规则库 3.8.1 + 排盘修复 + 评测体系

- 规则库 3.6→3.8.1（542→548 条）：剪枝上锁、R-杂-38 从格临界双轨、 R-事-10~13 冲刑方向判别、R-禁-23 自造必要条件之禁 - 排盘修复随 3.6.2 已发：港澳时区夏令时双向误拨、晚子时双重换日 - 新增评测/ 目录（BaziQA 题答分离、判分工具、记分卡、with/without-skill 对照）与 scripts/query_rules.py 条款检索；ZIPEC_EVAL 评测额度豁免 - manifest 重建并重新签名；VERSION 保持 3.6.2，不发版不打 tag
2026-10-13 12:34:36 +08:00
parent b20fd5c3c9
commit 1a0eed1999
338 changed files with 36533 additions and 30 deletions
@@ -0,0 +1,128 @@
+{
+ "model": "qoder-agent",
+ "skill_mode": "without-skill",
+ "skill_version": "无（裸模型对照）",
+ "eval_date": "2026-10-13",
+ "benchmark": "BaziQA-全球算命师大赛",
+ "overall": {
+  "total": 200,
+  "correct": 78,
+  "accuracy": 39.0,
+  "no_answer": 0
+ },
+ "baseline": {
+  "name": "官方基线（Gemini）",
+  "overall": {
+   "total": 200,
+   "correct": 69,
+   "accuracy": 34.5
+  },
+  "per_contest": {
+   "contest8_2021": 27.5,
+   "contest8_2022": 40.0,
+   "contest8_2023": 27.5,
+   "contest8_2024": 35.0,
+   "contest8_2025": 42.5
+  },
+  "per_dimension": {
+   "流年运势": 40.5,
+   "感情": 36.7,
+   "六亲": 27.6,
+   "事业": 24.0,
+   "性格": 45.0,
+   "学业": 41.7,
+   "财富": 30.8,
+   "健康": 23.1,
+   "其他": 50.0
+  }
+ },
+ "per_contest": {
+  "contest8_2021": {
+   "total": 40,
+   "correct": 18,
+   "no_answer": 0,
+   "accuracy": 45.0
+  },
+  "contest8_2022": {
+   "total": 40,
+   "correct": 17,
+   "no_answer": 0,
+   "accuracy": 42.5
+  },
+  "contest8_2023": {
+   "total": 40,
+   "correct": 13,
+   "no_answer": 0,
+   "accuracy": 32.5
+  },
+  "contest8_2024": {
+   "total": 40,
+   "correct": 14,
+   "no_answer": 0,
+   "accuracy": 35.0
+  },
+  "contest8_2025": {
+   "total": 40,
+   "correct": 16,
+   "no_answer": 0,
+   "accuracy": 40.0
+  }
+ },
+ "per_dimension": {
+  "流年运势": {
+   "total": 37,
+   "correct": 14,
+   "no_answer": 0,
+   "accuracy": 37.8
+  },
+  "感情": {
+   "total": 49,
+   "correct": 16,
+   "no_answer": 0,
+   "accuracy": 32.7
+  },
+  "六亲": {
+   "total": 29,
+   "correct": 15,
+   "no_answer": 0,
+   "accuracy": 51.7
+  },
+  "事业": {
+   "total": 25,
+   "correct": 8,
+   "no_answer": 0,
+   "accuracy": 32.0
+  },
+  "性格": {
+   "total": 20,
+   "correct": 6,
+   "no_answer": 0,
+   "accuracy": 30.0
+  },
+  "学业": {
+   "total": 12,
+   "correct": 7,
+   "no_answer": 0,
+   "accuracy": 58.3
+  },
+  "财富": {
+   "total": 13,
+   "correct": 7,
+   "no_answer": 0,
+   "accuracy": 53.8
+  },
+  "健康": {
+   "total": 13,
+   "correct": 4,
+   "no_answer": 0,
+   "accuracy": 30.8
+  },
+  "其他": {
+   "total": 2,
+   "correct": 1,
+   "no_answer": 0,
+   "accuracy": 50.0
+  }
+ },
+ "notes": "对照组：同模型不加载技能——无排盘脚本、无规则库、无 SKILL.md，仅给盲版题面的出生信息，模型手排四柱（自行决定时区/真太阳时口径并留档）作答；求解留档：评测/ai答案/qoder-agent/without-skill/（41 位命主）；判分：评测/工具/run_eval.py（与实验组同题集、同答案键、同脚本）"
+}