My thoughts are that we treat the supplied message like an unevaluated string.
> Each universal-character-name and each simple-escape-sequence in an unevaluated-string is replaced by the
> member of the translation character set it denotes. An unevaluated-string that contains a numeric-escape-
> sequence or a conditional-escape-sequence is ill-formed.
My suggestion would be to add “and is processed as an unevaluated-string.” To the end of p2 of the library wording.
That should handle making numeric and conditional escape sequences ill-formed, and map the universal names and simple escape sequences to their corresponding sequence of UTF-8 encoded code points.
AlisdairM