Even your concrete example may have hidden details which may lead to different results and as you said you're understand it and ready for surprises despite the fact that you had good experience with it.
But was that what was asked about?
Take the first question:
1. What is the ground wire referenced to / paired with?
Do you need reference to the full schematics, mechanical diagrams, and EMC plots, of some system, to judge the construction of a cable within that system, then?
It seems pretty answerable to me; but I guess there are mysteries that even lone cables contain. I bow to your wisdom, sir
...As long as we're here, then, does anyone else want to play?
2a. Is USB fully differential? If it depends, then also, when?
USB is not fully differential (at least USB1 and USB2). Data is sent differential, but it has some weird things (I think they are called "T-states") that are used to signal special conditions on the USB bus. In this regard it somewhat resembles the start and stop conditions on an I2C bus.
For the rest, I have always been confused of how to handle GND versus shield in cables.
Dingdingding! USB
is not a differential standard. While I don't know what the actual input structures look like (probably with good reason (that's IP)), it's plausible that they're just a pair of normal-mode CMOS input pins (for HID/low/full speed modes) or the low-threshold equivalent (high speed). It's also plausible that they use a differential receiver (perhaps the same one in all cases, i.e. ~mV precision and high speed capable), but also have normal or common mode receiver(s) to detect the special symbol.
The exact term I believe is
SE0, both lines low. The inverse is not used, and
J /
K denote the differential +/-1 (-/+?) states, I think it was.
A key insight: because of this, it is impossible to do more than a token amount of CM filtering on USBD_P/M; more than that, and you spoil the fidelity of the SE0 symbol, and line drops ensue. You (almost?) never see CMCs on Full Speed lines, and only a tiny amount on High Speed lines (usually ~100Ω at 100MHz -- nothing more than a ferrite bead).
Since Full Speed (and below) has the inputs open-circuit in receive mode, there's no impedance to filter against, anyway, just the tiny receiver input capacitance; and if you add your own onboard capacitance, you can only do it by loading each line to GND, reducing DM bandwidth as well (though not as quickly; Full Speed can tolerate some 10s of pF, perhaps at expense to maximum line length). But filtering below 30MHz or so disturbs or removes the SE0, and even then, the ISI is big, likely leading to line errors and dropped packets or links.
Remaining questions:
1. What is the ground wire referenced to / paired with?
(A: usually VCC; most USB cables I've taken apart, are double twisted pair, with foil on the data pair.)
2a. Is USB fully differential? If it depends, then also, when?
2a(i). If so/when, what is USB's common-mode range? How does it compare with commercial immunity levels?
2b. Is the power fully differential? (An odd way to put it, but more to the point: would there ever be a situation where we would worry about common-mode currents on the power pair? When, generally, are we ever concerned about that on a power line?)
(A: you've already implied that GND could be used as a signal reference, so the answer is no!)
3. If the answer to at least one of (2) is no, what solution would you suggest for it? (implied: a solid shield most likely, but on what justification; what possible alternatives?)
Concrete example:
https://www.eevblog.com/forum/beginners/on-devices-side-should-i-connect-usb-cable-shielding-to-the-black-wire/msg5510779/#msg5510779Diagram:
https://www.eevblog.com/forum/beginners/on-devices-side-should-i-connect-usb-cable-shielding-to-the-black-wire/?action=dlattach;attach=2232685;imageChallenge: what is the EMC situation for the USB cable, under variation of how we connect the shield, or connector in general, within the EUT?
Tim