That idea is definitely in the air, and for good reason. Anything (reasonable) we can do to prevent "overfitting" is probably worth considering. But here's one thing that took me a long time to realize: instead of showing content in forms X and Y (chosen randomly), we can just have two different cards, one each for X and Y. That is, in a good system, it should be cheap enough to add cards that we can get some of the benefit of random variation just by having lots of somewhat different cards.
This is better for stuff like "someone famous is known by several names and we want to learn their year of birth" than for stuff like "I want to learn how to do simple integrals of a given form," but it's worth keeping in mind.
After 630,000 or so reviews with spaced repetition ("SR") and hundreds of hours of making and thinking about SR memory tools, here are some strong opinions held loosely:
1. Algorithm quality is overrated.
When people praise SR, it's (approximately) because SR minimizes the time and effort cost of learning a given fact indefinitely.
Now, what is that cost? For concreteness, let's say you want to be disposed to remember a given fact with probability at least p for the next five years. Doing that will require:
It depends a lot on the structure of what I'm trying to learn, but very often 2 to 5.