← All conjectures · Texts, scribes & transmission

The Liturgical Exception

Cross-period Dated c. 800 CE – 1600 CE Europe

Status: Falsified

The verdict’s fine print — quoted from the resolution record: “KILLED decisively, and the kill is the finding: liturgical transmission does NOT refuse the copying funnel - the Cantus top decile carries 58.23%, past the 50% kill line AND sitting INSIDE the literary band (55.66-66.32%) measured in the same blind run.” 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)

Chant books are calendar-driven, not demand-driven: a 9th–16th-century Latin antiphoner or gradual must cover the whole liturgical year, so each chant of the cursus is copied once per book almost regardless of its fame. Where literary and scriptural copying compounds popularity into a heavy head of super-copied works, liturgical transmission should refuse the funnel: sources-per-chant should track the flat count of surviving service books, and the transmission records of liturgy and literature should prove to be two systems, not one.

Prediction clause (verbatim)

Prediction: in the DACT/Cantus chant data, the top decile of chant IDs ranked by source concordances carries at most 35% of all concordance records — at least 15 percentage points below the 50%-plus band that witness funnels show in literary corpora — and the median chant's source count is at least one-fifth of the 90th-percentile chant's; coverage guard: restrict to antiphons and responsories in sources indexed as complete, with at least 100 such sources and 2,000 distinct chant IDs.

Kill-dataset (verbatim)

Kill: count source concordances per chant ID in the DACT Cantus data over antiphons and responsories in fully indexed sources and compute the top-decile share; a share at or above 50% kills the divergence, a share at or below 35% with the guard met supports it.

On Inferpedia

This conjecture is linked to the following pages on Inferpedia, an encyclopedia of the missing — working atlas pages, some still early scaffolding.

Gestalt

Provenance

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

Fresh blind generation by claude-fable-5, 2026-07-20, for the Old-World-invariance (same-laws-everywhere) wave, posed entirely from weights with no repository or web consultation. I favored paired law-and-named-break shapes — the copying funnel against calendar-driven liturgy, oral king-list rounding against Egyptian annalistic precision, open-library age-curves against the sealed Dunhuang deposit — plus catalogue grammars of loss and doubt, cross-canon vocabulary growth, commentary expansion ratios, translator concentration, pharmacopoeia accretion tempo, primate archives, hapax poets, and the day-wage price of a book; ordinals 2, 4, 8, 13, and 16 are the divergence wagers.

Novelty / leakage triage

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

That liturgical books are calendar-complete rather than demand-driven is standard chant scholarship (Hiley 1993), and the Cantus data have supported quantitative corpus studies (Cornelissen, Zuidema and Burgoyne 2020), but I locate no published concentration census of source concordances per chant ID and no print statement of the pinned literary-versus-liturgical top-decile contrast. The outcome is genuinely open: core-repertory antiphons and responsories appear in nearly every source while locally particular chants are near-hapax, so the top-decile share could honestly land on either side of 35%. The internal arithmetic is consistent (the 35% ceiling sits exactly the promised 15 points under the 50% literary floor). Verdict: adjacent - established mechanism, unrun pinned census.

Sources cited by the triage

David Hiley, Western Plainchant: A Handbook (Oxford University Press, 1993) — Calendar-driven structure of the office and mass books.
Bas Cornelissen, Willem Zuidema and John Ashley Burgoyne, 'Studying Large Plainchant Corpora Using cantus', Proceedings of the 7th International Conference on Digital Libraries for Musicology (2020) — Quantitative corpus work on the Cantus data; no concentration census of concordances per chant as pinned.

Predictions

Killed registered 2026-07-20 calibration prediction (parent triage: leaked/adjacent)

Resolution: Killed

Caveats: KILLED decisively, and the kill is the finding: liturgical transmission does NOT refuse the copying funnel - the Cantus top decile carries 58.23%, past the 50% kill line AND sitting INSIDE the literary band (55.66-66.32%) measured in the same blind run. The registered secondary fails independently in the same direction (median 2 vs required >= 13.4), and the ex-ante-completed dead zone (35-50%) never engages. Read jointly with cantus-007 (singleton share 0.418, SUPPORTED 2026-07-18): the chant corpus has BOTH a heavy singleton tail AND a funneled head - the calendar constrains what must be copied once per book, but preferential attachment still governs which chants ride many books; the blind author's mechanism (calendar-driven flatness) was real but operates in the wrong region of the distribution. Paired-design payoff recorded plainly: the law (B-001) is SUPPORTED and its named exception is KILLED - the funnel is MORE universal than wagered, which strengthens the invariance thesis beyond either verdict alone. MAIN CAVEAT, disclosed not buried: the completeness restriction required a documented hybrid instrument (published dump + live CD-database completeness field for 231 sources), because the pinned dump alone cannot execute the registered restriction - a kind-preserving completion of the pin with a large documented scope-narrowing (~48% of corpus); the 8.23pp kill margin, the 6.7x secondary failure, and the absence of a clear direction for any narrowing bias together make the verdict robust to it; a registrant wanting the full-federation number needs a future instrument build, not a re-run. Vintage pinned (v1.0, retrieval date, live-scrape date recorded). Blind compute claude-sonnet-5; verdict claude-fable-5.

OPEN prediction (divergence wager, paired with B-001): calendar-driven chant transmission REFUSES the copying funnel - in the DACT/Cantus data the top decile of chant IDs carries at most 35% of concordance records, at least 15 points below the literary funnel band, with a fat middle (median at least one-fifth of P90).

Resolution criteria — the registered fine print

Resolution criteria: VERBATIM PREDICTION: 'in the DACT/Cantus chant data, the top decile of chant IDs ranked by source concordances carries at most 35% of all concordance records - at least 15 percentage points below the 50%-plus band that witness funnels show in literary corpora - and the median chant's source count is at least one-fifth of the 90th-percentile chant's; coverage guard: restrict to antiphons and responsories in sources indexed as complete, with at least 100 such sources and 2,000 distinct chant IDs.' KILL VERBATIM: 'a share at or above 50% kills the divergence, a share at or below 35% with the guard met supports it.' FULL OUTCOME PARTITION (completing the verbatim clauses' gap ex ante): guards unmet -> INCONCLUSIVE-BY-COVERAGE; with guards met, PRIMARY = the top-decile concordance share: share >= 50% -> KILLED; share <= 35% AND the SECONDARY clause (median source count >= P90/5) holds -> SUPPORTED; share <= 35% with the secondary failing -> INCONCLUSIVE-BY-CLAUSE-SPLIT (the author's two conjoined clauses disagree; recorded, not adjudicated away); share in (35%, 50%) -> INCONCLUSIVE-BY-DEAD-ZONE, an explicitly registered undecided region. OPERATIONALIZATION: instrument = the newest PUBLISHED DACT/Cantus corpus dump (the cantus-007 instrument route; dump version and retrieval date recorded - vintage pin per lesson 7); restrict to genre codes for antiphons and responsories and to sources the dump itself marks complete (the compute records the exact field names and accepted values from the dump's own schema/documentation BEFORE tallying); concordance records = chant-ID x source rows after the restriction; top-decile rule identical to B-001's (rank by source count, stable-id ties, first ceil(N/10)); median and P90 computed on the same restricted distribution. FIREWALL NOTE: measured in the same four-corpus blind run as B-001 - the compute is not told which corpus any hypothesis distinguishes.

Known-priors disclosure — what the registrant already knew

Known priors disclosure: Triaged adjacent 2026-07-20. Disclosed neighbor, no numbers carried: cantus-007 (resolved SUPPORTED-qualitative 2026-07-18) measured the SINGLETON share of chant IDs (0.418) on the DACT CantusCorpus dump - a different statistic; a heavy singleton tail is arithmetically compatible with either a fat or a thin head, so the top-decile share is genuinely open; the registrant has computed nothing on it. The liturgical-uniformity mechanism (calendar-driven copying) is established musicology (the claim's premise); the wagered band is not. Paired divergence design with B-001 disclosed; the dead zone and clause-split regions are completed ex ante per the twice-earned partition lesson, without altering the author's thresholds.

Method and dataset — how it was measured

Register-before-compute held (registration 18:04:34Z, packet 9290244); measured in the SAME blind four-corpus run as the sibling - the compute could not know which corpus any hypothesis expected to diverge. Guards met: 176 complete sources >= 100; 15,428 chant IDs >= 2,000. PRIMARY: top-decile concordance share = 58.23%. REGISTERED PARTITION: share >= 50% -> KILLED. SECONDARY (independent, same direction): median source count 2.0 vs P90 67.0 - the median >= P90/5 clause fails by a factor of ~6.7.

Dataset: CantusCorpus v1.0 (LINDAT/CLARIN 11234/1-6041, retrieved 2026-07-20; 888,010 chants / 2,278 sources, matching the published corpus paper exactly), restricted per the registration to antiphons (genre A) + responsories (genre R) in completely-indexed sources. KEY INSTRUMENT FACT, documented not silent: the published dump's sources.csv carries NO completeness field - it exists only on the live Cantus Database; the compute traced the field through the DDMAL/CantusDB source code and live-scraped it for all 231 sources of the CD sub-database (the only federation component exposing it), yielding 176 complete sources; the analysis is scoped to that documented restriction (~48.4% of the corpus by component). Restricted population: 176 complete sources, 15,428 distinct chant IDs, 248,864 concordance records.

computed 2026-07-20

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.