Toward a Phonetic Decryption of the Voynich Manuscript: Correction of Methodological Bias (Version 2)

Version 1 of this study presented a computational pipeline for Latin phonetic recovery of the Voynich Manuscript (Beinecke MS 408), claiming an 89.3% validation rate via the Perseus Digital Library and the identification of 33 pharmaceutical terms. This Version 2 documents a critical methodological bias that invalidates those initial claims. Analysis reveals that the sheer scale of the Perseus dictionary (exceeding 265,000 entries) allows for the statistical validation of almost any Latin-sounding syllable sequence, creating a "mirage" effect of linguistic coherence. Furthermore, the use of permutation testing demonstrates that the previously observed correlation with the Antidotarium Nicolai is a statistical artifact of Zipf’s Law rather than evidence of genuine semantic content. While the codicological and structural observations of the first version remain valid, the phonetic decryption claims are hereby formally retracted. This corrective emphasizes the necessity of rigorous cross-validation and independent data testing in Voynich research. It redirects future efforts toward a purely structural analysis of the text (published separately, Clement 2026b), which interprets the manuscript as a personalized system of Tironian shorthand.