That sounds like a pretty solid method for making progress. First outline something with spherical cows and refine the theory as data that doesn't quite fit keeps accumulating.