rebuild(screener): add controlled formulas and rolling backtests

next/tests/test_screener.py

@@ -6,12 +6,17 @@ import pytest
 
 from backend.database.connection import Database
 from backend.database.migrations import MIGRATIONS, MigrationRunner
+from backend.features.screener.backtest import (
+    attach_historical_estimate,
+    rolling_backtest,
+)
 from backend.features.screener.catalog import (
     CatalogError,
     factor_catalog,
     strategy_catalog,
     validate_formula,
 )
+from backend.features.screener.compiler import parse_compiled_formula
 from backend.features.screener.engine import execute_formula
 from backend.features.screener.repository import ScreenerRepository, decode_track
 from backend.features.screener.service import automatic_strategies
@@ -145,6 +150,65 @@ def test_formula_rejects_invalid_comparisons_and_duplicate_scores() -> None:
         validate_formula(duplicate)
 
 
+def test_natural_language_output_is_reduced_to_the_controlled_formula_schema() -> None:
+    compiled = parse_compiled_formula(
+        """```json
+        {
+          "universe": {"exclude_st": true, "listed_days_min": 120},
+          "filters": [{"field": "amount_billion", "op": ">=", "value": 3}],
+          "score": [
+            {"field": "return_20d", "weight": 60, "direction": "desc"},
+            {"field": "sector_strength", "weight": 40, "direction": "desc"}
+          ],
+          "limit": 20,
+          "min_score": 55,
+          "invented_instruction": "直接选择某只股票"
+        }
+        ```"""
+    )
+
+    assert [item["weight"] for item in compiled["score"]] == [0.6, 0.4]
+    assert compiled["min_score"] == 0.55
+    assert "invented_instruction" not in compiled
+    with pytest.raises(CatalogError, match="未知筛选因子"):
+        parse_compiled_formula(
+            json.dumps(
+                {
+                    **compiled,
+                    "filters": [{"field": "future_price", "op": ">", "value": 1}],
+                }
+            )
+        )
+
+
+def test_rolling_backtest_hides_small_samples_and_uses_t_plus_three() -> None:
+    formula = _formula()
+    snapshots = [
+        {
+            "trade_date": f"2026-07-{index + 1:02d}",
+            "coverage": {"market": 1},
+            "rows": [_row("000001.SZ", 10 + index)],
+        }
+        for index in range(23)
+    ]
+
+    small = rolling_backtest(snapshots[:10], formula)
+    assert small["sample_size"] == 7
+    assert small["stable"] is False
+    assert small["win_rate"] is None
+    assert small["average_return_3d"] is None
+
+    stable = rolling_backtest(snapshots, formula)
+    assert stable["sample_size"] == 20
+    assert stable["stable"] is True
+    assert stable["win_rate"] == 100
+    assert stable["average_return_3d"] == pytest.approx(16.99, abs=0.01)
+    enriched = attach_historical_estimate(
+        {"items": [{"score_display": 80.0}]}, stable
+    )
+    assert enriched["items"][0]["historical_estimate"] == 93.0
+
+
 def test_custom_strategies_and_tracks_are_account_isolated(tmp_path) -> None:
     database = Database(tmp_path / "screener.db")
     MigrationRunner(database).upgrade(MIGRATIONS)
@@ -173,6 +237,16 @@ def test_custom_strategies_and_tracks_are_account_isolated(tmp_path) -> None:
             missing_fields=[],
             result=[{**_row("000001.SZ", 10), "score": 1}],
         )
+        repository.save_backtest(
+            connection,
+            int(run["id"]),
+            {"sample_size": 7, "stable": False, "win_rate": None},
+        )
+        repository.save_backtest(
+            connection,
+            int(run["id"]),
+            {"sample_size": 99, "stable": True, "win_rate": 100},
+        )
         repository.add_track(
             connection,
             user_id=1,
@@ -185,6 +259,11 @@ def test_custom_strategies_and_tracks_are_account_isolated(tmp_path) -> None:
         assert len(repository.custom_strategies(connection, 2)) == 1
         assert len(repository.tracks(connection, 1)) == 1
         assert repository.tracks(connection, 2) == ()
+        assert repository.backtest(connection, int(run["id"])) == {
+            "sample_size": 7,
+            "stable": False,
+            "win_rate": None,
+        }
 
 
 def test_running_a_strategy_never_creates_tracking_rows(tmp_path) -> None:
@@ -219,7 +298,7 @@ def test_running_a_strategy_never_creates_tracking_rows(tmp_path) -> None:
         stored = connection.execute(
             "SELECT result_json FROM screener_runs WHERE id = ?", (run["id"],)
         ).fetchone()
-        assert len(json.loads(stored["result_json"])) == 1
+    assert len(json.loads(stored["result_json"])) == 1
 
 
 def test_tracking_statistics_and_milestone_events_are_persistent_and_idempotent(tmp_path) -> None: