Every surviving manuscript has a visible beginning.
The first page.
The first stroke.
The first drawing.
But the intellectual beginning may have happened somewhere else.
On another sheet.
In another book.
Inside a notebook that no longer survives.
In a workshop where one person drew and another copied text.
In the memory of a trained practitioner.
Or inside a plaintext, abbreviation system or cipher routine that was transformed while the surviving Voynich page was being made.
This is the exemplar problem.
Was the Voynich Manuscript the place where its information was first composed—or is the surviving codex already a copy, compilation, transformation or final-stage rendering of something earlier?
The question matters because several of the manuscript’s strongest physical observations can be explained in more than one way.
Text fits around drawings.
Pages can look carefully planned.
Obvious corrections are surprisingly sparse.
Several scribal hands or hand-like varieties may participate.
Visual regimes change.
Textual regimes change.
Some plant forms look as though they belong to inherited manuscript traditions rather than direct observation from life.
None of those facts proves a lost exemplar.
Together, they make the question unavoidable.
Quick Read
- An exemplar is a source used in producing another manuscript: a complete book, loose sheets, draft, model page, image source, text source or other copying reference.
- No complete direct exemplar of the Voynich Manuscript has been securely identified.
- That absence does not mean the manuscript was composed directly on the surviving parchment.
- Voynich page planning is compatible with a prepared exemplar because drawings generally precede the principal text and writing often fits around existing visual geometry.
- The apparent scarcity of conspicuous corrections is compatible with careful copying from a prepared source—but also with fluent composition, abbreviation, ciphering or constrained generation.
- Image and text need not come from the same source. A maker could copy plant or diagram traditions while composing, translating, abbreviating or enciphering new text.
- One manuscript can compile several exemplars or source traditions.
- Currier A/B, scribal variation and section boundaries may reflect different sources, different people, different work sessions, different document roles—or combinations of these.
- Comparator manuscripts show historical possibilities for copying, compilation and image transmission. They do not become Voynich sources merely because they resemble it.
- A credible exemplar hypothesis must explain what was copied, by whom, at what stage, and what measurable traces the copying process should leave.
What Is an Exemplar?
In manuscript studies, an exemplar is not necessarily a beautiful complete book sitting beside the scribe.
It can be any source used to produce another witness.
- a complete earlier manuscript;
- a damaged book;
- loose copied leaves;
- a rough draft;
- a model page;
- a text-only source;
- an image-only source;
- several different source books;
- a set of notes;
- a translation draft;
- a cipher plaintext or working table;
- oral dictation combined with a visual model.
This wider definition matters because “Did Voynich have an exemplar?” is not one yes-or-no question.
It contains several:
- Was the text copied?
- Were the pictures copied?
- Was the layout copied?
- Were only some sections copied?
- Were several sources compiled?
- Was readable source text transformed while being copied?
- Was a preliminary draft produced before the final parchment?
A manuscript can answer these questions differently on different folios.
Autograph, Copy and Compilation Are Different Production Models
Autograph or direct composition
The creator develops the content while making the surviving manuscript.
We might expect revisions, false starts, reorganisations or signs of thought changing on the page—although an experienced author can also compose cleanly.
Copy
The surviving manuscript reproduces an earlier source substantially enough that wording, sequence or imagery already existed.
Now the most revealing errors may be copying errors rather than authorial revisions.
Compilation
The maker combines material from several sources.
One source may supply herbal images.
Another may supply astronomical diagrams.
A third may supply practical records.
The compiler may translate, abbreviate, rearrange or rewrite everything into one visible system.
These models can overlap.
A compiler can copy some material directly and compose transitions.
A scribe can copy pictures but rewrite text.
A cipher clerk can transform a plaintext exemplar rather than reproduce it visually.
The production model therefore needs more precision than the single word “copy”.
Why Page Planning Makes the Exemplar Question Serious
On many illustrated Voynich pages, principal drawing outlines appear to precede the surrounding text.
Writing stops at plant stems.
It resumes in available spaces.
Short tokens occupy narrow gaps.
Longer lines use wider open regions.
This can be explained by direct composition around a finished image.
But it can also be explained by a copied or prepared layout.
Imagine that a rough paper draft once existed.
On the draft, the maker already solved the awkward geometry between plant and prose.
The final parchment is then copied from that plan.
The surviving page would look beautifully coordinated even though the real compositional struggle happened on a lost sheet.
A polished surviving page cannot tell us how messy the invisible draft was.
This is why When Text Meets Image makes the exemplar question stronger without answering it.
The Correction Problem Creates the Same Ambiguity
Voynich appears to contain comparatively few conspicuous corrections.
A clean copy from a prepared exemplar can explain that.
But so can:
- high scribal fluency;
- a forgiving orthography;
- a well-practised cipher routine;
- rule-based generation;
- careful preliminary drafting;
- erasures that are physically difficult to see.
Therefore the correction pattern is a clue to production method, not a certificate of copying.
What Would a Copied Text Leave Behind?
Copying introduces characteristic failure modes.
Eye-skip
A scribe looks away from the source, writes, looks back and resumes at the wrong similar sequence.
This can omit or duplicate stretches of text.
Dittography
A word, phrase or sequence is copied twice.
Haplography
Repeated or similar material causes one occurrence to disappear.
Misreading
An exemplar sign is mistaken for a similar sign.
Reordering
Source units are copied in the wrong sequence.
The difficulty is that without the source we cannot point to an odd Voynich passage and confidently call it a copying error.
We can only ask whether the statistical pattern resembles known copying behaviour strongly enough to generate a test.
The Exact Repetition article is particularly relevant because phrase-level repetition is much rarer than word-level repetition.
Could the Pictures Have an Exemplar While the Text Does Not?
Yes.
This is one of the most important models to keep separate.
A maker can copy a plant image from an inherited herbal tradition and then write new text beside it.
The new text could be:
- a translation;
- a commentary;
- a medical recipe;
- a mnemonic note;
- a ciphered version of source text;
- a newly composed description;
- generated filler;
- another specialised notation.
This matters because manuscript images often travel independently of wording.
A plant drawing can survive through repeated copying even when the text around it changes language, order or function.
Therefore an iconographic match does not automatically identify the textual source.
Could the Text Have an Exemplar While the Pictures Are New?
Also yes.
A scribe can copy a written source while an illustrator creates new diagrams or adapts older ones.
This is especially plausible in practical knowledge traditions where a compiler reorganises material for a new audience.
Now the text may preserve an older sequence while the image system belongs to the new manuscript.
That would make visual resemblance to another codex a poor guide to textual descent.
One Exemplar or Several?
The familiar six-section classification already suggests heterogeneity.
Plants.
Zodiac diagrams.
Human figures and pools.
Cosmological-looking diagrams.
Vessels and fragments.
Starred entries.
A compiler could have drawn from several source traditions.
That possibility is historically ordinary.
Medieval manuscripts are frequently composite intellectual objects even when physically produced as one codex.
But visual difference alone does not prove multiple exemplars.
One author can write on several subjects.
One workshop can use several page formats.
One source can itself contain several sections.
The new Six Sections That May Not Be Six Sections article explains why visual classification should not be promoted automatically into source history.
Currier A and B Can Be Read as Source Clues—but Need Not Be
Currier A and B are strong distributional regimes in the text.
Could they reflect different exemplars?
Possibly.
For example:
- one source tradition may use one notation convention;
- another source may use another;
- a compiler may preserve differences;
- different scribes may copy different sources.
But Currier difference can also arise without different source books.
- different scribes;
- different time periods;
- different document roles;
- different subjects;
- different encoding states;
- different stages in a writing system’s development.
The hidden variable remains open.
See Currier A and B.
Scribal Hands Do Not Equal Source Manuscripts
This distinction is easy to lose.
Five scribes can copy one exemplar.
One scribe can copy five exemplars.
A workshop can divide labour by section.
A single writer can change ductus across time.
Therefore palaeographic grouping and exemplar grouping are different inference problems.
If the same hand copies two sharply different textual regimes, that may suggest the difference belongs to source or content rather than writer.
If two hands produce nearly identical textual behaviour, that may suggest a shared underlying convention or source.
Those interactions are informative precisely because the categories are not identical.
See The Scribe Problem.
Could Voynich Be Copied From Plaintext While Appearing Nothing Like It?
Yes.
If the surviving script is transformed language, the visible Voynich page need not resemble its source text graphically.
A plaintext exemplar might be:
- Latin;
- Italian;
- German;
- another natural language;
- a technical shorthand;
- a mixed manuscript notation.
The Voynich scribe could then:
- translate;
- abbreviate;
- encipher;
- expand letters into verbose groups;
- insert nulls;
- change segmentation;
- use an artificial orthography.
Now “find the manuscript that looks like Voynichese” becomes the wrong search.
The source may look completely ordinary.
This is one reason the Cipher, Plaintext or Generated System? article matters for source history.
Could the Transformation Itself Have Happened Before the Final Copy?
Yes again.
There may have been an intermediate draft already written in Voynichese.
That would produce a two-stage pipeline:
- source language or information is transformed into Voynichese on a working copy;
- the polished surviving manuscript copies that Voynichese draft.
This would explain why a complicated underlying transformation could coexist with apparently fluent final writing and few conspicuous corrections.
The hard work happened earlier.
Again, this is possible rather than demonstrated.
The point is that production fluency on the surviving parchment cannot be used automatically to infer simplicity of the hidden mechanism.
Could It Have Been Dictated?
Dictation is another production model often forgotten because manuscripts look so visual.
A reader can speak from a source while a scribe writes.
This can create characteristic phonological or segmentation errors in ordinary language.
For Voynich, dictation would raise difficult questions:
- was Voynichese itself spoken aloud?
- was readable language dictated while the scribe transformed it?
- how were diagram labels coordinated spatially?
- why are word boundaries so visually stable?
- what error signature should dictation leave?
The model is possible but carries specific burdens.
Could the Makers Have Worked From Memory?
Not all knowledge requires a book on the desk.
Practitioners memorise forms.
Artists know conventional diagrams.
Medical workers know recipes.
Astrologers know standard zodiac structures.
A manuscript can therefore reproduce inherited conventions without a specific physical exemplar for every page.
This matters especially when a Voynich image resembles a broad tradition but not one uniquely identifiable manuscript.
The source may be cultural memory rather than one lost codex.
The Herbal Tradition Demonstrates Why Descent Is Hard
Medieval herbals often descend through long chains of copying.
An image can mutate.
Roots become stylised.
Leaves simplify.
Proportions drift.
One copyist misunderstands another.
A later illustrator corrects one feature while distorting another.
By the time two manuscripts survive, they may share ancestry without looking identical.
Conversely, two manuscripts can look similar because they belong to the same broad iconographic culture without sharing a direct parent.
Similarity can indicate a tradition. It does not automatically identify a genealogical link.
This is why comparisons with the Carrara Herbal, Egerton 747, Sloane 4016, Morgan M.873, Masson 116 and related manuscripts are useful as historical controls but must not be promoted into direct Voynich ancestry without a custody or copying mechanism.
A Comparator Is Not an Exemplar
This distinction deserves its own line.
Comparator: a manuscript that shows what was historically possible. Exemplar: a source actually used to make another manuscript.
One can be established by useful resemblance.
The other requires evidence of descent or use.
This is the same reason provenance cannot be inferred from visual resemblance alone.
What Would a Direct Exemplar Match Look Like?
Imagine a newly discovered fifteenth-century manuscript contains several plant drawings in the same unusual sequence as Voynich.
Not merely similar plants.
The same rare distortions.
The same accidental omission.
The same unusual foldout relationship.
The same ordering anomalies.
Now the case becomes much stronger because shared errors and arbitrary details can reveal copying descent.
A true exemplar relationship should explain coincidences that are unlikely to arise merely from common genre.
Shared Errors Are Often More Valuable Than Shared Beauty
If two manuscripts contain the same elegant zodiac sign, they may simply share a cultural convention.
If both contain the same strange misdrawn leg, missing sector or misplaced label, the relationship becomes harder to dismiss.
Why?
Because arbitrary mistakes are less likely to be invented independently.
This is one of the central principles of textual criticism and manuscript genealogy.
The Voynich problem is that we do not currently possess a candidate source showing enough such independent shared anomalies to establish direct copying.
What Would Multiple Exemplars Predict?
If different sections derive from different source traditions, we might expect boundaries where several properties change together.
- image style;
- page layout;
- vocabulary;
- glyph distribution;
- scribal hand;
- abbreviation behaviour;
- diagram conventions;
- correction pattern.
A boundary that affects only images may reflect illustrator choice.
A boundary affecting image, text, hand and layout simultaneously becomes more interesting as a source or production break.
Even then, several explanations remain possible.
Different exemplars are one.
Different work sessions are another.
Different scribes or subjects are others.
The strength comes from convergence, not from one boundary line.
Could the Present Binding Hide Source Units?
Yes.
The surviving binding order may not preserve original production or conceptual sequence perfectly.
Missing leaves complicate reconstruction.
Bifolia preserve physical relationships that modern page adjacency can obscure.
This means a source boundary should be tested against:
- bifolium structure;
- quire structure;
- missing leaves;
- scribal hands;
- text regimes.
A modern run of consecutive folio numbers is not automatically one original source unit.
See The Bifolium as a Production Unit and The Missing Leaves and Quire Reconstruction.
The Exemplar Could Be Younger Than the Knowledge It Contains
Source history can have several generations.
An early medical or botanical tradition may be centuries old.
A fifteenth-century exemplar may already be a late descendant of that tradition.
The Voynich Manuscript might then copy the fifteenth-century intermediary rather than the ancient original.
This is why cultural ancestry, textual ancestry and physical provenance must remain separate.
An idea can be old while the parchment is new.
An image tradition can be old while the visible drawing is newly executed.
A source can be lost while its descendant survives.
The Dangerous Sentence: “This Must Be the Source”
Voynich research repeatedly discovers something that looks right.
A herbal resembles a Voynich plant.
A bath manuscript resembles Quire 13.
A cosmological diagram resembles a foldout.
A jar resembles a pharmaceutical vessel.
Then resemblance becomes “source”.
That jump requires evidence that has not yet appeared.
For direct source attribution, ask:
- Does the candidate predate the Voynich production?
- Could the makers realistically have accessed it or a descendant?
- Are similarities specific rather than generic?
- Are arbitrary or erroneous details shared?
- Does ordering correspond?
- Do textual relationships support image relationships?
- Does the model explain differences as well as similarities?
- Is there a plausible copying chain?
Without that, we have a comparator.
Comparators are valuable enough.
What the Exemplar Problem Does to Decipherment Claims
A claimed decipherment often assumes the surviving manuscript directly records the author’s intended words.
But if Voynich is a copy, the visible text may include:
- copying errors;
- inherited source errors;
- scribal regularisation;
- abbreviations introduced during copying;
- layout-driven variants;
- changes made by different scribes;
- losses from missing folios;
- transformations made before the final copy.
A genuine decipherment should be robust to ordinary manuscript transmission.
It should not collapse whenever one copied form is imperfect.
At the same time, “copying error” cannot become a universal excuse for every failed translation.
Transmission explains limited noise. It cannot rescue an unlimited number of contradictions.
What Would Strengthen the Exemplar Hypothesis?
- shared arbitrary errors with another manuscript;
- matching rare image sequences;
- matching unusual text-image layout patterns;
- copying-error signatures in Voynich sequences;
- section boundaries aligning with identifiable source traditions;
- a discovered draft or intermediate witness;
- material or archival evidence connecting a candidate source to the relevant workshop or owner;
- predictable transformation from candidate source text into Voynichese.
What Would Weaken It?
- evidence that content was repeatedly revised during first composition on parchment;
- page structures generated uniquely from local geometry with no stable inherited template;
- visual forms that cannot be related to known copying traditions despite broad search;
- strong evidence that textual variation tracks individual scribes rather than source domains;
- a mechanism showing direct composition can reproduce all observed planning and correction behaviour more economically.
Primary School: Copy a Picture, Write a New Caption
Give a child a picture of a tree.
Ask them to copy the picture.
Then ask them to write a completely new sentence beside it.
The copied picture now has an exemplar.
The sentence does not.
This simple exercise explains why “the image came from somewhere” does not mean “the text came from the same place”.
Secondary School: Build Four Production Models
- direct composition;
- copy from one complete exemplar;
- compile several sources;
- copy images while transforming or composing text.
For each Voynich observation, ask which models explain it naturally and which need additional assumptions.
Students learn that the same evidence can support several causal histories.
JC and Adult Readers: Treat Source History as a Latent-Variable Problem
The surviving manuscript is observed.
The exemplar, if any, is latent.
Several hidden source histories can produce the same visible page.
The research task is therefore model selection.
Which hidden production history explains:
- page planning;
- correction scarcity;
- scribal variation;
- Currier regimes;
- visual inheritance;
- text-image coordination;
- repetition;
- section boundaries;
- physical gathering structure
with the fewest arbitrary assumptions?
The exemplar problem is therefore not solved by finding the prettiest comparison.
It is solved by recovering a causal history.
Reader Checklist: Before You Call Another Manuscript the Voynich Source
- Is it a comparator or a proposed direct exemplar?
- Does it predate Voynich?
- Are similarities generic or unusually specific?
- Are shared mistakes present?
- Does ordering correspond?
- Do text and image both correspond?
- Could both descend independently from an earlier tradition?
- Is physical or archival contact plausible?
- Does the proposal explain differences?
- Could several sources have been compiled instead?
- Does Currier or scribal evidence align with the source boundary?
- Does the hypothesis make a new prediction?
Frequently Asked Questions
Has the original Voynich exemplar been found?
No complete direct exemplar has been securely established. Many manuscripts provide useful comparisons, but resemblance alone does not demonstrate copying descent.
Does clean page planning prove there was a draft?
No. A prepared draft explains clean layout well, but an experienced scribe can also compose flexibly around pre-existing drawings.
Could different Voynich sections come from different sources?
Yes, compilation is historically plausible. Visual or textual differences alone do not prove it; the strongest case would involve several independent properties changing at the same boundary.
Could images be copied while text was newly written?
Yes. Image and text transmission can be partly independent. This is common enough historically that a visual source identification cannot automatically identify the text source.
Could the source have been normal language?
Yes. If Voynichese is transformed language, a readable plaintext source could have been translated, abbreviated or enciphered during an earlier or final production stage.
Research Foundations
- René Zandbergen — Origin and Production of the Voynich Manuscript
- Yale University Press — The Voynich Manuscript, edited by Raymond Clemens
- Voynich — eduKateSG Master Article
The Final Idea
The surviving Voynich Manuscript is not necessarily the place where Voynich began.
That sounds simple.
It changes almost everything.
A clean page may inherit a messy draft.
A strange plant may inherit a centuries-old pictorial tradition.
A fluent line of Voynichese may inherit an earlier transformation step.
A coherent codex may compile several older intellectual worlds.
None of those possibilities should be promoted into history without evidence.
But none should be erased merely because the earlier source is lost.
The exemplar problem asks us to distinguish the manuscript we possess from the information history that may have produced it.