[skyechurch]

>Some legitimate references were also lost, meaning they were not present in the metadata.

It's possible that some of the inconsistency between metadata and text could just be due to incompetence - it's harder to find a profit motive for dropping legitimate citations. Why wouldn't this sort of metadata auto-generated from the text (aside from enabling fraud, of course)?

[feoren]

Which is harder to detect: replacing reference 17 with the one you're trying to pump, or adding reference 35 when the bibliography in the original paper clearly stops at 34?

[neilv]

> it's harder to find a profit motive for dropping legitimate citations

Competitiveness for citation points, especially with someone in or adjacent to your niche?

Also, the non-profit: pettiness.