The Retraction Watch Database now holds records for 50,000 retracted papers. The figure is not a celebration; it is a bookmark in the history of scientific publishing. The database, launched in October 2018 by Ivan Oransky and Adam Marcus, began with roughly 18,000 entries and has grown at a rate few predicted when the pair started Retraction Watch in August 2010. It is maintained by the Center for Scientific Integrity, the nonprofit parent of Retraction Watch.
Retraction Watch began as a chronicle of notices journals would rather bury. The database made those notices searchable by author, country, journal, reason and date. That shift, from anecdote to dataset, changed what research integrity work can ask.
Oransky and Marcus did not invent retractions. They gave them an address.
A retraction is not the opposite of publication. It is publication correcting itself in public. The database has turned that correction into a durable record: 50,000 entries that say something went wrong and was acknowledged. Some notices are one line; others stretch for pages with legal hedging.
The milestone invites a mistake, though. Retraction counts are not a measure of how much fraud exists. They are a measure of how much fraud has been detected, documented and disclosed. The gap between those two quantities is where most of the scholarly record lives. A retraction is the literature's most honest sentence: it says the record was wrong and tells you how.
That is why the 50,000 mark matters less as a total and more as an infrastructure. The database allows questions that were once unaskable.
The cases behind the count
Behind the number are individual histories. Andrew Wakefield's 1998 Lancet paper on the MMR vaccine and autism was retracted in 2010 after a long investigation; it remains a reference point for how slowly flagship journals move. Diederik Stapel, a social psychologist, had 58 papers retracted for fabricated data. The anaesthesiology researchers Joachim Boldt and Yoshitaka Fujii account for more than 400 retractions between them, records that show the scale of harm in a field where patients were involved.
Some cases predate the database by decades. Jan Hendrik Schön, a Bell Labs physicist, had roughly two dozen papers retracted in 2002 for fabricated molecular-scale electronic results. The database collects these older records alongside new ones, which gives researchers a longer memory than any single institution or publisher keeps voluntarily.
The public record includes a range of reasons: data fabrication, image manipulation, plagiarism, authorship disputes, legal threats and honest error. The database tags each retraction with the reason stated in the notice, so fields can be compared. Retraction Watch continues to report new notices, while the searchable Retraction Watch Database remains free to use.
Photo by David Pupăză on Unsplash
Paper mills changed the arithmetic
For years, retractions were treated as rare individual failures. The past decade showed something different: industrial-scale fabrication. Paper mills produce fake or heavily templated manuscripts and sell authorship slots to researchers under pressure to publish. The scale became impossible to ignore in 2023, when more than 10,000 papers were retracted in a single year, according to a Nature analysis. Most came from Hindawi journals, an open-access publisher acquired by Wiley.
The database absorbed those notices, and the Hindawi episode pushed the total up sharply. Our earlier reporting on Wiley and Hindawi's mass retractions tracked how publishers respond when a journal list is compromised. Without a central ledger, a sudden wave of retractions across hundreds of journals becomes almost impossible for librarians, funders, hiring committees and research integrity officers to track.
Paper mills do not operate in one country or one discipline. They target journals with weak peer review and authors with strong publication requirements. The Committee on Publication Ethics has produced guidance on paper mill detection for editors, but the tools remain uneven. A retraction database is the backstop when those tools fail after publication.
What the leaderboards show
The database can be sorted by country, journal, author and reason. Raw country totals put China, the United States, India, Iran and Japan among the largest counts, but those rankings mirror publication volume as much as misconduct rates. A more careful reading uses several lenses.
- Retractions per paper published offer a fairer comparison across countries.
- Journal-level filters reveal repeat offending at specific titles.
- Author-level searches expose patterns of serial fabrication.
- Reason codes separate honest error from image duplication and data fabrication.
Some disciplines appear often because their methods were easy to fabricate. Psychology, anaesthesiology, cancer biology and the image-heavy life sciences are heavily represented in the older retraction literature. Newer waves have spread through materials science, plant biology, engineering and machine learning journals as paper mills learned to generate plausible datasets. The change is not that fraud moved; it is that the database made the movement visible.
Retracted papers still get cited years after the notice appears. Studies have repeatedly shown that many post-retraction citations fail to acknowledge the retraction. The database's integration with Crossref has made it easier for citation managers, reference checkers, preprint servers and systematic review tools to flag retracted work automatically. Our earlier article on Crossref retraction metadata reaching preprint servers explains how this works in practice.
Why institutions remain the slow part
A retraction can arrive years after the underlying misconduct. Universities often take longer. An investigation may sit in a dean's office while the paper remains in the curriculum, on a CV, in a clinical guideline or in a subsequent meta-analysis. The database shortens the distance between public notice and institutional memory, but it does not close it.
The 50,000-retraction milestone should put hiring committees and promotion panels on notice. A retraction is a searchable fact, and search committees already read the literature. For applicants, the database is not a punishment; it is a disclosure tool. A candidate who lists a retracted paper with an honest note has a better claim to integrity than one who hopes the notice stays buried. Research job postings increasingly name data integrity and responsible publication among expectations.
Funders and publishers have been slower to build shared infrastructure. The database filled a gap that commercial indexing services left open for decades. Clarivate's Web of Science and Elsevier's Scopus list retractions inconsistently, and their paywalled platforms make audit trails harder to follow. The Retraction Watch Database is free, which is a quiet institutional rebuke.
Photo by Mika Baumeister on Unsplash
The next 50,000 will look different
If the first 50,000 retractions were dominated by individual fabrication and paper mills, the next stretch will reflect more machine-assisted image manipulation, synthetic datasets, AI-written text and automated citation padding. Journals are deploying image-forensics tools and statistical checks, but the volume of submissions outpaces the number of people qualified to review them. Detection tools produce leads, not verdicts.
There is no reason to believe scientific misconduct is worsening at exactly the rate the database grows. The database grows because detection, journal willingness, disclosure and database infrastructure have all improved, unevenly. That is the better reading of the number: 50,000 is what transparency looks like when it works, not what fraud looks like when it runs wild.
The retraction notice remains a rare genre of honesty. It admits a specific failure in public and attaches it permanently to the literature. The database has now collected 50,000 of those admissions. The task ahead is to make the gap between what is wrong and what is acknowledged as small as the evidence will allow.