I agree, OTP is optimal, it's just not unique, even by a very broad definition of what an OTP is. Sending a Turing machine to the other party through a secondary channel, for example, would produce an equivalently (not more) secure mechanism. Or sending a normal real number as the key, and the ciphertext is an index into the base26 expansion.

Technically, either of these could be slightly more secure, in that they need not leak the length of the plaintext. For transmitting a single "yes" vs. "no" answer, that could be a fatal attack vector for a naive OTP.