The Voynich Text Examination asks whether a scribe of the years 1404 to 1438 could have produced a text with the manuscript's measured properties using only what was on a desk of the time. This page lets a reader follow one such procedure word by word, with one throw of a lot for every word, and watch a page take shape. A second page, Writing Voynich by the Recipe Read off the Page, does the same for the other procedure that stands at the held-out floor.
The procedure is a simulation, not a discovery. Nobody knows how the manuscript was written. What the examination found is narrower. A person equipped with a few rows of word parts written on parchment, a bag of numbered lots, a ruler and a small working sheet can produce text with most of the manuscript's measured properties at a pace a book hand could keep. The simulated text matches the manuscript on of the 23 statistics measured in the report, within three times each statistic's own tolerance, and it does so with about random acts a word and an apparatus of about page sides. What it misses is set out at the end of this page. Nothing here claims that the manuscript was made this way.
Every table is a row of cells written out once, and a throw picks one cell. A common part occupies several cells and a rare one few, so the rows are filled in proportion to how often each part is used in the manuscript's own words. There are two sets of rows, one for each of the two writing styles that Prescott Currier named A and B in 1976, and the scribe uses the set that belongs to the page being written; the rows of beginnings and of endings are shared by both. The sizes below are those of the simulation, counted cell by cell from the rows as written out.
A word is written like this. The scribe throws one lot for the word and reads it against a short list of cut-points. In cases out of 100 the word is copied from the page: with the ruler laid on the line above, one of the five words nearest the current place, or one of the last two words of the line being written (the line above wins times in 100 when both offer a word, and the ruler goes two lines up times in 100). A copy is changed before it is written, except times in 100. In cases the first part of one word in sight is joined to the last part of another. In cases the word is taken from the working sheet, from the column headed by the last glyph of the word just written, and changed times in 100. In the remaining cases the word is built: a beginning from the row for that last glyph, then the rest of the word from the row of bodies that belong to that beginning, and times in 100 a second middle and ending are joined on. A change is made either by swapping one part for the next different cell of its row or by a rule from the list, read on from a thrown line until one applies and gives a word whose parts the rows know; part changes cut the word short times in 100, and a second swap follows times in 100. Words longer than letters are made again. All the yes-or-no decisions about a word are read off the rest of the same lot, so most words cost one throw and a few draws; only a draw from a row, a rule line, a place on the page or a letter for a habit costs a throw of its own. The pen habits come last: a tall letter in front of the first word of a paragraph, a plain letter in front of the first word of a line, loops on some k and t of the top line, and a tail on the last stroke of a line.
Choose one of six pages of the book and a set of lots, and press Throw to take one step of the recorded run, or Next word, Finish line and Finish page to let it run ahead. Nothing is simulated in the browser: every step replays a run recorded from the procedure itself, so what you see is exactly what it wrote. The panel headed "This step" explains what each throw decided. The page keeps the manuscript's own line lengths and paragraph breaks, so the finished page can be set beside the real one.
The report scores every generator on 23 statistics of the running text and compares each with the manuscript's value, allowing the uncertainty that comes from measuring the text in halves. The table gives this procedure's value averaged over five runs of the whole book. A statistic counts as reproduced when it lies within one tolerance of the manuscript's value and as nearly reproduced within three.
| Statistic | Manuscript | Procedure | Result |
|---|
A match of the measurements is a production account, not a reading: it shows what kind of procedure could leave these statistics behind, and says nothing about what the text means or whether it means anything. One part of the procedure is also at odds with what the pages themselves show. This procedure takes a word from the page in cases out of 100, from the two words before it on the line or the five words nearest above it under the ruler, changes it in all but cases out of 100, and in another cases joins the first part of one word in sight to the last part of another. On the manuscript's own pages, knowing the words in sight helps to predict a word no more than the page's vocabulary does on its own, while in this simulated text the copies and joins can be found again by looking. The measurements match; the copying does not. The tables were filled from the manuscript's own words, so the simulation shows that such rows suffice and not how a scribe would have composed them. The report gives the full account, the earlier work it builds on (Gordon Rugg's tables of 2004, Torsten Timm and Andreas Schinner's copying of 2020, René Zandbergen's wheels of 2021), and what the result does and does not establish.