A Two-Stream Cipher Model for the Voynich Manuscript: Statistical Evidence for a Northwest-Semitic Consonantal Substrate.

We present a formally specified two-stream (SEM/STR) tokenisation of the Voynich Manuscript (Beinecke MS 408) and the quantitative evaluation of a single operational hypothesis: that the SEM stream carries a consonantal skeleton compatible with a North-West Semitic substrate. The operational claim is limited to a W3 (3-consonant sliding window) match rate on the Voynich SEM stream, projected via a bijective glyph operator (OPT9) onto the 9-letter output alphabet Σ₉ = {A,b,h,m,n,p,r,s,y}. Against a clean intersection of BDB (Biblical Hebrew) and Jastrow (rabbinic Aramaic) root-forms restricted to Σ₉, the Voynich corpus yields an instance-level W3 hit rate of 41.13% on N = 58,869 evaluable SEM instances (ACBR-v2). Three independent controls are reported: a random OPT9³ null (BDB enrichment 4.73×, z = 5.03, p < 0.001), a Latin-skeleton control (enrichment 3.52× over Latin), and an intra-Semitic Qumran Aramaic control (full-lexicon W3 rate 39.83%, size-matched enrichment 1.39×, reported as partially addressed). A pre-registered Blegen test on folio f103r returned FULL_FAIL for section-specific vocabulary prediction, reported without post-hoc revision. DictaBERT calibration (n = 100, dicta-il/dictabert) did not separate Voynich OPT9 forms from random (Welch t ≈ 1.16, p ≈ 0.25) and remains descriptive-only. Gap analyses include a scribe-hand STR control (section/hand ratio 4:1), a permutation-inversion accuracy test (E1: 86.9%), and extended cross-lexicon controls. All phase outputs, LaTeX source, and build parameters (SOURCE_DATE_EPOCH=1745500000, FORCE_SOURCE_DATE=1, three-pass pdflatex) are deposited for bit-identical reproduction. Concept DOI: 10.5281/zenodo.19686825. This is version 1.4 (2026-04-24); version 1.3 was deposited 2026-04-21.