ai-labeling: Prompt 1 v0.3(80%) + Prompt 3 v0.2(90%) + GT v0.4.2

Prompt 1 迭代 v0.1→v0.3：
- 多选字段加显著篇幅门槛+从属零件反例教学
- 前沿科技vs横切类比拆清（技术主角测试）
- 装备深解加同类体系/窄类别测试
- 跨域改为传播分类框（≥3域即标，非军事术语）
- 全对率 40%→80%

Prompt 3 开篇钩子 v0.1→v0.2：
- 阅读范围扩至3段（导视+主持人+首段解说）
- 强判定从紧迫感硬门槛改为6条独立路径
- 弱判定区分信息性提问vs悬念式提问
- 命中率 50%→90%

ground-truth v0.4.2：ep008补标跨域 + 20期opening_hook全标注
脚本：run_labeling/summarize 支持 opening_hook，summarize改从源GT读取

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

ai-labeling/scripts/run_labeling.py

@@ -44,6 +44,7 @@ ALL_EPISODES = list(range(1, 21))
 FIELD_PROMPT_MAP = {
     "narrative": "prompt2_narrative.md",
     "classification": "prompt1_classification.md",
+    "opening_hook": "prompt3_opening_hook.md",
 }
 
 
@@ -145,8 +146,8 @@ def main():
     parser.add_argument("--ep", type=int, help="单期编号")
     parser.add_argument("--all", action="store_true", help="跑全部")
     parser.add_argument("--model", default="mimo-v2.5-pro", help="模型键名")
-    parser.add_argument("--field", default="narrative", choices=["narrative", "classification"],
-                        help="打标字段: narrative(叙事结构) / classification(4分类)")
+    parser.add_argument("--field", default="narrative", choices=["narrative", "classification", "opening_hook"],
+                        help="打标字段: narrative(叙事结构) / classification(4分类) / opening_hook(开篇钩子)")
     args = parser.parse_args()
     if args.all:
         for ep in ALL_EPISODES:

ai-labeling/scripts/summarize.py

@@ -14,6 +14,16 @@ from pathlib import Path
 
 BASE_DIR = Path(__file__).parent.parent
 EXPERIMENTS_DIR = BASE_DIR / "experiments"
+GROUND_TRUTH = BASE_DIR / "benchmark-set" / "ground-truth.json"
+
+
+def load_ground_truth_map():
+    """从 ground-truth.json 源文件加载最新 GT（而非实验文件中的快照）。"""
+    try:
+        data = json.loads(GROUND_TRUTH.read_text(encoding="utf-8"))
+        return {ep["ep"]: ep for ep in data["episodes"]}
+    except Exception:
+        return {}
 
 
 def load_json(path):
@@ -57,6 +67,7 @@ def run(model, field="narrative"):
         return
 
     ep_files = latest_per_ep(files)
+    gt_map = load_ground_truth_map()  # 从源文件读最新 GT
     rows = []
     parse_fail = 0
 
@@ -67,12 +78,22 @@ def run(model, field="narrative"):
             rows.append({"ep": ep, "title": "?", "gt": "?", "pred": "解析失败", "hit": False, "conf": "?"})
             continue
 
-        gt = data.get("ground_truth", {})
+        # 优先用源文件 GT，回退到实验文件中的快照
+        gt = gt_map.get(ep, data.get("ground_truth", {}))
         result = data.get("result")
 
         title = gt.get("title", "?")
 
-        if field == "narrative":
+        if field == "opening_hook":
+            gt_val = gt.get("opening_hook")
+            pred_val = result.get("opening_hook") if result else None
+            conf = result.get("confidence", "?") if result else "?"
+            if gt_val is None:
+                rows.append({"ep": ep, "title": title, "gt": "(无标注)", "pred": pred_val or "解析失败", "hit": None, "conf": conf})
+            else:
+                hit = pred_val == gt_val if pred_val is not None else False
+                rows.append({"ep": ep, "title": title, "gt": gt_val, "pred": pred_val or "解析失败", "hit": hit, "conf": conf})
+        elif field == "narrative":
             gt_val = gt.get("narrative_structure", "?")
             pred_val = result.get("narrative_structure") if result else None
             conf = result.get("confidence", "?") if result else "?"
@@ -93,7 +114,24 @@ def run(model, field="narrative"):
             all_hit = all(v is True for v in sub.values() if v is not None)
             rows.append({"ep": ep, "title": title, "sub": sub, "all_hit": all_hit, "conf": conf})
 
-    if field == "narrative":
+    if field == "opening_hook":
+        for r in rows:
+            if r["hit"] is None:
+                mark = "-"
+            else:
+                mark = "✓" if r["hit"] else "✗"
+            conf_str = f'置信度:{r["conf"]}' if r["conf"] != "?" else ""
+            print(f'  ep{r["ep"]:03d} {r["title"]:<12} | 标准:{r["gt"]:<6} | {model}:{r["pred"]:<6} | {mark} {conf_str}')
+
+        scored = [r for r in rows if r["hit"] is not None]
+        hits = sum(1 for r in scored if r["hit"])
+        total = len(scored)
+        unlabeled = sum(1 for r in rows if r["hit"] is None)
+        print(f"\n  ===== {model} 命中情况 =====")
+        print(f"  opening_hook 命中: {hits}/{total} = {hits*100//total if total else 0}%")
+        print(f"  无标注(跳过): {unlabeled} 期")
+        print(f"  解析失败: {parse_fail} 期")
+    elif field == "narrative":
         # 打印每行
         for r in rows:
             mark = "✓" if r["hit"] else "✗"
@@ -139,7 +177,7 @@ def run(model, field="narrative"):
 if __name__ == "__main__":
     parser = argparse.ArgumentParser()
     parser.add_argument("--model", required=True, help="模型键名，如 mimo-v2.5-pro / deepseek-v4-pro")
-    parser.add_argument("--field", default="narrative", choices=["narrative", "classification"],
-                        help="打标字段: narrative(叙事结构) / classification(4分类)")
+    parser.add_argument("--field", default="narrative", choices=["narrative", "classification", "opening_hook"],
+                        help="打标字段: narrative(叙事结构) / classification(4分类) / opening_hook(开篇钩子)")
     args = parser.parse_args()
     run(args.model, args.field)