In the current reference implementation, when unpaired surrogate is read, it is returned as is (a character created by applying integer->char on the unpaired surrogate value).

RFC7159 permits such JSON but the behavior of software reading such JSON is undefined (8.2).

We can leave each srfi-180 implementation how to handle such case, but it's nice that it is explicitly mentioned in srfi document, e.g. after the list of mapping between JSON types and Scheme objects, say "If the JSON string contains unpaired surrogate, the behavior is implementation dependent. It may include such surrogate as a character with the codepoint of unpaired surrogate, or may raise a JSON error."