You're hitting on some extremely key insights IMO. Insights grounded in abstract fundamental principles useful all the way from the "hardest" of sciences (physics and even maths, arguably, as a formal science) to the much "softer" (human psychology etc. - where it's far more difficult to be directly quantitative for a whole host of reasons).
First, when it comes to engineering, the absolute best test is running the actual system. The "acid test" of a rocket is the launch of that rocket. And, even for all of our "computer-aided engineering" progress over the decades, a wind tunnel is still often a key step and can provide "better" and more reliable info regarding some characteristics of, say, rocket body shape performance, than Pro/ENGINEER or etc. can*. So, the best test when it comes to ADA-related issues is to engineer yourself, to the degree possible, into the position of someone with a "disability". The best work in these areas has involved people tying their limbs down etc. - because, even if you consciously work to not use one arm, say, you'll still involuntarily use it in many ways. For example, it'll naturally come up slightly to help regain balance in some situations.
Second, and this is actually, I'd argue, simply a more complete perspective partly covered just above - understanding critically depends on the degree to which one can be in some "position". Often enough, our minds can be adequate. In particular, we can "understand" abstractions that can't necessarily even have obvious "instantiations" - e.g., mathematical abstractions come to mind. There may be "exemplars", but, you can't literally "show" me "the number 3"**. That written, there are many cases where we CAN 'experience' some form of direct 'instantiation' and, for reasons both experiential and even statistical / logical, such an instantiation is pretty well guaranteed to do a better job of producing understanding in our overgrown monkey minds than any amount of sitting around and daydreaming can.
So, really, when it comes to the "hard(er)" (e.g., engineering) and "soft(er)" (e.g., psychology - including empathy, say), there's no substitute for "the actual launch" (to circle back to the language of the rocket example, above).
* Though, there may be cases / "regimes" that are too difficult [at least practically] to test, even in a wind tunnel, and where, especially these days, CFD modeling can at least give some info and potentially be even entirely adequate)
** Can't wait to see the replies that just say "3", kek