The interval between finishing a result and having it certified is measured in months to years, and much of that interval is pure overhead: an editor searching for referees, referees postponing, revision rounds serialized one after another, and finally a production queue [1, 2]. Surveys of authors put typical turnaround at several months even in fast fields, with first-decision times that have grown over the decades rather than shrunk despite the move to electronic submission [2]. In a survey of authors in conservation biology, the median reported turnaround was fourteen weeks against a perceived reasonable time of six, with individual horror stories running past a year [3]; the broader picture across fields is similar, and mathematics, computer science, and economics sit at the slow end [2, 4]. The cost of this delay is not merely inconvenience. Priority disputes, stalled dissertations, and grant and job timelines all hinge on the certification date, and knowledge that could be built on sits idle while it waits in a queue.

If review were slow but accurate, the delay might be a price worth paying. It is not especially accurate. The classic demonstration remains that of [16], who resubmitted twelve already-published articles to the same prestigious psychology journals that had printed them, with the authors’ prestigious affiliations replaced by fictitious low-status ones. Only three of thirty-eight editors and reviewers noticed the papers had appeared before; of the nine that then went through review, eight were rejected, most on grounds of “serious methodological flaws” in work the same journals had recently judged excellent. Two decades later, [17] found that agreement among reviewers in clinical neuroscience was scarcely greater than chance. The pattern recurs wherever it is measured directly: the 2014 and 2021 consistency experiments at the NeurIPS conference, which sent a subset of submissions through two independent committees, found that roughly half of accepted papers would have been rejected had the process simply been rerun [5]. A process whose accept/reject verdict is that sensitive to the luck of the referee draw is closer to a noisy lottery than to a measurement.

The noise is not even unbiased. In a controlled experiment at WSDM 2017, single-blind reviewers—those who could see author identities—bid for and favored papers from famous authors and prestigious institutions significantly more than double-blind reviewers evaluating the same papers [6]. The wider literature on reviewer bias documents effects tied to gender, nationality, language, and seniority [7]. A former editor of the BMJ, reviewing the evidence, concluded that peer review is slow, expensive, largely a lottery, and poor at detecting either error or fraud [8]. The point is not that referees are careless; it is that two or three of them, chosen under time pressure and reading in private, are a statistically thin instrument on which to hang a binary, career-defining verdict.

The system is also expensive, and the expense is decoupled from the service rendered. A small number of commercial publishers control more than half of all research output and sustain operating margins in excess of thirty percent [9, 10]—margins that persist precisely because the labor that creates the value, writing and refereeing, is donated by academics, while the resulting content is sold back to their institutions through subscriptions or to their funders through article-processing charges [10]. Universities pay twice, once in salaries to produce the work and again in fees to read it. Whatever one thinks of the profit, the architecture is hard to defend: the marginal cost of distributing a paper is now near zero, yet the price is not.

Finally, the reviewing effort the system extracts is spent in the wrong place. Every submission is refereed with roughly equal intensity before anyone knows whether it matters, which means enormous scrutiny is lavished on work that will never be read or cited, while the community’s collective attention is rationed by the accident of which journal accepted a paper rather than by the paper’s actual interest. Referee time is scarce and fatigue is real [3]; spending it uniformly, up front, and in private is close to the least efficient possible allocation. It would be better to let attention find the work that warrants it and to concentrate scrutiny there.

This is the strongest objection, and the evidence is sobering. Open commentary is a public good, and public goods are under-supplied: the private cost of writing a careful review is borne by the reviewer, while the benefit is diffuse. The current system papers over this by conscripting referees through editors; remove the conscription and voluntary participation may collapse. The data bear this out. A study of all 2020 bioRxiv and medRxiv preprints found that only about \( 7\%\) received even a single comment over roughly seven months [13], consistent with earlier estimates below ten percent [14]. A system in which ninety-plus percent of papers get no feedback has not solved the certification problem; it has merely relocated it.

The design must therefore treat participation as the central engineering problem, not an afterthought. Three levers help. First, make commentary and review count: attributed reviews and highly-upvoted comments should be citable, first-class research outputs that appear on a CV, so that the effort buys recognition rather than only costing time [15]. Second, retain a light editorial nudge—a small rotating pool that solicits a minimum number of reviews for each deposited paper—so that the baseline is not zero; this is a far cheaper editorial function than the current one and does not gate publication. Third, exploit the fact that, when comments do appear, they are substantive: the same bioRxiv analysis found that the comments left were predominantly criticisms, corrections, and suggestions of the kind traditional review produces [13]. The scarcity is of quantity, not quality, which is an incentive problem, and incentive problems are tractable.

The proposal requires real names, and real names cut two ways. They confer accountability and make credit assignable, but they also deter exactly the criticism the system needs most: a junior researcher who publicly faults a senior one risks retaliation at the next grant panel or hiring committee. It is not an accident that PubPeer, the most active venue for catching serious problems in the literature, permits anonymity. A pure real-name policy may therefore purchase civility at the price of candor.

The resolution is not to abandon real names but to be selective about where they are required. Authorship and forking, where credit is the point, should be attributed. Critical commentary can be offered under a persistent, verified pseudonym: an identity that is stable and reputation-bearing (so that a track record of sound criticism accrues, and abuse can be sanctioned) but not trivially linkable to a career. This preserves accountability at the level of the account while protecting the critic, and it lets the community weight a comment by the commenter’s history rather than only by their name.

Journal prestige, for all its faults, does real work: hiring committees, tenure cases, and grant panels use “published in \( X\) ” as a compressed, portable quality signal, and they use it precisely because they cannot read every candidate’s work closely. An archive with no accept/reject bit removes that signal, and the reform will not be adopted until something replaces it [9]. The design does supply replacements—vote counts, the number and standing of substantive reviews, fork counts and downstream citations in the provenance graph—but these are richer and noisier than a single journal name, and evaluators would have to learn to read them. There is also a genuine risk of trading one flawed proxy (journal brand) for another (a popularity score). The honest position is that the signal can be reconstructed, probably better, but that doing so requires deliberate metric design and a cultural shift in how committees evaluate, and that this is the reform’s true bottleneck—sociological, not technical.

The affirmative case assumes that interesting work attracts scrutiny and unimportant work is harmlessly ignored. Attention, though, is distributed by a rich-get-richer dynamic at least as much as by merit: famous authors, prestigious institutions, well-networked labs, and fashionable topics draw eyes regardless of quality, and a vote count amplifies this Matthew effect rather than correcting it. The bias documented for referees [6] does not vanish when the crowd replaces the panel; it may intensify, since the crowd sees author names by default. So “unimportant papers are simply ignored” is too sanguine: important work by unknown authors may also be ignored, and visible endorsement may track prominence rather than correctness. Mitigations exist—default-blinded presentation of author identity, vote-weighting that discounts pile-on and rewards reviewer track record, surfacing under-examined deposits—but they must be built in deliberately. Left to a raw vote, the system risks reproducing the inequities it set out to cure.

Any votable public forum invites gaming: coordinated up-voting of allies, down-voting of rivals, sock-puppetry, and the general fact that a popularity score measures agreement, not truth—and the crowd has been confidently wrong before. “Approved users” is the first line of defense: participation should require a verified scholarly identity (for example, an institutional or ORCID- backed account), which raises the cost of Sybil attacks and makes sanctions meaningful. Beyond that, votes should be weighted by reviewer standing rather than counted raw, coordinated voting should be detectable and penalized, and— crucially—the vote should be treated as an input to human judgment, not a verdict. The system’s job is to surface argument and evidence for readers to weigh, not to compute a truth value by majority.

Removing the pre-publication gate means flawed, fraudulent, or dangerous work reaches the public immediately, as the COVID-19 preprint episode illustrated. This is a real cost, and it should be stated plainly rather than waved away. Two things bound it. First, the gate the reform removes was never good at catching fraud in the first place [8]; open, attributed, post-deposit scrutiny by the whole community has a substantially better recent record at detection than closed pre-publication review did. Second, the design can retain targeted safeguards without a general gate: clear preprint-style labeling of review status, expedited community flagging, and heightened checks for categories where premature dissemination carries public-health or safety risk. The claim is not that an open archive eliminates bad work—nothing does—but that it exposes and corrects it faster and more visibly than the system it replaces.

Finally, the word “approved” hides a governance problem: whoever controls admission and moderation controls a gate, and a reform premised on removing gates cannot smuggle one back in unexamined. The answer is to keep the admission criterion broad, objective, and non-discretionary—a verified scholarly identity rather than a curator’s judgment—and to make moderation policy transparent, appealable, and ideally federated across institutions rather than vested in a single operator. Governance is where a system like this most easily goes wrong, and it deserves as much design attention as the commentary and forking mechanisms themselves.