Near the end of the Voynich Manuscript, something remarkable happens.
The pictures almost disappear.
No enormous plant dominates the page.
No zodiac animal sits at the centre of a ring.
No crowd of small figures stands inside green pools.
No row of elaborate vessel-like forms controls the eye.
Instead, the manuscript gives us writing.
Lots of it.
And beside that writing, again and again, a small star-like marker.
One star.
One short paragraph.
Then another star.
Another paragraph.
Across the surviving final section, hundreds of these markers turn the manuscript into a forest of entries.
This is Quire 20, commonly called the recipe section.
The name is understandable.
Short entries in a book already filled with plants, vessels and body-like imagery look exactly like the place where recipes ought to appear.
But “where recipes ought to appear” is not the same thing as “we have read a recipe”.
What Quire 20 gives us securely is not 324 decoded recipes. It gives us a highly segmented record system marked by 324 surviving star-like marginal signs.
That difference is where the interesting work begins.
Quick Read
One-sentence answer: Quire 20 is the text-heavy final gathering of the Voynich Manuscript in which 324 surviving marginal star-like markers segment the writing into many short entry-like units, strongly demonstrating an explicit record architecture while leaving the records’ exact function—recipes, notes, instructions, index entries, summaries, formulae or another structured list—undeciphered.
- Quire 20 occupies the final major text-heavy section of the manuscript.
- The surviving folios are conventionally numbered 103–108 and 111–116; the gap around 109–110 is consistent with a missing central bifolium.
- The original quire therefore contained more material than survives today.
- There are 324 marginal star-like markers on the surviving pages according to detailed folio analysis.
- Calling them “stars” describes their visual shape; it does not prove that they represent celestial stars.
- Most markers precede or accompany short blocks of text, creating a strong entry-delimiting pattern.
- The modern term “recipes” comes from the resemblance between these short segmented records and recipe entries in medical manuscripts.
- No accepted decipherment proves that each marked unit is a medical recipe.
- Star morphology varies: point counts, tails, dots and occasional colour differ across examples.
- Those variations may encode class, hierarchy, section, emphasis or nothing semantic; a stable legend has not been established.
- Some pages contain longer-running text alongside marginal stars, showing that the simple model “one star = one short paragraph” has exceptions.
- Folio 116r includes stars with varying relationships to adjacent writing, again warning against a rigid one-marker-one-recipe rule.
- An earlier text page, f58r, also uses marginal star-like markers, showing that star-delimited text is not unique to Quire 20.
- Historical medical manuscripts really do contain large collections of short recipes, formulae, antidotaries and practical instructions, making recipe interpretation plausible.
- But the strongest current conclusion is document-architectural: the stars act as repeated boundaries or addresses for records.
In a manuscript where so many divisions have to be inferred, Quire 20 does something unusually generous.
It shows us where many of its units begin.
It just does not tell us what the units are called.
The End of the Manuscript Changes the Reader’s Job
Earlier Voynich pages force image and text into the same interpretive field.
The plant can contaminate the translation.
The zodiac can tell us what we expect the labels to mean.
The pool can make every nearby word look medical.
Quire 20 removes much of that temptation.
Now the main evidence is textual architecture.
Marker.
Entry.
Marker.
Entry.
The page asks a different question:
What kind of knowledge is useful in hundreds of separately marked textual units?
That question is broader than recipes.
And because the architecture is so explicit, it may ultimately be one of the best places to learn how Voynich writing organises records.
Quire 20 Is a Physical Gathering, Not Just “The Last Pages”
As with Quire 13, physical structure matters.
The surviving final gathering includes folios 103–108 and 111–116.
The absent folio numbers 109 and 110 correspond neatly to one missing bifolium.
Reconstruction work therefore suggests an original structure larger than the surviving one.
This has an important consequence.
The 324 surviving star markers are not necessarily the original total.
Missing leaves may have carried additional entries.
So it is unsafe to build symbolic theories around “exactly 324 recipes” as though 324 were an original designed total.
The number is a count of surviving evidence.
It is not automatically a sacred, calendrical or pharmacological number chosen by the maker.
A Missing Bifolium Changes More Than the Count
Missing material can also interrupt sequence.
Perhaps entries were grouped by topic.
Perhaps one subsection began on the missing sheet.
Perhaps the missing bifolium contained a heading, index, transition or visual key.
We cannot assume any of these.
But we must lower confidence in theories that depend on uninterrupted current ordering.
A missing physical sheet is not merely missing words.
It may be missing structure.
This is why codicology belongs inside semantic reasoning.
We cannot reconstruct a classification sequence as though the page stack were complete when the book itself tells us otherwise.
324 Stars Is a Large Enough Pattern to Take Seriously
One star can be decorative.
Ten repeated stars suggest convention.
Hundreds of stars attached to textual units create architecture.
The surviving 324 markers therefore deserve to be treated as one of the strongest explicit segmentation systems in the manuscript.
The marker tells the reader:
something begins here.
That “something” may be:
- a recipe;
- a remedy;
- a formula;
- a note;
- a record;
- a definition;
- a prediction;
- an instruction;
- a summary;
- a cross-reference;
- another repeated textual unit.
Marker function is more secure than record semantics.
That is the right order of certainty.
The Word “Recipe” Is an Analogy, Not a Translation
Why are these pages called recipes?
Because the morphology makes sense.
A practical medical book can contain many short formulae.
Recipes are naturally segmented.
They often begin with a condition, purpose, ingredient or formula.
They can be copied as compact paragraphs.
And the wider Voynich Manuscript contains plants and pharmaceutical-looking material.
The analogy is therefore historically and visually reasonable.
But the crucial bridge remains missing:
we cannot yet read one entry and demonstrate that it gives ingredients, quantities and procedure.
Until that happens, “recipe section” should mean:
the final star-segmented text section conventionally interpreted as recipe-like.
The label remains useful as long as it does not hide the uncertainty.
Historical Medical Recipe Collections Make the Analogy Plausible
The medieval medical world really did produce large compilations of short practical recipes.
British Library Harley MS 2375 combines medical and astrological texts with many recipes.
Harley MS 2390 is another fifteenth-century medical compendium containing medical recipes alongside alchemy, astrology and prognostication.
Egerton MS 747 combines herbal material, the Antidotarium Nicolai, dosage, ingredient substitution, weights and measures, synonyms and calendrical material.
These known manuscripts teach us two things.
- A long practical manuscript can contain many compact records that modern readers reasonably call recipes.
- Recipes can coexist with herbs, astrology, antidotaries and other modules without the entire codex becoming one linear procedure.
This makes Quire 20’s conventional name plausible.
It does not decode a single star-marked entry.
Known manuscripts provide a function class.
Voynich must still prove membership.
A Recipe Has an Internal Grammar
If Quire 20 entries really are recipes, the hypothesis should eventually become visible in textual structure.
Recipes often need recurring functional slots.
- purpose or indication;
- ingredient;
- quantity;
- operation;
- administration;
- timing;
- warning or condition.
Not every recipe uses every slot.
But a large recipe corpus should produce recurring syntactic or lexical behaviour.
The same ingredient might recur.
The same preparation verb might recur in similar positions.
Quantities may form a restricted class.
Formulaic openings may appear.
This gives recipe theory a valuable property:
it can make predictions about the text before translation is complete.
A good recipe model should eventually discover repeated roles, not merely short paragraph lengths.
Short Paragraphs Are Compatible With Many Record Types
Here is the competing-model problem.
Recipes are short.
So are:
- dictionary entries;
- glosses;
- diagnostic notes;
- astrological judgements;
- calendar rules;
- inventory records;
- mnemonic prompts;
- index entries;
- observations;
- short prayers;
- formulae.
Therefore short length is not discriminating enough.
We need internal structure.
Recipe entries should have different statistical behaviour from name lists.
Index records should contain more repeated reference forms.
Astrological judgements may correlate with zodiac vocabulary if such vocabulary can be isolated.
A corpus does not reveal its genre merely by being divided into entries.
Segmentation is the first layer.
Genre is the next.
The Star Is Probably a Boundary Before It Is a Celestial Object
Inside the zodiac pages, a star-like form naturally invites celestial interpretation.
Inside Quire 20, the same visual instinct is less secure.
The stars sit in the margin beside text entries.
Their repeated placement behaves like a bullet, paragraph marker, entry sign or record delimiter.
This does not mean the star has no celestial meaning.
A symbol can carry both semantic and document-structural jobs.
But the strongest directly observable role is spatial:
the marker identifies or accompanies the beginning of a textual unit.
That is more secure than:
the marker means a star, planet or astrological influence.
Document function should be established before cosmological meaning.
The Zodiac and Quire 20 Stars May Be the Same Shape Doing Different Jobs
This is an important general principle.
A star on a map can mark a capital.
A star in a rating system can mean quality.
A star in a footnote can mark a note.
A star in astronomy can depict a star.
The silhouette is shared.
The function changes with context.
Therefore the recurrence of star-like forms between zodiac pages and Quire 20 does not prove shared celestial semantics.
It may reveal a shared visual marker vocabulary.
Or one semantic star concept may indeed have been reused structurally.
Distinguishing those possibilities requires correlations beyond shape.
Same symbol is evidence of reuse.
It is not automatic evidence of same meaning.
Star Morphology Varies
The marginal markers are not one mechanically identical stamp.
Many have seven or eight points.
Some include central dots.
Some carry red colouring.
Many have tails or stems extending from them.
This variation is tempting because it could encode metadata.
- entry class;
- importance;
- ingredient type;
- source;
- section;
- status;
- astrological category.
Or it may simply reflect handwriting variation and decoration.
A classification theory should therefore ask:
Do different star morphologies correlate with different textual patterns?
Different entry lengths?
Different positions in the quire?
Different recurring tokens?
Without such correlations, morphology is real and its meaning remains unresolved.
The Tail May Be More Structural Than the Star
Some markers appear as star-like heads attached to tails or stems.
This can make them look like flowers rather than stars.
Yale’s public description itself refers to “star-like flowers” in the final unillustrated section.
The ambiguity is useful.
It reminds us that even the noun “star” is a modern descriptive choice.
The maker may have thought of the marker as a flower.
A celestial sign.
An asterisk-like record marker.
Or an entirely conventional sign with no modern noun equivalent.
What survives all of those names is position.
The sign occupies the margin and coordinates with textual units.
Its document role is easier to observe than its pictorial identity.
One Star = One Entry Is Strong, but Not Perfect
Most of Quire 20 encourages a clean rule.
one marginal star → one short textual entry.
That is a useful first-order model.
Then the exceptions matter.
Some pages include longer running text alongside stars.
On f108v and f111, the simple short-entry morphology becomes less tidy.
On f116r, star markers do not all relate to running text in exactly the same way.
This does not destroy the entry-marker interpretation.
It refines it.
Perhaps stars mark records whose lengths vary.
Perhaps some text is heading or commentary outside the records.
Perhaps several record types coexist.
A robust model should explain the exceptions rather than quietly redefining them away.
Exceptions Are Where the Record Grammar May Reveal Itself
A perfect repetitive system tells us that a marker exists.
An imperfect system can tell us what the marker does.
Suppose a star appears without a normal entry.
Maybe the entry was omitted.
Maybe the star itself carries information.
Suppose long text spans several star positions.
Maybe stars mark subentries rather than paragraphs.
Suppose different star classes cluster around different entry lengths.
Maybe star morphology encodes record type.
These are examples of a broader method:
do not treat exceptions as noise before asking whether they distinguish the rule.
Quire 20 may be especially valuable because its repeated architecture makes exceptional cases visible.
Folio 58 Shows That Marginal Stars Are Not Unique to Quire 20
An earlier text-heavy folio also contains marginal star-like markers.
This is important because it weakens a simple equation:
star marker = Quire 20 recipe.
If the marker occurs outside the final section, it may belong to a wider manuscript convention.
Perhaps it means:
- entry start;
- special note;
- cross-reference;
- formula;
- record class;
- attention marker.
Then Quire 20 may be the section that uses the convention most intensively rather than the section that invents it.
This is a better cross-manuscript question than assuming all stars refer to astronomy.
Document conventions can travel across topics.
The same bullet can introduce a recipe, a note or a definition.
Entry Boundaries Give Us a Better Unit for Text Analysis
Much Voynich analysis uses pages, lines or space-delimited tokens as units.
Quire 20 offers another unusually explicit unit:
the star-marked record.
This is valuable because record boundaries may correspond to semantic units more closely than arbitrary page boundaries.
We can ask:
- Do entries share opening forms?
- Do they share closing forms?
- How does vocabulary repeat between entries?
- Are some entries near-duplicates?
- Do star morphologies correlate with textual clusters?
- Do entry lengths form distinct groups?
- Does physical bifolium structure affect vocabulary?
None of these questions requires knowing what the entries mean.
But any proposed meaning should eventually explain the answers.
The manuscript has effectively given us a segmentation annotation.
We should use it before inventing our own semantic boundaries.
Repeated Openings Could Reveal Formulaic Genre
If hundreds of entries belong to one genre, their openings may be formulaic.
Recipes might begin with equivalents of:
- for;
- take;
- against;
- when;
- to make;
- if.
Definitions might begin differently.
Astrological judgements differently again.
This means record-initial token distributions can discriminate genre hypotheses.
The same is true at entry endings.
A recipe may end with administration or duration.
An index entry may end with a reference.
A definition may end with a category statement.
Even without translation, repeated position can expose grammatical roles.
This makes Quire 20 a particularly promising place for structural linguistics.
Near-Duplicate Entries Would Mean Something Different From Repeated Words
Suppose two whole entries are structurally similar.
That is stronger than two entries merely sharing one common token.
Near-duplicate recipes could represent:
- variants of one remedy;
- same procedure for different ingredients;
- copying duplication;
- template-generated records;
- formulaic language.
A non-recipe record system can also generate near-duplicates.
The key is the level of repetition.
Entry-to-entry comparison may reveal hidden templates more clearly than manuscript-wide word frequency does.
This is another reason the star markers are valuable.
They give us candidate record boundaries before we know the record schema.
If These Are Recipes, Ingredients Should Connect Back to the Rest of the Book
A recipe interpretation gains enormous power if Quire 20 cross-references the herbal and pharmaceutical sections.
Imagine a recurring token appears beside a detached root in Quire 19.
The same token recurs inside several Quire 20 entries.
If the visual identification and textual relationship were established independently, the connection would become highly informative.
It could indicate an ingredient name.
A category.
A preparation.
A common grammatical word.
Further evidence would discriminate.
This is the kind of cross-manuscript relationship a genuine recipe theory should seek.
Not:
there are herbs earlier, therefore these must be herbal recipes.
But:
the same independently identified information unit recurs where a recipe model predicts it should.
If These Are Recipes, Quantities Should Form Their Own Class
Recipes need quantities more often than ordinary prose does.
A corpus of medical instructions may therefore contain repeated numeric, measure or dosage patterns.
Voynich numbers are not securely identified.
But a recipe hypothesis creates a testable expectation.
Some token class should behave like quantities or measures.
It may occur near ingredient-like terms.
It may occupy repeated syntactic positions.
It may be restricted in form.
Historical comparison with manuscripts containing doses and weights can help define what such positional behaviour might look like without assuming modern Arabic numerals.
A genre theory becomes scientifically useful when it predicts hidden grammatical classes.
Recipe theory has exactly that opportunity.
If These Are Recipes, Operations Should Repeat
Grind.
Boil.
Mix.
Drink.
Apply.
Wash.
Keep.
A medical recipe corpus should reuse a relatively small set of operations.
Those operations may generate repeated word-like families or positional patterns across many entries.
This offers another test.
Do Quire 20 entries show recurring internal templates consistent with operation words?
Could those same forms occur in pharmaceutical or bath contexts where an operation would make sense?
Again, no translation should be assigned before the distribution is established.
But the recipe hypothesis tells us what distribution to look for.
A Registry Hypothesis Makes Different Predictions
Suppose Quire 20 is not a recipe collection but a registry or catalogue.
Then the entries might record:
- name;
- category;
- location;
- status;
- cross-reference;
- one short property.
This model predicts more repeated entry structure and perhaps less action-oriented vocabulary.
It may predict strong links to labels elsewhere.
It may predict entries of more uniform length.
A recipe corpus and a registry can therefore fit the same star-paragraph morphology while diverging in internal textual behaviour.
This is the kind of competition we want.
Not:
recipe feels right.
But:
recipe and registry predict different distributions; which one does the corpus exhibit?
A Summary or Index Hypothesis Is Also Plausible Because This Is the End of the Book
Position at the end creates another possibility.
Perhaps the entries summarise or index earlier material.
This is attractive because the manuscript has already introduced plants, zodiacal systems, diagrams, human figures and pharmaceutical-looking components.
A final index could point backward.
A final summary could condense earlier knowledge.
A reference registry could connect modules.
But end position does not prove summary function.
The current binding is later.
Original sequence is not perfectly secure.
And a recipe collection naturally belongs at the end of a practical compendium too.
So the index hypothesis needs cross-references.
If many entries contain tokens strongly associated with earlier labelled objects, backward-reference function becomes more plausible.
Without such evidence, “final section = index” is another positional story.
Current End Position Is Not Guaranteed Original Function
This deserves emphasis.
We experience Quire 20 as the end of the surviving codex.
That can bias interpretation.
We expect conclusions at the end.
Indexes at the end.
Appendices at the end.
Recipes at the end of some compilations.
But the present codex has a later binding and missing leaves.
Physical position today is evidence.
Original functional position is an inference.
A strong theory should not need “it is at the end” as its main semantic support.
The Textual Regime Is Strongly B-Like but the Record Architecture Adds Another Dimension
Quire 20 is usually associated with Currier-B textual behaviour.
This links it statistically with other B-heavy material, including much of Quire 13.
It is tempting to use that as semantic evidence.
Quire 13 is medical, Quire 20 is B, therefore B is medical.
That is circular because Quire 13 itself is not securely decoded as medical.
What Quire 20 adds is a new independent variable:
record structure.
We can compare B-like text inside Quire 20 records with B-like running text elsewhere.
What changes because the genre changes?
What remains stable because the underlying writing system remains the same?
This is a more informative use of Currier classification than turning B into a subject label.
The Stars Give Us a Natural Corpus for Within-Section Comparison
Because the entries are repeatedly marked, Quire 20 can be treated as hundreds of comparable textual objects.
This is rare.
On a herbal page, it can be difficult to decide where one semantic unit ends.
In Quire 20, the margin repeatedly suggests an answer.
This lets us ask entry-level questions:
- How many tokens per entry?
- Which forms prefer first position?
- Which prefer last position?
- Which entries are unusually long?
- Which share rare words?
- Which belong to the same bifolium?
- Which star morphology accompanies them?
These measurements may reveal record classes even before their meanings are known.
A corpus of 324 surviving marked records is a gift to structural analysis.
The correct first response is to use the gift rather than rename all 324 entries as recipes.
Entry Length May Reveal More Than Page Length
A page is constrained by parchment size.
An entry is constrained by content or convention.
Therefore entry length may be a more semantically meaningful variable than page length.
If recipes have a typical structure, their lengths may cluster.
If there are several record types, multiple length families may appear.
If stars merely mark arbitrary notes, length may vary widely.
This is another way to make genre hypotheses compete without translation.
What appears visually as “short paragraphs” can become a quantitative distribution.
And distributions can expose subtypes that our eyes merge together.
Bifolium-Level Vocabulary Could Reveal Production Batches
The herbal section taught us that textual regimes can correlate with physical bifolia.
Quire 20 gives us another chance to connect record architecture with manufacture.
Do entries on one physical sheet share unusual vocabulary?
Star morphology?
Entry length?
Writing habits?
If yes, production batch may be influencing the record distribution.
If no, the entries may have been copied from a source whose organisation crossed physical sheets.
Either result is informative.
The important move is to stop treating record order, page order and physical production as one identical sequence.
A Cross-Reference Theory Would Make the Stars Addresses
What if the star is not merely a bullet?
What if variants of the star identify record classes or addresses?
Then the final pages could function as a lookup system.
A reader sees a plant label earlier.
That label points to a starred entry.
The entry contains a procedure or note.
This would make Quire 20 a textual back-end for the visual front-end of the manuscript.
It is an elegant architecture.
It also needs cross-reference evidence.
Do rare labels recur in Quire 20 disproportionately?
Do entry clusters correspond to earlier page families?
Do star variants correlate with earlier visual markers?
Without these links, “back-end” is metaphor.
With them, it could become a genuine document model.
The Starred Pages May Be a List of Outputs Rather Than Inputs
A grand medical pipeline often treats Quire 20 as the final recipe output.
Plant identified.
Ingredient prepared.
Body treated.
Zodiac time chosen.
Recipe recorded.
This makes Quire 20 the final output layer.
The manuscript does not establish that sequence.
Quire 20 could instead be:
- an input list;
- a summary;
- a parallel reference system;
- a registry assembled independently;
- a source corpus later used by other sections.
Current position at the end does not establish production direction.
This is why grand sequence theories need explicit cross-references rather than intuitive page order.
The Clean Plant → Jar → Body → Zodiac → Recipe Story Still Does Not Survive as a Proven Whole
Quire 20 makes the grand story more tempting than ever.
Here, at last, are the short instructions.
Surely these are the recipes generated by everything before them.
That is possible.
But the retained evidence still does not justify the entire pipeline as established architecture.
The links remain inferred.
Whole plant to fragment is not fully mapped.
Fragment to vessel is not a decoded operation.
Vessel to Quire 13 treatment is not demonstrated.
Zodiac to treatment timing is not demonstrated.
Quire 20 entry to recipe is not translated.
A chain is only as strong as its links.
The useful response is not to abandon the medical-compendium hypothesis.
It is to test the links separately.
A Recipe Theory Can Become Much Stronger Than It Is Today
Unlike many broad semantic hypotheses, recipe theory can generate detailed predictions.
- entry openings should be formulaic;
- ingredient-like tokens should recur across entries;
- quantity-like forms should occupy restricted positions;
- operation-like forms should recur in similar syntactic slots;
- entry endings may contain administration or condition formulae;
- some Quire 20 tokens should connect to labelled plant or pharmaceutical units elsewhere.
This is good news.
It means “recipes” need not remain a vague visual nickname forever.
It can become a model that risks failure.
If those predicted structures do not exist, the theory weakens.
If they do, and independent decipherment begins assigning historically coherent roles, the theory strengthens rapidly.
That is how a nickname can become scholarship.
What Survived From the Starred-Page Work
- Quire 20 is an explicit record-segmentation system.
- 324 marginal star-like markers survive.
- The original total was likely larger because a central bifolium is missing.
- Most star markers align with short textual units strongly enough to support entry-boundary interpretation.
- Star morphology varies and deserves classification.
- The stars are more securely document markers than literal celestial objects in this context.
- The one-star-one-short-entry model is strong but has meaningful exceptions.
- Earlier marginal-star use shows the convention is not unique to Quire 20.
- The star-marked record is a useful analysis unit independent of page and line.
- Historical medical recipe collections make recipe interpretation plausible.
- Recipe semantics remain unproven because no accepted reading recovers ingredients, quantities and operations from the entries.
- The section may function as recipe collection, registry, reference system, summary, formula corpus or another record architecture.
- Its greatest current value is that the manuscript itself exposes repeated boundaries that future textual models must explain.
This is one of the strongest structural gifts in the Voynich Manuscript.
We know where hundreds of units probably begin before we know what any one unit says.
What Did Not Survive as Established Knowledge
- A claim that there are exactly 324 original records.
- A claim that all 324 surviving markers correspond to medical recipes.
- A claim that every star represents a celestial object.
- A stable semantic legend for seven-point, eight-point, tailed, dotted or coloured stars.
- A perfect one-star-one-paragraph rule without exceptions.
- A proof that Quire 20 is the final output stage of a medical pipeline.
- A proof that the entries summarise earlier sections.
- A proof that the star convention has one identical meaning everywhere it appears.
- A deciphered set of ingredients, doses or operations.
- A demonstrated direct cross-reference system linking every record to earlier plant or vessel pages.
Keeping those claims out of the “known” column makes the actual known structure more valuable, not less.
A Better Quire 20 Analysis
1. Reconstruct the physical quire
Keep the missing central bifolium visible and avoid treating the surviving sequence as complete.
2. Treat every star-marked unit as a candidate record
Do this before naming the genre.
3. Classify marker morphology
Point count, tail, dot, colour and placement may or may not carry metadata.
4. Measure record structure
Length, opening forms, closing forms, recurrent tokens and near-duplicates.
5. Preserve exceptions
Long-running text and unusual star/text relations may expose subtypes.
6. Compare genre models
Recipe, registry, index, formula collection, summary and note systems should make different predictions.
7. Search for independent cross-references
Compare rare tokens with labels in herbal, zodiac, Quire 13 and pharmaceutical contexts.
8. Require held-out prediction
A proposed record grammar should predict an entry not used to build it.
Quire 20 is unusually suited to this kind of analysis because the manuscript has already done much of the segmentation work for us.
What Would Make the Recipe Theory Strong?
- Star-marked records show recurring formulaic openings.
- Distinct token classes behave like ingredients, quantities and operations.
- Rare ingredient-like forms cross-reference independently identified plant or pharmaceutical labels.
- Entry syntax remains stable across many records.
- Recipe variants produce understandable near-duplicate structures.
- Historical medical recipe conventions provide appropriate analogues.
- The model explains unusual long entries and marker exceptions.
- The grammar predicts unseen records.
- Independent researchers can reproduce the analysis without discretionary translation.
If those properties accumulate, “recipe” moves from visual nickname toward functional identification.
Until then, it remains a strong hypothesis rather than a reading.
What Would Make a Registry or Index Theory Strong?
- Entries have highly regular field-like structure.
- Rare labels from earlier visual sections recur disproportionately.
- Entries contain reference-like rather than operation-like forms.
- Star variants classify records consistently.
- Record order follows a recoverable sorting principle.
- Cross-references predict earlier folio classes or labelled components.
- The model explains why the corpus sits where it does physically without relying only on end position.
A registry model can therefore compete directly with a recipe model.
The manuscript should decide between them.
What Would Count as a Real Quire 20 Breakthrough?
Imagine that several independently identified pharmaceutical labels recur inside star-marked entries.
Their positions are consistent.
A second recurring token class occupies the position expected for quantity.
A third class behaves like operations.
The resulting grammar predicts an unseen entry.
The prediction produces a plant or process relationship independently supported elsewhere in the manuscript.
That would be an extraordinary result.
Or imagine instead that entries resolve into repeated reference fields that systematically point backward to earlier labelled units.
That would overturn the recipe interpretation and reveal a registry.
Either outcome would be progress because the record architecture is strong enough to make the competing models testable.
The breakthrough is not “we found a plausible translation”.
The breakthrough is when the same record grammar predicts meaning beyond the entries used to invent it.
Primary School: A Bullet Point Does Not Tell You the Topic
Show a child three lists.
One is a shopping list.
One is homework.
One is a recipe.
All use bullet points.
Ask:
Does the bullet symbol tell you which kind of list it is?
No.
It tells you where units begin.
The words tell you the topic.
That is the Quire 20 problem in one classroom exercise.
Lower Secondary: Same Marker, Different Genre
Give students four lists that all use stars as bullets.
- ingredients;
- people;
- rules;
- places.
Hide the words.
The star structure remains identical.
Ask what can still be known.
Number of entries.
Entry boundaries.
Relative lengths.
Maybe subgroups.
Not genre.
The learner sees how much document structure can be recovered without semantic reading—and where structure stops.
Upper Secondary: Compare Recipe and Registry Predictions
Give students two models.
Recipe: entries contain ingredients, quantities and operations.
Registry: entries contain identities, categories and references.
Ask what textual distributions each predicts.
Recipe predicts operation-like repetition and quantity-like slots.
Registry predicts field regularity and stronger links to labels.
Now the student is no longer guessing the genre.
They are designing a discriminator.
That is the core of serious model comparison.
JC and Adult Readers: Quire 20 Is a Natural Record-Level Dataset
At a higher level, Quire 20 solves a problem researchers often have to solve themselves.
Segmentation.
Where is one semantic record?
Where does the next begin?
The star markers provide a strong answer for hundreds of units.
This allows modelling at record scale rather than only character, token, line or page scale.
That matters because many genres are defined by record grammar.
A recipe has fields.
A catalogue entry has fields.
A diagnosis note has fields.
Comparing distributions across 324 surviving candidate records may therefore be more discriminating than comparing the manuscript as one giant text stream.
The page has already tokenised one level of its own architecture for us.
A Parent and Teacher Guide
The starred pages are an excellent way to teach the difference between structure and content.
- Find the boundaries. What visual sign tells you a new record begins?
- Count surviving records carefully. Do not turn a surviving count into an original designed total when pages are missing.
- Separate marker shape from marker meaning. A star-looking bullet need not be astronomical.
- Preserve exceptions. Unusual entries may reveal record subtypes.
- Compare genre hypotheses. Recipe, registry, index and note systems predict different internal structures.
- Search for cross-references. A real recipe system should reconnect with substances or labels elsewhere.
- Require prediction. A proposed grammar should work on an unseen entry.
This is not only medieval-manuscript reasoning.
It is how students learn to distinguish formatting from meaning in databases, forms, tables and modern documents.
Reader Checklist: Before You Call Quire 20 a Recipe Book
- Are you counting surviving stars or assuming an original total?
- Have you accounted for the missing central bifolium?
- Is “star” a visual description or a celestial interpretation?
- Does each marker truly correspond to one record?
- How are exceptions such as long-running text handled?
- Does star morphology correlate with record type?
- Does the same marker occur outside Quire 20?
- What internal grammar specifically supports recipe genre?
- Are ingredient-like, quantity-like and operation-like classes detectable?
- Do candidate ingredients cross-reference earlier labels independently?
- Could registry, index, summary or formula models explain the same segmentation?
- Is current end position being mistaken for original functional sequence?
- Does the theory predict an unseen record?
- What observation would make recipe theory fail?
Frequently Asked Questions
What are the Voynich starred pages?
They are the text-heavy pages of the final major gathering, Quire 20, where hundreds of star-like marginal markers divide or accompany short textual records.
How many stars survive?
Detailed folio analysis counts 324 surviving marginal star-like markers in Quire 20. Because a bifolium is missing, this should not be treated as the original total.
Are there 324 recipes?
Not established. There are 324 surviving star-like markers associated with entry-like text. “Recipe” is the conventional interpretation of the record genre, not a deciphered count of recipes.
Do the stars mean astronomy?
Not necessarily. In Quire 20 their repeated marginal position strongly supports a document-marker function. They may still carry additional semantics, but celestial meaning is not established.
Why are they called recipes?
The short segmented entries resemble recipe collections in practical medical manuscripts, especially given the Voynich Manuscript’s plant and pharmaceutical-looking pages. No accepted translation has confirmed the genre.
Are all entries the same length?
No. Many are short, but the section includes longer-running text and unusual star/text relationships. Those exceptions may reveal multiple record types or more complex marker rules.
Do stars appear elsewhere in the manuscript?
Yes. Star-like marginal markers also occur outside Quire 20, including on f58r, showing that the convention is not exclusive to the final section.
Could Quire 20 be an index rather than recipes?
Yes. Registry, index, formula collection, summary and note-system interpretations remain plausible. They can be tested against different predictions about record grammar and cross-references.
What is the strongest conclusion?
Quire 20 is a highly structured record system whose repeated marginal markers expose hundreds of textual boundaries. The semantics of those records remain undeciphered.
Related eduKateSG Reading
- Voynich | Everything eduKate Knows and Tested | The Vessels and Plant Fragments
- Voynich | Everything eduKate Knows and Tested | The Zodiac Pages
- Voynich | Everything eduKate Knows and Tested | What the Writing Does Before We Know What It Says
- Voynich | Everything eduKate Knows and Tested | Currier A and B
Research and Further Reading
- Yale Beinecke Library — The Beinecke Cipher (Voynich) Manuscript
- René Zandbergen — Quire 20 Structure and Starred Text
- René Zandbergen — Voynich Illustration and Marginal-Star Notes
- British Library — Harley MS 2375, Medical and Astrological Texts and Recipes
- British Library — Harley MS 2390, Fifteenth-Century Medical Compendium
- British Library — Egerton MS 747
The Final Idea
The final pages of the Voynich Manuscript are strangely generous.
They do not tell us what they say.
But they tell us how they are divided.
Star.
Record.
Star.
Record.
Hundreds of times.
That may be one of the clearest surviving pieces of the manuscript’s original information architecture.
The mistake would be to waste that clarity by naming the records too quickly.
Recipe?
Perhaps.
Registry?
Perhaps.
Index?
Perhaps.
The stronger thing we know comes first:
the maker wanted the reader to see hundreds of separate units.
Once we recover what one unit does, the other 323 surviving markers may stop looking like decoration and start looking like a dataset.
Continue Through the Voynich Research Map
This article is one specialist node in eduKateSG’s larger Voynich research library. Return to the canonical master to see the physical, visual, textual and historical evidence in one continuous argument.
Structure is evidence; structure is not translation. The master keeps resemblance, statistical structure, historical possibility and decipherment claims at separate evidentiary levels.
Next Best Routes
- Labels Versus Running Text — compare explicit record markers with shorter object-attached text.
- The Vessels and Plant Fragments — move from modular visual units into modular text records.
- Exact Repetition — test how repeated entry structure differs from repeated tokens and phrases.
Deep bridge: The Museum Label Is a Tiny Piece of Scholarship helps separate compact document structure from the semantic job the compact text eventually performs.