Fact-check: Crimes Against Science

← Back to the essay: the English original · the Arabic translation

This is a page of criticism, not a defence of the essay. It records what was checked and what was found, and nothing has been removed from it. The essay was published as written and was not altered to match this check. The scope is facts only: dates, names, numbers, quotations, attributions, and whether a cited source says what it is cited for. The thesis, the argument and the prose were not assessed.

This check covers the whole essay, the Introduction to the Conclusion, Parts I to X. It is in English and checks the English text. The Arabic version of the essay is a translation of that text.

Updated 13 October 2026. Parts IX and X were added to the essay, and this check was extended to them: new rows in the first table, finding 4, a second look at the Part V sentence on Boldt's starch trials, new entries under "Not established", the minor points and the source links, and a plagiarism scan of the two new parts. An Arabic translation of the essay was published. Everything that was on this page before stays as it was.

How it was checked

Each factual claim was searched on the web and compared with what named sources returned: encyclopedias and reference works, museum and university pages, Retraction Watch, Science, Nature, the Harvard Crimson, the Stanford Daily, Pesquisa FAPESP, elDiario.es, Chemistry World and others, each linked below where it is used. Earlier fact-checks on this site were searched first; the one for the Stanford 2% essay had already settled the 2019 Ioannidis database and the Nature analysis of self-citers.

The source pages themselves could not be opened in this pass. The evidence on this page is the text that search results showed from those pages, not a reading of the full articles or papers. Where a result gave only a paper's abstract or a summary of it, that is what is meant by "confirmed". Every link in the essay was also tested the same way.

A number of well-documented points in Parts V to VIII were not searched in this pass and are not counted as checked here: Summerlin, Darsee, Schön, Hwang, Wakefield, Stapel and the STAP case; the tobacco, sugar and Vioxx histories; publication bias (Turner 2008, ICMJE 2005, AllTrials); Beall's list; and the dates for Garfield, Hirsch, DORA, the Leiden Manifesto and CoARA. The same applies in Parts I to IV to the 1933 Civil Service law, Lenard and Stark, Mengele's twin research, the HeLa cells, and the wording of the ORI definition.

Parts IX and X were checked the same way, in a second pass on 13 October 2026. The claims in them had already been searched once, when the two parts were being written. This pass checked the published text against those results, searched again everything the first search had left uncertain or that the text states more firmly than the first search did, and re-searched a sample of what had been confirmed. The book details in the new reference entries and the new web links were checked too. As before, no source page was opened; the evidence is search-result text.

First: what held up

Most of the essay holds up. In Parts I and II, the Ptolemy debate, Galileo's trial, Beringer's planted stones, the 1712 Royal Society report on Newton and Leibniz, Fisher's verdict on Mendel, Babbage's four offences, Blondlot and Wood, the Piltdown exposure, Kammerer's inked toad and the Burt twin data all match the sources. In Parts III and IV, the Lysenko and Vavilov dates, the Nazi sterilisation figure, Verschuer and the Kaiser Wilhelm Institute, the Dachau and Ravensbrück experiments, the Doctors' Trial, Unit 731, Tuskegee, Guatemala, Beecher and Willowbrook, and the dates of the ethics reforms are all supported. In Parts V to VIII, the retraction records, the Gino, Macchiarini and Tessier-Lavigne cases, Bik's image study, the Clarivate suppressions, tortured phrases, sneaked references, coercive citation, the Springer Nature retractions and the Luque suspension are all supported. In Parts IX and X, the sleuths' cases and the lawsuits against them, the founding and sale of the Retraction Watch database, the PubPeer ruling, the institutional measures, the trust surveys, the DECREASE meta-analysis and the AI cases match the sources, with the exceptions below. The book details in the reading list (editors and authors, translator, publisher and year for all nine) match publisher and library records. The main points, with a source for each:

Second: what needs correcting or qualifying

1. Piltdown: what the 2016 study traced to a single orangutan. The essay says the De Groote study "concluded that the specimens came from a single orangutan and pointed to one forger, most plausibly Dawson himself." The study, as its summaries report it, traced the orangutan parts (the jaw and teeth) to one animal; the cranial pieces were human, from at least two individuals. "The specimens" reads as the whole find. The previous sentence in the essay does say the cranium was human, so a careful reader can piece it together, but a quick one may not. The second half is accurate: the authors point to one forger, most likely Dawson, and say it is uncertain whether he acted alone. Sources: LJMU repository; ScienceDaily. Both are summaries; the paper itself was not read.

2. Corchado: El País began reporting in 2024, not 2023. The essay says: "From 2023 the Spanish newspaper El País reported that he had instructed staff to insert citations to his work into their papers." The earliest El País article found is from March 2024 (Manuel Ansede's report on a four-paragraph paper that cited its author about 100 times). A whistleblower approached the paper in 2023, but that was a tip, not a report. The substance is supported as El País's reporting of an allegation: a Madri+d article states that El País reported Corchado allegedly asked staff to add citations to his own work. It was not shown who controlled the Google Scholar and ResearchGate profiles involved. Corchado is a living person; he denies wrongdoing and has called the reporting a campaign against him, and no finding of misconduct by him personally was seen in the sources. Earlier public criticism came from Retraction Watch in March 2022. Sources: Science; Retraction Watch, March 2022; elDiario.es (his denial).

3. Saudi affiliations: the drop came the same year, and its cause is stated more firmly than the sources state it. The essay says: "In 2023 El País reported on Spanish researchers ... The following year, the number of Saudi-affiliated researchers on Clarivate's list fell from 109 to 76 after Clarivate tightened its screening." The fall from 109 to 76 is between the 2022 list and the 2023 list, which came out in November 2023, the same year as the El País reports, not the year after. On the cause, Science says European institutions had raised concerns and that "tighter scrutiny from Clarivate appears to have played a role". The essay makes this a flat cause. Clarivate's 2023 exclusions, more than 1,000 researchers, were about citation practices in general. One SIRIS report gives 112 for 2022 rather than 109, so the starting figure depends on how affiliations are counted. Sources: SIRIS Academic; Science; Pesquisa FAPESP.

4. The PNAS piece on AI and the cost of fraud is not by Richardson's group. Part X says: "Richardson's group has argued in PNAS that it pushes the cost of fraud toward zero while checking stays costly", and links to 10.1073/pnas.2521606122. That article is "Confronting the inevitable: Harnessing technology to contain systemic scientific fraud", a single-author commentary by Philipp Singer of Roche Diagnostics, published on 27 October 2025. It responds to the Richardson study and makes the argument the essay gives: the cost of making a fake falls toward zero while peer review stays expensive. So the argument is real and the link points to it, but it is Singer's, not Richardson's group's. A reader would take it as the view of the researchers who measured the paper mills, which gives it more weight than a commentary carries. Richardson's team has said, as reported by the ASCO Post, that the growth they measured happened almost entirely before generative AI arrived in late 2022. Sources: PMC record of the commentary; ASCO Post. Both were seen as search results only.

Checked, and it holds: Boldt and hydroxyethyl starch (Part V). Part V says: "once his fraudulent studies were removed, meta-analyses showed that hydroxyethyl starch increased the risk of kidney injury and death in critically ill patients." Part X points back to it. While Part X was being written, a search turned up an updated meta-analysis of one starch, HES 130/0.4, that found no significant rise in deaths (RR 1.14, 95% CI 0.89 to 1.46), which suggested the sentence might be too strong. It is not. That analysis is by Wiedermann and Joannidis (2012): 13 trials and 1,131 patients, for one product, and the results seen do not say whether Boldt's trials were in it. The main meta-analysis, Zarychanski and colleagues in JAMA in 2013, left out seven Boldt trials and found higher mortality with starch (RR 1.09, 95% CI 1.02 to 1.17, 28 trials, 10,290 patients) and more need for dialysis (RR 1.32). The essay's sentence matches that. One shading, which does not change what a reader takes away: with Boldt's trials included the mortality estimate already leaned against starch (RR 1.07), so removing them made the harm clearer rather than revealing it from nothing. Sources: MDedge; Healio; Swiss Medical Weekly (Wiedermann and Joannidis). These are summaries; the papers were not read.

Third: not established

These could be neither confirmed nor ruled out from what search returned.

• That Lysenko "announced that the Central Committee had approved his report" in 1948. Search results credit Stalin with approving the session and do not mention the Central Committee.
• That the Dialogue went onto the Index in 1633 itself.
• Westfall's example of a fudged precession of the equinoxes. The sound-velocity example was confirmed; this one appeared only in a secondary source.
• That Lysenkoist close planting was used in China's Great Leap Forward.
• Wiley's price for Hindawi, "about $298 million". The Register, which the essay links, does not give it, and search found it in one blog only. The year the Hindawi brand was retired is also not pinned: sources differ between late 2023 and 2024.
• "About 15 years for science as a whole" in the PNAS paper-mill study. The 1.5-year doubling for paper mills is confirmed by press coverage; the 15-year figure was not seen in one of the two search passes, and the paper itself was not read.
• "Bogus upside" as a tortured phrase for "false positive". It was not in any list returned. "Counterfeit consciousness" for artificial intelligence was confirmed.
• That junior authors were "most likely to comply" in Wilhite and Fong. The paper's finding, as summarised, is that junior faculty were more often coerced; "comply" rests on one secondary summary.
• That the Shanghai ranking stopped counting secondary affiliations.
• That Study 329 "has never been retracted". This was true in the latest sources seen, from late 2025, when the journal declined a retraction request; nothing from 2026 was found.
• Where Gino's lawsuit stands after September 2026; Summerlin's ink being "washed off with alcohol"; and the author of the 2011 Science report on Saudi affiliations.
• Part IX: what came of the Marseille complaint against Bik. Nothing after 2021 was found.
• Part X: that the Surfaces and Interfaces paper "was later retracted". Coverage of the case was found, but the retraction notice was not.
• Part X: that Springer Nature screens images with Proofig or ImageTwin. Science's adoption of Proofig (2024) and ASM's of ImageTwin (a pilot reported in mBio) were found; Springer Nature's was not. The 2026 preprint the essay cites for all three was not returned by search either.
• Part X: the contents of the emails Tessier-Lavigne released in September 2026. The Stanford Daily's report of their release was found; no independent account of what they show.
• Part X: how much of England's fall in MMR coverage the Wakefield paper caused. The essay says this cannot be measured, and nothing found measures it.

Smaller points that do not change what a reader takes away: Beringer's book is dated 1726, while the hoax is usually placed in 1725; the 1933 law did not itself remove Einstein, who was abroad and resigned his posts, or Haber, who also resigned; Burt's figure is exactly 0.771, not "around" it, and Kamin's analysis is dated 1972 to 1974 in different sources; Babbage's "trimming" means smoothing readings toward the mean, close to but not quite "clipping outlying readings"; Chou's removal by Elsevier was published in late January 2020, not 2019, though the investigation was in 2019; the sneaked-references study concerned three journals of one publisher, not "some publishers"; and the 1969 Brown & Williamson memo's own words are "Doubt is our product since it is the best means of competing with the 'body of fact' that exists in the mind of the general public", which the essay paraphrases fairly and without quotation marks.

Smaller points in Parts IX and X: Abalkina's figures mix two versions of her study, since 1,063 offers is the count in the published Learned Publishing paper, which links at least 451 papers to the mill, while 434 comes from the March 2022 preprint (1,009 offers); "at least 434" is still true, only lower than the final count. Nature did not name the Problematic Paper Screener a key development of 2021; it named Cabanac, its builder, one of "Nature's 10" people of 2021. The Taloni dataset "would appear perfectly plausible to an untrained eye", as the essay says, but a 2025 follow-up in the same journal reports that the 2023 datasets failed forensic checks, and that 4 of 12 made by a refined tool passed them.

Source links

Each link in the essay was checked against search results. These are the problems found.

• Macchiarini (KSAT). An AP wire story copied on a local TV site; the facts are right, but the AP original would be the better link.
• Luque (WION). A syndicated copy on an embed. subdomain; Chemistry World reports the same facts directly.
• Corchado (elDiario.es). An AMP copy of the page; the standard URL ends _1_11820117.html. The article is about the rector attacking Nature over the retractions, and search did not show it as the source for the "factory" report or the El País reporting it is cited for.
• Hindawi (Retraction Watch). The linked piece is about the International Wound Journal and manipulated peer review, not Hindawi's special issues, so it does not support the sentence it is attached to.
• Hindawi (The Register). Confirms the 11,300 retractions and the end of the Hindawi brand, but not the $298 million figure.
• Richardson (blog). Search did not return this post. The PNAS paper, or Retraction Watch's report of 4 August 2025, would be a firmer source.
• Saudi purge (Pesquisa FAPESP, Spanish). The Spanish URL was not returned by search. The English edition exists and confirms 109 to 76.

The other links (Retraction Watch on Boldt and on the Clarivate suppressions, the Harvard Crimson, and the arXiv paper on Google Scholar) say what they are cited for. The reference list has no URLs or DOIs, and its page and volume details were not checked.

The links added with Parts IX and X were checked the same way. These are the problems found.

• PNAS (Part X). The link works and supports the argument, but the article is Philipp Singer's commentary, not a paper by Richardson's group. See finding 4.
• Bik feature (Yahoo News). A syndicated copy with a search-tracking tail (?fr=sycsrp_catchall). Search returned the same article on AOL, without a byline; it appears to be a 2021 BuzzFeed News story, but the original was not found. The Nature and Science links cited next to it carry the same facts.
• Gaming the Metrics (49th Shelf). A bookseller's catalogue page standing in for the book. The book is open access from MIT Press (DOI 10.7551/mitpress/11087.001.0001), which would be the better link.
• Image screening (arXiv, 2026). A preprint, not peer-reviewed, and search did not return it. The publishers' own announcements would be firmer sources; see "Not established".
• Fabricated citations (Enago). An editing company's blog standing in for the study. The 55% and 18% figures are from Walters and Wilder, Scientific Reports 2023 (DOI 10.1038/s41598-023-41032-5).
• Abalkina (arXiv). The reference list links an earlier preprint whose counts differ from the published Learned Publishing paper (DOI 10.1002/leap.1574). See the smaller points above.
• DECREASE (Forbes). Two contributor blog posts reporting the meta-analysis. They report it fairly; the paper itself, Bouri et al. in Heart, is free on PMC.
• Wikipedia (GRIM test, Theo Baker, Leonid Schneider). Encyclopedia entries where the original paper or report exists. The facts they are cited for match.

The other new links (Springer Nature, Retraction Watch, Crossref, Science, the Harvard Crimson, Courthouse News, STAT, Undark, the Stanford Daily, Pew, the University of Zurich, VICE, JAMA Ophthalmology, PMLR, PolitiFact) say what they are cited for, as far as search results show.

Plagiarism scan

Ten of the essay's most distinctive sentences were searched as exact phrases, among them "Piltdown teaches that fraud survives longest when it flatters the community that examines it", "Lysenkoism is the purest case of falsification by decree" and "a judge sitting on his own case". All ten returned no match. Search engines do not strictly enforce exact-phrase matching, so a zero is evidence that the wording is not copied, not proof. Ten case passages (Piltdown, Tuskegee, Wakefield, Schön, Lysenko and Vavilov, Summerlin, Nazi science, Burt, tobacco, and the cases Broad and Wade also cover) were compared with what Wikipedia, Britannica, Retraction Watch, the Natural History Museum, the CDC and similar sources return for the same cases. The site's 25 English essays and 38 Arabic essays were searched for the essay's distinctive phrases and case names.

Overall risk: LOW. The books the essay draws on (Broad and Wade, Oreskes and Conway, Reverby, R. R. Newton) were not compared with it directly, only with what search returns about them, so close paraphrase of those books is not established either way. Closeness between the English text and the Arabic essays was also not established, because the Arabic search did not work reliably on those pages.

The scan above covered the first version of the essay, Introduction to Conclusion without Parts IX and X.

Parts IX and X. Five distinctive sentences were searched as exact phrases: "A flag is a reason to look, not a verdict", "Most of this work is unpaid, and the legal risk falls on the person who speaks first", "A guideline built on false data turns one man's fraud into routine practice", "A retraction corrects the journal, but it does not reach the people the fraud already persuaded" and "A machine can now write the paper, but only a person can be made to answer for it". None returned a match. The two parts were also compared with the wording of the sources they were written from. Two phrases follow a source closely without quotation marks, and both name that source in the same sentence: the Taloni dataset "would appear perfectly plausible to an untrained eye", which is the authors' wording, and review text that "could have been substantially modified" by language models, which is Liang and colleagues' wording. Etzioni's "a hallmark of AI-generated citations" is quoted and credited. None of his other essays repeats this wording; the one shared phrase found on the site, "speaks first" in his essay on balance, is used there in an unrelated sense.

Overall risk for Parts IX and X: LOW.

Limits of this check

No source page, paper or book was opened in full; everything rests on search-result text, as described above. The points listed under "How it was checked" as not searched were not checked at all. Claims such as "has never been retracted" cannot be proved from a finite search. The essay's interpretations and argument were not assessed. For Parts IX and X, most claims rest on one round of search made while the parts were being written, with a fresh search for the uncertain ones and a sample of the rest; the nine books on the reading list were checked for their details, not read.

Verdict

The essay is accurate on the great majority of its dates, figures and names, across all eight parts. The three findings are a loose sentence about the 2016 Piltdown study, a date a year early on a living person's case, and a date a year late, with a firmer cause than the sources give, on the Saudi affiliations. The Hindawi sentence and the paper-mill growth figures should be read with reserve until the links behind them are replaced, and the Corchado paragraph is a report of allegations he denies.

Parts IX and X hold up the same way. Their names, dates, court outcomes and survey figures match the sources, and they are careful where the evidence is thin: they say that a flag is not proof, that nobody has shown misconduct caused the fall in trust, and that the MAHA report's use of AI was never confirmed. There is one finding, an argument in Part X credited to Richardson's group that belongs to a commentary by another author. The Part V sentence on Boldt's starch trials, questioned while Part X was being written, holds. The paragraph on publishers' image screening should be read with reserve, because its source is an unreviewed preprint that search did not find and one of the three publishers it names was not confirmed.

© 2026 Husain Alkhaldy — fact-check of this essay. Checked 13 October 2026.