A serious Voynich mechanism has to survive two very different kinds of evidence.
The first is static.
Frequency.
Entropy.
Word length.
Currier differences.
The second is dynamic.
How one visible form leads to another.
How boundaries behave.
How similar forms cluster.
How local history appears to persist through a line.
Many Voynich theories reproduce one class and fail the other.
That is why Vitaly Averyanov’s 2026 preprint A Workshop Cipher is worth treating seriously as a mechanism proposal even though it does not claim a reading. The paper presents a Naibbe-class verbose cipher augmented with scribal-workshop behaviours such as serial reuse, graphotactic choice, spelling noise and section-level variation, and reports that one compact model reproduces several static and dynamic signatures together.
The claim is strong.
So the evidentiary burden must be strong too.
A model that can imitate Voynich at several levels has earned attention. It has not yet earned the historical manuscript.
Quick Read
- Averyanov’s 2026 work is a preprint, archived on Zenodo and accompanied by a replication package.
- It proposes a Naibbe-class verbose cipher plus simple workshop-style scribal rules rather than a plaintext reading.
- The model is designed to reproduce both older static Voynich properties and newer dynamic boundary signatures.
- The paper also models distributionally interchangeable “twin” forms as a homophony-like layer and attempts to reproduce their within-line clustering.
- Section-level variation is treated as a change in cipher-table dialect rather than as a completely separate system.
- The reported fit is important because it raises the standard from matching one statistic to matching several layers jointly.
- The model does not recover the actual historical key.
- The associated public project reports null results for two key-recovery routes.
- The model does not provide a manuscript-wide accepted translation.
- Its strongest present role is as an executable historical-mechanism hypothesis and a demanding control for competing explanations.
What Is a Naibbe-Class Cipher?
The Naibbe cipher, published by M. A. Greshko in 2025, is important because it demonstrates capability.
A historically hand-operable cipher can transform Latin or Italian into ciphertext whose surface statistics resemble several Voynich properties.
This does not prove Voynich uses Naibbe.
It changes the null model.
Before such demonstrations, some researchers treated low conditional entropy or word-family behaviour as strong evidence against cipher.
After them, the safer statement is:
some historically plausible verbose ciphers can generate several Voynich-like statistics.
Averyanov’s workshop model builds on that capability class rather than starting from a modern neural generator.
Why “Workshop” Matters
The paper’s most interesting move is not the cipher table by itself.
It is the addition of scribal behaviour.
A real fifteenth-century manuscript is not produced by a perfectly memoryless algorithm.
Scribes repeat habits.
They prefer recently used spellings.
They avoid awkward repetitions.
Different hands can prefer different variants.
One section can inherit a slightly different working table or convention.
The workshop-cipher model tries to make those behaviours causal rather than treating them as noise around a cipher.
That matters because several modern Voynich results concern local history, not only global frequencies.
Statics: The Easy Part Is No Longer Easy
Voynich statistics often discussed as “headline” properties include:
- Zipf-like rank-frequency behaviour;
- low conditional character entropy;
- narrow token-length distributions;
- strongly constrained character sequences;
- Currier A/B differences;
- local vocabulary clustering.
Many generators can reproduce some of these.
The research problem has therefore moved.
Matching one statistic is no longer impressive enough.
The relevant question is joint survival across multiple independent constraints.
See The Joint Benchmark Problem.
Dynamics: Where the New Bar Is Higher
Christophe Parisel’s 2026 work raised the bar by asking about direction and boundaries.
The workshop-cipher preprint explicitly attempts to reproduce four newer boundary-related signatures reported in that research.
That matters because a generator can match token frequencies while still arranging those tokens incorrectly.
A real mechanism needs to know more than which forms are common.
It needs to generate the right dependencies across the places where forms meet.
This connects directly to The Directional Dissociation Problem and The Edge Problem.
The Twin Problem
The preprint also focuses on token pairs that are distributionally similar despite being visibly different.
In ordinary language, this can happen with synonyms.
In a cipher, it can happen with homophones: different ciphertext forms representing the same plaintext value.
Averyanov interprets these “twins” as a fingerprint of homophony and reports that the model reproduces their tendency to cluster within lines above chance.
This is an interesting mechanism-level claim because it connects two observations:
- contextual interchangeability;
- local serial clustering.
But homophony is not the only possible explanation.
Near-synonymous labels, formulaic alternation, scribal allography, generated substitution classes and morphological variants can also create contextually similar forms.
The twin result therefore strengthens the cipher hypothesis only if competing mechanisms fail the same test under fair controls.
Currier A/B as Table Dialect
One attractive feature of the workshop model is its treatment of Currier A/B.
Instead of requiring two languages, the model allows section- or workshop-level table variation.
This is historically plausible in principle.
A working cipher tradition can change conventions between hands or production batches without changing the underlying plaintext language.
But Currier has become more complicated than a simple binary.
Recent work on drift and the Currier switch asks whether the A/B distinction is one projection of a deeper control variable rather than two fixed tables.
A strong workshop model must therefore explain:
- classic A/B differences;
- mixed or transitional folios;
- continuous drift signals;
- hand-specific preferences;
- bifolium and quire structure.
The model should not be allowed to call every mismatch “another table dialect” after the fact.
Four Parameters Are Attractive—If They Stay Four
The preprint emphasises a compact set of interpretable parameters layered over the underlying cipher class.
That is a strength.
Historical mechanisms should preferably be compressible.
If every new Voynich anomaly requires a new parameter, the theory becomes an emulator rather than an explanation.
The critical future test is therefore parameter stability.
Freeze the model.
Then expose it to new independent metrics.
If the same compact parameterisation survives new evidence, confidence rises.
If the parameter count expands each time the manuscript surprises us, confidence falls.
The Missing Key Is Not a Minor Detail
The public project explicitly reports null results from two key-recovery attempts.
This is a good sign of evidentiary discipline.
A cipher mechanism can reproduce ciphertext statistics without allowing the actual historical key to be recovered from the surviving data.
But for decipherment, the missing key remains central.
Without the historical table or an externally anchored plaintext crib, the model remains a mechanism class.
It does not become a translation.
The project’s own statement—“the mechanism, not a reading”—is exactly the boundary we should preserve.
Why Null Key Recovery Is Useful
A null result can constrain the model.
If an external partition test fails, one class of easy recovery is eliminated.
If internal hill-climbing fails under declared assumptions, another easy route is eliminated.
That is better than quietly changing the key-search method until something readable appears.
Voynich has suffered from enough retrospective tuning already.
Could the Model Be Statistically Right and Historically Wrong?
Absolutely.
Two different mechanisms can occupy the same statistical region.
A modern generator can imitate medieval output.
A shorthand system can imitate a cipher.
A recursive generator can imitate morphology.
Historical identification therefore requires evidence from outside the text statistics.
- scribal ductus;
- known historical cipher practice;
- workflow plausibility;
- correction patterns;
- page planning;
- material chronology;
- external cribs;
- document function.
The workshop model is strongest when it invites those tests rather than claiming statistics have already replaced them.
Could the Same Observations Arise From Shorthand?
Yes.
The new Shorthand Problem exists because syllabic or tachygraphic compression can create low entropy, recurrent units and unusual vowel behaviour.
A strong cipher model should therefore be compared directly with shorthand under the same benchmark.
If both fit the same statistics, the discriminating evidence must come from where their predictions diverge.
Could It Be Self-Citation Plus Encryption?
Hybrid models are possible too.
A scribe can encrypt meaningful plaintext and still reuse recent ciphertext variants.
That would combine homophony with seriality.
The historical mechanism space is therefore not a set of mutually exclusive boxes.
This is why the canonical Cipher, Plaintext or Generated System? article treats hybrids explicitly.
The Most Important Test Is Out-of-Sample
The model’s future value depends on what happens after its publication.
New evidence appears.
New metrics are proposed.
Independent researchers reproduce the code.
Parameters are frozen.
Then the model is asked to predict structures not used in construction.
Examples include:
- hapax distribution;
- learned unit scale;
- visual/text section correspondences;
- label-specific structure;
- line and paragraph boundary grammar;
- scribe-transfer behaviour.
A model that keeps surviving becomes increasingly valuable.
A model that must be rewritten becomes a useful historical step rather than the answer.
What the Workshop Cipher Problem Does Not Prove
- It does not prove Voynich is a Naibbe cipher.
- It does not recover the historical key.
- It does not prove the plaintext language.
- It does not produce an accepted translation.
- It does not prove five hands equal five cipher-table dialects.
- It does not prove all twins are homophones.
- It does not eliminate shorthand, abbreviation, generated or hybrid explanations.
- It does not turn a preprint into consensus.
What It Can Give Us
- An executable historically plausible mechanism class.
- A joint test across static and dynamic text properties.
- A compact way to model serial reuse and local variation.
- A direct challenge to theories that explain only corpus-wide statistics.
- A public replication package and declared null key-recovery results.
- A new control that every competing Voynich mechanism should try to beat.
Reader Checklist
- Which Voynich statistics were used to construct the model?
- Which were genuinely held out?
- Are the dynamic signatures independently replicated?
- Does the model reproduce Currier variation without ad hoc retuning?
- Does it reproduce scribe variation?
- Does it reproduce label-specific behaviour?
- Does it survive Hapax, Unit-Scale and Edge tests?
- Can parameters remain fixed across new evidence?
- Is the historical key recoverable?
- Is mechanism fit being confused with plaintext recovery?
Research Foundations
- Vitaly Averyanov — A Workshop Cipher (2026 preprint).
- Voynich Studies — Publications and preprint status.
- Voynich Studies — Data, code and reported null key-recovery attempts.
- M. A. Greshko — The Naibbe Cipher.
- Christophe Parisel — Layered Positional and Directional Constraints (2026 preprint).
The Final Idea
A good Voynich mechanism should explain why the manuscript looks the way it does when we stop reading and start measuring.
The workshop-cipher proposal takes that challenge seriously.
Its next challenge is harder.
Explain the object.
Explain the hands.
Explain the labels.
And eventually recover something that survives as meaning.
A mechanism that reproduces the manuscript’s statistics has earned a seat at the table. The historical key is what would turn the seat into an answer.