Splitting off the duplicated author object could help a bunch for this one. Not sure how much though.

I just tried it, shaves it down to 77MB. removing redundant URL fields brings it down to 60MB. a lot of the rest will be all the file metadata

Related, I only mentioned that because I tried checking tsv size and it was even smaller (but couldn’t handle the nested author object).

yeah, the files would probably also screw with the TSV. That's why I originally made blocks-min.json so that 90% of the interesting info was available in a compact size.