I once found a statistician's write-up on Corsi (I just Googled, couldn't find it again to post it here), and while he was open-minded to Corsi he had his doubts, so he crunched some data in the proper fashion, complete with p-values.
Before doing so, however, he tried to look at the genesis and early days of Corsi to get a better understanding of the original intent of the metric, and like me, he could not logically understand why "5v5 Corsi" became a thing with respect to Team Corsi, because like me he also noted that doesn't really make sense. Net/net, he basically couldn't figure out how it evolved that way and basically decided it was just because, (paraphrasing) "conventionally that's just what most people seemed to initially do".
Anyway, his conclusion after his analysis, was that he determined that as a predictive metric, this "advanced statistic" was no better (and actually worse) than.....uhhhh.... "really simplistic statistics" for lack of a better term, such as just looking at goal differential.
In the 3 seasons he conducted the study, uber simplistic GF - GA was better predictive of NHL team's future success than 5v5 Corsi.
And I wasn't surprised in the least bit.