All flux calibrations in the pipeline are relative. Raw measurements of fluxes aren’t useful by themselves for accurate photometry over a field because the PSF varies over the field. Aperture corrections are an attempt to correct for these effects. We choose an aperture that is large enough to contain all (most) of the PSF variation as the reference. This reference aperture is too large for good signal-to-noise measurements of faint objects (or doesn’t have the right shape, weighting, etc.), so what we do is apply an “aperture correction”. This is a model of the magnitude difference between the aperture of interest and the reference aperture for point sources as a function of position.
For LSST, we calculate and apply such aperture corrections for all flux measurement algorithms (even the model-based ones: e.g., PSF magnitudes use an aperture with a weighting function equal to the PSF), and the reference aperture is (by default) a 12 pixel radius aperture. The only exceptions I’m aware of are the circular apertures that are intended to be used for profile studies (and therefore applying aperture corrections would not be helpful). For this reason, the difference of magnitudes from any two flux measurement algorithms should be a distribution centered around zero for point sources. This is what is used for the
base_ClassificationExtendedness algorithm which does simple star/galaxy separation: it thresholds on
psfMag - cModelMag.
Some other surveys attempt to do an aperture correction to infinity. I don’t believe this is necessary, because as long as the reference aperture is chosen so that it contains all the PSF variation as a function of position, the correction from the reference aperture to infinity is constant. So maybe you don’t count every single photon from your sources, but you count a consistent fraction of them, which is what matters for most of the applications our users care about.
So calibrated magnitudes reported by the pipeline are relative to what we would measure for the magnitude of a point source in the reference aperture, scaled to what the reference catalog would report for that source: it’s all relative.