A Candidate Decipherment of the Voynich Manuscript: Evidence for a Phonetic Transcription of Spoken Elu-Sinhala (V2)

Version 2 of the candidate decipherment of the Voynich Manuscript, identifying it as a 15th-century Sri Lankan Elu-Sinhala pharmaceutical text. Major changes since V1: V17 decoder canonical (resolves V1 §15 u-prefix anomaly: 13. 8% to 5. 0%) ; Bowern-Gaskell engagement with replicated random-forest classifier (V17-decoded Voynich crosses to meaningful while authentic Sinhala recipe text classifies as gibberish) ; hostile-reviewer cross-corpus test (3. 4× SL/pan-Indic ratio under 50, 000-token size-matched control) ; external grounding of two VPNS state-markers (leda + seda) ; two falsification probes passed (Vinaya 0/195K, Samantapasadika 0/265K VPNS markers) ; Cūḷavaṃsa 37. 146 primary-source documentation of King Buddhadāsa's medical compendium; Elu chronolect revalidation (81% Elu-native, 0% post-12c Sanskrit loans in LOCKED vocabulary) ; Bhesajjamañjūsā re-OCR completed via pdftotext (40. 5% type-match, 33. 9% size-matched) ; q-/ch- revised to phonologically-conditioned allomorphs of one deictic morpheme; register-specific grammar (formal BNF) ; polysemy disambiguation; parallel-recipe template matching (f75r line 38 matches Bodleian enumeration at Bonferroni-corrected P ≈ 10⁻⁷ across ~26, 000 line × template comparisons). P (Sinhala identification wrong) ≈ 5-8%. Licensing: ORIGINAL CONTENT (paper, code, analysis writeups) is CC-BY-4. 0. Bundled third-party reference corpora retain their licenses (GRETIL CC-BY-NC-SA; Wellcome and thesis materials under research fair-use). See LICENSE file in deposit. To redistribute under pure CC-BY-4. 0, remove references/medicalcorpus/iastclean/ and the catalogue/thesis OCR before re-upload.