Writing a fake Voynich by rules: copies, habits and the link between words

[quimqu]

To be clear about what my skribe knows and what it doesn’t, since that’s the key to the horseshoe.

From the real manuscript it takes the layout (pages per sheet, paragraphs, lines), which hand and section each page belongs to, and glyph tables per hand and section. Those tables are the same for every page of that hand and section. Nothing specific to any single page goes in. Each sheet gets its own habits drawn at random, and the words it copies are its own, from what it has already written on that page.

So two things need to be kept apart.

The big arc over the whole manuscript is expected. The skribe knows the hand and section of each page, and hands and sections are what place the bifolios along the arc. That’s why its overall order matches the real one so closely (0.86–0.92). That order comes from hands and sections, not from time.

Inside a quire it’s different. In Q13 and Q20 all bifolios share the same hand and section, so the skribe has nothing that tells one bifolio from another except its random sheet habits. Even so, the method finds a "secure" order there (0.65–0.76 in Q13), and it’s a different order from the real one, because it comes from noise.

So the part my skribe copies from the manuscript, hands and sections, which are styles, is enough to make the horseshoe without any chronology. And the part it can’t copy, the order inside a quire, is exactly where the method makes one up with the same confidence. The one place where the real text stands out is Q20 (0.95, above all my runs), and that’s the one worth a closer look.