How do you balance that [realism versus obscuring diegetic UIs]? Is that just a gut feeling? Is there a metric internally, or is it just on a case-by-case basis?

Trial and error. I think a lot of the time, the elevator ones we used in the video are a great example, because they're a list of very large, friendly buttons, I would say. They are perfect to press, right? Because that's all there is. And you go to something like a shopping kiosk and there's a lot going on.