But in fact translations from some arbitrary charset into normalized Unicode are not invertible.
which is why you can't expect all Unicode to be normalized, because it may have been translated from another charset.