The problem is, there are three choices for doing large aircraft UPRT training outside of a pure classroom environment.
(1) A simulator, which provides very poor representation of the physical motion and may well have a flight mechanics model that is at best guessing at the real behaviour of an aeroplane in a severe upset.
(2) A real part 25 aeroplane, which is likely to have a 2.5g structural g-limit, and is eyewateringly expensive to use for pure training purposes.
(3) A light aeroplane, with good aerobatic structural limits, but intertia two orders of magnitude below that of a large widebody, and a cockpit designed for single pilot operations.
Clearly, none of them are a perfect training tool. But, it is reasonable approach to use a light aircraft *if*
(1) It has a cockpit that lends itself to flying as if in a 2-crew part 25
(2) It will respond appropriately to the actions that you would use in a part 25
(3) It has limits that give a reasonable guarantee you won't break it whilst doing so.
This is the approach most UPRT training providers are using. It shouldn't actually matter that, say, the T67M260 I did my training in could be recovered from an upset much more efficiently in a different manner. What matters is that it is representative *enough* of the jet environment.
What I suspect however is not happening is much rigorous work to actually prove that, because nobody amongst either airline managers or training providers really want a robust and defensible answer to that question.
G