Scientific Institutions Through History

The framing under test is one almost nobody states explicitly and almost everybody assumes: that the institutions of science evolved toward better knowledge production. Journals, referees, academies, laboratories, doctorates and research teams are told as a sequence of improvements, each solving a problem the last one exposed. This page assembles the dates and the numbers, and the sequence does not survive them.

Established The load-bearing measurement is a rate that did not move. Across 6,665 Royal Society manuscript records between 1865 and 1965, the rejection rate held between 8.5% and 13.5% for the entire century, settling near 10–11% after the Second World War. Over the same span submissions roughly quadrupled, the referee corps tripled, participation among fellows rose from about 8% to about 30%, workload inequality climbed from a Gini of about 0.24 to nearly 0.45, and non-member submissions rose fivefold after 1890 until they vastly outnumbered fellows' papers. Everything about the system changed except the fraction of papers it turned away.

Established And the vocabulary is younger than almost anyone believes. The phrase “peer review” entered the scientific literature in 1965, in Science, describing NIH grant procedures; American newspapers of the 1960s and 1970s used it mainly for Medicare and Medicaid compliance review of medical practice; the New England Journal of Medicine used it in 1969 for Medicaid monitoring rather than for manuscripts; and it was applied to journal refereeing from about 1968. A term coined for American healthcare utilisation review is routinely projected back onto 1665.

Frontier What the page lands is that the epistemic machinery of science accreted and was rationalised afterwards. The journal began as a secretary's private correspondence venture; refereeing began as a committee convenience inside a members' club; the German research university was attributed to a man who did not design it, ninety years after the fact; the laboratory arrived sixty years after the university that supposedly invented it, to complaints that the institutes were “knowledge factories”; team science is visible only in retrospect through authorship counts. The two clearly designed cases in the record are state-founding acts.

This page is the historical institution and it stops in 1975. How peer review functions today as an allocation rule, what registered reports and replication policy do, and how research-integrity machinery performs all belong to Scientific Governance Models, which this page hands off to at Baldwin's 1975 NSF oversight hearings rather than summarising. Invention and its measurement belong to Innovation History; the epistemology belongs to Philosophy of Science and Scientific Revolutions.

Established The founding dates are solid and they come from the institutions themselves, which is worth saying out loud. The Royal Society dates its first meeting to 28 November 1660, following a lecture by Christopher Wren, then Gresham Professor of Astronomy, and growing out of informal gatherings of London and Oxford intellectuals; it was chartered in 1662 as “The Royal Society of London for Improving Natural Knowledge”, with the motto Nullius in verba adopted in that First Charter. The Society glosses the motto as “take nobody's word for it” and as a commitment “to withstand the domination of authority and to verify all statements by an appeal to facts determined by experiment”. That gloss is the Society writing its own history, and it is marked as such: the dates are reliable, the interpretation is the institution's own.

Frontier The journal is a private venture before it is an institution, and everything downstream depends on that distinction. Philosophical Transactions launched in March 1665, founded and edited by Henry Oldenburg, the Society's first Secretary, “who acted as publisher and editor”. Its publisher describes it as “the world's first and longest-running scientific journal” and describes Oldenburg as having spun “his wide-ranging network of European natural-philosophical contacts, the activity of the Royal Society, and his skills as a linguist and scientific editor into a new form of print intended to promote the enterprise of early modern science”. Frontier The “world's first” claim is the publisher's, about its own journal. The point that matters is not the primacy: Philosophical Transactions was a secretary's correspondence network converted into print, not an instrument the Society designed for quality control.

Established Six years later a second architecture appears, and it is the opposite one. The Académie des sciences was created by Colbert in 1666, first session 22 December 1666 in the King's library in Paris, “dedicated to the development of science and advising the government in this field”. On 20 January 1699 Louis XIV gave it its first regulations and his protection; the reorganised body sat at the Louvre with 70 members and 80 correspondents, appointed by royal authority. The Convention abolished all academies on 8 August 1793; the body was reconstituted in 1795 within the National Institute of Sciences and Arts and formally restored in 1816. Frontier Two founding institutions six years apart with opposite architectures — a self-governing gentlemen's society with a private journal, and a state organ with royally appointed pensioned members whose stated function includes advising government — and neither was selected over the other on evidence about knowledge-production quality. Both persist. A directional story has to explain a founding fork that never resolved.

Established Refereeing arrives 172 years after the journal, and the Society says so itself: “Refereeing of scientific papers commenced from 1832.” What began then was a Committee-of-Papers process inside a members' club. Frontier The definitive scholarly treatment of the prehistory — Moxham and Fyfe on the Royal Society and the prehistory of peer review, 1665–1965, in The Historical Journal (2018) — was not obtained for this pass: the repository copy is robots-blocked and the publisher page paywalled. Its existence and title are verified; its argument is not characterised here and it is not quoted.

Frontier What was obtained carries the argument on its own, and it is the strongest quantitative series in the history of scientific institutions. Fyfe, Squazzoni, Torny and Dondio analysed 6,665 manuscript records across the Royal Society's journals from 1865 to 1965. Submissions roughly quadrupled between 1865 and 1935, about 2% a year. Outcomes: 787 papers (11.8%) in Philosophical Transactions, 5,214 (78.2%) in Proceedings, 665 (10%) unpublished. Referee reports solicited rose from about 100 a year in the late 1800s to about 400 by the 1930s; referees from about 50 fellows a year in the 1870s to about 140 by the 1930s; participation from about 8% of fellows in the 1880s to about 30% by the 1930s. Workload inequality rose sharply: the Gini index from about 0.24 to nearly 0.45 since 1900, with the top 20% of referees writing about 37% of reports in the late nineteenth century and about 50% by mid-century, and their individual load nearly doubling from 3.1 reports a year in 1865 to 6.5 in 1965. And the author population changed character: about two-thirds of nineteenth-century submissions came from fellows, after which non-member submissions rose fivefold and eventually vastly outnumbered them.

Frontier The rejection rate through all of that held between 8.5% and 13.5%, settling near 10–11% after 1945. The reading is this page's and the numbers are theirs: a rate flat across a century in which submissions quadrupled, the referee corps tripled and the author population changed completely is a throughput constant, not a filter being tuned toward better knowledge production. Established The authors' own conclusion points the same way — today's complaints about submission growth, reviewer fatigue and unequal workload “are historically rooted phenomena, not modern inventions”, and embedding journals in scholarly communities gave accountability but “proved difficult to scale”.

Established The term itself is a Cold War American import, and Baldwin's documentation of it is the single best corrective in this brief. The phrase came out of medical and grant-making contexts rather than journals. Its first use in the scientific literature is in Science in 1965, by Joseph D. Cooper, describing NIH procedures: project grantees “were asked to make assessments…in which they were intimately involved both as grantees and as members of peer review groups”. The New England Journal of Medicine used it in 1969 for Medicaid monitoring. Franz Ingelfinger, editing the NEJM, was among the first to apply it to journal refereeing, writing in 1968 that “to be accepted the article must pass peer review”. And the crystallising moment was the 1975 NSF Peer Review Special Oversight Hearings, with Representatives John Conlan and Robert Bauman and Senator William Proxmire attacking NSF funding decisions — criticising, notably, that referee opinions were not weighted heavily enough, rather than the concept itself. Baldwin's argument is that 1975 is when peer review became central to scientific practice in American discourse.

Established The complementary thesis is retrievable and its evidence is not. Csiszar, in Nature (2016), argues that “pivotal moments in the history of academic refereeing have occurred at times when the public status of science was being renegotiated”. Only the article's metadata, thesis line and reference list were retrievable; the full text is paywalled, and no date or case on this page is attributed to it.

Established The most celebrated institutional design in the history of science is, on the best-sourced account available here, a retrospective attribution. Mitchell Ash argues that “the narrow linkage of 'the' German research university model to the name and ideas of Wilhelm von Humboldt is a myth, a tradition invented around 1900”, and that because Humboldt's own writings on university education “remained unpublished at first”, the claim that his ideas spread worldwide in the nineteenth century is “in a literal sense untrue”. Citing Paletschek, he notes that Humboldt was known in his own century as a founder of modern language studies, not as a university reformer. The University of Berlin was founded in 1810, but the Philosophical Faculty gained independent and co-equal status with the traditional faculties “not under Humboldt, but ten years later”. The seminar, the actual innovation, predates Berlin: it “originated in Göttingen before 1800”, where Humboldt studied and which he used as a model. And the laboratory arrived far later still: the natural-scientific and medical research institutes with their own lecture halls and teaching laboratories “emerged in the 1860s and 1870s”, and “contemporaries denounced them as 'knowledge factories', and they would surely have astounded Humboldt himself”. What Humboldt did do was the Prussian school reform establishing the Gymnasium certificate as the university entrance requirement. Frontier These are Ash's claims with his page references, and the thesis is contested within the field; it is carried here as a scholarly position rather than as settled fact.

Established Two things this brief needs are recorded as gaps rather than filled from memory, and the absence is itself a finding. Liebig's teaching laboratory at Giessen, conventionally dated 1824–25 and conventionally credited with turning research training into a reproducible institutional form, could not be sourced: the best candidate scholarly chapter returned metadata only. The emergence of the PhD, including the first American doctorate conventionally placed at Yale in 1861, could not be sourced either: the university's own graduate-school history page returned a 404 and the remaining results were blogs and content farms. No dates, student counts or firsts are asserted here for either. What can be said on the evidence in hand is Ash's: the research-institute laboratory in Germany is an 1860s–70s phenomenon, half a century after Berlin's founding, and contemporaries called the institutes knowledge factories. That places the laboratory as a late and contested arrival without requiring the Giessen story at all.

Established The growth of the literature has three regimes, not one exponential. Bornmann and Mutz ran segmented regression on Web of Science source items and cited references back to the mid-1600s and found each phase roughly tripling the previous rate: under 1% a year to the middle of the eighteenth century; 2–3% a year to the period between the world wars; and 8–9% a year to 2012. Frontier As a segmentation of a database this is solid; as a claim about science it inherits everything the database inherits, including changing coverage.

Established And the unit of production changed, decisively and measurably. Wuchty, Jones and Uzzi analysed 19.9 million papers over five decades and 2.1 million patents: science and engineering from 1955, social sciences from 1956, arts and humanities from 1975, US patents from 1975. Mean team size in science and engineering rose from 1.9 to 3.5 authors over 45 years; team-authored social-science papers from 17.5% in 1955 to 51.5% in 2000; patents from 1.7 to 2.3 inventors; while in arts and humanities solo authors still produced over 90% of papers. Relative team impact in science and engineering ran 1.7 times more citations in 1955 and 2.1 times by 2000, and a team-authored science and engineering paper is “currently 6.3 times more likely than a solo-authored paper to receive at least 1000 citations”. Speculative Citation advantage is uptake, not correctness, and the paper makes no claim that team science is better science.

Frontier The limit case is documented and it is a statement about credit rather than about size. The first joint ATLAS–CMS paper, in Physical Review Letters on 14 May 2015, carries 5,154 authors: 33 pages, of which 9 are text and data and 24 are the author list. The previous threshold was a 2008 CMS paper, the first past 3,000. Both collaborations list all contributors because they “believe there is no fair way to split credit for the project”. Frontier An institution whose credit-allocation mechanism its own participants describe as having no fair solution has reached a structural limit, not a stylistic one.

Established The ethos was stated in 1942 and first seriously tested in the 2010s, which is the seam this whole subject turns on. Merton's four institutional imperatives, in his words: universalism, “Truth-claims, whatever their source, are to be subjected to preestablished impersonal criteria: consonant with observation and with previously confirmed knowledge”; communism, “The substantive findings of science are a product of social collaboration and are assigned to the community…a common heritage in which the equity of the individual producer is severely limited”; disinterestedness, “a distinctive pattern of institutional control of a wide range of motives” supported by “the public and testable character of science” and explicitly not a personal moral quality; and organized scepticism, “the temporary suspension of judgment and the detached scrutiny of beliefs in terms of empirical and logical criteria”.

Frontier The only direct measurement available finds the norms endorsed and not followed. Bray and von Storch surveyed climate scientists — 286 valid responses from 4,491 solicited, a 7% response rate, on seven-point items for each norm and its counter-norm. Their finding, in their own words, is that Merton's CUDOS “remain the overall guiding moral principles” but “are not fully endorsed or present in the conduct of climate scientists: there is a tendency to withhold results until publication, there is the intention of maintaining property rights, there is external influence defining research and the tendency to assign the significance of authored work according to the status of the author rather than content of the paper”. About 86% were willing to accept peer-reviewed work on universalism; organised scepticism was strongly endorsed on all four measures; disinterestedness was mixed, with considerable acknowledgement that research aligned with funding. Frontier One field, 286 responses, a 7% response rate and self-report. It does not generalise on its own, and it is the only direct measurement there is. Established The earlier survey of NIH-funded scientists it compares against could not be fetched for this pass — the record returned a server error — and is cited here only as reported by Bray and von Storch. Established Seventy years separate the statement of the ethos from the first serious attempt to test it.

Frontier Was any of it designed? The evidence assembled here answers mostly no, and the exceptions are instructive. Accreted: the journal as Oldenburg's private correspondence venture; refereeing as a Committee-of-Papers convenience from 1832 that stabilised at about a 10% rejection rate for a century; “peer review” as a phrase imported from American healthcare utilisation review and retrofitted; the German research university as a model attributed to Humboldt around 1900, ninety years after Berlin's founding, whose actual components arrived respectively before it and sixty years after it; team science as an emergent reorganisation visible only in retrospect through authorship counts.

Established Designed: the Paris academy in 1666, a state instrument with appointed pensioned members and an explicit advisory function. And Vannevar Bush's Science, The Endless Frontier, delivered to President Truman on 5 July 1945, which proposed a federal agency to fund basic research at universities, argued that basic research drives applied advance, and argued that researchers must be protected from pressure toward immediate application — the report that led to the National Science Foundation. Frontier The copy fetched is hosted by the agency the report proposed, and the fetch returned paraphrase rather than verbatim text, so nothing is quoted from it here. Frontier The honest summary is that the two clearly designed cases are both state-founding acts, and that the epistemic machinery — journals, refereeing, laboratories, doctoral training, teams — accreted. That is an interpretation across the assembled cases and is flagged as one.

Frontier Is a flat rejection rate evidence of anything, or is it an artefact of one society? This is the sharpest open question the page raises and it has one data series behind it. A century of Royal Society records is a remarkable resource and it is a single institution with a specific membership structure, a two-journal outlet system and a Committee of Papers. A comparable series for a second society or a second national tradition would either establish the throughput constant as a property of learned-society refereeing or localise it to one club. No such series has been assembled.

Frontier Did refereeing ever function as a quality filter, or as a workload allocator? The Fyfe series shows the referee corps and the number of reports growing roughly with submissions while the rejection fraction stays fixed and the workload concentrates in a small group. That is what an institution scaling under load looks like, not what a tightening standard looks like. Speculative The alternative reading — that the constant rate reflects a stable underlying quality distribution being correctly sorted — is coherent and would predict the same numbers, and nothing in the series distinguishes them.

Speculative Do scientists follow the Mertonian norms outside climate science? One survey, one field, 286 responses, 7% response, self-report. The comparison study on NIH-funded scientists could not be obtained. The honest position is that the ethos most often quoted as describing how science works has been tested approximately once, and did not come through cleanly.

Frontier Has credit allocation actually broken, or only become unusual? The collaborations' own statement — that there is no fair way to split credit — is a claim by the people best placed to make it, and 24 of 33 pages given to the author list is a concrete cost. Speculative Whether contribution taxonomies, role statements or non-authorship credit mechanisms resolve it is a live design question and belongs, from 1975 onward, to Scientific Governance Models.

Handwave And the framing's own claim — that institutions evolved toward better knowledge production — is asserted far more often than it is tested. It requires a measure of knowledge-production quality that varies over the period, and no such measure appears anywhere in this literature. The available series measure throughput, participation, workload, growth rate, team size and citation uptake. None of them is quality, and the directional claim is doing its work by assertion.

Frontier The binding constraint in this subject is archival access, and it shaped this page visibly. The definitive treatment of the Royal Society's refereeing prehistory is robots-blocked in its repository copy and paywalled at the publisher, so it is named and not used. Csiszar's argument is retrievable as a thesis line and not as evidence. The comparison norms survey returned a server error. Three of the sources this brief most needed are recorded as not obtained rather than paraphrased from memory, and that is the honest version of the constraint.

Frontier The second bottleneck is that the institutions are the main historians of themselves. The founding dates, the charter, the motto's gloss, the 1832 refereeing claim and the “world's first journal” claim all come from the Royal Society and its publisher; the Paris chronology comes from the Académie; the Bush report is hosted by the agency it proposed. These sources are reliable for their own dates and are interested parties on every interpretive question, and this page marks which is which at each point.

Frontier The third is that there is one long quantitative series and it covers one society. Everything quantitative about the history of refereeing on this page — the flat rejection rate, the Gini, the participation share, the fivefold rise in non-member submissions — comes from 6,665 records at the Royal Society between 1865 and 1965. The finding is strong and its external validity is entirely untested, because the editorial archives that would supply a second series are held privately by publishers and societies and are not systematically open.

Established The fourth is a genuine hole in the record as it reached this page. Liebig's Giessen laboratory and the emergence of the doctorate are both standard components of any account of how research training became institutional, and neither could be sourced: the candidate scholarly chapter returned metadata only and the relevant university history page returned a 404, leaving blogs and content farms. A documented absence is a finding rather than an embarrassment, and it is typed as a requirement in section 13 instead of being written from general knowledge.

Frontier And the fifth is conceptual: nobody has proposed a measure of knowledge-production quality that varies across three centuries. Every quantity in this brief measures throughput, participation, growth, team composition or citation uptake. Without a quality measure the framing's directional claim cannot be tested even in principle, which is why it is flagged speculative wherever it appears rather than merely unsupported.

Established Nothing on this map produces a result this brief waits on, and no typed depends-on edge is claimed. It waits on two access conditions, both recorded as typed requirements below: editorial archives for a second learned society over a comparable span, which would test whether the flat rejection rate is a property of refereeing or of one club; and obtainable scholarship on the Giessen teaching laboratory and the emergence of the doctorate, which is the one substantive hole in this page's coverage.

Established The boundaries are stated so nothing is argued twice. Scientific Governance Models owns peer review as an allocation rule, registered reports, replication policy, integrity machinery and funding lotteries; the handoff point is explicit and it is the 1975 NSF hearings. Innovation History owns invention, patents and the measurement of innovation; where this page reaches the laboratory as an engine of technical advance it defers. Bell Labs and DARPA own twentieth-century industrial and agency research organisations; this page stops at Bush in 1945 and the founding of the NSF, and does not describe what those organisations did. Philosophy of Science and Scientific Revolutions own the epistemology; Merton's norms sit here rather than there, and the demarcation page points at this one.

Established Sources named as not obtained, so their absence does not read as an oversight: Moxham and Fyfe (2018), robots-blocked and paywalled, existence and title verified and argument not characterised; Csiszar's Nature piece, metadata and thesis line only; Anderson, Ronning, De Vries and Martinson's 2010 norms survey, whose record returned a server error and which is cited only as reported by Bray and von Storch; scholarly sources on Giessen and on the emergence of the PhD; and verbatim text of Bush's 1945 report, the fetched copy having returned paraphrase.

Frontier The single highest-value study in this subject is a second manuscript series. The Royal Society records give rejection rate, referee count, participation share and workload distribution across a century. Running the same analysis on the editorial archives of a second learned society — a continental academy, a national society in a different tradition, or a long-lived commercial journal — would establish whether a rejection rate flat between 8.5% and 13.5% is a property of learned-society refereeing or an artefact of one institution. The method is published, the outcome is interpretable either way, and the obstacle is access rather than technique.

Frontier Second: replicate the norms survey outside one field. The only direct measurement of whether scientists follow the Mertonian ethos covers climate science, 286 responses and a 7% response rate. The instrument exists, it is short, and the counter-norm items are the interesting half. Running it across three or four disciplines with different funding structures would test the finding that scientists endorse the norms while acting on the counter-norms, and would say whether disinterestedness tracks funding structure as the climate-science result suggests.

Frontier Third, and this is archival rather than experimental: obtain the two missing histories. Giessen and the emergence of the doctorate are not open research questions; they are well-studied topics whose scholarship this pass could not reach. The work is retrieval, not discovery, and until it is done this page carries a hole where the origin of research training belongs.

Frontier Fourth: test the throughput-constant reading directly against the quality-sorting reading. Both predict a flat rejection rate. They differ on what happens to the marginal paper: a throughput constraint implies the acceptance threshold drifts with submission volume, while stable sorting implies it does not. Referee reports with scores, held in the same archives as the outcomes, would separate them — and the Royal Society series has the outcomes without a published analysis of report content over time.

Speculative Fifth: measure what hyperauthorship does to credit in practice. The ATLAS and CMS collaborations state there is no fair way to split credit, and the consequences are observable — hiring outcomes, prize attribution, citation-based evaluation for individuals from papers with thousands of authors. A cohort study of early-career physicists from large collaborations against comparable small-collaboration cohorts would turn a structural claim into a measurement, and this pass located none.

Frontier The engineering object in this subject is a dataset, and its specification is legible from the one that exists. The Royal Society series required manuscript registers with dated submissions, named referees, report counts and final outcomes, linked across a century and reconciled with two publication outlets. That is what an editorial archive has to contain to be analysable, and it is the reason there is one such series and not five.

Frontier The second engineering fact is that the modern equivalents exist and are not accessible. Every manuscript-tracking system in use since the 1990s holds submission dates, referee identities, report texts, decision sequences and outcomes in structured form. The historical series that took years of archival work is generated automatically as a byproduct of every journal's operations and is held privately. How that data should be governed is Scientific Governance Models's question; that it exists is this page's point.

Frontier The third is that the growth series constrains what any institutional history can claim. A literature growing at 8–9% a year to 2012, after regimes of under 1% and 2–3%, means the twentieth-century record dwarfs everything before it by construction. Any institutional claim computed across the whole span is dominated by its last few decades unless it is explicitly rate-normalised, and most narrative institutional history is not.

Frontier And the fourth is that credit allocation is now an engineering problem with a stated failure mode. At 5,154 authors the author list occupies 24 of 33 pages and the collaborations state there is no fair way to split credit. Contribution taxonomies and role statements are the current answer and they are metadata attached to a byline that was never designed to carry them — a data-model problem rather than an ethical one, though it has ethical consequences.

Within this map: Scientific Governance Models, which owns everything from about 1975 onward — peer review as an allocation rule, registered reports, replication policy, integrity machinery and funding lotteries — and which this page hands off to at Baldwin's NSF oversight hearings rather than summarising; Innovation History, which owns invention and its measurement, including the world's-fair evidence and the patent-series audits; Philosophy of Science, which owns demarcation and points here for Merton's norms and the survey evidence on them; and Scientific Revolutions, which owns the account of change these institutions are supposed to have supported.

Also within it: Scientific Funding Models, whose subject begins where Bush's 1945 report ends; Scientific Advisory Institutions, which is the Paris architecture's direct descendant; Long-Term Institutions, for which a body founded in 1660 and still operating is a rare data point; and Institutional Design, where the accreted-versus-designed question is stated generally.

Outside this map: Bell Labs and DARPA, which own twentieth-century industrial and agency research organisations as method cases. And beyond the site: the history of science and of the book; the sociology of science, which supplies Merton and the norms surveys; scientometrics, which supplies the growth and authorship series; and the history of universities.

Established The interested parties on this page are named at every point where it matters, because in this subject the institution is usually the source. The Royal Society supplies its own founding dates, its own charter, its own gloss on Nullius in verba and its own claim that refereeing commenced from 1832. Its publisher supplies the March 1665 launch and the claim that Philosophical Transactions is the world's first and longest-running scientific journal. The Académie supplies the Paris chronology. The National Science Foundation hosts the report that proposed it. Each is authoritative for its own dates and interested on every interpretive question, and the two are separated at the point of use rather than in a note.

Frontier The load-bearing quantitative source is not in that category and the distinction is worth drawing. The Royal Society manuscript series is funded scholarship about an interested party rather than by one, published peer-reviewed, and its conclusion runs against the improvement story its subject institution tells about itself. That is why the flat rejection rate carries the weight it does on this page.

Frontier The institutional requirement this subject actually has is archival, and it is unmet. One society's editorial records produced the only long series in existence. Publishers and societies hold comparable material, modern manuscript systems generate it automatically, and none of it is systematically available for historical analysis. What is missing is not a method but a deposit arrangement — something like the arrangements that made census and parish records analysable — and no such arrangement exists for the editorial record of science.

Frontier And a chronological point that is itself an institutional finding. The Mertonian ethos was stated in 1942 and the first serious measurement of whether scientists follow it dates from the 2010s. Seventy years of confident institutional self-description preceded any test, and the test that eventually arrived covered one field with a 7% response rate. That gap is the clearest single illustration of how this subject has actually operated.

Frontier Projecting “peer review” onto 1665 is an anachronism of three centuries and it does specific work. It makes a nineteenth-century club procedure and a twentieth-century American grant-administration term look like a founding commitment of modern science, which raises the cost of proposing alternatives: reforming a three-hundred-year-old constitutional feature is a different political act from reforming a fifty-year-old administrative one. The dates matter because the argument from tradition is being made from the wrong date.

Established The credit problem at the limit is stated by the people it falls on. The ATLAS and CMS collaborations list every contributor because they believe there is no fair way to split credit, and the resulting paper gives 24 of its 33 pages to names. Frontier Every individual-level evaluation system in science — hiring, promotion, prizes, citation metrics — assumes a byline that carries attributable credit, and for a growing share of physics that assumption is known by the participants to be false.

Frontier The norms result is uncomfortable in a specific way and deserves stating without softening. Climate scientists endorsed Merton's imperatives as guiding principles while reporting a tendency to withhold results until publication, an intention to maintain property rights, external influence defining research, and a tendency to weight authored work by the author's status rather than the paper's content. Speculative One field, 286 responses, 7% response and self-report — and no reason to think the field surveyed is unusual, which is why the honest response is to run the study elsewhere rather than to explain the result away.

Frontier And the workload finding has an equity dimension the reform debate rarely reaches. Referee workload inequality at the Royal Society rose from a Gini of about 0.24 to nearly 0.45 across the twentieth century, with the top fifth of referees writing about half the reports by mid-century and their individual load doubling. That concentration was already a century-long trend before anyone described reviewer fatigue as a modern crisis, and the paper's own conclusion is that today's complaints are historically rooted rather than new.

Frontier The terminal position is that the institutions of science did not evolve toward better knowledge production; they accreted, were rationalised after the fact, and acquired their improvement narratives — and in one central case their vocabulary — long after the practices themselves. That is not a claim that they are bad or that they do not work. It is a claim about provenance, and the evidence for it is dates and rates rather than argument: a flat rejection rate for a century, a term coined in 1965 for healthcare administration, a university model attributed in 1900 to a man who left in 1810, and a laboratory that arrived sixty years after the university that supposedly invented it.

Frontier The founding fork never resolved and both branches are still running. A self-governing society with a private journal in London in 1660, and a state academy with appointed pensioned members and an advisory function in Paris in 1666. Neither was selected over the other on evidence about knowledge production, both survived revolution and war — the Paris body was abolished outright in 1793 and restored in 1816 — and every modern science system is a mixture of the two. A directional account of institutional evolution has to explain a three-hundred-and-sixty-year unresolved fork, and none of them tries.

Frontier The scaling finding is the one with the longest reach. A century of records shows an institution absorbing a quadrupling of submissions by growing its referee corps, raising participation from 8% to 30% of members, and concentrating workload into a fifth of them — while holding output composition and rejection rate essentially fixed. Embedding a journal in a scholarly community gave it accountability and, in the authors' words, “proved difficult to scale”. Every proposal to fix the present system is a proposal about that trade, and the historical record says the two properties have been in tension for a hundred and sixty years.

Speculative Credit allocation is the institutional feature most likely to break next, and it has already broken in one field. Solo authorship has been displaced across the sciences, team size has risen continuously for seventy years, and at the extreme the participants state plainly that fair attribution is not possible. If the trend continues, individual-level evaluation stops working before collective knowledge production does — which is an institutional failure of a kind the improvement narrative has no place for.

Handwave And the framing's own claim cannot currently be tested at all. “Better knowledge production” requires a quality measure varying over three centuries, and this literature has none: it has throughput, participation, workload, growth rates, team sizes and citation uptake. Until someone proposes a quality measure the directional story is not a weak hypothesis; it is not yet a hypothesis.

These horizons track access to records and the scaling of existing institutions, not discoveries:

• 10 yr: Frontier The tractable development is archival: a second society-length manuscript series would settle the external validity of the only long quantitative finding in this subject, and the obstacle is deposit permission rather than method. Frontier A norms survey outside climate science is equally cheap and equally unclaimed. Speculative Expect the Giessen and doctorate scholarship to be reachable well before either, because the work exists and only the access failed.
• 25 yr: Speculative If literature growth continues near the recent regime, the throughput problem the Royal Society solved by growing its referee corps and concentrating workload has no equivalent solution, because the referee population is drawn from the author population and both scale together. Speculative The plausible split is that allocation moves to rules and the historical model of community-embedded judgement survives only in small fields — a claim about mechanisms that Scientific Governance Models owns the present-day evidence for.
• 50 yr: Speculative The credit question resolves in some direction, because hyperauthorship makes individual evaluation from bylines unworkable and every hiring and funding system currently depends on it. Handwave Which direction — contribution taxonomies, collective credit, or the abandonment of individual attribution in large-collaboration fields — is an institutional prediction with no evidence behind it.
• 100 / 250+ yr: Handwave Beyond useful forecasting, though this subject has better anchors than most: two founding institutions are 365 years old, one of them was abolished and restored, and their two architectures never merged. Handwave Two cases in one civilisation is a history, not a base rate.

• Depends on Nothing on this map. No brief here produces a result this one waits on, and no typed depends-on edge is claimed. The constraints that bind are both access conditions on a documentary record, typed below.
• Requires (not on this map) Editorial archives of a second learned society, opened for a century-length manuscript series. The only long quantitative finding in this subject — a rejection rate flat between 8.5% and 13.5% from 1865 to 1965 while submissions quadrupled, referee participation rose from about 8% to about 30% of fellows and workload inequality climbed from a Gini of about 0.24 to nearly 0.45 — rests on 6,665 records from one institution with one membership structure and two outlets. A comparable series from a continental academy, a different national tradition or a long-lived commercial journal would establish whether the throughput constant is a property of learned-society refereeing or an artefact of one club, and the result is interpretable either way. The method is published and the obstacle is deposit permission: publishers and societies hold the material, modern manuscript-tracking systems generate the equivalent automatically, and none of it is systematically available for historical analysis. And obtainable scholarship on the Giessen teaching laboratory and the emergence of the doctorate: both are standard components of any account of how research training became a reproducible institutional form, and neither could be sourced for this page — the candidate scholarly chapter returned metadata only and the relevant university history page returned a 404, leaving blogs and content farms. This is a retrieval problem rather than a research problem, which is why it is typed as institutional; the page carries the gap openly instead of writing dates and student counts from general knowledge.
• Enables Scientific Governance Models takes the historical institution as given and asks how the present machinery performs, and the handoff is explicit at the 1975 NSF hearings; Scientific Advisory Institutions and Long-Term Institutions both inherit cases from this record. No typed enabling edge is claimed, because what this page supplies to them is context rather than a result: the one strong quantitative finding here covers a single society and its external validity is untested.
• Adjacent The history of science and of the book; the sociology of science, which supplies Merton and the norms surveys; scientometrics, which supplies the growth and authorship series; the history of universities; and on this site Bell Labs, DARPA, Innovation History and Philosophy of Science.

Handwave “Peer review is an ancient feature of science.” It is very widely believed and the dates are not close. The Royal Society dates its own refereeing to 1832, 167 years after its first journal issue. The phrase “peer review” enters the scientific literature in 1965, describing NIH grant procedures, and journal usage in 1968; the NEJM used it in 1969 for Medicaid monitoring rather than for manuscripts. Established Projecting it onto 1665 is an anachronism of three centuries, and the term's origin is American healthcare utilisation review.

Frontier “Refereeing was introduced to improve knowledge quality.” What the century-long series shows is a Committee-of-Papers process inside a members' club with a rejection rate flat between 8.5% and 13.5%, scaling under submission growth by recruiting more referees and concentrating the load rather than by tightening. Speculative The alternative reading — a stable quality distribution being correctly sorted — predicts the same numbers, and nothing in the published series distinguishes them. What can be said is that the improvement story is not what the data show on its face.

Established “The modern research university was designed by Humboldt.” On Ash's account it is “a tradition invented around 1900”: Humboldt's university writings remained unpublished at first, making the claim that they spread in the nineteenth century “in a literal sense untrue”; he was known in his own century as a founder of modern language studies; the Philosophical Faculty gained co-equal status “not under Humboldt, but ten years later”; and the seminar “originated in Göttingen before 1800”. Frontier It is a contested scholarly thesis rather than a settled correction, and it is the best-sourced account this page could obtain.

Established “The laboratory was an early feature of the German model.” The natural-scientific and medical research institutes with their own lecture halls and teaching laboratories “emerged in the 1860s and 1870s” — sixty years after Berlin's founding — and “contemporaries denounced them as 'knowledge factories'”. Frontier The conventional Liebig-at-Giessen origin story is deliberately not told here, because no scholarly source for it could be obtained, and a documented absence is preferable to a confidently recalled date.

Frontier “The Royal Society model won on epistemic grounds.” Two founding institutions six years apart on opposite architectures, one self-governing with a private journal and one a state organ with royally appointed pensioned members and an explicit advisory function. Both persist; the Paris body was abolished in 1793 and restored in 1816. Speculative No selection event on knowledge-production quality appears anywhere in the record, and every modern national science system is a mixture of the two.

Speculative “Team science produces better knowledge.” What was measured is citation advantage: 1.7 times more citations for team-authored science and engineering papers in 1955, 2.1 times by 2000, and 6.3 times the likelihood of reaching a thousand citations. Established That is uptake, not correctness, and the paper reporting it makes no such claim. The measured facts are that team size rose from 1.9 to 3.5 authors over 45 years and that arts and humanities remained over 90% solo.

Frontier “Scientists follow the Mertonian norms.” The only direct measurement — one field, 286 responses, a 7% response rate, self-report — found CUDOS endorsed as guiding principles but “not fully endorsed or present in the conduct” of those surveyed, with results withheld until publication, property rights asserted, funding shaping topics and status shaping credit. Established And Merton stated the ethos in 1942: seventy years separate the statement from the first serious test.

Established “The growth of science is a single exponential.” Segmented regression on Web of Science source items and cited references back to the mid-1600s finds three regimes, each roughly tripling the previous rate: under 1% a year to the mid-eighteenth century, 2–3% to the interwar period, and 8–9% to 2012. Frontier As a segmentation of a database it is solid; as a claim about science it inherits the database's changing coverage.

Frontier “Credit allocation scales.” At 5,154 authors the ATLAS and CMS collaborations state there is “no fair way to split credit for the project”, and the author list occupies 24 of the paper's 33 pages. Established That is one extreme case and it is the participants' own assessment of their own institution, which is the strongest form this kind of evidence takes.

Speculative “The epistemic machinery of science was designed.” The two clearly designed institutions in this record are state-founding acts — the Paris academy in 1666 and the National Science Foundation via Bush's report delivered to Truman on 5 July 1945. Frontier The journal, the referee, the laboratory, the doctorate and the research team accreted, and the interpretation across those cases is this page's rather than any single source's.

Handwave And the framing itself: “institutions evolved toward better knowledge production” is a directional claim with no measure attached to it. Every series available measures throughput, participation, workload, growth rate, team size or citation uptake. None of them is quality, and until someone proposes a quality measure that varies across three centuries the claim cannot be tested even in principle. Frontier What the record does show is accretion, retrospective rationalisation, and in one central case a vocabulary borrowed from healthcare administration three hundred years after the practice it names.