The Pharmacist's Cipher: Five Statistical Tests Supporting a Pharmaceutical Encoding of the Voynich Manuscript (MS 408)

We present five reproducible statistical tests on the text of the Voynich Manuscript (MS 408, Yale Beinecke Library), based on the IVTFF interlinear transcription corpus. All tests are computed from the transcription alone: (1) morphological prefix frequencies are stable across two independent transcribers (all 11 tested prefixes within 4% divergence), independently corroborated by the convergence of five non-supervised segmentation algorithms on the same morphological atoms; (2) a categorical morphological gap exists between the s- and sh- prefix families (sedy attested exactly 0 times against shedy at 427); (3) a 27:0 asymmetric lexical reuse pattern between the Pharma and Stars sections is compatible with a specific functional dependency; (4) paragraphs in the four major sections conform (78-88%) to a single procedural grammatical template, derived distributionally rather than from prior categorial assumptions; (5) six manuscript sections exhibit systematically distinct morphological profiles consistent with successive stages of a pharmaceutical workflow. A new discussion section summarizes the information architecture of the text calibrated against a genre- and size-matched real-language control (Nuovo Receptario Fiorentino, 1498): the sequential information carried by word order is roughly one tenth of the natural-language benchmark, while the information lives inside the token, including a measured gallows channel of about 1 bit per token that context cannot repair. Version 2.0 (July 27, 2026), corrected and updated edition. All counts recomputed on a corrected canonical extraction of the corpus (37,967 tokens; supersedes the 37,036-token extraction of v1.1-v1.3); every result replicates. Incorporates the previously unpublished v1.3 revision (May 2026): the former Result 6 (volume hierarchy via ee%) is withdrawn as evidence and preserved in an appendix with two additional refuting checks (the test does not survive removal of the da- family, and a token-length confound was identified); the former Result 4 (hand-assigned positional categories) is replaced by the distributionally derived paragraph grammar. New in this edition: an explicit frame non-uniqueness limitation (pharmacy is the most coherent domain reading, but tincture, alchemy and generic process notation survive the same distributional evidence); the ductus limitation of earlier versions is resolved by a paleographic catalogue of twelve gallows variants (6,143 traced glyphs, public dataset), whose variant grade behaves as an occurrence-level third information channel invisible to EVA-based analyses; companion references updated to the July 2026 status report and partial retraction (DOI: 10.5281/zenodo.21629182). Version history: v1.2 (March 26, 2026), seven targeted revisions. v1.1 (March 23, 2026), initial public release.