After the ISR step is completed, the pixel values and their variance are not modified any more.
This is probably incorrect; after ISR, a single-epoch image will have been flattened according to some SED, which may be:
- a flat or other simple analytic spectrum
- something that looks like a "typical" source for that band
- our best estimate of the spectrum of the sky
It is quite likely that we will un-flatten and re-flatten to other SEDs in later steps, because the first two are what we want to do photometry while the last is what we need to use when subtracting the background. Whether the SED we've flattened with coming out of ISR is the same as the one we end up with depends on the sequencing of different photometric calibration and background-estimation steps, and that's TBD (and ought to be below the level of detail of the DPDD).
As the above suggests, we'll also be subtracting the background from the images. While this (like any flattening) should be completely reversible, because of the interplay between flats and backgrounds, it's not as trivially reversible as it might seem.
In short, we should find a way to reword this section of the DPDD to not make unnecessary claims about when the pipelines will do what.