Voynich Manuscript: What We Actually Know

The most important thing we learned from spending a long time with the Voynich Manuscript was not how much we could explain.

It was how much explanation we had to throw away.

That sounds disappointing until you realise what kind of object we are dealing with.

The Voynich Manuscript is one of those rare things that punishes an intelligent reader for being intelligent too quickly. It gives you just enough familiarity to start building a story. Plants look like plants. Circular diagrams look astronomical. Human figures seem to bathe in green pools. Containers resemble jars. Stars appear beside blocks of writing. Repeated signs look temptingly like repeated words.

Then the mind does what minds are very good at doing.

It connects.

Plant → medicine.

Jar → preparation.

Stars → timing.

Bathing figure → anatomy or treatment.

Repeated sign → word.

Neighbouring pages → consecutive instructions.

And before long, a beautiful machine is humming in the reader’s head.

The problem is that a beautiful machine in our head is not yet the machine in the manuscript.

So this article begins somewhere less glamorous and much more useful.

What can we actually say after the attractive stories have been made to compete with the evidence?

Quite a lot.

Just not the thing everyone wants first.

Quick Read

One-sentence answer: the Voynich Manuscript is a genuine medieval or early fifteenth-century material object containing highly organised but still undeciphered writing and a complex illustrated programme; we can measure many regularities in it, but no accepted reading has yet identified its language, plaintext, complete function, author or place of production.

The manuscript is Beinecke MS 408 at Yale University’s Beinecke Rare Book and Manuscript Library.
Radiocarbon analysis of sampled parchment places the animal skin used for the manuscript in the early fifteenth century, commonly reported as 1404–1438 at 95% probability.
Materials analysis found inks and pigments compatible with historical manuscript production; this does not by itself date every stroke or prove when every layer was added.
The writing system remains unidentified and the text remains undeciphered.
The writing is not a simple random stream. Multiple statistical studies have found strong organisation, recurrence and distributional structure.
Those statistical properties do not prove that the visible word-like units are ordinary words in a natural language.
A 2025 study showed that a historically plausible hand-operable cipher can reproduce several Voynich-like statistical properties while preserving meaningful plaintext underneath, keeping sophisticated ciphertext hypotheses alive.
The familiar labels “herbal”, “astronomical”, “biological”, “cosmological”, “pharmaceutical” and “recipes” are useful descriptive aliases for visual regions, not decoded medieval chapter titles.
Our retained analysis is better described by overlapping physical, scribal, textual, visual and geometrical structures than by one perfectly clean set of semantic sections.
Comparisons with the Carrara Herbal, Masson 116, Sloane MS 4016, Egerton MS 747 and other medieval manuscripts are valuable because they show what manuscript makers could do. Resemblance does not establish Voynich identity, provenance, ownership or direct influence.
Several attractive total explanations become weaker when required to account for the entire object rather than a few spectacular pages.
The responsible position is not “nothing is known.” It is “many properties are known, while meaning remains unresolved.”

That distinction is the doorway into serious Voynich study.

The First Mistake Is Thinking “Unsolved” Means “Empty”

People often meet the manuscript through a dramatic sentence:

Nobody knows what it says.

That sentence is broadly true and intellectually dangerous.

It can make the manuscript sound like a blank wall.

It is not.

We can know the dimensions of an object without knowing why it was made. We can identify a pigment without identifying the painter. We can discover that two textual regions behave differently without knowing what either region means. We can recognise that certain visual components recur without assigning them a medieval name. We can reconstruct parts of an ownership trail without reconstructing the circumstances of manufacture.

This is a basic lesson in research:

Unknown meaning does not erase known structure.

The Voynich Manuscript is unusually rich in this middle zone.

We know enough to rule things out.

We know enough to distinguish strong observations from weak stories.

We know enough to compare it with other manuscripts.

We know enough to discover that many features are organised across more than one axis.

What we do not have is the final bridge from structure to semantics.

And that is exactly where a responsible account should leave the gap visible.

Start With the Object, Not the Legend

The Voynich Manuscript is held at Yale as Beinecke MS 408. The Beinecke Library provides high-resolution digital images of the complete surviving manuscript and a catalogue record describing its collation, foldouts, missing leaves and later binding.

The surviving codex is made of parchment. It contains roughly a hundred folios, depending on how one counts foldouts and surviving leaves, and numerous pages carry illustrations alongside the unidentified script. Several leaves fold out beyond ordinary page dimensions. Some leaves are missing. The binding is later than the manuscript’s earliest material layers.

Already, three different dates have appeared in that paragraph:

when the animal skin became parchment;
when the manuscript was written and illustrated;
when the surviving leaves were bound in their present form.

They need not be identical.

This matters because “the manuscript dates to X” is a compressed sentence. A codex is not manufactured in one atomic instant. Material preparation, ruling, drawing, writing, painting, assembly, correction, foliation, rebinding and later annotation can occur at different times.

When we ask when the Voynich Manuscript was “made”, we should therefore ask which layer of manufacture we mean.

The Parchment Gives Us a Powerful Boundary

Radiocarbon testing performed at the University of Arizona on parchment samples is commonly reported as placing the material between 1404 and 1438 with 95% probability.

This is one of the strongest anchors in the entire problem.

It makes certain chronologies difficult or impossible.

But notice the wording:

the parchment

Radiocarbon dating does not translate the writing. It does not identify the author. It does not tell us whether the text was encoded. It does not prove a city. It does not automatically tell us how long prepared skins waited before being used. It does not transform every stylistic resemblance into a dated historical connection.

A date can be strong evidence and still have a bounded job.

This is a recurring Voynich lesson.

Ink and Pigment Tell Us About Materials, Not Meaning

McCrone Associates examined and sampled the manuscript in 2009. Their published report describes brownish-black writing consistent with quill use and identifies iron-gall inks in sampled areas, alongside historically familiar pigment and binder materials.

The report is useful because it gives us a physical vocabulary.

Ink.

Binder.

Pigment.

Stroke.

Layer.

It also teaches restraint. A material that is consistent with fifteenth-century manuscript production is not a signature saying “made in Padua by person X on Tuesday”. Material compatibility narrows the world. It does not collapse the world into one workshop.

Many historical materials travelled widely. Recipes for inks circulated. Parchment production was not unique to one city. Pigments could be traded. Scribes moved. Books moved. Exemplars moved.

Good evidence often reduces possibility without eliminating it.

The Provenance Is Better After the Seventeenth Century Than Before It

The manuscript’s later history becomes substantially clearer in Central Europe.

Yale’s current overview links it to the court of Rudolf II and then to Jacobus Horčický de Tepenec, Georg Baresch, Johannes Marcus Marci and Athanasius Kircher. Marci sent the book to Kircher in the seventeenth century with a letter describing earlier beliefs about its ownership and supposed authorship.

Then comes an enormous historical gap.

In 1912 the antiquarian bookseller Wilfrid M. Voynich acquired it from the Jesuit collection at Villa Mondragone near Frascati. It later passed through Ethel Voynich, Anne Nill and bookseller H. P. Kraus before Kraus gave it to Yale’s Beinecke Library in 1969.

The famous name therefore belongs to a twentieth-century bookseller, not to the unknown medieval maker.

That is worth remembering.

The object existed for centuries before it became “the Voynich Manuscript”.

The name is part of its modern biography, not its original identity.

A Catalogue Description Is Not a Decipherment

The Beinecke catalogue divides the illustrations into six familiar regions that generations of writers have called botanical, astronomical or astrological, biological, cosmological, pharmaceutical and recipe-like.

These labels are convenient.

Convenience is not the same as proof.

Calling a page “herbal” because it prominently depicts a plant is a reasonable visual description. It does not prove that the accompanying text is a herbal entry in the ordinary medieval sense.

Calling a circular diagram “astrological” may be useful when zodiac signs are visible. It does not prove that every nearby token is an astrological instruction.

Calling pages with containers “pharmaceutical” does not prove that the containers hold medicines, nor that plant fragments beside them are ingredients, nor that the text gives recipes.

The distinction seems small until you see how much theory can be smuggled inside a noun.

Description can quietly become interpretation when the label is repeated often enough.

So throughout this series we will use the conventional labels when they help orientation, but we will keep quotation marks around their certainty.

The Manuscript Does Not Obey One Perfect Map

One of the strongest lessons from our retained work is that the manuscript becomes harder to understand when we insist that every kind of boundary must be the same boundary.

A physical boundary is not necessarily a textual boundary.

A change in illustration type is not necessarily a change in writer.

A change in recurring vocabulary is not automatically a change in subject.

A foldout is a physical design decision, but it may support several possible semantic jobs.

Once we kept those layers separate, a more durable picture emerged:

some structures follow the physical construction of leaves and bifolios;
some follow writing behaviour;
some follow page morphology;
some follow distributions of recurring textual forms;
some follow diagram geometry;
some overlap strongly;
and some cut across one another.

This is less satisfying than a single master key.

It is more faithful to the object.

A city has streets, sewer lines, electoral boundaries, property lots, postal districts and transit zones. All are real. None is the unique map of the city.

The Voynich Manuscript may need to be approached in the same spirit.

Different maps can capture different truths without one map being “the meaning”.

The Writing Is Organised

The writing is one of the reasons the manuscript refuses to collapse into a casual doodle book.

Even before meaning, there is behaviour.

Signs recur.

Sign combinations recur.

Some forms prefer particular positions.

Some “word”-like units are locally common and globally uneven.

Textual regions differ from one another.

Line and paragraph positions matter.

The traditional Currier A and Currier B distinction—originally based on differences in character and word frequencies—remains a useful descriptive observation even though we should be careful about turning those distributions directly into two identified languages, dialects or subjects.

Statistical studies published in PLOS ONE in 2013 found non-random, language-compatible organisation at several scales. Montemurro and Zanette reported long-range patterns in the distribution of word-like units. Amancio and colleagues found that network and frequency measures distinguished the manuscript from shuffled versions and were in important respects compatible with natural-language texts.

Those are real findings.

Then comes the dangerous jump:

Therefore Voynichese is an ordinary natural language written in an unknown alphabet.

That conclusion is stronger than the measurements warrant.

A text can inherit statistical structure from plaintext after transformation. An encoding system can impose structure of its own. A generative procedure can create regularity. Spaces may not correspond exactly to lexical word boundaries. A visible glyph may represent more or less than a phoneme. A repeated token may not behave like a modern dictionary word.

Pattern narrows the problem.

Pattern does not finish it.

The 2025 Cipher Result Matters Because It Restores an Alternative

For years, one popular line of argument went roughly like this:

The manuscript looks statistically language-like, therefore a complex cipher is increasingly implausible.

A 2025 Cryptologia paper by Michael Greshko complicated that inference. The paper introduced a verbose homophonic substitution system called the Naibbe cipher and showed that it could be executed by hand with historically plausible materials while producing ciphertexts from Latin and Italian that reproduced several unusual statistical properties associated with the Voynich text.

This did not decipher the manuscript.

It did something more modest and more scientifically useful.

It demonstrated existence.

A historically plausible cipher family can generate a bundle of Voynich-like statistics without destroying recoverable meaning.

Therefore the statistical argument cannot simply be:

language-like statistics → not ciphertext.

The hypothesis space remains wider.

This is exactly what good research should do. It should not only accumulate support for favourite explanations. It should revive alternatives when new evidence shows that we closed them too early.

Five Scribes? One Evolving Hand? The Right Answer Is Still “Contested”

In 2020 palaeographer Lisa Fagin Davis published an influential digital palaeographic study identifying five scribal hands in the manuscript. That model has been widely adopted in later discussion because handwriting variation can help reconstruct production: who wrote where, whether collaborators alternated, and how textual regimes relate to physical gatherings.

It is an important hypothesis.

It is no longer safe to present it to readers as uncontested fact.

In 2026 Torsten Timm published a detailed critique arguing that the diagnostic variation may form a continuum rather than five discrete hands and that some of the apparent categories may partly track already known textual or layout distinctions.

The correct public sentence is therefore not:

Five people wrote the Voynich Manuscript.

It is closer to:

A prominent palaeographic model identifies five hands; that model has supporting work and remains influential, but recent criticism disputes whether the variation requires five discrete scribes.

That sentence is longer.

It is also more useful.

Research literacy often means resisting the urge to shorten a live argument into a dead fact.

The Pictures Are Evidence Before They Are Explanations

The illustrations are where human pattern-recognition becomes most dangerous.

We are built to recognise.

A root resembles an animal.

A leaf resembles a known species.

A tube resembles plumbing.

A circular arrangement resembles a cosmological diagram.

A woman standing in liquid resembles bathing, medicine, reproduction, ritual or metaphor depending on what the viewer already expects.

The image itself cannot stop us from importing our own categories.

So the first useful questions are deliberately boring:

What lines are actually present?
Which components recur?
Where do they attach?
Which parts vary?
What is labelled?
What is adjacent?
What is repeated on another page?
Which page types share the same motif?
Which apparent resemblance disappears when scale or orientation changes?

Those questions postpone meaning.

Postponement is not cowardice.

It is how you give the object a chance to surprise you.

Why the Plants Are Harder Than “Identify the Species”

It is tempting to imagine a botanical identification contest.

Find the modern species.

Find its medieval name.

Translate the label.

Open the manuscript.

Real medieval image traditions make this much harder.

Images can be copied from exemplars rather than from living plants. Parts can be emphasised because they are diagnostically useful. Images can degrade or transform across copying generations. A scribe or painter may combine inherited conventions with observation. Scale can be sacrificed. Roots can become visually elaborate. Different artists can render the same named substance differently. A plant name can refer to a substance, drug or medicinal material rather than map neatly onto modern taxonomic species concepts.

And the Voynich plants sometimes look as though recognisable botanical features have been reorganised, exaggerated or fused.

That creates at least three possibilities:

we are seeing poor or transformed botanical representation;
we are seeing deliberate information-rich diagrams rather than naturalistic portraits;
we are seeing objects whose intended category is not exactly the modern botanical category we are imposing.

There may be other possibilities as well.

What we should not do is force a species match and then use the forced match to “confirm” the theory that generated it.

The Carrara Herbal Was a Powerful Comparison—and a Dangerous Temptation

The Carrara Herbal, British Library Egerton MS 2020, matters because it sits close in time to the Voynich parchment range and comes from a sophisticated northern Italian medical and artistic environment. The British Library dates it approximately 1390–1404 and describes it as an Italian translation of a work associated with Serapion the Younger, richly illustrated with plants and connected to the Carraresi court of Padua.

For us it became valuable for a specific reason.

It demonstrates capability.

People in that historical world could make a learned vernacular medical book in which image, substance identity and explanatory text operated together at high visual resolution.

That is useful when asking whether some Voynich distortions reflect artistic inability. If naturalistic and diagnostically careful plant depiction existed nearby in time and region, then “they simply could not draw plants” becomes a weaker universal explanation.

But capability is not identity.

The Carrara Herbal does not make the Voynich Manuscript a Carrara book.

It does not prove Padua.

It does not prove a medical purpose.

It does not prove shared ownership.

It does not prove that the two manuscripts share an exemplar.

This was one of the most important corrections in our work:

A good historical comparison can make a function plausible without making a provenance true.

Masson 116 Teaches a Different Lesson

Paris, École nationale supérieure des Beaux-Arts MS Masson 116 is another extraordinary control object. Biblissima describes it as a Tractatus de herbis image collection containing natural-history illustrations with simple Latin captions and no continuous text.

That matters because modern readers often assume that if an image is useful, explanatory prose must sit beside it.

Medieval manuscript practice is more diverse.

An image can participate in recognition.

It can preserve a visual tradition.

It can act as an index or retrieval surface.

It can be copied into another codex and recontextualised.

And indeed the British Library catalogue identifies Sloane MS 4016 as an Italian herbal in the North Italian group and as a copy of Masson 116, with full-page plant miniatures, captions and occasional people or animals.

Here is a known transmission relationship.

That gives us something precious: a control for what visual inheritance can look like when a relationship is historically supported.

It also makes casual Voynich comparison harder.

If we want to argue that Voynich imagery descends from another tradition, we should expect more than a vaguely similar root or leaf. Known copying traditions give us a standard against which to judge the strength of alleged inheritance.

Egerton MS 747 Shows How Much Can Live in One Medical Codex

British Library Egerton MS 747 is older than the Voynich Manuscript, but it is useful for a different reason.

Its catalogue describes a codex that includes the Tractatus de herbis, additional plant images, a forecasting fragment with lunar calendar, the Antidotarium Nicolai, material on doses, ingredient substitutions, weights and measures, synonyms for medicinal substances and other additions.

This is a warning against modern disciplinary boxes.

Herbal knowledge.

Remedy preparation.

Measurement.

Substitution.

Synonym management.

Lunar or forecasting material.

These could coexist in one manuscript world.

Therefore the mere coexistence of plants, stars and containers in the Voynich Manuscript is not inherently absurd as a practical knowledge combination.

But once again, possibility is not identity.

Egerton 747 shows that such combinations existed.

It does not tell us what Voynich combinations mean.

One Attractive Total Story Did Not Survive

At one stage it was tempting to imagine a clean pipeline running through the manuscript:

plant identification → preparation → bodily application → celestial timing → compound recipe.

It is a beautiful idea.

It also has historical plausibility in the abstract because medieval medical books could connect substances, preparations, doses, calendars and therapeutic knowledge.

The problem appears when the whole manuscript is asked to cooperate.

The boundaries do not all line up.

The text does not provide a decoded directional sequence.

Visual recurrence does not yet establish function.

Current folio order cannot safely be treated as guaranteed original operational order.

Some apparently convenient modules are inherited descriptive categories rather than demonstrated medieval categories.

So the integrated pipeline had to be demoted from explanation to possibility.

That demotion is progress.

A theory becomes scientifically valuable not when it is emotionally satisfying, but when it survives the parts of the object that do not want to fit.

Another Attractive Story: “The Jars Tell Us the Process”

The container-like drawings are seductive because containers are functional objects.

If a drawing looks like a vessel, perhaps the vessel type tells us what happened to the plant beside it.

Powder.

Infusion.

Distillation.

Storage.

Dosage.

Again, the hypothesis is not silly.

What failed was the leap from resemblance to a reliable universal code.

A container can indicate many things. It may be a vessel. It may be stylised. It may function decoratively. It may classify. It may distinguish groups. It may be copied from an exemplar. It may represent something we would not call a jar at all.

Until visual categories make repeatable predictions elsewhere in the manuscript, “jar type = process” is not a decoding rule.

This is a useful general lesson for students:

A plausible interpretation becomes stronger when it predicts something you did not use to invent it.

The Zodiac Does Not Automatically Control the Rest of the Book

Zodiac imagery is among the manuscript’s most recognisable material.

That recognition invites a powerful inference:

If zodiac signs appear, perhaps celestial timing governs medicine, plants, bathing or reproduction throughout the manuscript.

Historically, such relationships are conceivable. Medieval and early modern medicine often lived beside astrological thinking, calendars and theories of timing.

Yet the Voynich Manuscript still has to demonstrate its own relationship.

A zodiac page proves that a zodiac-like visual system is present.

It does not by itself prove that every plant page is scheduled by it.

It does not prove that every nude figure is a patient.

It does not prove a medical calendar.

The link must be earned through internal evidence.

This is where many interpretations become circular:

The manuscript is medical because the zodiac is for medical timing, and the zodiac is for medical timing because the manuscript is medical.

Two uncertain claims cannot become certain by leaning against each other.

Current Page Order Is Not a Free Gift

Modern readers naturally read books in the order they are bound.

Historical manuscripts complicate this habit.

The Voynich codex contains foldouts, missing leaves, difficult collation and a later binding. The Beinecke catalogue itself notes inconsistencies and losses.

Therefore page adjacency today is evidence of current physical adjacency.

It is not automatic proof of original composition sequence.

This destroys a surprising number of easy stories.

If Page A looks like an ingredient and Page B looks like a process, we cannot simply declare A→B an instruction sequence because the two pages face each other today.

We need to know whether the leaves belong together physically, whether the current binding preserves original order, and whether other independent features support the same relationship.

Codicology—the study of books as physical constructions—is therefore not a side subject in Voynich research.

It is one of the brakes that prevents semantic imagination from outrunning the object.

What Survives After the Grand Stories Are Removed?

Something much more interesting than “nothing”.

The manuscript contains real organisation at multiple scales.

Physical construction matters.
Textual distributions matter.
Line and paragraph positions matter.
Recurring page morphologies matter.
Some visual components recur.
Some textual regimes cluster.
Some relationships are strong within bifolios or gatherings while others cut across them.
Different descriptive partitions overlap without becoming identical.

That suggests a manufactured information object, not a pile of independent curiosities.

But “manufactured information object” is deliberately broad.

A prayer book is one.

A medical compendium is one.

An encoded notebook is one.

A teaching manual is one.

A catalogue is one.

A deliberately generated pseudo-text can be one.

The surviving organisation tells us that there is a design problem worth solving.

It does not yet tell us which historical job the design served.

The Most Useful Word in Voynich Research Is “Because”

Anyone can propose an identification.

This plant is aloe.

This diagram is a city.

This token means water.

This jar indicates distillation.

This page comes from Padua.

This script encodes Latin.

The serious question is:

Because what?

Because three shapes look similar?

Because the proposed reading works on one label?

Because the theory predicts unseen cases?

Because a known manuscript lineage shows the same transformation?

Because material evidence narrows the historical region?

Because a pattern persists after the folio order is changed?

Because an independent source documents ownership?

Not all becauses are equal.

That is why the manuscript is such a good classroom for reasoning.

Resemblance Is Cheap

Put enough medieval manuscripts beside one another and you will find similarities.

Plants.

Stars.

Circles.

Containers.

Nude bodies.

Waves.

Rays.

Heraldic forms.

Architectural motifs.

A resemblance can be a clue.

It can also be the result of shared visual culture, common geometry, functional convergence, generic iconography, copying several generations apart, restoration, or our own desire to see a match.

This is why historical controls matter.

When Sloane MS 4016 is catalogued as a copy of Masson 116, we can inspect what a supported relationship looks like.

When the Carrara Herbal is documented within the Carraresi court environment, we can see what provenance evidence looks like.

When Egerton 747 demonstrably contains herbal, calendrical and pharmaceutical materials in one codex, we can see what actual functional coexistence looks like.

The more we know about known manuscripts, the higher the standard becomes for claims about the unknown one.

What We Do Not Know

A mature knowledge article should have an explicit empty shelf.

Here is ours.

We do not have an accepted plaintext translation.
We do not have an accepted identification of the underlying language, if there is one.
We do not have an accepted cipher or encoding scheme.
We do not know whether spaces correspond exactly to words.
We do not know the intended names of most illustrated objects.
We do not know the complete original folio order with certainty.
We do not know the exact place of manufacture.
We do not know the author or authors.
We do not know with finality how many scribes participated.
We do not know whether writer and illustrator were the same people.
We do not know the complete intended function of the manuscript.
We do not know whether it was intended for one owner, a workshop, a school, a professional network or some other receiver.
We do not know whether the imagery is primarily representational, mnemonic, classificatory, operational, symbolic or some combination.
We do not know whether the text is plaintext, ciphertext, encoded notation, constructed language, constrained generation or another system.
We do not have evidence that proves Carrara ownership, Paduan production or direct descent from any particular comparator manuscript.

That list is not a confession of failure.

It is the edge of the current map.

The Difference Between a Mystery and a Research Object

A mystery asks:

What is the answer?

A research object asks a larger set of questions:

What exactly was observed?
How stable is that observation?
What alternative explanations produce the same observation?
Which historical comparisons are valid?
Which comparisons are only superficially similar?
What would a hypothesis predict beyond the page that inspired it?
What evidence would make us abandon the hypothesis?
Which uncertainties are material rather than annoying?
What can another researcher reproduce without inheriting our assumptions?

Once those questions enter, the emotional shape of Voynich research changes.

You stop trying to win.

You start trying to be wrong in useful ways.

That is a much stronger form of curiosity.

Failed Ideas Are Part of What We Know

This deserves emphasis because education often gives students the wrong picture of research.

School exercises are usually designed so that a correct answer already exists.

Research is often the opposite.

You may spend days building an explanation and discover that it collapses on a control case.

You may find that a visual match was caused by selective attention.

You may discover that the page order you assumed was not secure.

You may prove that a statistical signature can be produced by a mechanism you had ruled out.

You may learn that two variables correlate because both track a third feature.

That is not wasted work if the failure shrinks the space of bad explanations.

In the Voynich problem, a failed total theory can be more valuable than an untested successful-sounding theory.

The failed theory leaves a boundary.

It tells the next reader where not to walk casually.

A Better Way to Read One Voynich Page

Suppose you open a folio and see a large plant, a block of script and several repeated forms.

Do not begin with:

What plant is this?

Begin with layers.

Layer 1: Physical

Which leaf is this? Is it part of a bifolium? Is the page intact? Is it near a missing leaf? Is it a foldout? Are there stains, folds, repairs or later marks?

Layer 2: Layout

Where does text begin? Does it wrap around the drawing? Are labels isolated? Are there paragraph markers? Does the illustration interrupt text or appear to have been planned first?

Layer 3: Visual

What shapes are actually drawn? Which components recur elsewhere? Which details are unusual? Which apparent identities depend on imagination rather than line evidence?

Layer 4: Textual

Which word-like forms recur? Which are local? Which occur at line boundaries? Does the page resemble one known statistical regime more than another? Are labels drawn from the same distribution as running text?

Layer 5: Comparison

Which known manuscripts offer a relevant comparison? Is the comparison about drawing style, page function, copying process, medical content, book manufacture or something else? Do not use one type of similarity to prove another.

Layer 6: Interpretation

Only now propose meanings—and keep several alive.

This sequence feels slower.

It is usually faster than spending a month defending the first thing the picture reminded you of.

What Would Count as a Real Breakthrough?

The internet has announced the solution to the Voynich Manuscript many times.

A real breakthrough would need to behave differently from a clever reading of a few labels.

For a proposed textual solution, we would want something like:

a clear and reproducible mapping from manuscript signs to the proposed system;
rules established before cherry-picking convenient passages;
successful reading of substantial unseen text;
consistency across different textual regions or a principled explanation for differences;
grammatical or structural coherence in the proposed underlying language;
semantic relationships to images that are predictive rather than retrofitted;
historical plausibility for the writing or encoding system;
independent replication by researchers who did not build the theory;
clear accounting for failures and exceptions.

For a proposed provenance, we would want documentary, codicological, palaeographic, material or transmission evidence capable of distinguishing that place or workshop from plausible alternatives.

For a proposed function, we would want the explanation to predict how multiple page types, textual behaviours and physical structures work together—not merely rename familiar illustrations.

The standard should be high because the number of possible stories is enormous.

Primary School: Learn the Difference Between Seeing and Guessing

The Voynich Manuscript can be useful even for a young learner if we reduce the problem to one simple distinction.

Show a child one strange plant drawing.

Ask:

What can you see?

Then:

What do you think it might be?

Two different columns.

In the first column:

three large leaves;
a branching root;
blue paint;
a line of writing beside the stem.

In the second:

maybe a medicinal plant;
maybe a composite;
maybe a symbolic drawing.

That child has just learned one of the deepest habits in science and history:

observation is not the same thing as interpretation.

Lower Secondary: Build Competing Explanations

By Secondary 1 and Secondary 2, the learner can take one recurring feature and generate alternatives.

Suppose a star appears beside a paragraph.

Possible functions might include:

bullet marker;
topic marker;
astronomical meaning;
decorative separator;
retrieval cue;
entry delimiter;
something else.

Now ask what each hypothesis predicts.

If it is an entry delimiter, does it appear consistently at entry boundaries?

If it is astronomical, do nearby textual or visual behaviours differ from non-starred passages?

If it is decorative, is its placement weakly constrained?

The student is no longer guessing.

They are turning guesses into testable expectations.

Upper Secondary: Learn How Evidence Gets Overstated

By Secondary 3 and Secondary 4, students can examine the language of claims.

Compare:

The manuscript contains zodiac imagery.

with:

The manuscript is an astrological medical manual.

The first is close to direct observation.

The second bundles several interpretive steps.

Or compare:

The parchment dates to the early fifteenth century.

with:

The manuscript was written in 1420.

One is supported by a dating interval for sampled material.

The other invents precision.

This is excellent training for essays, science reporting, history and everyday media literacy.

JC and Adult Readers: Separate Model Fit From Model Identity

At a higher level, the Voynich problem becomes a lesson in underdetermination.

More than one model can explain the same evidence.

Language-like statistics can arise from natural language.

They can also survive some forms of encoding.

A repeated visual component can represent an object.

It can also be a structural motif, copied convention, classification mark or drawing habit.

A manuscript that combines plants, stars and human bodies can be medical.

It can also combine those domains for a different purpose.

The research task is not merely to find a model that fits.

It is to find observations that discriminate between models.

That is a much harder question.

It is also the question that turns mystery-hunting into scholarship.

Why “Everything We Know” Must Include Uncertainty

There is a cultural pressure to make knowledge look cleaner than it is.

Dates become single years.

Hypotheses become facts.

Disputed readings become captions.

Convenient labels become historical categories.

Weak provenance becomes “from”.

Statistical compatibility becomes “is”.

The result is a prettier article and a worse map of reality.

For the Voynich Manuscript, uncertainty is not empty space around knowledge.

It is part of the knowledge.

Knowing that a five-scribe model is contested is better than memorising “five scribes”.

Knowing that parchment dating constrains material age but not exact inscription date is better than memorising “written in 1420”.

Knowing that text is highly organised but not translated is better than saying either “gibberish” or “natural language” as if the matter were closed.

Precision sometimes means saying less.

A Parent and Teacher Guide: Use the Voynich Manuscript to Teach Epistemic Discipline

You do not need to teach children cryptography to make the manuscript educationally useful.

Use it to teach claim control.

Ask the learner to separate what they can point to from what they infer.
Ask for two alternative explanations before accepting a favourite.
Ask what evidence would make the favourite explanation weaker.
Ask whether a comparison demonstrates identity or merely possibility.
Ask whether the date belongs to the material, the writing, the binding or the ownership record.
Ask whether a repeated visual form has been named too early.
Ask whether the same interpretation predicts anything on a page the learner has not yet examined.
When a hypothesis fails, preserve what was learned from the failure instead of quietly changing the story.

This is not only Voynich training.

It is training for science.

History.

News.

Statistics.

Research.

And the ordinary human problem of being certain before reality has earned our certainty.

Reader Checklist: Before You Believe a Voynich Claim

Is the claim an observation, inference, historical fact or proposed interpretation?
Does the source distinguish parchment date from writing date?
Does it treat the conventional “sections” as descriptions or decoded functions?
Does a plant identification depend on one spectacular resemblance?
Does a proposed translation work on substantial unseen material?
Does the method explain failures as openly as successes?
Does the argument assume current folio order is original?
Does it use a neighbouring manuscript as proof of origin rather than comparison?
Does a statistical pattern actually distinguish the proposed mechanism from alternatives?
Does the writer acknowledge current disputes, including the scribe question?
Does the theory make new predictions?
Can somebody else reproduce the result without adopting the theory first?

If the answer to several of those questions is no, the claim may still be interesting.

It should not yet be called solved.

Frequently Asked Questions

Has the Voynich Manuscript been decoded?

No decipherment has achieved broad scholarly acceptance. Many proposed solutions have been announced, but the manuscript’s text remains officially and academically unresolved.

Is it definitely a natural language?

No. Its text has many statistical properties compatible with natural language, but compatibility is not identity. Ciphertext and other structured-generation hypotheses remain viable.

Is it definitely a cipher?

No. The unidentified script may encode language in several possible ways, but no accepted cipher system has been recovered.

Is it a medical book?

Medical and herbal interpretations are historically plausible because of the imagery and comparable manuscript traditions. They remain interpretations rather than established function.

Was it made in Padua?

Not established. Paduan and northern Italian manuscripts provide unusually useful comparisons, but comparison does not prove provenance.

Did the Carrara family own it?

There is no documentary evidence currently sufficient to establish literal Carrara ownership. The Carrara Herbal is a comparison object, not an ownership certificate for the Voynich Manuscript.

Are there five scribes?

Lisa Fagin Davis’s influential 2020 study proposed five hands. A 2026 critique argues that the variation may be continuous and attributable more parsimoniously to one evolving hand. The issue should currently be taught as contested.

Are the six sections real?

The visual groupings are useful for navigation, but their familiar names are modern descriptive labels. They should not automatically be treated as six proven medieval semantic programmes.

Could it be a hoax?

A deliberate meaningless production remains part of the broad hypothesis space, although the manuscript’s material investment and complex textual organisation must be explained by any serious version of that hypothesis. “Hoax” is not an explanation until a mechanism accounts for the observed regularities.

What is the strongest thing we can say?

The strongest safe statement is that the Voynich Manuscript is a deliberately manufactured, highly structured illustrated codex whose physical, textual and visual regularities are measurable, but whose semantic system remains unresolved.

Research and Further Reading

The Final Idea

The Voynich Manuscript is often introduced as a book that has defeated us.

There is another way to see it.

It is a book that keeps forcing us to improve the quality of the question.

At first we ask:

What does this mean?

Then the manuscript pushes back.

So we learn to ask:

What did I actually observe?
What am I assuming?
What else could produce this pattern?
What would make my favourite explanation fail?
What remains after it fails?

Those questions reach far beyond one mysterious medieval book.

They are how a curious mind becomes a reliable one.

Continue the Voynich Map

Voynich is the ten-part master synthesis. Use it to return from this evidence floor to the complete knowledge map.

The Manuscript Before the Mystery — move from claim control to the physical codex.
What a Real Voynich Decipherment Must Survive — see how the evidence floor becomes a solution standard.

Deep bridge: How Scientific Research Works explains why bounded claims, alternative explanations, negative results and replication matter here.

Use the next link to narrow the question, not to replace evidence.

Voynich | Everything eduKate Knows and Tested | What We Actually Know