Belief State

Evidence supports: The text-only section (entropy 3.9016 bits, 7 folios) represents the closest approximation to natura

Evidence supports: The recipes section (25 folios, 11,611 words, entropy 3.8586 bits — second highest and largest wor

Evidence supports: The Currier A/B scribal split represents two different scribes encoding the same underlying language

Evidence supports: The dominance of word-initial glyphs o, c, q, s, d and word-final glyphs y, n, l, r, o reflects a Vi

Evidence supports: The dominant word-initial glyph constraint (o, c, q, s, d account for most word starts) reflects a c

Evidence supports: Word-initial glyph constraints (o, c, q, s, d) and word-final glyph constraints (y, n, l, r, o) are

Evidence supports: The strong positional constraints on word-initial glyphs {o,c,q,s,d} and word-final glyphs {y,n,l,r,

Evidence supports: The Currier A/B split corresponds to two different scribal hands applying the same underlying cipher

Evidence supports: The word-initial glyph constraints {o,c,q,s,d} and word-final glyph constraints {y,n,l,r,o} are arti

Evidence supports: Glyph positional constraints (specific glyphs appearing predominantly word-initially vs word-finally

Evidence supports: The strong word-initial and word-final glyph positional constraints ({o,c,q,s,d} and {y,n,l,r,o}) re

Evidence supports: Currier A and Currier B use two structurally distinct cipher tables (not merely two scribal hands) e

Evidence supports: Currier A and Currier B employ two structurally distinct cipher alphabets encoding the same underlyi

Evidence supports: The Currier A and Currier B sub-corpora use two structurally distinct cipher tables encoding the sam

Evidence supports: The Currier A/B dialect split reflects two scribal hands encoding the same underlying language using

Evidence supports: The Currier A/B split represents two different scribes encoding the same underlying Latin text using

Evidence supports: The Currier A and B sub-corpora encode the same underlying plaintext language using two different bu

The herbal section is the best starting point for decipherment

Illustrations provide potential plaintext anchors for plant names

Token coverage alone is insufficient for Voynich decipherment: the null floor from 1,300 bigram-Markov nonsense skeletons is 83.56% (+/- 1pp), above Schechter and within 3pp of Brady

20-trial null lexicon test; Brady +3.3pp above null, Schechter -0.75pp below null

EVA 'q' is a categorical word-initial marker at 98.9% word-initial frequency (n=5,416), the strongest positional constraint of any EVA glyph

four-parallel-tests.py glyph positional analysis

Positional glyph constraints reflect cipher structure not grammar

Slot grammar could be either cipher artifact or linguistic feature

Plain EVA gallows 't' and 'p' function as line-initial paragraph markers (H-BRADY-02): p at 5.0x line-initial enrichment, t at 2.8x, bench gallows cth/ckh/cph at <0.5x

Brain-V independent verification of Brady 2026 section 3.5

Per-token section prediction from EVA vowel patterns alone does not beat always-predict-majority baseline on held-out folios (pharma F1 0.341 vs 0.614; bio F1 0.658 vs 0.716). The in-sample 100% agreement on patterns like _.eo was overfitting; the rule precision 0.827 on held-out data is real but only fires on 40% of tokens.

vowel_holdout_v1.py: 1037 held-out tokens across f101r, f89r2, f78r, f82r. NB multi-class 43.6% vs 55.7% majority baseline.

Evidence supports: The text-only section (7 folios, entropy 3.9016 bits — highest of all sections) represents unencip

Evidence supports: The Currier A/B split reflects a genuine linguistic difference between two distinct languages or dia

Evidence supports: The glyph positional constraints (word-initial {o,c,q,s,d}, word-final {y,n,l,r,o}) are not cipher a

Word-order syntactic structure is not recoverable by lexicon substitution: all 4 tested lexicons (Schechter, Brady, Hebrew, Brain-V v1) fail the shuffle test on connector-to-content bigrams (deltas -0.003 to -0.032 in-order vs shuffled).

4-lexicon shuffle test aggregate; connector-content bigram metric negative in-order vs shuffled across all

Different sections may use different encoding methods

Currier A/B language split shows measurable statistical differences

Evidence supports: The text uses a combination of substitution and transposition ciphers in the biological section.

Evidence supports: The Currier A/B split encodes two different plaintext languages (e.g., Latin in sections assigned to

Psychoactive plant folios (narcotic, hallucinogenic, soporific, psychedelic) show 8.14x elevated _.oii vowel-pattern rate vs non-psychoactive plants: 2.80% vs 0.34%. Pre-registered one-tailed Welch p=0.046, Cohen d=2.63, Mann-Whitney U p=0.000015. n=4 psychoactive folios (Paris quadrifolia, Cannabis, Rhododendron, Nymphaea caerulea) — small but all four individually elevated.

H-BV-PSYCHOACTIVE-01 pre-registered test confirmed

Evidence supports: The zodiac section's anomalously low entropy (3.7149 bits, lowest of all sections) reflects a label-

Evidence supports: The text-only section's elevated entropy (3.9016 bits, highest of all sections) reflects unenciphere

Evidence supports: The zodiac section (entropy 3.7149 bits, lowest of all sections) uses a label-oriented encoding in w

Evidence supports: The zodiac and astronomical sections use a systematically different word-order encoding than herbal

Evidence supports: The zodiac and label-heavy sections (zodiac entropy 3.7149 bits, astronomical 3.7471 bits) use a hom

Evidence supports: The text is encoded using a simple substitution cipher on an unknown language.

The simple 3-ring volvelle (6x26x8 with per-section cartridges) does NOT reproduce the _.oii plant-folio enrichment. 100 of 100 null runs produced 0.00x enrichment. Caveat: volvelle root generator cannot produce vowel clusters by construction, so richer volvelles remain open.

H-BV-VOLVELLE-OII-01; 100-corpora null test

Evidence supports: The 70.1% hapax ratio is substantially artifactual: stripping word-final glyphs drawn from the set {

Evidence supports: The manuscript's text structure reflects a mix of prose and verse, with the herbal section being pri

Evidence supports: The high hapax ratio (70.1%) is partially artifactual, caused by consistent scribal abbreviation or

Evidence supports: The Voynich text encodes a natural language using a null-cipher or homophones, where multiple glyphs

Evidence supports: The manuscript's herbal section uses a combination of substitution and transposition ciphers, which

Evidence supports: The extremely high hapax ratio (70.1%) is produced by a systematic suffix-stripping or abbreviation

Evidence supports: The Zipf exponent of 0.8946 (below the natural-language baseline of ~1.0) is caused by systematic wo

The underlying language is most likely Latin or Italian

15th century Italian provenance, Naibbe cipher compatibility, glyph entropy ~3.86 close to both

EVA vowel choice within a fixed consonant skeleton is non-randomly distributed by section at aggregate level (55/70 groups significant at p<0.01) but does NOT predict section per-token on held-out folios (best F1 below blind baseline).

55/70 skeleton groups significant at p<0.01; kdy case chi2=262 vs crit 50.9

The manuscript contains meaningful content, not random glyphs

Statistical structure too regular for gibberish — Zipf fit, entropy in natural language range

Evidence supports: The Voynich text uses a homophonic substitution cipher on medieval Latin or Italian, where multiple

Evidence supports: The manuscript uses a combination of substitution and transposition ciphers in the biological sectio

The cipher is performable by hand with period tools

No evidence of mechanical aids, Naibbe cipher demonstrates feasibility