FWIW I’ve been heavily using Claude Code for review as well. I love your quiz idea which I have not tried. re:SFL review, I found the biggest gains was in its ability to quickly make (counter) examples for diagrams, since I am pretty bad at making examples in general. So when I hit a conditional I don’t quite understand, I can just ask and it seems to nail it 100% of the time (maybe after long backtracking thought strings where it realizes the examples don’t hold, but it gets there!).
In general, asking LLMs verifiable things is the biggest gain, so things like counter-examples fit in really nicely in that paradigm.