The STIS CCD response is reasonably flat, and appears to be extremely stable. Ground testing showed short-term pixel-to-pixel stability at better than 0.1% over several wavelengths spanning the CCD range. In flight, broad-band flat-field exposures spanning a month constrain any changes in the pixel-to-pixel response to less than 0.2% rms. The most prominent features in broadband imaging flats , and in spectroscopic flats shortward of ~ 750 nm, are "dust motes:" shadows of specks of dust on the CCD faceplate. The f ratio of the calibration lamp beam is slightly different from that of an external source, so it is possible that sky flats will remove these features better.
These dust motes are flagged as 1024 (small blemish) in the DQ extensions of recent pixel-to-pixel flat field files (_pfl.fits). This data quality flag gets propagated to the science file during calibration, if FLATCORR is performed, which makes it look as if all these regions are bad, even though the effect of the dust motes on flat-fielded data is typically less than 2%. One reason these regions are flagged is that the effect of the dust motes varies with wavelength, so the flat field can't correct really well at all wavelengths.