A New Interpretation of the Voynich Manuscript: A Statistical Evolutionary Code

[Simona Vargiu]

Hi RobGea,
thank you for your interest in my work and for raising these important questions regarding reproducibility. Let me clarify the methodological steps and also share the reasoning that initially led me to this approach.
My starting point was quite simple: I first tried to understand how many distinct symbols composed the manuscript in order to see whether it behaved more like a natural language or a coded system. While examining the manuscript — mainly through the digitized images available online — I was also struck by the illustrations: some of the shapes resemble musical notations or complementary geometric patterns. This inspired me to test whether the text could be mapped onto a code-like system, eventually leading me to explore a correspondence with the four DNA bases (as a metaphor for a minimal, structured code).

• EVA transcription: I used [specify version – e.g. the Interlinear Voynich Transcription available on René Zandbergen’s site / Takahashi transcription / another source]. This served as my reference corpus.
  
• Uncertain spaces and unknown characters (eva-?, gaps, uncertain readings): these were either removed from the analysis or mapped to a neutral placeholder that was excluded from statistical calculations, so as not to introduce artificial structure.
  
• Rare characters: they were included if present in the EVA corpus, but given their very low frequency they did not significantly affect the statistical results. In some tests I grouped them under a generic “other” category.
  
• Control texts (Latin and vernacular medieval): I used
  • . The choice was made to match manuscript length and structure as closely as possible.
  • Treatment of non-A–Z characters (ligatures, diacritics, abbreviations): these were normalized to their base Latin letters (e.g. æ → ae, diacritics removed, abbreviations expanded when possible). This ensured consistency across corpora.
    
  • Use of ChatGPT-5: I did not use it to generate results, but as an assistant for data preprocessing (e.g. cleaning inconsistent transcription entries, normalizing characters) and for generating reproducible code snippets (Python) that carried out the statistical analyses. All methodological decisions, mappings, and final results were made by me.
    
  I hope this clarifies the methodology and also the intellectual context that guided my choices.
  Best regards,
  Simona Vargiu