Ars Inquirendi

AI-generated conjecture · below the evidence/publication boundary

← All conjectures · Literature & poetics

Moerbeke's dividend

Status: Anticipated · untested

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)

Moerbeke's dividend. Joins the chronology of Thomas Aquinas's writing career to the arrival curve of the new Greek-Latin Aristotle: early on, Aquinas met much of Aristotle through florilegia, commentary lemmata, and older versions, quoting at second hand; as William of Moerbeke's literal translations and Italian court access accumulated, direct consultation became cheap. Citation behaviour should show a career-long conversion from mediated to direct contact for Aristotle specifically, while authorities with no new translation supply, such as Augustine, hold flat.

Prediction clause (verbatim)

Across composition-date windows in the judged citation-contact corpus, primary clause: the direct-quotation share among Aristotle contacts rises with date strongly enough that Spearman's rho between window midpoint and direct share exceeds +0.5, with the late-career direct share at least 15 percentage points above the early-career share. Secondary: the same statistic for Augustine contacts stays within 5 points of flat. The verdict follows the primary clause.

Kill-dataset (verbatim)

Kill: the in-house Aquinas citation corpus (1,424 judged citation contacts with cited authority, direct-vs-mediated quotation, and composition-date windows).

Nobody has run this test. The kill-data is named above. If you can run it — or you know the paper that already settles it — claim the kill or submit the prior scholarship. Kills and prior scholarship are credited here, by name, as they come in.

On Inferpedia

This conjecture has been linked to the following subject pages on Inferpedia — an encyclopedia of the missing, now in limited preview.

Provenance

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

Generated by a fresh Fable-tier instance (claude-fable-5) at maximum effort with generation-first blindness (no repo reads, no web searches, no DB queries; single packet Write was the only tool use); title list supplied at launch, titles only, no verdicts or dossiers seen (sidecar docs/generated/conjecture_fresh_fable_w01_titlelist_20260708.md); prompt pre-committed in docs/generated/conjectures_1001_wave_ledger.md and docs/generated/conjecture_fresh_fable_w01_prompt_20260708.txt. Wave W01 of docs/GOAL_CONJECTURES_1001_CAMPAIGN_20260708.md. Novelty unverified by construction.

Novelty / leakage triage

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

The chronology of Aquinas's progressive adoption of Moerbeke's translations is minutely documented (Gauthier's Leonine prefaces; Hankey's dating tables), anticipating the direction; but no study computes a career-long mediated-to-direct conversion curve with a flat Augustine control. The Spearman-rho operationalization on a judged contact corpus is un-run.

Predictions

Inconclusive registered 2026-07-11 calibration prediction (parent triage: leaked/adjacent)

Resolution: Inconclusive

Caveats: The point estimate is stated plainly first: under the registered rule every Aristotle contact in every window classifies DIRECT (100.0 / 100.0 / 100.0), RISE = 0.0 pp, and the registered KILLED trigger reads RISE <= 0. I am not applying that trigger, and here is why, without varnish: the operator's instrument note shows the corpus's via_work_title labels are canonical translation references, not historical channels — 'Moerbeke's revised Nicomachean Ethics' appears as the via-work for 20 Scriptum contacts dated c.1254, decades before that translation existed. The field the registered rule quantifies over carries zero channel variance by construction; the classifier is degenerate, and RISE = 0 is a constant of the labeling scheme, not a measurement of Aquinas's translation uptake. A kill on this number would claim falsification of a mechanism the instrument never touched. The registered rule's domain presupposed historically valid channel labels, and that presupposition failed, so the case falls outside the rule rather than under its kill clause. Stated against myself: this choice is the verdict-favorable one for the shepherd's own conjecture; the registration should have carried a via-label-validity guard and did not — that omission is mine, and INCONCLUSIVE-by-instrument-failure is a weaker outcome than the registration promised the public. The Augustine control (0.0 pp delta, formally PASS at n=33 and n=185) is tainted by the same constant-label problem and certifies nothing. A better corpus build would record the historical versio actually underlying each contact (vetus vs. Moerbeke revision, per the Leonine apparatus) rather than a canonical reference edition; the residual dating-table imperfections the operator noted elsewhere (the escaped 'vulgate' subset, the crude authority normalization) do not bear on this row's failure, which is entirely the via-label anachronism.

Public prediction (kept in substance): 'Across composition-date windows in the judged citation-contact corpus, primary clause: the direct-quotation share among Aristotle contacts rises with date strongly enough that Spearman's rho between window midpoint and direct share exceeds +0.5, with the late-career direct share at least 15 percentage points above the early-career share. Secondary: the same statistic for Augustine contacts stays within 5 points of flat. The verdict follows the primary clause.' Restated for THIS corpus as it exists (2,245 judged contacts, not the 1,424 named in the public kill text; the registration binds to the corpus as built): the composition-date windows are exactly three — Scriptum super Sententiis Lib. II (c. 1252-56, midpoint 1254), De veritate QQ. XIII-XXIX (c. 1257-59, midpoint 1258), Summa Theologiae Ia (c. 1266-68, midpoint 1267). With three windows, Spearman's rho > +0.5 is arithmetically possible only as rho = +1, so the rho clause is registered as: the Aristotle direct-contact share strictly increases across the three host works in composition order. 'Early-career share' = the Scriptum II window; 'late-career share' = the ST Ia window, and I register NOW that this is a truncation: the public text presumed windows running to 1273 (the literal-commentary period, where the mechanism predicts its fullest expression), while this corpus's last window closes c. 1268, at the early-adoption stage of the Moerbeke versions. The 15-point magnitude clause is therefore truncation-exposed and cannot kill on its own (see resolution criteria); the direction clause is NOT truncation-exposed, because Aquinas's uptake of Moerbeke's translations is documented as already under way by ST Ia. OPERATIONALIZATION OF 'DIRECT', fixed before computation (translation-channel reading): an Aristotle contact counts as DIRECT iff its mediation coding is 'direct', OR it is coded 'mediated' but the via_work_title is itself a Latin translation of the same Aristotelian work being cited (translator recorded; the via-channel is translational, e.g. a Moerbeke-era or older versio of the cited text) — a translation is a channel to Aristotle's text, not an intermediary authority. A contact counts as MEDIATED iff it is coded 'mediated' with a via-work that is a distinct authorial intermediary (Averroes, Avicenna, the Liber de causis, Boethius, a florilegium, the Lombard, a patristic quotation, etc.). Cited-authority attribution follows the corpus's recorded authority field as-is, with no reclassification (e.g. Liber de causis rows count as Aristotle only if the corpus recorded them as Aristotle). Denominator per window: all Aristotle contacts in that host work with non-missing mediation coding. Secondary (control): the same statistic for Augustine contacts, end-to-end |ST Ia share minus Scriptum II share| <= 5 percentage points; reported, does not flip the verdict.

Resolution criteria: Compute per host work: d_w = (Aristotle contacts classified DIRECT under the registered translation-channel rule) / (Aristotle contacts with non-missing mediation coding), for w in {Scriptum II, De veritate XIII-XXIX, ST Ia}. Let RISE = d_STIa - d_ScriptumII (in percentage points). SUPPORTED: d strictly increasing across the three works in composition order (d_Scriptum < d_DeVer < d_STIa, equivalent to rho = +1 > +0.5) AND RISE >= 15 pp. KILLED: RISE <= 0 (flat or falling end-to-end across the available 1252-68 window; the mechanism predicts visible adoption by ST Ia, so no rise at all within this window falsifies). INCONCLUSIVE: every other pattern — i.e. RISE > 0 but < 15 pp (magnitude clause truncated by the corpus ending c. 1268, before the 1268-73 commentary period the public text presumed; an undersized rise cannot separate 'mechanism false' from 'onset later than the corpus reaches'), or RISE >= 15 pp with a non-monotone middle window (rho = +0.5, failing the public rho clause; windows 1 and 2 sit only ~4 years apart, within dating noise, so the inversion is not treated as evidence against the mechanism either). DATA GUARDS (any triggers INCONCLUSIVE): fewer than 30 Aristotle contacts with non-missing mediation coding in any single host work; or mediation coding missing on more than 20% of Aristotle contacts in any window. SECONDARY CONTROL, reported alongside but never flipping the verdict: PASS if |Augustine d_STIa - d_ScriptumII| <= 5 pp (guard: >= 30 Augustine contacts per endpoint window, else control 'not evaluable'). Verdict follows the primary clause, per the public text.

Known priors disclosure: Shepherd-adjacent. The direction of this prediction is strongly anticipated by minutely documented scholarship the shepherd knew when authoring: Gauthier's Leonine prefaces and Hankey's dating tables chart Aquinas's progressive adoption of Moerbeke's translations work by work. What is un-run is the quantified career-long mediated-to-direct conversion curve on a judged contact corpus, with the Spearman operationalization and a flat-Augustine control. Additional adjacency: the shepherd authored the conjecture, oversaw the pipeline that built the judged corpus, and is authoring this registration; no resolution statistic has been computed or seen — the direct-contact operationalization (translation-channel rule), the truncation handling, and all thresholds above are fixed here, before computation.

Per host work, Aristotle contacts were classified DIRECT under the registered translation-channel rule: coded 'direct', or coded 'mediated' with a via-work that is itself a Latin translation of the cited Aristotelian text. Direct share d_w was computed per window over Aristotle contacts with non-missing mediation coding; RISE = d_STIa - d_Scriptum; Augustine served as the registered flat control. The registered KILLED trigger was RISE <= 0.

Dataset: In-house judged citation-contact corpus of 2,245 contacts across three dated Aquinas host works: Scriptum super Sententiis II (midpoint 1254), De veritate QQ. XIII-XXIX (midpoint 1258), and Summa Theologiae Ia (midpoint 1267).

computed 2026-07-11

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.