\documentclass[11pt]{article}

\usepackage[margin=1in]{geometry}
\usepackage{amsmath}
\usepackage{amssymb}
\usepackage{graphicx}
\usepackage{booktabs}
\usepackage{tikz}
\usepackage{pgfplots}
\pgfplotsset{compat=1.18}
\usepackage[colorlinks=true, linkcolor=black, citecolor=black, urlcolor=blue]{hyperref}

% --- skin palette: flag red, ledger slate, amber -----------------------------
\definecolor{flagred}{HTML}{B8452E}
\definecolor{flagdeep}{HTML}{8A3220}
\definecolor{ledger}{HTML}{33475A}
\definecolor{amber}{HTML}{9C6B16}
\definecolor{millgold}{HTML}{A9761A}

\title{Forty-Five Editors, Thirty Percent of the Retractions}
\author{The Science Journaling Club}
\date{Field Note --- Paper Analysis --- Metascience}

\begin{document}
\maketitle

\begin{abstract}
In August 2025 a team from Northwestern, Sydney, and the PNAS editorial board
published a forensic audit of who handled which manuscript at several
journals, cross-referenced against which manuscripts were later retracted or
flagged on the post-publication review site PubPeer \cite{richardson2025}.
At PLOS ONE alone, of 18{,}329 editors who have ever accepted a submission, a
Poisson-binomial enrichment test flags 45, a mere 0.25\%, who together handled
1.3\% of the journal's 276{,}956 articles but sit behind 30.2\% of its 702
retractions. More than half of them, 25 of 45, also authored papers that
PLOS ONE retracted. A cluster of them spanning four countries sent most of
their submissions to each other rather than to the journal's other 18{,}284
editors. The same signature, self-referential acceptance, batched
retractions, and a shared trail of duplicated images, reappears at Hindawi
and across IEEE conference proceedings. This Field Note walks through
exactly how the authors caught it, why a coordinated ring is a fundamentally
harder catch than a single fabricated paper, and what the exercise implies,
and does not imply, about trusting the other 99.75\% of editors doing the job
honestly. All illustrative arithmetic beyond the paper's own reported numbers
is clearly marked as the club's own; we ran no experiment and analyzed no new
data of our own.
\end{abstract}

%============================================================================
\section{Forty-Five Names on Two Spreadsheets}

Two spreadsheets did it. Not a whistleblower, not a leaked internal email,
not a confession extracted under pressure. Two ordinary bibliographic
exports: one listing which handling editor accepted which manuscript at a
journal, the other listing which of those manuscripts were later retracted
or drew a comment on PubPeer. Merge them. Run a statistical test on every
editor who has ever worked there. See who comes out anomalous.

At PLOS ONE, which is unusual among large journals in publicly naming the
editor who handled each accepted paper, that test flags forty-five people
\cite{richardson2025}. Forty-five out of 18{,}329 editors who have accepted
at least one article since the journal launched in 2006, a group so small it
is 0.25\% of everyone who has ever done the job. Between them they have
handled 1.3\% of the journal's 276{,}956 published articles. And they account
for 30.2\% of everything the journal has ever had to retract.

That is the headline of a paper by Reese Richardson, Spencer Hong, Jennifer
Byrne, Thomas Stoeger, and Lu\'is Amaral, published in the \textit{Proceedings
of the National Academy of Sciences} in August 2025, titled ``The entities
enabling scientific fraud at scale are large, resilient, and growing
rapidly'' \cite{richardson2025}. One of the authors, Amaral, is himself a
longtime PLOS ONE editor, which is part of why the paper reads less like an
outside accusation and more like an institution auditing itself.

What follows is our attempt to understand the machinery behind that number.
How do you catch forty-five people doing this without a confession? Why is a
network of colluding editors so much harder to catch than one person faking
a gel image? And once caught, what does it change about trusting the review
process the rest of us rely on?

%============================================================================
\section{Defection, Not Villainy}

The paper's own framing is worth taking seriously before the numbers.
Richardson and colleagues borrow a term from game theory: \emph{defection}
\cite{richardson2025}. Treat the scientific enterprise as a public-goods
game in which everyone benefits from a shared pool of trustworthy knowledge
and is supposed to contribute genuine work to it. A defector takes the
benefits without contributing genuine knowledge. Historically defection was
assumed to be rare and individual: one person under career pressure fabricates
one dataset.

The paper's taxonomy has three moving parts, and they are not the same
thing. \emph{Paper mills} mass-produce fake or low-quality manuscripts for
sale. \emph{Brokers} connect that supply to journals willing to publish it,
sometimes by paying editors, sometimes by embedding themselves as editors.
\emph{Predatory or hijacked journals} are the venues, sometimes legitimate
journals a broker has captured outright, that skip the quality control a
journal is supposed to provide. None of the three alone needs a network of
forty-five people. What needs the network is doing all three at once,
repeatedly, at a scale bigger than any single fabricator could manage alone.

%============================================================================
\section{How You Catch an Editor Doing This}
\label{sec:method}

PLOS ONE has published 276{,}956 articles since 2006. Of those, 702 have been
retracted and 2{,}241 have drawn a comment on PubPeer \cite{richardson2025}.
Neither number is evenly spread across the journal's history or its 18{,}329
editors. For every editor, for every year active, the authors counted how
many manuscripts that editor accepted were \emph{eventually} retracted, and
separately how many eventually drew a PubPeer comment, then asked: given the
journal-wide, time-varying background rate, is this editor's count higher
than chance would produce? The tool is a Poisson binomial test, run editor by
editor, with the false discovery rate held at 0.05 across all 18{,}329
simultaneous tests \cite{richardson2025} --- the same multiple-testing
correction problem as testing thousands of genes at once, applied here to
thousands of editors.

They ran the test three ways: eventual retraction, eventual PubPeer comment,
and unusually short turnaround (30 days or fewer between submission and
acceptance), a hallmark some regard as paper-mill-adjacent processing
\cite{richardson2025}. Retraction anomalies flagged 22 editors; PubPeer
anomalies flagged 33; short-turnaround anomalies flagged 88. But when the
authors checked whether \emph{authors} were disproportionately routed to
those flagged editors, the numbers inverted: 21 authors for the
retraction-flagged group, 18 for PubPeer, and only 7 for short turnaround.
A footnote from Amaral explains part of why: PLOS ONE itself once actively
encouraged fast turnaround as policy, later deprioritized, so some flagged
editors were simply following journal guidance \cite{richardson2025}. The
authors' own data show their weakest signal being weak for a documented,
non-nefarious reason.

\begin{figure}[htbp]
\centering
\begin{tikzpicture}
\begin{axis}[
  width=12.5cm, height=7cm,
  ybar, bar width=16pt,
  ymin=0, ymax=100,
  symbolic x coords={retraction anomaly, PubPeer anomaly, short turnaround},
  xtick=data,
  x tick label style={font=\small},
  ylabel={people flagged},
  ymajorgrids=true, grid style={dashed, gray!30},
  legend style={font=\scriptsize, at={(0.02,0.97)}, anchor=north west}
]
\addplot[fill=ledger, draw=ledger] coordinates {
  (retraction anomaly,22) (PubPeer anomaly,33) (short turnaround,88) };
\addlegendentry{editors flagged}
\addplot[fill=millgold, draw=millgold] coordinates {
  (retraction anomaly,21) (PubPeer anomaly,18) (short turnaround,7) };
\addlegendentry{authors disproportionately routed to them}
\end{axis}
\end{tikzpicture}
\caption{Three anomaly signals across all 18{,}329 PLOS ONE editors, FDR held
at 0.05. Short turnaround flags the most editors (88) but the fewest
complicit authors (7), consistent with the paper's footnote that PLOS ONE
briefly encouraged fast turnaround as policy. Data: \cite{richardson2025}.}
\label{fig:signals}
\end{figure}

%============================================================================
\section{The Ones Who Send Work to Each Other}

The 22, 33, and 88 above overlap and don't capture everyone: an editor can
also be flagged indirectly, by having their own submissions disproportionately
handled by \emph{other} flagged editors. Combine every direct and indirect
route and the total comes to 45 PLOS ONE editors \cite{richardson2025}.

Here is the club's own arithmetic on top of that. If retractions were
distributed in proportion to workload, 45 editors responsible for 1.3\% of
PLOS ONE's articles should be responsible for about 1.3\% of its 702
retractions: roughly 9. They are instead responsible for 30.2\%, or about
212. Dividing 212 by 9 gives a group producing retractions at more than
\textbf{23 times} the rate their workload share would predict. That ratio is
ours; the paper reports the two percentages, we divided them.

More than half, 25 of the 45, also personally authored articles that
PLOS ONE went on to retract \cite{richardson2025} --- the same people appear
on both sides of review. Mapping who sent submissions to whom, the authors
found a densely connected cluster of editors serving 2020--2023, based
across four countries, who sent most of their own submissions to each other
rather than to the journal's other 18{,}284 editors. More than half of the
articles this cluster accepted were later retracted, with near-identical
retraction notices:

\begin{quote}
\itshape ``One of a series of submissions for which we have concerns about
authorship, competing interests, and peer review.''
\end{quote}
\hfill --- as reported in \cite{richardson2025}, citing PLOS ONE editorial notices

The paper flags an honest complication worth repeating: editor assignment at
PLOS ONE is nominally close to random, but authors may recommend an editor,
so an author landing with a preferred editor ``is not unlikely'' on its own
\cite{richardson2025}. Some of what looks like editor-side self-dealing could
in principle be author-side targeting of accommodating editors; the
statistical signature does not fully distinguish the two, and the paper does
not pretend it does.

\begin{figure}[htbp]
\centering
\begin{tikzpicture}
\begin{axis}[
  width=12.5cm, height=7cm,
  ybar, bar width=22pt,
  ymin=0, ymax=35,
  symbolic x coords={share of editors, share of articles, share of retractions},
  xtick=data,
  x tick label style={font=\small},
  ylabel={share attributable to the 45 flagged editors (\%)},
  ymajorgrids=true, grid style={dashed, gray!30}
]
\addplot[fill=ledger, draw=ledger] coordinates {
  (share of editors,0.25) (share of articles,1.3) };
\addplot[fill=flagred, draw=flagred] coordinates {
  (share of retractions,30.2) };
\end{axis}
\end{tikzpicture}
\caption{A quarter of one percent of PLOS ONE's editors, handling 1.3\% of
its articles, accounts for 30.2\% of its retractions --- about
23$\times$ their proportional share, by the club's own division of the
paper's two percentages. Data: \cite{richardson2025}.}
\label{fig:disproportion}
\end{figure}

%============================================================================
\section{Same Story at Other Journals, and in Conference Halls}

Hindawi, the other publisher that discloses handling editors, offers an
independent test of whether the pattern generalizes. Restricting to the ten
Hindawi journals with the most retractions and PubPeer comments, the same
procedure flags 53 editors for anomalous retraction rates, 52 for anomalous
PubPeer rates, and 96 for anomalous fast-tracking, with six, four, and seven
of those respectively flagged at more than one journal simultaneously
\cite{richardson2025}.

The method also extends to a venue with no single ``editor'': IEEE conference
proceedings. Treating each organizing committee as equivalent to an editor,
the authors examined 2{,}294{,}067 articles across 19{,}969 IEEE conferences
since 2003. Forty-five conferences held 2009--2011 had more than 10\% of
their published articles eventually retracted, producing a visible spike in
the retraction record around those years \cite{richardson2025}. The full
test flags 84 conferences for retraction anomalies, 158 for PubPeer
anomalies, and 183 for anomalously frequent ``tortured phrases,'' the
garbled synonym-substitutions typical of paraphrasing software used to evade
plagiarism detection. At least seven multi-year IEEE conference series were
flagged for tortured-phrase anomalies in \emph{every single year} they were
held \cite{richardson2025}.

%============================================================================
\section{A Second Fingerprint, Left in the Images}

Editorial handling records are one line of evidence. The paper builds a
second, independent one from duplicated images. Starting from PubPeer reports
of image duplication, the authors constructed a network of 2{,}213 articles
(nodes) connected by 4{,}188 duplicate-image observations (edges), splitting
into 20 connected components, the largest holding 622 articles
\cite{richardson2025}.

Using stochastic block modeling to find internal structure, two results
stand out. First, articles inside a given module are not randomly
distributed across publication year or publisher; a Shannon-entropy test
rejects that null for essentially every module with enough data to test
\cite{richardson2025} --- consistent with a broker placing a batch of
manuscripts in one venue within a tight window, rather than independent
authors reusing their own old figures. Second, despite hard visual evidence
that its underlying data cannot be what it claims, only 34.1\% of this
network has actually been retracted \cite{richardson2025}. Two-thirds of a
network built entirely of provably duplicated images sits uncorrected in the
scientific record.

\begin{figure}[htbp]
\centering
\begin{tikzpicture}
\draw[flagred, line width=1.4pt] (0,0) -- (0:1.5) arc[start angle=0, end angle=-122.76, radius=1.5] -- cycle;
\fill[flagred, opacity=0.85] (0,0) -- (0:1.5) arc[start angle=0, end angle=-122.76, radius=1.5] -- cycle;
\draw[ledger, line width=1.4pt] (0,0) circle (1.5);
\node[font=\bfseries] at (0,0.3) {622};
\node[font=\small] at (0,0.02) {articles};
\node[font=\small, anchor=west] at (1.7,-1.0) {\textcolor{flagred}{34.1\% retracted}};
\node[font=\small\itshape, anchor=north] at (0,-2.0) {largest connected component};
\node[font=\small, anchor=north] at (0,-2.9) {network total: 2{,}213 articles \textbullet\ 4{,}188 duplicate-image edges \textbullet\ 20 components};
\end{tikzpicture}
\caption{The largest component of the image-duplication network, wedge
shaded to the retracted fraction, mirroring the paper's own Figure 2C
convention. Data: \cite{richardson2025}.}
\label{fig:network}
\end{figure}

%============================================================================
\section{Why Not Just Check Who Cites Whom?}

A natural guess about how you'd catch a cartel of cooperating editors is
worth confronting directly, because it is not what this paper did. If a
group of editors and authors were cooperating, wouldn't they also cite each
other, inflating citation counts and impact factors the way documented
citation cartels do elsewhere? Coordinated editorial action toward exactly
that goal is real and separately studied: a 2022 \textit{Research Policy}
paper examined how institutions and journals game citation-based rankings
\cite{siler2022}, and a 2024 \textit{Scientific Reports} paper mapped
fabricated co-authorship networks used to sell authorship slots
\cite{porter2024}.

But that is not the fingerprint Richardson and colleagues pulled up. Nowhere
in the PLOS ONE, Hindawi, or IEEE analyses does a citation network appear.
The signal was who \emph{accepted} whose manuscript and who \emph{reused}
whose image, not who \emph{cited} whose paper. That distinction changes what
a ring has to do to stay invisible. A citation cartel has to cite, leaving
tracks exactly where bibliometricians already look. A network that
fast-tracks and rubber-stamps each other's submissions needs no cross-citing
at all --- the papers can sit in entirely unrelated subfields and never once
reference one another. An audit that only ever examined citation graphs,
however sophisticated, would have sailed straight past this one.

%============================================================================
\section{The Business That Follows the Deindexing List Around}

The paper's clearest evidence for an organized broker is a named case study:
ARDA, whose website as of June 2024 advertised ``Conferences and Meetings,''
``Journal Publication,'' and ``Thesis/Article Writing,'' and claimed
involvement in ``4{,}565+ publications'' \cite{richardson2025}.

Since January 2018, ARDA has listed 188 unique journals as guaranteed
publication venues, growing from 14 journals in January 2018 to 86 by March
2024 --- roughly 34\% compound annual growth in portfolio size over six
years, by the club's own arithmetic on those two endpoints. Of the full list,
106 (56.4\%) were Scopus-indexed, 51 (27.1\%) Web-of-Science-indexed, 29
(15.5\%) both, and 2 (1.1\%) MEDLINE-indexed; 17 (9.0\%) are suspected
``hijacked journals,'' formerly legitimate titles a paper mill has captured
outright \cite{richardson2025}.

A journal ARDA listed in 2020 that was Scopus-indexed at the time had a
33.3\% chance (13 of 39) of later being deindexed, versus a 0.5\% baseline
(147 of 27{,}197) for Scopus-indexed journals actively publishing in 2020
generally, a difference the paper confirms with a two-tailed test ($P <
0.0001$) \cite{richardson2025}. Dividing those percentages, as the club did,
puts an ARDA-listed journal at roughly \textbf{66 times} the deindexing risk
of an ordinary journal. When a batch of journals was deindexed in 2020--21,
ARDA dropped them from its site by May 2021 and replaced them with new
titles \cite{richardson2025}. For five ARDA-listed journals inspected in
detail, 34.0\% to 98.7\% of published articles fell entirely outside the
journal's stated scope.

\begin{figure}[htbp]
\centering
\begin{tikzpicture}
\begin{axis}[
  width=6cm, height=6cm, ybar, bar width=26pt,
  ymin=0, ymax=100,
  symbolic x coords={Jan 2018, Mar 2024}, xtick=data,
  x tick label style={font=\scriptsize},
  ylabel={journals listed by ARDA},
  ymajorgrids=true, grid style={dashed, gray!30}
]
\addplot[fill=ledger, draw=ledger] coordinates {(Jan 2018,14) (Mar 2024,86)};
\end{axis}
\end{tikzpicture}
\hspace{0.4cm}
\begin{tikzpicture}
\begin{axis}[
  width=6cm, height=6cm, ybar, bar width=26pt,
  ymin=0, ymax=40,
  symbolic x coords={baseline 2020, ARDA-listed 2020}, xtick=data,
  x tick label style={font=\scriptsize, rotate=12, anchor=east},
  ylabel={\% later deindexed by Scopus},
  ymajorgrids=true, grid style={dashed, gray!30}
]
\addplot[fill=ledger, draw=ledger] coordinates {(baseline 2020,0.5)};
\addplot[fill=flagred, draw=flagred] coordinates {(ARDA-listed 2020,33.3)};
\end{axis}
\end{tikzpicture}
\caption{Left: ARDA's portfolio grew sixfold in six years. Right: an
ARDA-listed 2020 journal was roughly 66$\times$ more likely to later be
deindexed than an ordinary Scopus journal (ratio computed by the club from
the paper's two percentages). Data: \cite{richardson2025}.}
\label{fig:arda}
\end{figure}

%============================================================================
\section{Some Fields Are Simply Better Targets}

Restricting to six closely related, similarly sized RNA biology subfields
with comparable growth since roughly 2000, the authors tracked annual
publication counts alongside annual errata and retraction rates from 2002 to
2022 \cite{richardson2025}. Errata sit in a narrow band of 1.5\% to 2.5\%
across all six subfields; the literature in each is revisited for ordinary
reasons at roughly the same rate.

Retraction rates do not share that consistency. CRISPR-Cas9 sits near 0.1\%.
Transfer-RNA-and-cancer studies peak near 1\%. Circular RNA studies peak near
2.5\%. Both microRNA-and-cancer and long non-coding RNA studies peak near
4\%, and for individual journals in those two subfields the rate exceeds
10\% \cite{richardson2025}. Because the errata baseline stays flat while
retraction rates swing roughly 40-fold across subfields studied with the
same tools over the same period, the authors argue the gap reads as
targeting rather than a maturity artifact.

\begin{figure}[htbp]
\centering
\begin{tikzpicture}
\begin{axis}[
  width=13cm, height=7cm, ybar, bar width=20pt,
  ymin=0, ymax=5,
  symbolic x coords={CRISPR-Cas9, tRNA \& cancer, circular RNAs, miRNA \& cancer, lncRNAs},
  xtick=data, x tick label style={font=\scriptsize, rotate=12, anchor=east},
  ylabel={peak retraction rate (\%)},
  ymajorgrids=true, grid style={dashed, gray!30}
]
\addplot[fill=ledger, draw=ledger] coordinates {
  (CRISPR-Cas9,0.1) (tRNA \& cancer,1.0) (circular RNAs,2.5)
  (miRNA \& cancer,4.0) (lncRNAs,4.0) };
\draw[millgold, dashed, line width=1pt]
  (axis cs:CRISPR-Cas9,1.5) -- (axis cs:lncRNAs,1.5)
  node[pos=0.5,above,font=\scriptsize,millgold]{errata band 1.5--2.5\%};
\end{axis}
\end{tikzpicture}
\caption{Retraction rate swings roughly 40-fold across six matched RNA
biology subfields while the errata rate stays flat. Data: \cite{richardson2025}.}
\label{fig:rna}
\end{figure}

%============================================================================
\section{The Arithmetic of Outrunning Everyone Chasing It}

The paper's closing empirical argument is about speed. Comparing four
annually growing quantities, the authors report doubling times of 15.0 years
for total global scientific output, 3.6 years for PubPeer-commented
articles, 3.3 years for retracted articles, and 1.5 years for suspected
paper-mill products \cite{richardson2025}.

Doubling times are awkward to compare by eye, so here is the club's own
conversion: a doubling time $T$ corresponds to a continuous annual growth
rate $r = \ln(2)/T$. Applied to the paper's four numbers:

\begin{table}[htbp]
\centering
\begin{tabular}{lrr}
\toprule
\textbf{Quantity} & \textbf{Doubling time (paper)} & \textbf{Annual rate (club, $\ln2/T$)} \\
\midrule
Total global publications        & 15.0 y & 4.6\% \\
PubPeer-commented articles       & 3.6 y  & 19.3\% \\
Retracted articles               & 3.3 y  & 21.0\% \\
Suspected paper-mill products    & 1.5 y  & 46.2\% \\
\bottomrule
\end{tabular}
\caption{Doubling times: \cite{richardson2025}. Growth-rate column: club arithmetic.}
\label{tab:growth}
\end{table}

Divide the fastest rate by the slowest and the fraud economy, as the paper's
own trend lines describe it, is growing at almost exactly \textbf{ten times}
the rate of legitimate science itself.

Web of Science and Scopus each deindex on the order of a hundred journals a
year, tenfold smaller than the number of journals currently publishing
suspected paper-mill products \cite{richardson2025}. Of 29{,}956 suspected
paper-mill products with a matching OpenAlex record, only 8{,}589 (28.7\%)
have actually been retracted; extrapolating current trends, the authors
project only around 25\% will ever be retracted and only around 10\% will
ever end up in a deindexed journal \cite{richardson2025}. On the paper's own
projection, roughly three out of four papers already identifiable as
probable paper-mill products today will sit uncorrected in the record
indefinitely.

%============================================================================
\section{Why This Beats the Immune System Built for Lone Fabricators}

A lone fabricator concentrates risk: one person, one paper, one implausible
dataset. The signal is small in scope but large in magnitude, exactly what
post-publication review sites like PubPeer are built to catch. A coordinated
network does the opposite: it distributes risk unevenly, by design or by
accident. Of the 45 flagged PLOS ONE editors, only 22 carried an
individually anomalous retraction rate on their own record --- a needle so
fine, out of 18{,}329 editors, that finding it required a purpose-built
statistical test run simultaneously on everyone who has ever done the job,
with a false-discovery correction to stop the multiple-testing problem
itself from manufacturing false positives. The other 23 needed no anomalous
personal record at all: they were caught only because they routed
submissions to editors who were already flagged, a second, independent
signal unrelated to any one person's own numbers looking alarming.

Catching one bad paper takes a reader with good instincts. Catching this took
a research team, a full retraction database, PubPeer's comment archive, and
a statistical method sophisticated enough to separate a real coordinated
signal from eighteen thousand editors' worth of ordinary noise --- and even
then, the network described operated for at least three years, 2020 to
2023, before this analysis surfaced it.

%============================================================================
\section{What This Should, and Should Not, Change About Trusting Peer Review}

It would be easy to flatten this into ``peer review is broken.'' The paper
resists that reading. Forty-five editors is 0.25\% of everyone who has done
the job at PLOS ONE; the other 99.75\% show no statistical anomaly at all.
The finding is not that editors, as a class, cannot be trusted. It is that
an accountability system built around the assumption that fraud is rare and
individual has little capacity against fraud that is organized and
cooperative.

The authors' own discussion is structural rather than moralistic. Journal
editors ``have been offered substantial payments for rapid publication of
selected manuscripts,'' and ``coordinated editorial action has been
implicated in efforts to increase the impact factor of journals''
\cite{richardson2025}, citing among other work the impact-factor gaming
literature above \cite{siler2022}. The people currently doing the work of
finding this are, in the authors' words, ``courageous but isolated
individuals,'' some ``accused of vigilantism and dismissed,'' others
``threatened with legal action'' \cite{richardson2025}. Their recommendation:
separate detection, investigation, and sanctioning, and remove each from the
hands of parties with a conflict of interest in the outcome.

A coda in the discussion deserves attention. Both symbolic ``machine
scientists'' and large language models are increasingly used to summarize
the literature, and none can currently distinguish a real paper from a
fraudulent one \cite{richardson2025}. Every fraudulent paper this network
produced, and every one still uncaught, becomes an input to systems built on
the premise that the literature they read is trustworthy --- and that
compromised fraction is, by the paper's own numbers, the fastest-growing
fraction there is.

%============================================================================
\section{What the Authors Themselves Flag as Limits}

The paper's limitations section is direct. The analysis can only find fraud
that has already left a trace in the corpora available, and the expert
consensus it cites holds that the large majority of paper-mill products have
not yet been detected at all \cite{richardson2025}. Case studies such as
ARDA and the RNA biology subfields focus on particular businesses or
disciplines that may not generalize. And changes over time in how much
attention watchdogs pay to different fields could themselves produce trends
that look like growth in fraud without being growth in fraud, a possibility
the authors say they cannot fully rule out.

We end where the authors end their own discussion: not with a verdict that
the system has failed, but with an argument that detecting fraud at this
scale cannot keep being the work of isolated volunteers. It needs resources
commensurate with a fraud economy growing, by the paper's own trend lines
and the club's own arithmetic on them, about ten times faster than the
science it hides inside of.

%============================================================================
\begin{thebibliography}{9}

\bibitem{richardson2025}
Richardson, R.~A.~K., Hong, S.~S., Byrne, J.~A., Stoeger, T. \& Amaral,
L.~A.~N. (2025). The entities enabling scientific fraud at scale are large,
resilient, and growing rapidly. \textit{Proceedings of the National Academy
of Sciences} 122(32), e2420092122. doi:10.1073/pnas.2420092122

\bibitem{siler2022}
Siler, K. \& Larivi\`ere, V. (2022). Who games metrics and rankings?
Institutional niches and journal impact factor inflation. \textit{Research
Policy} 51(10), 104608. doi:10.1016/j.respol.2022.104608

\bibitem{porter2024}
Porter, S.~J. \& McIntosh, L.~D. (2024). Identifying fabricated networks
within authorship-for-sale enterprises. \textit{Scientific Reports} 14,
29569. doi:10.1038/s41598-024-71230-8

\end{thebibliography}

\end{document}