Ha.  I wonder what the rationale is for this choice.

I would think the goal is to estimate the artifact in the data before pre-processing, and so all labs I've been in have only ever input the raw images before any preprocessing.