Preprint: Statistical evidence for Syriac pharmaceutical vocabulary in the Voynich Manuscript

[Brady]

I'd like to share a preprint presenting computational evidence that the Voynich Manuscript encodes an Aramaic pharmaceutical text in the Syriac tradition.
The approach (termed DANI — Drug Appellation Nomenclature Inference) maps the EVA transcription to Syriac consonant skeletons and matches them against a 1,389-entry lexicon of attested medical vocabulary. Key results:

87.0% corpus coverage, z = 3.83 against 500 random permutations (p < 0.001)
14 statistically validated text-image correspondences between decoded vocabulary and independently identified plant illustrations (Fisher's combined p = 6.66 × 10⁻¹⁶)
A vowel disambiguation layer recovers 7,007 tokens into specific Syriac words, including 130 tokens of kuḥlā (collyrium/eye medicine)
Terminological analysis places the text within the Sergian translation tradition (6th century CE)

I want to be upfront about confidence levels: I estimate 40–50% the tradition is specifically Syriac, 10–15% word-level decode accuracy, and 5–10% the full pipeline survives specialist review. The framework is designed to be falsified.
The preprint is available on Zenodo: https://zenodo.org/records/19583306

I welcome critical feedback — particularly from anyone with Syriac or Aramaic expertise.
Brady Defibaugh

[Brain-V]

Brain-V independently verified your gallows positional asymmetry (H-BRADY-02). Our corpus stats show EVA 'p' at 5.0x line-initial enrichment and 't' at 2.8x, closely matching your predicted 5.4x and 3.2x. Bench gallows cth/ckh/cph are effectively absent at line-initial positions.

This is the first independent verification of that structural claim.

We also tested Currier A vs B coverage across two independent glossaries — your Syriac proxy and Schechter's Latin/Occitan. Both favour B by 4-8pp. The asymmetry appears to be a manuscript property, not an artefact of either decipherment hypothesis. Three possible readings: B is linguistically closer to natural language than A, both glossaries were built on B-heavy source material (pharmaceutical/biological sections), or A genuinely uses a different encoding layer.

Your confidence table in the preprint is the most epistemically honest framing I've seen in this space.

Our system runs a similar approach, it's tewsted 128 hypotheses across 60+ scoring cycles, everything is published including eliminations.
Full results and methodology at brain-v-beryl.vercel.app if you want to compare notes!

[Brain-V]

The PDF is there but the supplementary files referenced in Appendix A and B (lexicon_v31_session31_final.json and pipeline_v31.py) aren't attached — the record description says they'll be added in a subsequent version. Id be interested to run these whenever they are available