I want to share a short commentary I just published, on “The Modal Mode of Thinking about Scholarly Publishing,” as part of the Journal of Electronic Publishing’s 30th anniversary special issue (edited by Alyssa Arbuckle and Janneke Adema).

Here’s the abstract from the piece (OA and free to download):

The essay argues that the study of scholarly communication would benefit from attending to a “modal” sensibility—that is, a self-conscious sensitivity to the differences that different mediums make in understanding published works of scholarship. The essay critiques the unreflective textualism that dominates the conversation on publishing. The claim is that the primacy of text, as the sovereign medium of academic communication, is a largely invisible parochialism. The essay points to examples and traditions of multi-modal publishing as an entry point to taking the medium-specificity of publishing formats as an object of analysis. Such experimentation has followed, sometimes closely, the emergence of new mediums of storage and transmission within the societies that scholars work. The mid-twentieth century birth of the modern medium concept made multi-modality a conceivable, self-conscious project. Even so, the discourse on academic publishing has rarely registered the implications, including for inherited text-based formats. The essay concludes with a call for media scholars, curiously underrepresented in the discourse, to take up this task, with reference to pioneering works in the field.

The other three essays in the issue—by Arbuckle and Adema, Chérifa Boukacem-Zeghmouri (en français), and John W. Maxwell—are excellent, and well worth reading.

Speaking of ‘publish or perish’: Kelly-Ann Allen, writing in Nature earlier this year (paywalled, alas):

In my view, we could make academic life more rewarding by systematically scoring the underappreciated behaviours and conduct that researchers value. The product might be called the ‘G+ index’ — G representing generosity, giving and other ‘good things’ in academia. This index could either stand alone or complement existing citation measures.

Allen proposes a bunch of potential measures, with the proviso that the full spread should be decided by the academic community. And she thinks the G+ index should be mostly private:

Rather than creating a public ranking system, individual scores could be kept private, sharable with funders and institutions for evaluation purposes. Shifting control to individual researchers might go some way to reduce pressure to game the system.

It’s a thoughtful proposal, resonant with the great work of the HuMetrics group. There are also resonances with the longstanding effort (not too successful, alas) to promote alternatives to the GDP measure.

Vladimir Moskovkin, writing for LeidenMadtrics, posted a delightful essay on the origins of the “publish or perish” phrase. His sleuthing traced the expression back to at least the late nineteenth century, in an attribution (undated) to Swiss historian Jacob Burckhardt:

Since Jacob Burckhardt died in 1897, the literary proven reference to the slogan "publish or perish" should be pushed back from the early 20th century (William Morris Davis, 1904) to at least the late 19th century. It would be interesting to initiate historical and archival research using the memoirs of Jacob Burckhardt and the works of his biographers in order to find more detailed information on the description of this phenomenon.

I’m posting a short commentary here, a preprint under review at Yearbook on Data Autonomy (YoDA), a new OA journal from the University of Groningen Press.

Download PDF

---

In the last year, the giant AI companies have released tools with indistinguishable names: Deep Research (Google), Research (Anthropic), deep research (OpenAI), Researcher (Microsoft), and—yes—Deep Research (Perplexity).¹ In the highly profitable world of scholarly publishing, meanwhile, some of the biggest players have launched their own products, based on their proprietary databases of the academic literature: Scopus AI (Elsevier), Web of Science Research Assistant (Clarivate), and Dimensions AI Assistant (Digital Science).² What these two bundles of tools share is the promise to summarize and cite. They each return paragraphs of distillations, complete with links to apparent sources.

The AI companies and publishers draw on overlapping bodies of knowledge. The products from Elsevier and company leverage the peer-reviewed literature, while the various Deep Researches cite not just academic work but reports, Wikipedia, and posts on the open web. But the tools arrive at a similar place: summary and citations.

Even the marketing is hard to distinguish. Google pitches its Gemini-based tool as “your personal research assistant.” OpenAI touts its model’s ability to produce reports at the “level of a research analyst,” with “clear citations.” Anthropic’s Research promises “high-quality, comprehensive answers in minutes” and “easy-to-check citations.” All of that is echoed in the scholarly publishers’ marketing boasts. Elsevier’s Scopus AI generates “reliable and digestible research summarization.” The offering from Clarivate, the Web of Science Research Assistant, lets you “synthesize the literature more efficiently.” The Dimensions AI Assistant, from Springer Nature’s corporate sibling Digital Science, will return “immediate and more relevant summary results.”³ Silicon Valley and big publishing are even converging on product language: Elsevier’s Scopus AI recently released its own “Deep Research”-branded feature.⁴

At the core of all these tools is a series of verbs: to surface, to rank, to summarize, and to recommend. In this sense the AI products are already acting as knowledge arbitrators, by picking winners and losers according to what they make visible. There are two big and interlocking problems with this role: the models are trained on the scholarly past, and their filtering logic is inscrutable. As a result, they may smuggle in the many biases that mark the history of scholarship, around gender, geography, and other lines of difference.

In this context it’s useful to revive an old concept in the sociology of science. According to the Matthew effect—named by Robert Merton decades ago—prominent and well-cited scholars tend to receive still more prominence and citations. The flip side is that less-cited scholars tend to slip into further obscurity over time. These dynamics of cumulative advantage have, in practice, served to amplify the knowledge system’s patterned inequalities—with, for example, diminished citations for female scientists, a development aptly labeled the Matilda effect by Margaret Rossiter.⁵

The deployment of AI tools in scholarship, as their use picks up, may produce a Matthew effect on the scale of Scopus, and with no paper trail. The problem is analogous to the well-documented bias-smuggling with existing generative models; image tools trained on, for example, mostly white and male photos reproduce the skew in their prompt-generated outputs. With our bias-laden scholarship as training data (“120 years of research,” in Web of Science’s boast), the research models may spit out results that, in effect, double-down on inequality.⁶

What’s worse is that we won’t really know, due to the models’s black-boxed character. Thus the tools may act as laundering machines—context-erasing abstractions that disguise their probabilistic “reasoning.” Existing biases, like male academics’ propensity for self-citation, may win a fresh coat of algorithmic legitimacy. Or consider center-periphery dynamics along North-South and native-English-speaking lines: Gaps traceable to geopolitical history, including the legacy of European colonialism, may be buried still deeper. The models, in short, could serve as privilege multipliers. Early evidence suggests that’s exactly what’s happening.⁷

The Matthew Effect in (Human) Scholarship

The Columbia sociologist Robert Merton named the Matthew effect in 1968. Merton sourced the idea—that the already well-known accrue still-more reputational lucre—to the New Testament: “For to every one who has will more be given, and he will have abundance; but from him who has not, even what he has will be taken away” (Matthew 25:29). The reason that the scholarly rich get richer, according to Merton, is familiarity. The eminent scientist is already lodged in other scholars’ memories; when it comes time to cite or reference or read, these are the names that are ready at hand. The subsequent spread of the famous name leads to still-stickier impressions, seeding even more mentions. The result is a visibility cascade.⁸

Merton framed the dynamic as a “misallocation” of recognition—an injustice to those (often junior) scientists whose credit is diverted to the undeserving few. He invoked what he called the “class structure” of science, which the Matthew effect helps underwrite. Akin to the broader society’s stratification system, the academic class structure distributes life-chances unevenly. The already famous don’t just win honor and esteem; they also secure “access to the means of scientific production”—and thus the chance for still more prestige and more resources.⁹

Merton heavily qualified his critique, however. Yes, the Matthew effect may be unfair to individual scientists, but it serves a healthy, system-wide purpose. Given the crushing volume of research (even in 1968), the rich-get-richer dynamic acts to tame the literature, by “focalizing” attention on important findings. This “communication function” helps to clear the field—to concentrate attention—and thereby “accelerate” the march of science.¹⁰ He also downplayed his core point—cumulative advantage—by singling out the unique “character structure” of renowned scientists. These luminaries really are different from their peers: They’re more confident and “self-directed,” and they’re better at recognizing, and then framing, key findings. So perhaps it’s a good thing, after all, that they’ve won the spotlight.¹¹

The takeaway from Merton’s famous paper is that, whatever the injustice to individuals, the overall system of science benefits. Merton’s upbeat functionalism was soon challenged, within a broader 1970s critique of his sociology-of-science program. The problem wasn’t with the Matthew effect thesis itself; indeed, most subsequent studies furnished empirical support for the claim. The issue was with Merton’s claim that the system benefits, even if individuals suffer.¹²

It’s not just isolated individuals who suffer, after all. The Matthew effect helps mete out the rewards of recognition in patterned ways—along lines of difference already inscribed in the postwar U.S. (and world) academic system. Consider gender: The Matthew effect helped to sideline already marginal female scientists. Dynamics of cumulative dis-advantage worsened, and thereby perpetuated, the patterned exclusion of academic women. This was Margaret Rossiter’s charge in her 1993 “The Matilda Effect,” named for Matilda Gage, a late nineteenth-century U.S. suffragist who both described and experienced the spiraling of silence. Rossiter, the late historian of science, cited the history of male-colleague credit accrual. The Matilda effect names a process of erasure that, in traceable ways, adds insult to reputational injury.¹³

One of these Matildas, arguably, was Merton’s graduate student and future wife, Harriet Zuckerman. Merton’s 1968 paper (cited over 12,000 times) was largely based on Zuckerman’s 1965 dissertation on Nobel laureates. In a self-exemplifying irony, Zuckerman’s 1967 journal-article writeup, like her subsequent book, have attracted far less attention—and she is almost never credited as the concept’s co-author. Hence Rossiter’s half-serious suggestion to call it the “Harriet Effect” in honor of Merton’s “invisible collaborator.”¹⁴

Rossiter’s point about doubling-down on patterned exclusion has been extended to other “have-nots,” who, like women, have been “ignored, denied credit or otherwise dropped from sight.”¹⁵

Too Much to Know

One of Merton’s claims for the upsides of the Matthew effect was that it helped the scientific community cope with the publication onrush. In conditions of published abundance and attention scarcity, we can’t help but sift and summarize. As Ann Blair has brilliantly shown, there was always too much to know—leading, in the early modern European case she studies, to various filtering and synopsis devices like the index and the reference work. The more recent emergence of the journal-article abstract is another concession to information overload.¹⁶

The scholarly citation is also a summary device. On the textbook construal, references are the linchpin of the collective enterprise of science; when we cite, we stand on the shoulders of select and worthy predecessors. But the humble citation is much more than this; in-text and footnoted references to the “literature” are flexible, indeed baggy, symbols, doing all kinds of diverse work for citing authors. Citations can signal allegiance, call out a rival, establish precedence, and spread the peacock feathers. They are legitimacy set in type. And they serve to summarize and direct—to offload the burdens of evidence and support.¹⁷

In the last half-century, we have come to collate and count citations—to distill them into lean and precise-seeming metrics. The bibliographic labor of scholars is, of course, the raw material from which tallies and indexes are made. In this second layer of abstraction—by which symbols are fashioned from symbols—a single integer stands in for each of us. A journal’s citations are counted too, and rendered with ordinal clarity to encourage citation-seeking submissions that—in an autopoietic coupling—run up both scores in addition to publisher margins. All to answer the legibility demands of rushed administrators.¹⁸

The metric tide has, among many other things, added another recursive layer to the Matthew effect. Circa 1968, the effect hinged on scholars’ memories—on what amounted to an availability heuristic repeated across the academic system. The quantified proxies of the h-index and Journal Impact Factor focalize—to borrow Merton’s verb—visibility and attention. The scores themselves act as direct signals to fellow scholars. And they’re also fed into ranking algorithms that power, for example, first-page Google Scholar results.¹⁹ Thus to every one who has will more be given.

The Handoff to Computational Probability

The example of Google Scholar is half the way to Deep Research–style AI summary. The algorithmic logic that generates the ten blue citations is a closely guarded secret. Even the Google engineers who manage the complex recommendation engine are unlikely, as individuals, to have a full grasp on how it works. Old-school Matthew effects, by contrast, could be traced and measured; the task was to follow the bibliographic paper trail left by humans.

The AI-driven summary-and-cite products are flat-out inscrutable. They’re all using foundation models to power their reference-studded summaries—even the scholarly publishers, who use the big AI models to mine their proprietary databases.²⁰ Anthropic and Google and OpenAI aren’t inviting outsiders to audit their LLMs, of course, but it wouldn’t much matter if they did. At the level of any particular output, it is near-impossible to look under the hood—to scrutinize the matrix of probabilistic calculations that result in well-ordered text. Nor is it possible to glean a model’s actual “sources”—a word that, within the vast oceans of training data and retrieved information, has no human-legible analogue.

All AI models, one way or another, use the past to make probabilistic predictions about the future. In the deep research case, the tools—when they turn to the academic literature—inherit the whole sordid legacy of patterned exclusion. With each probabilistic bet on which literature to surface, they are likely to re-inscribe those legacies. Yet we won’t have any easy way to know.

The pre-digital Matthew effect helped to stratify science and scholarship—to reproduce an academic “class structure” through visibility and all its attendant rewards. The system’s structural inequalities—around race and gender, the English language, and global center-periphery dynamics—have their own roots and causes. But they’re all propped up by dynamics of cumulative advantage.

Silicon Valley and the big commercial publishers are likely, in the coming years, to fight over who has rights to mine the scholarly record. The publishers are re-fashioning themselves as data-analytics companies, and they think they’re sitting on a copyrighted gold mine. There will be drama, including lawsuits, licensing deals, and (perhaps) acquisitions.²¹

In some ways all the deck-shuffling to come is a side show. The bigger story is the embrace of foundation models as knowledge distillers by both industries, the one chasing profit and the other appallingly lucrative. Of all the hyped promises of AI, LLMs’ knack for summary is disarmingly impressive. The summaries are pretty good, they’re easy to generate, and they come with citations. Yes, the models hallucinate, and that’s a problem in itself. But the Matthew effect dynamics traced here depend, ironically, on *real *references—and there are lots of them.

The key point is that summary-and-cite AI tools are performative. Since visibility begets visibility, they help enact the world they purport to merely describe. It’s true, as OpenAI puts it, that the “ability to synthesize knowledge is a prerequisite for creating new knowledge.”²² This task was a vexed one already, when humans were doing the summarizing. If we hand off the work to computational probability and the profit motive, we may reproduce, or perhaps worsen, an academic “class structure” with all of its accumulated—and now hidden—injustices.

Bibliography

“A New Way of Exploring Dimensions Data.” Dimensions: A Digital Science Solution, accessed November 9, 2025. https://www.dimensions.ai/discover-dimensions-ai-assistant.

Barlo, Daniele, Chiara Valentin, Fariba Karimi, Luis Galárraga, Gonzalo G. Méndez, Lisette Espín-Noboa “Whose Name Comes Up? Auditing LLM-Based Scholar Recommendations.” arXiv preprint, September 10, 2025. https://doi.org/10.48550/arXiv.2506.00074.

Bazerman, Charles. Shaping Written Knowledge: The Genre and Activity of the Experimental Article in Science. University of Wisconsin Press, 1988.

Bellen, Simon van, Juan Pablo Alperin, and Vincent Larivière. “Scholarly Publishing’s Hidden Diversity: How Exclusive Databases Sustain the Oligopoly of Academic Publishers.” PLoS One 20, no. 6 (2025): e0327015. https://doi.org/10.1371/journal.pone.0327015.

Blair, Ann M. Too Much to Know: Managing Scholarly Information before the Modern Age. Yale University Press, 2010.

“Claude Takes Research to New Places.” Claude Blog, April 15, 2025. https://www.claude.com/blog/research.

Demeter, Márton. Academic Knowledge Production and the Global South: Questioning Inequality and Under-Representation. Palgrave Macmillan, 2020.

Dempsey, Lorcan. “Libraries, AI and Training Collections.” Against the Grain, July 5, 2024. https://www.charleston-hub.com/2024/07/libraries-ai-and-training-collections.

DiPrete, Thomas A., and Gregory M. Eirich. “Cumulative Advantage as a Mechanism for Inequality: A Review of Theoretical and Empirical Developments.” Annual Review of Sociology 32, no. 1 (2006): 271–97. https://doi.org/10.1146/annurev.soc.32.061604.123127.

“Elsevier Lunches ScienceDirect AI to Transform Research With Rapid Mission-Critical Insights From Trusted Content.” Elsevier, March 12, 2025. https://www.elsevier.com/about/press-releases/elsevier-launches-sciencedirect-ai-to-transform-research-with-rapid-mission.

Fares, Rudolf, and Tobias Wolbring. “Matthew Effects in Science and the Serial Diffusion of Ideas: Testing Old Ideas with New Methods.” Quantitative Science Studies 2, no. 2 (2021): 505–26. https://doi.org/10.1162/qss_a_00129.

Fyfe, Aileen. “The Surprising History of Abstracts.” Nature Reviews Physics 6, no. 8 (2024): 464–65. https://doi.org/10.1038/s42254–024–00741–0.

“Gemini Deep Research.” Gemini, accessed November 9, 2025. https://gemini.google/overview/deep-research.

He, Jiangen. “Who Gets Cited? Gender- and Majority-Bias in LLM-Driven Reference Selection.” arXiv preprint, August 2, 2025. https://doi.org/10.48550/arXiv.2508.02740.

“Introducing Deep Research.” OpenAI Release, February 2, 2025. https://openai.com/index/introducing-deep-research.

Howell, Anthony, Jieshu Wang, Luyu Du, Julia Melkers, Varshil Shah. “Prestige over Merit: An Adapted Audit of LLM Bias in Peer Review.” arXiv preprint, September 18, 2025. https://doi.org/10.48550/arXiv.2509.15122.

King, Molly M., Carl T. Bergstrom, Shelley J. Correll, Jennifer Jacquet, and Jevin D. West.“Men Set Their Own Cites High: Gender and Self-Citation Across Fields and Over Time.” Socius 3 (2017): 2378023117738903. https://doi.org/10.1177/2378023117738903.

Kozlowski, Diego, Thema Monroe‐White, Vincent Larivière, and Cassidy R. Sugimoto. “The Howard‐Harvard Effect: Institutional Reproduction of Intersectional Inequalities.” Journal of the Association for Information Science and Technology 75, no. 8 (2024): 869–82. https://doi.org/10.1002/asi.24931.

“Latest Release: Deep Research (August 2025).” Elsevier, accessed November 9, 2025. https://www.elsevier.com/products/scopus/scopus-ai/product-release-notes.

Ma, Lai. “Generative AI for Academic Publishing? Some Thoughts About Epistemic Diversity and the Pursuit of Truth.” KULA 7, no. 1 (2024): 1–5. https://doi.org/10.18357/kula.287.

Ma, Lai. “Information, Platformized.” Journal of the Association for Information Science and Technology 74, no. 2 (2023): 273–82. https://doi.org/10.1002/asi.24713.

Merton, Robert K. “The Matthew Effect in Science, II: Cumulative Advantage and the Symbolism of Intellectual Property.” Isis 79, no. 4 (1973): 606–23. https://doi.org/10.1086/354848.

Merton, Robert K. “The Matthew Effect in Science.” Science 159, no. 3810 (1968): 56–63. https://doi.org/10.1126/science.159.3810.56.

Pardo-Guerra, Juan Pablo. The Quantified Scholar: How Research Evaluations Transformed the British Social Sciences. Columbia University Press, 2022.

Perplexity Team. “Introducing Perplexity Deep Research.” Perplexity Blog, February 14, 2025. https://www.perplexity.ai/hub/blog/introducing-perplexity-deep-research.

Pooley, Jefferson. “Surveillance Publishing,” Journal of Electronic Publishing 25, no. 1 (2022): 39–49. https://doi.org/10.3998/jep.1874.

Pooley, Jefferson. “Large Language Publishing: The Scholarly Publishing Oligopoly’s Bet on AI.” KULA 7, no. 1 (2024): 1–11. https://doi.org/10.18357/kula.291.

Rajkó, Andrea, Csilla Herendy, Manuel Goyanes, and Márton Demeter. “The Matilda Effect in Communication Research: The Effects of Gender and Geography on Usage and Citations Across 11 Countries.” Communication Research 52, no. 2 (2025): 209–32. https://doi.org/10.1177/00936502221124389.

Restivo, Sal. “The Theory Landscape in Science Studies: Sociological Traditions.” In Handbook of Science and Technology Studies, rev. ed., edited by Sheila Jasanoff, Gerald E. Markle, James C. Petersen, and Trevor Pinch, 95–110. Sage, 1995. https://doi.org/10.4135/9781412990127.n5.

Rigney, Daniel. The Matthew Effect: How Advantage Begets Further Advantage. Columbia University Press, 2010.

Ross-Hellauer, Tony, Stefan Reichmann, Nicki Lisa Cole, Angela Fessl, Thomas Klebel, and Nancy Pontika. “Dynamics of Cumulative Advantage and Threats to Equity in Open Science: A Scoping Review.” Royal Society Open Science 9, no. 1 (2022): 211032. https://doi.org/10.1098/rsos.211032.

Rossiter, Margaret W. “The Matilda Effect in Science.” Social Studies of Science 23, no. 2 (1993): 325–41. https://doi.org/10.1177/030631293023002004.

Rubin, Hannah, and Cailin O’Connor,. “Discrimination and Collaboration in Science.” Philosophy of Science 85, no. 3 (2018): 380–402. https://doi.org/10.1086/697744.

“Scopus AI: Think Bigger. Move Faster. Act with Confidence.” Elsevier, accessed November 9, 2025. https://www.elsevier.com/products/scopus/scopus-a.

Small, Henry G. “Cited Documents as Concept Symbols.” Social Studies of Science 8, no. 3 (1978): 327–40. https://doi.org/10.1177/030631277800800305.

Spataro, Jared. “Introducing Researcher and Analyst in Microsoft 365 Copilot.” Microsoft 365 Blog, March 25, 2025. https://www.microsoft.com/en-us/microsoft-365/blog/2025/03/25/introducing-researcher-and-analyst-in-microsoft-365-copilot.

Tian, Yifan, et al. “Who Gets Recommended? Investigating Gender, Race, and Country Disparities in Paper Recommendations from Large Language Models.” arXiv preprint, December 31, 2024. https://doi.org/10.48550/arXiv.2501.00367.

“Web of Science Research Assistant.” Clarivate, accessed November 9, 2025. https://clarivate.com/academia-government/scientific-and-academic-research/research-discovery-and-referencing/web-of-science/web-of-science-research-assistant.

Zuckerman, Harriet. “Nobel Laureates in the United States: A Sociological Study of Scientific Collaboration.” PhD diss., Columbia University, 1965.

Zuckerman, Harriet. “Nobel Laureates in Science: Patterns of Productivity, Collaboration, and Authorship.” American Sociological Review 32, no. 3 (1967): 391–403. https://doi.org/10.2307/2091086.

Zuckerman, Harriet. Scientific Elite: Nobel Laureates in the United States. Transaction Publishers, 1977.

Zuckerman, Harriet E. “The Matthew Effect Writ Large and Larger: A Study in Sociological Semantics.“ In Concepts and the Social Order: Robert K. Merton and the Future of Sociology, edtied by Yehuda Elkana, György Lissauer, and Andras Szigeti, 121–64. Central European Press, 2011.

Delta Think’s Dan Pollock and Heather Staines, with data on APC prices and inflation:

Overall, OA prices are increasing below inflation. The data suggest that authors continue to get (modestly) increasing value for money for their OA fees.

APCS, of course, have never had any meaningful relationship with “costs”—not ever, not since their inception in the early 2000s. So the idea of increasing “value” is a bit off-key—the unscalable wall for the vast majority of the world’s scholars is a few inches shorter but still comically high.

However, the headlines are skewed by the large number of hybrid journals. When looking only at fully OA (“gold”) journals, prices are rising slightly faster than inflation. Our market sizing data suggest that significantly more articles are published in fully OA journals than in hybrid, so the inflationary effects will be more prevalent than the real-terms cost savings for hybrid. Although it’s worth noting that fully OA prices are cheaper than those for hybrid, so they are growing from a lower base.

“lower base.”

As discussions continue about the increasing cost and affordability of OA, it’s important to keep an eye on the nuances of real terms effects.

“Real terms” effects are basically irrelevant in the discussions around cost and affordability of closed-authorship OA: Standard APCs, this year or last year or next year, are so unaffordable that tinkering around the inflation-edges is meaningless.

I missed this excellent 2024 UKSG Insights piece, by Tasha Mellins-Cohen. In a studiously neutral, mapping-the-landscape fashion, Mellins-Cohen classifies the torrent of OA models into five core types. From the abstract (with the types in bold):

The proliferation of open access (OA) business models has been rapid, presenting challenges for stakeholders in communicating and working effectively with one another. […] It presents five core types, each with distinct characteristics and implications for funding, equity, and implementation. Operating at the level of individual pieces of content, transactional models expose authors to the financial implications of their decisions to make content OA; they often must pay out of their own funds. Driven by negotiations between libraries or consortia and publishers, bundle models typically operate at the level of the institution. In many cases, authors are not exposed to the financial aspects of their OA decisions. In cooperative models, the community acts in concert to spread the cost of OA as widely as possible. These models are typically initiated or driven by publishers. Sponsored models generally require a single actor to take responsibility for OA funding within a specific remit or title. Alternative OA models offer either immediate access to something other than the version of record (alternative OA: repository), or delayed access to the version of record (alternative OA: delayed).

The essay is great—thorough and taxonomically satisfying, especially if you, like me, have completionist tendencies. The fifth and last type, “alternative,” is a grab-bag of leftovers, but the other four categories are genuinely useful, especially as Mellins-Cohen—paragraph-by-paragraph—fits in existing, branded “models” within the bigger tents that she’s named. It’s real service to the community.