More than 10,000 research papers were retracted in 2023, a record reported by Nature. Most attention goes to the journal article. Yet many of those papers began as preprints, and the preprint version often stayed online without a machine-readable link to the retraction notice. Crossref has been building the plumbing to fix that, not through dramatic announcements but through metadata fields that preprint platforms can query the same way they query citations.
The Digital Object Identifier, or DOI, is the persistent link that makes a scholarly record followable. Crossref, the nonprofit DOI registration agency, registers journal articles, book chapters, datasets, and a category called posted content for preprints. A retraction notice is itself a DOI-bearing record. If a publisher deposits the relationship, the notice points to the article it retracts, and the article points back. The metadata is boring. That is the point.
How Crossref Retraction Metadata Works
Journal publishers have been able to deposit retraction relationships in Crossref for years. The 2023 record of more than 10,000 retractions made the gap between publisher practice and the metadata visible. When a publisher issues a retraction, the workflow is supposed to look like this: the retraction notice receives its own DOI; the notice metadata includes a relation to the original article; the original article record carries the inverse relation; anyone calling the Crossref API can find it. Each step can fail silently. An incomplete relation leaves a retraction notice that exists but has no visible path back to the preprint or to the version of record.
In Crossref metadata, a relation of the type is-retracted-by runs from a retracted article to its retraction notice, and the notice itself points back with a corresponding relation. Preprint servers can query that API with a DOI and receive the relationship in a structured response rather than relying on a manual search.
What Preprint Servers Could Not See
Preprints occupy an awkward editorial position. They are public but typically not peer-reviewed. A journal retraction applies to the version of record, not to the earlier draft. That does not mean the draft is vindicated. If the underlying data were fabricated, the preprint shares the problem. But until recently, many preprint servers had no automated way to know that a journal article later linked to one of their posted files had been retracted. Some servers added warnings by hand after being alerted by authors or readers. Most did not.
For servers that assign DOIs through Crossref, such as bioRxiv and medRxiv, the technical route already exists. A query for the preprint DOI can return the relationship to the published article, and a query for the published article can return is-retracted-by. The server still has to decide whether to display the notice on its landing page.
The result was a fork in the record: one DOI said the work had been withdrawn, while another DOI, for the preprint, gave no signal at all. Readers arriving through Google Scholar or a direct link could miss the retraction entirely.
Two Workflows Compared
The distinction matters in practice. A journal article can be retracted, while a preprint is normally corrected or withdrawn. Crossref integration does not retract the preprint; it supplies a relation to the original article so the preprint server can decide what to display.
| Feature | Journal article after retraction | Preprint after the journal article is retracted |
|---|---|---|
| DOI record | Can carry an explicit retraction relation to the notice | Can carry a link to the original DOI, but no retraction status of its own |
| Landing page | Typically displays a retraction banner from the publisher | Shows a warning only if the server queries Crossref and chooses to display it |
| Who deposits the signal | The journal publisher responsible for the retraction notice | The preprint server, if it chooses to update its posted-content record |
| Failure mode | Missing or incorrect Crossref relations | No automatic propagation from the publisher retraction notice |
NISO CREC and the Standardization Push
The National Information Standards Organization published the Communication of Retractions, Removals, and Expressions of Concern recommended practice to impose common metadata for these events. The acronym CREC is short, unglamorous, and exactly the kind of standard that determines whether a machine can separate a full retraction from a correction, a removal, or an expression of concern.
Crossref documentation now explains how preprint servers and reference managers can retrieve retraction records through the REST API. The Center for Scientific Integrity, which runs the Retraction Watch Database, has compiled tens of thousands of retraction notices that can be matched to DOIs. Nature reported the 2023 figure of more than 10,000 retractions in a single year.
- Retraction: the work is no longer considered credible.
- Correction: a fix that does not invalidate the study.
- Expression of concern: an interim notice while an investigation continues.
- Removal: the record is taken down for legal or safety reasons.
Paper Mills and the Strain on Metadata Systems
The strain became impossible to ignore during the Hindawi journal retractions, which followed paper mill activity. Notices arrived faster than many systems knew how to process. A retraction notice that is not linked to the preprint and not picked up by an indexer has limited effect. Readers find the preprint through a DOI and never see a warning. That is not a metadata detail; it is the difference between a retraction that changes the literature and one that merely tries to.
Other platforms have begun pulling the same Crossref signals. ResearchGate now displays retraction notices on article pages, a separate integration that depends on the relationships publishers deposit. The Committee on Publication Ethics guidance is clear about the preprint question: a preprint should be linked to a retraction notice, not quietly removed.
What This Changes for Hiring and Evaluation
Universities scan publication lists and promotion dossiers. A retracted journal article should disqualify a contribution from being counted, but hiring committees often do not know about the retraction if their systems do not consume retraction metadata. The Crossref integration makes the signal available to the tools they use; it does not force those tools to surface it. A dossier that lists only the preprint remains a known loophole.
For faculty job applicants, the practical implication is straightforward: a preprint linked to a retracted article is not a neutral artefact. Search committees increasingly use the same APIs that preprint servers do, and a missing relation can quietly misrepresent a record.
The Unanswered Question
Retraction metadata works only if publishers deposit it, preprint servers consume it, and users understand what it says. Each step is voluntary. The remaining ambiguity is one that standards bodies have not fully resolved. Does a retraction relation attached to a preprint DOI mean the preprint is retracted, or only that it points to a retracted article? Server policies still differ on the answer, and that is the question the press releases leave open.
Photo by Google DeepMind on Unsplash