How does NumPy internally implement masked array operations?

Masked arrays use a companion boolean mask array to track invalid or ignored elements.

NumPy masked arrays wrap a standard ndarray and attach a boolean mask of the same shape. Operations propagate masks alongside data, ensuring that invalid or missing values are excluded from computations. Unlike NaN-based masking, this approach allows masking for non-floating types. Internally, ufuncs are overridden to respect mask propagation rules.