I just realized that we only have an arxiv metadata field, which for some datasets is empty, because there isn't a paper on arxiv! This often the case for the older datasets, but also for a few new ones.
I therefore suggest to add the inspirehep.net reference and remove (?) the arxiv field; we should be able to get the arxiv field from the inspirehep entry. What do you think?