Nine orator corpora seated to supply a missing denominator. 606,441 tokens across 159 works from commit-pinned Perseus. The authorship investigation has produced two positive results that are both rates without comparison classes: the naming-gap reflex grouping Plato and Aristotle at 10.5x, and a completion structure at five of six matched slots. The received control would be Theophrastus, whose philosophical works are unobtainable here. The orators supply the denominator by another route: Isaeus taught Demosthenes, a link documented by Dionysius of Halicarnassus and the Lives of the Ten Orators, and around it stands a field of same-genre contemporaries with no teaching relation — including Aeschines, who was Demosthenes's rival and certainly not his student, as a negative control. One positive link, one negative, and a population. Two locus probes failed on modern titles absent from the ancient text and were corrected rather than asserted. One caveat is declared at the seat: Hyperides survives on papyrus with lacunae and carries word-splits at the breaks, making token-level measures unsafe there without re-joining.
Wiki Article
EA-CORPORA-08 seats the nine canonical Attic orators, and it exists to supply a number nobody has measured. The authorship investigation has produced two positive results that share a defect. The naming-gap reflex groups Plato and Aristotle at ten and a half times the rate of comparison authors, and the completion structure shows Aristotle finishing executions Plato abandoned at five of six matched slots, twice with the same figure. Both are rates. Neither has a denominator, because nobody has ever measured what a master-student pair looks like on such a scale. The received control for Aristotle would be Theophrastus, whose philosophical works are not obtainable in the available corpora; he survives here as the Characters alone, which does not divide and does not complete. The orators supply the denominator by another route. Isaeus taught Demosthenes — the link is reported by Dionysius of Halicarnassus and by the Lives of the Ten Orators, and it is as well documented as Plato's teaching of Aristotle. Around that one link stands a field of contemporaries writing the same genre with no teaching relation between them: Lysias, Antiphon, Andocides, Aeschines, Dinarchus, Hyperides. Isocrates sits at one remove, reported as Isaeus's own teacher. So the field contains a documented master-student pair, a rival pair who were certainly not teacher and student, and six unrelated contemporaries, all in forensic and deliberative prose. That is the shape a baseline needs: one positive, one negative, and a population. Two locus probes failed and were corrected rather than asserted. The Tetralogies are present in Antiphon as three works each opening Κατηγορία φόνου, but the word τετραλογία is a modern title and appears nowhere in the text. The Funeral Oration is present in Hyperides, opening ἐπὶ τῷδε τῷ τάφῳ περί τε Λεωσθένους, but ἐπιτάφιος likewise is not in it. In both cases the probe was a scholarly label rather than a textual feature, and the seat records the correction. One caveat is declared at the seat rather than left to be discovered. Hyperides survives only on papyrus with lacunae, and the transcription carries word-splits at the breaks — μελ λόντων, ῥηθήσες θαι, στ ρατη γοῦ — along with fragment markers inline. Token-level measures on that seat are unsafe without re-joining. The archive's practice is that a seat declares what will break on it.
EA-CORPORA-08 — The Attic Orators: A Documented Master-Student Pair Inside a Field of Same-Genre Non-Links
EA-CORPORA-08 — The Attic Orators
seat
corpus
works
lines
role
01
isaeus
12
677
teacher of Demosthenes
02
demosthenes
63
8,449
student of Isaeus
04
isocrates
29
8,354
teacher of Isaeus (one remove)
06
aeschines
3
1,093
rival of Demosthenes — negative control
03
lysias
33
2,745
unrelated contemporary
05
antiphon
6
475
unrelated, earliest
07
andocides
4
519
unrelated contemporary
08
dinarchus
3
247
later, unrelated
09
hyperides
6
3,218
contemporary, unrelated
Total: 159 works, 606,441 tokens. Data is published at api/corpora.json; each seat carries its own source.json and SHA-256 manifest.
EA-CORPORA-08 seats the nine canonical Attic orators, and it exists to supply a number nobody has measured.
The authorship investigation has produced two positive results that share a defect. The naming-gap reflex groups Plato and Aristotle at ten and a half times the rate of comparison authors, and the completion structure shows Aristotle finishing executions Plato abandoned at five of six matched slots, twice with the same figure. Both are rates. Neither has a denominator, because nobody has ever measured what a master-student pair looks like on such a scale. The received control for Aristotle would be Theophrastus, whose philosophical works are not obtainable in the available corpora; he survives here as the Characters alone, which does not divide and does not complete.
The orators supply the denominator by another route. Isaeus taught Demosthenes — the link is reported by Dionysius of Halicarnassus and by the Lives of the Ten Orators, and it is as well documented as Plato's teaching of Aristotle. Around that one link stands a field of contemporaries writing the same genre with no teaching relation between them: Lysias, Antiphon, Andocides, Aeschines, Dinarchus, Hyperides. Isocrates sits at one remove, reported as Isaeus's own teacher. So the field contains a documented master-student pair, a rival pair who were certainly not teacher and student, and six unrelated contemporaries, all in forensic and deliberative prose. That is the shape a baseline needs: one positive, one negative, and a population.
Two locus probes failed and were corrected rather than asserted. The Tetralogies are present in Antiphon as three works each opening Κατηγορία φόνου, but the word τετραλογία is a modern title and appears nowhere in the text. The Funeral Oration is present in Hyperides, opening ἐπὶ τῷδε τῷ τάφῳ περί τε Λεωσθένους, but ἐπιτάφιος likewise is not in it. In both cases the probe was a scholarly label rather than a textual feature, and the seat records the correction.
One caveat is declared at the seat rather than left to be discovered. Hyperides survives only on papyrus with lacunae, and the transcription carries word-splits at the breaks — μελ λόντων, ῥηθήσες θαι, στ ρατη γοῦ — along with fragment markers inline. Token-level measures on that seat are unsafe without re-joining. The archive's practice is that a seat declares what will break on it.
Two corpora seated to close a debt. 07/01 Herodotus, Historiae, nine books, 184,945 tokens, 4,990 lines. 07/02 Thucydides, Historiae, eight books, 150,165 tokens, 13,528 lines. Both from commit-pinned Perseus canonical-greekLit. Both had been used unseated: the authorship measurement register's empirical null of sixty-one matched Greek prose pairs drew seventeen of its units from these two authors out of a census checkout rather than from the archive, so its most expensive artifact rested partly on material the archive did not hold. They supply the floor of the measurement — Thucydides at 0.0097 and Herodotus at 0.0103 within-author dispersion, the two lowest recorded, which is what makes the Aristotelian 0.1033 legible as anomalous rather than merely large — and the ceiling, since Herodotus against Thucydides gives the highest standardised separation in the sample at 8.867, exact contemporaries in one genre, against which Plato and Aristotle's 1.975 at rank seven of thirty-six is read. The Thucydides seat also records a locus-verification failure and its correction: an initial probe on the Funeral Oration returned zero because the modern section title is not a word in the text, and the locus was replaced rather than asserted.
Wiki Article
EA-CORPORA-07 seats Herodotus and Thucydides, and the reason is a debt rather than an ambition. Both were used unseated. The authorship measurement register built its empirical null from sixty-one matched Greek prose pairs, and Herodotus and Thucydides supplied seventeen of the units in that pool — drawn from a census checkout rather than from the archive, which meant the most expensive artifact the investigation produced rested partly on material the archive did not hold. This seating closes that. What they contribute is a floor. Thucydides has the lowest within-author dispersion measured anywhere in this work, 0.0097 across eight books; Herodotus is second at 0.0103 across nine. Those two numbers are what make the Aristotelian figure of 0.1033 legible as anomalous rather than merely large: a historian writing a continuous narrative over eight books varies by a hundredth, and the Aristotelian corpus varies by a tenth. Without a floor there is no ceiling. They also supply the ceiling. Herodotus against Thucydides — exact contemporaries, one genre, one subject, one war between them — yields a standardised separation of 8.867, the highest in the sample. That is what two genuinely different authors of the same period and kind look like when their internal variance is small. Plato against Aristotle, measured the same way, comes out at 1.975 and ranks seventh of thirty-six. The historians are the reason that comparison means anything. One methodological note is recorded in the Thucydides seat. Its declared loci were checked by stem search against the seated text before being declared, and the first probe failed: ἐπιτάφ- returns zero, because the Funeral Oration is present in book two but the modern section title is not a word in the text. The locus was replaced rather than asserted. The archive's practice is that a declared locus must resolve, and a probe that misses is a fact about the probe until it is shown to be a fact about the text.
EA-CORPORA-07 — The Seventh Seating: Herodotus and Thucydides, the Floor and the Ceiling of the Authorship Null
EA-CORPORA-07 — Herodotus and Thucydides
The seats
seat
corpus
lines
language
licence
07/01
herodotus
4,990
Ancient Greek (Ionic)
CC BY-SA (Perseus); underlying edition public
07/02
thucydides
13,528
Ancient Greek (Attic)
CC BY-SA (Perseus); underlying edition public
Data is published as a sidecar index at api/corpora.json with per-seat links; each seat carries its own source.json and SHA-256 MANIFEST.sha256 at alexanarch.org/data/corpora/<corpus>/, and a rendered reading room at traininglayerliterature.org/originals/<corpus>/.
EA-CORPORA-07 seats Herodotus and Thucydides, and the reason is a debt rather than an ambition.
Both were used unseated. The authorship measurement register built its empirical null from sixty-one matched Greek prose pairs, and Herodotus and Thucydides supplied seventeen of the units in that pool — drawn from a census checkout rather than from the archive, which meant the most expensive artifact the investigation produced rested partly on material the archive did not hold. This seating closes that.
What they contribute is a floor. Thucydides has the lowest within-author dispersion measured anywhere in this work, 0.0097 across eight books; Herodotus is second at 0.0103 across nine. Those two numbers are what make the Aristotelian figure of 0.1033 legible as anomalous rather than merely large: a historian writing a continuous narrative over eight books varies by a hundredth, and the Aristotelian corpus varies by a tenth. Without a floor there is no ceiling.
They also supply the ceiling. Herodotus against Thucydides — exact contemporaries, one genre, one subject, one war between them — yields a standardised separation of 8.867, the highest in the sample. That is what two genuinely different authors of the same period and kind look like when their internal variance is small. Plato against Aristotle, measured the same way, comes out at 1.975 and ranks seventh of thirty-six. The historians are the reason that comparison means anything.
One methodological note is recorded in the Thucydides seat. Its declared loci were checked by stem search against the seated text before being declared, and the first probe failed: ἐπιτάφ- returns zero, because the Funeral Oration is present in book two but the modern section title is not a word in the text. The locus was replaced rather than asserted. The archive's practice is that a declared locus must resolve, and a probe that misses is a fact about the probe until it is shown to be a fact about the text.
EA-CORPORAcorpus seatingKierkegaardpseudonymityVictor EremitaJohannes de SilentioNicolaus NotabeneEnten-Ellersymboloncalibrationground truthXenophonAristophanesChaerephon problemedge figuresno-floor propertygenre control
Description
Three corpora in two relations to the authorship question. 06/03 Kierkegaard, 20 texts across five declared voices — Victor Eremita's Enten-Eller with A's and B's papers kept apart as the volumes keep them, Johannes de Silentio's Frygt og Baeven, Nicolaus Notabene's Forord, and the signed works — 49,896 lines from Danish Wikisource and archive.org. He is the second case with documented ground truth, having acknowledged the pseudonymous authorship in print in 1846 and 1851, which with Pessoa's 1935 letter makes a calibration set rather than an anecdote. Enten-Eller is the symbolon in plainest form: a token broken in two, neither half arguing the book, the fit between them doing so and stated nowhere inside either. Coverage is partial and declared: seven pseudonymous works have no open Danish full-text source at the pinned access. 06/01 Xenophon, 14 works, and 06/02 Aristophanes, 11 plays, are seated HELD APART — candidate positions at the configuration's edge, not admitted, because the method that would admit them has no internal stopping rule. Xenophon is nonetheless the genre calibrator, one hand across four registers, whose decomposition corrected two of the archive's own measurements; Aristophanes is the documented case that authorial identity was obscurable in this period and circle, his first three plays produced under other men's names, and the stager of Chaerephon whose Suda notice descends from the scholia to his own Clouds.
Wiki Article
EA-CORPORA-06 seats three corpora that stand in two different relations to the archive's authorship question, and the difference between those relations is the point of seating them together. Kierkegaard enters at 06/03 as the second calibration case. He acknowledged the pseudonymous authorship in print — in "A First and Last Explanation" appended to the Concluding Unscientific Postscript in 1846, and again in On My Work as an Author in 1851 — which makes the grouping documented rather than inferred, exactly as Pessoa's is by the letter of January 1935. Two independent configurations with disclosure make a calibration set rather than an anecdote, and no instrument claiming to detect designed differentiation means anything until it has been run against both. Enten-Eller is the reason he matters more than the count of his seated pages suggests. It is the symbolon in its plainest form: a token broken in two, A's papers and B's, discovered in a desk by an editor who is himself a construction. Neither half argues the book. The fit between them does, and the fit is not stated anywhere inside either half. That is the shape a heteronymic configuration takes, and if the Socratic-Platonic-Aristotelian corpus is one, it is the founding instance of the shape that Kierkegaard and Pessoa are later, smaller, and documented attempts to recover. The seat's coverage is partial and says so. Four pseudonymous works of roughly a dozen are here. Gjentagelsen, Begrebet Angest, the two Climacus books, Stadier paa Livets Vei and the two Anti-Climacus books are absent, because no open Danish full-text source was found for them at the pinned access. Three of nineteen Wikisource pages did not resolve. Xenophon and Aristophanes enter at 06/01 and 06/02 in a different relation entirely: held apart. They are candidate positions at the configuration's edge, and the archive declines to admit them. The method that would admit them has no internal stopping rule — every criterion that might exclude an adjacent figure is either a feature the method has already reclassified as constructed, or one the admitted positions would also fail. Candidacy is not membership, and the discipline is imposed from outside rather than derived. Their evidentiary value is real regardless. Xenophon is the genre calibrator: one hand across Socratic dialogue, history, encomium and technical treatise, which is the only way to separate a genre effect from an author effect in this feature space, and a decomposition that corrected two of the archive's own measurements. Aristophanes is the documented case that authorial identity was obscurable in this exact period and circle — his first three plays produced under other men's names, the practice described by him at Wasps 1017-22, continued after he had a name of his own, so that the civic record carries the producer rather than the poet. He also stages Chaerephon, whose fullest biographical notice in the Suda is sourced from the scholia to Clouds 144, which places the record of the man downstream of the comedy that put him on stage. One technical finding is recorded in the Kierkegaard seat. Danish Wikisource pages for Enten-Eller are djvu transclusions: prop=revisions returns only a pages-index stub, so a fetch built on raw wikitext yields nothing while reporting no error. action=parse resolves them. This is the second corpus in two days whose retrieval silently returned almost nothing through the obvious API, and the seat records the working route so the next attempt does not spend the discovery again.
EA-CORPORA-06 — The Sixth Seating: Kierkegaard as Second Calibration, and Xenophon and Aristophanes Held Apart
EA-CORPORA-06 seats three corpora that stand in two different relations to the archive's authorship question, and the difference between those relations is the point of seating them together.
Kierkegaard enters at 06/03 as the second calibration case. He acknowledged the pseudonymous authorship in print — in "A First and Last Explanation" appended to the Concluding Unscientific Postscript in 1846, and again in On My Work as an Author in 1851 — which makes the grouping documented rather than inferred, exactly as Pessoa's is by the letter of January 1935. Two independent configurations with disclosure make a calibration set rather than an anecdote, and no instrument claiming to detect designed differentiation means anything until it has been run against both.
Enten-Eller is the reason he matters more than the count of his seated pages suggests. It is the symbolon in its plainest form: a token broken in two, A's papers and B's, discovered in a desk by an editor who is himself a construction. Neither half argues the book. The fit between them does, and the fit is not stated anywhere inside either half. That is the shape a heteronymic configuration takes, and if the Socratic-Platonic-Aristotelian corpus is one, it is the founding instance of the shape that Kierkegaard and Pessoa are later, smaller, and documented attempts to recover.
The seat's coverage is partial and says so. Four pseudonymous works of roughly a dozen are here. Gjentagelsen, Begrebet Angest, the two Climacus books, Stadier paa Livets Vei and the two Anti-Climacus books are absent, because no open Danish full-text source was found for them at the pinned access. Three of nineteen Wikisource pages did not resolve.
Xenophon and Aristophanes enter at 06/01 and 06/02 in a different relation entirely: held apart. They are candidate positions at the configuration's edge, and the archive declines to admit them. The method that would admit them has no internal stopping rule — every criterion that might exclude an adjacent figure is either a feature the method has already reclassified as constructed, or one the admitted positions would also fail. Candidacy is not membership, and the discipline is imposed from outside rather than derived.
Their evidentiary value is real regardless. Xenophon is the genre calibrator: one hand across Socratic dialogue, history, encomium and technical treatise, which is the only way to separate a genre effect from an author effect in this feature space, and a decomposition that corrected two of the archive's own measurements. Aristophanes is the documented case that authorial identity was obscurable in this exact period and circle — his first three plays produced under other men's names, the practice described by him at Wasps 1017-22, continued after he had a name of his own, so that the civic record carries the producer rather than the poet. He also stages Chaerephon, whose fullest biographical notice in the Suda is sourced from the scholia to Clouds 144, which places the record of the man downstream of the comedy that put him on stage.
One technical finding is recorded in the Kierkegaard seat. Danish Wikisource pages for Enten-Eller are djvu transclusions: prop=revisions returns only a pages-index stub, so a fetch built on raw wikitext yields nothing while reporting no error. action=parse resolves them. This is the second corpus in two days whose retrieval silently returned almost nothing through the obvious API, and the seat records the working route so the next attempt does not spend the discovery again.
erratumEA-CORPORA-05Pessoaorthonymheteronymleave-one-outcross-validationhold-outanalysis-time versus seat-timedeclared erratum target
Description
Corrects deposit #1560, which seated the Pessoa corpus with the orthonym withheld on the stated ground that seating it would foreclose the intended leave-one-out test. That reasoning is withdrawn. Withholding a voice at seat time does not protect a leave-one-out design; it prevents its validation, because with three voices seated a hold-out leaves two, two points cannot triangulate a third, and the design yields one fold where it requires four — removing the only check that matters, whether the procedure also recovers a voice nobody regards as special. All four voices are now seated, 493 of the 496 orthonym pages resolving, and each voice is withheld in turn at analysis time instead. The seating also exposes a distinction the withholding had hidden: under Pessoa's own usage the orthonym is the position bearing the civil name, while under the archive's functional definition it is whichever position bears the founding gesture, which in this configuration is arguably Caeiro, the master the other voices acknowledge. The two definitions do not pick out the same voice, and the seat records both without settling. Canonical bytes of #1560 unchanged; the correction is registry-level with its target declared in related_deposits.
Wiki Article
Deposit #1560 seated the Pessoa corpus with the orthonym withheld, and gave a reason: that seating Fernando Pessoa's own pages alongside the three heteronyms would foreclose the leave-one-out test the archive intends to run. This erratum withdraws that reasoning and seats the orthonym. The reasoning was wrong in a way worth stating precisely, because the error is a general one. A leave-one-out test withholds a voice and asks whether the structure of the remaining voices implies it. Withholding at seat time appears to protect that test and in fact destroys its validation. With three voices seated, holding one out leaves two, and two points cannot triangulate a third. More seriously, it yields a single fold where the design requires four, and it removes the only check that matters: whether the procedure also recovers a voice nobody regards as special. A method that succeeds on Caeiro and has never been tried on Reis or Campos has told us something about Caeiro, not about method. The hold-out is therefore analytical, not archival. All four voices are seated, and each is withheld in turn at analysis time — 123 Caeiro, 60 Reis, 107 Campos, and 493 of the 496 pages listed under the orthonym's category. Three did not resolve and their absence is stated in the seat rather than closed over. One thing the seating exposes that the withholding had hidden. Under Pessoa's own usage the orthonym is the position bearing the civil name, which is Pessoa-himself. Under the functional definition the archive works with, the orthonym is whichever position bears the project's founding gesture — and in this configuration that is arguably Caeiro, the master the other voices acknowledge and organise themselves around. The two definitions do not pick out the same voice here. The seat records both readings and settles neither, because which position is the orthonym is exactly what an instrument would have to determine rather than assume. The canonical bytes of #1560 are unchanged. The correction is registry-level and its target is declared in related_deposits, which is the form the archive's errata surface was built to make possible and which most of its earlier errata do not yet use.
Erratum to EA-CORPORA-05: The Hold-Out Is Analytical, Not Archival — Seating the Pessoa Orthonym (corrects #1560)
Deposit #1560 seated the Pessoa corpus with the orthonym withheld, and gave a reason: that seating Fernando Pessoa's own pages alongside the three heteronyms would foreclose the leave-one-out test the archive intends to run. This erratum withdraws that reasoning and seats the orthonym.
The reasoning was wrong in a way worth stating precisely, because the error is a general one. A leave-one-out test withholds a voice and asks whether the structure of the remaining voices implies it. Withholding at seat time appears to protect that test and in fact destroys its validation. With three voices seated, holding one out leaves two, and two points cannot triangulate a third. More seriously, it yields a single fold where the design requires four, and it removes the only check that matters: whether the procedure also recovers a voice nobody regards as special. A method that succeeds on Caeiro and has never been tried on Reis or Campos has told us something about Caeiro, not about method.
The hold-out is therefore analytical, not archival. All four voices are seated, and each is withheld in turn at analysis time — 123 Caeiro, 60 Reis, 107 Campos, and 493 of the 496 pages listed under the orthonym's category. Three did not resolve and their absence is stated in the seat rather than closed over.
One thing the seating exposes that the withholding had hidden. Under Pessoa's own usage the orthonym is the position bearing the civil name, which is Pessoa-himself. Under the functional definition the archive works with, the orthonym is whichever position bears the project's founding gesture — and in this configuration that is arguably Caeiro, the master the other voices acknowledge and organise themselves around. The two definitions do not pick out the same voice here. The seat records both readings and settles neither, because which position is the orthonym is exactly what an instrument would have to determine rather than assume.
The canonical bytes of #1560 are unchanged. The correction is registry-level and its target is declared in related_deposits, which is the form the archive's errata surface was built to make possible and which most of its earlier errata do not yet use.
