HSC is facing the same issue, with two i-band filters. For our most recent production run, we decided to simply throw everything in together, combining all data taken with an i-band filter regardless of which generation it was. Certainly this is not optimal, but the result isn’t too bad — we haven’t identified any problems so far. Actually, we were kinda already combining data taken with subtly different i-band filters because the filter curve varies over the focal plane, so combining images at the center of the field and at the edge of the field has a similar effect.
I believe @rhl and @jbosch have plans to track not just the filter of each observation but the transmission curve as a function of position as well. I’m not sure what can be done to correct images to a common bandpass given that it’s not just a scaling but a color-dependent scaling that needs to be applied, but perhaps techniques being developed for image subtraction in the presence of DCR could find utilisation here. However, I think that coadds sweep enough details under the rug (PSF, clipping) that the filter issue isn’t such a big problem. Coadds are principally intended for detection and rough estimation of quantities for faint objects. If precision measurements are required, you’re better off looking at the individual exposures (e.g., multifit), which can account for the different bandpasses given a model SED.