Another tenet is that work is work. You attempt to break the work into chunks that are close enough to previous work efforts to use your model. If you can't do that, then you don't actually have any idea of what you're doing yet. Which is fine, but you certainly can't estimate in that state.