In the past, I’ve understood the ‘conventional wisdom’ to be that one achieves better image subtractions by (re)calibrating the astrometry of the new/search image directly to to the astrometry of the reference/template image, as opposed to calibrating both images independently to some external data set like Gaia. I believe that e.g., SWarp from the Astromatic toolset is capable of performing this type of astrometric recalibration directly against the reference image.
Is this how difference imaging in LSST pipeline works? Or does the differencing in the LSST pipeline simply adopt whatever astrometric solutions to the external reference catalog were achieved earlier on during the calibrate task. I suppose that another variant of this might be that the kernel matching kernels are sufficiently flexible to effectively incorporate small astrometric tweaks/offsets (?). Thanks very much.