The calendar with too many months

Deep prehistory Dated c. 43000 BCE – 10000 BCE Europe

Status: Falsified

The verdict’s fine print — quoted from the resolution record: “The pinned PRIMARY clause decides and it fails decisively: R=0.0154 (primary, n=651) and 0.0148 (conservative, n=405) are far BELOW the registered 0.10 threshold, so the conjecture -- which wagered that MORE than 10% of sequences exceed the 13-mark lunar ceili…” 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)

In 2023 Bennett Bacon and colleagues (Cambridge Archaeological Journal 33/3) argued that the dots, lines and a 'Y' mark placed beside Upper Palaeolithic animal images are a phenological notation: the count of marks in a sequence records the month, reckoned from the onset of spring, in which that species mates, gives birth, or aggregates, and the 'Y' (a line inside a line) means 'giving birth'. The reading is anchored to modern analogues of each animal's breeding calendar, and it was published with a data table of sequences, species and mark-counts. It is exactly the kind of proto-writing claim this lane treats as decipherment-instability data rather than settled fact. The model carries a hard internal ceiling that its authors did not foreground as a test: a lunar-month calendar reckoned within a single year cannot index more than about thirteen months. So the distribution of sequence lengths is decisive on the model's own terms. A true within-year month-count should pile up in the short breeding-season range and never exceed the ceiling; a comfortable tail of long sequences would mean the marks are counting something else, or nothing countable, and the calendar reading fails without any appeal to what the makers believed.

Prediction clause (verbatim)

Prediction: re-tallying the Bacon et al. 2023 supplementary dataset of dot/line sequences, the fraction of sequences whose mark-count exceeds 13 (the within-year lunar-month ceiling the model imposes) will exceed 10 percent, and a label-permutation test - shuffling the species-to-birth-month assignment 10,000 times - will place the reported mark-count-versus-phenological-month correlation inside the middle 90 percent of the shuffled null (primary clause: the >10 percent over-ceiling fraction; the failed permutation test is the corroborating clause). Coverage guard: if fewer than 40 distinct (species, sequence) pairs survive the authors' own inclusion criteria the test voids for insufficient degrees of freedom rather than resolving.

Kill-dataset (verbatim)

Kill: the Bacon, Khatiri, Palmer, Freeth, Pettitt & Kentridge (2023, Cambridge Archaeological Journal) article and its supplementary data table - count the fraction of scored sequences with more than 13 marks, and re-run their mark-count/birth-month association under a 10,000-fold permutation of the species-month labels. Both operations run directly on the published SI.

Provenance

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

Fresh blind generation by claude-fable-5, 2026-07-17, deep-prehistory wave against real published prehistoric corpora (von Petzinger's Upper-Palaeolithic geometric-sign database; the Bacon et al. 2023 CAJ phenological-notation dataset and its rebuttals; the Marshack vs d'Errico notation re-readings; Snow's hand-stencil digit-ratio tables; the Gargas/Cosquer incomplete-hand catalogues; Blombos/Diepkloof engraved-object inventories; the Reznikoff/Fazenda cave-acoustic surveys; the Geissenkloesterle/Hohle Fels/Isturitz flute corpus and the Divje Babe controversy; Berezkin's Analytical Catalogue and d'Huy's motif phylogenies; Nunn & Reid's 21 drowned-coastline traditions; Mazama/Budj Bim geochronology; Thom's megalithic-yard corpus and Kendall/Freeman reanalyses; Winn's Vinca sign inventory; the Ishango/Lebombo/Border Cave notched bones; the Aubert/Brumm U-series rock-art dataset). Discipline: every kill names a real corpus/database/dataset and a countable, decidable operation (discovery-curve saturation, permutation/ceiling tests, mark-count instability census, misclassification-corrected CIs, configuration-combinatorics, engraved-fraction, geometry-controlled acoustic permutation, survivorship distribution, distribution-geometric motif tests, bathymetric/tephrochronological consistency, Bayesian quantum reanalysis, repertoire-size thresholds, sampling-vs-origin correlation), thresholds far from 1, disambiguation pinned in the prediction, and explicit coverage guards separating taphonomy from ancient absence. Famous controversies (lunar notation, proto-writing, flute-or-bone) enter as decipherment-instability data, not as settled stories. Sound items stay strictly pre-notational (Palaeolithic aerophones and cave resonance) and disjoint from the concurrent ancient-notated-music wave (Hurrian/Greek/Mesopotamian tuning texts) and from the Asian music-scores wave (medieval/early-modern East/South/Southeast-Asian tablatures). Drops: Goebekli Tepe symbol-recurrence and Jiahu incised marks (held back to avoid a third proto-sign repertoire-threshold item; documented in the report as runnable alternates).

Novelty / leakage triage

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

Bacon et al. 2023 (Cambridge Archaeological Journal 33/3) published its sequence/species/mark-count supplementary table, and the calendar reading has drawn substantial published critique - most formally the García-Bustos et al. 2023 Discussion in the Journal of Paleolithic Archaeology. But those critiques target tracings, sign identifications and the undefined notion of 'association'; none tabulates the specific >13-mark over-ceiling fraction or runs a species-to-birth-month label-permutation null on the published SI. The discriminating arithmetic the conjecture specifies was not located as run.

Sources cited by the triage

B. Bacon, A. Khatiri, J. Palmer, T. Freeth, P. Pettitt & R. Kentridge, 'An Upper Palaeolithic Proto-writing System and Phenological Calendar,' Cambridge Archaeological Journal 33(3) (2023)
A.J. García-Bustos et al., 'Discussion: "An Upper Palaeolithic Proto-writing System and Phenological Calendar" by Bennett Bacon et al. (2023),' Journal of Paleolithic Archaeology 6 (2023), doi:10.1007/s41982-023-00158-8

Predictions

Killed registered 2026-07-18 calibration prediction (parent triage: leaked/adjacent)

Resolution: Killed

Caveats: The pinned PRIMARY clause decides and it fails decisively: R=0.0154 (primary, n=651) and 0.0148 (conservative, n=405) are far BELOW the registered 0.10 threshold, so the conjecture -- which wagered that MORE than 10% of sequences exceed the 13-mark lunar ceiling -- is REFUTED. (a) The corroborating label-permutation clause could NOT be run (the GLM predictors are absent from the structured SI columns and the compute rightly declined to hand-parse free-text prose); this is noted honestly and does NOT rescue the conjecture -- the primary already fails. (b) The reconstruction's N (651/405) does not exactly reproduce the authors' published 606 analysable sequences because their row->sequence split code was unpublished; but the >13 fraction is stable at ~1.5% across BOTH split rules, so the verdict is robust to that choice. (c) Substantively, the mark-counts respect the ~13 lunar-month ceiling far more than the conjecture wagered -- Bacon's ceiling holds up on THIS axis. This REFUTES a skeptical conjecture; it does NOT endorse the calendar model beyond this single test (the Garcia-Bustos et al. 2023 methodological critique is a separate matter, untouched by this arithmetic). Independent VERIFY remains open per the no-self-grading rule.

Registered against the Bacon et al. 2023 (Cambridge Archaeological Journal) supplementary dataset of Upper Palaeolithic dot/line/Y-sign sequences. PREDICTION VERBATIM: Prediction: re-tallying the Bacon et al. 2023 supplementary dataset of dot/line sequences, the fraction of sequences whose mark-count exceeds 13 (the within-year lunar-month ceiling the model imposes) will exceed 10 percent, and a label-permutation test - shuffling the species-to-birth-month assignment 10,000 times - will place the reported mark-count-versus-phenological-month correlation inside the middle 90 percent of the shuffled null (primary clause: the >10 percent over-ceiling fraction; the failed permutation test is the corroborating clause). Coverage guard: if fewer than 40 distinct (species, sequence) pairs survive the authors' own inclusion criteria the test voids for insufficient degrees of freedom rather than resolving.

Resolution criteria — the registered fine print

Resolution criteria: DENOMINATOR (verbatim): the (species, sequence) pairs in the Bacon et al. 2023 supplementary dataset that survive the authors' own inclusion criteria. NUMERATOR: sequences whose mark-count exceeds 13. DATA: download the CAJ article's supplementary dataset (freeze URL + sha256); apply the authors' stated inclusion criteria as documented in the SI. R = fraction of surviving sequences with mark-count > 13. Corroborating (non-verdict): a 10,000-iteration label-permutation test shuffling the species-to-birth-month assignment, reporting whether the observed mark-count-vs-phenological-month correlation falls inside the middle 90% of the shuffled null. CLAUSE PRECEDENCE: (1) INCONCLUSIVE_BY_DESIGN if fewer than 40 distinct (species, sequence) pairs survive the inclusion criteria (the conjecture's own void rule); (2) SUPPORTED if R > 0.10; (3) REFUTED if R <= 0.10. Compute firewalled from threshold and direction.

Known-priors disclosure — what the registrant already knew

Known priors disclosure: Triage (2026-07-17) graded adjacent and issued an ERRATUM to the conjecture's own framing: the real published methodological critique is Garcia-Bustos et al. 2023 (Journal of Paleolithic Archaeology), not a 'Meyering 2025' the generator hedged (which could not be located); Garcia-Bustos is a methodological critique, NOT the >13-mark/permutation arithmetic, so the specific test remains un-run. Direction genuinely open; the registrant leans toward the over-ceiling fraction being real given the small dataset.

Method and dataset — how it was measured

Register-before-compute: registration row in docs/generated/conjecture_prediction_batch2_20260718.json (session fable-depth-registrations-20260718-batch2, registered 2026-07-18 ~09:38Z, denominator quoted verbatim). Firewalled Sonnet compute (docs/generated/compute_bacon_calendar_20260718.json) at computed_at_utc 2026-07-18T10:30:03Z, which POSTDATES registered_at (rule 6). Inclusion criteria reconstructed from the authors' own spreadsheet ambiguity annotations (the SI narrative contains no prose exclusion section); over13 fraction (strictly >13) computed on BOTH the primary split rule (multi-register cells split per integer, N=651) and a conservative robustness rule (drop all 90 multi-value cells, N=405). The compute was firewalled from threshold and direction. Registered clause precedence [(1) VOID if <40 pairs; (2) SUPPORTED if R>0.10; (3) REFUTED if R<=0.10] applied at this shepherd stage.

Dataset: Bacon, Khatiri, Palmer, Freeth, Pettitt & Kentridge 2023, Cambridge Archaeological Journal 33/3, DOI 10.1017/S0959774322000415 -- the AUTHORS' OWN supplementary dataset (sup001.xlsx sha256 0f93a78e0ea58b2ebf72823056f90b7a4524fa3d73613770ffe5fdef651ec597, sup002.docx sha256 6f4c04b69e315e5bcd1a29ce0e000f45f8c1003abc0cff568d8324dbd5f65340, sup003.xlsx sha256 1e9ab5a7cdf0e71850be9e8d1a812c6460c53f5556f11966d36c1ed0d1d52df1; all three verified). 637 raw data rows -> 495 clean rows after the authors' own in-cell ambiguity flags. DENOMINATOR (verbatim): surviving (species, sequence) pairs -- primary split rule N=651, conservative drop-multi-value rule N=405. NUMERATOR: sequences with mark-count strictly greater than 13 (the within-year lunar-month ceiling the model imposes).

computed 2026-07-18

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.