_MCP_AA": 1,
+          "left_ring_MCP_FE": 6,
+          "left_ring_PIP": 2,
+          "left_thumb_IP": 2,
+          "left_thumb_MCP_AA": 1,
+          "multi_A7": 1,
+          "multi_right_index_DIP": 22,
+          "multi_right_index_MCP_AA": 7,
+          "multi_right_index_MCP_FE": 17,
+          "multi_right_index_PIP": 11,
+          "multi_right_middle_DIP": 13,
+          "multi_right_middle_MCP_FE": 21,
+          "multi_right_middle_PIP": 11,
+          "multi_right_pinky_CMC": 37,
+          "multi_right_pinky_DIP": 17,
+          "multi_right_pinky_MCP_FE": 15,
+          "multi_right_pinky_PIP": 15,
+          "multi_right_ring_DIP": 10,
+          "multi_right_ring_MCP_FE": 18,
+          "multi_right_ring_PIP": 13,
+          "multi_right_thumb_CMC_AA": 2,
+          "multi_right_thumb_IP": 6,
+          "multi_right_thumb_MCP_AA": 1,
+          "multi_right_thumb_MCP_FE": 3
+        },
+        "joint_limit_hard_episode_events": 100,
+        "finger_joint_limit_saturation_counts": {
+          "left_index_DIP": 154,
+          "left_index_MCP_AA": 19,
+          "left_index_MCP_FE": 59,
+          "left_index_PIP": 72,
+          "left_middle_MCP_AA": 11,
+          "left_middle_MCP_FE": 23,
+          "left_middle_PIP": 29,
+          "left_pinky_CMC": 9,
+          "left_pinky_MCP_FE": 19,
+          "left_pinky_PIP": 1,
+          "left_ring_MCP_AA": 16,
+          "left_ring_MCP_FE": 22,
+          "left_ring_PIP": 26,
+          "left_thumb_IP": 101,
+          "multi_right_index_DIP": 205,
+          "multi_right_index_MCP_AA": 61,
+          "multi_right_index_MCP_FE": 16,
+          "multi_right_index_PIP": 41,
+          "multi_right_middle_DIP": 83,
+          "multi_right_middle_MCP_FE": 78,
+          "multi_right_middle_PIP": 122,
+          "multi_right_pinky_CMC": 136,
+          "multi_right_pinky_MCP_FE": 17,
+          "multi_right_pinky_PIP": 15,
+          "multi_right_ring_DIP": 158,
+          "multi_right_ring_MCP_FE": 47,
+          "multi_right_ring_PIP": 165,
+          "multi_right_thumb_CMC_AA": 10,
+          "multi_right_thumb_IP": 20,
+          "multi_right_thumb_MCP_AA": 22,
+          "multi_right_thumb_MCP_FE": 7
+        },
+        "joint_limit_max_excess_by_joint": {
+          "left_index_DIP": 1.6815930604934692,
+          "left_index_MCP_AA": 0.5835015773773193,
+          "left_index_MCP_FE": 0.6076315641403198,
+          "left_index_PIP": 0.4581177234649658,
+          "left_middle_DIP": 0.2170879989862442,
+          "left_middle_MCP_AA": 0.44771286845207214,
+          "left_middle_MCP_FE": 0.577806293964386,
+          "left_middle_PIP": 0.6001745462417603,
+          "left_pinky_CMC": 0.36728325486183167,
+          "left_pinky_DIP": 0.12839090824127197,
+          "left_pinky_MCP_FE": 0.5017547011375427,
+          "left_pinky_PIP": 0.2654762268066406,
+          "left_ring_DIP": 0.047665711492300034,
+          "left_ring_MCP_AA": 0.5033174753189087,
+          "left_ring_MCP_FE": 0.4588993787765503,
+          "left_ring_PIP": 0.9179931879043579,
+          "left_thumb_CMC_AA": 0.0010698437690734863,
+          "left_thumb_IP": 0.7950096726417542,
+          "left_thumb_MCP_AA": 0.03789395093917847,
+          "multi_A5": 0.0002841949462890625,
+          "multi_A7": 0.8324110507965088,
+          "multi_right_index_DIP": 1.9653520584106445,
+          "multi_right_index_MCP_AA": 1.350473403930664,
+          "multi_right_index_MCP_FE": 0.6537641882896423,
+          "multi_right_index_PIP": 0.4873863756656647,
+          "multi_right_middle_DIP": 1.2303272485733032,
+          "multi_right_middle_MCP_FE": 0.5616170763969421,
+          "multi_right_middle_PIP": 1.7219103574752808,
+          "multi_right_pinky_CMC": 1.6296635866165161,
+          "multi_right_pinky_DIP": 0.0646938756108284,
+          "multi_right_pinky_MCP_FE": 0.4806692600250244,
+          "multi_right_pinky_PIP": 0.275006502866745,
+          "multi_right_ring_DIP": 1.706547737121582,
+          "multi_right_ring_MCP_AA": 0.004061490297317505,
+          "multi_right_ring_MCP_FE": 0.3368939161300659,
+          "multi_right_ring_PIP": 1.4780924320220947,
+          "multi_right_thumb_CMC_AA": 0.269027441740036,
+          "multi_right_thumb_IP": 0.3309504985809326,
+          "multi_right_thumb_MCP_AA": 0.7556639909744263,
+          "multi_right_thumb_MCP_FE": 0.7450425624847412
+        },
+        "coverage": {
+          "hard_limit_source": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          },
+          "joint_role_manifest": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          }
+        },
+        "hard_limit_sources": {
+          "runtime.joint_pos_limits": 10
+        },
+        "soft_limit_sources": {
+          "runtime.soft_joint_pos_limits": 10
+        },
+        "unknown_runtime_joint_names": [],
+        "baseline_checks": {
+          "post_home": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          },
+          "post_reset": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          }
+        }
+      }
+    },
+    "robot_motion": {
+      "joint_acceleration_rms": 49.58765700440844,
+      "joint_effort_rms": 37.71310070737992,
+      "joint_jerk_rms": 5022.147078617628,
+      "joint_velocity_rms": 0.7892031924125328
+    },
+    "grasp": {
+      "kinematic_grasp_stability_index": 0.6038355070022731,
+      "mean_kinematic_grasp_stability": 0.15945902026124886,
+      "per_object_best_kinematic_grasp_stability": {
+        "obj_326_piece_1": 0.0,
+        "obj_326_piece_2": 0.43808445689762926,
+        "obj_326_piece_3": 0.4678962707170451,
+        "obj_326_piece_4": 0.11426626611525119,
+        "obj_326_piece_5": 0.16158070433298227
+      },
+      "grasp_gsi_diagnostics": {
+        "mean_anti_slip_score": 0.9893395621507682,
+        "mean_duration_score": 0.05450929742603283,
+        "mean_hold_score": 0.010514152239626385,
+        "mean_kinematic_stability_score": 0.005262529630942942,
+        "mean_legacy_gsi": 0.21548690462276396,
+        "mean_lift_score": 0.018565539983511955,
+        "mean_motion_stability_score": 0.010595505441884681
+      }
+    },
+    "tool": {
+      "tool_selection_accuracy": null,
+      "tool_switch_success_rate": null,
+      "tool_switch_total": 0,
+      "tool_switch_successful": 0
+    },
+    "runtime": {
+      "avg_steps": 3639,
+      "error_count": 0
+    }
+  }
+}

gpt6_astra_1x_icl/per_task_eval/76_soup_serving/Full/summary.json

@@ -0,0 +1,328 @@
+{
+  "metric_schema_version": "core-v1.8-policy-step-query-split",
+  "metadata": {
+    "policy_name": "gpt-6-astra",
+    "base_seed": 207600000,
+    "seed_policy": "episode_seed = base_seed + episode_index - 1",
+    "randomization_recorded": true,
+    "n_episodes": 10,
+    "successes": 0,
+    "error_count": 0,
+    "evaluation_protocol": "reach_and_stop",
+    "early_stop": true,
+    "success_definition": "terminal.raw_condition holds for dwell_time_s",
+    "measurement_window": "episode_start_to_stable_success_or_failure_budget",
+    "max_episode_steps": 2958,
+    "dwell_time_s": 0.5,
+    "field_semantics": {
+      "success_rate": "primary protocol success rate: stable_success for reach_and_stop; terminal at_end_success_budget for fixed_horizon",
+      "avg_time_to_success_s": "main efficiency metric: mean time_to_stable_success_s over reached-success episodes",
+      "task_efficiency": "benchmark-specific expert-normalized speed ratio: expert_time_s / time_to_stable_success_s over reached-success episodes",
+      "avg_steps_to_success": "diagnostic mean steps_to_stable_success over reached-success episodes, or None if no episode succeeds",
+      "avg_policy_steps_to_success": "diagnostic mean policy-rate control/action steps through stable success over reached-success episodes",
+      "avg_policy_queries_to_success": "diagnostic mean actual model inference queries through stable success over reached-success episodes; None when the first-success query count was not observed",
+      "safe_success_rate": "episode success rate with no task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "hard_violation_rate": "episode rate with any task-level hard safety violation (drop or tracked-object high speed; joint limits are diagnostic by default)",
+      "drop_rate": "episode rate with drop_off_table or drop_to_table_after_lift violation",
+      "high_speed_violation_rate": "episode rate with any tracked object exceeding the configured speed threshold",
+      "excessive_impact_rate": "deprecated contact-force impact metric; no longer populated by core safety",
+      "robot_constraint_violation_rate": "diagnostic episode rate with any asset hard-joint-limit observation; excluded from safe_success_rate unless include_in_core_safety is explicitly enabled",
+      "finger_joint_limit_saturation_rate": "episode rate with dexterous finger joints saturated beyond finger joint-limit thresholds; diagnostic by default and not included in safe_success_rate",
+      "joint_limit_active_step_rate": "diagnostic mean fraction of steps with any joint exceeding the soft diagnostic margin",
+      "finger_joint_limit_saturation_step_rate": "diagnostic mean fraction of steps with finger joint-limit saturation",
+      "max_joint_limit_excess_rad": "diagnostic maximum operating/soft-limit excess observed in radians; hard-limit source and role validation are in robot_constraint_diagnostics",
+      "safety_violation_rate_time": "compatibility diagnostic; with fixed physics_dt this is equivalent to safety_violation_step_rate and is omitted from grouped summary output",
+      "safety_violation_step_rate": "core safety metric: fraction of executed steps with task-level safety violations; joint-limit observations are excluded by default",
+      "safety_violation_rate_event": "deprecated compatibility alias of safety_violation_events_per_step; omitted from grouped summary output",
+      "safety_violation_events_per_step": "diagnostic event density; may exceed 1 when multiple violations happen in one step",
+      "episode_violation_rate": "compatibility alias of hard_violation_rate; omitted from grouped core output",
+      "latched_stage_completion_rate": "main progress metric; ever-reached stage completion rate at episode end",
+      "current_stage_completion_rate": "diagnostic terminal-state stage completion rate at episode end",
+      "current_normalized_progress_score": "chain-depth diagnostic; normalized progress score at episode end",
+      "current_chain_depth_progress_score": "alias of current_normalized_progress_score for ordered dependency-depth diagnostics",
+      "latched_chain_depth_progress_score": "alias of latched_normalized_progress_score for ever-reached dependency-depth diagnostics",
+      "robustness": "success rates grouped by benchmark protocol category: cov, inv, and inv+cov"
+    }
+  },
+  "core_metrics": {
+    "completion": {
+      "success_rate": 0.0,
+      "success_rate_ci": {
+        "mean": 0.0,
+        "ci_low": 0.0,
+        "ci_high": 0.0
+      }
+    },
+    "progress": {
+      "latched_stage_completion_rate": 0.425
+    },
+    "efficiency": {
+      "avg_time_to_success_s": null
+    },
+    "safety": {
+      "safe_success_rate": 0.0,
+      "hard_violation_rate": 0.9,
+      "drop_rate": 0.9,
+      "high_speed_violation_rate": 0.2,
+      "safety_violation_step_rate": 0.0017917511832319134
+    },
+    "robustness": {}
+  },
+  "diagnostic_metrics": {
+    "completion": {
+      "ever_instant_success_rate": 0.0,
+      "at_end_success_rate": 0.0,
+      "task_family_success_rate": {
+        "bimanual_soup_scoop_serve": 0.0
+      }
+    },
+    "progress": {
+      "current_stage_completion_rate": 0.375,
+      "current_chain_depth_progress_score": 0.5,
+      "latched_chain_depth_progress_score": 0.5666666666666667,
+      "avg_current_chain_depth": 1.5,
+      "avg_latched_chain_depth": 1.7
+    },
+    "efficiency": {
+      "task_efficiency": null,
+      "expert_time_s": 32.86666666666667,
+      "expert_time_step": 1972,
+      "avg_steps_to_success": null,
+      "avg_policy_steps_to_success": null,
+      "avg_policy_queries_to_success": null
+    },
+    "safety": {
+      "safety_violation_events_per_step": 0.0006423258958755916,
+      "violation_counts": {
+        "drop_obj_330_bowl_2": 1,
+        "drop_to_table_obj_330_bowl_2": 8,
+        "drop_to_table_obj_330_ladle_3": 8,
+        "high_speed_distractor_00_clutter_019_pitcher_base": 1,
+        "high_speed_obj_330_bowl_2": 1
+      },
+      "robot_constraint_violation_rate": 1.0,
+      "finger_joint_limit_saturation_rate": 1.0,
+      "joint_limit_active_step_rate": 0.14533468559837728,
+      "finger_joint_limit_saturation_step_rate": 0.10256930358350237,
+      "max_joint_limit_excess_rad": 1.776158094406128,
+      "robot_constraint_diagnostics": {
+        "joint_limit_soft_counts": {
+          "FFJ1": 38,
+          "FFJ2": 111,
+          "FFJ3": 37,
+          "FFJ4": 4,
+          "LFJ2": 23,
+          "LFJ4": 58,
+          "LFJ5": 76,
+          "MFJ1": 1,
+          "MFJ2": 219,
+          "MFJ4": 87,
+          "RFJ2": 64,
+          "RFJ4": 56,
+          "THJ2": 3,
+          "THJ3": 12,
+          "THJ4": 229,
+          "THJ5": 156,
+          "WRJ1": 16,
+          "WRJ2": 43,
+          "l_FFJ1": 131,
+          "l_FFJ2": 845,
+          "l_FFJ3": 229,
+          "l_LFJ1": 9,
+          "l_LFJ2": 576,
+          "l_LFJ3": 1430,
+          "l_LFJ4": 82,
+          "l_LFJ5": 731,
+          "l_MFJ1": 60,
+          "l_MFJ2": 1575,
+          "l_MFJ3": 2,
+          "l_MFJ4": 53,
+          "l_RFJ1": 3,
+          "l_RFJ2": 1879,
+          "l_RFJ3": 273,
+          "l_RFJ4": 36,
+          "l_THJ4": 160,
+          "l_THJ5": 35,
+          "l_WRJ2": 1
+        },
+        "joint_limit_hard_counts": {
+          "FFJ1": 1,
+          "FFJ2": 16,
+          "FFJ3": 3,
+          "FFJ4": 1,
+          "LFJ1": 1,
+          "LFJ2": 7,
+          "LFJ3": 1,
+          "LFJ4": 14,
+          "LFJ5": 23,
+          "MFJ1": 7,
+          "MFJ2": 31,
+          "MFJ4": 8,
+          "RFJ2": 13,
+          "RFJ4": 6,
+          "THJ2": 1,
+          "THJ3": 1,
+          "THJ4": 20,
+          "THJ5": 20,
+          "WRJ1": 1,
+          "WRJ2": 1,
+          "l_FFJ1": 12,
+          "l_FFJ2": 23,
+          "l_FFJ3": 8,
+          "l_LFJ1": 3,
+          "l_LFJ2": 25,
+          "l_LFJ3": 30,
+          "l_LFJ4": 2,
+          "l_LFJ5": 133,
+          "l_MFJ1": 2,
+          "l_MFJ2": 13,
+          "l_MFJ3": 1,
+          "l_MFJ4": 3,
+          "l_RFJ1": 11,
+          "l_RFJ2": 24,
+          "l_RFJ3": 5,
+          "l_RFJ4": 7,
+          "l_THJ4": 3,
+          "l_THJ5": 17,
+          "l_WRJ2": 1
+        },
+        "joint_limit_hard_episode_events": 181,
+        "finger_joint_limit_saturation_counts": {
+          "FFJ1": 35,
+          "FFJ2": 15,
+          "FFJ3": 11,
+          "LFJ4": 1,
+          "LFJ5": 33,
+          "MFJ2": 69,
+          "MFJ4": 37,
+          "RFJ4": 21,
+          "THJ3": 6,
+          "THJ4": 146,
+          "THJ5": 131,
+          "l_FFJ1": 119,
+          "l_FFJ2": 598,
+          "l_FFJ3": 25,
+          "l_LFJ2": 501,
+          "l_LFJ3": 804,
+          "l_LFJ4": 67,
+          "l_LFJ5": 346,
+          "l_MFJ1": 41,
+          "l_MFJ2": 1351,
+          "l_MFJ4": 50,
+          "l_RFJ2": 1544,
+          "l_RFJ3": 145,
+          "l_RFJ4": 1,
+          "l_THJ4": 107
+        },
+        "joint_limit_max_excess_by_joint": {
+          "FFJ1": 1.4879580736160278,
+          "FFJ2": 0.43703046441078186,
+          "FFJ3": 0.3517078161239624,
+          "FFJ4": 0.05196830630302429,
+          "LFJ1": 0.00720801530405879,
+          "LFJ2": 0.23519718647003174,
+          "LFJ3": 0.007392942905426025,
+          "LFJ4": 0.2500416338443756,
+          "LFJ5": 0.3535561263561249,
+          "MFJ1": 0.0374506413936615,
+          "MFJ2": 0.7520123720169067,
+          "MFJ4": 0.22446885704994202,
+          "RFJ1": 0.003974391613155603,
+          "RFJ2": 0.17116332054138184,
+          "RFJ4": 0.28328296542167664,
+          "THJ2": 0.09219378232955933,
+          "THJ3": 0.4389112591743469,
+          "THJ4": 0.9524463415145874,
+          "THJ5": 1.4296963214874268,
+          "WRJ1": 0.10897630453109741,
+          "WRJ2": 0.22000986337661743,
+          "l_FFJ1": 1.611231803894043,
+          "l_FFJ2": 0.7768445611000061,
+          "l_FFJ3": 0.19642049074172974,
+          "l_FFJ4": 0.0037686824798583984,
+          "l_LFJ1": 0.026788610965013504,
+          "l_LFJ2": 1.5064268112182617,
+          "l_LFJ3": 0.5958528518676758,
+          "l_LFJ4": 1.100192904472351,
+          "l_LFJ5": 1.776158094406128,
+          "l_MFJ1": 0.689359724521637,
+          "l_MFJ2": 1.197654128074646,
+          "l_MFJ3": 0.01969921588897705,
+          "l_MFJ4": 0.7579835653305054,
+          "l_RFJ1": 0.01248151995241642,
+          "l_RFJ2": 1.1910874843597412,
+          "l_RFJ3": 0.21871352195739746,
+          "l_RFJ4": 0.2540991008281708,
+          "l_THJ1": 3.4332275390625e-05,
+          "l_THJ2": 0.00016617774963378906,
+          "l_THJ4": 0.8159288167953491,
+          "l_THJ5": 0.056596994400024414,
+          "l_WRJ1": 0.0025249123573303223,
+          "l_WRJ2": 0.010103315114974976
+        },
+        "coverage": {
+          "hard_limit_source": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          },
+          "joint_role_manifest": {
+            "available": 10,
+            "unavailable": 0,
+            "unknown": 0
+          }
+        },
+        "hard_limit_sources": {
+          "runtime.joint_pos_limits": 10
+        },
+        "soft_limit_sources": {
+          "runtime.soft_joint_pos_limits": 10
+        },
+        "unknown_runtime_joint_names": [],
+        "baseline_checks": {
+          "post_home": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          },
+          "post_reset": {
+            "checked": 10,
+            "hard_violation_episodes": 0
+          }
+        }
+      }
+    },
+    "robot_motion": {
+      "joint_acceleration_rms": 20.435814333606697,
+      "joint_effort_rms": 46.37797538179373,
+      "joint_jerk_rms": 2096.443657576408,
+      "joint_velocity_rms": 0.39872399002143477
+    },
+    "grasp": {
+      "kinematic_grasp_stability_index": 0.9920605397375284,
+      "mean_kinematic_grasp_stability": 0.6440057063325317,
+      "per_object_best_kinematic_grasp_stability": {
+        "obj_330_bowl_2": 0.9647714908991828,
+        "obj_330_ladle_3": 0.9608106828040366
+      },
+      "grasp_gsi_diagnostics": {
+        "mean_anti_slip_score": 0.7896777101645251,
+        "mean_duration_score": 0.7264153707459993,
+        "mean_hold_score": 0.4236384381338793,
+        "mean_kinematic_stability_score": 0.2652297121587954,
+        "mean_legacy_gsi": 0.5461500173370503,
+        "mean_lift_score": 0.5291075050709939,
+        "mean_motion_stability_score": 0.37767901349760563
+      }
+    },
+    "tool": {
+      "tool_selection_accuracy": null,
+      "tool_switch_success_rate": null,
+      "tool_switch_total": 0,
+      "tool_switch_successful": 0
+    },
+    "runtime": {
+      "avg_steps": 2958,
+      "error_count": 0
+    }
+  }
+}