I have to say that I really like this idea of mixing the tests and making sure people are tested in different ways as part of the process of making sure they know the words.

I've found the generative reviews much more challenging than simply identifying the definition at the upper levels, especially when you have to ask yourself -- "is this neutral tone or not"? How exactly to work this into the current SRS approach is a good question -- we want to strike a balance between being able to intuit what people know and avoid questioning EVERYONE on all of these words multiple times. That gets irritating pretty fast, especially for upper-level students who find easier words in their lists for various reasons.