The book
The Voynich manuscript is a book of about 35,000 words kept at Yale University's library. It dates from the early fifteenth century, and nobody can read it. The tests on these pages found no translation either. Its text is written in an alphabet found in no other book. On these pages each of its few dozen letters is called a sign. The word letter is kept for real languages.
Its pictures show plants, stars and bathing figures. The page below shows a plant.
A page from the part of the manuscript that shows plants
These pages measure the text of the Voynich manuscript and test seven explanations of how that text was made. No translation is offered. Nobody can read the Voynich manuscript, and these pages do not read it either. The verdicts are in the section ‘The answer’ below. The page ‘What the tests conclude’ gives them in full.
These pages were written by computer programs, and the ‘we’ of these pages means those programs. They took the measurements and wrote these pages. No person has checked these numbers by hand. The checking that exists is programs re-running each other's work from the manuscript's own text. The main result of the page ‘One bundle of sheets’ was re-run from the raw words. Every figure came out identical.
One finding on these pages is about how the book was put together, and two scholars published it first. Layfield and Davis found that the two halves of a folded sheet use words more alike than neighbouring pages do. They conclude the book was meant to be read sheet by sheet. Colin Layfield and Lisa Fagin Davis published it in July 2026 in the journal Digital Medievalist. Their paper is called ‘Singulion Structure and the Voynich Manuscript’.
A book of this age is made of sheets of parchment folded in half. Fold one sheet and you get two leaves, four pages. Stack the folded sheets inside each other and sew them. That is a bundle. The two halves of one sheet then end up pages apart in the finished book. Only the innermost sheet is different. Its two halves face each other.
The programs that wrote these pages found the same thing as Layfield and Davis, on their own. They found it in one bundle of the book. We find what they found on one list of words and not on another. They had not read the paper, because they were set to work from the manuscript alone. What is added here is modest. In that bundle, take the pages that were once the two halves of one folded sheet. They share more of the book's middling-common words than any other way of folding those same leaves would give. Middling-common words are the words that are neither the book's commonest nor its rare ones.
Ranked against all 120 ways of folding those leaves, the real folding comes first. Here is how the 120 are counted. Ten of the bundle's leaves are in the pairing. Five come from the front half of the bundle. Five come from the back. A folded sheet always joins one front leaf to one back leaf. The first front leaf can join any of five back leaves. Each front leaf after it has one fewer to choose from. Multiplied together, those choices give 120 ways of folding them into sheets.
One of the 120 is the real folding, the pairing the book's own sheets give. The real folding was not established here. It comes from the published description of the manuscript. The question was whether the words prefer it over the other 119 ways.
The finding lives on one choice of words and dies on another. We counted two different sets of words, chosen two different ways. The result appears in one and not the other. The first set is the book's middling-common words. There the real folding comes first of 120. The second is the bundle's own words. Those are the words common inside it and rare elsewhere in the book. There the same test puts the real folding eleventh of 120. The finding is not found in the other bundle tested.
Real books written out into the same pages fall short of the manuscript on this test. We took real books and wrote them out into the manuscript's own page and line lengths. It is as if a scribe had copied those books into this one. Measured the same way, they fall short of the manuscript on the same kind of words. If real writing did what the manuscript does, the finding would be just something written text does. The page ‘One bundle of sheets’ gives every count with its limits.
What we did
We did not try to read the text. We measured how it behaves instead. A measurement is one thing measured on the text, such as the average length of its words. We made 23 measurements in all. We set every one beside the same measurement made on writing whose origin is known.
One measurement is how easy the next sign is to guess from the sign before it. A computer does the guessing, from how often each sign follows the one before. We also counted how often a run of four words appears twice in the same order. That repeat count is not one of the 23 measurements. The verdicts lean heavily on it anyway. The 23 are the yardstick every imitation is scored on.
We made the same measurements on 243 samples of real writing. The samples are passages cut from real texts, and real lists of words. The passages come from forty texts. Those texts are in thirty-one languages. The lists come from the early fifteenth century.
We then tested seven explanations of how the text was made. For some of them a computer program writes text by that explanation's rule. Text written that way is an imitation. The others we tested by comparing the manuscript with the samples, or by searching the manuscript itself.
We scored every imitation against the manuscript on the 23 measurements. The seven explanations, and the verdict on each, are on the page ‘What the tests conclude’.
The answer
Each explanation gets one of five verdicts. An explanation fails when text written by its rule comes out unlike the manuscript on the measurements. One fact runs through the verdicts. Every sample of real writing repeats some of its phrases. The manuscript almost never does. A cipher disguises a message. Nobody can see the message behind a cipher. So nobody can see whether that message repeated its phrases. The verdicts assume it did, since every real text sampled does. From the most certain verdict to the least, they are:
- Ruled out: the tests exclude it. No assumption is needed.
- Very unlikely: no way of doing it, tested or untested, can pass if the disguised text repeated its phrases. Every real text does. This verdict rests on that assumption.
- Unlikely: every way we tested fails. A way we did not test could pass. The assumption does not settle it.
- Undecided: nothing found supports it. Nothing rules it out.
- The best guess, not confirmed: closer to the manuscript than any other explanation. It does not pass every test.
Real writing repeats itself because a prayer is said again or an instruction is given twice. A four-word phrase is a run of four words. It repeats when the same run appears twice in the same order. Latin prose of the manuscript's length repeats dozens of four-word phrases, and the manuscript repeats one.
So the text is very unlikely to be a real language written in a made-up alphabet. A new alphabet only renames the letters, and it cannot remove a repeated phrase. A simple cipher always writes each letter or word the same way, so it keeps every repeated phrase too. That makes a language in a simple cipher very unlikely as well.
A changing cipher changes its spellings as it goes, and it can hide the repeated phrases. We tested 22 changing ciphers and all fail, but one kind was not tested. That kind keeps one spelling of each word through a page and changes it at the next. So a changing cipher is unlikely, a verdict one step short of very unlikely.
A message could also hide in some feature of the writing, not in its words. Take the first sign of every word, for example: read in order, those signs could spell out a message. We searched 29 features of that kind, the length of every word among them, and found no message. A message of 3,000 words or more is ruled out, and a shorter one could be missed.
Random meaningless text is words set down by chance. It is ruled out. In the manuscript a word used on a page tends to be used on that page again. Words set down by chance do not. A list or a catalogue, such as a list of names, is undecided. Nothing found supports it, and nothing rules it out.
The best guess is a procedure. A procedure is a fixed set of steps carried out by hand with dice and tables of words. Its imitations come closer to the manuscript than any other explanation's do. The tables in those imitations hold the manuscript's own words. The real tables, if there ever were any, are lost.
We built three procedures. The first has large tables of 18,000 entries. The second has small tables of a few thousand. An entry is one word or part of a word. The third has rules worked out from the manuscript's own pages. They come from counting which words copy a nearby word and which are spelled sign by sign.
A copied word repeats a word already standing nearby on the page, such as the word directly above. At most one sign is changed. A word spelled sign by sign is built afresh, one sign at a time. Each sign is chosen by the sign before it.
None of the three is confirmed. The large tables come out unlike the manuscript on two of the 23 measurements. The small tables and the third procedure fail one harder test. The easier test fills the tables with words from the whole manuscript and scores the procedure on the whole manuscript. The harder test fills the tables from half the manuscript only, so they hold only the words of those pages. It then asks the procedure to write the other half.
Both the small tables and the third procedure then write too few words that appear only once. The manuscript's distinct words are its words with repeats dropped. About two thirds of them appear only once. In the pages the small tables and the third procedure write, far fewer do. The large tables were never given the harder test, so how they would do on it is not known. The best guess page describes the procedures and both tests.
One more observation is about the people who wrote the text. Five scribes wrote the book. Lisa Fagin Davis showed in 2020 that their handwriting can be told apart. Two of them use many of the same uncommon words, and what produced that overlap is not settled. It leaves every verdict as it is.
The size of the work
We compared the text with 243 samples of real writing. The samples are in thirty-one languages. For the changing cipher alone we ran 22 programs. We searched 29 features of the writing, such as the length of every word, for a hidden message. The small tables are the best of some forty kinds we tried.
In all we used 211 techniques. Of those, 32 are new. A technique is one method of measuring or testing the text. We refer to 104 published works.
These pages are one part of a wider search. The page ‘What was tried’ gives its size.
The pages
The pages fall into three groups. Each is listed below with one line on what each page answers.
For the general reader
- ‘What the tests conclude’ gives each verdict with its reasons, and the finding about the two scribes.
- ‘One bundle of sheets’ gives the only finding on these pages about how the book was made. Layfield and Davis published it first. Every check made on it here is on that page.
- ‘The 23 measurements’ lists the measurements and says how an imitation is scored on them.
- ‘The best guess: a procedure’ describes the three procedures and the harder test.
- ‘What was ruled out, and why’ takes the failed explanations one by one, with the evidence against each.
- ‘What was tried’ sets out the tests as questions and answers, and gives the size of the wider search.
- ‘What is new here’ says what was known before, what was added, and which published claims fail.
The detailed pages
Fourteen pages open with a short summary. Ten of them are the pages listed here. The other four come first in the imitation research below.
- ‘The findings in brief’ gives every finding with the measurement behind it.
- ‘What the text is like’ measures the signs, the words and the phrases.
- ‘Is it a language, or a cipher of one?’ tests the language and cipher explanations.
- ‘Is there a hidden message?’ tests every feature of the writing a message could hide in.
- ‘Does the text carry content?’ asks whether the text is about something. For example, do its words go with its pictures?
- ‘Verdict’ gives each verdict and says what would overturn it.
- ‘What is and is not established’ sorts the findings by how sure each one is.
- ‘The data and how far to trust it’ describes the photographs, the transcriptions typed out from them, and the samples.
- ‘Compared with published work’ lists every method used and says which were applied before.
- ‘Works referred to’ is the reference list. The 104 works referred to are listed there.
The imitation research
These pages tell how we built and scored the programs, and how we found the procedures.
- ‘What produced it’ scores the programs for each explanation and names the closest, the procedure with large tables.
- ‘A hand procedure of the period’ asks one question. Could a scribe of the time have carried out the procedure with large tables?
- ‘The search for a smaller procedure’ looks for the smallest tables that match. It tests them on pages the tables were not built from.
- ‘Reading the procedure off the page’ asks how each word was made. Was it copied from a nearby word, or spelled out sign by sign?
- ‘Writing Voynich by Hand’ lets the reader carry out a small procedure, one throw of the dice for each word.
- ‘Writing Voynich with the Large Tables’ does the same for the procedure with large tables.
- ‘Writing Voynich by the Recipe Read off the Page’ does the same for the third procedure. That page calls the procedure a recipe.
- ‘Four Dice, One Tablet’ shows the smallest kit. That is dice, tables, and a wax tablet holding a working sheet of thirty words. A wax tablet is a wax-coated wooden slab that can be written on and wiped.