Guard against division by zero in apply_per_asset_coefficients

synth/validator/moving_average.py

@@ -121,7 +121,10 @@ def apply_per_asset_coefficients(
         df.loc[df["asset"] == asset, "prompt_score_v3"] *= coef
         sum_coefficients += coef * len(df.loc[df["asset"] == asset])
 
-    df["prompt_score_v3"] /= sum_coefficients
+    if sum_coefficients == 0:
+        df["prompt_score_v3"] = 0.0
+    else:
+        df["prompt_score_v3"] /= sum_coefficients
 
     return df["prompt_score_v3"]
 

tests/test_asset_coefficients.py

@@ -0,0 +1,43 @@
+import pandas as pd
+
+from synth.validator.moving_average import apply_per_asset_coefficients
+
+
+def test_zero_coefficients_no_nan():
+    """When all assets have coefficient 0 (unknown assets), scores should be 0, not NaN."""
+    df = pd.DataFrame(
+        {
+            "asset": ["UNKNOWN_A", "UNKNOWN_B", "UNKNOWN_A"],
+            "prompt_score_v3": [0.5, 0.3, 0.7],
+        }
+    )
+    result = apply_per_asset_coefficients(df)
+    assert not result.isna().any(), f"Got NaN: {result.tolist()}"
+    assert (result == 0.0).all()
+
+
+def test_normal_coefficients_unchanged():
+    """Normal assets with non-zero coefficients still work."""
+    df = pd.DataFrame(
+        {
+            "asset": ["BTC", "ETH"],
+            "prompt_score_v3": [0.5, 0.3],
+        }
+    )
+    result = apply_per_asset_coefficients(df)
+    assert not result.isna().any()
+    assert result.sum() > 0
+
+
+def test_mixed_zero_and_nonzero_coefficients():
+    """Mix of zero-coef (unknown) and nonzero-coef assets."""
+    df = pd.DataFrame(
+        {
+            "asset": ["BTC", "UNKNOWN_X"],
+            "prompt_score_v3": [0.5, 0.3],
+        }
+    )
+    result = apply_per_asset_coefficients(df)
+    assert not result.isna().any()
+    # BTC has non-zero coef so total should be > 0
+    assert result.sum() > 0