Voynich Currier Switch: Is A/B One Hidden Control Setting?

Currier A and B are among the strongest recurring facts in Voynich research.

Different pages prefer different character groups.

Different word families rise and fall.

The distinction survives character counts, bigrams and whole-token statistics.

But what if A and B are not the deepest layer?

What if they are the visible result of one hidden control setting acting on a more complicated shared system?

That is the question raised sharply by Christophe Parisel’s April 2026 preprint.

The paper first attempts something useful and conservative: recover the Currier distinction quantitatively without simply feeding the original labels into the model.

It reports that a two-component mixture is strongly preferred and that held-out folio labels can be predicted at high accuracy.

Then it goes further.

Its proposed dominant A/B mechanism is surprisingly compact.

A binary folio-level state governs which vowel-like unit tends to follow the digraphs ch and sh.

The paper also reports that token template explains much of the remaining variance.

In other words:

Currier A/B may be a low-resolution projection of a richer generative system in which one folio-level switch changes part of the token grammar while local template controls much of the rest.

That is an elegant hypothesis.

It is also a 2026 preprint, not settled Voynich knowledge.

And the word “switch” is dangerous because it can sound as if a cipher key has been discovered.

It has not.

A statistical binary state can arise from language, dialect, orthography, shorthand, cipher setting, scribal preference, exemplar tradition or another production variable.

The Currier Switch Problem is therefore not whether a two-state model exists. It is what historical mechanism could create that state, why it is stable at folio scale, and whether the same mechanism predicts the rest of the manuscript.

Quick Read

Direct answer: Currier A/B is a robust descriptive distinction. Parisel’s 2026 preprint reports that the split can be recovered quantitatively and proposes that a major component behaves like a binary state set once per folio, especially affecting what follows ch/sh, while token-template identity explains most additional variation. This is a promising mechanism model, not proof of a cipher switch, two dialects or one specific historical process.

Currier’s A/B labels describe statistical text regimes, not proven languages.
The distinction has been independently recoverable through many kinds of quantitative analysis.
Parisel’s 2026 preprint uses character-pair substitution ratios across folios and reports a strong two-component structure.
The model predicts held-out A/B labels at roughly 89% accuracy under the paper’s analysis.
The paper argues that the A/B contrast is not primitive but one projection of a higher-dimensional system.
Its dominant proposed component is a binary folio-level switch governing the vowel-like unit following ch and sh.
The paper reports that token-template identity accounts for most of the variance in the relevant environments.
Most folios are assigned strongly under the switch model.
This does not identify the historical meaning of the switch.
A binary state can be linguistic, orthographic, cryptographic, scribal or procedural.
RZ C and other intermediate material complicate a simple two-box model.
Drift evidence raises the possibility that discrete and continuous variation coexist.
Bifolium coherence makes physical production scale important.
Scribal hands must be controlled because text state and hand are correlated but not identical.
A real control-state model should predict unseen folios and independent physical or scribal variables.

Currier’s Original Warning Still Matters

Prescott Currier used the word “language” loosely.

He explicitly warned that A and B were labels for marked statistical differences.

They did not prove two underlying spoken languages.

This is still the correct starting point.

The modern question is not:

Which language is A and which language is B?

It is:

what latent variable or variables generate the reproducible difference we call A/B?

The Difference Exists at Several Resolutions

Currier differences are visible at:

single-character frequencies;
character groups;
word families;
whole-token frequencies;
final forms;
initial constructions.

This matters because a one-variable explanation has a high bar.

If one switch drives A/B, its consequences must propagate across these different scales naturally.

A switch that explains one bigram but leaves the rest unrelated is incomplete.

What Parisel’s 2026 Model Adds

The preprint approaches the problem in two stages.

First, recover the distinction quantitatively.

Second, ask whether A/B can be decomposed into simpler underlying factors.

The reported result is that a two-state structure is strongly supported, but the states themselves appear to be generated by more local token templates plus one dominant folio-level binary choice.

This is conceptually important.

A category can be real without being fundamental.

The ch/sh Environment Is the Proposed Lever

At the centre of the proposed switch is a particular environment involving ch and sh and the vowel-like material that follows them under the paper’s representation.

One folio-level state favours one outcome.

The other favours another.

The switch is not applied independently to every word.

It is proposed as a page-scale setting.

That makes it historically interesting.

A reader or scribe could plausibly operate under one convention for a folio and another elsewhere.

But the historical identity of the convention remains open.

“Vowel-Like” Is Not “Decoded Vowel”

The wording matters.

Calling one character family vowel-like describes its statistical role in the model.

It does not prove phonological vowel identity.

The Vowel Problem already showed that manuscript-wide vowel/consonant separation remains unresolved.

Therefore the Currier switch can be modelled over visible categories without assuming spoken sounds.

If later evidence connects the states to phonology, that becomes a second claim.

Template Identity Explaining Most Variance Is a Major Constraint

The preprint reports that token-template identity accounts for a very large share of variation in the relevant contexts.

This means the binary switch is not operating on a blank slate.

Voynich tokens already belong to strongly constrained structural families.

The switch modulates those families rather than generating words from nothing.

This aligns with decades of observations that Voynich tokens have slot-like internal grammar.

The deeper architecture may therefore be:

template grammar + folio state + local variation.

That is a mechanism proposal worth testing independently.

A Cipher-State Interpretation Is Tempting

A cryptographic system can have states.

One key table for one folio.

Another table later.

One homophone preference in A.

Another in B.

This could create exactly the kind of persistent local difference Currier saw.

But calling the statistical state a cipher state merely renames the unknown unless an encryption rule is supplied.

A real cipher-state model should encrypt plausible plaintext forward and reproduce the A/B switch automatically.

A Dialect or Orthographic Interpretation Is Equally Possible

Two scribes can spell the same language differently.

One dialect can favour one syllable or ending.

A shorthand school can use different conventions.

A folio-level switch can therefore arise without secrecy.

The historical mechanism must be inferred from cross-axis evidence rather than the existence of two states alone.

Scribal Hands Are an Immediate Confounder

Currier originally correlated language states with handwriting styles.

Lisa Fagin Davis later proposed a more detailed five-scribe palaeographic model.

The modern hand model does not map trivially onto Currier A/B.

This is useful.

If the folio switch crosses proposed hand boundaries, a purely individual scribal habit weakens.

If it aligns almost perfectly with one hand, scribal convention strengthens.

Text state and hand should therefore be estimated separately and compared.

The Bifolium Is a Better Production Test Than the Page Alone

Currier observed that herbal bifolia tend to remain within one language regime.

That is important for a folio-level switch model.

The physical maker handled sheets, not isolated digital pages.

If the control state belongs to production, the paired surfaces of a bifolium should show coherent behaviour.

A switch that changes arbitrarily between recto and verso would be harder to reconcile with sheet-level preparation.

Bifolium structure can therefore test the natural scale of the state.

RZ C Complicates the Binary Picture

Zandbergen’s newer RZ classification adds an intermediate C-like state, especially in astronomical and cosmological material.

This immediately asks a hard question of a binary switch.

Is C:

a mixture of A- and B-like templates;
a third state;
an intermediate parameter value;
a document-role effect that projects between the two?

A good switch model should predict where C-like material lands rather than ignore it because Currier did not classify it originally.

The Drift Problem Complicates It Further

Other recent work argues that Voynich vocabulary and handwriting may change continuously across production.

Continuous drift and a binary switch can coexist.

A typewriter has a shift key and still wears gradually.

A dialect can evolve while one spelling convention remains binary.

A cipher can change homophone frequencies gradually within two larger key states.

Therefore the correct model may contain both discrete and continuous dimensions.

Forcing them into an either/or choice may be another false binary.

Document Role May Create the Apparent Switch

If A and B pages perform different textual jobs, token templates may change accordingly.

Running prose.

Labels.

Records.

Diagram annotations.

One folio can then appear to choose a state because document role determines the relevant grammar.

A real control-setting theory must therefore compare matched document roles across regimes.

The Switch Should Predict More Than Its Discovery Feature

This is the central validation rule.

Suppose the switch is discovered from ch/sh follower behaviour.

Then ask what else it predicts.

word-family frequencies;
suffix distributions;
line-start forms;
edge coupling;
rare glyph use;
label grammar;
hand or bifolium grouping.

If the state predicts only the exact feature used to define it, the explanatory gain is small.

If it predicts independent dimensions, the hidden-control interpretation strengthens.

A Switch Has to Be Operational

What would a fifteenth-century user actually do?

Choose spelling convention A for this folio?

Use cipher table B?

Write one shorthand dialect?

Copy one exemplar tradition?

A statistical state becomes historical explanation only when a plausible operation produces it.

The Almost-Correctionless Manuscript Is Relevant Again

A binary state is easy for a human to maintain.

Choose A.

Write.

Choose B elsewhere.

This is operationally cheaper than a continuously changing complex key.

That makes folio-scale binary control plausible from a human-factors perspective.

It still does not identify what the state means.

A Good Switch Model Should Classify Unseen Folios

Learn the state definition on one corpus.

Freeze it.

Now take an unseen folio.

The model should assign its state before consulting the traditional Currier label.

Then the assigned state should predict other independent properties.

This converts classification into mechanism testing.

A Better Test Uses Bifolia Held Out Together

Holding out random words leaks page-specific information.

A stronger test withholds entire physical production units.

Train on some bifolia.

Predict the state of unseen bifolia.

Then compare with scribal hand, visual role and codicological placement.

The physical manuscript becomes part of the validation.

The Switch Must Eventually Connect to Meaning—or Explicitly Not

A hidden control state can be purely orthographic.

Purely cryptographic.

Or reflect different underlying language.

A future decipherment should tell us which.

If A and B decrypt to the same grammar and vocabulary under different surface mappings, control-state/cipher or orthography strengthens.

If they decrypt to genuinely different language systems, the old “two languages” interpretation returns with evidence.

Until then, the switch is structural rather than semantic.

What Survives the Currier Switch Work

Currier A/B is a robust statistical distinction.
Currier himself did not claim that A/B proved two spoken languages.
Parisel’s 2026 preprint independently recovers a strong two-state structure.
The paper proposes a dominant folio-level binary switch affecting ch/sh follower behaviour.
It also reports that token-template identity explains most additional variance in the relevant analysis.
This suggests A/B may be a projection of a richer generative system.
The historical interpretation of the switch remains unknown.
RZ C, scribal hand, bifolium structure, document role and continuous drift all constrain the model.
A successful switch theory should predict independent features and unseen folios.
No accepted historical mechanism has yet been identified as “the Currier switch”.

What Does Not Survive as Established Knowledge

Currier A and B are proven two languages.
Parisel’s switch is a proven cipher key.
The ch/sh follower is a decoded vowel.
A/B is explained entirely by one binary variable.
RZ C is irrelevant to the switch model.
Scribal hand and Currier state are the same thing.
Continuous drift disproves discrete states.
Discrete states disprove continuous drift.

A Better Currier-Switch Analysis

Recover state labels without using Currier’s labels during fitting.
Freeze the switch definition.
Withhold entire bifolia or folios for prediction.
Test independent character and word-family features.
Control token template explicitly.
Compare RZ C and other intermediate material.
Compare proposed scribal hands.
Control document role and illustration family.
Fit hybrid models containing both switch and drift.
Require a historical operational mechanism before naming the state cipher, dialect or shorthand.

What Would Count as a Real Currier-Switch Breakthrough?

Imagine the binary switch is replicated independently under another transcription.

The switch is learned on one subset and predicts unseen bifolia.

Without using their labels, the state predicts several other A/B properties.

RZ C is explained as a known mixture or third parameter regime rather than an exception.

The state crosses or aligns with scribal hands in a reproducible way.

A historically documented shorthand, orthographic or cipher process reproduces the same state transition from meaningful source text.

Then a future decipherment shows the predicted underlying relationship between A and B.

That would convert a statistical switch into production history.

The Currier switch becomes explanatory when one compact state variable predicts features it was not built from and connects to an operation a historical writer could actually perform.

Primary School: One Rule, Two Modes

Invent a word game.

In Mode A, every “ch” is followed by one marker.

In Mode B, it is followed by another.

Everything else stays mostly the same.

A reader can classify whole pages by one small recurring difference.

This is the intuition behind a hidden switch.

Secondary School: Category or Cause?

Create two groups of synthetic texts that differ under one switch plus several shared templates.

Students can cluster the texts into A and B.

Then ask whether discovering the clusters automatically tells them why the switch exists.

It does not.

Classification and mechanism are separate scientific jobs.

JC and Adult Readers: Latent State Plus Template

At a higher level, the proposed architecture is a hierarchical latent-state model.

A folio receives a global state.

Each token belongs to a local template.

The observed glyph outcome depends on both.

The statistical task is to show that this hierarchy predicts held-out observations better than purely discrete A/B, purely continuous drift, hand-only and document-role-only alternatives.

Reader Checklist: Before You Call A/B a Hidden Switch

Was A/B recovered without using Currier labels?
What exact feature defines the switch?
Is it genuinely folio-level?
Does it remain stable across recto/verso?
Does it respect bifolium coherence?
How much variance is already explained by token template?
What happens to RZ C?
Does the switch align with scribal hand?
Does document role explain the same pattern?
Can switch plus drift outperform either alone?
Does the state predict independent features?
Can a historical mechanism generate it?

Frequently Asked Questions

Are Currier A and B real?

Yes as reproducible statistical regimes. Their historical cause and semantic meaning remain unresolved.

What is the 2026 Currier-switch hypothesis?

Christophe Parisel’s preprint proposes that a dominant part of the A/B distinction behaves like a binary state set once per folio, especially affecting the unit following ch/sh, while token template explains much of the remaining variation.

Does this prove a cipher key?

No. A statistical state does not identify whether the historical cause was cipher, dialect, orthography, shorthand, scribe or another mechanism.

What about RZ C?

Intermediate material is an important challenge and opportunity. A mature switch model must explain whether C is mixture, third state or another independent axis.

What is the strongest current conclusion?

A/B may be compressible into simpler latent factors than two monolithic “languages”, but the historical identity of those factors is not yet known.

Research and Further Reading

The Final Idea

Currier gave us two boxes.

The boxes remain useful because the manuscript really does behave differently inside them.

Now the deeper question is whether the boxes are shadows.

A page-level setting.

A token template.

A scribal convention.

A continuous drift.

Several axes layered together.

The next Currier breakthrough will not be a better name for A and B. It will be the smallest historical mechanism that can generate both states, the intermediate material between them, and the physical manuscript that carries them.

Voynich | Everything eduKate Knows and Tested | The Currier Switch Problem: Is A/B One Hidden Control Setting?