"Peer-reviewed" functions in public debate as a stamp of established truth. Inside science it means something narrower: two or three unpaid experts, working in spare hours, judged a manuscript plausible, competent and interesting enough to publish. That filter has real value, and knowing what it does not do has become essential literacy for anyone who reads science news.
Reviewers assess logic, methods as described, and fit with existing knowledge. They almost never see raw data, do not re-run analyses, and cannot detect a fabricated spreadsheet that looks like a real one. Peer review is a plausibility check, not an audit, and its designers never claimed otherwise. Its known biases are also documented in its own literature: novelty and positive results pass more easily than careful null findings, famous names and prestigious institutions get read more generously, and genuinely revolutionary work is periodically rejected by referees invested in the paradigm it disturbs.
The strain is quantitative as much as qualitative. Millions of papers now flow annually through a system whose reviewing labour is donated, and career incentives reward publishing far above the invisible work of checking others' publishing. Editors report begging dozens of researchers to secure two reports. Into that overload has stepped organised abuse: paper mills selling authorship on fabricated manuscripts, fake reviewer identities recommending acceptance of their own products, and journals compromised at industrial scale, followed by mass retractions running into the thousands and the delisting of entire titles.
The repair work
The interesting story is what science is building alongside the old machine. Preprints separate distribution from certification: findings post immediately, and evaluation follows in public, a shift the pandemic normalised at speed, with well-known costs when raw claims meet headlines. Registered reports attack publication bias at the root, with journals accepting studies on the strength of the question and methods before results exist, so null findings get published rather than shelved. Open data and code requirements make the audit possible that review never was. And post-publication scrutiny has acquired teeth: statistical sleuths and image-forensics volunteers, coordinating on platforms built for the purpose, have driven retractions of prominent results and unseated senior figures, functioning as the investigative journalism of the research world.
None of the reforms replaces the core insight that expert eyes catch errors authors cannot see; they redistribute when and how the eyes arrive. For the reading public, the practical calibration is this: a single peer-reviewed paper is a claim entered into evidence, not a verdict. The verdict is replication, convergence across groups, and survival of exactly the scrutiny described above, and headlines built on one exciting paper are reporting the opening statement as if it were the judgment.
