feat: 足球 LLM 预测服务初始提交

Profeto — 给 LLM 提供数据,让 LLM 预测足球比分。

核心模块:
- FastAPI 后端 + PostgreSQL (SQLAlchemy async)
- 多 Agent LLM 预测 (5 专家 + 终裁)
- 数据采集 (bzzoiro / understat / injuries)
- React 前端 (Vite + Tailwind)

包含:
- 数据源抽象 (DataSource 协议 + 注册表)
- Alembic 数据库迁移
- Prompt 模板 (单/多 Agent)
- 核心路径单元测试

src/llm/eval.py

@@ -0,0 +1,81 @@
+"""评估:赛后回填 + 统计。"""
+from __future__ import annotations
+
+import logging
+
+from sqlalchemy import select
+
+from src.db.base import AsyncSessionLocal
+from src.db.models import Prediction
+
+logger = logging.getLogger(__name__)
+
+
+async def settle_prediction(prediction_id: int, home_goals: int, away_goals: int) -> Prediction:
+    """回填实际结果。"""
+    async with AsyncSessionLocal() as db:
+        pred = await db.get(Prediction, prediction_id)
+        if pred is None:
+            raise ValueError(f"prediction {prediction_id} not found")
+        pred.actual_home_goals = home_goals
+        pred.actual_away_goals = away_goals
+        pred.settled = True
+        await db.commit()
+        await db.refresh(pred)
+        return pred
+
+
+def _actual_1x2(home: int, away: int) -> str:
+    """根据实际比分返胜平负。"""
+    if home > away:
+        return "1"
+    if home == away:
+        return "X"
+    return "2"
+
+
+async def get_eval_summary() -> dict:
+    """按 provider × 模型聚合评估。"""
+    async with AsyncSessionLocal() as db:
+        stmt = (
+            select(Prediction)
+            .where(Prediction.settled == True)
+        )
+        result = await db.execute(stmt)
+        rows = list(result.scalars().all())
+
+        from collections import defaultdict
+        buckets: dict[tuple[str, str], dict] = defaultdict(lambda: {
+            "total": 0, "correct_1x2": 0, "score_errors": [], "conf_sum": 0.0, "conf_count": 0,
+        })
+        for p in rows:
+            key = (p.provider, p.model)
+            b = buckets[key]
+            b["total"] += 1
+            if p.actual_home_goals is None or p.actual_away_goals is None:
+                continue
+            actual = _actual_1x2(p.actual_home_goals, p.actual_away_goals)
+            if p.pred_1x2 == actual:
+                b["correct_1x2"] += 1
+            if p.pred_home_goals is not None and p.pred_away_goals is not None:
+                err = ((p.pred_home_goals - p.actual_home_goals) ** 2 +
+                       (p.pred_away_goals - p.actual_away_goals) ** 2) ** 0.5
+                b["score_errors"].append(err)
+            if p.confidence is not None:
+                b["conf_sum"] += p.confidence
+                b["conf_count"] += 1
+
+        summary = []
+        for (prov, model), b in sorted(buckets.items()):
+            acc = (b["correct_1x2"] / b["total"] * 100) if b["total"] else 0
+            avg_err = (sum(b["score_errors"]) / len(b["score_errors"])) if b["score_errors"] else None
+            avg_conf = (b["conf_sum"] / b["conf_count"]) if b["conf_count"] else None
+            summary.append({
+                "provider": prov,
+                "model": model,
+                "total": b["total"],
+                "accuracy_1x2": round(acc, 1),
+                "avg_score_rmse": round(avg_err, 2) if avg_err is not None else None,
+                "avg_confidence": round(avg_conf, 2) if avg_conf is not None else None,
+            })
+        return {"summary": summary}