If your converter purports to produce any one of the Unicode encoding forms, then it cannot conformantly produce malformed Unicode as a result.
If, of course, it does not purport to do that, it can do anything it wants to.
Then, as long as the documentation states that on error condition the output buffer is not guaranteed to contain UTF-whatever, then the converter would be conformant.
A./