Merge pull request #292 from kevin9327/fix/composite-single-candidate-neutral

fix(strategy): 叠加策略单候选子策略改用中性分, 与回测合并同口径

backend/app/strategy/composite.py

@@ -80,7 +80,12 @@ def merge_results(
             ordered = sorted(symbols, key=lambda s: res.scores[s], reverse=True)
             count = len(ordered)
             for rank, sym in enumerate(ordered, start=1):
-                norm[sym] = 1 - (rank - 1) / max(count - 1, 1)
+                # 单候选无法排名, 必须用中性分: 当成"最优=1"会凭空抬高融合分,
+                # 而回测合并 (merge_signal_matrices 的 n <= 1 分支) 用的是中性分,
+                # 两条路径同一天同一标的会给出不同评分与排序。
+                norm[sym] = (
+                    _NEUTRAL_NORM if count <= 1 else 1 - (rank - 1) / (count - 1)
+                )
         else:
             # 子策略未产出 score: 命中即中性分, 不奖励也不惩罚。
             for row in res.rows:

backend/tests/test_composite_strategy.py

@@ -399,6 +399,54 @@ def test_composite_no_scores_uses_neutral(tmp_path):
     assert all(abs(s - 50.0) < 0.01 for s in merged.scores.values())  # 中性分 0.5*100
 
 
+def test_composite_single_candidate_child_matches_backtest_merge():
+    """子策略当天只选出一只票时, 选股合并与回测合并必须给出同一套评分。
+
+    单候选无法排名, 只能用中性分 0.5; 若当成"最优=1"会凭空抬高该票的融合分,
+    与 merge_signal_matrices (n <= 1 → 中性分) 分叉 —— 同一天同一标的在选股页
+    和回测里评分与排序都不一样, 正是本模块要防的口径分裂。
+    """
+    from app.backtest.matrix import make_signal_matrix
+    from app.strategy import composite as composite_mod
+    from app.strategy.engine import StrategyResult
+
+    as_of = date(2026, 1, 2)
+    child_a = StrategyResult(as_of=as_of, strategy_id="a", scores={"X": 7.0})
+    child_b = StrategyResult(
+        as_of=as_of, strategy_id="b", scores={"X": 1.0, "Y": 3.0, "Z": 2.0}
+    )
+    merged = composite_mod.merge_results(
+        [child_a, child_b], [1.0, 1.0], "union", 0, as_of=as_of, strategy_id="blend"
+    )
+
+    shape = (1, 3)  # 一个交易日, 三只标的 X/Y/Z
+
+    def _sig(entry: list[int], score: list[float]):
+        return make_signal_matrix(
+            shape,
+            entry=np.array([entry], dtype=np.uint8),
+            exit=np.zeros(shape, dtype=np.uint8),
+            score=np.array([score], dtype=np.float32),
+        )
+
+    matrix = composite_mod.merge_signal_matrices(
+        shape,
+        [_sig([1, 0, 0], [7.0, 0.0, 0.0]), _sig([1, 1, 1], [1.0, 3.0, 2.0])],
+        [("a", 1.0), ("b", 1.0)],
+        "union",
+        0,
+        max_hold=1,
+    )
+    backtest_scores = dict(zip(("X", "Y", "Z"), matrix.score[0], strict=True))
+
+    # X 只被单候选子策略 a 命中: 中性分 0.5 与 b 的最差名 0 融合 → 25 分
+    assert abs(merged.scores["X"] - 25.0) < 0.01
+    for symbol in ("X", "Y", "Z"):
+        assert abs(merged.scores[symbol] - float(backtest_scores[symbol])) < 0.01, symbol
+    # 排序也一致: Y > Z > X
+    assert merged.scores["Y"] > merged.scores["Z"] > merged.scores["X"]
+
+
 def test_composite_empty_children_returns_empty(tmp_path):
     """空子结果列表 → 返回空 StrategyResult。"""
     from app.strategy import composite as composite_mod