Bayesian methods force you to actually think about how your data and model parameters are distributed, and explicitly specify a model.

Yep. I know some very senior scientists who get a lot of mileage out of finding places where "model-free" methods are implicitly assuming a bloody stupid model, and then attacking them.