Technically speaking the clauses on either side of the "or" aren't mutually exclusive. You can have a "full, correct, fully compliant, reference implementation" that is also a closed-source implementation!
Well, unless the implication that Circle isn't "full, correct, [and] fully compliant", in which case I feel I should ask "with respect to what?" and "why do you need those requirements?"
Eh, bit of a mixed bag, I think, depending on the context in which the words are used. "Circle" can refer to the compiler/toolchain or the set of C++ extensions the compiler implements, whereas Safe C++ is either the proposal or the extensions the proposal describe. As a result, you can say that you can compile Safe C++ using Circle, and you can also describe Safe C++ as a subset of the Circle extensions. I wouldn't exactly describe the lines as well-defined, for what it's worth.
> There are presumably differences between them, and I do not know what those differences are, and I do not know if those differences were documented somewhere.
They're sort of documented indirectly, as far as I can tell. Compare the features in the Safe C++ proposal and the features described in the Circle readme [0]. That'll get you an approximation at least, albeit somewhat shaded by the old docs (understandable given the one-man show).
> I cannot find any occurrences of "reference implementation" in the Safe C++ draft.
The exact words "reference implementation" may not show up, but I think this bit qualifies (emphasis added):
> Everything in this proposal took about 18 months to design and implement in Circle.
[0]: https://github.com/seanbaxter/circle/blob/master/new-circle/...
Like, Python and Javascript both have many "implementations", and those are some of the most popular languages. Python does not have an ISO specification. But Javascript does have an Ecma standard, ECMAScript.
Rust is getting another implementation in the form of gccrs. And there is work on a specification for Rust https://rustfoundation.org/media/ferrous-systems-donates-fer... . Arguably not a standard, but still helpful.
