It wouldn't really be much different if we had ISO8601.1 and ISO8601.2, would it? The standard can define two different types, and it makes it clear what it means to have a time without an offset specifier. It also defines formats for durations, intervals, and repeating intervals.

Is the distinction in representation too subtle?

We accept this subtlety elsewhere; we are used to 0 and "0" being different things, and expect them to behave differently under operators. Few would find the following surprising:

    0 + 0 == 0
    "0" + "0" == "00"

Then why would we expect these to behave the same?

   20200710T010000Z + `3 hours`
   20200710T010000 + `3 hours`

The first is a fixed timestamp in UTC, the result of the second depends on timezone.

Good point, which is why ISO 8601 does have a very different duration format (though the options for it such as calendar weeks make it a lot more obvious why it isn't just HH:MM:SS).

Maybe there should be a "float" offset marker. Another thing I was reminded of is that +0000 offset should be Z (UTC), but it is often application dependent if -0000 offset is also Z as some applications use -0000 for "user local time, regardless of user". Which is related to "floating", but yet another semantic difference.

I haven't run into that use of -0000, that's insane.

I suspect a side-effect of libraries lacking support for the full range of ISO8601 (for example, refusing to parse a value without an offset, forcing people to use such hacks). Wouldn't surprise me; iOS doesn't even have a consistent way to parse ISO8601 values with and without milliseconds.

[edit:

Apparently this is a RFC3339 thing, did not know that: https://tools.ietf.org/html/rfc3339#section-4.3

Negative zero offset is invalid in ISO8601, but valid and carries the meaning you describe in RFC3339. I misread what that meaning actually is from your comment - it is different to a "floating" or unqualified local time, it always represents a UTC time but with an unknown local time offset. ]