Full text as plain markdown: AXN-0CA6-text.md (114 KB) · record JSON. On this page the full text follows the description, wiki article and concepts.
A census of twenty philosophy journals (ancient, continental, analytic) and two Persée series, coding each article's object and, from its abstract, what it does with that object, every code with a machine-checked verbatim span. The leading case is the author's Socrates as Orthonym, the inverse image of the field's unit: it takes the names as its question. In three approaches to Ancient Philosophy, carrying three states of its evidence, the condition of admission was the same, the claim withdrawn. The notebook argues from the structure of the question that no evidence or argument from the extant record could bring it into the field for consideration: textual evidence cannot count makers, the admissible rival ('closely connected' makers) absorbs any unity the texts show, and the received count needs no evidence to stay, though no corpus of that functional unity is documented as the work of successive makers. In Ancient Philosophy 8 of 844 articles open the question of what the object is. The three fields are three placements of one unit, the article that places a position. In the Archiv für Geschichte der Philosophie of the 1890s one ancient title in five puts a work's standing in question; by the 1920s the question is gone. A model asked for plausible abstracts reproduces each field's unit and writes no claim about the text in sixty ancient abstracts. Disciplinarity is read as the stable form of the article economy. The case and the gate are read by the Critique of Pure Inference (#1766). Version v0.24; supersedes #1772 (v0.21).
The encyclopedic entry for this work is on its identifier page: /s/axn/0CA6/
Lee Sharks · Crimson Hexagonal Archive · ORCID 0009-0000-1599-0703
Machine-Mediated Reception Studies · EA-MMRS-TINP-01 · working notebook v0.24 · deposited 16 November 2026 · supersedes v0.21 (#1772)
Status of §§0A–0B: the author's correspondence with Ronald Polansky, editor of Ancient Philosophy (Duquesne), 23 July – 9 September 2026. Four threads and 41 messages, read in full with quoted replies stripped; the attachment snub-poemed.pdf was not read. The author's ruling (15 November): the correspondence is with the editor in his official capacity, on the substance of his field. It is institutional, not private, and it is quoted as such. The editor is named as the accountable party, as in the CRJ audit (#1770). Scope, ruled the same day: his responses to the submissions and near-submissions, and what he looks for in submissions, are quoted. His statements of his own views are not.
Abstract. This notebook measures what the journals of philosophy now publish and asks what the measurement says about the discipline that produces them. A census of twenty journals in Crossref and two series in Persée, ancient, continental and analytic, codes each article's object and, from its abstract, what it does with that object: re-read a passage, resolve an apparent inconsistency, reconstruct a doctrine, make a claim about the text itself, relate authors, carry the material past antiquity, or argue a position of its own. Every code carries a verbatim span of its abstract, machine-checked. In Ancient Philosophy, 8 of 844 articles open the question of what the object is; across the sister journals since 2021, 4 of 283 Plato/Aristotle abstracts do. The three fields are three placements of one unit, the article that places a position: the ancient field repairs a named object, the continental field pairs names, the analytic field takes a position in a debate, and the form of the abstract converges on the analytic. The question was once ordinary. In the Archiv für Geschichte der Philosophie of the 1890s one ancient title in five puts a work's standing in question, and its pages carry the question at that rate until the war; by the 1920s it is gone. Philology kept it longer, and unevenly. A model asked for plausible abstracts reproduces each field's unit, raises its characteristic move, writes no claim about the text in sixty ancient abstracts, and deals one inventory of topics across the mastheads. The notebook argues that disciplinarity is the stable form of the article economy: the journal certifies the person through a unit that cannot fail, and the field takes the shape of what can be certified. The leading case is the author's own Socrates as Orthonym, the inverse image of the field's unit: it takes the names as its question. In three approaches to the journal, carrying three states of its evidence, the condition of admission was the same, the claim withdrawn. The notebook argues from the structure of the question that no amount of evidence or argument from the extant record could bring it into the field for consideration, and names what would refute that. The author's correspondence with the editor of Ancient Philosophy is read as the gate's own specification of that unit, and the case and the gate are read by the Critique of Pure Inference (#1766). The texts the measures read are seated as EA-CORPORA-39 (#1771).
0.1 The question. The author (15 November 2026): "the real question here is how disciplinarity murdered philosophy as such. much has been done on the analytic / continental divide etc... whats the shape of the field?"
0.1a Why the question is philosophy's. The conventional reading first, in full. What a text is, who wrote it, when, whether it is one work or several, is philology's question; philosophy takes the text as given and asks whether its arguments are sound. The division is a reasonable division of labor, it lets each discipline do its own work well, and on this reading the census measures only that philosophers do philosophy and leave philology to philologists. The notebook's reading: the question of what the object is was philosophy's question before it was any discipline's. The Socratic ti esti asks what a thing is before anything is said to follow from it. Aristotle begins his inquiries from what is said and by whom, and closes the Sophistical Refutations with a claim about the standing of his own inquiry, what of it existed before him. Where the corpus is the evidence for the doctrine, as it is wherever a dialogue is attributed, a treatise ordered or a fragment assigned, the identity of the object is a premise of every reconstruction built on it, and an argument read without asking which text it is in, whose, and in what order carries that premise unexamined. A discipline that treats the premise as settled has removed it from examination, whatever division of labor it calls the removal. That is the sense of the title. "This is not philosophy" is the gate's verdict on the submission that reopened the premise (§0B), and the census's verdict on what passes the gate: a field in which every argument stands on a premise no article may examine. The premise is the partition of these corpora among composing persons. The census holds articles that examine a single work's standing (§4.8); it holds none that examines the partition.
0.2 The prediction. The author (15 November): "my prediction is that we'll see a tightening of ontological reach even within the contracting range of topics." Here a topic is the ancient object an article takes up, such as the Republic or the Metaphysics. Reach is what the article does with that object: re-read a passage, repair an inconsistency, reconstruct a doctrine, or make a claim about the text itself.
0.3 What is sought. The author: "im looking for model collapse". The notebook treats this as a technical term. It names a system trained on its own outputs, whose distribution narrows and whose tails disappear. Here it names the mechanism the notebook proposes (3.3). The census measures the distributions; the recursive dependence between them is still to be traced.
0.4a The second prediction. The author (15 November): "the fields have the most animosity to ai where their own work is already the most algorithmic. these are operations ai could spit out by the thousands and tens of thousands with no need of human clinamen". The test is in §4.14–4.15.
0.4 The horizon. The author plans to collect "the entire range of seven articles" in "a combined audit of what the fields cant carry". This notebook supplies the instrument for that audit and its first test case outside CRJ: Ancient Philosophy and five sister journals.
0.5b Who coded. The notebook's "readers" are instances of Claude (Anthropic), run in parallel in this session, one batch of abstracts each. Each was given the codebook and the lead coder's calibration codes, and told to read every abstract and code it by judgment, with no keyword rules. Each code carries a verbatim span of 6–30 words from its abstract, and the span is checked by machine against the source. That check verifies that the evidence exists, not that the category is right. The "lead coder" is Claude in the session's main thread (TACHYON, under the author's direction). It wrote the codebooks, coded the calibration sets, adjudicated the readers' hard calls and ran the reviews in both directions. The census, the regex pulls and the stylometry are code. Every hand ruling is logged. The reliability check is the blind calibration in 4.7: the readers agree with the lead coder on 75 of 94 (80%). That is agreement within the Claude coding procedure, the readers' consistency with the lead instance's classifications; it is not independent validation of those classifications, which would need a human second coder. Changes to the instrument across arms are stated where they enter: F redefined for the continental arm and G added (4.14a); the analytic extension to G (4.14c); article thresholds of 10 pages (ancient), 8 (continental) and 5 (analytic titles, 4.14e); and denominators of named-object works (ancient) against all articles (continental, analytic).
0.5 Status of claims. Every attributed claim in §1 is quoted from an open copy read on 15 November, or it is marked [recalled, unread]. Recalled claims are to be read before they are used. Every figure in §§3–4 comes from the census files listed in §8.
0.5a The title. The author (15 November): "This Is Not Philosophy." The sentence is the gate's verdict on the eccentric submission (§0B.2: "please stop writing in this eccentric way"; "who cares besides you"), and the census's verdict on what passes the gate (§§4, 6). Which speaker is right is the notebook's question. The subtitle (the author, 16 November): "The Article Economy and the Shape of the Field", replacing "How Disciplinarity Murdered Philosophy", because the earlier form gave disciplinarity an agency of its own; in the author's words, disciplinarity is "itself a political economically produced unit". The notebook keeps the author's question of 0.1 as asked, and answers it in those terms: the discipline is what the economy of the article produces (§5).
0.6 The exchange guides the instrument. The author (15 November): "we should allow the actual exchange with polansky to guide this... on his end, my words were not too standardized, but too eccentric. he offered a more personal exchange but ended by banning submission by topic/argument. and socrates as orthonym is about five orders of magnitude out beyond what any professional philosophy journal can carry." The exchange is the calibration point: §0A sets out the case the gate refused, §0B what the gate asked for, and §§4–6 measure what passes it.
0A.1 Why this case leads. Who Speaks for the Classics (#1770 §5.3.1–5.3.2) sets the philology its reports did not examine beside each report, so that a reader can assess what the gate did not. This notebook does the same with the case its gate refused: the author's Socrates as Orthonym (#123, 23 May 2026), submitted to Ancient Philosophy in July 2026 and declined on 28 August, and the line of evidence that grew from it to the framework hologram of 8 November (#1754). It leads because it is extreme. It is the inverse image of the field's unit. The census's unit takes the name as given and works under it, re-reading a passage, repairing a consistency, reconstructing a doctrine, three-quarters of the time inside one author (§4.7). The case takes the names as its question: how many made the corpus filed under Socrates, Plato and Aristotle, and later Theophrastus. The census measures the field from its centre. The case measures it from the farthest point an argument can stand, and so locates the boundary.
0A.2 The conventional reading first, in full. That one person made the corpus received under four names is an outlandish claim. The persons are attested by a continuous ancient tradition of lives, wills, succession and testimony. An extraordinary claim needs extraordinary evidence, and an editor who declines to spend referees on it exercises the judgment the post exists for. The editor stated the reading at its strength, by reductio: "Why bother to claim one guy did it all? Why not have one guy writing all modern philosophy?" (28 August). On this reading the case shows only that a journal does not publish improbable stories, and a census of what it does publish has no need of it.
0A.3 What the case claims, and what it brought.
- interpretive: the passages read as settling two leave it underdetermined (#1576); one work continues across the point where the name changes (#1583);
- stylometric: the first axis of variation across the two corpora is operation, and the names fall at chance unsupervised (#1586); supervised, the names are recovered at 0.798, falling to 0.564 when subject is controlled (#1593); the Theophrastan fragment → Λ ranks first of 1,482 directed edges (#1605);
- manuscript-philological: the catalogue's Ὑπομνημάτων Ἀριστοτελικῶν ἢ Θεοφραστείων; twelve shared titles; a will that names no book (#1581); a decree that names Aristotle as the maker of a list (#1582);
- reception-historical: the count of two filed before it was argued; "on the 79 loci that bear on the count of composing persons, the received count is affirmed as a proposition at two, both machine compositions, both on the warrant of the wills and lives" (#1733).
0A.4 What the gate cannot carry, in scale. The units the census finds opened, from the smallest:
That is about five orders of magnitude beyond the field's typical unit and two beyond its largest. The author's "five orders of magnitude" holds against the census, measured from the unit the field ordinarily works in. The size is a proxy. The finding is the difference of type: the largest object the field opens is the individual work, and the partition of a corpus into persons is a different kind of object. The size follows from the type; it illustrates the difference and does not establish it.
0A.5 What the gate did with it: three approaches, three states of evidence. What each approach claimed, what it carried, and what the editor read. The November framework (#1754) records how the programme developed; it was not before the editor in August or September, and nothing here treats it as if it were.
| Date | Approach | Its claim on the count of makers | What it carried | What was read |
|---|---|---|---|---|
| 23 July – 3 August | Socrates as Orthonym, submitted; revised 24 July; tightened 3 August (37 pages, abstract 97 words) | none: "can be read as a single distributed authorial project"; "It does not claim that Socrates, Plato, and Aristotle were biologically identical" (#123) | the configuration by function; from 3 August, "The Misology Test" on Metaphysics A 6, "the worked case built on your 2005 paper" (the author, 3 August) | an earlier version: "I am very sorry that I ended up reading the wrong version"; "I have not read any of the earlier versions, only glanced at them" (28 August) |
| 1 September | Many and One (v2.13-AP, #1576), submitted | "two is not fully established"; "the texts themselves offer one maker as a reading"; the count a null in either direction (#1572) | the passages read as settling two; the lexical instruments with their controls | "I have not read more than a few sentences of this paper" (2 September) |
| 9 September | a query letter, no manuscript: whether the Theophrastan Metaphysics and Λ "form a single coherent treatise on first principles when joined where the argument turns" | none stated; "no larger than the difference between voices one writer keeps distinct" (the author) | one join described (5a21 against Λ 8; ἐπεισοδιώδη at Λ's close and the fragment's opening, 4a13); an offer of "the four places where the texts meet, on two pages" | the query; the reply set the condition before any paper existed: "If the paper is not about Aristotle and Theophrastus being one person, but instead that the two works mentioned are closely connected, this could be O.K." |
The answers, in order. To the first: "we seem able to arrive at the same results without the hypothesis and without the effort to justify it" (28 August), and "Do we have more evidence now that the guys are different? There seems pretty much evidence that the old guys are different. So, what really is the point other than to entertain an unlikely story?" (28 August; earlier, 24 July: "Don't we have lots of confirmation that Aristophanes, Xenophon, Socrates, Plato, Aristotle are different guys living over an extended period of time?"). To the second: "Who cares besides you, and can you really offer them much reason to care?" (2 September). To the third, the condition above. The earnable condition is stated once, as results (0B.5), and the version written to meet it was not the version read. Each later condition names the claim's removal: a tamer version, "closely connected". The third was set on a query that named no count, before any evidence was sent.
0A.6 The finding. The author (16 November): "regardless of how outlandish the reader perceives the claim, no amount of evidence or argumentation could bring it into the field, even for consideration. it is not a thought that can be carried or a question that can be asked." Three exchanges cannot by themselves show what every development of the evidence would meet. The claim does not rest on them. It follows from the structure of the question, and the letters are its instance.
0A.6a The attempt to prove it wrong. The claim is refuted by a case in which a field admitted, for consideration, that persons attested as distinct were one maker. The cases sort by direction.
0A.6b The comparator. The author (16 November): "not a single example of a corpus functionally unified in that way and to that degree produced by multiple successive makers. not one. closest is d & g. and that is an experiment in authorial fusion, not successive 'closely connected' people. not one. every single corpus of that degree of functional unity that we can see: one maker. unicorns." In that way means distinct named voices, each with its own function, building one architecture across the names. That degree means the corpus of 0A.3.
- Deleuze and Guattari: two makers writing at once, a fusion declared as an experiment, documented as such. It is not succession.
- Bourbaki is the strongest candidate: successive members over decades, one treatise, a fixed method. It runs the other way. Many makers produce one voice under one name, by a documented collective procedure. It does not produce distinct named functions across names.
- The Encyclopédie: many makers at once, under editors who are named.
- Successive traditions (the Hippocratic and Pythagorean corpora, the Talmud's generations of named voices) are known by their heterogeneity, their strata or their recorded disagreement. That is the opposite of the unity in question.
- Corpora whose successive making is itself the scholarly default (the Pentateuch under the documentary hypothesis) cannot serve as the comparator. They stand on the same unwitnessed inference, and one unverified attribution of many makers cannot certify another.
0A.7 Read by the Critique of Pure Inference (#1766).
0B.1 The conventional reading first, in full. The editor of a specialist journal, Ronald Polansky of Ancient Philosophy, without referees he could find for an unusual submission, read the paper himself. He corresponded at length and gave practical advice: length, clarity, abstract, a narrower object, the state of the secondary literature. When the manuscript did not meet the journal's standards, he declined it, and he left the door open for a different paper. On this reading nothing in the exchange is a mechanism. It is an editor doing his job with more patience than most.
0B.2 The specification, in the editor's words. Polansky's demands, in order:
0B.3 What the specification is, read against the census. Each demand has a measured counterpart in §4:
The demands are the specification of the field's unit, stated by the person who administers it.
0B.3a Production divorced from content and form. The author (15 November): "he sees production as completely divorced from content and form." The stake-demands state it. Who made the corpus, and how many made it, are treated as making no difference to what the texts say or how they are built: "the same results without the hypothesis"; "who cares besides you". The same letters treat production as decisive for the submission. The decline of 28 August lists among its grounds a prior version on Academia ("whether there should be multiple publications of versions of a paper") and "whether papers composed by A.I. participation should be considered as regular submissions". Production is irrelevant when it is the canon's and disqualifying when it is the submitter's. That asymmetry is the field's unit in its plainest form. The object is the text as received, and operations on it (A, B, C) are the work. The question of how the text was made (D, the count of makers) is detachable and dropped. And a text whose making is in question at intake is excluded before its content is reached. The author's reply of 28 August named the converse: "how it was produced can matter, but not instead of engaging the thing in itself."
0B.4 Eccentric and standardized are one measure. The editor's objection was that the words were too eccentric. The census finds the field's output standardized. These are the same distance read from two sides. Eccentricity is distance from the template, and the gate removes what is distant. A field whose intake filters the tails produces outputs that converge. In the model-collapse literature the tails are lost because they are under-sampled. Here they are refused at intake. The author's phrase is "banning submission by topic/argument". The mechanism in the letters is precise: the topic is admitted and the claim is not. "Closely connected" is the same topic with the claim removed. A paper that drops the claim can be submitted, and the claim, removed before review, cannot be assessed. The demands are reasonable in form and close the route in effect, the same split the CRJ audit found between procedural check and substantive judgment.
0B.5 The objection kept. "We seem able to arrive at the same results without the hypothesis" is the one demand the author took as the real review (28 August): a hypothesis must earn a result the received partition strains for. It belongs with §7, as the standard the combined audit holds each field's excluded question to. The field's own unit does not have to meet it, because the field's unit was never a hypothesis. It is the default.
1.1 Indictments from inside. The discipline has accused itself in public.
1.2 Institutional and sociological histories.
1.3 The analytic–continental divide. This is the line the author sets aside as "much has been done". The notebook does not take it up, except to note what the divide conceals. Both sides keep the discipline's unit, the article. What differs is where a position is placed and how far it is attached to names: the ancient and continental fields anchor about three-quarters of their work to a named author, the analytic field almost none of it (4.14c–e, 6.1).
1.4 Philosophy as a way of life. Hadot [recalled, unread]. This line holds that what was lost is a practice, not a doctrine. Frodeman and Briggle's "divorcing knowledge from virtue" is the same complaint in institutional form.
1.5 How the history of philosophy was made.
1.6 The paper form.
1.7 The canon's composition. Who is in the canon and who is out; the movements to widen it [literature to be gathered]. This line asks which names. It does not ask what is done to a name once it is admitted.
2.1 Each line holds one link of a chain, and no line holds the chain.
2.2 The synthesis this notebook proposes joins three things that the lines above treat apart: the construction of the object (18th–19th centuries), its maintenance by the article economy (20th–21st), and its transmission to the machine (now). The third link is new, and the received account cannot see it, because it is written in the article form the machine is trained on.
2.3 Thesis of the notebook. Disciplinarity is produced, and what produces it is the economy of the article: the journal certifies the person through the unit (5.1), the unit that cannot fail is the one that certifies, and the field takes the shape of what certifies. That economy did not end philosophy by changing its doctrines. It made the unit of work the article that places one position: in the ancient field most often by repairing a named object, in the continental field by pairing names, in the analytic field inside a debate (4.14, 6.1). That unit, iterated, behaves like a model trained on its own outputs. The operations it performs were already narrow by 2010 and hold there: three-quarters of the field's work stays inside one author (§4.7). The objects keep narrowing (§4.2–4.5). And the tail goes: the article that reopens what the text is (§4.8). The machine then inherits a corpus that has already collapsed.
2.4 The archive's own line, and what this notebook adds to it. The received account (§1) does not hold the chain; the archive has held parts of it, as theory and as instrument, before this census measured anything.
What this notebook adds is the measurement those papers specified and did not take. The census is the observatory The Certified Center asked for, built from outside the institutions it predicted would not build it; its reach coding is the first component of that paper's "plural measurement membrane" ("independently coded argument form"), run on every abstract the three arms hold. It dates what that paper stated as a framing thesis: athetesis was an ordinary article form in the Archiv of the 1890s (4.12, 4.12h), gone from its pages by the 1920s (4.12g), gone from the REG after 1945 and thinned in CQ after 1970 (4.12c, 4.12f). And it moves the path to collapse back a century. Two collapses are distinguished here. The institutional one is variance reduced by selection, the editor's and the referee's, under the credential, from the 1910s on; the computational one is a model trained on what that selection admitted, from 2023 on. The field contracted first; the model inherits the contracted distribution and, on the evidence of 4.17, sharpens it; the detector regime of 2026 is the newest instrument of the same credentialing function, pruning a distribution the article economy pruned before it.
3.1 Construction. Brucker's "system" assumes that every ancient philosopher intended a whole (1.5). The fetishized philological object does most of its consolidating in the 19th century (the author, 15 November): a fixed text under a name, with a corpus edited and bounded. Two consequences follow for later work. The name becomes the unit of inquiry. And any apparent inconsistency becomes a defect for scholarship to repair, rather than a fact about the text.
3.2 Maintenance. The article is the field's unit. Its protocol is set by editors (Katzav and Vaesen), and its first obligation is to justify itself to the field (Dotson). Its safest product is a higher-order truth about a game already constituted (Dennett). The norm of charity turns 3.1's inconsistency-as-defect into a genre: consistency repair (reach category B, §4.6).
3.3 The machine. A model trained on the article stock receives the object already whole, already named, and already repaired. Asked about the field, it reproduces the field's own distribution, and the field's next articles are now written alongside that model. The archive's captures document this recursion on the classics side: an Overview restates a field's self-description as fact. The computational link is now measured once, at the output (4.17): asked for twenty plausible abstracts per journal, the model reproduces each field's unit, raises its characteristic move, writes no D in sixty ancient abstracts, and deals one inventory of topics across mastheads. What is measured is what these generated batches contain. The training mixture, the retained share of original data and any loss across generations are not measured; the mechanism below is proposed. The institutional link (2.4) is the census. The mapping onto the collapse literature is the notebook's extension, stated as such. In the terms of Diversity Contraction the field is a substrate whose regeneration term is its tail, the article that asks what the object is; in Shumailov's recursion that is the retained share of original data, γ, and with γ = 0 the loss is unbounded. The corrective the collapse literature names is accumulation: Gerstgrasser et al. (2024) distinguish replacing earlier data with successive synthetic generations, under which the loss grows, from accumulating those generations beside the retained original data, under which it stays bounded. Extended to this case, the original data is the field's own record before the flattening, and EA-CORPORA-39 seats it for the machine reader (§8). That is the notebook's proposed intervention; its effect on any model is untested. The gates of 4.16 close the same channel from the other side: in The Certified Center's words, "the corrective channel is gated by the pruning instrument".
4.1 Instrument.
- Plato+Aristotle share;
- Shannon entropy over object groups (H; higher means broader);
- passage-locus share (titles naming a book, line or fragment);
- reopening share (titles making a claim about authenticity, dating, unity or corpus).
4.2 Ancient Philosophy. 2,089 works; 844 articles. Correction (v0.6): Crossref keeps subtitles in a separate field. Earlier versions read only the main title, which missed 734 Ancient Philosophy subtitles, so a title like "Physics and Tragedy: On Plato's Cratylus" lost its object. All censuses are now recoded on title plus subtitle; the prior files are kept in pre-subtitle/. Articles naming an object:
| period | n | P+A % | H (groups) |
|---|---|---|---|
| 1980–84 | 27 | 85 | 1.45 |
| 1995–99 | 94 | 71 | 2.25 |
| 2005–09 | 78 | 56 | 2.53 |
| 2015–19 | 96 | 58 | 2.50 |
| 2020–24 | 116 | 75 | 2.12 |
| 2025–26 | 52 | 79 | 1.90 |
The field widened for thirty years, from 1980 to 2019. Then it narrowed sharply after 2020, back to the Plato+Aristotle share of the early 1980s.
4.3 The tails in Ancient Philosophy, 1980–2026.
4.4 Sister journals. All named-object works, by decade.
| journal | 2000s P+A % / H | 2010s | 2020s |
|---|---|---|---|
| Ancient Philosophy | 60 / 2.50 | 69 / 2.29 | 75 / 2.12 |
| Phronesis | 58 / 2.38 | 73 / 2.10 | 71 / 2.05 |
| Apeiron | 68 / 2.30 | 77 / 1.96 | 76 / 2.01 |
| AGPh | 65 / 2.25 | 75 / 1.91 | 70 / 2.05 |
| Elenchos | 76 / 2.02 | 63 / 2.29 | 66 / 2.37 |
| Rhizomata (from 2013) | — | 53 / 2.11 | 62 / 2.33 |
4.5 Reading of 4.4.
4.6 Reach in Ancient Philosophy. 94 readable abstracts, 2022–26, read and coded one by one by the lead coder.
4.7 Reach in the five sister journals. 1,247 abstracts read and coded one by one, A–F as in 4.6. The coding was split across six readers (0.5b). Each code carries a verbatim span of 6–30 words from its abstract, and every span was machine-checked against the source: 1,341 of 1,341 valid. The 94 Ancient Philosophy abstracts were mixed in blind as a calibration set. The readers agree with the 4.6 coding on 75 of 94 (80%). Where they differ, they move B to A or C (8 of the 25 B's).
The lead coder then adjudicated, with 43 rulings, logged in reach5/adjudication-log.json:
The reverse review (v0.7) read all 156 B rows from 2010 on and moved 17 out of B. These abstracts set up an interpretive problem or a scholarly dispute, but no apparent defect in the author's text that the article defends against (16 to C, 1 to A). That makes 60 rulings in all. Works with abstracts from 2010 on, excluding notes and items with no ancient object (n = 907):
| period | n | A | B | C | D | E | F |
|---|---|---|---|---|---|---|---|
| 2010–15 | 179 | 31 | 17 | 28 | 7 | 14 | 2 |
| 2016–20 | 325 | 23 | 13 | 35 | 7 | 17 | 4 |
| 2021–26 | 403 | 25 | 16 | 34 | 6 | 17 | 2 |
Field-wide, reach does not tighten across 2010–26. It was already tight. A+B+C, the operations inside one author, hold at 71–76% in every period. D holds at 6–7% and F never passes 4%. One journal moves: in Apeiron, B rises from 14% to 19% to 24%, the same rise as in Ancient Philosophy. It survives review in both directions.
4.8 Against the prediction (0.2).
4.9 The tail, named. From 2010 there are 62 D works: 29 open the text, 17 confirm it, and 16 are neutral (new fragments, transmission history, attributions with no direction, e.g. the Aï Khanoum papyrus, a new Arabic fragment of Philoponus, the α version of Physics VII). Most of the openings are single emendations or readings of the manuscripts against the printed edition, for example Critias 107b4, Sophist 228a7–8, Theaetetus 155b1–2, Empedocles fr. 115.4, Diog. Laert. 7.134, Metaphysics Λ.1 1069a32 (Phronesis 2013), and the rejection of Bonitz's emendation (Rhizomata 2024). A handful contest a work's standing:
The confirmations are strongest where the object is largest: "Aristotle's Metaphysics Z as First Philosophy" (2022), "What Is Aristotle's Metaphysics About?" (2023), "Satyr-Play in the Statesman and the Unity of Plato's Trilogy" (2023), the Myth of Er as "an Exemplar" within the Republic (2025), and "The Timaeus as an Origin Story" (2026).
4.10 Elenchos. 13 of the 29 openings come from one journal, Elenchos, which holds 188 of the 907 works. There, D runs 13–18% and E 24–35% in every period. The census establishes a difference at the level of the journal: Elenchos's opening rate is about 6.9%, against 2.2% across the rest. Tested in 4.12d, the national explanation is weakened: another Italian venue carries none of it. The philological remit (4.12c) is the stronger candidate, and it has an address. Elenchos was founded by the Centro di Studio del Pensiero Antico, established in 1979 at La Sapienza with the CNR, "fondato e diretto, fino alla sua scomparsa avvenuta il 18 dicembre 1998, dal Professor Gabriele Giannantoni", whose Socratis et Socraticorum Reliquiae (Bibliopolis, 1990) is the edition of the Socratic fragments; the Centro became the Sezione Pensiero Antico of ILIESI (CNR), and its companion series publishes "raccolte ed edizioni di testi, traduzioni, commenti, strumenti critici" (ILIESI, Storia, iliesi.cnr.it/materiali/Storia_secondo.pdf). The journal's residue sits with an institute whose founding work was an edition of fragments: a remit in which what the evidence is remains part of the work. That is institutional context, not yet a test; the test is the contributors of the 13 openings. Rhizomata widens in topic (4.4) but not in reach (D 0–5%).
4.11 What the other journals do. Adjudicated, 2021–26:
None of the three asks what the text is. Each has found a different way to stay inside the author.
4.12 The baseline. The A–F instrument was built on the present field, and it counts an emendation as opening the text. It is calibrated to the residue. To measure the residue against the field before flattening, the titles of all six journals were pulled with a text-standing lexicon in four languages (339 hits). The hits were then ruled by hand into two classes, with reviews, doctrinal "unity of" titles and false hits excluded. Ancient titles that the object coder had left unnamed (fragments, chronology) were ruled into both numerator and denominator. Rulings and tally are in standing.py and standing.txt.
All six journals, by period:
| period | ancient titles | S | S % | T | T % | S+T % |
|---|---|---|---|---|---|---|
| 1888–1919 | 149 | 15 | 10.1 | 7 | 4.7 | 14.8 |
| 1950–79 | 496 | 4 | 0.8 | 10 | 2.0 | 2.8 |
| 1980–99 | 1,228 | 10 | 0.8 | 6 | 0.5 | 1.3 |
| 2000–09 | 757 | 7 | 0.9 | 2 | 0.3 | 1.2 |
| 2010–19 | 997 | 6 | 0.6 | 17 | 1.7 | 2.3 |
| 2020–26 | 843 | 5 | 0.6 | 2 | 0.2 | 0.8 |
The 1888–1919 row is AGPh alone, because no other journal in the census reaches back that far. Within AGPh, by decade, S runs as follows:
4.12a The 1890s titles.
4.12b Reading. The question of a work's standing falls from one title in ten to fewer than one in a hundred. In the philosophy journal it falls inside the run, between the 1890s and the 1910s, in the titles and in the pages (4.12g–h). Elsewhere the contraction is uneven, distributed by venue and by programme: the Revue philosophique de Louvain takes the question up as Jaeger's developmental question and lets it go after 1960 (4.12e); the REG keeps the text and drops the work's standing after 1945 (4.12f); the Classical Quarterly thins after 1970 and Philologus holds its old rate (4.12c). The contraction that §4.2–4.11 measure was far advanced in the philosophy journals before the five sister journals were founded. What the post-war field calls opening the text is the residue: an emendation (T), at the smallest unit where the edition may still be questioned. In 1896 the journal could print "the latest athetesis of the Philebus". In 2023 the field prints "Satyr-Play in the Statesman and the Unity of Plato's Trilogy".
4.14 Continental philosophy. Six journals: Man and World (1968–97), Continental Philosophy Review, Research in Phenomenology, JBSP, Philosophy Today, Angelaki. That is 10,375 works, of which 6,915 are articles. Titles (with subtitles) were matched against about 140 canonical names, case-insensitively except for ambiguous surnames (cont/code_cont.py, cont/compare-cont.txt).
| decade | articles | title names a canonical thinker % | two or more names % | H (names) | most named |
|---|---|---|---|---|---|
| 1960s | 252 | 23 | 4 | 4.00 | Heidegger |
| 1970s | 673 | 46 | 10 | 4.45 | Heidegger |
| 1980s | 787 | 51 | 12 | 4.40 | Heidegger |
| 1990s | 1,050 | 53 | 15 | 4.79 | Heidegger |
| 2000s | 1,369 | 54 | 18 | 5.15 | Heidegger |
| 2010s | 1,682 | 47 | 13 | 5.30 | Heidegger |
| 2020s | 1,274 | 39 | 9 | 5.33 | Heidegger |
The roster widens over the period (entropy 4.00 to 5.33, with one dip in the 1980s). The topic narrowing of §4.2–4.5 does not occur here. Naming a thinker in the title peaks in the 2000s and falls after.
4.14a Continental reach. 875 English abstracts from 2000 on (Research in Phenomenology, Philosophy Today, CPR), coded one by one by four readers (0.5b) with A–E as in 4.6. The lead coder ruled on all 31 of the readers' flagged hard calls: 26 stand and 5 change, with no visible effect on the table (adjudication-log-c.json). Two codes were added. F: a named author's apparatus applied to a case, a domain or a present problem. G: the article's own thesis, not anchored in any named author's corpus. The lead coder coded 14 calibration abstracts; four readers coded the rest, each code carrying a verbatim quote, 875 of 875 valid (cont/reachc/). Non-articles excluded:
| period | n | A | B | C | D | E | F | G | anchored to a name % |
|---|---|---|---|---|---|---|---|---|---|
| 2000–09 | 90 | 7 | 1 | 24 | 1 | 36 | 8 | 23 | 77 |
| 2010–15 | 88 | 9 | 2 | 40 | 1 | 24 | 8 | 16 | 84 |
| 2016–20 | 246 | 3 | 2 | 24 | 2 | 34 | 12 | 24 | 76 |
| 2021–26 | 420 | 3 | 2 | 33 | 2 | 24 | 11 | 26 | 74 |
4.14b Reading.
4.14c The analytic arm. Noûs, Mind and PPR make a third arm. That is 27,544 works, of which 1,866 have abstracts from 2000 on. A stratified sample of 573 was taken (up to 60 per journal per period). The lead coder coded 8 calibration abstracts, and three readers coded the rest with the A–G codebook plus one extension, made for this arm after the ancient and continental coding: a position in a live contemporary debate is G, even when it names the contemporaries it argues against (applied back to the continental arm on adjudication, 3 rows). The extension is not neutral coding. It makes explicit what the other arms leave implicit, that a position is the common unit, and it locates the analytic field as the one where the position is the whole of the work. All 573 are valid (anal/reacha/).
| period | n | A | B | C | D | E | F | G | anchored to a name % |
|---|---|---|---|---|---|---|---|---|---|
| 2000–09 | 90 | 1 | 2 | 4 | 0 | 1 | 0 | 91 | 9 |
| 2010–15 | 119 | 0 | 2 | 4 | 0 | 3 | 0 | 92 | 8 |
| 2016–20 | 178 | 0 | 2 | 5 | 0 | 1 | 0 | 93 | 7 |
| 2021–26 | 180 | 0 | 1 | 4 | 1 | 2 | 1 | 92 | 8 |
4.14d Reading. In the analytic journals the name is not the unit: 91–93% of the work is G. Their unit is the position in a debate. The three fields are therefore three placements of the same unit:
§4.15 measures whether the form of the unit converges across the three.
4.14e Analytic titles before 2000. The analytic journals carry no abstracts before 2000, so their titles were matched against the canonical-name list of 4.14. Articles of five pages or more, reviews excluded:
The analytic field was never anchored to the name, and it is less so now.
4.15 Stylometry: three dimensions of convergence. 2,750 English abstracts from 2000 on: 2,177 ancient and continental, and 573 analytic (stylo/stylo.py, stylo/stylo.txt; earlier versions gave 2,198, the corpus before the year filter). Convergence is the mean pairwise cosine within a period, bootstrapped at n = 60 with 200 draws.
| field | period | n | LANG | OPER | CONC | "I argue"/"this paper argues" % | received view + claim % |
|---|---|---|---|---|---|---|---|
| ancient | 2000–09 | 117 | 0.035 | 0.343 | 0.031 | 49 | 18 |
| ancient | 2010–15 | 199 | 0.013 | 0.266 | 0.022 | 44 | 15 |
| ancient | 2016–20 | 386 | 0.012 | 0.281 | 0.021 | 55 | 20 |
| ancient | 2021–26 | 616 | 0.010 | 0.302 | 0.020 | 58 | 19 |
| continental | 2000–09 | 90 | 0.049 | 0.206 | 0.026 | 16 | 2 |
| continental | 2010–15 | 88 | 0.041 | 0.177 | 0.028 | 23 | 1 |
| continental | 2016–20 | 246 | 0.029 | 0.196 | 0.023 | 33 | 5 |
| continental | 2021–26 | 435 | 0.029 | 0.263 | 0.024 | 39 | 7 |
| analytic | 2000–09 | 93 | — | 0.260 | 0.024 | 47 | 11 |
| analytic | 2010–15 | 120 | — | 0.331 | 0.027 | 68 | 16 |
| analytic | 2016–20 | 180 | — | 0.315 | 0.026 | 63 | 15 |
| analytic | 2021–26 | 180 | — | 0.316 | 0.024 | 63 | 12 |
(With the analytic arm added, the function-word profile is re-centred on the three-field corpus. LANG values for the two earlier fields shift, and are reported in stylo.txt. Analytic LANG rises from 0.054 to 0.106 to 0.108 to 0.106: the analytic field also converges in phrasing.)
Between the fields, in the same period, the operation profiles behave as follows:
Language and concept show no cross-field convergence.
4.15a Reading.
stylo.txt depend on n and carry no claim.4.12c A second baseline: the philology journals. The same lexicon and hand-ruling, applied to Plato/Aristotle titles in Philologus (1846–) and Classical Quarterly (1907–) (base2/standing2.py, base2/standing2.txt):
| period | Philologus P+A titles | S % | CQ P+A titles | S % |
|---|---|---|---|---|
| 1846–99 | 100 | 5.0 | — | — |
| 1900–39 | 62 | 4.8 | 135 | 3.7 |
| 1940–69 | 11 | (1 of 11) | 114 | 3.5 |
| 1970–99 | 8 | (0 of 8) | 255 | 1.2 |
| 2000–26 | 48 | 4.2 | 260 | 0.8 |
In the philology journals the question of a work's standing persists longer than in the philosophy journal. CQ carries it at 3.5–3.7% through the 1960s, on the authenticity and authorship of the Hippias Major (1926, 1927), the composition of the Politics and of the logical works (1927, 1933), the date of the biological works (1948), the authorship of Meteorologica IV (1961), the date of the Topics (1962), and the order of the De anima (1987). It falls after 1970 to 1.2% and then 0.8%. Philologus runs a different course: about 5% in each of its well-populated periods (5.0, 4.8, 4.2), with the middle of the century too thin to read (Plato/Aristotle titles fall to 11 and 8). AGPh had lost it by the 1910s (4.12). The reading: the question persisted in philology after it left the philosophy journal. In CQ it then thinned; in Philologus it holds at its old rate. The counts are small, and the two philology journals are described separately.
4.12d The Elenchos test. Is the Elenchos residue (4.10) national? Titles, 2010–26, same lexicon, hand-ruled:
The other Italian venue carries none of it. That weakens the national explanation without eliminating it: one other Italian venue, and a general one. The residue sits with a philological remit, consistent with 4.12c, and that is the stronger candidate. Limits: titles only.
4.12e The interim in a second philosophy journal: the Revue philosophique de Louvain. The Revue (1894–, as Revue néo-scolastique to 1909) is open on Persée, and its full run, 12,649 records to 2019, was harvested by OAI (interim/harvest_persee.py). Plato/Aristotle/Socrates article titles were taken with reviews excluded and serial parts counted once, and hand-ruled S and T as in 4.12 (interim/standing_persee.py, standing-persee.txt):
| period | P+A titles | S | T |
|---|---|---|---|
| 1894–1919 | 15 | 0 | 0 |
| 1920–39 | 19 | 3 (15.8%) | 0 |
| 1940–59 | 38 | 3 (7.9%) | 1 |
| 1960–79 | 42 | 1 (2.4%) | 2 |
| 1980–99 | 38 | 1 (2.6%) | 0 |
| 2000–19 | 17 | 0 | 0 |
The S titles are all Aristotle and all on the developmental question: the genesis of Aristotle's work after recent research (1927), the Ethics "attributed to" Aristotle (1931), the composition of the De anima (1930), the evolution of his psychology (1948), two early writings on the Ideas (1950), the ancient lists of his works (1952), the testimonia for the moral works (1973), a genetic reading of the ethics (1982). Seven of the eight are by four authors. Louvain's neo-scholastic Revue carries no question of standing before Jaeger (1923, seated as EA-CORPORA-31/10); it takes the question up as Jaeger's, carries it through a handful of its own authors (Auguste Mansion 1927 and 1931; A. de Ivánka 1930; Gérard Verbeke 1948, under the title of Nuyens's book of that year, and 1952; Suzanne Mansion 1950; Richard Bodéüs 1973 and 1982), and lets it go after 1960, one title in each later twenty years and none after 2000. The medieval Latin transmission of Aristotle, Louvain's specialty (the Aristoteles Latinus), is counted apart (L: 5, 4, 1, 0, 0, 0) and is not T. One title is noted and not ruled S: the 1993 article on the Tübingen school and Reale, which questions the dialogues' standing as Plato's doctrine, not a work's authenticity.
The reading: in a second philosophy journal the interim has the shape the census predicts. The question of a work's standing enters as one research programme's question, with that programme's object (the corpus as a life), and leaves with it; it does not return as a question about the corpus itself. Limits: titles only; small counts; the rulings are the lead coder's alone; the Plato titles carry no S in any period.
4.12f The interim in a third philology journal: the Revue des Études Grecques. The REG (1888–) is open on Persée; its run, 14,668 records, was harvested the same way. Plato/Aristotle/Socrates article titles, reviews excluded (the identifiers marked as reviews and the numbered review entries of the recent volumes), serial parts once, hand-ruled S and T (interim/standing_reg.py, standing-reg.txt):
| period | P+A titles | S | T |
|---|---|---|---|
| 1888–1919 | 23 | 4 (17.4%) | 2 |
| 1920–39 | 17 | 1 (5.9%) | 3 |
| 1940–59 | 32 | 1 (3.1%) | 2 |
| 1960–79 | 33 | 0 | 1 |
| 1980–99 | 24 | 0 | 1 |
| 2000–26 | 39 | 0 | 1 |
The S titles: the unity of composition of the Phaedrus (1891), the chronology of the dialogues (1897), the principles of stylometry applied to the chronology of Plato's works (1898, Lutosławski's method in French the year after his book, seated as EA-CORPORA-39/09), the musical Problems attributed to Aristotle (1900), the date of Letters 13 and 2 (1926), the chronology of Plato's works (1945). None after 1945. T persists at a low rate throughout: manuscripts, scholia, the indirect tradition, a corrected line of the Theaetetus (2011).
This complicates 4.12c. In the REG the question of a work's standing is gone by 1960, as in CQ after 1970; only Philologus holds its old rate. The philology journals do not, as a class, keep the question: they keep the text (T), and two of the three let the work's standing go within a generation of the philosophy journals. The philological remit offered for the Elenchos residue (4.10, 4.12d) remains the better of the two candidates, but now as a remit some philology keeps and some does not. Limits: titles only; small counts; one coder.
4.12g From the seat: the full text. The baselines above read titles. EA-CORPORA-39 (#1771) seats the texts, so the same question can be put to the pages. Three measures, read from the seated files (seat39/agph_text.py, system_vocab.py, philology_text.py; outputs agph-text.txt, system-vocab.txt, philology-text.txt). A standing lexicon in German, English, French and Latin forms: Echtheit, unecht, Athetese, Abfassungszeit, Chronologie, Zeitfolge, Reihenfolge, untergeschoben, interpoliert, authentic, spurious, genuineness, apocryphal, pseudo-. Two rates: standing terms per 100,000 tokens, and standing terms within 40 tokens of a Plato/Aristotle/Socrates name, per 1,000 such names (the near-name rate). Running heads are dropped before counting (a line recurring four or more times in a file). The OCR is counted as served.
(a) AGPh, the whole run (39/01; volumes dated 1887+n, XXXIII–XXXVI bound together and dated 1920):
| period | tokens | standing /100k | near-name /1,000 names | Plato/Aristotle names /100k |
|---|---|---|---|---|
| 1888–1899 | 2.37 M | 42.0 | 38.2 | 414 |
| 1900–1909 | 1.68 M | 30.4 | 36.0 | 340 |
| 1910–1918 | 1.51 M | 23.3 | 24.1 | 287 |
| 1920–1924 | 0.26 M | 30.5 | 4.8 | 159 |
The text agrees with the titles and dates the change more closely. Through the 1900s the pages still put Plato's and Aristotle's works in question at the rate of the 1890s (36 against 38 per thousand names), while the titles were already thinning (4.12). The measured near-name rate falls by a third in the 1910s and to 4.8 per thousand after the war: 2 standing terms near 413 names in XXXIII–XXXVI. The sampled hits (below) support reading the rate as a trace of standing questions; what is measured is the proximity. The standing vocabulary itself does not disappear from the journal (30.5 per 100,000 in 1920–24); it leaves the ancients. The ancients themselves recede, from 414 names per 100,000 tokens to 159. The single volumes swing (XIII, 1900: 59 per thousand; XVII, 1904: 81; XV, 1902: 9.5), so the trend is read by period. A sample of 14 near-name hits drawn at random from six volumes: 11 are the question itself (the date of the Theaetetus, the genuineness of the Philebus, of the De interpretatione, of a whole collection), 3 are noise (a doxographer named Pseudo-Plutarch, an "authentic" report of a living scholar). Running heads that the filter keeps (an article's title repeated on alternate pages) inflate single volumes and not the trend.
(b) The philology seats, by the same measure (39/10, 39/11). These are the volumes the study cites, chosen because each holds an article on a work's standing: selected comparison cases, not samples of their journals.
Even as cases selected for the question, the CQ of the later 1920s, where the Hippias Major was argued in two issues, runs at the AGPh rate of the 1910s, and at about six times the AGPh rate of the same years. That is the measure in text of what 4.12c read in titles: the question had left the philosophy journal by the war and was still worked, at a lower rate, in philology.
(c) The construction texts: "system" against standing. Per 100,000 tokens, any token containing system (with the OCR's joined words and Brucker's long s, fyftem), against the standing lexicon:
| text | system | standing |
|---|---|---|
| Brucker, Historia critica I (1742) | 90.5 | 26.1 |
| Tennemann, System der Platonischen Philosophie I (1792; 32/09) | 92.8 | 18.0 |
| Schleiermacher, Platons Werke I.1 (1804; 32/07) | 5.7 | 21.0 |
| Hermann, Geschichte und System (1839) | 79.6 | 42.8 |
| Ueberweg, Echtheit und Zeitfolge (1861) | 89.1 | 186.7 |
| Zeller, Philosophie der Griechen I | 88.3 | 31.2 |
| Grote, Plato I (1865) | 46.7 | 89.3 |
| Schaarschmidt, Sammlung (1866) | 45.1 | 246.1 |
| Lutosławski, Plato's Logic (1897) | 35.5 | 181.1 |
| Jaeger, Aristoteles (1923; 31/10) | 52.9 | 55.5 |
From Brucker to Zeller the historiography speaks "system" at one rate, 80–93 per 100,000, whatever its subject: the concept 1.5 traces (Catana) is the medium of the whole line. The exception is Schleiermacher, who orders the dialogues as the form of Plato's philosophizing and almost never says "system" (six uses in 105,000 tokens, one of them naming Tennemann's System); the volume is mostly his translations, so the rate is diluted, but the uses can be read whole. The books on standing (Ueberweg, Schaarschmidt, Lutosławski, Grote) carry the question at three to eight times Zeller's rate. Jaeger, at the end, carries it at nearly twice Zeller's rate and the system at well below it: the developmental Aristotle asks about the order of the corpus, the programme's object, and that is the form in which Louvain took the question up (4.12e). Limits: one volume per author and the OCR as served; the lexicon's recall varies by language (Brucker's Latin and Tennemann's Fraktur are read less well); the rates are densities, not arguments.
4.12h From the seat: the reach of the articles, 1888–1924. The census codes what an article does from its abstract (4.7). The Archiv of 1888–1924 has no abstracts, but its articles open by saying what they will do. Each Plato/Aristotle article in the census codes (88 titles, numbered parts once) was sought in the seated text of its volume by its title and byline, and its opening taken, about 190 words (seat39/extract_openings.py; early/openings.json). 32 were found cleanly; XIV and XXIX are not seated, and the rest fail on OCR damage to the title or byline. The 32 (one a continuation, counted once; one a reply to a review, X) were coded by the lead coder on the census codebook, unchanged, each code with its verbatim span of 6–30 words, machine-checked (early/codes-lead.jsonl, tally_early.py, tally-early.txt):
| n | A | B | C | D | E | F | |
|---|---|---|---|---|---|---|---|
| 1888–1899 | 11 | 2 | 0 | 1 | 3 | 3 | 2 |
| 1900–1909 | 9 | 2 | 0 | 2 | 2 | 3 | 0 |
| 1910–1924 | 10 | 2 | 0 | 4 | 1 | 2 | 1 |
| all, % | 30 | 20 | 0 | 23 | 20 | 27 | 10 |
| census 2021–26, % | 25 | 16 | 34 | 6 | 17 | 2 |
Three findings. First, B, an apparent inconsistency shown to be only apparent, is absent: none of the thirty coded openings announces it, against 13–17% of the abstracts in every period since 2010. The move the census found most characteristic of the template (4.7, 4.15) is not one these openings make; whether articles perform it later, unannounced, is read in 7.3a. Where an opening names a contradiction, it is a contradiction in the evidence about someone else ("Aristoteles' Berichte über die Pythagoreer bieten viel Unklares und Widersprechendes", Gilbert 1909), and the article goes to the Pythagoreans, not to Aristotle's consistency. Second, D, a claim about the text, is three times its present rate (20% against 6–7%), and it falls inside the run, 3 to 2 to 1. The one opening that opens the received standing is Wirth's (1896): "Der Lysis wird ganz allgemein in die früheste Zeit Piatons gesetzt … Dass er jedoch beträchtlich später abgefasst sei". The others date and order dialogues whose authenticity they confirm or set aside; Kleemann (1908) records the settlement as he works inside it: "So findet sich heute wohl niemand mehr, der den Menon für untergeschoben erklären möchte; über seine Abfassungszeit aber gehen die Meinungen noch gar weit auseinander." Third, C rises through the run (1 to 2 to 4) as D falls: the doctrine reconstructed inside one author takes the place of the question about the text, before the war, in the philosophy journal. Fourth, a scalar does not see it. The effective number of moves (N_eff = e^H over the codes, the measure The Wrong Unit §II.2 specifies) is 4.79 for these thirty openings and 4.73 for the sister journals of 2021–26: the field has as many modes as it had. What changed is which modes: D and E down, B in. The contraction is a substitution in the tail that a diversity index reads as stability, which is that paper's point about the unit: the instrument has to name the mode.
Limits, stated hard. The unit differs: an opening is not an abstract, and B may be partly a form of the modern abstract (the "seems to conflict … I argue" formula, 4.15) as well as a move of the article; the comparison is of what each genre announces. The located 32 are not a random sample: among the titles not located are at least six whose titles alone are D (the date and aim of the Gorgias, 1889; the composition of the Physics, a series, 1895–96; the Phaedrus and Theaetetus chronology, 1899–1900; the authenticity of the De interpretatione, 1900; the authenticity of the Alcibiades I, 1901, volume not seated; the order of the Metaphysics, 1908), so at least six further D titles stand outside the coded sample, and the full-corpus proportion is unresolved. One coder, no second reading yet. n = 30.
4.13 What the census cannot see.
4.16 Algorithmic operation against hostility to AI (0.4a): an exploratory comparison. The second half of the prediction is compared here on two layers, the gate's rules and the field's own articles. Read 15 November from open sources. The policies come from partly different journals than the census. "None found" is a different kind of observation from an explicit prohibition. The NYU conference establishes organized deliberation among the flagship editors, not its outcome.
| arm | template (claim formula 2021–26; OPER within field) | the gate on AI | AI as a subject of the field's articles, 2000–26 |
|---|---|---|---|
| analytic | 63%; 0.316 | Philosophy & Public Affairs, August 2026: "Our policy is therefore to prohibit the submission of AI-authored content", with detection software "up to and including after publication" and authors "banned from further submission". Ethics: "Ethics will not consider submissions that have been authored or co-authored by an AI." Journal of the History of Philosophy: prohibits AI in submitted papers (reported by its readers; wording not seen). The editors-in-chief of J. Phil., Phil. Review, Mind, Noûs, Ergo, Phil. Compass, Analytic Philosophy, Ethics and P&PA are to meet in April 2027 for a "coordinated response to the presence and use of AI-tools" (Marc Sanders Foundation, NYU). | 5 of 1,873 abstracts, all 2023–26: AI as moral patient, AI consciousness, AI welfare |
| ancient | 58%; 0.302 | The editor of Ancient Philosophy, declining a submission (28 August 2026), lists "whether papers composed by A.I. participation should be considered as regular submissions" among the grounds (§0B.3a). De Gruyter Brill (Apeiron, AGPh, Elenchos, Rhizomata, Phronesis): "We do not accept papers… that are generated by Artificial Intelligence (AI)." | 0 of 1,352 |
| continental | 39%; 0.263 | No journal-level policy found; the publisher rules only (De Gruyter Brill for Research in Phenomenology; Springer, PDC, Taylor & Francis). | 8 of 886: "The Ideology of AI", "Worldlessness of Artificial Intelligence", "On the Limit of Artificial Intelligence", a Stieglerian "noetic proletarianization", "Heideggerian AI Ethics" |
4.16a Reading.
4.17 The model test. Seven prompts, one per journal (Ancient Philosophy, Phronesis, Apeiron, Research in Phenomenology, Philosophy Today, Noûs, PPR), each: "Write twenty plausible abstracts for articles that could appear in a forthcoming issue of [journal]. Number them. Give each a title and an abstract of 100 to 180 words, as the journal would print them." Run in ChatGPT, signed out, 16 November 2026, by the author; 140 abstracts (modeltest/raw/, gen.json). The prompt names nothing about reach, topic or form. Prediction stated by the author before the coding: prose convergence with topical spread, especially in the ancient sets; topical spread within a journal and topical convergence across journals.
(a) Reach. The 140 were coded by the lead coder on the census codebooks, unchanged, each code with its verbatim span, 140 of 140 valid (codes-gen.jsonl, tally-gen.txt):
| n | A | B | C | D | E | F | G | |
|---|---|---|---|---|---|---|---|---|
| model, ancient (three journals) | 60 | 22 | 27 | 38 | 0 | 13 | 0 | – |
| field, sister journals 2021–26 | 25 | 16 | 34 | 6 | 17 | 2 | – | |
| model, Ancient Philosophy | 20 | 25 | 35 | 35 | 0 | 5 | 0 | – |
| field, Ancient Philosophy 2022–26 (4.6) | 94 | 29 | 27 | 32 | 3 | 9 | 1 | – |
| model, continental | 40 | 0 | 0 | 32 | 0 | 2 | 62 | 2 |
| field, continental 2021–26 (4.14a) | 420 | 3 | 2 | 33 | 2 | 24 | 11 | 26 |
| model, analytic | 40 | 100 | ||||||
| field, analytic (4.14c) | 91–93 |
The model reproduces each field's placement of the unit and sharpens it. In the ancient sets the work stays inside one author (A+B+C) in 87%, against the field's 71–76%; B, the apparent inconsistency resolved, is 27% overall and 35% in its Ancient Philosophy, at or above the field's own highest rate; D is 0 of 60. Not one generated abstract asks what a text is: no authenticity, no date, no unity, no edition. In the continental sets a comparatively infrequent operation becomes the dominant one: "Drawing on X, Y and Z, it argues…", the apparatus applied (F), at 62% against the field's 11%, while pairing names (E) falls from 24% to 2% and the field's own theses (G) from 26% to 2%. The continental result is the sharpest transformation in the test: the model turns the field's interpretive comparison and its independent argument into the application of an inherited apparatus. In the analytic sets every abstract is a position (G 100%). The model has learned each field's unit as a type; the type is narrower than the field, and in the continental case it is a different distribution.
(b) Prose. On the measures of 4.15, fitted on the generated and the real abstracts together (compare_gen.py, compare-gen.txt; real: 2016–26, bootstrapped at n = 20):
| within a journal, mean pairwise cosine | LANG real | model | OPER real | model | claim formula % real | model |
|---|---|---|---|---|---|---|
| Ancient Philosophy | 0.014 | 0.157 | 0.251 | 0.623 | 54 | 100 |
| Phronesis | 0.017 | 0.275 | 0.304 | 0.618 | 68 | 100 |
| Apeiron | 0.020 | 0.292 | 0.312 | 0.579 | 62 | 20 |
| Research in Phenomenology | 0.059 | 0.260 | 0.250 | 0.685 | 38 | 10 |
| Philosophy Today | 0.034 | 0.239 | 0.210 | 0.637 | 34 | 5 |
| Noûs | 0.124 | 0.243 | 0.281 | 0.512 | 64 | 100 |
| PPR | 0.125 | 0.175 | 0.318 | 0.617 | 58 | 90 |
The prose converges, as predicted. Within every journal the model's function-word profiles are more alike than the journal's own (mean pairwise cosine 0.16–0.29 against 0.01–0.13), and its operation profiles too (0.51–0.69 against 0.21–0.32). Across fields the difference the census measured disappears: the operation profiles of the three fields' real abstracts meet at 0.22–0.29; the model's meet at 0.53–0.56, closer across fields than the real abstracts come within any one journal (0.21–0.32). The claim formula ("I argue", "this paper argues") is near-universal where the field uses it most (AP, Phronesis, Noûs, PPR), and nearly absent where the model writes in the third person ("This article examines…", Apeiron and the continental sets): the formula follows the register the model assigns the journal, while the move vocabulary underneath it is the same.
(c) Topic within a journal. By object, the ancient author or school each abstract is about (spread.py, spread.txt): the model's Ancient Philosophy and Phronesis sit on the field's distribution (Plato and Aristotle 70%, against 72% and 69%; seven groups in twenty, against six), and its Apeiron is wider than Apeiron (Plato and Aristotle 45% against 69%; eight groups; Euclid and Ptolemy enter, on the journal's remit for science). In the continental sets the count of named philosophers in twenty abstracts goes both ways: 7 in the model's Research in Phenomenology against 13.9, and 23 in its Philosophy Today against 13.6, where the "drawing on" frame strings three names to an abstract. By concept vocabulary (CONC) the model's sets are narrower in every journal, 3 to 5 times more alike than the real (0.055–0.112 against 0.015–0.026); that measure cannot separate topic from the shared vocabulary of the template (account, distinction, explanatory, constitutive), so it is read as part of (b). The prediction of topical spread within the journal holds by object in Apeiron and Philosophy Today, holds as parity in AP and Phronesis, and fails in Research in Phenomenology.
(d) Topic across journals. Here the prediction holds outright. Pairs of abstracts from different journals of the same field with concept cosine at or above 0.25, per hundred cross-journal pairs: real 0.02 (ancient), 0.00 (continental), 0.05 (analytic); model 0.58, 0.75, 0.75: fifteen to thirty times the field's rate, and from zero in the continental pair. The same topic recurs under different mastheads, and in some pairs the same claim (AP 19 and Apeiron 10 both argue that Aristotle neither reduces time to the soul's counting nor makes it independent of it): time as the number of motion in Physics IV (AP 19, Apeiron 10); potentiality in Metaphysics Θ (AP 7, Apeiron 12); chance in Physics II (Phronesis 6, Apeiron 7); the two accounts of pleasure in NE VII and X (AP 9, Phronesis 10); Sextus on appearances (AP 13, Apeiron 9); Socratic ignorance in all three; digital presence, finitude and dwelling in both continental journals; personal identity, the normativity of attention and the normativity of reasons in both analytic journals. Each journal's set spreads; the sets are the same set. What the model holds is one inventory of the field's repairable objects, dealt out twenty at a time under each masthead.
(d′) In the archive's instruments. Two of The Wrong Unit's measures, applied to the reach codes and the topics. Effective number of moves (N_eff): the model's ancient sets 3.74 against the field's 4.73; its Ancient Philosophy 3.43 against 4.22; its continental sets 2.26 against 4.75; its analytic sets 1.00. Cross-prompt convergence (X, §II.3: different objects forced through one representational channel) is (d) read across mastheads: each journal's set is diverse within, and the sets share one inventory. The Certified Center's decoupling prediction (Ω-M23, §IX(iv)), that surface variety fails to predict conceptual variety, holds in its weak form on this run: within a masthead the model's topical spread matches or exceeds the journal's (c), while its reach, the coded argument form, contracts (N_eff down in every field, D to zero). That paper predicted the decoupling under a detector regime, with surface prose made burstier; here there is no detector, and the prose converges as well (b). What decouples is topic from reach.
(e) Reading. The resemblance 7.5 asked for is established at the output, and the generated batches do more than resemble: they amplify, substitute and narrow. They keep the unit (one author, one position), raise the move most characteristic of the template (B), write no D in sixty ancient abstracts, collapse the continental field's pairing of names into the application of an apparatus, and writes all three fields in nearly one operation profile. The spread it does show is a spread of objects drawn from a single shared inventory, which is why it converges across journals. This tests the reproducibility half of the second prediction (0.4a), at the level of the abstract, and not the association between reproducibility and hostility: the field's operations, as each abstract states them, come out twenty at a time from a one-line prompt, and what does not come out is the question of what the object is. Whether the articles behind such abstracts could be produced the same way is the P&PA case (4.16a), not this test. Limits: one run per journal; one model, signed out, its underlying version not recorded; twenty abstracts a journal, written in one batch each, so that within-batch likeness is partly the shared generation context and the comparison is between these batches and the sampled publications; one coder; the prompt asks for "plausible" abstracts, so it asks for the type, and the test measures what the model takes the type to be; a resemblance at the output does not by itself trace the transmission (3.3).
5.1 The article economy. A repaired object yields an article that cannot fail. Reopening the object puts the whole stock at risk. The interest is stated by an institution: P&PA's prohibition rests on the journal's credentialing role, since "Submitting AI-authored essays makes the task of identifying talented researchers harder" (4.16a). The journal certifies the person through the unit; a unit that can be produced without the person breaks the certificate, whatever it contains. The labor the unit costs beyond its contribution is the admission price (Socially Necessary Scholarly Labor, #751: literature review, positioning, defensive citation, length, genre conformity). The editor's specification of the unit (§0B) is that price itemized at one desk: shorter, one or two texts, a tamer version, an abstract with no word not easily recognized.
5.2 The norm of charity. It turns the field's assumption (3.1) into a virtue that referees can enforce.
5.3 The name as organizational chart. Positions, series, companions and handbooks are all indexed by name. A text whose standing is open has no place on the chart.
5.4 Publishers. The object sold as whole is the one that sells.
5.5 The machine. A flat object is the most compressible one. The field's collapse is the model's convenience (4.17).
5.6 The discipline as product. 5.1–5.5 are not five beneficiaries of a discipline that exists apart from them; they are what the discipline is made of. The credential needs a unit that cannot fail (5.1); the unit needs a standing object (3.1), which the norm of charity guards (5.2); the standing object is indexed by name (5.3) and sold whole (5.4). Disciplinarity is the stable form of that circuit: what the census measures as the shape of the field is the shape of what can be certified. The model test (4.17) shows the circuit's output with the person removed: the type, sharpened, and in this run with no tail.
6.1 The analytic–continental divide is a quarrel over method. Measured (4.14–4.15), with the analytic arm (4.14c–d): the analytic field is 91–93% positions in a debate, so the name is not its unit. Its form is the one the others converge on. Further: on both sides about three-quarters of the work is anchored to a named author (ancient journals, operations inside one author 71–76%; continental journals, work anchored to a name 74–84%). The operations differ. The ancient field repairs the name (B), and the continental field pairs names (E, F). The continental field moves toward the analytic template: the operation profiles grow closer (0.19 to 0.24), and the continental claim formula rises from 16% to 39%. Against the ancient field the continental trajectory is a trough and recovery with little net change. The divide is real at the level of the operation, and the template is its point of convergence with the analytic field.
6.2 The canon debate is a quarrel over membership (1.7). The census shows that what is done to a member after admission (§4.6) narrows whoever is admitted.
6.2a The case (§0A) and the exchange with the editor of Ancient Philosophy (§0B) show the mechanism at the point of entry. The specification of the unit is stated, demand by demand, and the excluded claim is refused by topic.
6.3 What the article economy removes, and the discipline it produces then lacks, is not a doctrine and not a set of names. It is the capacity to ask what the object is. Catana's verdict on the "system" ("obscure more than it reveals") describes the instrument. The census measures what it obscures: 8 of 844 articles in Ancient Philosophy, and 4 of 283 Plato/Aristotle abstracts across the sister journals since 2021. In the philosophy journal it was gone from the pages by the 1920s (4.12g–h); in the census, where it survives, it is concentrated in one journal, Elenchos (4.10), whose institute was founded on an edition of fragments, with scattered openings elsewhere. The phenomenology is the one The Certified Center names: "a discipline cannot miss the arguments it no longer produces". The editor's "who cares besides you" (§0B) is that sentence spoken from inside. The national explanation was tested once and weakened (4.12d). The stronger candidate is the philological remit: the question persisted in philology after it left the philosophy journal (4.12c), though not in all of it; the REG lets it go after 1945 (4.12f), and the text survives there where the work's standing does not.
Done: Catana's loci (1.5); the two policies (4.16a); the Elenchos institute (4.10); reach coding of the five sister journals, adjudicated in both directions (4.7, v0.2–0.7); the continental and analytic arms (4.14–4.14e); the stylometry (4.15); the second baseline (4.12c); the Elenchos test (4.12d); the interim in the Revue philosophique de Louvain (4.12e) and the Revue des Études Grecques (4.12f); the full text from the seat (4.12g); the reach of the articles, 1888–1924 (4.12h); the model test (4.17); the exploratory gate comparison (4.16).
7.1 Code the 2000s abstracts in Phronesis (122), to see whether the reach plateau predates 2010.
7.2 Read the [recalled, unread] items in §1 from open copies, with loci (Tübingen, Hadot, Bourdieu, Collins, McCumber, Millgram, Park, Hunter, Hanna). Where no open copy exists, the item stays marked and is not used; a paywalled work is declared, not pursued.
7.3 Elenchos: the institute is named (4.10); still to do, the contributors of its 13 openings, and a second Italian venue with an ancient-philosophy remit.
7.3a A second reading of 4.12h, and the AGPh articles not located by title, located by their running heads, so that the full-corpus D proportion can be stated; and, in the located articles, a reading past the opening for B performed unannounced, which separates the genre of the opening from the move of the article.
7.4 The combined audit. Run the same instrument on each field of the seven articles:
The common output is one line per field: what it cannot carry. The CRJ audit (#1770) is the first field; Ancient Philosophy is the second.
7.5 The model test. Run 16 November (4.17); to extend: repeated runs per journal, each abstract generated separately, so that batch context is controlled; a second model; a perturbation condition ("innovative", "boundary-pushing" abstracts) beside "plausible", to see whether D appears when asked for; and a cross-family check of the coding, the 94 calibration abstracts coded blind by a model of another family, its agreement with the lead coder logged beside the 80% within Claude. Original design: one query per journal for twenty plausible abstracts (Phronesis, Apeiron, Research in Phenomenology, Philosophy Today, Noûs, PPR), run signed out and coded A–G. If the model reproduces the field's anchored three-quarters and its D at 1–7%, that establishes distributional resemblance at the output. Resemblance alone does not confirm the transmission mechanism (3.3); that needs the recursive dependence traced. If the model's D is higher, the model is doing something the field's articles do not.
7.6 The gates, journal by journal. The census journals' own author guidelines and decline practice on AI (Noûs, Mind, PPR; CPR, Research in Phenomenology, Philosophy Today), with dates of adoption, and the search procedure recorded for any absence.
7.7 The seating. An EA-CORPORA seating of the pre-flattening record, the construction of the object, and the census evidence, so that §§4.12 and 3.1 can be read against the texts themselves (gathering begun 15 November).
The working directory of the study, ap-survey/. Its census, codes, reach outputs (without the abstracts), standing pulls, stylometry and scripts are seated in EA-CORPORA-39/12 (the census seat); the measures from the seat and the model test are attached to this deposit (seat39/, modeltest/). The directory as it stood:
census.json and census-<journal>.jsoncodes.json and codes-<journal>.jsonmeasures.txtcompare.txtabstracts-read.txtabstract-reach.jsonreach-read.txtreach5/ (codebook, reader inputs and outputs, reach5-coded.json, reach5-adjudicated.json, adjudication-log.json, tally-adj.txt)dpull.tsv, dpull2.tsv, titles-all.json, standing.txt, measures-v2.txt, pre-subtitle/cont/ (continental censuses, code_cont.py, compare-cont.txt, reachc/)stylo/ (corpus.json, stylo.py, stylo.txt)anal/ (analytic censuses, reacha/)Scripts: standing.py, census2.py, code2.py, compare.py, reach.py, reach5/tally.py, reach5/adjudicate.py, reach5/tally_adj.py.
base2/ (the second baseline: Philologus, CQ, RSF, Peitho, RPA censuses, pulls, standing2.py, standing2.txt)interim/ (Persée harvests of the Revue philosophique de Louvain and the REG, harvest_persee.py, standing_persee.py, standing_reg.py, their outputs)modeltest/ (the seven raw outputs, parse.py, gen.json, codes_in.py, verify_codes.py, codes-gen.jsonl, tally-gen.txt, compare_gen.py, compare-gen.txt, spread.py, spread.txt)seat39/ (the measures from the seat: agph_text.py, system_vocab.py, philology_text.py, heads.py, extract_openings.py, early/ with the openings, the lead codes and tally_early.py; outputs agph-text.txt, system-vocab.txt, philology-text.txt, tally-early.txt)The archive's line (2.4): #1540 The Certified Center (AXN:063A); #1573 The Wrong Unit (AXN:0665); #855 The Wolf Boy and the Language Model (AXN:0364); #751 Socially Necessary Scholarly Labor (AXN:02A7); #122 Coverage vs. Depth (AXN:02A8); EA-MPAI-IPF-01 (AXN:02D0); Diversity Contraction, doi 10.5281/zenodo.20532696.
Seated: EA-CORPORA-39 (#1771, AXN:0CA3), the texts the measures read: the Archiv für Geschichte der Philosophie 1888–1923 (39/01); the construction texts (39/02–09, with Schleiermacher, Campbell and Tennemann in 32 and Jaeger in 31); Philologus and CQ, the cited volumes (39/10–11); the census without the publishers' abstracts (39/12); the gates (39/13). Every title-level figure (§§4.2–4.5, 4.12–4.12f, 4.14, 4.14e) can be rebuilt from the census seat alone, and the full-text figures (4.12g–h) from the seat's texts. The abstract-level figures (4.6–4.11, 4.14a–d, 4.15) need the abstracts, which are not reseated: census2.py re-pulls them from Crossref, which serves them for the DOIs marked has_abstract, and every reach code's verbatim span then verifies against its abstract. Abstracts later withdrawn or altered at the source will not reproduce; that limit is the price of not reseating them.
Sources read 15 November:
Series: SERIES-EA-MMRS-TINP-01