The National Institutes of Health (NIH) unveiled a five-year plan totaling $174 million, branded as “R3PEATS,” with a stated aim of tightening the rigor, replicability, and reproducibility of NIH‑funded research. The initiative was introduced by NIH Director Jay Bhattacharya in a Science editorial and at a September press briefing, marking the agency’s response to a two‑decade‑old claim that a large share of published findings may be unreliable.
The acronym’s full expansion—“Rigor, Replicability, and Reproducibility to Promote Excellence, Accuracy, and Translation in Science”—is meant to signal a comprehensive push. Yet, much like Oreos whose nickname foregrounds a cream that isn’t really there, and Cheez Whiz’s claim to cheese that is more myth than dairy, R3PEATS risks failing to deliver real replication due to resistance from those it seeks to supervise.
The $174 million earmarked for R3PEATS breaks down to roughly $35 million per year, representing about 0.07 percent of NIH’s annual budget of $47.2 billion. During a confirmation hearing in January 2025, Health and Human Services (HHS) Secretary Robert F. Kennedy Jr.—the department that oversees the NIH—asserted that the NIH should allocate at least 20 percent of its funding to replication efforts. That would equate to about $9.4 billion annually, vastly exceeding the amount proposed for R3PEATS by a factor of about 270.
We have a fairly solid sense of replication research costs from private efforts that have operated for years on foundation or government money. In 2013, the Reproducibility Project: Cancer Biology received a $1.3 million grant from the Laura and John Arnold Foundation to conduct 50 high‑impact replication studies, ultimately spending around $50,000 per paper. More recently, psychologist Brian Nosek’s Center for Open Science published replications of 164 social‑science papers, with 49 percent of them reproducing. That project was funded by a $7.6 million grant from the Pentagon’s Defense Advanced Research Projects Agency (DARPA), amounting to under $50,000 per paper even if all funds were devoted to replication.
At those price points, $174 million could finance a few thousand replication attempts—enough to cover a random sample of several hundred NIH‑funded findings each year. Yet NIH is not devoting its entire budget to replication as the plan would suggest.
The money is divvied into five categories: roughly $44.75 million for “network and community” initiatives, $42.4 million for “grassroots rigor projects,” about $45 million for “multi‑team triangulation,” around $38.25 million for “rigor scholar cohort awards,” and $3.75 million reserved for NIH staff and workshops. Crucially, there is no allocation earmarked specifically for reproducing published results.
Why this happened is telling. In January, NIH staff proposed establishing three to five replication centers that would redo studies selected by a committee. The idea was routed to the NIH’s Council of Councils—a panel largely comprised of researchers who receive NIH funding—and the council pushed back. Members worried that a small cadre would decide which studies to replicate, that replications might overturn established science, and that such efforts could discourage innovative research. Science noted an additional concern: redoing studies could unfairly cast doubt on some scientists’ work.
Ultimately, the NIH’s move toward replication was replaced with a broader agenda of promotive, intangible outcomes. The plan’s “anticipated outcomes” describe a change in culture, new knowledge about methodological and biological variability, stronger mentorship and incentive systems, and a self‑sustaining U.S. network for reproducibility. Yet the document contains no baseline replication rate, no explicit targets or metrics, and no deadline by which the program’s success would be measured. It also does not attempt to quantify the magnitude of the reproducibility problem it is meant to address.
Bhattacharya does venture an estimate in his Science editorial, though he embeds a crucial misstatement. He writes that in studies re‑evaluating established medical practices, about 40 percent showed the newer approach was no better than the old one or was ineffective, citing a 2013 article in the Mayo Clinic Proceedings. He then adds that “preclinical cancer research also yields troubling rates of failure,” linking to a 2016 audit in PLOS Biology by epidemiologist Shareen A. Iqbal and colleagues.
That cited PLOS Biology paper does not examine preclinical cancer research nor does it attempt to quantify a failure rate. Instead, it surveys 441 PubMed articles for transparency; among 268 papers that included data, none provided complete raw data and only one described a prespecified protocol. While the study does not support Bhattacharya’s claim about preclinical cancer failure rates, it does underscore an ongoing issue: scholars like me would be happy to audit NIH‑funded studies if the NIH would simply require the data and protocol information it has claimed since 2003 to be available.
Almost every element of R3PEATS could be undertaken by a private nonprofit or academic coalition, and many already are. Nosek coordinates replication efforts; psychologist Dorothy Bishop founded the U.K. Reproducibility Network with minimal funding; journals can host dedicated replication sections, universities can broaden postdoc training, and any researcher with a laptop can assemble a PubMed browser for cross‑checking papers.
Just last month, the NIH introduced one such browser—Linked Discoveries—which color‑codes around 200 related papers by citation counts and retraction status. Bhattacharya acknowledges that it does not determine whether a finding is correct, and he is right.
In his Science editorial, Bhattacharya points toward actions for other players—“research institutions should,” “publishers should,” “funders should,” and “scientific leaders should.” He, who leads the world’s largest funder, emphasizes that “funders should recognize” rather than stating that “the NIH will require.”
The NIH could condition funding on stricter data practices. It could demand preregistration of confirmatory analyses, require raw data and full protocols to be posted, make grant renewals contingent on publishing data from prior awards, audit a random sample of funded results each year and publish the findings under the grantee’s name, and score future grant requests based on whether the applicant’s past findings endured scrutiny. All of these measures would cost far less than $174 million, yet they remain unpopular with grantees, which is why such measures are more feasible when imposed by a funder alone.
Whether the NIH is prepared to adopt such a stance remains unclear. Data sharing has been “expected” since 2003, and a 2007 federal statute requires registered clinical trial results to be posted within a year, under penalties exceeding $10,000 per day. Yet a 2020 Lancet study found that only 31 percent of government‑funded clinical trials (mostly NIH) met the deadline, a performance even worse than the 41 percent rate for government‑funded studies overall.
The FDA waited until April 2021 to issue its first noncompliance notice regarding the data mandate and, to date, has not been known to levy fines. The FDA’s broad enforcement push arrived with a March 2026 letter to 2,200 research sponsors reminding them that the law exists.
An HHS inspector general review found that roughly half of NIH‑funded trials with data due in 2019 or 2020 were reported on time. The Government Accountability Office reports that NIH had never suspended grant funding for noncompliance with the mandate until October 2021—more than a decade after the requirement was introduced.
In January 2016, NIH added mandatory “rigor and transparency” sections to every grant application. A decade later, that requirement often results in a boilerplate paragraph rather than meaningful change.
Cases of fraud highlight the insufficiency of these half‑hearted safeguards. Charles Piller, the Science journalist who documented failures in trial reporting and who wrote the book on scandals involving manipulated Alzheimer’s disease images, described a culture where doctored papers were uncovered by unpaid data sleuths working after hours, not through funded auditing programs. One of the researchers named in Piller’s book ran the neuroscience division of the NIH’s own National Institute on Aging. A robust, well‑funded audit program is crucial to alter the culture Piller depicts—one where people fear speaking out, institutions shield their stars, and journals take years to retract misleading research.
It’s essential to note that an ideal replication rate would not be 100 percent. Scientists should not postpone publication until every finding is proven beyond any doubt, and even the strongest studies are sometimes superseded by future work. The core problem is not constructive work that is revised or overturned as others build on it—it’s the existence of utterly useless research conducted by scientists who hide data, obscure methods, and misrepresent results.
There are real challenges in rooting out poor research without treating researchers like criminals, stifling innovation, or wrecking careers. Some portion of audit results will be incorrect, leading to legitimate findings being questioned and a tendency to focus on past errors rather than future breakthroughs. Yet R3PEATS seems to settle for mere ceremonial action, lingering on the breakfast‑meeting table along with Froot Loops and Krispy Kreme, rather than delivering substantive reform.
