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

[quimqu]

Can you please refresh us on what makes the output of your generator have such long-range (page to page) correlations?

Hi Jorge. It doesn’t have any memory from page to page. When it copies a word, it only takes it from the few lines above on the same page, so nothing is carried to another page.
What makes pages look alike is that they share settings. Each hand and each section has its own glyph habits, so pages of the same hand and section resemble each other wherever they are. And each bifolium gets its own preferences for the free choices (k or t, ch or sh, q or not, dy or y), shared by its four pages, a bit like one writing session. That alone gives the Davis–Layfield pattern, without any notion of page order.
The real text still has more page to page similarity than the skribe  (0.28 against 0.15), so something is missing there (measured with LSA method).