feat: compare legacy and live ranking universes

tests/unit/test_backtest_service.py

@@ -1166,12 +1166,29 @@ def test_slim_replay_can_retain_shorts_for_ranking_universe(monkeypatch):
     full_ranking_universe = bt._replay_candidates_for_period(
         "AAA", columns, {}, {}, None, date.min, "daily", True
     )
+    dual_ranking_replay = bt._replay_candidates_for_period(
+        "AAA", columns, {}, {}, None, date.min, "daily", True, True
+    )
 
     assert [row["direction"] for row in long_only] == ["long"]
     assert {row["direction"] for row in full_ranking_universe} == {
         "long",
         "short",
     }
+    assert len(dual_ranking_replay) == 2
+    assert sum(
+        bool(row.get("_universe_rank_observation"))
+        for row in dual_ranking_replay
+    ) == 1
+
+    monkeypatch.setattr(bt, "_window_setups", lambda *_args, **_kwargs: [])
+    rank_only = bt._replay_candidates_for_period(
+        "AAA", columns, {}, {}, None, date.min, "daily", True, True
+    )
+    assert len(rank_only) == 1
+    assert rank_only[0]["direction"] == "rank_only"
+    assert rank_only[0]["_rank_only"] is True
+    assert rank_only[0]["_universe_rank_observation"] is True
 
 
 def test_daily_replay_uses_exact_date_ranking_periods():

tests/unit/test_daily_reentry_matrix.py

@@ -1,6 +1,12 @@
-from datetime import date
+from datetime import date, timedelta
 
-from scripts.run_daily_reentry_matrix import PrecomputedDailyEngine, ReentryPolicy
+import pytest
+
+from scripts.run_daily_reentry_matrix import (
+    PrecomputedDailyEngine,
+    ReentryPolicy,
+    _live_universe_rank_map,
+)
 
 
 RANKING_KEY = "strategy_rank"
@@ -95,3 +101,72 @@ def test_two_session_confirmation_excludes_stop_day_close():
     emitted = _call(policy, ORD + 2, state, 2)
     assert emitted is not None
     assert emitted["_reentry_reason"] == "two_qualified_post_stop_closes"
+
+
+def _rank_observation(
+    symbol: str,
+    *,
+    raw: float,
+    residual: float,
+    volatility: float,
+) -> dict:
+    return {
+        "symbol": symbol,
+        "date": date.fromordinal(ORD).isoformat(),
+        "ranking_period": ("date", ORD),
+        "momentum": raw,
+        "residual_momentum": residual,
+        "vol_6m": volatility,
+    }
+
+
+def test_live_universe_rank_uses_each_ticker_once_and_residual_when_available():
+    observations = [
+        _rank_observation("AAA", raw=0.1, residual=0.3, volatility=0.1),
+        _rank_observation("BBB", raw=0.3, residual=0.1, volatility=0.2),
+        _rank_observation("CCC", raw=0.2, residual=0.2, volatility=0.3),
+    ]
+    first_benchmark_day = date.fromordinal(ORD) - timedelta(days=300)
+    benchmark = {
+        first_benchmark_day + timedelta(days=offset): 100.0
+        for offset in range(252)
+    }
+
+    ranks = _live_universe_rank_map(observations, benchmark, 0.8)
+
+    assert ranks[("AAA", date.fromordinal(ORD).isoformat())] == {
+        "momentum_percentile": 100.0,
+        "volatility_percentile": 0.0,
+        "strategy_rank": 80.0,
+    }
+    assert ranks[("BBB", date.fromordinal(ORD).isoformat())][
+        "momentum_percentile"
+    ] == 0.0
+    assert ranks[("CCC", date.fromordinal(ORD).isoformat())][
+        "strategy_rank"
+    ] == 60.0
+
+
+def test_live_universe_rank_uses_raw_fallback_before_benchmark_is_ready():
+    observations = [
+        _rank_observation("AAA", raw=0.1, residual=0.3, volatility=0.1),
+        _rank_observation("BBB", raw=0.3, residual=0.1, volatility=0.2),
+    ]
+
+    ranks = _live_universe_rank_map(observations, {}, 0.8)
+
+    assert ranks[("AAA", date.fromordinal(ORD).isoformat())][
+        "momentum_percentile"
+    ] == 0.0
+    assert ranks[("BBB", date.fromordinal(ORD).isoformat())][
+        "momentum_percentile"
+    ] == 100.0
+
+
+def test_live_universe_rank_rejects_duplicate_ticker_date():
+    observation = _rank_observation(
+        "AAA", raw=0.1, residual=0.2, volatility=0.1
+    )
+
+    with pytest.raises(ValueError, match="one observation"):
+        _live_universe_rank_map([observation, dict(observation)], {}, 0.8)