Imagine a researcher whose paper has been retracted. The publisher issues a retraction notice, and in the citation database that notice shows up in the researcher's author profile – as a publication authored by the researcher. If the researcher disputes the retraction, that is a bitter detail: the paper is retracted, and the notice announcing it carries their name as if they had written it.
How can a database that follows its own rules end up doing this? Because sometimes a solution faithfully follows an operational rule while undermining the very policy that rule was meant to implement.
The trickiest content problems are often those that reach senior management directly, or prompt complaints from official bodies. Responding takes careful coordination between several departments. Fortunately, our databases are underpinned by clear, defensible policies, which allow us to explain our decisions without appearing arbitrary. But policies only protect us if our own solutions follow them.
Whose statement is it?
It must always be fully clear whether Elsevier made a statement or someone else did. In a citation database this is particularly important, because it includes articles from many publishers who are of course not bound by our policies. And since these publishers are located around the world, their views on sensitive topics may differ fundamentally from ours.
Let's look at this from the perspective of author profiles. The word "profile" may give the impression that authors own these profiles, in the way they own their LinkedIn profiles. But they do not: an author profile is not a personal page owned or controlled by the researcher. It is our representation of their publication record. If someone authored a paper, that paper belongs in the record, even if the author would rather forget about it. Researchers can of course provide evidence to correct errors, but that is not the same as having the authority to rewrite their publication history.
This clear policy is undermined by any wording in our solutions that refers to "your profile" – correct your profile, check your profile. When things are right this comes across as friendly, but in a dispute it doesn't help.
What's on the tin
Who determines that someone authored a paper? Our policy is that publishers determine authorship, and we follow their lead. In practical terms we say that we "capture what's on the tin", or that we "faithfully present the publisher's version". This works remarkably well, but the policy and the operational rule are not equivalent. The publisher's version is evidence of the publisher's decisions, not necessarily a flawless representation of them. And sometimes the publisher itself provides the evidence that its metadata is wrong.
As so often, we have confused the policy with the practical operational rule – a rule that works in almost all cases, but not all.
An example is an erratum informing us that person A was not one of the authors of paper P. By longstanding convention, paper P itself often isn't corrected: it still shows A as an author. Yet removing A from the authorship in the database is not a violation of our policy. After all, we don't take anyone's word for it; we carry out the correction that the publisher communicated through the erratum.
The original publication remains unchanged, as part of the historical record. But our knowledge graph should reflect the corrected authorship. We are not rewriting the publication; we are updating what we know about it, based on authoritative evidence from the publisher.
Our prime objective is not to follow the publisher's metadata dogmatically, but to create an accurate knowledge graph of scientific publications based on what the publishers tell us.
Who wrote the notice?
This is especially tricky if what the publisher tells us is confusing. Many years ago the layout of errata was standardized, following best practices to make it very clear what the erratum is about. We still see errata that are simply called "Erratum" – not very informative. At the same time, the convention was introduced to list the authors of the original article as the authors of the erratum. This may seem convenient, but it is fundamentally wrong if the text wasn't written or issued by those authors at all.
For instance, many retraction notices are issued by the Editor and/or the Publisher. Nevertheless, many publishers, Elsevier included, list the original authors as the authors of the notice. PLOS ONE, on the other hand, is upfront about this and uses "The PLOS Editors" as the authors.
Especially in cases where the authors dispute the retraction decision, this is an ongoing source of customer complaints. The authors feel doubly punished: their article is retracted, but they also have to walk the streets with a notice telling the world about it, issued by the publisher, and because they are listed as its authors, the notice appears in their author profile.
Should we remove the retraction notice from their author profiles? I think so. Not because the notice should disappear from the database – quite the opposite, it is an essential part of the scientific record – but because attributing it to people who did not issue it is wrong.
Our policy is not to decide independently who authored a publication. It is to represent authorship accurately on the basis of authoritative evidence from the publisher. When the publisher's own notice makes clear that it was issued by the editor or publisher, but the metadata attributes it to the original authors, blindly copying that metadata is not being faithful to the publisher.
Of course, this doesn't mean that we should simply overwrite the original metadata and forget about it. The metadata supplied by the publisher remains part of our evidence. Our knowledge graph should make a separate, provenance-backed assertion about who issued the notice. Where the responsible editor or publisher is explicitly identified, we should attribute the notice accordingly. Where it isn't, we should not invent an author merely to fill the field.
That the publisher structured the notice in a confusing, or sometimes mistaken way, is irrelevant to our obligation to represent the scientific record accurately.
It is also too easy to tell the customer to direct their complaint to the publisher. That makes sense if the publication record gives us no indication that the customer's claim is correct. But it doesn't make sense if the publication record itself provides all the information we need. Asking the customer to persuade the publisher to correct its metadata, when the publisher's own notice already makes the situation clear, is hardly a satisfactory response.
Correct the representation, change the practice
So we should make these corrections. The retraction notice should remain in the database, properly linked to the original article, but it should not appear as a publication authored by people who did not create or issue it. This is a correction to our representation of the scientific record, not a change to the historical publication itself. As long as we are bound to our old formats, where we can't always make the distinction clearly, the accuracy of the knowledge graph should take precedence.
However, the publishing practice itself should change as well. Attributing works to people or organizations that did not create or issue them is wrong, even if it follows a longstanding convention. Being precise about provenance is not an administrative nicety; it is fundamental to the integrity of the scientific record. And that applies to the entire corpus, not just the articles arriving today.
Faithfully representing the publisher's decisions is not the same as faithfully copying the publisher's metadata.
Related posts
- 7 October 2026Bloodlines What tagging given names tells us about company history (and about data modeling)
- 2 August 2026Are we helping book authors and editors? Filling coverage gaps doesn't make a good data product if the underlying model is built for journals
- 2 August 2026Is an editorial an editorial? Model the content, not the words – the central lesson of content modelling, applied to document types