Voynich Scribes: Five Hands, One Hand and the Production Problem

There is a very seductive equation in Voynich research.

different handwriting = different person.

It sounds reasonable.

Sometimes it is right.

But the moment we turn it into:

different person = different author = different language = different topic = different production phase

we have crossed several evidentiary bridges without checking whether any of them actually exist.

The Voynich Manuscript makes this especially dangerous because handwriting variation overlaps with other strong kinds of variation.

Currier A and Currier B.

Different visual page families.

Different physical bifolia.

Different distributions of labels and paragraph text.

Different stages of a book whose original sequence is not fully secure.

In 2020, manuscript scholar Lisa Fagin Davis published one of the most influential modern palaeographic analyses of Beinecke MS 408 and proposed that five distinct scribal hands could be identified in the principal Voynich writing.

That model immediately changed the conversation.

One mysterious author became a possible collaborative production.

One writing stream became several hands.

And because some hand assignments correlated with Currier distinctions, the five-hand model began to influence linguistic and computational work as well as palaeography.

Then the argument became more interesting.

In 2026, Torsten Timm published a direct critique arguing that the supposed diagnostic differences may occur more continuously than discrete five-scribe categories suggest, and that some hand classifications may partly reproduce distinctions already visible in Currier language or page type.

The result is not that one side has made handwriting useless.

The result is that the real research question has become better.

What exactly does handwriting variation measure in a manuscript whose text, page roles and production batches also change?

That is the scribe problem.

Quick Read

One-sentence answer: the Voynich Manuscript clearly exhibits handwriting variation, and Lisa Fagin Davis’s peer-reviewed 2020 study proposed five distinct scribal hands, but the number of underlying people remains debated because handwriting can vary by time, pen, speed, context and text type, while a 2026 critique argues that some of the variation may form a continuum rather than five discrete categories.

Palaeography studies handwriting forms, habits and recurring diagnostic features.
A “hand” is an analytical attribution to a writing style; it should not automatically be equated with author, composer or intellectual owner.
Davis’s 2020 article proposed five principal scribal hands in the Voynich Manuscript.
Her model uses recurring details in glyph construction and scribal habit rather than merely broad page appearance.
The proposal is peer-reviewed and has become influential in later Voynich work.
Some hand assignments correlate strongly with Currier A/B and with particular page families.
Correlation is useful but creates a confounding problem: are handwriting, textual regime and page function independent variables or several views of one underlying production difference?
A 2026 critique by Torsten Timm argues that the diagnostic features used to separate hands are distributed more continuously and may not require five discrete people.
That critique is current dissent, not a peer-reviewed replacement consensus.
Even if five hands are real, five hands do not prove five authors.
One author can dictate to several scribes; one scribe can copy several authors; a workshop can divide labour.
Even if one hand were ultimately responsible for most text, the manuscript can still contain multiple sources, registers, topics and production states.
Bifolium-level consistency is especially important because physical sheets often preserve coherent textual or handwriting states.
The strongest current conclusion is that production was organised enough for handwriting variation to interact systematically with physical and textual structure.

The number “five” matters.

The deeper question matters more:

what kind of production system would generate the handwriting variation we see?

What Is a Scribal Hand?

In manuscript studies, a hand is a recurring pattern of handwriting attributed to one writer or one distinguishable scribal practice.

Researchers compare details such as:

how a stroke begins;
how a loop closes;
where a crossbar meets a stem;
how a flourish is formed;
pen angle;
relative proportions;
spacing;
preferred ligatures;
habitual variants.

The important word is recurring.

One unusual letter is not normally enough to define a scribe.

People vary.

Ink changes.

Pens wear down.

Hands become tired.

Writing size changes with available space.

Formal and informal writing differ.

A useful hand attribution therefore depends on a constellation of habits strong enough to separate natural variation within one writer from variation between writers.

Why the Voynich Script Makes Palaeography Harder

Ordinary medieval scripts can be compared with thousands of dated examples.

Voynichese has no securely identified external script tradition.

That changes the problem.

A palaeographer can still compare handwriting within the manuscript, but it is harder to know which differences are normal variants of a script convention and which are personal scribal signatures.

If five medieval scribes all learned the same ordinary Gothic hand, we know roughly what shared schooling and personal variation look like because enormous comparison corpora survive.

If the Voynich script was invented for this project, the surviving manuscript may be nearly the whole population.

We are trying to distinguish writers while also learning what “normal handwriting variation” means for the script itself.

That makes internal controls especially important.

What Davis Proposed in 2020

Lisa Fagin Davis’s 2020 Manuscript Studies article, “How Many Glyphs and How Many Scribes? Digital Paleography and the Voynich Manuscript,” used digital images to compare recurring glyph forms across the codex and proposed five scribal hands.

The model mattered for several reasons.

It brought professional manuscript palaeography directly into a field often dominated by linguistic and cryptanalytic arguments.
It treated handwriting as a manuscript-production problem rather than merely a character-transcription problem.
It suggested the codex may be collaborative.
It gave later researchers a new metadata layer with which to compare vocabulary and statistics.

Most importantly, the proposal can be tested against the physical book.

Do proposed hand changes occur cleanly at bifolium boundaries?

Do two halves of one sheet normally share the same hand?

Do textual regimes correlate with hands?

Do exceptional pages exist?

This is exactly the kind of model Voynich research needs: one that can interact with several independent evidence layers.

Five Hands Does Not Mean Five Authors

This distinction is foundational.

A medieval manuscript can be:

dictated by one author to several scribes;
copied by several scribes from one source;
compiled by one person from multiple sources;
copied by one scribe from several authors;
assembled by a workshop whose intellectual and physical labour are divided.

Therefore the chain:

five hands → five authors

is not automatic.

Nor is:

five hands → five languages.

The handwriting layer tells us about production.

Authorship and language require additional evidence.

One Person Can Produce Several Handwriting States

The reverse mistake is also possible.

Handwriting is not a biometric fingerprint fixed perfectly for life.

One writer can change with:

age;
speed;
fatigue;
pen;
ink;
page scale;
formality;
time between production sessions;
deliberate script style.

That is why a “one hand” hypothesis cannot be dismissed merely because pages do not look identical.

The real comparison is not:

are the pages different?

They are.

It is:

is the observed variation better explained by several stable hands or by continuous within-writer variation plus contextual effects?

That is a model-comparison question.

The 2026 Critique Changes the Shape of the Debate

Torsten Timm’s 2026 critique argues that some of the glyph variants used to distinguish hands occur much more broadly across the manuscript than a discrete five-scribe model implies.

He also argues that certain hand groupings overlap strongly with pre-existing text categories such as Currier A/B or with pages dominated by labels, raising the possibility that some apparent scribal distinctions are not independent of textual context.

The critique proposes a more continuous alternative: handwriting may evolve gradually alongside vocabulary and other manuscript properties.

This is a serious challenge worth including in any 2026 account of the scribe problem.

It should also be calibrated correctly.

Davis’s five-hand proposal is peer-reviewed palaeographic scholarship.

Timm’s critique is a recent independent challenge deposited publicly with supporting data.

The existence of dissent does not turn the five-hand model into a mistake.

It turns “five scribes” from a fact-like public shorthand back into what it should always have remained:

a strong, influential hypothesis about the manuscript’s production that still has discriminating work to do.

The Key Confounder: Currier A and B

Suppose Hand 1 writes mostly Currier A and Hand 2 writes mostly Currier B.

Then a computational classifier can separate their pages using vocabulary.

Has it identified handwriting?

Or language regime?

Or both because the two variables genuinely correlate?

This is a classic confounding problem.

One latent production variable may generate several observable differences at once.

For example, a different scribe could use a different orthography.

Or two scribes could be copying two source traditions.

Or one scribe could change textual regime when copying a different source.

To separate the causes, we need pages where variables cross.

The most informative cases are not the pages where hand and Currier regime agree perfectly.

They are the pages where they do not.

Bifolia Are a Powerful Independent Axis

The physical sheet gives us something vocabulary cannot fake.

A bifolium is one piece of parchment folded to create two leaves.

If a proposed hand normally stays consistent across both halves of a bifolium, that supports a sheet-level production model.

If handwriting switches in the middle of one physical sheet, the model needs a reason.

Likewise, Currier’s observation that herbal bifolia tend to preserve one text regime is powerful because it connects statistics to manufacture.

When handwriting, Currier regime and physical sheet all align, the evidence for distinct production states becomes stronger.

But even then the hidden cause remains open.

Different writer?

Different source packet?

Different working session?

Different encoding table?

Physical alignment narrows production.

It does not by itself identify the human explanation.

A Workshop Model Can Explain More Than Either Extreme

The public debate often feels binary.

Five scribes.

Or one scribe.

Real manuscript production can be more complicated.

Imagine a workshop with:

one compiler;
two or three principal scribes;
one illustrator;
one person adding colour;
shared training in an invented script;
several source packets.

The resulting book can show both discrete and continuous variation.

Scribes imitate one another.

A compiler occasionally writes a page.

A principal hand changes over time.

Labels may be added in a different pass from prose.

This is why the correct answer may not be a clean integer discovered from one clustering algorithm.

The production system may contain hierarchy and overlap.

Hand Does Not Equal Topic

A scribe may specialise in one section.

That makes hand and topic correlate.

But correlation can arise from labour division rather than intellectual authorship.

One scribe is assigned plant folios.

Another copies Quire 13.

The vocabulary differs because the topics differ.

The handwriting differs because the scribes differ.

Now a statistical classifier sees two clusters.

It cannot know which part of the separation comes from topic and which from scribe unless the variables cross elsewhere.

This is why page families should not be used as automatic confirmation of hand assignments.

Hand Does Not Equal Language

The same applies to Currier A/B.

A scribe can copy two languages.

Two scribes can copy one language.

One writer can use two orthographies.

A workshop can assign different notation states to different scribes.

Therefore a mapping such as:

Scribe 1 = Language A

may be descriptively useful without proving a natural-language identity.

The hidden variable could be source, orthography, code state or workflow.

The scribe model should constrain language hypotheses.

It should not silently decide them.

Hand Does Not Equal Chronology

If one proposed hand appears earlier in the current codex and another later, chronology becomes tempting.

But present page order is not guaranteed to be production order.

Gatherings may have been produced separately.

Missing leaves interrupt sequence.

The current binding may not preserve every original adjacency.

Even with perfect hand identification, chronology needs physical or material ordering evidence.

Five hands can work simultaneously.

One hand can work across years.

Hand attribution tells us who-like.

It does not automatically tell us when.

The Strongest Pages Are the Crossing Cases

If we want to know whether hand drives vocabulary, we need cases where one proposed hand writes more than one textual regime.

If we want to know whether topic drives handwriting perception, we need one proposed hand across multiple page families.

If we want to know whether Currier and hand are independent, we need pages where they disagree.

These crossing cases are statistically less tidy.

That is exactly why they are valuable.

Perfectly aligned categories cannot tell us which category is causal.

Exceptions can.

The page that refuses to fit both maps at once may be more informative than the hundred pages that fit them perfectly.

Could One Scribe Deliberately Change Style?

Yes.

Scribes can write formally and informally.

They can imitate exemplars.

They can adopt a display hand for labels.

They can enlarge or simplify forms according to page scale.

This possibility becomes especially relevant in a manuscript containing radically different text geometries.

Running prose.

Short labels.

Circular text.

Radial text.

Starred records.

A hand classification should therefore control for writing task where possible.

A label hand may look different because labels are written differently, not because a different person wrote them.

Could Several Scribes Learn One Invented Script?

This is the mirror possibility.

If Voynichese uses an invented or specialised script, several people could share its conventions.

That would imply a transmission event.

Someone had to teach or specify:

glyph inventory;
stroke order;
spacing;
ligature conventions;
positional rules;
possibly encoding rules.

If five writers can use the system coherently, the writing system is not a purely idiosyncratic private doodle.

It becomes a teachable technology.

This would be historically important even before decipherment.

A multi-scribe solution implies a community of practice, however small.

What a Five-Scribe Model Would Predict

A strong discrete-hand model should eventually predict several things.

Diagnostic handwriting variants should cluster by proposed scribe more strongly than expected from page type alone.
Independent palaeographers should assign unseen samples similarly.
Within-scribe variation should be narrower than between-scribe variation after controlling for text role and page geometry.
Physical bifolia should usually respect hand assignments or reveal interpretable production exceptions.
Computational features based on stroke shape should classify hands without relying mainly on vocabulary.

The final condition is important.

If a “hand classifier” succeeds because it is actually detecting Currier vocabulary, it has not independently validated handwriting.

A handwriting model should be able to see hands when the words are hidden.

What a One-Hand Continuum Model Would Predict

A strong one-hand or continuum model makes different predictions.

Supposed diagnostic variants should overlap broadly rather than form clean scribe-exclusive sets.
Handwriting distance should change gradually across production sequences or related physical units.
Apparent clusters may weaken after controlling for Currier regime, page type and label-versus-prose differences.
Exceptional pages should bridge rather than sharply separate categories.
One writer’s evolving habits should explain multiple proposed scribe features under fewer assumptions.

The model gains strength if one smooth trajectory predicts both handwriting and textual change.

It weakens if truly discrete stable signatures appear repeatedly across widely separated pages without intermediates.

Again, the important feature is prediction.

“One hand” and “five hands” should become competing generative models, not identities defended by impression.

The Answer Could Be Three

Or four.

Or five principal hands plus occasional secondary intervention.

Or one principal hand plus one specialist.

This sounds trivial, but it matters because public debate often hardens around named endpoints.

Five.

One.

The true production system does not care about our debate format.

A better model-selection process should allow intermediate answers if they explain the evidence with fewer exceptions.

The point is not to defend an integer.

It is to recover the workflow.

Production Workflow Is the Bigger Prize

Imagine we eventually know the exact number of hands.

Why would that matter?

Because it changes the social history of the manuscript.

One writer suggests one kind of production story.

Several coordinated writers suggest another.

A multi-scribe codex implies organisation.

Who allocated sheets?

Who controlled the script?

Who supplied source material?

Were text and images coordinated before or after writing?

Were labels added in a second pass?

Were Currier states tied to people or source packets?

The scribe count is therefore not merely a census.

It is an entry point into production architecture.

A Multi-Scribe Manuscript Would Change the Cipher Question

If several scribes used the same encoded writing system, the system had to be transferable.

That creates practical constraints.

A highly personal encoding method that depends on one author’s improvisation becomes less plausible.

A shared table, rule set, exemplar or training process becomes more plausible.

If different scribes correspond to different encoded states, the workshop may have used several tables or conventions.

If one scribe wrote everything, a continuously evolving private notation becomes easier to imagine.

The scribe problem therefore changes the space of plausible decipherment mechanisms.

It does not solve them.

A Multi-Scribe Manuscript Would Also Change the Authorship Question

Modern books teach us to imagine one authorial mind.

Historical manuscripts are often more collaborative.

The intellectual source may be one person.

The compiler another.

The scribes several.

The illustrator another.

The patron another.

Therefore even a perfect five-hand map might leave the “author” question fundamentally misframed.

Perhaps the better historical question is:

what network of people and sources was required to make this object?

That question scales better to the evidence we actually have.

What Survives the Current Debate

The principal Voynich writing shows real palaeographic variation.
The variation is systematic enough to support serious hand classification.
Davis’s five-hand model is an influential peer-reviewed account and should be treated as a major hypothesis, not casual speculation.
Recent dissent makes it unsafe to present “five scribes” as completely settled.
Some handwriting divisions correlate with textual and page-type divisions.
Those correlations are valuable but create confounding that must be controlled.
Bifolium-level physical evidence remains one of the strongest independent production axes.
Handwriting variation does not automatically determine author, topic, language or chronology.
The most productive question is the production system that generated the variation.

That is a strong conclusion even without knowing the final number of people.

What Does Not Survive as Established Knowledge

Exactly five historical individuals as an uncontested fact.
A direct equation of Currier A/B with different people.
A direct equation of each hand with one topic.
A direct equation of each hand with one language.
A proven chronological sequence from Hand 1 to Hand 5.
A proof that one writer produced the entire manuscript.
A proof that several scribes imply several authors.
A proof that handwriting alone identifies the manuscript’s institution or place of origin.

Those are exactly the places where future evidence can make a real difference.

A Better Scribe Analysis Has Four Independent Axes

1. Stroke and glyph formation

Analyse handwriting without using vocabulary as the primary signal.

2. Physical sheet

Map each proposed hand onto bifolia and gatherings.

3. Textual regime

Track Currier and other distributional differences separately.

4. Page role

Control for labels, running text, circular text, starred records and visual section.

Only after the four axes are mapped separately should they be recombined.

If the same clusters appear independently on several axes, confidence rises dramatically.

If they separate, the hidden workflow becomes more interesting.

What Would Count as a Real Scribe Breakthrough?

Imagine a blind palaeographic study in which several independent manuscript specialists receive cropped Voynich writing samples stripped of page identity and vocabulary context.

They consistently recover the same five clusters.

The clusters align strongly with bifolium boundaries.

Material or pen-stroke analysis independently supports the divisions.

That would strengthen the five-hand model substantially.

Or imagine the reverse.

When samples are stripped of topic and textual context, proposed hand clusters dissolve into continuous variation, while page order or production sequence predicts the changing stroke forms better than discrete identities.

That would strengthen a continuum model.

Either outcome would be progress because the alternatives make different predictions.

Primary School: Same Writer, Different Pen

Ask a child to write the same sentence three times.

Once carefully.

Once very quickly.

Once with a thick marker.

The three samples look different.

Same person.

Now ask two different children to imitate the same model alphabet.

The samples may look surprisingly similar.

Different people.

The learner immediately sees why handwriting identification is a pattern problem rather than a one-letter problem.

Lower Secondary: Separate Hand From Topic

Give one student two tasks: write a formal essay and write a shopping list.

The vocabulary and layout change.

The handwriting may change slightly too.

Then give two students the same shopping list to copy.

Now topic is constant and hand varies.

This simple exercise teaches confounding.

If both hand and topic change together, we cannot know which variable caused a statistical difference without another comparison.

Upper Secondary: Design the Crossing Case

Give students a table:

Writer A / Topic 1
Writer A / Topic 2
Writer B / Topic 1
Writer B / Topic 2

Ask which comparisons isolate writer and which isolate topic.

This is exactly the kind of design Voynich needs conceptually when hand, Currier regime and page type overlap.

Students learn that the strongest evidence often comes from the combinations that break an easy correlation.

JC and Adult Readers: Think in Latent Production Variables

At a higher level, the scribe problem is another latent-variable problem.

We observe glyph shape, vocabulary, physical sheet, layout and illustration.

We do not directly observe the production causes.

One hidden variable—scribe—could affect handwriting and orthography.

Another—source exemplar—could affect vocabulary.

Another—genre—could affect both vocabulary and layout.

Another—chronology—could affect handwriting gradually.

The goal is not to find the clustering scheme with the prettiest separation.

It is to identify which hidden production model best generates all observed axes together.

A Parent and Teacher Guide

The scribe debate is a useful way to teach a child not to mistake classification for cause.

Find the repeated handwriting differences.
Ask whether one writer could vary this much.
Ask whether several writers could imitate one shared script.
Separate handwriting from vocabulary.
Separate hand from topic and page type.
Use physical sheets as independent evidence.
Look hardest at crossing cases and exceptions.
Do not turn a hand count into an author count automatically.

The wider lesson is powerful:

two categories can line up perfectly and still represent different causes.

Reader Checklist: Before You Say “Five Scribes” or “One Scribe”

Is the claim based on handwriting or on vocabulary?
Which glyph features are supposed to be diagnostic?
Do those features occur outside the assigned hand?
Does the classification survive when page topic is hidden?
Does it survive when Currier regime is controlled?
Does it respect bifolium structure?
Could one writer’s speed, pen or chronology explain part of the variation?
Could several scribes share one trained script?
Are label pages being compared fairly with running prose?
Do independent palaeographers reproduce the groups?
Is the number of hands being converted into a number of authors?
What observation would favour a continuum over discrete clusters, or the reverse?

Frequently Asked Questions

How many scribes wrote the Voynich Manuscript?

An influential 2020 peer-reviewed study by Lisa Fagin Davis proposed five principal scribal hands. The exact number of historical writers remains debated, especially after a 2026 critique arguing for more continuous variation.

Does five hands mean five authors?

No. Scribal copying, intellectual authorship, compilation, illustration and patronage can be divided among different people.

Could one person have written the whole manuscript?

It remains a live dissenting hypothesis. One writer can vary over time and by task, but a one-hand model must explain systematic recurring differences that the five-hand model treats as scribal signatures.

Are Currier A and B the same as different scribes?

They correlate with scribal classifications in important places, but they should remain separate analytical axes. Currier A/B describes textual distribution, not a proven headcount.

Why do bifolia matter?

Because one bifolium is one physical sheet. Consistency of hand or textual regime across a sheet ties abstract classifications to actual manuscript production.

Does a multi-scribe manuscript imply a workshop?

It would make coordinated production more plausible, but “workshop” can describe many levels of organisation, from two collaborators to a more formal scribal environment.

What is the strongest current conclusion?

The manuscript contains systematic handwriting variation that interacts with textual and physical organisation. The exact number of people and the causal relationship among hand, Currier regime, source and workflow remain open research questions.

Research and Further Reading

The Final Idea

The question “How many scribes?” sounds like a request for a number.

It is really a request for a production model.

Five people?

One evolving person?

A smaller collaborative group whose styles overlap?

The final answer will matter because it changes what the Voynich Manuscript was socially.

A private notebook.

A copied compilation.

A workshop product.

A shared technical system.

That is why the best current position is neither to erase Davis’s five-hand breakthrough nor to ignore the new challenge to it.

Keep both models alive.

Then ask the manuscript to choose between them.

The number of hands is interesting. The workflow that made those hands look different is the deeper discovery.

Voynich | Everything eduKate Knows and Tested | The Scribe Problem: Five Hands, One Hand, and the Production Question