The Voynich Ninja - Analysis of the text

Relation between word types and vocabulary density.

Sun, 25 Oct 2026 11:13:56 +0000

There have been interesting discussions lately about @npcompl33t's seriation research. Admittedly, some of that is over my head, since my brain prefers things that are more concr33t. Some of the graphs reminded me quite a bit of my MATTR research. This is a while ago (2019) so I went digging in my notes, and used Claude to make some new graphs. Marco has been kind enough to explain some things that weren't clear to me yet and to suggest some improvements.

In my attempt to vulgarize the discourse, which is another purpose of this thread, I will no doubt introduce inaccuracies, so please do correct me if I'm wrong.

Let me first illustrate how Davis & Layfield's and npcompl33t's seriations reminded me of things I had seen before:

q20-orders-equal-mattr-ranking (1).png (Size: 288.52 KB / Downloads: 280)

I will try to explain in my own words what I think is happening here. If we assume bifolios are sequential and not supposed to be nested (singulions vs quire structure), then three seemingly different approaches point to the same order.

Davis & Layfield look at which words occur together. The technique they used, LSA, works at the word level and is interested in what kinds of words can be used to order folios. They remove the most common words, and then find connections between folios based on shared specific vocabulary.
npcompl33t's technique does not consider whole words at all, but only looks at which glyphs tend to follow each other. Pages that use glyph pairs in similar proportions sit close together.
MATTR looks at TTR, type-token ratio. It expresses how rich the vocabulary of a given text is, without caring about the specifics. You simply divide the number of different words (types) by the total word count (tokens). The higher the number, the richer the vocabulary.

All three seemingly different things lead to the same proposed bifolio order. This next graph compares MATTR to npcompl33t's MDS1, which is the main "line" of his analysis. You can think of this as the strongest way to spread out all folios based on their preferred glyph pairs. It basically sorts folios according to Currier A - B, since it focuses on the low-hanging fruit: prevalence of edy-type words typical for Currier B, and cho-type words.

bifolios-split-mds1-vs-mattr100.png (Size: 340.98 KB / Downloads: 280)

What we observe is that the more extreme B-folios (left) and the more extreme A-folios (right) also have the lowest vocabulary density. No doubt, all of these stats are related. If a page is more repetitive, this is felt in its "core" word types. Or the other way around, if a page falls back more on its more typical A vs B word types, it also becomes more repetitive.

I personally remain skeptical of the viability of seriation based on so little text, but that is another discussion. What does stand out for me here is that we can see some very clear divisions between and within sections.

Q13 stands out due to its relatively low density of vocabulary. You can see it all the way at the bottom of the graph. Some Q13 folios are notably worse than others.
Q20 splits very clearly. Two bifolios are almost as poor in vocabulary as Q13, and three bifolios are among the richest in the MS. But it all takes place on the B-side.
There is a lot of variability within Herbal A, but this may be due to its limited amount of text per page.
Some pages are strong outliers. What is even going on with f33/f40? Interestingly, the closely bound f34-f39 shows the largest distance between folios that are on the same sheet.
The two folios of Q8, especially 58-65, are near the middle on the MDS1, which is no surprise since Q8 is an inconsistent mess, distant from typical A and B.

On the evolution of bench gallows

Thu, 22 Oct 2026 15:35:21 +0000

This thread is a sibling to the one about P/F's evolution, but this one is focused specifically around Bench gallows.

Looking at bench gallows use across the manuscript arranged in my proposed chronological order shows some striking patterns. You can view this yourself You are not allowed to view links. Register or Login to view.(scroll down to see the graphs by page).

This chart shows the per 10k occurrences of ckh (first set of bars) and cth (second set of bars) across the manuscript, in chronological order.

  image - 2026-10-22T101433.733.png (Size: 600.08 KB / Downloads: 101)

I've annotated the chart to show the interesting features. First, note the void in section 1b. This occurs in the 18:23 bifolio. The first page of the bifolio, 18r, has above average cth compared to all prior usage. But this plummets on the two interior pages and the final page.

If we look at 18r, it seems the scribe was really struggling with the bench gallows forms. Look how many 'mistake' or 'aberrant' bench gallows are on this page.

  image - 2026-10-22T095733.267.png (Size: 1.95 MB / Downloads: 101)

Did the scribe just get frustrated and abandon bench gallows entirely for the rest of the bifolio? The first page on the next bifolio (f44r), cth is also is absent, although the following page You are not allowed to view links. Register or Login to view. contains the most cth occurrences in the entire MS.

The 2nd drop in pharma, also occurs mid bifolio. This one is a fold out, 87:90. There is a substantial drop from 87v to 90r, that persists throughout the remainder of the manuscript. This change is not subtle or gradual, it happens all at once, and persists. Ckh does not appear to be affected like cTh.

Looking at 87v, we can again see several apparent issues the scribe had writing the bench gallows, even more than in 18r. It seems they were consistently having trouble with the h glyph, with many missing or showing signs of corrections. Several sh's look abnormal as well, like the scribe may have started to use a bench gallows but then stopped and tried to correct it into an sh.

  image - 2026-10-22T100859.731.png (Size: 1.73 MB / Downloads: 101)

It seems extremely odd that cTh would experience such a sharp decline, mid bifolio, that persists over all subsequent pages. Did the scribe end up deciding that cTh simply was too much trouble?

both pages immediately after the cTh decline (18v and 90r), show substantial increases in q words. Both are the top 2 pages by number of q uses in the entire Currier A portion of the MS. The fact this occurs after both declines would seem to indicate the Q words are directly replacing the cth words.

  Screenshot 2026-10-22 101821.png (Size: 53.26 KB / Downloads: 98)

Whats more, it appears to specifically be qok, NOT qot words that are elevated.

  Screenshot 2026-10-22 102314.png (Size: 66.15 KB / Downloads: 97)

This strongly seems to suggest that cTh -> qok. I have not looked to see if word forms are comprehensively preserved. cthy for instance is one of the common cth forms that disappears from 18r->18v, and qoky does indeed appear in 18v. But other qok word appear that do not seem to have any cth equivalent, like qokchy, which is predominantly an A word, and possibly what evolves into qokedy if my previous hypothesis that chy->edy.

Much like the P/F discussion, this i think raises important questions about bench gallows, whether they can be expressed in other combinations of glyphs, and what forms replace them. Like P/F, there does not appear to be a consistent rule, which i think is troubling and supports different forms undergoing different mutations.

The evolution of P/F vs T/K raises important questions

Wed, 21 Oct 2026 19:21:04 +0000

I want to discuss P/F gallows, and their relationship to K/T. Throughout this post I am going to refer to my proposed 'chronological gradient'. If you disagree such a gradient exists, there are other places to argue that.

LFD's explanation for P/F is that they represent Ke/Te, based on the fact K is depleted in line openings. However, this explanation alone doesn't seem to line up with what we see in the MS.

K is only depleted in line openings in Herbal A, which almost entirely lacks any ke or te words. Pch is incredibly common, but tech and kech is nonexistent.

The charts below show the per 10k occurrences, arranged by my chronological gradient of

Ke or Te
Kch or Tch
pch

Screenshot 2026-10-21 124305.png (Size: 45.95 KB / Downloads: 249)

Note, the apparent gradual increase in Ke/te from the 1.5-2a sections, coincides almost exactly with the decline of tch and kch. Pch is slightly less common in 2a, but it's use is much flatter over the entire course of the MS.

It's clear from the above charts P/F cannot simply be Ke or Te, as Ke/Te did not become popular until pharma.

To better see the differences between P/F bearing words and K/T bearing words, i created the same MDS plots, but I filtered to only Gallows bearing words, and I replaced all gallows characters with . For each bifolia, i then pooled the

gallows, and the gallows.

For comparison, I also created a 2nd version, that uses Ke/Te instead of K/T. In this version, Ke/Te are collapsed to , so we can compare P/F to ke/te words directly.

This allows us compare word construction patterns between P/F, K/T and Ke/Te, along the MDS dimensions. When we do this, something very interesting happens.

In the chart below, each arrow points from the same bifolio's P/F words, to its K/T words.

  kt_vs_kete_mds.png (Size: 676.72 KB / Downloads: 239)

Immediately something is very obvious. P/F virtually always sit to the right of K/T. What's more, the distance between the P/F words and the K/T words increases over the course of the manuscript, as the chart below shows. Comparing P/F to Ke/Te does not resolve this issue -- in fact this creates substantially greater distance between the points in 1a/1b, while not particularly lowering the later distances.

  curve_distance_readable_full.png (Size: 236.03 KB / Downloads: 243)

If i turn this (pf)->(kt) into a vector, I can compare it directly against the local path for all words. Basically, this is seeing if the direction of the pf->kt matches the direction of the overall gradient. If the directions align, the arrow is green. If they point in the opposite direction, it is orange. The charts at the bottom report the actual angle of alignment.

  kt_vs_kete_path_alignment.png (Size: 690.45 KB / Downloads: 248)

We can see that overwhelmingly, P/F and K/T are separated in the direction of the gradient. This strongly implies that P/F are more conservative or 'lagged' K/T. In other words: at any given point, P/F are following older word construction patterns than K/T.

This raises many important questions.

Are P/F an 'archaic' or conservative 'K/T'?
Why did P/F not eventually get the same E-treatment as K/T?
What does this mean for Ke/Te words as a whole? Is simply some kind of shorthand or alternative notation that never is integrated with P/F?
Are the apparent coeval drop of ch and rise of e related? ch already strongly resembles ee.

f57v and f66r – The Primer Analogy (German: Fibel)

Wed, 21 Oct 2026 18:44:17 +0000

Hello everyone, please be patient with me, this is my first time posting here :-)

f57v and You are not allowed to view links. Register or Login to view. – The Primer Analogy (German: Fibel)

I am aware that many of these approaches have already been examined individually. What is particularly important here is the physical connection between You are not allowed to view links. Register or Login to view. and f66r. These pages were therefore not randomly selected to fit the idea; they already share a historical/physical connection (bifolium).

What I could not find in my research was an analogy between a primer (German: "Fibel", a traditional first-grade reading textbook) and the diagram on f57v. I therefore want to emphasize that I am not trying to take credit for other people's ideas, but rather to combine existing ideas with my own.

In this primer analogy, You are not allowed to view links. Register or Login to view. could represent a second-grade language textbook. The two pages would not necessarily have to be directly connected through readable text, but they could still be connected in terms of their meaning. What do I mean by this?

The following could therefore be possible: a didactic guide can contain repetitions, variations, isolated characters, combined units, and complete sentences right next to each other, without all of them serving the same function.

What the primer actually demonstrates (with german example)

→ P p
= Basic forms

The purpose here is to introduce characters. The same letter appears in uppercase and lowercase, as well as in different writing styles (print and cursive).

→ Pa – pa
= Syllable formation

The purpose here is to demonstrate construction. One character is combined with another already familiar character. This creates a readable unit.

→ Papa
= Word formation

This demonstrates the principle of repetition and combination. Two instances of "Pa" form a complete word.

→ Papa ist am …
= Application

Here, context is introduced. The same building blocks are then used in larger structures.

In the photographic example of a first-grade primer, the right-hand column shows the variations Papa, Pap, Pa, P, Pa, Pap, Papa. This demonstrates how something can be broken down and put back together, showing the logic behind the letters, syllables, and the word. However, "Pap" represents an intermediate element that only serves the formation of the word later on, without having a meaning of its own in the language. We only know this because we can read the language. With f57v, this is precisely the knowledge we are missing.

The difference between my interpretation of the rings and Matsuta's interpretation is, in my view, that he would already assign the procedures according to the positions of the rings. I see it more as a compact way of showing how the rings might interact to form a kind of "instruction manual".

If we consider the four rings in a way similar to Llull, several possibilities arise:

- One ring could introduce individual building blocks.- Another could show variations or possible combinations.- A further ring could contain more complex constructions.- The last could represent applications or another stage of processing.

BUT: The primer example also shows that these functions do not necessarily have to be distributed across four separate rings. Several functions can already appear within a single learning unit.

Nevertheless, one could conclude from this that repetitions and deviations within the ring sequences are not necessarily errors or meaningful textual variants. They could simply be systematically demonstrating the different possibilities.

One could also ask whether the characters are only combined linguistically, or whether they might also serve as a basic graphical arrangement or a construction principle for gallows and benched gallows. However, to investigate this, we would probably first need to understand the basic forms.

One logical, didactic consequence could be:

R2 (sequence of forms) → R4 (basic elements) → R3 (process of transformation) → R1 (formation of variants).

This sequence follows more of a logical progression if we stay with the primer analogy. But why a different order? If R1 is the ring of variants, it might need more space because several variants would have to be shown in greater detail. R2, on the other hand, needs space for its 68 character positions.

The idea of placing the sequence of forms before the basic elements comes from the primer analogy. It does not necessarily have to begin with the traditional ABC. It is conceivable that a limited selection of characters and their sequence could first be introduced, before explaining their structure, transformation, or combination.

However: This would first have to be confirmed by a statistical or paleographic investigation.

Theoretically, the unusual features identified by Matsuta might even support this idea ({&169} on f57v, {&170} on f57v, {&171} on f57v, and {&172} on both You are not allowed to view links. Register or Login to view. and f66r).

The three characters described only on You are not allowed to view links. Register or Login to view. could indicate a special function of this diagram page. The additional occurrence of {&172} reported on f66r, on the other hand, could be a possible concrete connection between the two pages. Taken together, this might also point towards an "instruction manual" or a "learning path".

f66r

If we now consider You are not allowed to view links. Register or Login to view. as a possible example of application, the question arises whether the three text areas might also serve different functions. Similar to the primer, they would not necessarily have to show the same content, but could represent different steps or variations of a common procedure.

Lisa Fagin Davis's study (2020) is also interesting in this context. Although You are not allowed to view links. Register or Login to view. and You are not allowed to view links. Register or Login to view. belong to the same bifolium, Davis assigns the two pages to different scribal hands (Hand 1 and Hand 5). This could point towards collaboration or at least different stages of work. Whether this also has any significance for the content naturally remains an open question.

Koen G. also pointed out the unusual characteristics of Quire 8 on the Voynich Ninja forum (October 2025). What stands out is that the connection often observed between scribal hand, Currier language, and page type does not entirely follow the usual pattern here. You are not allowed to view links. Register or Login to view. in particular occupies an unusual position. Whether this has any significance for a possible connection to You are not allowed to view links. Register or Login to view. would first need to be investigated.

Another interesting aspect of this folio is the large number of simple relationships between variants.

The five paragraphs on You are not allowed to view links. Register or Login to view. could also be compared structurally. Could different exercise or application steps be identified in connection with my assumption?

To what extent a tabular arrangement in three columns establishes a relationship between them is something I cannot demonstrate as a fact. It is possible that columns 1 and 2 form a kind of construction kit that serves as a reading or construction aid for column 3.

Alternatively, even a fixed relationship between horizontal divisions and vertical columns would be conceivable. This could be based on the five graphically apparent paragraphs (attached PDF, page 2, illustration "A"), or on further subdivisions within those five paragraphs involving characters from the respective column(s) (attached PDF, page 2, illustration "B").

It would also be conceivable that, under the assumption of a "Frame-Core Hypothesis" (German: "Rahmen-Kern-These"), the page might demonstrate by example how the definition of columns 1 and 2 affects the context in column 3.

My personal opinion is that the arrangement of the three columns was deliberate and was not added afterwards.

To test this hypothesis, however, a statistical analysis and, ideally, a paleographic investigation would be necessary. I would cautiously point out that purely statistical comparisons could potentially misinterpret the context of the primer analogy, since a direct repetition or correspondence between instructions and application does not necessarily have to be present. The important question would therefore be which relationships should actually be tested.

Interesting historical examples related to the primer analogy

1. Donatus, Ars minor, and Alexander de Villa Dei, Doctrinale

These works taught basic rules of grammar. They did not simply present reading texts, but organized learning around grammatical rules.

2. Ramon Llull – Ars brevis, 1308

Perhaps the most interesting candidate in this context: learning through regulated positions in circular diagrams. The diagram was therefore not merely used as an illustration, but as a tool.

Furthermore, it could have been known to an author (or authors) of MS 408. Other aspects of its content are also remarkable when considered in relation to MS 408.

Source: You are not allowed to view links. Register or Login to view.

Already discussed in the Voynich Ninja threads "The Lullian Apophatic Machine" (03/2026) and "A Cusanus Ladin Hypothesis" (07/2022).

Further sources

René Zandbergen – Voynich MS, Quire 8
Documentation of the shared bifolium f57/f66, the ring structure on f57v, and the division of the text on f66r.

Jorge Stolfi – Interlinear Transcription of the Voynich Manuscript
Transcription comparisons and descriptions of the unusual character forms and column structure on f66r.

Lisa Fagin Davis (2020) – "How Many Glyphs and How Many Scribes? Digital Paleography and the Voynich Manuscript." Manuscript Studies, Vol. 5, No. 1.

Mitsuro Matsuta (2026) – "Structural Analysis of the Voynich Manuscript: Elucidating a Knowledge System through Statistical Methods." Preprint, March 2026. DOI: 10.5281/zenodo.19058612.

Koen G. (2025) – Post #14, October 9, 2025, in the Voynich Ninja thread "Looking at You are not allowed to view links. Register or Login to view. in regards to the line as a functional unit." Discussion of the unusual combinations of scribal hand, Currier language, and page type within Quire 8, particularly You are not allowed to view links. Register or Login to view. and the f57/f66 bifolium.

You are not allowed to view links. Register or Login to view.

Translation note: This post was originally written in German and translated into English with the assistance of AI. The ideas, hypotheses, interpretations, and conclusions are my own.

example_57v_66r.pdf (Size: 718.26 KB / Downloads: 2)

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

Sun, 18 Oct 2026 21:34:03 +0000

@nablator asked me to explain the generator I mentioned, so here it is. Sorry, it got a bit long...

What it is

I call it "the scribe". It's a small program that writes a complete fake Voynich: same pages, same sheets, same paragraphs, same number of lines and words per line, same gaps for the drawings, Currier A and B, the five Davis hands. There is no meaning anywhere in it. It only follows writing rules that I measured first on the real text (ZL3b).

Why I did it

For months I collected rules about how the text is written. At some point you have to check if they are right, and the best way I found is this: if the rules are good, someone who follows them blindly should write a text you can't tell apart from the real one, at least in everything we know how to measure. If the fake text fails somewhere, a rule is missing or wrong.

How it writes

First it picks the habits of the sheet. Every sheet (the folded piece of vellum, four pages) has its own taste for some free choices: k or t, ch or sh, putting q in front or not, ending in dy or in y, and using words with e or without. The pages of the same sheet look alike in this, and inside a page it drifts a little from line to line.

Then, word by word, it decides to copy or to write something new.

About a third of the time (35%) it copies a word that is already on the page. It prefers the words close by, the chance halves every three lines or so, and it likes the word just above. In almost half of the copies it changes one letter, so qokedy can come back as qokeedy or okedy. One thing I learned the hard way: that change has to be something the hand would write anyway. More on that below.

The rest of the time it writes a new word letter by letter, and to choose the next letter it only looks at the two before. That is enough to keep the words in shape. They follow a sort of template: maybe a start (q, o, d, s or y), then a gallows or a bench, sometimes some e, and an ending. Words with e end in y or dy (qokedy, chedy) and words without e end in aiin, ar, ol (qokaiin, dar). It never puts two gallows in one word and never mixes e and i.

How a word starts depends on how the previous one ended. After a word in y, a word whose core starts with o usually gets q (shedy qokedy). A core starting with a gets d (daiin).

Then the line. The first word of a line likes to start with d, s or y, and avoids the same first letter as the line above (with ch it always avoids it, with s almost never). The last word gets "closed": daiin becomes dam, chol becomes choly, chor becomes chory. It does this more or less strongly depending on the hand.

And the paragraph. It starts with a gallows, the first line uses p and f instead of k and t, and the first line almost never copies.

That's it. It's seven steps, but some of the tables behind them are big, and they are learned from the text, so it's not a ten line rulebook.

How it looks

Real f78r, first lines:

Code:
tshedor shedy qopchedy qokedy dyqokol oky
qokeedy qokedy shedy tchedy otar olkedy dam
qckhedy cheky dol chedy qokedy qokain olkedy
yteedy qotal dolshedy qokedar chcthey otor dor or

The scribe, same layout:

Code:
qokod chol qokeey chedy qopar otom
o ral ydain shedy qody chody chals
qokedy qotedy qokedy qokeey qolal shedytedy ytam
tchdy shor qokeey qokaiinsh qokedy shedy qokaiin qoted

You can still spot some bad words in it, like shedytedy or qokaiinsh (two word endings glued together). I'll come back to that.

The measures

To compare the fake text with the real one I use 64 measures. Each one is a number you can compute on both texts the same way. Some examples, real value first and the scribe second:

How often a word is repeated right after itself, compared with what chance would give: 3.1 times chance in the real text, 3.3 in the scribe.

How many words have a "twin" earlier on the same page, the same word or one letter different: 64% real, 68% scribe. The nearest twin is usually two lines up in both.

How often the last word of a line ends in m: 15.9% real, 16.7% scribe. Closed endings in general at the end of a line: 27.1% and 27.3%, against about 2% inside the line.

How often a paragraph starts with a gallows: 93% and 92%. How many gallows in the first line are p or f: 31% and 31%, against less than 3% in the other lines.

How much the pages differ from each other in their habits, for example in putting q after y. If every page was the same this number would be about 1. Real 6.3, scribe 5.7.

The link between the end of a word and the start of the next (more below).

And an "eye test": how many words have a piece of three letters that appears 3 times or less in the whole real text. Those are the words that look wrong when you read them. Real 4.6%, scribe 4.1%.

A measure counts as reproduced if the scribe is within 25% of the real value, average of 5 runs.

What comes out

The scribe reproduces 58 of the 64 measures.

For comparison, a simple generator with three rules that I made before gets 33 of 64. So the rules definately make the difference, it's not that any generator gets there.

The test I like most: I let the scribe learn its tables only from the odd folios, and then write the even folios, wich it has never seen, and compare with the real even folios. It gets 51 of 64. That number alone doesn't say much untill you see this: the real odd folios against the real even folios match on only 55 of 64, because some measures change a lot from one part of the book to another. So 51 should be read against 55, not against 64. It means the rules describe how the text is made, they don't just memorize it.

And if I take out any single rule, measures go wrong. Without the copying, all the repetition measures fail. Without the habit of the first line, p and f dissapear from the first lines. Every piece is needed.

The 6 tests it fails

Exact repeats. A word written twice in a row is 0.9% of the word pairs in the real text and 1.2% in the scribe. Compared with chance it's fine (3.1 against 3.3 times chance), but in raw numbers it repeats a little too much.

Repeats at distance 5, the same word coming back five words later: 2.9 times chance in the real text, 3.8 in the scribe.

Words that avoid repeating. In the real text 16% of the common words almost never come right after themselves (dal, shey, saiin), while okaiin does it five times more than chance. The scribe has 21%. It only knows families of words, not single words.

The m at the end of long lines. Real lines of 10 words or more end in m 3.4 times more often than lines of 1 to 4 words. The scribe only gets 1.9 times.

q at the start of a paragraph: 4.2% real, 6.2% scribe. The numbers are small here, it's just over the limit.

Long words, 8 letters or more: 5.3% real, 7.4% scribe. Some of them are the two endings glued together you saw in the example (shedytedy, qokaiinsh).

Why I think it's interesting

The recent Harvard paper (Schwartz) says that no generator so far, theirs included, reproduces one thing: the link between the end of one word and the start of the next. They leave it as one of the few things unexplained.

In the scribe that link is just the habit I described above: the start of a word depends on how the previous one ended, and on nothing else from it. Not the whole word, not the word before, not the position in the line. Even the last pen stroke alone carries most of it, so it looks like a thing of the hand and not of grammar. Measured roughly the way the paper does, the real text gives 0.115 bits for hand 1 and 0.224 for hand 2 (the paper says 0.12 and 0.21). With the habit, the scribe gives about the same strength (after removing what a shuffled text would give: 0.097 against 0.089 for hand 1, and 0.171 against 0.205 for hand 2). Without the habit it falls to almost zero.

Two things I didn't expect

No dictionary needed. An earlier version made too many wrong looking words. Almost all came from the copies with a changed letter, because the change was random (daiin became daain, chedy became chddy). Giving it a list of real words fixed the look but killed the variety. What worked was making the change follow the same two letter rule as new words. Wrong looking words went from 8.7% to 4.1%, real is 4.6%.

The kind of word follows the sheet, not the plant. Words with e (qokedy, shedy) against words without (qokaiin, dar, ol): their share is shared by the pages of the same sheet, even with the other half of the sheet that sits far away in the book, and not with the next page if it belongs to another sheet. In the herbal, where every page has a different plant, the other half of the sheet matches at z = -7.0 and the next page at -0.7. So it behaves like the habit of the day it was written, not like the subject.

What it doesn't show

It doesn't prove the Voynich has no meaning. It shows that a procedure with no meaning is enough for most of what we can measure.

And the tables are learned from the text, so it says how the pieces combine, not where the letter rules come from.

Any questions, I'll be happy to reply.

[split] Does Voynichese evolve from A to B?

Sat, 17 Oct 2026 04:16:50 +0000

MODERATOR NOTE - Posts 1 to 23 were originally part of the You are not allowed to view links. Register or Login to view. and were moved to be their own thread]

(16-10-2026, 11:36 PM)pfeaster wrote: You are not allowed to view links. Register or Login to view.From the standpoint of potential parallels with Voynichese, I'm particularly interested in their finding that the writer of the Rohonc Codex seems to have been working out the details of his writing system as he went along, with rules that evolved over time.

But I don't get that impression from Voynichese. Yes there are differences between sections, but since we have no independent way of determining the order of writing or the internal structure, we cannot tell whether they are the result of the Author gradually evolving the language/script/encryption, as in the Rohonc Codex, or just determined by different topics/structure/style, possibly with one discrete but relatively modest "spelling reform" at some point.

(And yes, @npcompl33t claims otherwise. We are still fighting over that.)

By the way, the style of the nymphs tells me that Zodiac was drawn before Bio. And that Aries and Taurus were drawn before Pisces and the rest of Zodiac.

And the way some Pharma sketches were badly copied and expanded in Herbal tells me that Pharma is earlier than Herbal.

Also the bifolios of the Starred Parags section may have been scrambled and/or flipped over, but pages You are not allowed to view links. Register or Login to view. and You are not allowed to view links. Register or Login to view. were written last. (But, as I discuss in another thread, You are not allowed to view links. Register or Login to view. was all but completely erased by a water spill and wrongly "restored" by a later owner.)

All the best, --stolfi

Chronological Path Stability/Uncertainty Analysis

Fri, 16 Oct 2026 21:19:06 +0000

This is a follow-up analysis to my You are not allowed to view links. Register or Login to view. All of the code for this work can be found in the You are not allowed to view links. Register or Login to view..

Problem / Goal

The goal of this analysis was to answer "how uncertain / unstable are points on the proposed chronology". In particular, some follow-up analysis I have been conducting shows high sensitivity around the inclusion of f58, suggesting the placement of the missing H quire could change the ordering.

The core problem is, when tracing a path, the ends of the path have the least certainty. This is because points in the middle of the path have preceding and following points, which also have preceding / following points. These multiple overlapping observations provide much more certainty about the ordering.

However, points on the end do not have this. By definition, the first and final points would only have either a preceding or following connection. Depending on the direction and amount of movement, this can create uncertainty around which group is the first or last.

f58 and the missing H bifolio add this same uncertainty in the middle of the manuscript. We are missing part of the path, which creates additional uncertainty, specifically in the Herbal B ordering.

Method

I tested the ordering stability using both 207 individual pages and 50 pooled bifolia, holding the transition-distance/MDS representation fixed.

I utilized the ensemble procedure described in You are not allowed to view links. Register or Login to view. which involves fitting curves to 100 random approximately 90% subsets, using eight starting routes and two regularization settings. A whole-bifolium sampling control brought the total to 4,800 fits.I then compared recurring positions, order reversals and competing routes’ scores within each sample.

I ran You are not allowed to view links. Register or Login to view. to attempt to identify any discontinuous regions along the gradient.

To establish toe potential impact of the missing pages, and the creation of the the H-position MDS charts below, I used the You are not allowed to view links. Register or Login to view., applied derivative-based continuation at λ=.01, checking assignment changes and different movement paths.

A You are not allowed to view links. Register or Login to view. of exactly how each methodology was applied can be found on the voynich-clusters web page. All code is You are not allowed to view links. Register or Login to view.

Results

Instability Across the Gradient

The chart below shows the variance in the ordering of the individual pages based on different ensemble simulations described in ElPiGraph's paper. This shows where across the 4,800 different fit variants each page and bifolia fell. This allows us to determine how sensitive the positioning is to different starting conditions, assumptions, and modeling parameters. Darker regions indicate more variants placed the page in that area.

ordering_ensemble.png (Size: 161.66 KB / Downloads: 238)

Observations from the chart:

the ordering with respect to 1a and 1b is ambiguous, however 1a pages generally exhibit a slight initial ordering preference over 1b
1.5a and 2a's placements are generally stable.
f58, astro and cosmo are generally stable as a group. The only exception is Rosettes, which is the single isolated band not with the others in 5a, on the page level chart. It clearly was not made at the same time as the other 5a group, probably instead being produced alongside the Herbal B sections, but possibly even being made as late as stars 3c.
Herbal 3a and 3b's ordering is not stable, and will be discussed below in the next section.
At the bifolium level Stars 3c is generally stable, with f85r1 and r2 (the two unassigned bars ) also having the strongest preference for the same timing.
Bio / Stars 4c are more ambiguous, with fits generally preferring the ordering Stars 4c-> Bio

Checking for discontinuity

scEGRET is a method described by Zheng et al that applies directly to our problem. It is a toolkit designed to follow paths and identify discontinuous regions, allowing us to more firmly establish if we are missing any sections of the overall curve. The code is freely available on the authors github.

The general idea is this, along a well sampled progression, successive groups should generally change gradually. An unusually large change indicates the trajectory crosses a gap, for example, because intermediate data is missing.

scEGRET identifies no fully discontinuous regions across the gradient. Note, this does not mean certain areas along the curve are not missing pages. It simply means scEGRET was unable to detect one.

F58 does show a modest local signal. The scEGRET scores measuring relative change jump from .184 to .271, around a 1.5x increase. Calibration places this within the 95% confidence region. Removing f58 along with adjacent astro material entirely does produce discontinuity flags, but it depends on certain model settings and is therefore not conclusive.

So there appears to be no obvious gaps in the gradient, f58's region appears suspicious but is not enough to claim the region is discontinuous.

Impact of Missing Quire H on Herbal B

The chart below shows the impact the missing Quire H bifolia could have on the ordering of Herbal B's 3a/3b sections.

This is using the same MDS dimensions as the plots in the previous chronology theory,
The star shows the current placement of f58, the only remaining portion of quire H.
The colors indicate which ordering of 3a and 3b is preferred, based on where the missing pages of quire H would shift the current Quire H centroid on the MDS dimensions
Red regions mean if the missing pages shift the centroid there, the preferred ordering is 3a->3b. If the missing bifolia instead shift the centroid into the blue regions, then the preferred ordering is 3b->3a.
The black line divides the frontier between these regions

cassidy_H_regions.png (Size: 167.79 KB / Downloads: 236)

You can see the current f58 placement is almost directly on this red/blue frontier. This means without those missing pages, it is probably not possible to say for certain which Herbal B group comes first, because any slight changes from the current f58 point could alter the ordering.

Conclusion

The overall gradient is clear, but how confident we can be around the relative ordering of certain sections along that gradient varies. The largest uncertainty is:

The ordering of Early Herbal A sections 1a and 1b, although 1a is generally preferred as the starting point
The ordering of 3a and 3b, basically indeterminant without the missing Quire H
The ordering of 4c and 4a / Stars-T and Bio is indeterminant, although bio is generally preferred as the endpoint.

eva-m on f3/f6

Wed, 07 Oct 2026 08:19:23 +0000

I have been trying to think about eva-m a lot recently and I realised that almost all of the instances of this letter within Herbal A is on the one singulion which is f3/f6 (as per Lisa's analysis).

I'm not sure if this helps us figure out what this glyph actually is but it almost looks the scribe was testing it out or had some other reason to use it over and over again specifically here.

Also of interest is that it is used non line-final in multiple places in these pages.

For those who don't believe that 'm' is an line-final positional glyph (i.e. it doesn't encode for any information) or those who don't believe it to be an abbreviation for 'iin' what other theories are out there (that don't just say that it is meaningless)?

"rdy" is weird.

Sat, 03 Oct 2026 14:34:20 +0000

I think rdy is weird because of the apparent rules around where it is allowed to be used on a page / within text.

rdy (and to a lesser extent ldy) are almost only ever found ending lines, next to images, as labels or first word of page.
More simply, it does not seem to hardly ever exist in bulk text.

"So what?"
I have been wondering for a couple of days if all these cases are, or are part of, a label. Also how did the writer know this rdy containing word would land where it did? - Was it written first? Careful planning? - Random chance?
And also if this marker could be relatively reliable at finding where text starts/ends in unclear circular writing. If so, are there more that could aid in this.

Of 23 matches for rdy (Some of these could be in 2 category's),

2 are the first word of a page

  rd1.JPG (Size: 22.03 KB / Downloads: 146)

9 are in very close proximity to an image

  rd2.JPG (Size: 75.13 KB / Downloads: 153)

5 are the last word of a line

  rd3.JPG (Size: 36.54 KB / Downloads: 146)

4 are labels

  rd4.JPG (Size: 18.16 KB / Downloads: 145)

1 could be "bulk text" or a "long label(?)"

  rd5.JPG (Size: 7.22 KB / Downloads: 146)

1 could be "bulk text" or "line end" or "first word of a page", there's no indication on where the text starts/ends on this page.
It is found at "12 o'clock" on the page, so it's possible it's where the text starts or ends.

  rd6.JPG (Size: 7.03 KB / Downloads: 146)

1 is certainly "bulk text"

  rd7.JPG (Size: 6.71 KB / Downloads: 148)

So which ever way we slice the cake certainly 80%+ of cases are not in bulk text with 90%+ being possible.

"Choe" is weird.

Fri, 02 Oct 2026 20:49:10 +0000

I find Choe weird as e* (e followed by anything) happens 14263 times (Voynichese.com), egallows accounts for 833 - 5.84%

1.  choees - This one is actually "chores", discounted it from %
2.  choety
3.  tchoep
4. ychoees
5. choees
6. choety
7. choeky
8. choetchy
9.  choek
10. choekeey
11. choer
12 . choeea
13. choekcheey
14. choekey
15. choetey
16. choekeeos
17. choeky
18. choetchaldy
19. choeteedy
20. choepchey
21. qotechoep
22. choeey
23. choeesy
24. choeeer
25. choefy
26. pchoetal

When e is proceeded by cho the frequency of a gallows glyph following e jumps from 5.84% to 72%

It is also strange that 3 of 25 examples are double gallows words, which are extremely rare.

Harvard physicist's take on the VMS

Fri, 02 Oct 2026 12:04:12 +0000

Yesterday, Harvard physicist Matthew Schwartz unveiled a very powerful LLM harness for quantitative science called You are not allowed to view links. Register or Login to view., as well as a whopping 36 papers he has authored or coauthored using the tool. Among them is Schwartz's take on how the VMS was written, which is that it's iteratively generated gibberish:

You are not allowed to view links. Register or Login to view.

His abstract:

The Voynich manuscript is an illustrated book of the early fifteenth century in a script no one has been able to read. Since Rugg, and then Timm and Schinner, showed that text with many of its statistics can be produced mechanically, quantitative study has come to favor a writing procedure over a cipher or an unknown language. Here we make that picture precise with quantitative tools: information theory, efficient methods for mixture-model computation using exact arithmetic, and solvers validated on planted test cases. Readings and text generators become probability models, fitted on half a scribe’s pages and judged on the rest. In six candidate languages, readings by simple or verbose substitution, consonantal writing, abbreviation, codebooks and the Naibbe cipher all fail on withheld pages. Such failures exclude a reading only where the test detects disguised real text of the same kind, so far only for scripture. Knowing a word barely helps predict the next, less than in any comparison text, glossaries included, yet the edge glyphs of adjacent words are coupled as strongly as in a recipe collection. The scribes reused their words, occasionally copied a recent one, and coined others by letter habit; this keyless procedure reproduces nearly all summary statistics of both principal hands. No proposed generator, ours included, reproduces that coupling. Two weak traces of content remain: distinctive label words and a faint dependence of word choice on subject. The low cost of the writing and the genres the pictures imitate suggest a book made to be looked at, not read.

Traces of a wheel or similar

Thu, 01 Oct 2026 06:15:42 +0000

Hi,

I have been studying the text statistically the last days, and want to know if anyone else have noticed this.
I know tavie and MarcoP have shown that a line's first glyph depends on the line above, what I have found is that there is some sort of direction to this. Among d, q and y, consecutive line starts step d -> q -> y -> d more often than the reverse.

Example :
dain.ar.olshey.dytain.qokain.checthy.okeedy.qokeedy.ror
qokeey.qokedy.okeey.qokol.sheedy.qokeedy.rolchey.qokeedy
yteedy.qokeedy.qokain.olkeey.cheokain.dy,teey.qokain
dol.sheey.qol.olkshey.qokeedy.olkeedy.qol.okaiin.oly

These are the counts for the whole manuscript:
forward: d→q 125 q→y 75 y→d 119 (total 319)
backward: q→d 75 y→q 35 d→y 81 (total 191)

Total the wheel moves forward 63% of the time (expected 50%). The difference is about 5.7 standard deviations from the expected. Permutation tests give the same result (p<0.0001).

Frequencies alone can not produce this, if it had fixed probablities (50% d, 30% q, 20% y) the count would be expected to be the same in both directions (0.5*0.3==0.3*0.5).

This is valid through the whole document, not just parts of it. It could appear everywhere.

It also seems like ONLY the line preceding the line counts. Both lines need to have d,q or y to see this pattern. If one line appears between them, the pattern disappears (or so my initial investiagation shows). It seems like the scribe looks at the preceding line ONLY when dealing with d,q,y changes. There is no hidden "counter" keeping track.

The first line of a paragraph is outside (often other glyphs).

What I have checked:
Two transcriptions give the same result (ZL3b and voynichese.com)
Currier A,B and different scribes all follow the "rule".
It goes from start to start, not a seam between line breaks.
It is JUST the first glyph - no matter what the rest of the word is.
Currier A also has a similar rule, it seems : y-o-q-y
I can not find any direction anywhere else line to line.

What does it mean?
It could mean that some rule is governing these letters. It *could* be a wheel that is cycled, or some sort of weighted random variable. It seems like they move forward 2 of 3 times, roughly. Backwards 1/3 of the time.

Evidence for 2 rules linking words in Herbal A->B

Thu, 24 Sep 2026 02:20:32 +0000

In my opinion, the differences between Herbal A and Herbal B represent one of the most significant challenges in understanding the VM. Each contains words that are very common in one, but completely absent in the other. These changes go beyond simple word selection, as for example the bigram 'ed' is almost non-existent in Herbal A.

You are not allowed to view links. Register or Login to view. that Herbal A and B are not really different languages, and the differences can be explained by scribes encoding or spelling a group of common words in different ways.

For the last few days, I have been attempting to discover one of these 'transformation rules' that would repair the severed word distributions between the two different Herbals, and I think I have found a few that have compelling evidence.

I should also be clear -- these results are ONLY between Early Herbal A and Herbal B, the Later Herbal A / Pharma sections are not included. Some other work I've done, has led me to believe that Herbal B and other 'Currier B' sections contain significant differences comparable to those between Herbal A and B, so including them muddies the analysis. A natural follow-up is to see if similar transformations exist between Herbal A / Herbal B and other sections.

The rules:

interior (ie non-initial) ch -> ed, followed by y (qokchy becomes qokedy, otchy becomes otedy, etc)
interior (ie non-initial) ch -> chd, followed by y ( qotchy becomes qotchdy, qokchy becomes qokchdy)

Below, for convienience, i am going to refer to these rules as chy->edy, and chy->chdy, but i want to be clear that these are within-word-edits, not suffix swaps. This is not asserting that everything to the left of the edited portion is a common stem.

Lets go through the evidence:

They transform "Vanishing" A words into "Novel" B words, and the transformed words occur at roughly the same frequencies as the original.

I will explain this fully for the chy-> edy transformation, along with all of the data.

chy-> edy creates 17 pairs of A -> B words, each listed below.

For each pair of transformed words, we test (a) whether the frequency ORDER of the A words matches the order of their B forms, and whether the MAGNITUDE matches (the attested/predicted ratio, in both directions)

A word (rate in A | rate in B) ->B word (rate in A | rate in B), rates in per 10k words:

[*]
FORUM_chart_ed_rule.png (Size: 53.31 KB / Downloads: 366)
Across these 17 pairs, the A frequencies correlate with the B frequencies: Spearman ρ = .758, permutation p = .0004, remaining significant after Holm correction across 14 comparisons (adjusted p = .0047). So the frequencies and rates do overlap, statistically.

The main takeaway's here are that common A words map to common B words. Rare A words map to rare b forms. This is necessary but not sufficient, as to some extent this is showing how common the initial portions of words are, but just bear with me and I will address all of this. Crucially, you can also see that B versions have no A correspondence, while some A versions have a small B correspondence.

You should also note that some expected -edy forms occur at roughly 1.65x the rate of the -chy forms, which i will also discuss below. The 2nd transformation from chy->chdy does not have this same property.

The same conditions all hold true for the chy->chdy transformation, these frequencies and rate overlaps are statistically significant, even after holm correction.

FORUM_chart_d_rule.png (Size: 68.64 KB / Downloads: 371)

Alternate forms exist in both directions, where grammar permits

As i mentioned in the previous section, -edy forms occur at roughly 2x -chy. This seems problematic, until you run the operation in reverse, try to replace -edy with -chy, then examine the missing forms.

Take for example, chedy. Running this operation in reverse, produces chchy, which fails the established slot grammar in Zattera 2022.

Chedy, shedy, cheedy and sheedy collectively account for about 40% of the missing token mass, and these can be matched to corresponding A words with a single operation, d insertion

chedy can connect to chey by inserting d; it need not come from chchy.
shedy can connect to shey.
cheedy and sheedy can connect to cheey and sheey through d-insertion.

B has 0 tokens containing ped, fed, ded, led, red or yed. ped and fed will be discussed in the next section. Ded, led, red, and yed are excluded by Zattera’s proposed formal grammar (2022, Figure 4). How then, would dchy move from Herbal A to B? We can see that it uses d insertion, chy->chdy, and NOT the chy->edy.

dchy 23 → dchdy 4; no dedy.
opchy 6 → opchdy 3; no opedy.
qopchy 5 → qopchdy 3; no qopedy.
olchy 2 → olchdy 2

So now we have an explanation for why chy can go to either -edy or -chdy, as well as an explanation for the missing reversed forms of -edy in A. These rules appear to respect the known grammar of voynich, and attempt to avoid producing invalid words.

The transformations interact with other Voynich patterns as expected

This is where things get really interesting, and you may have already caught this in some of the examples of the previous seciton.

In her 2022 keynote, Lisa Fagin Davis proposed what I like to refer to as 'Davis expansion', that consists of the following rules:

P and F represent abbreviated forms of ke and te. BUT no abbreviation can be used before e, and no abbreviations can be used before d.

So for example using P to abbreviate keedy is strictly forbidden. How then, might an aspiring scribe, seek to use a P form for such a word? Is there simply no way?

Let us consider one of the rules from the previous section: chy->edy

What if we did something like keedy -> kechy -> pchy? This is now perfectly valid. In fact, P and F seem to affect the rest of the word, beyond Davis' proposed Te / Ke abbreviation, because these P/F forms appear to condition the word to use the ch over ed form.

In general, Herbal B authors tend to prefer ed over ch, except when it comes to P and F. So in certain contexts, the bench form appears to represent a preferred choice over edy, even for those that prefer edy.

A possible 3rd Rule

About 1/3 of the -chy tokens have no attested B form under either rule.

The higher count of these cases are:

chokchy 13
chotchy 12
chochy 5
ychy 4,shokchy 4; dchokchy 3.
shtchy, shytchy, chopchy, torchy, chykchy, otorchy, {ct}chy and dpchy: 2 each.

These are fairly unique voynich words, and interestingly enough, they do have fairly obvious equivalents, involving 2 edits: changing the first o following a bench to an e, followed by one of the rules from the first section.

For example:

chokchy → chekchy → chekedy (5) or chekchdy 1.

In fact, this pathway seems to exist for virtually all of these words.

chokchy 13 → chekedy 5, or chekchdy 1.
chotchy 12 → chetedy 1, or chetchdy 1.
chochy 5 → cheedy 2.
shochy 1 → sheedy 8.
chopchy 2 → chepchdy 1.
shopchy 1 → shepchdy 1.
chofchy 1 → chefchdy 1.

This initial e swap, after a bench, appears to have more examples, with different suffixes, and warrants more investigation to determine its full valid grammatical context:

choky 19 → cheky 22.
chokaiin 7 → chekaiin 5.
chokey 3 → chekey 3.
chotaiin 6 → chetaiin 1.
shoky 2 → sheky 5.

Words without a bench, like ychy, seem to not be valid targets for this rule, and remain unexplained.

Idea to find the word Root, Leaf, Stem, Flower & Seed.

Thu, 17 Sep 2026 09:45:56 +0000

Eg. Root:
1) find the labels representing root
2) find paragraphs that only mention those labels and not others
3) find words in those paragraphs that are unique to those paragraphs regarding roots
4) cross reference those words found in the paragraphs of other categories and narrow it down like that.

Please let me know what you think and if you can help me at all

Independent Analysis Identifying Languages Beyond Currier A/ B, compared to René Zand

Thu, 17 Sep 2026 00:01:43 +0000

Long time lurker but first time poster here. Professionally, I am a data scientist and for fun I enjoy digging into the VM.

First off, I'd like to start out by saying I have nothing but respect for RZ and the work he has done with respect to Voynich. I frequently use many of the tools he has created, and while I disagree with a few of his classifications and methods, we arrived at many of the same conclusions. I'm posting this mainly because his page on the subject asks for independent review of his parameters. <3 <3

I recently conducted my own analysis to understand if the VM contains any 'languages' beyond Currier A and B, and compared my results against those from You are not allowed to view links. Register or Login to view.(note languages here does not necessarily imply fully different languages, I'm simply using Currier's terminology).

My main motivation for undertaking this analysis is the decryption attack I’m pursuing. I have a substantial historical corpus from which I can obtain reliable botanical cribs in different languages, but the greatest barrier to applying this attack to the Voynich manuscript is the shortage of shared, botanically enriched candidates across Herbal A and Herbal B compared with the historical training works. Understanding the differences between Herbal A and Herbal B is therefore one of my top priorities.

I have two main concerns with RZ's approach. The first is that reliance on a few bigram rules could result in a false continuous gradient, that largely disappears when other features are included. The second is that the selected bigrams could completely miss large outliers, for example, as is the case with f58. While i can see the utility of using bigram segmentations as a language classifier, at least for my purposes, that type of segmentation doesn't capture enough of the underlying differences between the different language clusters.

Method

The approach i took was different than that of RZ.

I represented each page by its within-word transition counts and derived conditional transition probabilities. G² log-likelihood-ratio statistics were used to quantify feature contrasts and construct descriptive axes. Overall profile differences were measured using context-weighted Jensen–Shannon divergence, with clustering cross-checked using a Hellinger representation.

I tested candidate manuscript populations against their within-sample variation and whole-sheet comparisons, and examined whether similar groupings emerged when pages were clustered without supplying those population labels. My final results came from this combination of tests, it was not simply the output of one clustering algorithm. Classification rules similar to those supplied by RZ could be obtained from this process, but my goal was more focused around identifying the populations as opposed to developing ways to classify them.

This method produces segmentations that broadly are in agreement with RZ, although there are some meaningful differences that I'll discuss below. The fact that two different statistical approaches recover much of the same structure supports the theory that the Currier A / B classifications are insufficient to describe the variation between the text across the manuscript.

Results

My segments are below. The percentages below express JS divergence relative to the early Herbal A–Herbal B split, set to 100%.

1. Early Herbal A
RZ designation: A
Divergence from Herbal B: 100%

2. Late Herbal A + Pharma
RZ designation: Ae
Divergence from early Herbal A: 50%; from Herbal B: 80%

3a. Herbal B¹
RZ designation: B
Divergence from early Herbal A: 100%

3b. Stars / Recipes
RZ designation: broadly Bs
Divergence from early Herbal A: 115%; from Herbal B: 35%

4. Biological / Balneological
RZ designation: Bb; You are not allowed to view links. Register or Login to view. receives Cb at page level
Divergence from early Herbal A: 155–160%; from Herbal B: 60%

5. Celestial / hand 4²
RZ designation: overlaps C/Ce
Divergence from early Herbal A: 105–125%; from Herbal B: 60–80%

6. f58
RZ designation: A
Divergence from early Herbal A: 100%; from Herbal B: 75%

¹ Herbal B hand 2 supplies the comparison baseline.
² The range covers astronomical, cosmological and zodiac circle/radial text separately.

Some notes:

(3a+3b) are not sufficiently different to clear the bar for an individual 'language', but represent differences significant enough that they should not be considered the same and may confound some types of analysis

f58, as a single folio, does not enough evidence to say it is fully its own language, but its divergence from the rest of the text is extreme, far more than any other page in the MS, as I will explain more below.

Compared To RZ

RZ separates languages into A, B, or C using a single bigram, 'ed'. While this is indeed an extremely simple and efficient way of segmenting Currier A vs B, relying on it as the primary classification mechanism leads to some misleading and possibly erroneous conclusions like 'C is intermediate to A and B'. A page can be intermediate on the ed diagnostic while differing substantially on other construction patterns. An intermediate diagnostic value does not establish that some of the text is an intermediate form of Herbal A and Herbal B.

Ce as an independent dialect, in my opinion, is not supported. My threshold is a between group divergence of at least 2x, and a whole sheet permutation test with a p-value <= .01. Ce is 1.18x and .80, so it is not close, at least in my analysis. Certainly you can segment it with the rule eo > 16.7%, but this alone does not establish such a distinction is the result of different languages. The difference between approximately 22% and 12% could reflect two distinct settings, a continuous gradient, differing mixtures of constructions, or section-specific vocabulary. Group averages alone cannot distinguish those possibilities, especially with the limited text for that section.

Similarly, the single page in Cb, f80r, is not sufficiently different from other bio text to warrant its own dialect. You are not allowed to view links. Register or Login to view. remains firmly within the larger Bio profile (RZ does acknowledge the evidence for this section is lower).

The classifications of f58 and You are not allowed to view links. Register or Login to view. are examples of the limitations of this bigram threshold approach. f58 gets an A designation despite its large departure from ordinary Herbal A, while You are not allowed to view links. Register or Login to view. receives Cb at page level while remaining broadly Bio-like. There are, in fact, are several such outliers that are roughly about as divergent as f80r: 1r, f66r, f86v5 and f99v.

The outlier score below is calculated as JS divergence from the nearest page on another physical sheet divided by the JS divergence between the page’s own odd/even text-line halves. Below 1, another page is closer to it than its own two halves are to each other. Around 1: its distance from another page is comparable to its internal half-to-half difference. Above 1: even its closest external neighbor is farther away than that internal difference. This is a rough benchmark, it does not automatically establish a page as an 'outlier'.

Outlier ranking among 187 eligible prose pages:

- f58r: rank 1; score 1.763
- f58v: rank 2; score 1.673
- f86v5: rank 3; score 1.314
- f66r: rank 4; score 1.222
- f80r: rank 11; score 1.070
- f1r: rank 22; score 0.969
- f99v: rank 25; score 0.942

In my opinion, f58 is the only folio that is an obvious 'outlier', although the You are not allowed to view links. Register or Login to view. score does intrigue me. If only we had the other missing folios from this quire.

All of the above aside, i want to be clear, I do agree with most of RZ's proposed expansions to Currier A/B. In general, the amount of text divergence across sections exceeds that of other historical works and needs further analysis and explanation. For my attack, I would distinguish at least four main groups, with celestial material and f58 bringing the working inventory to six.