← All conjectures · Byzantine book culture

The ninth-century needle

Medieval Dated c. 850 CE – 1000 CE Aegean & Greece

Status: Falsified

The verdict’s fine print — quoted from the resolution record: “Killed on clause B's concentration pin alone, and the kill is unusually instructive.” Read the full caveats ↓

Status is derived only from the shepherd-authored triage/prediction data above -- community submissions and claims are a separate overlay and can never change it (see the participation panel below).

This is a conjecture imagined by a language model — drawn from its trained weights and held to falsifiability, novelty, and plausibility, not to any one method: it may join two or more fields, or none. It is not an article and not evidence: it sits below the evidence/publication boundary. A quantitative prediction and a named kill-dataset are attached (when registered) so the claim stays falsifiable rather than merely evocative.

Claim (verbatim)

Between roughly 850 and 1000 the whole ancient Greek inheritance passed through a needle: the transliteration (metacharakterismos) of texts from majuscule into the new minuscule bookhand. What was transcribed joined the future; what was not was, with vanishing exceptions, finished - majuscule books stopped being read, were dismembered, scraped or shelved to rot, and Alphonse Dain's observation stands that most classical traditions descend from a single transliterated exemplar. Photios and the early Macedonian scholars sit exactly at the needle's eye, which is why he could still read what we cannot. The needle is also why, for profane literature, the earliest surviving witness of almost everything ancient is a ninth-, tenth- or eleventh-century book, while the great pre-800 codices that do survive are overwhelmingly biblical (Sinaiticus, Alexandrinus), with a handful of profane outliers like the Vienna Dioscorides and the palimpsest undertexts. The claim is that this funnel is directly visible as a spike in the Pinakes date-of-earliest-witness distribution - a one-century demographic pinch through which fifteen hundred years of writing had to pass - and that its sharpness can be measured: not 'the ninth century mattered' but a modal century holding a third of all earliest witnesses, with the pre-800 tail nearly empty and made of palimpsests.

Prediction clause (verbatim)

Prediction: restricting to profane Greek works composed before 300 CE with at least one dated codex witness in Pinakes (biblical, patristic and liturgical corpora excluded; papyrus finds excluded), fewer than 10% will have any surviving codex witness dated before 800, and the modal century of earliest-witness dates will fall within 800-1099 and contain at least 35% of all earliest witnesses (primary clause: the under-10% pre-800 share plus the modal location; the verdict follows it); disambiguation: palimpsest undertexts count at their undertext date and are expected to dominate the pre-800 tail; dating follows Pinakes century-precision fields; coverage guard: void if fewer than 500 qualifying works.

Kill-dataset (verbatim)

Kill: the in-house Pinakes extracts (works joined to earliest dated witnesses) filtered to profane pre-300 CE authors - an earliest-witness date histogram with a modal-century test.

Provenance

Run: Fresh agent generation · model: claude-fable-5

Fresh blind generation by claude-fable-5, 2026-07-17, Byzantine instrument-anchored wave against DBBE/Pinakes and the empire's self-indexing censuses (Photios' Bibliotheca, the Suda, the Excerpta Constantiniana): every kill names an open instrument and a countable operation (review-length-vs-survival quartiles, witness-count contrasts, condemned-partition survival ratios, title-list censuses, stemmatic agreement rates, collection-witness geometry, fragment-carrier attribution, book-block contiguity, earliest-witness histograms, double-notch production comparison, catena-only shares, calendar-gradient retention, per-book anthology retention, renewal-formula date lags, destroyed-carrier joins). Disjointness checked against the full corpus and all 2026-07-16/17 waves: the w11 Byzantine wave (35 items) owns the seal/PBW joins, DBBE patron-network/meter-rank/gender/disaster/debasement/gift operations, the 1204 deluxe-dedication survival discontinuity (w11-012) and post-1453 copying geography (w11-035) - the double-notch item here compares aggregate dated production rates instead; w11-028 owns seal-conditioned pre-metaphrastic retention - the Metaphrastes item here partitions by the Menologion's own calendar-density gradient (per steer, kept witness-geometric via Pinakes); fablemax-batch2-004 owns the Pinakes singleton/Chao1 census and batch3-004 the Pinakes-DBBE distribution collapse - the planned classical-vs-Byzantine witness-curve item was DROPPED for that ownership, replaced by earliest-witness-date geometry (the transliteration needle) which no item owns; the palimpsest lower-text-rate candidate was DROPPED (owned trio: w25-001 price barometer, w10-017 Sinai undertext languages, w11-027 sponsorship odds) with the Vat. gr. 73 palimpsest fact folded into the Excerpta single-witness item; Greek catena dismemberment is distinct from w02-013/019 (Pinakes polemic/council-premium operations) and w02-020 (Aquinas' Latin Catena aurea); DBBE operations here (renewal-formula chronology; destroyed-carrier join) collide with none of w07-018/019, w08-010/037, w09-039, w23-010..016, w26-008, minds-w01-011, minds-w02-006, batch2-005. Corpus 'Suda' hits outside this wave are 'Sudanic Africa' false positives; Photios, the Excerpta and the Palatine/Planudean comparison were wholly unowned. Honesty flags: MODERATE on the Photios quartile ratio, Suda title-survival shares and Planudean retention figures (thresholds cushioned in the safe direction); item 15 carries the wave's explicit Kill-(partly-not-yet-built) fallback (Pasini-catalogue back-collation). All predictions carry pinned disambiguation rules and coverage guards; thresholds far from 1 throughout.

Novelty / leakage triage

anticipated in the literature — this exact test has never been run

The needle is textbook - Dain's dictum that most classical traditions descend from a single transliterated exemplar, Irigoin's account of the ninth-century revival, Reynolds & Wilson's chapter - and the qualitative shape (earliest witnesses massing in the ninth to eleventh centuries, the pre-800 profane tail a handful of palimpsests and the Vienna Dioscorides) is stated everywhere; but no one has computed a work-level earliest-witness histogram over Pinakes with a modal-century share and an under-10% pre-800 clause at 500-work scale. The shape leaks; the histogram is new counting.

Sources cited by the triage

A. Dain, Les manuscrits (3rd ed., Paris, 1975)
J. Irigoin, 'Survie et renouveau de la litterature antique a Constantinople (IXe siecle)', Cahiers de civilisation medievale 5 (1962)
L.D. Reynolds & N.G. Wilson, Scribes and Scholars (4th ed., Oxford, 2013), chapter on the Greek East

Predictions

Killed registered 2026-07-19 variant: v1 calibration prediction (parent triage: leaked/adjacent)

Resolution: Killed

Caveats: Killed on clause B's concentration pin alone, and the kill is unusually instructive. Clause A held resoundingly: 2.80% of 1,287 works have any pre-800 witness, far under the 10% ceiling -- and the disclosed Siecle-parse noise (8 rows with unmatched parens creating spurious 1st/2nd-century dates) biases AGAINST clause A, so the clean figure is nearer 2.2%; the noise carries no weight in the verdict. Clause B's LOCATION was exactly right: the modal century is the 10th, inside the predicted 800-1099 window, and the three-century window holds 53.6% of all earliest witnesses. But the pinned single-century concentration failed: 28.98% < 35%. The registration flagged this ex ante as 'the risky half', and the reported-only latest-bound sensitivity reading (34.58%) misses the bar too -- by 0.42 points -- so the kill stands even on the conjecture-favorable unpinned parse. CALIBRATION: first kill in the proportion class, and specifically a CONCENTRATION overprice: the transliteration-bottleneck story is right (direction, location, tail-thinness all held); what died is the wager that one century alone carries >=35% of a 1,287-work histogram. Conjunctive pins multiply failure odds; concentration shares over wide histograms behave like the dead multiplicative class, not like the reliable single-proportion class. Palimpsest expectation unadjudicated per registration (instrument lacks an undertext-date field). Data source credit: Pinakes (IRHT-CNRS).

Registered against the IN-HOUSE, operator-authorized Pinakes CSV export (obtained 2026-06-18 via Pinakes' own published export feature per docs/GOAL_SCRIPTOME_PINAKES_INGEST_20260618.md; reuse basis 'Pinakes published export; operator-authorized reuse' with IRHT/Pinakes attribution; derived aggregates only, no raw-row republication): file /home/steph/pinakes_unique_rows.csv, sha256 19fed82e7b91674d90c40986de23f5c5da93c8dff31f442160f1bf73c712ae4a, 260,182 parsed content rows, columns incl. Siècle, Auteur, Oeuvre, Fonds, Cotes, BHG. The live pinakes.irht.cnrs.fr remains robots-FORBIDDEN to this agent and is NOT touched; this registration realizes the conjecture's own phrase 'using the in-house Pinakes extracts'. PREDICTION VERBATIM: "Prediction: restricting to profane Greek works composed before 300 CE with at least one dated codex witness in Pinakes (biblical, patristic and liturgical corpora excluded; papyrus finds excluded), fewer than 10% will have any surviving codex witness dated before 800, and the modal century of earliest-witness dates will fall within 800-1099 and contain at least 35% of all earliest witnesses (primary clause: the under-10% pre-800 share plus the modal location; the verdict follows it); disambiguation: palimpsest undertexts count at their undertext date and are expected to dominate the pre-800 tail; dating follows Pinakes century-precision fields; coverage guard: void if fewer than 500 qualifying works." FULL OPERATIONALIZATION PINNED: SCOPE ('profane Greek works composed before 300 CE', realized as a closed author roster over the export's controlled Auteur vocabulary — every roster member a securely pre-300-floruit, non-Christian-corpus author; anonymous category heads, all Christian/biblical/liturgical authors, and all '(pseudo)'/'quidam' Auteur strings excluded by construction): the 101 pinned exact Auteur strings ["Homerus", "Hesiodus", "Pindarus", "Theognis elegiacus", "Aeschylus", "Sophocles", "Euripides", "Aristophanes comicus", "Theocritus poeta", "Callimachus poeta", "Apollonius Rhodius", "Nicander Colophonius", "Oppianus Anazarbensis", "Oppianus Apamensis epicus", "Lycophron", "Valerius Babrius", "Dionysius Periegeta", "Herodotus historicus", "Thucydides historicus", "Xenophon historicus", "Polybius historicus", "Diodorus Siculus", "Dionysius Halicarnassensis", "Appianus", "Arrianus Flauius", "Dio Cassius", "Herodianus historicus", "Flauius Iosephus", "Strabo", "Pausanias periegeta", "Lysias", "Isocrates Atheniensis", "Isaeus Atheniensis", "Aeschines orator", "Demosthenes", "Antiphon orator", "Andocides", "Lycurgus", "Hyperides", "Dinarchus", "Demades", "Aelius Aristides", "Dio Chrysostomus", "Hermogenes rhetor", "Polyaenus rhetor", "Maximus Tyrius sophista", "Philostratus Flauius Maior", "Philostratus iunior", "Lucianus", "Plato philosophus", "Aristoteles philosophus", "Theophrastus philosophus", "Epicurus", "Epictetus", "Marcus Aurelius Antoninus imp.", "Sextus Empiricus", "Plotinus", "Porphyrius", "Alexander Aphrodisiensis", "Cebes philosophus", "Albinus (Alcinous)", "Numenius philosophus", "Cassius Longinus", "Hippocrates medicus", "Galenus medicus", "Dioscorides Pedanius", "Soranus Ephesius medicus", "Rufus Ephesius medicus", "Aretaeus Cappadox medicus", "Euclides", "Archimedes", "Apollonius Pergaeus", "Nicomachus Gerasenus mathematicus", "Diophantus Alexandrinus mathematicus", "Theon Smyrnaeus philosophus", "Cleomedes", "Autolycus astronomus", "Aristarchus Samius astronomus", "Aratus Soleus", "Claudius Ptolemaeus", "Heron Alexandrinus", "Achilles Tatius astronomus", "Artemidorus Daldianus", "Apollodorus mythographus", "Aelianus Claudius", "Aelianus tacticus", "Onasander tacticus", "Athenaeus Naucratensis", "Athenaeus mechanicus", "Iulius Pollux", "Harpocration grammaticus", "Phrynichus Atticista", "Moeris Atticista", "Dionysius Thrax", "Apollonius Dyscolus", "Herodianus Alexandrinus", "Achilles Tatius scriptor eroticus", "Chariton scriptor eroticus", "Longus scriptor eroticus", "Xenophon scriptor eroticus", "Plutarchus"]. Deliberate borderline exclusions, stated ex ante: Philo Iudaeus (scriptural-exegetical corpus), Menander comicus (Sententiae are later compilations), Aesopus (recension dating), Heliodorus scriptor eroticus (4th-c. dating live), Zosimus alchemista and Menander rhetor (c. 300), Iamblichus/Themistius/Libanius (post-300), Aristophanes Byzantius (transmitted layer is Byzantine-epitomized), all '(pseudo)' strings incl. Longinus (pseudo). WORK UNIT: distinct (Auteur, Oeuvre) pair with Auteur in the roster and non-blank Oeuvre not matching the exclusion regex /scholia|commentari|paraphras|$pseudo|spuri/i (later derivative layers filed under ancient names are not pre-300 compositions). WITNESS ROWS: the pair's export rows; the export's '\t' placeholder counts as blank. PAPYRUS/OSTRACA EXCLUSION: drop rows where Fonds+' '+Cotes matches /papyr/i, /(^|[^a-z])pap\.?($|[^a-z])/i, or /ostrac/i. CENTURY PARSE (pinned, realizing 'dating follows Pinakes century-precision fields'): from the row's Siècle string, first delete fraction markers /\(\d+\/\d+$/ (e.g. '13 (1/2)'), then extract all integers; keep those in 1..21 as century candidates; a row with no candidate is undated (excluded, counted). Row earliest-bound century = min of candidates (a range '8-9' thus ADMITS a pre-800 date — the reading unfavorable to the conjecture's <10% clause, stated ex ante). QUANTITIES: for each qualifying work w (>=1 dated, non-papyrus row): E(w) = min over its dated rows of the row earliest-bound century. Clause A: share of works with E(w) <= 8 is < 0.10 ('any surviving codex witness dated before 800' = some row century <= 8). Clause B: the modal century M of the E(w) distribution satisfies M in {9, 10, 11} AND count(E=M)/N_qualifying >= 0.35; if the mode ties, every tied century must lie in {9,10,11} and carry >= 0.35. PALIMPSEST NOTE (instrument limit, stated ex ante, non-voiding): the export carries no undertext-date field; witnesses count at Pinakes' recorded Siècle, so the prediction's palimpsest expectation is realized only to the extent Pinakes' own century field records undertext dates.

Resolution criteria — the registered fine print

Resolution criteria: COMPUTE RECIPE (blind agent, zero web access): (1) verify sha256 of /home/steph/pinakes_unique_rows.csv equals the pinned value; (2) parse with the std-lib csv module (utf-8-sig, DictReader; note the header spells 'Röle'); (3) apply the pinned roster, work-unit rule, exclusions, and century parse exactly as registered; (4) compute N_qualifying, the E(w) distribution, clause A's share, and clause B's mode and modal share. CLAUSE PRECEDENCE: (1) INCONCLUSIVE_BY_ACCESS if the file is absent or its sha256 differs; (2) INCONCLUSIVE_BY_DESIGN if N_qualifying < 500 (the prediction's own coverage guard, quoted verbatim; registration-time scout puts the mass at ~1,288, a 2.6x clearance); (3) SUPPORTED iff clause A AND clause B both hold; (4) REFUTED otherwise. REPORTED, NON-VERDICT: the full E(w) century histogram; the pre-800 work list with shelfmarks (the conjecture expects palimpsests to dominate this tail — reported for the write-up, not adjudicated, the instrument lacking an undertext field); per-author qualifying-work counts; excluded-row tallies (papyrus/ostraca, undated, exclusion-regex); and a sensitivity line recomputing A and B with row century = MAX of candidates (latest-bound reading) — stated ex ante as reported-only, so neither parse direction can be shopped post hoc. Compute firewalled from Dain/Irigoin/Reynolds-Wilson expectations, from the registrant's roster-mass scout beyond the numbers disclosed here, and from this batch's other rows. ATTRIBUTION REQUIREMENT: any published surface of this resolution credits Pinakes (IRHT-CNRS) as the data source.

Known-priors disclosure — what the registrant already knew

Known priors disclosure: Triage 2026-07-17 graded ADJACENT: the ninth-century transliteration bottleneck is textbook direction (Dain, Les manuscrits; Irigoin on the ninth-century revival; Reynolds & Wilson, Scribes and Scholars) and 'the shape leaks', but no one has computed a work-level earliest-witness histogram with a modal-century share and an under-10% pre-800 clause at 500-work scale — the histogram is new counting. The registrant states the textbook prior openly: direction expected pro-conjecture; the live uncertainty is whether a SINGLE century holds >=35% of ~1,288 works' earliest witnesses (clause B's concentration is the risky half) and whether the pre-800 tail stays under 10% when range-dates are read at their earliest bound (the pinned, conjecture-unfavorable reading). SCOUTED DENOMINATOR MASSES (in-house file reads, disclosed per lesson 5): 260,182 export rows; 2,860 distinct Auteur values; Siècle non-empty on every row; Siècle format-shape inventory ('N' 171,143 rows; 'N (N/N)' 27,445; 'N-N' 24,609; 'N med./ex./in.' ~25,000; comma-lists ~2,900; the parse rule above covers all shapes); the 101 roster authors carry 31,114 rows and 1,288 distinct qualifying works post-exclusion. NOT SCOUTED (effect quantities, untouched): any Siècle value distribution for roster rows, any per-work earliest-witness century, any pre-800 membership. Roster construction saw ONLY per-author row/work masses, never dates. PROVENANCE/PERMISSION: the live Pinakes site is robots-forbidden to this agent (lesson 4 register) and was not contacted; the in-house export is operator-authorized reuse of Pinakes' own published export (goal doc 2026-06-18), used derived-only with attribution. Known instrument caveats from the goal doc, disclosed: the export carries no Diktyon/Pinakes ID numbers (identity rests on controlled Auteur/Oeuvre label strings) and the Identifiants column is empty; the Röle column is empty export-wide, so no role filtering is possible. CALIBRATION CLASS: clause A is a PROPORTION pin and clause B a LOCATION(mode)+PROPORTION pin — both in the historically reliable class; no multiplicative pin anywhere in the verdict (lesson-6-preferred shape). House DBBE compute (renewal-lag, Zenodo dbbe_archive.sqlite) shares no quantities with this item.

Method and dataset — how it was measured

Firewalled blind compute (claude-sonnet-5) executed the registered operationalization verbatim (compute artifact docs/generated/compute_pinakes009_histogram_20260719.json): closed Auteur roster, (Auteur,Oeuvre) work units minus the derivative-layer regex, papyrus/ostraca row exclusion, earliest-bound century parse with the pinned fraction-marker deletion. Verdict authored by claude-fable-5 per the registered clause precedence: (1) access guard passed (sha match); (2) coverage guard passed (1,287 >= 500); (3) SUPPORTED iff clause A (pre-800 share < 0.10) AND clause B (modal century in {9,10,11} with modal share >= 0.35); (4) otherwise REFUTED.

Dataset: In-house operator-authorized Pinakes CSV export (IRHT-CNRS attribution), sha256 verified 19fed82e7b91674d90c40986de23f5c5da93c8dff31f442160f1bf73c712ae4a, 260,182 rows; 101-author pinned profane roster; N_qualifying=1,287 works (>=1 dated non-papyrus witness) against the registered 500-work coverage guard (2.6x clearance).

computed 2026-07-19

Weigh in

No community feedback yet.

New here? Create an account first

Create an account or sign in and your feedback is tied to you — you can track it, get replies, and claim this conjecture so others know you’re working on it. Prefer not to? Just leave your take below as a guest — only the name you type is shown.

Add your take

Posted immediately (spam is removed). Community feedback is never an adjudicated verdict and never changes this conjecture's triage label or status above.