Scientific Revolutions

Thomas Kuhn's The Structure of Scientific Revolutions (1962) proposed that science alternates between long stretches of puzzle-solving inside an accepted framework and short episodes in which the framework is replaced wholesale. The book gave English the phrase “paradigm shift”, and it is the single most cited work of twentieth-century philosophy of science. The framing under test on this page is the one the phrase carries in ordinary use: that science advances through periodic paradigm shifts, and that the Kuhnian account describes how.

This brief is a merged slot, and the merge is worth stating on the page because it is a worked example of the brief's own subject. The topic map carried X-11 Scientific Revolutions and X-25 Scientific Paradigm Shifts as two separate commissions. They are not two subjects; they are the same Kuhnian brief written twice under two names that entered English from the same book. X-25 now cross-lists here, and this page is the whole subject — Kuhn's 1962 argument, the 1969 Postscript, incommensurability's forty-year narrowing, Lakatos, Laudan, and the quantitative question.

Established The interesting part is how the duplication survived a tool built to catch it. The corpus's collision detector works on pairwise co-occurrence: it flags two slots as overlapping when their characteristic strings turn up together on pages that already exist. That test needs pages to co-occur on. At the time of the merge the corpus contained none of the relevant vocabulary — no page said “incommensurability”, none said “normal science”, none said “Kuhn”. Two slots sharing their entire vocabulary registered zero overlap, because the shared vocabulary appeared nowhere. The detector was not wrong. It was measuring co-occurrence in a corpus with no occurrences, which is a method that fails hardest exactly where a field is newest.

Frontier That failure mode is not a bug report; it is the ordinary condition under which taxonomies get revised. A category system reorganises when a term that named one thing is found to name two, or when two terms are found to name one. Kuhn's central late claim — the one that survived thirty years of retreat — is that scientific revolutions are taxonomic revisions of exactly this shape. A brief about how science reorganises its own categories found the live case in its own map. The analogy is this page's, not Kuhn's, and it is an analogy rather than evidence.

Frontier What the page lands is a tie broken by one measurement and one methodological scandal. The punctuated picture has cases, narrative fit, and no quantitative support. The one direct attempt to measure conceptual change across many fields recovered gradualism and named Lakatos rather than Kuhn. And the decade's most-cited quantitative claim about the tempo of scientific change is under a specific, arithmetically stated attack showing it drifts toward its conclusion as reference lists lengthen. The quantitative turn was supposed to settle Kuhn. It undercut the theory and its own best instrument in the same decade.

Established What Kuhn actually argued in 1962, stated narrowly. Science alternates between normal science — puzzle-solving inside a framework that is not itself under test — and revolutionary science, where the framework is replaced rather than corrected. Three theses did most of the work and most of the damage: theory-ladenness of observation, that what scientists see depends on the framework they hold; progress without convergence, that science progresses from problems rather than toward a final true description; and incommensurability, that theories separated by a revolution suffer “certain deep kinds of failure of comparability”. He also rejected the positivists' separation of discovery from justification. Established Every attribution on this page runs through the peer-reviewed encyclopedia entries rather than through Kuhn's text, which was not obtained for this pass; the page says so at each point and quotes no primary text it did not read.

Established The 1969 Postscript conceded a term, not a thesis, and the difference is the first correction the page owes its readers. The second edition (1970) carries a Postscript—1969 in which Kuhn accepted that his central word had been used in incompatible ways and split it: paradigm as exemplar, a concrete exemplary puzzle-solution that guides later work by perceived similarity rather than by rule, which Kuhn called “the most novel and least understood aspect of this book”; and paradigm as disciplinary matrix, the whole constellation of shared beliefs, values, instruments and techniques held by a community. Frontier The concession was forced by Margaret Masterman's contribution to the 1965 London colloquium, the famous count of distinct senses of “paradigm” in the 1962 text. The number usually quoted is 21, and this pass could not obtain Masterman's own paper — treat the count as unverified here.

Established The Postscript also contains the whole of Kuhn's answer to the irrationality charge, and it is five words long. Theory choice is constrained by five shared values — accuracy, consistency, scope, simplicity, fruitfulness — which are held across the sciences, do not compose into an algorithm, and are weighted differently by different scientists, who therefore reach different conclusions from the same evidence. That is the non-algorithmic reading, and it is Kuhn's own. Frontier Whether it answers the charge is a live philosophical question with no empirical test.

Established Incommensurability spent forty years getting smaller, and its final form is a claim about classification. The 1962 version is global: whole paradigms fail to make contact. Feyerabend introduced a cognate notion independently the same year, defining incommensurable theories as deductively disjoint through meaning variance. From the 1980s Kuhn narrowed it three ways: local incommensurability, the untranslatability confined to “a few, typically interconnected concepts” rather than to whole theories; taxonomic incommensurability, in which what changes across a revolution is the classification of kinds and successive lexicons cross-classify the world; and the no-overlap principle, that kind terms may not overlap unless one is a species of the other, so categories nest or are disjoint. A revolution violates the incumbent taxonomy's no-overlap structure, which is why two lexicons cannot be merged into one.

Established The most misquoted point in the whole subject: incommensurability does not imply incomparability. The reference literature states this explicitly and repeatedly, and attributes the rejection of the conflation to Kuhn and Feyerabend themselves — theory comparison “is merely more complicated than imagined by some philosophers of science”. Comparison proceeds pragmatically and without a neutral observation language; it does not stop. Established The early objectors got there first and were answered on different ground: Dudley Shapere's 1964 review pressed the relativist consequences and framed the subsequent debate, and Israel Scheffler's Science and Subjectivity (1967) ran a referentialist counterattack — reference to worldly things can survive a change in the sense of the terms, so two theories can be about the same objects while meaning different things by their names. Frontier Whether Scheffler's reply succeeds is undecided and untestable.

Established Lakatos is the rival with an edge, and the edge is novel prediction. His methodology of scientific research programmes keeps Kuhn's historical realism and restores a rational criterion. A programme has a hard core immunised by decision: “The negative heuristic of the programme forbids us to direct the modus tollens at this 'hard core'. Instead, we must use our ingenuity to articulate or even invent 'auxiliary hypotheses', which form a protective belt around this core.” The positive heuristic is “a partially articulated set of suggestions or hints on how to change, develop the 'refutable variants' of the research programme”. A programme is theoretically progressive if each new theory predicts novel unexpected facts and empirically progressive if some of those predictions are corroborated; failing both, it is degenerating. His own illustration is blunt: “Has…Marxism ever predicted a stunning novel fact successfully? Never!” — theoretically progressive at first, then empirically degenerate, then theoretically degenerate, with auxiliary hypotheses “all cooked up after the event”.

Frontier The edge is duller than advertised, on three counts the entry states and this page does not soften. Feyerabend's revival objection is that a degenerating programme can come back, and the embarrassing case is Urbach's: the environmentalist IQ programme declared degenerating in 1974 and then spectacularly vindicated by the Flynn effect. Progressiveness is legible only in hindsight, so the criterion tells you what was rational rather than what to pursue now. And Lakatos conceded that playing the game of building progressive programmes carries no guarantee of reaching truth, which would require the “whiff of inductivism” he elsewhere refused. A criterion that only scores completed matches is not a criterion for choosing a match.

Established Laudan's contribution here is historiographical rather than demarcational, and it is corrosive to the whole enterprise. His 1983 survey shows that the criteria philosophers have used to characterise science have themselves turned over repeatedly — apodictic certainty in Aristotle, certainty without causal explanation in Newton, method in the nineteenth century, then verifiability, then falsifiability — and that at the method stage “there was no agreement about what the scientific method was”. Frontier A discipline whose meta-level criteria are as unstable as its object-level theories is poor material for a single account of how change happens. His positive programme, problem-solving effectiveness as the unit of appraisal in Progress and Its Problems (1977), is the third rival to Kuhn and Lakatos; no primary source for it was obtained, so it is named here and not characterised further. The demarcation argument itself belongs to Philosophy of Science.

Frontier The best recent textual intervention argues that the received Kuhn is not the written Kuhn. Samuel Schindler, in Synthese (2024), argues on textual grounds that Kuhn never applied “dogmatic” or “uncritical” to practising scientists — only to scientific education — and that commitment to a paradigm under anomaly is epistemically justified rather than dogmatic, because paradigms earn commitment by demonstrated problem-solving success and abandoning one on an anomaly would remove the basis for productive work. His cases are Mercury's perihelion, unresolved for over two hundred years; stellar parallax, 295 years between prediction and observation; and the Higgs boson, 52 years between prediction and discovery. In each, scientists actively pursued the anomaly rather than ignoring it. Established That Schindler argues this, on these cases, is a fact; that the reading of Kuhn is correct is a 2024 move in an open dispute.

Frontier And then somebody measured it. Ju, Zhou, Blevins, Lydon-Staley, Kaplan, Tuma and Bassett built 38 concept networks from Wikipedia hyperlink structure across mathematics, the natural sciences and philosophy, ranging from 77 nodes (Boolean algebra) to 3,174 (law), with discovery dates parsed from the articles. Paradigm shifts are operationalised as temporal modularity change across multilayer networks, knowledge gaps as topological cavities via persistent homology, and epochs via changepoint detection. The recovered signature is four epochs — “(i) an initial, short epoch of little change, (ii) a long epoch of moderate and lasting change, (iii) a short epoch of much greater change, (iv) a long epoch of stability” — and the authors' own conclusion is that shifts in the structure of concepts occur “not as abrupt Kuhnian shifts but as gradual Lakatosian modifications”. The rival a Kuhn-centred page would present as a footnote is the one the data picked.

Frontier The authors state their own limits and this page carries them rather than burying them. Network structure “is different from the process used to obtain that structure”; multiple processes can produce the same network; and the analysis captures conceptual relationships only, not theoretical commitments or practices. It is one operationalisation, one data source, and that source is an encyclopedia written by volunteers in the present about the past. This is not “Kuhn refuted”. It is the only direct measurement anyone has made, and it came out the other way.

Frontier The cautionary case sits next door and it is the reason to distrust the next bibliometric answer as well as this one. Park, Leahey and Funk reported in Nature in 2023 that papers and patents are becoming less disruptive over time — 45 million papers and 3.9 million patents across six decades and six datasets, scored on a CD index that characterises how a paper or patent changes networks of citations. The reported declines are enormous: 91.9% in the social sciences and 100% in the physical sciences from 1945 to 2010, and for patents 78.7% (computers and communications) to 91.5% (drugs and medical) from 1980 to 2010. Frontier Petersen, Arroyave and Pammolli then showed the index is biased by citation inflation: CD has a bounded numerator and an unbounded denominator containing an extraneous citation term, so as reference lists lengthen the index drifts toward zero regardless of actual disruptiveness. Their figures: mean references per article rising from about 9 in the 1960s to about 23 in the 2000s, a 2.6-fold increase; citation-network growth of 5.1% a year, doubling every 13.6 years; a −0.0253 coefficient on log reference count; and, contrary to the received result, a positive team-size coefficient.

Frontier The point of the pair is not which side wins. It is that the single most-cited quantitative claim about the tempo of scientific change turned out to be substantially a property of how citation practice has changed, and that any bibliometric operationalisation of “revolution” inherits that hazard by construction. Innovation History runs the same dispute as evidence about innovation rates; here it is used only as a warning about proxies, and the two pages do not restate each other. Established The bibliographic details of the Petersen critique could not be confirmed — the fetched page returned an internally inconsistent venue and year, so it is cited by URL rather than by a reconstructed citation.

Frontier Is conceptual change punctuated at all? This is the live question and it currently has one answer, from one method, on one corpus. Ju and colleagues find four epochs dominated by a long stretch of moderate lasting change followed by a long stretch of stability. A second measurement on a different corpus, with a different operationalisation, would be the most valuable single result anyone could produce in this subject — and nobody has produced it.

Frontier Is the taxonomic version of incommensurability testable? The no-overlap principle is a claim about the structure of kind terms across a revolution, and it is closer to an empirical claim than anything else in Kuhn: it says the incumbent and successor lexicons cross-classify, and cross-classification is visible in a sufficiently careful reconstruction of what a discipline's terms picked out before and after. Frontier No such reconstruction has been built at scale. The obstacle is not conceptual; it is that historical term-extension data does not exist in machine-readable form for any discipline.

Frontier Does Lakatos's criterion work forward? Novel-prediction success is checkable in retrospect and the IQ-programme case shows that a degenerating verdict can be reversed decades later. Whether a research programme's trajectory can be scored in near real time — on preregistered novel predictions rather than on retrospective narrative — is an open and directly testable question that nobody is testing.

Frontier Was normal science ever dogmatic? Schindler's 2024 reading says Kuhn never said it was of practising scientists, and offers three long-anomaly cases in which the anomaly was actively pursued. The counter-position — that commitment under anomaly is exactly what dogmatism looks like from inside — is held widely and argued textually rather than measured. Frontier A behavioural version of the question is available and unasked: how much of a field's effort goes to its acknowledged anomalies, and does that share change before a theory turns over.

Speculative Are the 1962 Kuhn and the post-1969 Kuhn effectively different theorists? Kuhn's own remark on his reception was that he was “tempted to posit the existence of two Thomas Kuhns”, and the volume that quotes it holds there are now dozens. That is a reading of a reception history, not a result, and it is carried here because the page's readers usually hold one Kuhn and do not know which.

Speculative Can bibliometric indices detect revolutions at all? The whole scientometric-revolution literature assumes yes. The CD-index episode is direct evidence that an index in this exact area can be substantially a property of citation practice, and no index currently in use has been validated against an independently established revolution. Handwave The step where this literature does its work by assertion is the identification of a network signature with a conceptual event; nothing in any of these papers establishes that link, and the strongest of them says so.

Frontier The first bottleneck is that “paradigm” has no agreed referent, and this is not a rhetorical complaint. The term's author accepted in print that he had used it in incompatible ways and split it into two. Any measurement of paradigm shifts must first choose one of those senses, or a proxy for one, and the choice is not neutral: exemplars are concrete puzzle-solutions and disciplinary matrices are community-wide constellations of commitments. A method that detects change in one is not detecting change in the other.

Frontier Every quantitative instrument in this subject measures the record rather than the science. Ju and colleagues measure Wikipedia's hyperlink structure and say so. The CD index measures citation networks and is demonstrably sensitive to reference-list length. Co-word and topic-model approaches measure publication vocabulary, which is edited for publication. There is no instrument that touches a theoretical commitment directly, and there is no obvious way to build one.

Established The second bottleneck is arithmetic and it is specific. The CD index has a bounded numerator and an unbounded denominator containing an extraneous citation term; reference lists lengthened by a factor of about 2.6 between the 1960s and the 2000s; the citation network grows 5.1% a year. Under those conditions the index falls whether or not anything about science changes. Any index with the same shape has the same failure mode, and the shape is common because bounded-over-unbounded is what you get when you normalise a count by a growing network.

Frontier The third is that the historical record thins exactly where the theory needs it. Kuhn's cases are drawn from a small number of well-documented episodes in physics, chemistry and astronomy. The taxonomic thesis needs the extensions of technical terms before and after a change, which survives for a handful of episodes and for essentially nothing else. A theory of change built from the best-documented cases and tested against the best-documented cases is not being tested.

Frontier The fourth is that the field's own reference literature is load-bearing in a way that is rarely stated. Almost every claim about what Kuhn, Feyerabend, Lakatos, Shapere or Scheffler held reaches a page like this one through peer-reviewed encyclopedia entries rather than through the primary texts, because the primary texts are books in copyright. That is a real epistemic dependency and it belongs in a section about bottlenecks rather than in a footnote.

Established Nothing on this map produces a result this brief waits on, and no typed depends-on edge is claimed. What it waits on is a measurement nobody has built and an access condition nobody has supplied, both recorded as typed requirements below. The measurement is a second, independent operationalisation of conceptual change that does not run on citation or hyperlink structure. The access condition is machine-readable, licensed corpora of primary scientific texts across a discipline's turnover, which is what a taxonomic test would need and what copyright and archive terms currently prevent.

Established Five primary sources this subject rests on were not obtained for this pass, and each is named rather than paraphrased from memory: Kuhn's Structure itself in either edition; Masterman's 1970 paper and the count of senses of “paradigm” usually given as 21; Shapere's 1964 review; Scheffler's Science and Subjectivity (1967); and Laudan's Progress and Its Problems (1977). Everything attributed to them here is attributed through the encyclopedia entries, and no direct quotation from any of them appears on this page.

Frontier The one genuine research dependency is on scientometrics, and it points the wrong way. Any future attempt to detect revolutions quantitatively will use the same citation and text infrastructure that produced the disruption-index episode. Until an index in that family is validated against an independently established case of theory replacement — rather than calibrated on the assumption that it detects one — results from that infrastructure cannot settle the question this brief asks.

Frontier The cheapest high-value experiment is to rerun the concept-network analysis on a corpus that is not Wikipedia. The method is published, the pipeline is described — multilayer temporal modularity, persistent homology for cavities, changepoint detection for epochs — and the obvious second corpora exist: textbook editions across revisions, discipline-specific thesauri and classification schemes with dated revisions, and encyclopedia editions from before the web. If the four-epoch signature survives a change of corpus it becomes a finding about knowledge; if it does not, it was a finding about Wikipedia. Neither answer has been obtained.

Frontier Second: recompute the disruption literature with the correction already specified. Petersen and colleagues did not merely allege a bias; they named the mechanism, quantified the reference-list growth (a factor of about 2.6), the network growth rate (5.1% a year) and the log-reference coefficient (−0.0253), and reported a sign reversal on team size. The corrected series across the same six datasets is a computation, not a research programme, and until it is published the field has a headline finding and a specific reason to distrust it.

Frontier Third: build the taxonomic test Kuhn's late position actually implies. Take a discipline with a documented terminological turnover and reconstruct what its kind terms picked out before and after — then check the no-overlap structure directly. If successor lexicons cross-classify the incumbent's kinds, the strongest surviving version of incommensurability has its first empirical support. If they nest cleanly, it does not. This is the one test in the subject that could come out either way on evidence, and nobody has built it.

Frontier Fourth: score research programmes forward rather than backward. Lakatos's criterion turns on novel predictions that were unexpected when made. Preregistration infrastructure now records exactly that — what a group expected, before it looked. A prospective register of programme-level predictions in a small number of contested fields would convert an entirely retrospective criterion into a scoreable one within a decade. Speculative Whether the resulting scores would track anything anyone recognises as progress is the open part.

Frontier Fifth, and cheapest of all: measure anomaly effort. Schindler's textual case and its opposite make different behavioural predictions about how a field allocates work to its acknowledged anomalies before a theory turns over. Funding records, conference programmes and preprint volumes are all dated and all counted. The measurement has not been attempted and does not require anything that does not exist.

Frontier The toolchain that produced the only direct measurement is worth stating precisely, because its parts are borrowed and each one imports an assumption. Concept networks are built from hyperlink structure with node dates parsed from article text. Change is scored as temporal modularity variation across a multilayer network, which assumes that community structure in a link graph tracks conceptual organisation. Knowledge gaps are scored as topological cavities via persistent homology, which assumes a cavity in a simplicial complex built from the graph corresponds to something missing in the knowledge. Epochs come from changepoint detection, which assumes the underlying series is piecewise stationary. Any of those three assumptions failing would produce the observed result for reasons unrelated to science.

Frontier The engineering requirement for a better instrument is a dated, versioned, machine-readable record of what a field's terms meant. Not what was published — what was classified as what. Classification schemes exist and are revised on dated schedules; discipline thesauri carry term histories; taxonomic databases in the life sciences carry explicit synonymy. These are the closest thing to a lexicon time series that exists, and none of them has been used to test the taxonomic thesis.

Established The third engineering fact is that indices normalised by a growing network need a growth correction as a matter of construction. The CD-index episode is the demonstration. A citation network doubling every 13.6 years will drive any bounded-over-unbounded statistic toward its floor on a timescale that looks exactly like a secular decline in whatever the statistic is named after. The correction is arithmetic and known; what is missing is a convention requiring it before publication.

Frontier And a fourth, which limits everything above: the unit of analysis in this field is almost always a document. Papers, patents, articles and links are the things that can be counted. Commitments, exemplars and disciplinary matrices are not documents. Every instrument here reaches its subject through a proxy chosen for availability, and the honest statement is that no engineering path from documents to commitments has been proposed, let alone built.

Within this map: Philosophy of Science, which owns demarcation, confirmation, realism and explanation — Popper appears in both pages and the seam is explicit: that brief asks whether falsifiability separates science from non-science, this one asks whether change-by-refutation matches the historical record, and hands the answer to Lakatos. Laudan is split the same way: the demarcation argument is filed there, his place in the post-Kuhnian historiographical revolt is filed here. Innovation History owns the disruption literature as evidence about innovation; this page borrows it only as a warning about bibliometric proxies. Scientific Institutions Through History owns the journals, academies, referees and norms that the community in Kuhn's account is made of.

Also within it: Scientific Governance Models, which owns how the present machinery allocates and polices; Artificial Scientists, where the question of whether a machine could execute normal science or only normal science becomes concrete; Scientific Funding Models, which owns the money that decides which programmes get to be progressive; and Intelligence Measurement, which owns the Flynn effect that supplied Lakatos's most embarrassing counterexample.

Outside it: history of science, which supplies the cases; scientometrics, which supplies the instruments and the cautionary tales; and the sociology of scientific knowledge, whose appropriation of Kuhn he spent the last decades of his life rejecting.

And the map itself is adjacent in an unusual sense. The merge that produced this page — X-25 Scientific Paradigm Shifts cross-listing to X-11 Scientific Revolutions — is a category revision of the kind the page is about, performed on the page's own index, invisible to the tool built to detect it because that tool measures co-occurrence and there was nothing to co-occur.

Frontier This subject has an unusual institutional dependency: its evidence base is books in copyright, and its working reference layer is an encyclopedia. The peer-reviewed entries this page runs on are excellent and are cited as what they are — entries, not the philosophers. But a field whose primary literature is commercially published monographs from the 1960s and 1970s, and whose secondary access is a single freely available reference work, has a concentration risk that nobody in it names.

Frontier The measurement side has the opposite problem: too many instruments, no validation regime. Scientometrics publishes indices faster than it validates them, and the disruption-index episode is what that looks like when a headline result reaches Nature. What is missing institutionally is not a better index but a convention: no tempo-of-science index should be published without a stated behaviour under reference-list growth. That convention would have caught the CD index before 45 million papers were scored with it.

Established The interested parties on this page are named where they matter. The two Kuhn quotations that anchor the reception history — Lakatos's “mob psychology” and Kuhn's “two Thomas Kuhns” — are taken from a university press blog post promoting that press's own volume on interpreting Kuhn. It is used for two quotations it reproduces and for nothing else. The Ju analysis is cited from an author-hosted PDF whose venue is not stated on the copy fetched, and is named as an author-hosted PDF. The Petersen critique's own venue line could not be confirmed.

Frontier And the corpus this page sits in is itself an institution with a taxonomy problem. The collision detector that failed to see this merge is a real tool that works as designed on a mature corpus and fails silently on a new category. Recording that on the page rather than in a maintenance log is a deliberate choice: a map that hides its own revisions is asking readers to trust a category system it will not show them.

Frontier “Paradigm shift” is the most successful piece of vocabulary philosophy of science has ever exported, and its export destroyed its content. In ordinary usage it means a large change that the speaker approves of, and it is used to claim exemption from the standards a field applies to smaller claims. That usage is traceable to a book whose author conceded in print that his own use of the term was inconsistent. The ethical issue is not linguistic drift; it is that the phrase is deployed to move a claim out of the reach of evidence.

Established The irrationality charge does real damage and it is not Kuhn's. It originates largely with Lakatos's characterisation that on Kuhn's view scientific revolution is “irrational, a matter for mob psychology” — a polemical formulation, not a quotation from Kuhn. Kuhn denied the position explicitly and repeatedly, gave five shared evaluative values that constrain theory choice, and distanced himself from the social-constructivist appropriation of his book, insisting in the essays collected as The Road Since Structure (2000) that the primary determinants of scientific outcomes are internal to science rather than socio-political. Frontier A reader who takes “Kuhn showed science is irrational” from popular sources is holding a position its supposed author spent thirty years rejecting.

Frontier There is a symmetric error on the other side and this page will not commit it. Kuhn's rejection of the constructivist reading is a fact about Kuhn. It is not an argument against the sociology of scientific knowledge, whose own case stands or falls on its own evidence and is assessed in Philosophy of Science. Using Kuhn's denial as a refutation of a programme he did not evaluate would be exactly the move this page objects to elsewhere.

Speculative And there is a practical cost to the punctuated picture that nobody has measured. If change is gradual and Lakatosian, then a funding culture organised around identifying and backing the next revolution is optimising for an event structure that may not exist. That is a hypothesis about incentives, not a finding; the allocation evidence lives in Scientific Funding Models and does not currently address it.

Frontier The terminal position is that “periodic paradigm shifts” is not currently a measured feature of the historical record, and that the one serious attempt to measure it found something else. That is weaker than “there are no revolutions” and stronger than “the jury is out”. Kuhn's account has cases, narrative fit and enormous explanatory reach. It does not have a quantitative signature, its author narrowed its strongest thesis repeatedly over thirty years, and the only direct network measurement reports gradual modification.

Frontier If the gradualist reading is right, the practical consequence is about time horizons rather than about epistemology. A field that changes by accumulated modification of a protective belt gets better slowly and continuously, and the interesting institutional question becomes how long a programme should be funded before its degeneration is called — a question Lakatos's criterion answers only in retrospect and the Urbach case shows can be answered wrongly for twenty years. Speculative On that reading the civilisational risk is not missing a revolution; it is terminating a programme in its slow phase.

Speculative The taxonomic version, if it survives testing, has a sharper implication. If revolutions are reclassifications that violate an incumbent no-overlap structure, then the durable unit of scientific change is the category system, not the theory — and the systems civilisation has built for storing knowledge at scale are all classification schemes. Subject headings, patent codes, gene ontologies and disease nosologies would then be where revolutions actually happen, and their revision histories would be the record. Nobody has looked.

Frontier The measurement lesson generalises beyond this subject and is the most transferable thing on the page. A community adopted an index, scored 45 million papers with it, and published the result in its most visible venue before anyone checked the index's behaviour under the largest secular trend in its own input data. The correction, when it came, was arithmetic. Any civilisation-scale claim computed from a growing record inherits this hazard, and the defence is a stated behaviour under record growth rather than a larger sample.

Speculative And the reflexive point, offered as a hypothesis rather than a conclusion. The reason two slots on this map named one subject is that a term for a thing and a term for the thing's most famous instance drifted into looking like two things. Category systems do that continuously, under everyone's nose, and the tools built to catch it look for co-occurrence in a record that has not been written yet. If Kuhn's late position is right, that is not a maintenance failure. It is the mechanism.

These horizons track what could be measured and when, not what science will do:

• 10 yr: Frontier The corrected disruption series is a computation with a published specification, so the most likely near-term development is that the CD-index dispute resolves one way or the other on the same six datasets. Frontier A second concept-network measurement on a non-Wikipedia corpus is equally cheap and rather less likely, because there is no career in replicating a result nobody has cited enough to defend. Speculative Expect the taxonomic test to remain unbuilt: the data it needs is dated classification-scheme revisions, which exist, and the field that would use them does not currently talk to the field that maintains them.
• 25 yr: Speculative If validated instruments arrive at all, they arrive from scientometrics rather than from philosophy, and they will measure the record rather than the commitments — which means the honest form of the answer will still be about publication and classification behaviour. Speculative A plausible split: gradualism confirmed for concept structure, punctuation confirmed for instrument-driven episodes, and no single answer for “science” because the question was never well posed at that level.
• 50 yr: Speculative The interesting long-run possibility is that machine-executed research changes the object of study rather than the answer — if programme articulation becomes cheap and continuous, the distinction between normal and revolutionary science may stop carving anything, and the question becomes historical. Handwave That is an extrapolation from a capability that does not yet exist to a conceptual distinction that is already contested.
• 100 / 250+ yr: Handwave Beyond useful forecasting. The one datum at that horizon is that the vocabulary of this debate is itself about sixty years old, and the debate it replaced — logical empiricism against its critics — lasted about forty. Handwave Two turnovers is a story, not a base rate.

• Depends on Nothing on this map. No brief on this map produces a result this one waits on, and no typed depends-on edge is claimed. The dependencies that bind are a measurement nobody has built and an access condition nobody has supplied, both typed below.
• Requires (not on this map) A measure of conceptual change that does not run on citation or hyperlink structure. Every instrument in this subject reaches its object through the publication record: the only direct measurement of paradigm change uses Wikipedia hyperlink structure and says so, and the most-cited quantitative claim about the tempo of science was shown to drift toward its conclusion as reference lists lengthened by a factor of about 2.6 in a citation network doubling every 13.6 years. Until a second instrument exists that fails differently, a single answer from a single corpus is what the field has. This is not something anyone can choose to fund into existence — it is a missing result. And machine-readable, licensed corpora of primary scientific texts across a discipline's turnover: the strongest surviving version of Kuhn's thesis is taxonomic, saying that successor lexicons cross-classify the incumbent's kinds in violation of a no-overlap structure, and testing it requires knowing what technical terms picked out before and after a change. That evidence sits in monographs in copyright, in journal archives behind paywalls, and in classification schemes whose revision histories are held by standards bodies. It is an access problem rather than a discovery problem, which is why it is typed as institutional: a licensing arrangement could supply it and none has been sought.
• Enables Philosophy of Science and Scientific Institutions Through History both assume an account of how scientific change happens, and both are written so that they do not depend on this one being right. No typed enabling edge is claimed, and the reason is a finding: on the evidence assembled here the punctuated account has no measured signature, so nothing should be built on it.
• Adjacent History of science, which supplies the cases; scientometrics, which supplies the instruments and the cautionary tales; the sociology of scientific knowledge; and within this map Innovation History, Scientific Governance Models and Artificial Scientists.

Established “Kuhn showed science is irrational.” He denied it explicitly and repeatedly. The Postscript gives five shared values — accuracy, consistency, scope, simplicity, fruitfulness — that constrain theory choice without determining it, and his defence of exemplar-based similarity judgement is that being non-rule-governed is not the same as being irrational: family resemblance cannot be reduced to rules of rationality, so the charge is a category error. Frontier The charge originates largely with Lakatos's “mob psychology” formulation, which is a polemical characterisation and not a quotation from Kuhn. Whether Kuhn's defence works is open; whether he held the position attributed to him is not.

Established “Incommensurable theories cannot be compared.” The reference literature states the opposite as the position of both Kuhn and Feyerabend, in terms: comparison “is merely more complicated than imagined by some philosophers of science”. Frontier The conflation of incommensurability with incomparability is the most common single error in the subject, it is the load-bearing step in the relativist reading, and it is a misreading rather than a disputed interpretation.

Established “Kuhn retracted under criticism.” He disambiguated a term in 1969 and narrowed one thesis from the 1980s. Splitting “paradigm” into exemplar and disciplinary matrix is a concession about vocabulary; moving incommensurability from global untranslatability to local taxonomic restructuring is a narrowing of scope. Neither is a retraction of the account, and treating them as one is how readers end up believing the whole thing collapsed.

Frontier “Paradigm shift is a technical term with agreed content.” Masterman's critique and Kuhn's own Postscript concession are precisely that it was not, in the text that introduced it. Established The count of distinct senses usually given is 21 and this pass could not obtain Masterman's paper, so the number is printed here as unverified rather than repeated as fact. The concession itself is in the Postscript and is not in doubt.

Frontier “The punctuated picture has quantitative support.” It has cases and narrative fit. The one direct network measurement across 38 concept networks reports gradualism and names Lakatos. The disruption-index literature, which is where most people have met a number about the tempo of scientific change, is in open dispute over whether its central trend is an artefact of reference-list growth. Frontier Anyone citing a figure for how often paradigms shift is citing something nobody has measured.

Frontier “Science has become less disruptive, and that proves revolutions are ending.” Two problems, and the second is fatal to the inference even if the first resolves. The CD index is under a specific quantified attack showing it drifts to zero as reference lists lengthen, with a corrected team-size coefficient that reverses sign. Established And even a clean disruption series would be a series about how papers change citation networks, which is not a measurement of theory replacement and was never claimed to be one by the paper that introduced it.

Established “Kuhn licenses social constructionism about science.” He rejected the appropriation in The Road Since Structure, insisting the primary determinants of scientific outcomes are internal to science rather than socio-political. Frontier That is a fact about Kuhn and not an argument about the strong programme, whose own case is serious, is routinely misrepresented in both directions, and is assessed in Philosophy of Science rather than dismissed here by proxy.

Frontier “Normal science means dogmatic science.” Schindler's 2024 textual argument is that Kuhn applied “dogmatic” and “uncritical” to scientific education and not to practising scientists, and that his three long-anomaly cases — Mercury's perihelion over two centuries, stellar parallax at 295 years, the Higgs boson at 52 — show anomalies being actively pursued rather than suppressed. Frontier It is a live intervention in an open dispute, not a settled correction, and is carried here as one.

Frontier “Lakatos solved what Kuhn left irrational.” His criterion is real and checkable, and its author conceded it carries no guarantee of reaching truth without the “whiff of inductivism” he refused. Feyerabend's revival objection has a concrete case: a programme declared degenerating in 1974 and vindicated later by the Flynn effect. A criterion that can be wrong for twenty years about a live programme is a scoring rule for history, not a decision rule for research.

Handwave “Bibliometrics will settle this.” This is where the whole quantitative literature does its work by assertion. Identifying a network signature with a conceptual event is the load-bearing step and no paper in this area establishes it; the strongest of them states plainly that network structure is different from the process that produced it and that multiple processes yield the same network. Speculative The CD-index episode is direct evidence that a headline finding in this exact area can be substantially a property of citation practice, and no revolution-detecting index has been validated against an independently established revolution.

Established And one about this page rather than about Kuhn. Two slots existed on this map for one subject, because “revolution” and “paradigm shift” sound like different things. The tool built to catch that could not, because pairwise co-occurrence needs pages to co-occur on and the corpus contained none of the shared strings. The duplication was found by reading, and it is recorded here rather than quietly deleted, because a brief about how science reorganises its own categories should show its own.