Yeah - my understanding (and I could be completely wrong) is that Opta uses the entire league (or some very large, broad sample) to establish baseline/good xG and xA and then apply it… that way you get meaningful comparisons and interesting trends over long period… games are useless for virtually any football stat b/c of the nature of the game…
I mean we look at things that have averages of 1.5 in a game that is 20 field players where one player is involved “directly” in play that affects stats like 3-5% maybe.
Anyway, I tend to take the view that these things can be useful in the right context… I can see why someone might find them virtually useless though.
Since I have a deep background in stats (I suspect you do too) but have mostly applied in business setting (where it is massively noisy), i am more willing to take a semi-scientific approach.