Bug: EventFeatures minmax-standardizes constant columns — produces all-NaN in 3 of 48 input dims

[bradsmithmba]

Summary

EventFeatures.calculate sets all three output columns (days_to_earnings, days_to_fomc, days_to_opex) to single scalar constants across the entire date range. The final standardization loop then calls minmax on each constant column: (x - min) / (max - min) = 0 / 0 = NaN. All three event feature columns become all-NaN, infecting 3 of the 48 dimensions fed to RegimeDetector.

Evidence

src/features/regime_features.py lines 574-580: constants 30.0, 21.0, 10.0 assigned to all rows.
src/features/regime_features.py lines 583-584:

for col in features.columns:
    features[col] = self.standardize(features[col], method='minmax')

minmax on constant series → division by zero → NaN for every row.

Fix

Skip standardization for constant columns (check series.nunique() <= 1), or return the raw values without standardizing. The real fix is to implement actual event detection (earnings calendar, FOMC dates) so the columns are not constant.