Answer · Voynich Manuscript

Latin or Italian under a chancery mixed key

The text is Latin (or an Italian vernacular) enciphered with an Italian chancery mixed key of the Tranchedino type: several glyph-string signs per letter (more for vowels), signs for common syllables, a nomenclator of whole-word signs, and nulls; written either one plaintext word per written word, or with plaintext words split across written words, or as a continuous stream re-spaced.

Ruled outAnswers 1 questionAbductiveTested0

Where it stands

Ruled outNone pieces of evidence for, three against.

What it predicts

P1 fingerprint: a simulator of this family, fitted on the training quires, reproduces both the weak word order (order gain within 0.25 of 0.42; held-out bigram gain <= 0) and the topical page locality (z within 2 of 16) that no earlier procedure reproduced together, and the form-function coupling (rho >= 0.10)
P2 collapse: merging spelling variants that behave alike in context (edit-1 pairs with high context similarity) raises the manuscript's word-order gain and repeated 4-grams by more than merging the same number of random edit-1 pairs, and by more than the same procedure does to Latin; on the simulator the context-guided collapse recovers part of the oracle (true-plaintext) order gain
P3 choice dynamics: successive occurrences of the same variant family within a page show a same-variant rate different from the page-shuffled expectation (below it = a human cycling homophones; above it = sticky choice or copying; equal = random choice), and the rate is the same for each scribe
P4 solver: a mixed-unit solver with nulls that recovers planted keys of this family on Latin at >= 50 % held-out word accuracy (or >= 70 % of the plaintext's dictionary coverage) decodes the manuscript's held-out sheets to dictionary coverage above its own permutation null by more than the control's margin; if the solver fails its controls at some rung of the ladder, the hypothesis is reported as untested beyond that rung

Evidence none for · 3 against

Endings are more concentrated than Italian's

Established · weighs some · F-033

AGAINST

No Tranchedino-type mixed key reproduces the text

Established · weighs heavily · F-031

AGAINST

Spelling variants depend on the word's place in the line

Established · weighs some · F-032

AGAINST

A mixed key whose signs are glyph strings cannot be recovered by the element solver even…

Established · weighs little · F-034

NEUTRAL

Tested by 27 runs

Run	Measured	Result	Against	Decision	Date
RUN-081	held-out element score, real minus spelling-permuted null (nats/element), word spacing, learned units	Voynich Latin +0.27 / +0.18 (seeds 3, 4); Italian +0.44	Planted-key control (L5-type) +0.60 / +0.50 · content-free texts with Voynich word structure: slot grammar +0.00, self-copying +0.02 · stream: control -0.36, Voynich -0.22. Dictionary coverage real <= permuted for every text incl. the control (0.33 vs 0.36)	Untested beyond rung L4 (registered rule); exploratory gap = sequential structure inside words that the content-free generators lack, not Latin words	22 Oct
RUN-066	planted mixed key on Latin: held-out word accuracy / key accuracy by use / dictionary coverage (plaintext's)	0.112 / 0.398 / 0.344 (0.850)	Seed 11	Fails (simulator signs ambiguous; superseded)	22 Oct
RUN-064	planted mixed key on Latin: held-out word accuracy / key accuracy by use / dictionary coverage (plaintext's)	0.6 / 0.854 / 0.607 (0.850)	Seed 11	Passes (simulator signs ambiguous; superseded)	22 Oct
RUN-065	planted mixed key on Latin: held-out word accuracy / key accuracy by use / dictionary coverage (plaintext's)	0.125 / 0.357 / 0.278 (0.850)	Seed 11	Fails (simulator signs ambiguous; superseded)	22 Oct
RUN-067	planted mixed key on Latin: held-out word accuracy / key accuracy by use / dictionary coverage (plaintext's)	0.861 / 0.974 / 0.764 (0.850)	Seed 11	Passes	22 Oct
RUN-068	planted mixed key on Latin: held-out word accuracy / key accuracy by use / dictionary coverage (plaintext's)	0.892 / 0.977 / 0.792 (0.850)	Seed 11	Passes	22 Oct
RUN-070	planted mixed key on Latin: held-out word accuracy / key accuracy by use / dictionary coverage (plaintext's)	0.814 / 0.948 / 0.754 (0.850)	Seed 11	Passes	22 Oct
RUN-071	planted mixed key on Latin: held-out word accuracy / key accuracy by use / dictionary coverage (plaintext's)	0.527 / 0.859 / 0.723 (0.850)	Seed 11	Passes	22 Oct
RUN-074	planted mixed key on Latin: held-out word accuracy / key accuracy by use / dictionary coverage (plaintext's)	0.087 / 0.062 / 0.228 (0.850)	Seed 11	Fails (lexicon segmentation; superseded by unit discovery)	22 Oct
RUN-075	planted mixed key on Latin: held-out word accuracy / key accuracy by use / dictionary coverage (plaintext's)	/ 0.082 / 0.235 (0.849)	Seed 11	Fails (lexicon segmentation; superseded by unit discovery)	22 Oct
RUN-076	planted mixed key on Latin: held-out word accuracy / key accuracy by use / dictionary coverage (plaintext's)	0.087 / 0.245 / 0.332 (0.850)	Seed 11	Fails (unit discovery, cold rounds; superseded by warm-started rounds)	22 Oct
RUN-077	planted mixed key on Latin: held-out word accuracy / key accuracy by use / dictionary coverage (plaintext's)	/ 0.0 / 0.338 (0.849)	Seed 11	Fails (unit discovery, cold rounds; superseded by warm-started rounds)	22 Oct
RUN-078	planted mixed key on Latin: held-out word accuracy / key accuracy by use / dictionary coverage (plaintext's)	0.087 / 0.254 / 0.356 (0.850)	Seed 11	Fails	22 Oct
RUN-079	planted mixed key on Latin: held-out word accuracy / key accuracy by use / dictionary coverage (plaintext's)	/ 0.203 / 0.315 (0.849)	Seed 11	Fails	22 Oct
RUN-057	planted mixed key on Latin: held-out word accuracy / key accuracy by use / dictionary coverage	0.012 / 0.0 / 0.476	Plaintext coverage 0.771	Fails (character-level objective; superseded by the element solver)	22 Oct
RUN-058	planted mixed key on Latin: held-out word accuracy / key accuracy by use / dictionary coverage	0.017 / 0.162 / 0.367	Plaintext coverage 0.771	Fails (character-level objective; superseded by the element solver)	22 Oct
RUN-059	planted mixed key on Latin: held-out word accuracy / key accuracy by use / dictionary coverage (plaintext's)	0.68 / 0.889 / 0.659 (0.850)	Seed 11	Passes (simulator signs ambiguous across positions; superseded (rows 68-72))	22 Oct
RUN-060	planted mixed key on Latin: held-out word accuracy / key accuracy by use / dictionary coverage (plaintext's)	0.6 / 0.854 / 0.607 (0.850)	Seed 11	Passes (simulator signs ambiguous across positions; superseded)	22 Oct
RUN-063	conditional MI of variant and line position within frames vs within-page shuffles; paragraph-first/last lines excluded	median z 4.5 over 10 classes (6 above 3)	Latin median z 0.0 over 10 classes (0 above 3) · simulators word median z -0.8 over 3 classes (0 above 3), split median z 0.2 over 10 classes (0 above 3), stream median z 0.6 over 10 classes (1 above 3) · scribe rows with z>3: 7/32	See note	22 Oct
RUN-069	conditional MI of variant and line position within frames, plaintext lines = verses	key: median z 0.3, 0/10 classes above 3	Latin letters: median z 1.4, 0/10 above 3	See note	22 Oct
RUN-080	same-variant rate of successive family occurrences within page, observed/expected; families from the top-20 context alternations	1.041	Scribes 1.02-1.08 · Latin 1.009 · simulated keys with random/cycling/sticky choice: 1.01/0.995/1.015 (true families), 1.06/1.07/1.06 (inferred)	Instrument fails calibration: cannot separate choice modes	22 Oct
RUN-056	same-variant rate of successive family occurrences within page, observed/expected	1.093	See out/choice_dynamics.csv for scribes, Latin and simulator controls	See note	22 Oct
RUN-062	variant choice vs line position and previous glyph, top alternation classes: median z against within-page shuffles	line 33.3; previous glyph 17.6	Simulators and Latin in out/position_blind.csv	See note	22 Oct
RUN-073	held-out statistics failed by the fitted mixed key, per spacing practice (\|z\|>2 of 12)	word: 12; split: 10; stream: 9	20 simulations each · section-matched Latin plaintext	See note	22 Oct
RUN-055	change in word-order gain after merging 300 context-similar edit-1 pairs (random-pair merge; Latin)	-0.224	Random -0.058 · Latin context -0.483	See note	22 Oct
RUN-061	mean excess of context-chosen over random merges at equal type count: repeated 4-grams / held-out bigram gain	Voynich 23.053 / 0.022	Latin 2.392 / 0.012 · mixed key word None / None · split 0.6 / 0.014	See note	22 Oct
RUN-072	Simonetta 1474 diagnostics: word-final units above 1% / share of five commonest endings; q followed by o	Voynich 8 / 0.91; q->o 0.976	Italian 11 / 0.83 · Latin 12 / 0.71	Endings more concentrated than plain Italian, the opposite of vowel homophones	22 Oct

The questions it answers

Is it a cipher of a known language?

Part of What kind of text is it?

Most likelyNo answer ahead

Cite it

Mysteries Department. (2026). Latin or Italian under a chancery mixed key [Answer ANS0D8FZZM071JE]. https://mysteriesdept.com/r/ANS0D8FZZM071JE

@misc{mysteriesdept_ans0d8fzzm071je,
  title     = {Latin or Italian under a chancery mixed key},
  author    = {{Mysteries Department}},
  year      = {2026},
  note      = {Answer, catalog key ANS0D8FZZM071JE},
  url       = {https://mysteriesdept.com/r/ANS0D8FZZM071JE}
}