In a pair-coding session with Lang, he got the Tractor to fit galaxies to multiple overlapping images simultaneously. This problem is a hard one just because the (high-resolution) galaxy properties are fixed in celestial coordinates, while the point-spread function is different for every image and defined in image-pixel coordinates.
In another session with Holmes, we got ubercalibration completely working for the fake Euclid data Holmes has generated in the last few days. This is not a trivial feat, as it involves a thousands-of-parameters non-linear (bi-linear) least-squares fit. Our goal for tomorrow is to compare different deep-field observing strategies in terms of the quality of the inferred flat-field and photometric measurements. If we can get that done by tomorrow, we will be on our way to influencing survey strategy for Euclid and any number of other upcoming imaging projects.