Possible bug
I was benchmarking the pregrading with precise positioning and seem to have stumbled upon a bug. I have two cases where the scan was 4 blank boxes but somehow there was a false positive. only thing is that when looking at the problem isolated in a jupyter notebook, it comes back as empty. Interestingly enough, when going through the pdf of scans, the scan that comes before the false positive when going top to bottom has that box filled. I think there might be a database commit going wrong somewhere when the current page being checked is empty while the next one has a checked box.