A Locked Decoder for Beinecke MS 408: Statistical Evidence for Hebrew Encoding Under a Domain-Transparent Substitution Cipher

This work proposes a locked, deterministic substitution-cipher decoderfor Beinecke MS 408 and reports its behaviour on the target corpus andon four control corpora. The decoder consists of nineteen substitutionrules, locked on 28 February 2026 against an SHA-256 manifest, appliedby an approximately 180-line Python script with no human interventionor contextual disambiguation. The principal empirical finding is a four-corpus differential. Appliedto the Takahashi EVA transcription of MS 408, the decoder producesoutput whose consonantal skeletons match documented medieval Hebrewtriconsonantal roots in 59% of decoded tokens. Applied to medieval Latin(Vulgate Genesis 1-25, forced into EVA-shape characters) the rate fallsto 4%; applied to medieval Hebrew prose (Mishneh Torah, Sefer ha-Madda,similarly transliterated) it falls to 8%; applied to pseudo-Voynichesegenerated by a third-order Markov chain trained on the manuscript's ownn-gram statistics, it reaches 17%. The factor-of-3.5 separation betweenthe target and the strongest control is the principal quantitative claimof the work. The work does not claim a completed decipherment. It claims a structured,falsifiable hypothesis with a pre-registered held-out test on thebalneological folios (f75-f84) that constitutes the principal next steptoward confirmation or rejection. Reproducibility package available as related deposit. Released underCC BY 4.0.