The cliff moves when the language changes

Claim (verbatim)

Three of antiquity's great technical works break off in Greek at a cliff - and in each case the Arabic translation keeps going past it, because ninth-century Baghdad translated from exemplars upstream of the Byzantine bottleneck. Apollonius' Conics: eight books; Greek transmits I-IV, exactly the books Eutocius of Ascalon re-edited with commentary around 510, while the Banu Musa's Arabic (Thabit ibn Qurra translating V-VII) carries seven (Heiberg's edition for the Greek; G. J. Toomer, Apollonius: Conics Books V to VII, 1990); book VIII is lost everywhere, reconstructed by Halley in 1710. Diophantus' Arithmetica: thirteen books by its own preface; the Greek line kept six; in 1968 a manuscript in Mashhad yielded four more - books IV to VII of Qusta ibn Luqa's Arabic version, which by their internal difficulty and cross-references sit BETWEEN Greek III and the remaining Greek books (Jacques Sesiano's 1982 edition; Roshdi Rashed's 1984 Budé). Galen's Anatomical Procedures: fifteen books; the Greek breaks off partway through IX; the Arabic transmits the rest (Max Simon's 1906 edition of the last seven books; Duckworth's 1962 English; Garofalo's editions). The mechanism is institutional: a Greek scientific work's surviving block is what late-antique teaching editions stabilized for the classroom, and the cliff falls at the commentary's edge; the translation movement, arriving before the final Greek narrowing finished its work, caught the fuller text and preserved the far side.

Prediction clause (verbatim)

Prediction: compiling a book-by-book two-column carrier table from the named editions' transmission registers for the three works, every work will show books 1-3 extant in Greek while the translation-only books - at least two per work and at least twelve across the three - all lie beyond book 3 (primary clause: the positional split, >=2 translation-only books per work all beyond book 3; the verdict follows it); secondarily, where a named late-antique commented edition exists (Eutocius on Conics I-IV) the Greek block will coincide with its coverage exactly; a book counts Greek-extant only on continuous direct transmission, not fragments or Renaissance retro-translation, Anatomical Procedures book IX is assigned per Garofalo's break-point, and the test voids if any edition's register cannot assign every book of its work.

Kill-dataset (verbatim)

Kill: the transmission registers and prolegomena of Heiberg's Apollonius (1891-93) with Toomer's Conics V-VII (1990), Tannery's Diophantus (1893-95) with Sesiano (1982) and Rashed (1984), and Garofalo's Anatomical Procedures with Simon (1906) - a per-book carrier census with a positional (beyond-book-3) test.

Novelty / leakage triage

already answered in the literature

Each cliff is stated with its position in the standard editions: Toomer's prolegomena state that Greek transmits Conics I-IV, the books of Eutocius' commented edition, with V-VII Arabic-only and VIII lost; Sesiano and Rashed state the six Greek books against the four Arabic-only books and argue their placement after Greek III; Simon, Duckworth and Garofalo state that the Greek of Anatomical Procedures breaks in book IX with the remainder Arabic-only. The primary positional clause (>=2 translation-only books per work, all beyond book 3, >=12 in total) is guaranteed list-arithmetic on printed per-book registers, and the Eutocius-coverage coincidence is Toomer's own stated explanation; assembling the three into one table runs no new measurement.

• G.J. Toomer, Apollonius: Conics Books V to VII. The Arabic Translation of the Lost Greek Original in the Version of the Banu Musa (New York, 1990), introduction
• J. Sesiano, Books IV to VII of Diophantus' Arithmetica in the Arabic Translation Attributed to Qusta ibn Luqa (New York, 1982)
• R. Rashed, Diophante: Les Arithmetiques, livres IV-VII (Paris, 1984)
• M. Simon, Sieben Buecher Anatomie des Galen (Leipzig, 1906), with W.L.H. Duckworth, Galen on Anatomical Procedures: The Later Books (Cambridge, 1962)

Predictions

No prediction registered yet.