Coming from a statistics background, this makes a lot of sense to me. Before any model, we always spend most of the time just cleaning and validating data. the actual modeling is usually the smaller part.
I think that's why the medallion architecture caught my attention when I started learning about DE. It basically forces you to be honest about data quality at each step before moving forward.
One thing I'm still trying to understand as someone new to this: how much of the silver layer work ends up being automated vs decisions that someone has to make manually? That part seems like where things get tricky.