arXiv: Repository and Archive ID not always populated

[dstillman]

zotero/zotero#5680 (comment)

https://arxiv.org/abs/2512.20702

[alex-ter]

So this happens when there's a non-default DOI (listed as "Related DOI" on the page), which doesn't resolve properly. That particular paper is probably too new, so the DOI is not yet active, even though the paper was accepted for publication.

Currently the translator only has the logic for "non-default DOI exists" and "non-default DOI doesn't exist", and this is the third case - "exists, but doesn't [yet] work".

What do we want to do then? Execute the else branch in full, overwriting the non-default DOI, as if it didn't exist?

BTW, in general I find this overriding logic to be a step too far IMHO, we are saving the arXive piece after all, and it's a preprint by definition, so substituting that one with the fetched DOI seems like overreaching. The preprint may not be exactly the same as the published version, for example, so such substitution is IMHO wrong.

I have a similar case I want to implement for the IACR ePrint translator, they have a similar DOI field intended for the published version of the preprint, but I planned on doing something like "multiple" in that case, allowing the user to select which one to save: preprint or the published one (or both), as that's IMHO a much cleaner approach. In the cryptography community it's frequently the "full version on ePrint + short version published and referenced via the DOI field" case, that's why such substitution logic won't fly there at all.

Should I maybe implement something like this here as well, instead of this substitution?

[AbeJellinek]

Execute the else branch in full, overwriting the non-default DOI, as if it didn't exist?

That's what I'd think. If it doesn't resolve, it isn't a valid DOI (yet) and shouldn't be cited.

The preprint may not be exactly the same as the published version, for example, so such substitution is IMHO wrong.

That's true, but the preprint itself could also have multiple versions. Just a problem inherent to pre-publication items. We label the preprint PDF as such, so hopefully nobody will mistake it for the peer-reviewed print version. I don't think we've really gotten complaints about this behavior.

When we have Update Metadata merged (zotero/zotero#5120), we'll need translators like this one and IACR ePrint to return single items with DOIs so items can be updated to their print versions.

[alex-ter]

That's what I'd think. If it doesn't resolve, it isn't a valid DOI (yet) and shouldn't be cited.

Thanks, that make sense. Let me work on that.

We label the preprint PDF as such, so hopefully nobody will mistake it for the peer-reviewed print version.

Fair enough, though with this approach there's no way of citing the preprint version when that "see also"/"published version" DOI is added, which IMHO is an unnecessary limitation, don't you think?

Not sure how it works for the active arXiv users and what's typical approach in e.g., the physics community, but as mentioned, the ePrint ones are sometimes expanded versions, with additional proofs and such, having additional value over the published ones, so the argument of "preprint always leads to a publication, so becomes useless afterwards" doesn't work there, as those preprints could still be cited. All in all, it's of course not a blocker either way and is easily corrected manually, that's why I guess there's no active complaints about that.

On that Updated Metadata item, by quick glance it doesn't seem like that approach of making DOI-containing items a multiple with the preprint+published records would conflict with that, as at the end of the day there will be single records for each type, so Update Metadata would treat them separately.

As for the complaints, quick look at the forum turns up this, though in all fairness, there are some other ones, mentioning a direct opposite preference of having some sort of auto-merging :)

On a side note and back to arXiv translator specifically, when there's a valid external DOI, it's first detected as journalArticle and that type is later updated using whatever the DOI translator returns, leading to some funny cases like this one in one of the tests. Do you think a difference between detected and saved types is ok, or is this something we should fix too?