Reply to post:

I can certainly see the application of a more generalised system, but at the moment it goes to the heart of the problems with black box magic predictors - it needs to be fast enough to provide a useful part of the design - test loop.