Currently PhotoCalib.calibrateImage() computes the variance plane as:

fluxVar = instFlux.variance * scaleVar +  instFlux.variance * scale**2 + instFlux.image**2 * scaleVar

where scale and scaleVar are my own names for the scale factor that the photocalib applies and the measured uncertainty on that photocalib.

This equation assumes that the scale factor variance and the image variance are independent, and this isn't obviously true.

Whether this is the correct variance equation is irrelevant, however. It is a change to how the variance plane in the zeropoint-normalized warps is currently computed, and I suspect it is the primary source of the photometry differences we're seeing in coadds produced with jointcal. While we are testing the effects of using jointcal vs. meas_mosaic's calibration on coadd (DM-16596), it is important that we are able to test these two changes (jointcal's calibrations and the variance plane equation change) separately.

Implementation: This could be done with an extra argument like photoCalib.calibrateImage(Image, addCalibVar=False or with a state switch like Calib's Calib.setThrowOnNegativeFlux(False).