This is more product feedback than a question, you’ve also probably already thought of it, but I thought I’d pitch it anyway.
It’s great that the question numbers change, but in story problems, the context of the numbers doesn’t. The result is that practicing the math is quite efficient but practicing diagnosing the question doesn’t work so well.
After just a week of studying I find my brain keying in on the scenario rather than the structure of the problem to pick the solving strategy. Rather than looking for variables in the answer choices, I find myself looking for arbitrage entrepreneurs and gizmos. (algebraic expressions). What if sometimes the question was about a Realestate mogul and properties?
Even better would be using variables for the verbs as well, In addition to thinking about the distance between 4 sisters living in a rectangular state, can we think about the distance between magnets on a fridge, muffins on a tray, or a quarterback and receivers on a field? Of course, more templates are great, but even just more variability within the existing templates could be very useful.
Generating lists of related scenario words seems like the kind of task you could hire undergrads or M turk to do, coding them in would of course require a bit more effort, but considering the importance of question diagnosis in the achievable method, it seems worthwhile.