feat(PX): 自训练v1.5 P0准备 — NHN模式 + 总分法胜率解析 + 评估管线

CB-01: Program.cs 加 --nhn 模式 [NEURAL,ISMCTS,NEURAL]
- 3人跑得快 N插花坐排列
- Rotations 数组补全为4种排列 (IIN/NII/INI/NHN)
- --nhn 参数: 不生成CSV, 直接评估

CB-02: 胜率解析统一总分法
- TrainingCollector 加 ScoreAccumulator/BotWins/BotGames 静态累加器
- GameOver 无条件调用 AccumulateScore (eval + 训练模式均记录)
- stress_summary.txt 输出 #BOT_SCORES: bot=wins/games,wr=XX%,score=YYY
- orchestrator eval 解析改用 #BOT_SCORES 格式, 正则回退逻辑移除

管线: orchestrator eval→stress_summary→总分法解析, 闭环可验证

scripts/training_orchestrator.py

@@ -209,13 +209,21 @@ def train_v(prev_winrate=None):
             with open(sf) as f:
                 summary = f.read()
             log(f"  {summary.strip()}")
-            # parse Neural win rate from summary
-            m = re.search(r'NEURAL.*?(\d+)\s*wins?.*?(\d+)\s*games?', summary, re.IGNORECASE)
-            if not m:
-                m = re.search(r'position.*?(\d+).*?wins?.*?(\d+)', summary, re.IGNORECASE)
-            if m:
-                wins, games = int(m.group(1)), int(m.group(2))
-                winrate = wins / games * 100 if games > 0 else 0.0
+            # parse Neural win rate from #BOT_SCORES (总分法: 每局 Score>0 算赢)
+            in_bot_scores = False
+            for line in summary.splitlines():
+                line = line.strip()
+                if line == "#BOT_SCORES":
+                    in_bot_scores = True
+                    continue
+                if in_bot_scores and line.startswith("NEURAL="):
+                    # Format: NEURAL=wins/games,wr=XX.X%,score=YYY
+                    m = re.search(r'wr=(\d+\.?\d*)%', line)
+                    if m:
+                        winrate = float(m.group(1))
+                    break
+            if winrate == 0.0:
+                log("  ⚠️ BOT_SCORES not found, fallback to legacy parse")
         log(f"  Neural winrate: {winrate:.1f}%")
     except Exception as e:
         log(f"  eval failed: {e}")