diff --git a/src/portfolio.py b/src/portfolio.py
index c79c1dd..ba4554a 100644
--- a/src/portfolio.py
+++ b/src/portfolio.py
@@ -2,6 +2,7 @@
 
 from typing import Iterable
 
+import numpy as np
 import pandas as pd
 
 
@@ -16,11 +17,16 @@ def select_portfolio(
     if eligible.empty:
         return eligible
 
+    weighting = weighting.lower()
+    if weighting in {"inverse_vol", "inverse-vol", "inv_vol"} and "vol20" in eligible.columns:
+        # Missing risk estimates cannot produce a trusted inverse-vol weight.
+        # Exclude those candidates before top-N selection so one bad row does
+        # not turn every selected target weight into NaN.
+        eligible = eligible.loc[np.isfinite(eligible["vol20"])]
     selected = eligible.sort_values(score_column, ascending=False).head(top_n).copy()
     if selected.empty:
         return selected
 
-    weighting = weighting.lower()
     if weighting in {"inverse_vol", "inverse-vol", "inv_vol"} and "vol20" in selected.columns:
         inverse_vol = 1.0 / selected["vol20"].clip(lower=0.05)
         selected["target_weight"] = inverse_vol / inverse_vol.sum()
@@ -43,4 +49,3 @@ def calculate_turnover(previous_weights: pd.Series, next_weights: pd.Series) ->
     previous_weights = previous_weights.reindex(next_weights.index).fillna(0.0)
     next_weights = next_weights.reindex(previous_weights.index).fillna(0.0)
     return float(0.5 * (next_weights - previous_weights).abs().sum())
-
diff --git a/tests/test_backtest_accounting.py b/tests/test_backtest_accounting.py
index a1b8cb2..a2c2791 100644
--- a/tests/test_backtest_accounting.py
+++ b/tests/test_backtest_accounting.py
@@ -109,6 +109,18 @@ def test_valid_frozen_empty_selection_produces_cash(self) -> None:
         self.assertEqual(len(result.returns), len(self.dates))
         self.assertTrue(result.returns.eq(0).all())
 
+    def test_inverse_vol_skips_top_score_with_missing_volatility(self) -> None:
+        panel = self.panel(("A", "B"))
+        panel["vol20"] = 0.2
+        panel.loc[(slice(None), "A"), "final_score"] = 2.0
+        panel.loc[(slice(None), "A"), "vol20"] = np.nan
+        panel.loc[(slice(None), "B"), "final_score"] = 1.0
+        config = {"strategy": {**self.config["strategy"], "weighting": "inverse_vol"}}
+
+        result = run_single_backtest(panel, "final_score", config)
+
+        self.assertEqual(result.trades["symbol"].tolist(), ["B"])
+
     def test_nonfinite_eligible_scores_are_not_cash(self) -> None:
         for bad_score in (np.inf, -np.inf):
             with self.subTest(bad_score=bad_score):