We report results from a pre-registered hypothesis-testing investigation of the Voynich Manuscript (Beinecke MS 408) conducted with the Brain-V cognitive architecture, in which every statistical claim was locked to a falsifiable threshold before its test was executed. Across more than fifty hypotheses, we identify a set of robust structural findings: a Hand-A plant-folio vowel-pattern marker (_. oii) at 5. 04x corpus-wide enrichment surviving 5-fold cross-validation; three Hand-B plant markers (_. e. e, _. eeo, _. o. eo) with zero stem overlap with the Hand-A marker; a (scribe, content-type) classifier reaching 74. 2% five-fold accuracy across six classes; and a confirmed architectural difference between the two hands (Hand A's _. oii is a 3-stem closed class, Hand B's markers are productive). We also report substantive negative results of equal value: per-plant content decoding is refuted under a 0. 586-correlation null on token-count vs entry-length, twelve language-substitution hypotheses fail under shuffle-test or chi-square thresholds, three species-level identification tests fail, and four families of volvelle architectures (1, 400+ runs) cannot mechanically reproduce the _. oii enrichment. The contribution is not a decipherment but a structural map of what the manuscript cannot be (a per-plant illustrated dictionary, a simpleletter-substitution of any tested European or Semitic language) together with what it does appear to be at the surface level (natural-language-like prose with scribe-specific topical markers, decoded without illustration support).
Benjamin Horne (Thu,) studied this question.