It works better if you “split the axis”; (imo) good code is easy to modify, bad code is hard to understand. Everything else is, as you say, trade offs, and so only evaluatable within a specific context.