The Voynich manuscript, an illustrated fifteenth-century codex written in an undeciphered script, has long resisted linguistic and cryptographic analysis. Despite more than a century of research, no framework has produced consistent, interpretable readings across the entire manuscript. This preprint introduces the Codex EVA–Romance v2.0, a novel methodological approach that combines stable phonetic-syllabic correspondences with a rhythm-based principle of reading. Building on the EVA transcription system, recurrent tokens were mapped to Romance or Latin lexical roots (e.g., shedy → herba, chedy → radix, chol → folia, daiin → amen, qokedy → facit). Crucially, meaning only emerges when these mapped tokens are parsed in three- to four-word “arrhythmic cycles,” which mirror the cadence of medieval ritual or herbal formulae. Without rhythmic grouping, the text collapses into incoherent lists; with it, recognizable micro-clauses such as herba radix folia facit amen become visible. To ensure reproducibility, the framework incorporates coherence scoring and negative controls. Shuffled and masked text sequences failed to produce coherent clauses, while original sequences consistently exceeded the 0.60 coherence threshold. Across ten botanical folios, 78% of tested lines yielded functional micro-formulae. Statistical analysis (Table 3, Figure 3) confirmed that coherence depends on sequential order rather than random alignment. These findings suggest that the Voynich manuscript is not a meaningless artifact or insoluble cipher, but rather a rhythm-dependent Romance text. While provisional, the Codex EVA–Romance v2.0 establishes the first reproducible and falsifiable pathway to functional readings, opening a new research direction in Voynich studies. This Version 2 expands upon the preliminary preprint Voynich Preprint in English (v1), now including a complete EVA–Romance lexicon (~12,000 tokens), arrhythmic principle, connectors, morphological variants, and statistical validation across the full corpus.