Formulate and test the model
What can you choose? What limits those choices? What are you trying to improve?
We turn those questions into a mathematical formulation, examine the available evidence, and test against a simple baseline. We separate errors in the data and code from weaknesses in the model itself. We shave with the Occam’s razor and practice the scientific method in our work.
Added complexity has to improve a result that matters. Sometimes the right answer is a better model. Sometimes it is better data, a simpler calculation, or a clear statement that the evidence is insufficient. Not every phenomenon can be modeled quantitatively.