major segmentation fail, not clear why, but this one needs to be trapped about cosmos

[iross]

We have a check in place to try and recognize cases where the page is chopped up too much (e.g. tables where each cell is a separate object). If a page has more than 19 objects on the page, it repeats the proposal process while requiring a slightly larger whitespace gap to define objects.

Likely because of the short paragraphs and a couple of equations, the above page has 21 objects after the first pass, triggering the re-proposal. It does this a few times, until the whitespace gap to define a break is so large that no divisions between objects are found on the page, and the whole thing becomes one object.

Boosting the number of allowed objects on the page seems to be the best path forward.

Interesting note for posterity: gap division sizes of 15, 20, 25, and 30 all create 21 objects on the page. 35 makes it 1 object. A gap of 31 makes it settle on 12 objects.

[cambro]

Another one> This is a preprint, clearly, so this illustrates two issues that need to be addressed. One fetching one in getting into COSMOS.
http://cosmos3.chtc.wisc.edu:8081/search?id=3374581
https://onlinelibrary.wiley.com/doi/abs/10.1002/cmdc.202000223