Yes that's exactly right. My library isn't doing anything super smart (just statistics 101) but if you have to do some awkward heuristic in your code (to "eyeba

Yes that's exactly right. My library isn't doing anything super smart (just statistics 101) but if you have to do some awkward heuristic in your code (to "eyeball" some approximate parameters) it's better to just supply some input/output examples as part of your code (which GoodGuesser autogenerates) then just have a computer approximate the ideal parameters using multiple linear regressions. That will be a lot easier to maintain than some sort of hacky heuristic. Plus, once you've refactored your logic as input/output pairs, It'll be easier in the future to do fancier stuff with your code if you want (deep neural nets, etc)