Basically, at least to remove the instrumental and atmospheric signature, and present results in as nearly physical units as possible. For example, beginning with a set of CCD frames containing a possib ly distorted registration of a galaxy spectrum in counts per exposure, yield a corrected array of intensity in energy units per wavelength as a function of position in arcseconds, with foreground night-sky emission removed as well as defects of the CCD. Optionally, measuring a velocity and linewidth profile along the slit tracking statistical errors. Another example would be taking a set of CCD images, normalizing from the median of stars to remove atmospheric changes, and doing time-series photometry of stars within the series. As another poster said - any of the processes going from raw data to physically interpretable results.

This is what most of the time of many practicing astronomers gets taken up with, which the practically religious obsession in discussions of particular platforms or approaches.