Nontrivial, yes, but for said crypto-supporting crowd I don't think it would be a stretch to presume that a lot of them are the kind that would know how.
Barring that, a community-curated repo of interfaces for common applications would also be great. You could just have a wrapper that automatically downloads the applicable config files on the fly.
The second rule is that if you think you're in the club, you probably aren't. The third row is that if you only think you're in the club to lessen your chances of being in the club, you're more likely than someone who doesn't even know about the club to be in the club.
Nitpicks: there are lots of grammatical errors, egregiously out-of-place usage of emojis and emoticons, and it could be argued that Gödel's incompleteness theorem is hardly "obscure."
Current Transformer models are looking pretty good at complex end-to-end tasks (at least, better than the shallow regression with hand-picked features that ETS probably uses). In a few years, complete end-to-end evaluation may not be so impossible, especially with so much data.
English is NOT an easy language to "just read". Compared to many other languages, it is very very hard to tell how a word should be pronounced from its spelling owing to the sheer irregularity. Just because you, as an adult who has spent decades reading English, find it easy doesn't make it easy for children.