Purpose: document the preprocessing policy used by the analysis pipeline.

• NaN indicates structural non-applicability, not random missingness.
• Typical cases: conditional descriptors (for example WDM-specific fields outside WDM scenarios).

1. Compute mean/std per feature using non-NaN values.
2. Apply z-score on valid entries.
3. Replace remaining NaN with 0 in standardized space.

This is a methodological decision to keep scenarios comparable in a common geometry.