I would argue that the puzzle-solving paradigm could be useful if you go in understanding that the nature of the "puzzle" is to solve for every possible edge ca

I would argue that the puzzle-solving paradigm could be useful if you go in understanding that the nature of the "puzzle" is to solve for every possible edge case and understand why those edge cases materialized in the first place. There are times where working backwards into code that fits a model is a reasonable way to do it. But this does, at least, presuppose that you understand your code to be deterministic, not some voodoo you copy/pasta.