{
  "edition": "2026.10.27",
  "sourceVersion": "Sample Gallery HY v2.0",
  "counts": {
    "tasks": 13,
    "families": 6,
    "base": 169,
    "astra": 65,
    "hy": 104,
    "hintExcluded": 15
  },
  "overall": {
    "astra": 30.425193999855384,
    "hy": 24.277978567766347,
    "gap": 6.1472154320890375
  },
  "cases": [
    {
      "id": "020_LNG_receiving_terminal",
      "family": "01",
      "astra": {
        "mean": 67.47994,
        "min": 57.0465,
        "max": 75.5241,
        "n": 5,
        "runs": [
          {
            "slot": 1,
            "score": 75.5241,
            "source": "trajectory/020_LNG_receiving_terminal/gpt-6-astra/rollout_codex_1/reward.json",
            "sha256": "14c36d742da82d4c246697b7bd39308f09b8d66f347bac08617faf0040e37658"
          },
          {
            "slot": 2,
            "score": 71.9967,
            "source": "trajectory/020_LNG_receiving_terminal/gpt-6-astra/rollout_codex_2/reward.json",
            "sha256": "98c26a26aee8df6b1f94f16fefdee31502897daffc60ead8138f411150fdb95d"
          },
          {
            "slot": 3,
            "score": 57.0465,
            "source": "trajectory/020_LNG_receiving_terminal/gpt-6-astra/rollout_codex_3/reward.json",
            "sha256": "c3d3702bdbd702f38171a779087cb5ad5773ff4af151e52b09500e2b5ff3a515"
          },
          {
            "slot": 4,
            "score": 75.5241,
            "source": "trajectory/020_LNG_receiving_terminal/gpt-6-astra/rollout_codex_4/reward.json",
            "sha256": "14c36d742da82d4c246697b7bd39308f09b8d66f347bac08617faf0040e37658"
          },
          {
            "slot": 5,
            "score": 57.3083,
            "source": "trajectory/020_LNG_receiving_terminal/gpt-6-astra/rollout_codex_5/reward.json",
            "sha256": "58041f46a4e0b416c2470e0098500d3087f99e1ad9e4f528887fa04127e5bcac"
          }
        ]
      },
      "hy": {
        "mean": 28.07705,
        "min": 10.0,
        "max": 64.6164,
        "n": 8,
        "runs": [
          {
            "slot": 1,
            "score": 10.0,
            "source": "trajectory/020_LNG_receiving_terminal/Hy4-preview/rollout_codebuddy_code_1/reward.json",
            "sha256": "1a40aa3a69afbb0fbc113cd07a5e7b4961710bfcc68930d8e644a489f14adabd"
          },
          {
            "slot": 2,
            "score": 25.0,
            "source": "trajectory/020_LNG_receiving_terminal/Hy4-preview/rollout_codebuddy_code_2/reward.json",
            "sha256": "60d377ff1f5757e90d7bcd09710cd96e7dbc7f7048569c854659717ad4103ebe"
          },
          {
            "slot": 3,
            "score": 25.0,
            "source": "trajectory/020_LNG_receiving_terminal/Hy4-preview/rollout_codebuddy_code_3/reward.json",
            "sha256": "60d377ff1f5757e90d7bcd09710cd96e7dbc7f7048569c854659717ad4103ebe"
          },
          {
            "slot": 4,
            "score": 25.0,
            "source": "trajectory/020_LNG_receiving_terminal/Hy4-preview/rollout_codebuddy_code_4/reward.json",
            "sha256": "60d377ff1f5757e90d7bcd09710cd96e7dbc7f7048569c854659717ad4103ebe"
          },
          {
            "slot": 5,
            "score": 25.0,
            "source": "trajectory/020_LNG_receiving_terminal/Hy4-preview/rollout_codebuddy_code_5/reward.json",
            "sha256": "60d377ff1f5757e90d7bcd09710cd96e7dbc7f7048569c854659717ad4103ebe"
          },
          {
            "slot": 6,
            "score": 25.0,
            "source": "trajectory/020_LNG_receiving_terminal/Hy4-preview/rollout_codebuddy_code_6/reward.json",
            "sha256": "60d377ff1f5757e90d7bcd09710cd96e7dbc7f7048569c854659717ad4103ebe"
          },
          {
            "slot": 7,
            "score": 64.6164,
            "source": "trajectory/020_LNG_receiving_terminal/Hy4-preview/rollout_codebuddy_code_7/reward.json",
            "sha256": "81ff63362c008e77549720dbf7652ee0dce4883891acf34a8cbe2b8a6d0b96cf"
          },
          {
            "slot": 8,
            "score": 25.0,
            "source": "trajectory/020_LNG_receiving_terminal/Hy4-preview/rollout_codebuddy_code_8/reward.json",
            "sha256": "60d377ff1f5757e90d7bcd09710cd96e7dbc7f7048569c854659717ad4103ebe"
          }
        ]
      },
      "gap": 39.40289,
      "name": "LNG 接收站",
      "label": "项目建模与估值",
      "number": "01",
      "question": "项目经营、融资与估值，能否在同一模型中保持联动？",
      "value": "把资本开支、运营假设与财务输出放进同一业务场景，观察模型是否能稳定处理跨期关系。",
      "suggestion": "建议优先验证长链条财务关系：将经营假设、融资安排和终值计算拆成中间检查点，再测试输入变化能否持续传导。"
    },
    {
      "id": "T04_177",
      "family": "04",
      "astra": {
        "mean": 52.63116199812,
        "min": 19.9024813258,
        "max": 100.0,
        "n": 5,
        "runs": [
          {
            "slot": 1,
            "score": 19.9024813258,
            "source": "trajectory/T04_177/gpt-6-astra/rollout_codex_1/reward.json",
            "sha256": "098c12008a0de587d9670e09fcd76279a55a61d2cd22a8b2933eb21d9daae54c"
          },
          {
            "slot": 2,
            "score": 21.6266643324,
            "source": "trajectory/T04_177/gpt-6-astra/rollout_codex_2/reward.json",
            "sha256": "b4f0a9cec4e4b94b8578894516a8eba629564b53fef73bd576a997cb74354266"
          },
          {
            "slot": 3,
            "score": 21.6266643324,
            "source": "trajectory/T04_177/gpt-6-astra/rollout_codex_3/reward.json",
            "sha256": "b4f0a9cec4e4b94b8578894516a8eba629564b53fef73bd576a997cb74354266"
          },
          {
            "slot": 4,
            "score": 100.0,
            "source": "trajectory/T04_177/gpt-6-astra/rollout_codex_4/reward.json",
            "sha256": "159eb8ada264672ec3e9e5c41399dd6711eb5d41d0aa09f62c2075f33308615a"
          },
          {
            "slot": 5,
            "score": 100.0,
            "source": "trajectory/T04_177/gpt-6-astra/rollout_codex_5/reward.json",
            "sha256": "159eb8ada264672ec3e9e5c41399dd6711eb5d41d0aa09f62c2075f33308615a"
          }
        ]
      },
      "hy": {
        "mean": 30.6168338809625,
        "min": 18.4348414494,
        "max": 49.0,
        "n": 8,
        "runs": [
          {
            "slot": 1,
            "score": 18.4348414494,
            "source": "trajectory/T04_177/Hy4-preview/rollout_codebuddy_code_1/reward.json",
            "sha256": "7bd978e9918a5d8fda546c059d4746c46ae3032c282ab35bdc6ae4d5a1b628a3"
          },
          {
            "slot": 2,
            "score": 49.0,
            "source": "trajectory/T04_177/Hy4-preview/rollout_codebuddy_code_2/reward.json",
            "sha256": "2dd58674786e563b460f087e66716b9089b8129bb07cc4652b70cce4f25ebf5a"
          },
          {
            "slot": 3,
            "score": 20.2733029878,
            "source": "trajectory/T04_177/Hy4-preview/rollout_codebuddy_code_3/reward.json",
            "sha256": "c3be94b11e0a2e6d7e934f9fd6bff30e979a50406690e826fc4b6917831976aa"
          },
          {
            "slot": 4,
            "score": 20.3090172736,
            "source": "trajectory/T04_177/Hy4-preview/rollout_codebuddy_code_4/reward.json",
            "sha256": "cc4c73504b6425befc4ba878e6b7064ed9158356b48643db9f7edc92810d74fc"
          },
          {
            "slot": 5,
            "score": 19.0048506069,
            "source": "trajectory/T04_177/Hy4-preview/rollout_codebuddy_code_5/reward.json",
            "sha256": "b8ab9a9bd6f7a491a55f732f33d4dad93ad6830de63464bf02f19f7e9e1ca229"
          },
          {
            "slot": 6,
            "score": 49.0,
            "source": "trajectory/T04_177/Hy4-preview/rollout_codebuddy_code_6/reward.json",
            "sha256": "2dd58674786e563b460f087e66716b9089b8129bb07cc4652b70cce4f25ebf5a"
          },
          {
            "slot": 7,
            "score": 19.91265873,
            "source": "trajectory/T04_177/Hy4-preview/rollout_codebuddy_code_7/reward.json",
            "sha256": "74279fa8eec8ba45025dd75f167e4de4b8bb0a8000c9f98dab6513ef549b06b5"
          },
          {
            "slot": 8,
            "score": 49.0,
            "source": "trajectory/T04_177/Hy4-preview/rollout_codebuddy_code_8/reward.json",
            "sha256": "2dd58674786e563b460f087e66716b9089b8129bb07cc4652b70cce4f25ebf5a"
          }
        ]
      },
      "gap": 22.0143281171575,
      "name": "容量与工单约束",
      "label": "业务规则与约束执行",
      "number": "02",
      "question": "多份业务规则叠加后，模型能否持续守住约束边界？",
      "value": "将容量、工单、审批与资金条件组合为可执行场景，让规则理解与实际数值响应一起接受检验。",
      "suggestion": "建议围绕约束优先级、审批有效期和可用容量建立专项样本，并以边界扰动检查规则是否真正进入计算。"
    },
    {
      "id": "S2E2_441",
      "family": "03",
      "astra": {
        "mean": 43.610079999999996,
        "min": 36.3544,
        "max": 53.98729999999999,
        "n": 5,
        "runs": [
          {
            "slot": 1,
            "score": 37.367,
            "source": "trajectory/S2E2_441/gpt-6-astra/rollout_codex_1/reward.json",
            "sha256": "7464c2a81c652f8731e4205f517726134468a276e3c17c949e4d0c5cef0ceb63"
          },
          {
            "slot": 2,
            "score": 53.98729999999999,
            "source": "trajectory/S2E2_441/gpt-6-astra/rollout_codex_2/reward.json",
            "sha256": "4c315b2b8c290355256cb6d53ceb149aec4a2fa567f0af70321c19dc4aaeff97"
          },
          {
            "slot": 3,
            "score": 53.98729999999999,
            "source": "trajectory/S2E2_441/gpt-6-astra/rollout_codex_3/reward.json",
            "sha256": "4c315b2b8c290355256cb6d53ceb149aec4a2fa567f0af70321c19dc4aaeff97"
          },
          {
            "slot": 4,
            "score": 36.3544,
            "source": "trajectory/S2E2_441/gpt-6-astra/rollout_codex_4/reward.json",
            "sha256": "11dcb1137d5e61829917c64c3348c9982f1d19a120bd7967822518021a520a7d"
          },
          {
            "slot": 5,
            "score": 36.3544,
            "source": "trajectory/S2E2_441/gpt-6-astra/rollout_codex_5/reward.json",
            "sha256": "11dcb1137d5e61829917c64c3348c9982f1d19a120bd7967822518021a520a7d"
          }
        ]
      },
      "hy": {
        "mean": 30.136824999999998,
        "min": 5.0,
        "max": 47.5,
        "n": 8,
        "runs": [
          {
            "slot": 1,
            "score": 40.0,
            "source": "trajectory/S2E2_441/Hy4-preview/rollout_codebuddy_code_1/reward.json",
            "sha256": "6447cdaafc3b70f17e63a3b75dd7aa32751bfab547affd928402e035adbe164e"
          },
          {
            "slot": 2,
            "score": 5.0,
            "source": "trajectory/S2E2_441/Hy4-preview/rollout_codebuddy_code_2/reward.json",
            "sha256": "1444e2de42b5bc09e6800929914177c4574591b3620b8330d6c07b392adbf0ac"
          },
          {
            "slot": 3,
            "score": 25.619999999999997,
            "source": "trajectory/S2E2_441/Hy4-preview/rollout_codebuddy_code_3/reward.json",
            "sha256": "3dfeb85e9ee5312381dd22b66486eefe714a002b71d0b162dc9fd3a2a6a6044a"
          },
          {
            "slot": 4,
            "score": 38.9873,
            "source": "trajectory/S2E2_441/Hy4-preview/rollout_codebuddy_code_4/reward.json",
            "sha256": "59b1bc4956e909bc597971f56e8e79397399df2169480ac3fc9344ade36c213a"
          },
          {
            "slot": 5,
            "score": 5.0,
            "source": "trajectory/S2E2_441/Hy4-preview/rollout_codebuddy_code_5/reward.json",
            "sha256": "1444e2de42b5bc09e6800929914177c4574591b3620b8330d6c07b392adbf0ac"
          },
          {
            "slot": 6,
            "score": 47.5,
            "source": "trajectory/S2E2_441/Hy4-preview/rollout_codebuddy_code_6/reward.json",
            "sha256": "6abb1e48f8c496627c394f626780b36f8e4a46479953051d14a9665f9981e447"
          },
          {
            "slot": 7,
            "score": 38.9873,
            "source": "trajectory/S2E2_441/Hy4-preview/rollout_codebuddy_code_7/reward.json",
            "sha256": "59b1bc4956e909bc597971f56e8e79397399df2169480ac3fc9344ade36c213a"
          },
          {
            "slot": 8,
            "score": 40.0,
            "source": "trajectory/S2E2_441/Hy4-preview/rollout_codebuddy_code_8/reward.json",
            "sha256": "6447cdaafc3b70f17e63a3b75dd7aa32751bfab547affd928402e035adbe164e"
          }
        ]
      },
      "gap": 13.473254999999998,
      "name": "复杂现金流联动",
      "label": "多交易财务传导",
      "number": "03",
      "question": "多笔交易同时变化时，现金流与报表结果能否一致响应？",
      "value": "从交易识别延伸到组合场景和输入扰动，为财务逻辑的一致性提供更细的检验切面。",
      "suggestion": "建议把单笔交易、组合交易与独立输入扰动连成训练和复测序列，重点观察现金流分类与报表联动是否稳定。"
    }
  ],
  "method": {
    "score": "原始 reward × 100，称为得分率；不是通过率。",
    "overall": "每题先分别计算各模型全部基础运行的均值，再对13题等权平均。",
    "selection": "全部满足 Astra 平均得分率 − HY 平均得分率 ≥ 10 个百分点的任务，共3题；各题保留全部5次/8次基础运行。",
    "conditions": "Astra：codex；HY：codebuddy_code。历史运行条件不同，不是统一执行框架的控制实验。",
    "exclusions": "15次Astra hint不纳入基础比较。缺少一致计时和用量数据，不比较耗时、Token或成本。"
  },
  "sources": [
    {
      "name": "RECORDS.json",
      "sha256": "1fba42312b4db0c1a2e66312adcac417e97c6f37f75901660030c9b91a7db6e4"
    },
    {
      "name": "SUMMARY.json",
      "sha256": "1c2646151d59ffc3d2d82540b5bc702ef2f836e1766f668040843db3bf5b41ca"
    },
    {
      "name": "TASK_INDEX.json",
      "sha256": "ea2312ada4ea0f02a37ed0ef12bad29737c5ff93dbc2f4d052705c3590db7f8c"
    },
    {
      "name": "CURRENT_SCORING_DECLARATIONS.json",
      "sha256": "d0a92ee85d7ff0914ad7b98b6a7f423b5d6719f0edf4ab6ce5908823e754f4a7"
    }
  ],
  "traceExample": {
    "case": "020_LNG_receiving_terminal",
    "rule": "各模型选择最接近本题基础得分中位数的slot；距离相同取最小slot。",
    "records": [
      {
        "instance_id": "020_LNG_receiving_terminal",
        "model": "gpt-6-astra",
        "harness": "codex",
        "condition": "base",
        "slot": 2,
        "reward": 0.719967,
        "score_percent": 71.9967,
        "step_count": 21,
        "tool_call_count": 15,
        "has_timestamps": true,
        "trajectory": "trajectory/020_LNG_receiving_terminal/gpt-6-astra/rollout_codex_2/trajectory.json",
        "trajectory_sha256": "98666901fd2e96f1097ae5b3635f51527bf82719bbfa2531182a720eb9ecb2a9",
        "reward_file": "trajectory/020_LNG_receiving_terminal/gpt-6-astra/rollout_codex_2/reward.json",
        "reward_sha256": "98c26a26aee8df6b1f94f16fefdee31502897daffc60ead8138f411150fdb95d",
        "evaluation": "trajectory/020_LNG_receiving_terminal/gpt-6-astra/rollout_codex_2/evaluation.json",
        "evaluation_sha256": "24d1c2f7c92f7277120144f14e212fc1d1448ea124997fb023b66906401c5738",
        "median_reward": 0.719967,
        "milestones": [
          {
            "step": 6,
            "label": "读取公开数据与模型接口"
          },
          {
            "step": 10,
            "label": "编写联动模型"
          },
          {
            "step": 11,
            "label": "LibreOffice重算"
          },
          {
            "step": 14,
            "label": "修订公式与再次重算"
          },
          {
            "step": 20,
            "label": "公开接口校验通过"
          }
        ],
        "validation_observed": "step20工具返回：OK: finance-strict-v1; exact keys and LibreOffice recalculation passed."
      },
      {
        "instance_id": "020_LNG_receiving_terminal",
        "model": "Hy4-preview",
        "harness": "codebuddy_code",
        "condition": "base",
        "slot": 2,
        "reward": 0.25,
        "score_percent": 25.0,
        "step_count": 41,
        "tool_call_count": 23,
        "has_timestamps": false,
        "trajectory": "trajectory/020_LNG_receiving_terminal/Hy4-preview/rollout_codebuddy_code_2/trajectory.json",
        "trajectory_sha256": "893ad4e2fdc1b18a84bf046602ea87bdb8f8520f921b4d399464e9904d17f3b0",
        "reward_file": "trajectory/020_LNG_receiving_terminal/Hy4-preview/rollout_codebuddy_code_2/reward.json",
        "reward_sha256": "60d377ff1f5757e90d7bcd09710cd96e7dbc7f7048569c854659717ad4103ebe",
        "evaluation": "trajectory/020_LNG_receiving_terminal/Hy4-preview/rollout_codebuddy_code_2/evaluation.json",
        "evaluation_sha256": "30f0511447c81469113cc11d76b3bd6925b4b9a4caa0d2238d7914bf5e56a006",
        "median_reward": 0.25,
        "milestones": [
          {
            "step": 3,
            "label": "读取公开文件与模型接口"
          },
          {
            "step": 27,
            "label": "编写联动模型"
          },
          {
            "step": 31,
            "label": "初次校验发现公式错误"
          },
          {
            "step": 36,
            "label": "修订与再次检查"
          },
          {
            "step": 40,
            "label": "公开接口校验通过"
          }
        ],
        "validation_observed": "step40工具返回：OK: finance-strict-v1; exact keys and LibreOffice recalculation passed."
      }
    ],
    "interpretation": "步骤按原始轨迹编号展示，不代表耗时；两侧均通过公开接口校验，深层业务评分仍有差异。"
  }
}