Identifying the Language of the Voynich Manuscript: Rabbinic Hebrew, its Grammar and Syntax, under Pre-Registered Controls

Beinecke MS 408, the Voynich manuscript, has resisted decipherment for over a century, defeating a long line of confident solutions. Working from the v101 transcription under a pre-registered protocol, with predictions fixed before computation and tested against scrambled-key and other null models, we identify its language as Hebrew of a rabbinic, post-biblical register, written left to right in an invented, sound-based abjad and read on the consonantal skeleton. The grammar is Hebrew above the word: a one-letter proclitic article and conjunction in their Hebrew places, the adjective after its noun against a within-line shuffle, and a glyph-defined verb class that refuses the definite article and concentrates in the instructional herbal section, behaving as the rabbinic periphrastic present and reproducing on independent halves of the corpus. A frequency-matched medieval Latin corpus pushed through the same key reproduces none of these signatures. Word-level recognition under a blind Hebrew pipeline is consistent with the identification. The manuscript carries a written vowel layer that does not carry a full per-word Hebrew vocalisation but and its reconstructed values are consistent with an Ashkenazi reading tradition of the kind current in the manuscript's likely region of origin, and which resolves a consonant-ambiguous word to a single reading in a majority of cases on a held-out test. A standard objection also dissolves: on the glyph-level v101 transcription the manuscript repeats less than natural Hebrew, its reputed extreme repetitiveness an artefact of the EVA transcription. What we report is a language identification and a controlled decipherment of the grammar, not a translation of the text; a continuous translation is the work of a second paper.