The thing that I'm impressed by is how rapidly my ability drops off as more angles are added to the set; at 5, 15, 25, 35 I can perform with essentially perfect accuracy; the only non-trivial decision is between 25 and 35 degrees. Angles of 0, 45, and 90 are trivial to recognize absolutely as such, so I never include them. With the set 5,10,15,20,25,30,35,40 my accuracy drops considerably, and almost every question I'm thinking hard about whether it's one thing or another 5 degrees different.
This is in no way a careful experimental setup, but I'm not trying to run a careful experiment, so whatever. One thing I noticed is differential effects from one query to the next seem to help a lot; obviously if the angle doesn't change from one to the next, I can easily tell that, and I feel myself thinking "hmm... it only moved up about that much from the last one, so I guess it's probably five degrees greater".
The peculiar thing about playing this compared to the sort of kana flashcard games I did way back when I was trying to learn japanese a few years ago is the strength of interference effects seems to be much greater. Fr the most part adding new kana to the set I was testing myself on would only produce bad performance on the new ones. There was only minor interference with the existing well-memorized ones, but here I do worse on the "old" angles just as much as the "new". There's something very Saussurean going on here. I'm not learning that "ok, this is what 25 degrees looks like" once and for all. I'm learning how to perform a four-way distinction, a parcelling-up of a crowded continuous space, which only barely helps in learning a larger eight-way distinction.