Preprint Overlay Journals Are Redrawing Peer Review

What an Overlay Journal Actually Does

An overlay journal does not own or host the papers it evaluates. It sits on top of an existing preprint server, typically arXiv, bioRxiv, medRxiv, or an institutional repository. The article stays on the repository. The overlay journal adds what a traditional journal would add: peer review, a recommendation, a citable marker of quality, and a place in the academic conversation that a bare preprint does not provide.

Discrete Analysis, launched in 2016 by mathematician Timothy Gowers, is a classic early case. It publishes only articles that first appeared on arXiv and adds referee reports rather than producing new PDFs for a publisher's site. The production cost is close to zero; the editorial judgment is the product. That separation matters because distribution and evaluation stop being bundled into a single black box.

The Models Are Not All the Same

Post-publication peer review is an umbrella term wide enough to hide real differences. It covers formal review after posting, comments left by any reader, journal-independent review that happens before a paper reaches a journal, and the low-stakes note that never gets written. Conflating these is how publishers sell old practices as radical change.

• eLife's Reviewed Preprints. Every paper that clears editorial screening is published, followed by public reviews and an eLife Assessment. Authors revise, and the same article accumulates versions. The journal no longer issues a thundering reject after peer review.
• Platform publishing with invited reviews. F1000Research, launched in 2012, publishes within days, then assigns named reviewers. Two positive reviews are enough for an article to enter indexing databases such as PubMed.
• Community recommendation. Peer Community In runs communities that review preprints already on repositories. A PCI recommendation is a public signal of quality that authors can use with journals or funders.
• Journal-independent review. Review Commons, a joint project of EMBO and ASAPbio, provides peer review before submission so authors can walk the same reports into multiple journals.

Each version rearranges the same jobs: hosting, review, validation, and the billing that used to tie them together. eLife moves validation after publication. F1000Research makes publication the first step and review the second. Peer Community In removes the journal from the loop entirely until a recommendation exists. Review Commons moves review before the journal but keeps the choice of venue with the author.

Photo by Emily Park on Unsplash

Here's the Catch

The model assumes reviewers will do the same unpaid work, just later and in public. That assumption is wearing thin. Reviews on preprints remain scarce compared with the number of preprints posted. eLife's own model sidesteps the shortage by assigning reviewers before publication, which means review still happens on a publisher's timeline, not a reader's. The main change is that the rejection signal disappears; an eLife Assessment that calls a study incomplete still functions as a judgment, but the paper stays citable.

That last part is what Clarivate found troublesome. The company's explanation for withholding the Journal Impact Factor was not that eLife's science got worse. It was that eLife's process no longer fit the company's definition of a journal. The episode illustrated a point this publication reported earlier: journal indexation is not a quality guarantee. Losing an index badge can matter for careers while changing nothing about a paper's validity.

Funder platforms such as Wellcome Open Research and Gates Open Research show the same pattern. Publication is fast. Review is not. An article may be indexed after two approving reviews, but if one reviewer asks for major changes and the authors disagree, the paper sits in limbo with a public record that looks unfinished. For readers, a preprint with two signed opinions attached is not quite a verified finding.

Cheap Is Not the Same as Funded

Overlay journals promise to strip out the cost of typesetting, hosting, and most of the production chain. That is true, as far as it goes. Discrete Analysis runs on volunteer labour and arXiv infrastructure, and Peer Community In has no author fees. A system that depends on unpaid coordination tends to scale through burnout, not funding. This publication has covered the parallel problem in diamond open access, where no author or reader fees still leaves someone holding the bill for the platform, the legal work, and the staff time. The overlay model faces the same arithmetic.

Some platforms run on service fees from universities or consortia; others run on grant support. The cheapest version of a journal is not automatically the most durable one.

Hiring Committees Still Have Their Own Clock

The hiring committee does not care about the production chain. It cares about the eleven minutes per file. A reviewed preprint may be better evidence than a paywalled article, but if the committee cannot place it in a familiar ranking, it is treated as risk. The fix is not another indicator. Promotion and tenure guidelines need to name reviewed preprints, PCI recommendations, overlay journal articles, and the public review reports attached to them as legitimate evidence. Committees need to say what reviewers should actually read instead of count.

That shift is hard because it asks committees to do more reading, not less. For early-career candidates, the practical move is to present overlay and post-publication review work with a one-line explanation of what the recommendation means. A PCI recommendation on a preprint is not the same as a page number in a subscription journal, and pretending otherwise does not help.

Photo by Philip Strong on Unsplash

Generative AI Makes Post-Publication Review Harder

Publishing first and collecting review later produces a bigger surface for manipulation. The same generative AI tools that can write referee reports for traditional journals can write comments on preprints. Public review does not automatically mean trustworthy review. Earlier reporting on generative AI in peer review found policies still under construction. Any post-publication system needs to say who counts as a reviewer, whether identities are verified, and how conflicts of interest are handled.

ASAPbio has spent years pushing for open preprint review, and its work is a useful test of whether funders will pay for the human part of the process. The technology is not the bottleneck. The incentive to write a careful review for an article that may never be read remains the bottleneck.

What Happens Next

The next twelve months will show whether funders extend recognition, indexes adapt, hiring committees change their checklists, and university promotion codes remove the asterisk from preprints. The overlay journal is no longer a fringe idea, and the post-publication review model is no longer a thought piece. Both now have enough real cases — eLife's Reviewed Preprints, F1000Research's versioned papers, PCI's recommendations, Review Commons reports — that the conversation can move from whether they exist to how they should be governed.

Michael Eisen, the geneticist who edited eLife when its new model was being built, spent years telling institutions to judge the science and not the venue. He is no longer in that chair. The institutions he criticised are still hiring, still promoting, still sorting, still asking for the familiar badge. A model that separates evaluation from publication will only matter when those institutions decide the separation is legitimate.