How much is the difference in colourbar/whisker length scale (0.40 vs. 0.27) influencing the visual appearance here?

I don't think the color scale effects the size of the whiskers. So I don't think it makes a difference in this case.

I have checked the correlation of the ellipticity residuals on the individual visits for tract 8766 and 8767 using both pipelines. The results agree well and LSST is actually doing slightly better.

This means the problem is coming from the coaddition process. Here is the same plot, but for the same coadd tract:

Jim Bosch suggested there may be a problem with the validPolygon being set in the individual visits. To test this, I reran using only the ccds from the central region of the camera. This did not make any improvement.

I found that the ccd positions were not being updated in python once the meas_mosiac fit had finished due to the pybind11 wrapping. This meant that the incorrect solutions were written to disk. This has been fixed on DM-10688. I have confirmed that this fixes the problem for a limited set of visits.

Waiting for Tim Morton [X] to do an RC dataset run with the DM-10688 fix in place to verify that this fixed it.

(Tim, is that ticketed? Is it happening?)

Here is the result of tract 8766 from the RC dataset with the pybind11 fix compared to the S16A HSC data:

This looks good. I am closing this.