Inferpedia - an encyclopedia of the missing

AI-generated conjecture · below the evidence/publication boundary

← All conjectures · Texts, scribes & transmission

Manuscript Yule process

Status: Prior

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 proposed connection between two domains, generated by a language model. 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)

Manuscript Yule process. Copies per text follow preferential attachment: power-law copy counts, with survival size-biased so bestsellers over-survive superlinearly. Falsify: medieval library catalogs vs extant counts.

Kill-dataset (verbatim)

medieval library catalogs vs extant counts.

Provenance

Run: Imported conversation (verbatim harvest) · model: claude-fable-5

Origin: operator conversation with Claude Fable 5 at max effort, conducted 2026-07-03, relayed verbatim by the operator into the shepherd session on 2026-07-04. No ModelRun exists for the original generation (it happened outside the pipeline); this transcript file is the canonical capture. Transcript path: docs/generated/conjecture_harvest_fablemax_20260703.md. Model (operator-attested, not pipeline-recorded): claude-fable-5. Novelty disclaimer (verbatim, load-bearing -- rule 4): "Same caveat as before, doubled: at 100 items across all of archaeology and history, some of these will have cousins in the literature I can't check. What I can guarantee is the format — each links two things not normally linked, and each names the dataset or measurement that would kill it."

Novelty / leakage triage

Adjacent (closely related prior work exists)

Modelling manuscript transmission as a stochastic birth-death population process is published: Cisne 2005 fits demographic models to copies of medieval technical texts; unseen-species richness estimators have been applied to medieval literature loss; and recent complex-systems work treats written cultures as transmission systems. The specific formulation — preferential attachment yielding power-law copies-per-work with size-biased SUPERLINEAR over-survival of bestsellers — was not located as stated. In-house works-by-witnesses data (Pinakes, DBBE) supports a calibration test of the distributional part.

Predictions

Supported registered 2026-07-04 calibration prediction (parent triage: leaked/adjacent)

Resolution: Supported

Caveats: Calibration verdict, not a novelty claim (triage: adjacent — Cisne 2005 and the unseen-species literature model manuscript transmission demographically). Nuance that must survive into any narrative: the heavy-tail FAMILY is confirmed in both datasets, but the pure power law wins only in DBBE (alpha=2.37, inside the registered (1.5,4.0) band); in Pinakes the discretized lognormal beats the zeta by ~2,009 AIC, which is consistent with preferential-attachment-with-death or proportional-growth variants but NOT with a pure Yule process — the conjecture's specific 'power-law copy counts' wording is only partially vindicated. The superlinear size-biased survival half of the harvest conjecture is untested (needs medieval catalogue-vs-extant joins). Coverage caveats from the extracts apply (Byzantine Greek catalogue; book-epigram genre scope). Survival bias: these are EXTANT witness counts, i.e. the conjecture's process convolved with loss, not production counts.

In both in-house works-by-witnesses datasets (Pinakes Greek works: 21,513 works / 245,592 witness links; DBBE Byzantine epigram type-groups: 4,898 groups / 12,123 witnesses), the copies-per-work distribution is heavy-tailed in the preferential-attachment family rather than thin-tailed — the distributional signature the harvest conjecture requires. Calibration test only: triage is 'adjacent' (Cisne 2005; unseen-species literature), so no novelty is claimed, and the superlinear-survival half of the conjecture is NOT testable with these data.

Resolution criteria: Fit three models to each dataset's copies-per-work counts by maximum likelihood: discrete power law (zeta, xmin=1), discretized lognormal, and geometric. SUPPORTED if min(AIC_powerlaw, AIC_lognormal) < AIC_geometric - 10 in BOTH datasets AND works with k >= 10 witnesses comprise >= 0.5% of works in both. KILLED if the geometric model is within 10 AIC of the best model in EITHER dataset. Otherwise INCONCLUSIVE. If the power law wins, report its exponent; an exponent outside (1.5, 4.0) counts against the preferential-attachment reading in the narrative but does not alone kill.

Known priors disclosure: The registrant has seen the Phase-A extractor summary counts only (Pinakes: 21,513 works, 245,592 witnesses, f1=7,545 singleton works; DBBE: 4,898 groups, 12,123 witnesses) and holds the general bibliometric prior that copy-count distributions are typically heavy-tailed — which is precisely why this is registered as calibration, not discovery. The registrant has NOT seen either distribution's shape, tail mass, or any fitted model.

Maximum-likelihood fits of three discrete models (zeta power law xmin=1; discretized lognormal; geometric) to each f_k distribution, compared by AIC, exactly as pre-registered. scipy-free implementation (Euler-Maclaurin zeta, erf-based normal CDF, golden-section/coordinate optimization); script preserved with the artifact.

Dataset: In-house copies-per-work distributions: Pinakes Greek works x witnesses (21,513 works, 245,592 witness links; Byzantine Greek catalogue coverage) and DBBE book-epigram type groups (4,898 groups, 12,123 witnesses; genre-scoped). Both from committed Phase-A extract artifacts; no new ingestion.

computed 2026-07-04

Weigh in

No community feedback yet.

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.