At training, a feature was named country. In production it's named region (same data, new name). Your skew check only compares the number of features. What happens, and what's the better check?