A Structural Process Analysis of the Voynich Manuscript Cross Transcription and Nuisance Controlled Validation

This study evaluates a frozen set of page-level features derived from a pre-existing symbolic process ontology against multiple scholarly transliterations of the Voynich Manuscript. Version 4 extends an earlier 226-folio analysis through three transliteration corpora, 1,000-permutation leave-one-quire-out tests, cross-transliteration stability analysis, and adjustment for Currier language and scribal hand. MST-only balanced accuracy under whole-quire holdout was 0.459 in RF1b, 0.408 in ZL3b, and 0.476 in the independently originated Takahashi-derived IT2a transcription. Each observed value exceeded all 1,000 corresponding shuffled-label results under the stated unconditional grouped null. Median page-level feature correlation across corpus pairs was 0.989 to 0.991. Adding MST features to language and hand metadata increased balanced accuracy by 0.169 to 0.213. The result is a cross-transcription distributional and methodological finding, not a decipherment. It does not establish plaintext, pronunciation, grammar, historical language, unique translation, or verified semantic meanings for the English process labels. A previously perfect biological-versus-herbal result is retained as a documented failure because it is fully confounded by quire. A protected blinded visual-validation protocol is frozen for future independent execution.