feat: compare legacy and live ranking universes

app/services/backtest_service.py

@@ -1040,15 +1040,19 @@ def _replay_candidates_for_period(
     start_date: date,
     cadence: str = DEFAULT_BACKTEST_CADENCE,
     include_short_candidates: bool = False,
+    include_universe_rank_observations: bool = False,
 ) -> list[dict]:
     """Slim picklable replay used by local event studies.
 
     Unlike the full report worker it skips factor-series construction and only
     evaluates setup dates on or after ``start_date``. Long-only remains the
     compatibility default. Set ``include_short_candidates`` when the caller
-    needs the production-faithful cross-sectional ranking universe; shorts can
-    then contribute to percentiles while the portfolio simulator still trades
-    only qualified longs.
+    needs the legacy full-backtest candidate-ranking universe; shorts can then
+    contribute to those historical percentiles while the portfolio simulator
+    still trades only qualified longs. ``include_universe_rank_observations``
+    additionally marks exactly one row per ticker/session for a live-like rank
+    across tickers rather than across directional setup candidates. If no setup
+    exists on that session, a non-tradeable rank-only row is emitted.
     """
     date_ords, opens, highs, lows, closes, volumes = columns
     bars = [
@@ -1079,10 +1083,19 @@ def _replay_candidates_for_period(
         )
         vol_6m = _realized_vol_6m(window_closes, len(window) - 1)
         iso = bars[i].date.isocalendar()
-        for setup in _window_setups(window, config, activation):
-            if not include_short_candidates and setup["direction"] != "long":
-                continue
-            candidates.append({
+        raw_momentum = (
+            window_closes[-22] / window_closes[-253] - 1.0
+            if len(window_closes) >= 253 and window_closes[-253] > 0
+            else None
+        )
+        setups = [
+            setup
+            for setup in _window_setups(window, config, activation)
+            if include_short_candidates or setup["direction"] == "long"
+        ]
+        observation_emitted = False
+        for setup in setups:
+            candidate = {
                 "symbol": symbol,
                 "date": bars[i].date.isoformat(),
                 "iso_week": (iso[0], iso[1]),
@@ -1101,6 +1114,24 @@ def _replay_candidates_for_period(
                 "meets_core": setup["meets_core"],
                 "action": setup["action"],
                 "risk_level": setup["risk_level"],
+            }
+            if include_universe_rank_observations and not observation_emitted:
+                candidate["_universe_rank_observation"] = True
+                observation_emitted = True
+            candidates.append(candidate)
+        if include_universe_rank_observations and not observation_emitted:
+            candidates.append({
+                "symbol": symbol,
+                "date": bars[i].date.isoformat(),
+                "iso_week": (iso[0], iso[1]),
+                "ranking_period": _ranking_period(bars[i].date, cadence),
+                "direction": "rank_only",
+                "momentum": raw_momentum,
+                "residual_momentum": residual_momentum,
+                "vol_6m": vol_6m,
+                "meets_core": False,
+                "_universe_rank_observation": True,
+                "_rank_only": True,
             })
     return candidates