eduKateSG · TANGENTIAL VOYNICH · THE OBSERVER RESET
Perhaps the next useful move is not another explanation. Perhaps it is learning how to look again without quietly carrying every previous explanation into the room.
Voynich Research Library · The Tangential Method · The Mathematics and Science
Put the Voynich Manuscript on an imaginary table. Around it, place everything we have already brought to the investigation: medical books, astronomical diagrams, manuscript production, language statistics, places, people, copying traditions, inventories, machines, networks and the many useful worlds that the Tangential series has deliberately borrowed.
Now ask a slightly uncomfortable question. How much of what we see belongs to the manuscript, and how much belongs to the furniture we have arranged around it?
The question does not invalidate the earlier work. It makes that work accountable. A comparison can teach us something and still become an obstruction when we forget that it was a comparison. A promising hypothesis can organise a search and then quietly begin choosing what the search is allowed to notice. An unfamiliar mark can acquire a familiar name; the name can acquire a function; the function can acquire a birthplace. Eventually, a chain of suggestions can feel like a chain of evidence.
This continuation proposes a paired correction. First, look with an imagined alien eye: not because extraterrestrials made the manuscript, but because an outsider is a useful way to interrupt our automatic recognition. Then look with a historically informed period eye: not because we can become a person from the fifteenth century, but because the manuscript should not have to conform to the visual expectations of our own.
The two movements pull in opposite directions. One removes familiarity. The other restores a different familiarity. Between them lies a more disciplined form of curiosity.
Reset the claim, not the evidence. Keep the observations, sources, failed tests and unresolved questions. Suspend the inherited explanation long enough to discover which parts of it the evidence can actually rebuild.
This is a research-method essay and a historically grounded comparison, not a new decipherment or a report of a newly completed experiment. It revisits the public research architecture and selected recorded checkpoints; it does not claim that every published article, transcription or manuscript image has been independently re-audited for this edition. The exercises below are proposals or explicitly constructed examples. None is presented as a newly observed Voynich result.
Reading route: Reset without amnesia · The alien eye · The clue mechanism · The period eye, 1350–1450 · Return to the object · Testing the reset · Sources and continuations
I. Reset Without Amnesia
Zero is a working position, not an empty mind
There is a careless version of starting again. It declares that previous research has become too complicated, sweeps the table clean, and allows the next attractive idea to enjoy the privileges of innocence. That is not a reset. It is the loss of an audit trail.
A serious reset does almost the opposite. It preserves the earlier record so carefully that an old mistake cannot return disguised as a new discovery. What it removes is the automatic authority of the current interpretation. The manuscript remains the same object. The photographs remain the same photographs. A recorded negative result remains a negative result. What changes is the permission we give ourselves to reopen a question that the accumulated narrative had made seem closed.
Zero therefore means a deliberately modest starting statement: this is the feature that can be observed through this source, at this scale, under these limitations. It does not mean that all explanations are equally plausible. It does not mean that established material evidence should be discarded. It does not mean that a speculative origin story deserves the same evidential standing as a documented physical observation.
We are reducing unearned interpretation, not reducing knowledge. The difference matters because the most seductive fresh starts often retain the favourite conclusion while throwing away the controls that made it difficult to defend.
Imagine a researcher who has come to think of a particular illustrated form as a root. The reset does not order the researcher to forget that roots exist. It asks for a second description that does not depend on the identification: a thick central form, several branching extensions, a particular attachment pattern, an outline of uncertain completeness. The word root can remain in the hypothesis column. It simply stops doing the work of an observation.
That small separation is the whole movement in miniature. What we know about the world may generate a candidate. What we can establish about the object determines whether the candidate advances.
What the earlier investigation must be allowed to keep
The public Voynich Research Library separates several questions that are easily flattened into one: the physical object, the writing system, the visual programme, the page as an interface, historical comparators, localisation and the conditions under which an interpretation should fail. That separation is not administrative decoration. It prevents success in one room from being reported as success in every room.
The recorded segmentation work offers a useful example. Its surviving lesson was not that one set of familiar section names had finally captured the manuscript. It was that physical, scribal, textual, visual and geometrical divisions need not coincide. A boundary can matter on one axis without defining a complete semantic province on every other axis. The public Hand × Bifolio × Section treatment explains why those coordinates should remain distinct.
A reset must preserve that restraint. We should not return to zero and immediately redraw the manuscript as a single programme merely because a new metaphor makes the arrangement look elegant. Nor should we promote an entry-like layout into an inventory, a recurring component into an ingredient, or a statistical boundary effect into a translated operator. Those are separate claims with separate burdens.
The same applies to place. The Padua programme is useful precisely because it distinguishes a historical environment capable of producing certain kinds of knowledge from evidence that this particular codex was produced there. A text can travel. A trained scribe can travel. A picture tradition can travel. The place where a book was assembled need not be the place where its exemplar was made or its later owner learned to read it. The Coordinate Stack keeps these possibilities from collapsing into one address.
Nothing about the alien eye should dissolve those distinctions. Nothing about period fashion should bypass them. This article adds a way to examine the observer; it does not reopen every previously withheld claim as though the withholding never happened.
The parallax trap
In this essay, parallax is a methodological metaphor. We are interested in how an apparent relation changes when the observer changes position. We are not claiming that interpretive bias is literally an optical displacement, or that moving between metaphors supplies the mathematical guarantees of a calibrated physical measurement.
The useful comparison is this: the object can remain unchanged while its apparent organisation changes with the standpoint from which it is inspected. To a reader searching for botany, branching forms become salient. To a reader searching for machinery, connections become salient. To a reader searching for ritual, repetition becomes salient. The manuscript has not acquired three new sets of marks. Three different selection rules have been applied to the marks already there.
That movement can be productive. The Tangential manifesto deliberately uses incompatible systems to produce questions that a single familiar frame might never ask. But the movement becomes a trap when the new arrangement is mistaken for newly acquired evidence. A diagram does not become more likely to be hydraulic simply because a hydraulic vocabulary makes it easy to describe.
There are two mistakes to avoid. One is refusing to move at all, allowing the first plausible interpretation to define the whole search. The other is moving constantly and treating the number of available interpretations as progress. A gallery of imaginative explanations can grow indefinitely while the evidential state remains unchanged.
The corrective question is not merely, “What can I see from here?” It is, “Which measurable distinction becomes available from here, and does it survive when I leave?” A useful tangent returns with a test. A decorative tangent returns with another vocabulary.
How one clue becomes five apparent confirmations
Consider a constructed chain. A form resembles a root. The root suggests a plant. The plant suggests a herbal. The herbal suggests medical practice. Medical practice suggests a particular city with a strong medical culture. It is tempting to count five mutually reinforcing observations. In fact, the chain may contain one visual resemblance and four dependent inferences.
Searching within the suggested city then produces a medical manuscript with similar imagery. The resemblance now seems to have geographical support. Yet the geographical search was selected by the original resemblance. Unless the alternative environments have been given a fair search, the new result may tell us more about the route taken through the archive than about the origin of the object.
This is why a record of discovery matters. Which feature was noticed first? Which label generated the next search? Was the comparator selected because it was independently appropriate, or because it illustrated the preferred explanation? Could the same search procedure have produced an equally persuasive case elsewhere?
The estate already addresses this problem in The Evidence Independence Problem. The observer reset makes the dependency visible before the next round begins. We do not delete the chain. We mark which links are observations, which are interpretations, and which depend on an earlier choice.
A claim should not gain weight merely because it has acquired more sentences.
II. The Alien Eye
“It is aliens” can mean two very different things
The phrase can name a literal origin hypothesis: nonhuman visitors produced the manuscript or supplied its contents. It can also name a thought experiment: let an imagined observer, unfamiliar with our inherited categories, describe the object before being told what humans think it resembles.
These uses must not be allowed to exchange credentials. The second may be a useful intellectual exercise. Its usefulness supplies no evidence for the first. A fictional outsider can help us notice the assumptions inside the word plant without making the depicted form extraterrestrial.
This article does not present evidence for alien authorship. Its alien is an intentionally imagined observer, not a proposed historical participant. The literal hypothesis remains unsupported here. That is not a declaration that curiosity must stop at familiar explanations. It is a statement about the evidence actually being offered.
The distinction also protects the more interesting question. We do not need an extraordinary maker to explain why a modern reader might misunderstand an old representation. We only need a difference between the conventions the maker expected and the conventions the reader supplies. The Tangential series has already explored that missing connection in The Original Receiver Is Missing.
A reader can be culturally alien to a human-made object. That is enough to make the exercise worthwhile.
An explanation that can imitate everything distinguishes nothing
Suppose the literal theory is allowed to say that an advanced visitor could make any material, imitate any historical style, encode any language, leave no distinctive trace, and deliberately create every apparent inconsistency. Such a theory can accommodate almost any possible observation. That flexibility is not automatically explanatory strength.
For a hypothesis to help discriminate, it must make some outcomes more expected than others in a way its competitors do not. What should we find under the proposed account that an ordinary human production history would not also lead us to expect? What observation would count against the account? What restriction prevents a disappointing result from being explained away by adding another invisible capability?
The problem is not unique to extraordinary theories. A human cipher hypothesis can become equally evasive when every inconvenient token is declared a null, every failed reading an abbreviation, and every exception a new rule. A workshop hypothesis can become equally protected when every mismatched feature is attributed to a travelling specialist. Unbounded rescue is a methodological problem regardless of the story’s atmosphere.
Nor does an unexplained feature automatically favour an extraordinary origin. It establishes a gap in our account. Moving from that gap to a specific cause requires additional evidence. “We cannot presently identify this form” and “this form depicts a nonterrestrial organism” are not neighbouring grades of the same statement. They are different claims.
The alien tangent is therefore strongest when it disciplines the observer rather than expands the producer’s imagined powers. We deliberately borrow the stranger’s lack of recognition. We do not borrow unlimited explanatory freedom.
Give the outsider a smaller vocabulary
Imagine asking the outsider to inspect a bounded image region. Do not begin with “What species is this?” Begin with questions that can be answered without an object name. Where do contours meet? Which elements repeat? Which features are attached and which merely overlap? Is a line continuous at the available resolution? Which arrangements change when the region is viewed with its surrounding page?
The answer might describe an axial form, a branching event, a terminal shape, a band, an enclosure or an attachment relation. Such descriptions are not perfectly theory-free. Choosing an axis, deciding where a component begins, and treating two shapes as comparable are already analytical decisions. Neutrality here is a practical reduction in semantic commitment, not a claim to perception without assumptions.
That reduction is still valuable. “Three extensions leave a central region” can be checked separately from “three roots emerge from a bulb.” The second sentence may eventually become defensible. But the first gives a second observer a better chance to agree or disagree without first accepting a botanical interpretation.
The public Components Before Objects branch develops this ordering. First ask whether a recurring visual component can be recovered reproducibly. Only later ask what the component represents. Recurrence can be a structural result even when identity remains open.
The imagined outsider is therefore not asked to be more brilliant than a historian. It is asked to perform a narrower job. Its temporary ignorance is useful only when paired with disciplined description.
The alien must not secretly become a modern engineer
There is an amusing failure mode in this exercise. We remove the medieval plant name and replace it with “sensor array,” “fluid manifold,” “data port” or “control node.” The vocabulary sounds less familiar to the manuscript, so the description feels more neutral. It is not. We have exchanged one specialised world for another.
A tube-like outline is not automatically a conduit. A repeating shape is not automatically a symbol with a stable semantic value. A central form with radiating branches is not automatically an information hub. These may be productive models, but their nouns bring functions with them.
The same danger appears when the outsider assumes that a useful representation must minimise redundancy, that a well-designed diagram must use consistent scale, or that a sophisticated maker must avoid ambiguity. Those are expectations about design, not observations of this object. Even an apparently universal word such as efficient needs a specified task and a specified user.
The remedy is to keep two records. One contains the low-commitment description. The other contains the borrowed interpretation and the prediction it generated. If the interpretation says that one form acts as a connector, the next question is not whether it looks connector-like. It is whether its placement and relationships distinguish the connector model from alternatives.
Remove the borrowed world at the end. Can the proposed observation still be stated without the language of machinery, botany, ritual or software? If not, the tangent has not yet returned with a result that belongs to the manuscript.
What the alien eye is allowed to accomplish
Its first achievement may be nothing more dramatic than a better question. Perhaps the feature we had been naming is actually a relation between two components. Perhaps the apparent object boundary was created by a crop. Perhaps a repeated element deserves to be measured across several page types rather than identified inside one striking image.
None of these possibilities should be reported as a discovery until it has been checked. But the exercise can improve the design of that check. It can expose an assumption that previously had no label and therefore no obvious place in the failure conditions.
A second achievement is a cleaner disagreement. Two researchers may discover that they agree about the visible geometry but disagree about what it depicts. That is progress in the organisation of the question. There is no need to force agreement at the semantic level when the disagreement can be located precisely.
A third achievement is a justified stop. An image may simply be inadequate for the proposed distinction. Calling the mark a plant, a machine or an alien organism does not restore the missing detail. The correct output can be “not distinguishable in this representation.” The outsider has done useful work when it helps us reach that boundary honestly.
III. From Recognition to Discrimination
A clue should narrow the right question
The central mechanism is simple to state and demanding to apply: observe a feature, describe it independently of the preferred explanation, decide which scale of question it can address, generate plausible alternatives, and seek the feature that would distinguish those alternatives. Commit only at the scale the evidence supports. Otherwise, hold.
Scale is not a minor detail. A clue about how one glyph was formed need not identify the language. A clue about the language, even if securely established, need not identify the place where this copy was produced. A clue about a diagram’s source tradition need not date the parchment. Each step to a broader or more precise claim needs its own connecting evidence.
Imagine a feature that separates illustrated manuscripts from unillustrated account books. It may perform that broad job well while being useless for distinguishing two neighbouring illustrated traditions. Its success at the first task must not be transferred to the second. A good broad classifier can still be a poor localiser.
This is the precision cliff. A researcher has genuine evidence for a broad category and then crosses into a narrow identification without noticing that the evidence stopped at the edge. The result can sound especially persuasive because the first half of the reasoning was sound.
A disciplined clue record therefore includes a ceiling: this feature may support a question about layout, production practice or comparison class; it does not yet support an answer about author, place, date or meaning. The ceiling is not an apology. It is part of the clue.
The nearest lookalike matters more than the easiest alternative
A theory does not become strong merely because it outperforms a weak rival. A manuscript with regular text and diagrams will naturally look unlike a random arrangement of unrelated marks. That comparison may establish something useful about organisation. It does not tell us which organised production process generated it.
The more demanding test admits the nearest plausible confusions. If a feature is proposed as evidence for a medical book, compare it with other books that also contain bodies, calendars, diagrams or plant imagery. If a feature is proposed as regional, compare it with neighbouring and mobile traditions capable of carrying the same feature. If a token pattern is proposed as linguistic, test it against constrained nonlinguistic or encoded alternatives appropriate to the particular claim.
The aim is not to multiply alternatives forever. It is to make sure the preferred explanation wins a meaningful contest. The best next observation is often the one on which the leading candidates disagree, not the one that adds another attractive resemblance to the leader.
Ask what would reverse the ranking. If there is no conceivable observation that would do so, the ranking may be a preference rather than an evidential conclusion. The public Expected Evidence Problem makes this a concrete discipline: an explanation should tell us not only what fits, but what we ought to encounter if it is doing the work we assign to it.
Confidence needs a place to be corrected
In a known-answer exercise, an identification can be compared with a withheld answer. The observer can discover that a supposedly distinctive feature is common elsewhere, that a memorable success was lucky, or that a broad judgement was reliable while an exact one was not. Repeated correction creates an opportunity for calibration.
Voynich does not provide that same answer reveal for its disputed meaning and origin. Consequently, confidence learned in a task with frequent external correction cannot simply be carried into the manuscript. The transferable part is the discipline of observation, competing candidates, discriminators, failure recording and abstention. The confidence itself must be earned again.
This is why known manuscripts are useful before the unknown one. Hide the catalogue identity of a suitable control, state the question at a defensible scale, and test whether a proposed feature actually helps recover the withheld information. The public Calibration Problem and Blind Localisation Test describe this direction of travel.
Even then, a control success is not a Voynich identification. It shows that a method can perform a specified job on the tested class of material. Transfer requires another argument: is the unknown object sufficiently comparable, and are the same relevant features actually available?
IV. The Period Eye: What Was Fashionable Around 1400?
Begin with the right century, but do not pretend it had one face
The proposed envelope is approximately 1350–1450: fifty years on either side of 1400. That is a comparison window, not a claim that every part of Voynich was made within it. It is wide enough to include earlier models, nearby practices and plausible continuities, but also wide enough to contain substantial differences. A century is not one moment.
Nor is a period eye a universal medieval consciousness. A courtly reader, a university-trained reader, a workshop practitioner, a devotional reader and a person with limited access to books need not bring the same expectations. Their different training and uses are questions to investigate, not characters whose inner lives we can confidently invent.
For this essay, the historical examples are a deliberately bounded group of western European manuscript witnesses, especially French and northern Italian material near the turn of the fifteenth century. They are not a census of Europe, still less of the whole world. They supply concrete counterweights to modern assumptions. Their inclusion does not establish the geographical limits of Voynich’s origin.
The practical question is therefore more precise than “What did people think in 1400?” It is: which expectations can we document for which readers, objects and settings, and which of those expectations would change the way we formulate a test?
Historical imagination is welcome at the entrance. At the evidence desk, it needs an object, a source and a boundary.
Keep the parchment clock separate from the writing clock
The often-cited 1404–1438 range concerns the parchment. A particularly useful source is not merely the conference report that repeated the result, but a correction to that report. On 1 June 2012, Yale conservator Paula Zyats clarified, on the American Institute for Conservation’s blog, that Gregory Hodgins’s radiocarbon work gave 1404–1438 at the reported 95% level and 1411–1430 at the reported 68% level. The original report on the same page had incorrectly described a 1450s date. Read Zyats’s correction beneath the report, rather than reproducing its uncorrected sentence.
That distinction is a small demonstration of the method. The authority of the hosting institution does not remove the need to identify who made a claim, which part of the page contains it, and whether it was corrected. A source is not one undifferentiated block of certainty.
More importantly, a radiocarbon range for the material is not an independently measured date for the act of writing. The University of Arizona’s contemporary account also distinguishes dating the parchment from the difficulty of directly dating the ink in this case. Its release is available through ScienceDaily’s reproduction of the university announcement. This is a report of the same research, not a second independent dating experiment.
A manuscript can carry several clocks: the biological material, its preparation, the model from which an image was copied, the copying itself, the addition of colour, assembly, later annotations and subsequent binding. We should not assign all of them the same interval because one layer has been measured. In particular, 1438 must not silently become a proved latest possible writing date.
The period eye uses chronological evidence to structure comparison. It does not squeeze every layer of the object into a single birthday.
Fashion includes clothing, but the clue is in the detail
There is a direct answer to the question about fashionable appearance. Getty’s The Medieval Bookshelf exhibition identifies a French Romance of the Rose illustration, Music in the Garden, as about 1405. Its discussion specifically notes long hanging sleeves and hems sweeping the ground as features of fashionable early-fifteenth-century dress. That is a dated example tied to an identified object, not a generic image labelled “medieval.” See Getty’s object discussion.
The methodological value lies in that specificity. A long garment alone is a broad resemblance. The relation between sleeve, body, hem, layering, setting and depicted activity may support a more useful comparison. Even then, the comparison concerns a depicted convention in a particular witness. It is not automatically a precise date for another drawing.
If a Voynich figure seems oddly proportioned or oddly dressed to us, a period-informed question would ask what the drawing is doing before calling it an error or an anomaly. Is the relevant feature a garment, a head covering, hair, a compositional shorthand, a damaged outline or something not resolvable in the available image? We should not decide between those possibilities merely because one interpretation is visually memorable.
Nor should we use a courtly example as the uniform of a continent. The fashionable appearance in one luxurious illustration does not tell us what every worker, scholar or household member wore. A period eye becomes more accurate when it becomes more differentiated.
The next useful question is not “Does this look old?” It is “Which documented convention, among its nearest alternatives, does this particular detail actually distinguish?”
The picture is not a wardrobe photograph
Getty’s broader exhibition on fashion in manuscript art emphasises that clothing can communicate status and identity while also being idealised, fanciful or shaped by the story being illustrated. That framework is useful, but its featured objects span different dates, including examples outside our 1350–1450 envelope. It should be used as an interpretive warning, not treated as a uniformly period-matched comparison set. See Fashion in the Middle Ages.
From that warning follows a crucial distinction. We can study the date of a depiction, the date of the fashion it appears to reference, the setting of the story, and the date of the surviving copy. These need not be identical. A copied image can preserve a convention older than the material on which it now appears.
We also need to ask what the genre permits. A didactic figure may omit details irrelevant to its task. A symbolic figure may exaggerate the feature that makes its role recognisable. An illustration may mix recognisable objects with an arrangement that is not intended as a literal scene. These are candidate explanations to test against known examples, not automatic excuses for every mismatch.
This is where historical familiarity should make us more exacting rather than more permissive. “Medieval artists were different” is too loose to explain anything. The stronger claim specifies which practice is documented, in which material, and how it predicts the particular difference under discussion.
Period context should reduce the number of unsupported assumptions. It should not become an unlimited licence to explain them away.
A beautiful book could also be a working object
The Metropolitan Museum of Art dates the Belles Heures of Jean de France, duc de Berry, to 1405–1408/1409, probably in Paris. The Limbourg brothers made this private devotional manuscript for the duke. The museum records tempera, gold and ink on vellum. These details establish a concrete early-fifteenth-century case in which elaborate visual production and a defined devotional purpose belong to the same object. See the Met’s collection record.
The inference for our method is limited but important. We should not force “beautiful,” “prestigious” and “useful” into mutually exclusive boxes before examining the evidence. An object can carry several kinds of value. Its visual investment may help explain ownership, attention or use without exhausting its purpose.
Conversely, an object’s uneven finish does not by itself tell us that its maker was incompetent, fraudulent or unconcerned with function. To make that argument, we would need to know which features mattered for the proposed task and which production standards are appropriate to the comparison. Judging a working representation by the finish of an elite commission can manufacture an anomaly.
The existing Courtly Luxury Object tangent explores the social possibilities. The present article does not claim that Voynich belongs to that category. It uses the known object to break a modern false choice: either a book is a tool or it is a display. Before deciding, ask what forms of use could coexist and what traces would distinguish them.
The fashion of owning and using books
When we widen the word fashion, we need to keep its meanings clear. A practice documented in one prestigious object is not necessarily widespread. A visual convention is not automatically a short-lived craze. Here, the useful broader question concerns the choices through which books could become desirable, recognisable or appropriate to their users.
The Met’s account of medieval books of hours describes prayer books for lay use, organised around devotional texts and often accompanied by calendars and other materials. It also discusses the role of illustrations and the variety of individual examples. The early-fifteenth-century Belles Heures provides our period anchor; the broader essay supplies a context that spans a longer history. See The Book of Hours in Medieval Christianity.
For Voynich, this suggests a question about the relationship between visible organisation and an expected user. A book may not explain every convention from first principles. Its reader may already know how to enter, navigate and interpret part of the surface. That is a possibility, not evidence that Voynich had a specific devotional or professional audience.
The modern reader’s expectation of an explicit title, a contents page, a legend and a self-contained explanation should therefore be treated as an expectation. Their apparent absence needs comparison with appropriately chosen objects. It cannot, by itself, prove that a book is meaningless or deliberately deceptive.
At the same time, the missing reader must not become an invisible helper who can explain everything. A proposed user should create constraints. What prior knowledge would that user need? Which conventions would have to be stable? What would the object make easier? Which features would remain inexplicable even after the user is imagined?
One period joke breaks a surprisingly large assumption
The Met’s account of Jean de Berry’s patronage records an especially useful counterexample: the Limbourg brothers gave the duke a mock book made from a decorated block of wood. It looked like a book without supplying an ordinary book’s pages of readable content. The episode belongs to the duke’s early-fifteenth-century collecting world, not to a modern conceptual-art thought experiment. See Patronage of Jean de Berry.
Its relevance is not that Voynich must be a joke. That conclusion would repeat the very error we are trying to prevent. The counterexample has a smaller job: it defeats the universal assumption that an object with the form and social presentation of a book must derive all its value from conventional reading.
A counterexample to a universal claim is not positive identification of an unknown object. Once the universal claim has been weakened, several possibilities reopen. Each still needs evidence. An object might be read, displayed, exchanged, demonstrated, treasured, joked about or used in more than one way. The list is a candidate space, not a verdict.
The discipline is worth stating plainly. A historical curiosity may show that a possibility existed. It does not show that the possibility is common, likely in this case, or sufficient to explain the whole manuscript. The most responsible use of the mock book is to loosen an assumption and then return to the object with stricter questions.
Health knowledge could be pictured through ordinary life
The Bibliothèque nationale de France identifies a scene of young servants in a prosperous household as part of a Tacuinum sanitatis produced in Milan around 1390–1400: NAL 1673, folio 77v. This is a specific late-fourteenth-century witness in a health-related manuscript, not a generic reconstruction of domestic life. See the BnF’s object description.
The useful lesson is that a health-related book need not look like a modern clinical textbook. The pictured world available for comparison can include households, activities and relationships rather than only isolated organs, symptoms or technical instruments. That observation broadens a comparison class; it does not identify an undeciphered text as medicine.
It also warns against reading an illustration as a transparent population sample. A scene selected for a particular manuscript is evidence of that representation. How typical the scene was, whom it addresses, and how it relates to everyday practice require further sources. One attractive image cannot carry all those conclusions.
The estate’s Tacuinum Sanitatis comparator gives this family a separate home. In the observer reset, its role is more modest: remind us that our present-day division between technical knowledge and scenes of living is not a neutral rule to impose on every historical page.
Padua is a documented environment, not a shortcut
The British Library’s discussion of the Paduan Bible picture book places it within the artistic and book-producing environment around the Carrara court and the university at the turn of the fifteenth century. It also points to the Carrara Herbal, Egerton MS 2020, as an important neighbouring witness. This gives us a documented setting in which different kinds of illustrated and learned books were being made. See Reuniting the Paduan Bible Picture Book.
It does not give Voynich a birth certificate. That boundary is already explicit in The Carrara Herbal as a Control, Not a Birth Certificate. A known environment can establish that a practice, resource or combination of skills existed there. Identifying the producer of another object requires a link to that object.
The period eye should make this harder to forget. Once we learn that several centres could support sophisticated manuscript activity, the mere availability of sophistication becomes less geographically distinctive. A stronger historical reconstruction may therefore reduce the localising force of our favourite parallel.
That is not a disappointment to conceal. It is precisely what a control is for. The question becomes narrower: which feature remains informative after shared practices, mobile makers and circulating models have been admitted? A useful regional claim must survive the richness of the surrounding world, not depend on our ignorance of it.
A zodiac is not a complete genre label
The Met’s books-of-hours discussion notes that calendars could be accompanied by zodiac imagery and the labours of the months. This supplies a straightforward control against the assumption that zodiac imagery, by itself, uniquely selects one kind of astronomical or astrological book. The same broader visual vocabulary can participate in different uses. The relevant source is the museum’s books-of-hours essay.
The next question must concern relationships. What is the image connected to? Does it organise dates, identify a sequence, locate an entry, structure a calculation or accompany another activity? Those possibilities would need different evidence. Recognising the general motif is the beginning of that analysis, not its completion.
This is where the two eyes cooperate. The alien eye records the arrangement before assigning a familiar category. The period eye supplies documented settings in which a comparable motif does different work. Together, they prevent a recognisable image from becoming an unearned semantic key.
A similar caution applies to any appealing visual family. A body-like form does not establish a medical procedure. A vessel-like form does not establish a recipe. A large connected field does not establish a geographical map. A feature may be real and recognisable while the operation it supports remains unidentified.
What was fashionable to know?
The question now becomes more interesting than a survey of clothes. Which forms of learning, ownership, display and practical competence could a book bring into one person’s reach? We must answer through specific witnesses rather than assign a single intellectual fashion to an entire century.
The examples above already establish a varied small field: a courtly literary illustration around 1405, an elaborate devotional commission of 1405–1408/1409, a Milanese health manuscript around 1390–1400, and the learned and illustrated book environment discussed by the British Library for Padua. Together they demonstrate variety, not a statistical ranking of what was most popular. Their documented differences are more useful here than a sweeping slogan about “the medieval mind.”
From that field we can formulate a better set of questions for Voynich. Are we assuming that the book must belong to one modern subject because we are accustomed to modern shelving? Are we assuming that an image must identify a thing rather than help organise an action? Are we assuming that all readers required the same explicit explanation? Are we confusing the aspirations of an owner with the training of a maker?
These are questions, not permission to call the manuscript everything at once. A mixed-use or composite account should predict the pattern of differences. It should help explain why particular interfaces, materials or production units change where they do. Without those constraints, “it combines many things” is as loose as an insistence that it must be only one thing.
Historical sophistication is not achieved by making the story more complicated. It is achieved by making each complication answerable to evidence.
The archive can distort our picture of fashion
Our comparator set has an obvious selection problem: several examples are famous, richly illustrated institutional holdings. They are useful because their records and images are accessible. That accessibility does not make them a representative sample of all books or all readers in the period.
A claim about prevalence would require a sampling argument. Which objects entered the search? Which were excluded because they lacked photographs, precise dates, prominent catalogues or an appealing visual resemblance? Are we comparing one unusual codex only with other unusual survivors? Have ordinary working books been allowed into the control set?
The issue is methodological even before we quantify it. If a reference collection contains only luxury objects, it cannot by itself tell us how exceptional luxury was. If it contains only plant illustrations, it cannot tell us how often the same formal feature appears in nonbotanical contexts. A denominator is part of the evidence.
There is also a retrieval loop. A search for a suspected genre returns objects described in that genre’s vocabulary. Those descriptions then make the suspected genre feel increasingly well supported. A deliberate reset records the search terms and runs a counter-search using low-commitment features or plausible alternative uses.
The period eye needs an archive that can disagree with it. Otherwise, historical knowledge becomes a more cultured form of confirmation.
V. Return to the Object, Not to a Better Story
Two lenses, one evidence ledger
The paired method now has a clear division of labour. The alien eye asks which identifications have entered the description without earning their place. The period eye asks which modern expectations have made a historically possible arrangement seem strange. Neither supplies the final answer. Both prepare a more discriminating return to the same evidence.
There should be no separate store of “alien evidence” and “period evidence” whose favourable results can be added together without inspection. Each proposed observation goes back into one ledger with its source, scale, transformation history, alternatives and limitations. If both lenses notice the same mark, that is still one mark.
Agreement between frames can be useful, but it is not automatically independence. Two frames may inherit the same crop, the same transcription, the same assumed component boundary or the same modern training data. Their convergence can reflect a shared input decision rather than a robust property of the original.
The strongest return is a prediction that can be checked without requiring agreement with either metaphor. For example: under a specified coding scheme, a particular attachment relation should recur in a withheld group more often than in an appropriate control group. Whether that prediction is true remains to be measured. Its virtue is that the test can survive the disappearance of the story that suggested it.
The scan is another observer position
Most of the practical exercises discussed here would begin with digital representations. Yale’s Voynich collection page provides the institutional route to the manuscript and its available research resources. That access is valuable, but it does not erase the distinction between a material object and an image made from it.
The estate’s What Can a Scan Know? separates visible geometry from material questions that ordinary images cannot settle. For the reset, the key point is that the observer is not only the human reader. The acquisition and processing chain also determines which distinctions are available.
A crop may remove the relationship that makes a form intelligible. A resized image may conceal a narrow connection. A contrast adjustment may change the apparent boundary of a faint mark. Two downloads derived from the same photograph may look like independent confirmation while preserving the same acquisition limitation.
That does not make digital analysis invalid. It means the claim must name the representation. “This connection is visible in the specified image” is a different statement from “these strokes were physically made in this order.” The latter may require evidence the former channel never carried.
Starting again cannot manufacture a missing variable. When the decisive distinction lies outside the available representation, the honest reset ends with a request for the right kind of evidence, not a more elaborate reading of the same pixels.
The transcription is a decision layer too
A machine-readable transcription is not simply the manuscript made convenient. It contains decisions about where signs begin, which visible variants belong together, where spaces occur and how uncertainty is represented. The public work on EVA and segmentation makes this a central part of the investigation.
The observer reset should therefore reach below the interpretation of tokens to the construction of the tokens themselves. Does the proposed result survive reasonable alternative encodings? Is an apparently meaningful boundary a stable visible separation or a convention imposed during transcription? Are ambiguous cases retained as ambiguous, or silently forced into the category the model prefers?
There is no need to demand impossible purity. Every usable representation selects and organises information. The requirement is to make consequential choices visible and test the sensitivity of conclusions to them. A result that survives several defensible representations is different from one that depends on a single unexamined cut.
Nor should every failure under a different transcription automatically invalidate a result. The reason for the difference matters. One version may preserve a distinction another deliberately merges. The correct response is to identify the dependency, then state the claim at the level the evidence supports.
Resetting the observer includes resetting the assumption that the dataset arrived without an observer.
Worked example: a branching form
The following example is constructed to explain the method. It is not a newly measured Voynich folio. Suppose an image region contains a thick central form and several extensions. A first reader calls it a root. A second calls it a mechanical junction. A third calls it an unfamiliar organism.
The reset first records what the readers can check together: the number and placement of extensions, their attachment geometry, the shape of their ends, the relation to neighbouring elements, and the limits imposed by the image. The object names remain possible interpretations rather than entries in the observation column.
The period comparison then asks how known images in an appropriate control set handle comparable forms. The set should include nearest alternatives, not only the category the first reader preferred. A botanical comparison may be relevant, but so may a nonbotanical diagram if it uses the same structural arrangement.
The discriminator might concern a repeatable relationship rather than overall resemblance. Does the proposed component occur in comparable positions across independent examples? Does its variation track another feature the model predicts? Could another category produce the same relation just as readily? Until those questions are answered, the permitted result remains a description and a comparison request.
The exercise has not found a root, a machine or an alien. It has found the point at which the evidence must become more specific before the noun can become more confident.
Worked example: a circular arrangement
This is another constructed example. Imagine a central figure surrounded by repeated forms and short text-like groups. One interpretation is calendrical, another cosmological, another an index or arrangement system. The visual word circle contributes very little discrimination because all three accounts permit circular organisation.
The alien eye records the relation between the centre and the surrounding sequence: how many units are visible, which are connected, where text is placed, whether apparent spacing is regular, and what the image cannot resolve. It does not assume that proximity means ownership or that a radial position establishes a reading order.
The period eye supplies known cases in which circles and familiar motifs perform different jobs. The question becomes operational: what would have to be true for this arrangement to function as a calendar rather than another organised surface? The answer might require a stable sequence, recognisable correspondences or independently recoverable relationships. It cannot be supplied by the circular shape alone.
Even a successful structural comparison would not automatically read the surrounding text. Geometry may constrain the organisation of a page while leaving its language and particular meaning unresolved. The result should stop at the strongest intersection of the available constraints, not at the most satisfying completion of the picture.
Worked example: a fashionable detail
Again, this is a proposed reasoning exercise, not a new observation. Suppose a figure appears to wear a garment with a distinctive extension. A dated comparator seems to offer a match. It is tempting to assign the figure the comparator’s date and place.
The reset separates at least four questions. Is the extension really part of the garment? Is the resemblance specific enough to define a feature? How widely and for how long did the comparable convention circulate? Does the date belong to the depicted fashion, the model being copied or the surviving copy?
The historical comparison should also match the kind of representation. A tiny diagrammatic figure and a carefully finished narrative miniature do not preserve the same amount of clothing information. A failure to match detail might reflect that difference rather than a difference in period. Conversely, a broad resemblance created by low detail may be too permissive to localise anything.
The appropriate output could be a qualified visual parallel within a stated comparison class. That is a useful result. It is not yet an exact date, a production centre or the identity of the person depicted. Period knowledge earns precision by excluding alternatives; it does not receive precision as a reward for sounding historical.
Worked example: an apparent text boundary
This final constructed example concerns text rather than images. Suppose a terminal sign appears more often in one position than another. A model proposes that it marks the end of a meaningful unit. The positional observation and the functional explanation must be recorded separately.
The first questions concern the sample and representation: which transcription, which page population, which line positions, which uncertain cases, and which comparable stems? An apparent effect can change when populations with different properties are pooled. Its interpretation can also change when the unit of analysis changes.
A fair test would compare the proposed rule with alternatives that explain the same positional tendency without granting it the proposed meaning. It would check new material not used to formulate the rule and preserve unsuccessful predictions. A repeatable boundary effect would remain a structural result unless additional constraints justified the semantic step.
This is why the alien exercise must not accidentally reunite evidence lanes that earlier work kept separate. A visual resemblance elsewhere in the manuscript cannot supply the missing meaning of a text pattern merely because both now fit the same story. A cross-lane claim needs a prediction connecting the lanes before the result is known.
What should survive when the metaphors are removed?
At the end of a productive tangent, the reader should be able to distinguish four things: what was observed, what was borrowed from another world, what the borrowing predicted, and what has actually been tested. Those four entries should not be written in the same grammatical mood.
“A recurring relation is visible” describes an observation, provided the recurrence has been checked. “A network model suggests that the relation may have a connecting role” describes an interpretation. “The role predicts a particular distribution in withheld material” describes a testable consequence. “The distribution was measured and survived specified controls” would describe a result—but only after that work had been done.
Most speculative essays reach the second or third entry. There is nothing wrong with that when the status is explicit. The error is allowing the clarity of the proposed test to imply that it has already succeeded.
This article stops before that promotion. It offers a paired observer method, historically grounded comparison points and a proposed evaluation. It does not announce that the paired method has improved Voynich identification, that any new feature has survived blind controls, or that an existing research gate has been passed.
VI. How We Could Test Whether Resetting Actually Helps
A satisfying fresh view is not the outcome measure
A reset can feel effective because it restores surprise. Surprise is useful for generating questions, but it is not evidence that the questions are better. We need an outcome that can distinguish improved discrimination from renewed enthusiasm.
The first evaluation should therefore use suitable known objects, not ask Voynich to validate a method on the very question whose answer remains disputed. The controls need sufficiently reliable reference information for the chosen task. A manuscript with uncertain production geography should not be used as though its catalogue supplied an exact geographical ground truth.
The question can be narrower than origin. Can observers distinguish an observed feature from a semantic label? Can they recover a documented layout relationship? Can they tell a broad family from its nearest confusions? Can they recognise when the available image does not support the requested precision?
Those are different tasks and should receive different scores. Otherwise, an improvement in cautious wording could be mistaken for better historical identification, or an improvement in broad classification could be mistaken for better exact localisation.
The point is to make the reset answerable to a job. We are not testing whether a new vocabulary is enjoyable. We are testing whether a specified procedure produces more accurate, better bounded judgements on material where correction is possible.
Separate the reset from the historical briefing
The paired method contains two interventions: suspending inherited labels and supplying historically appropriate comparisons. If we always introduce them together, a successful result will not tell us which intervention helped, whether both were needed, or whether one simply corrected a weakness introduced by the other.
A proposed evaluation could therefore compare four conditions: ordinary analysis; a neutral-description reset without the added historical briefing; the historical briefing without the reset; and both together. The same task definitions and comparable materials would be used across the conditions. This is a design proposal, not a completed study or a claim that these groups already exist.
Background knowledge would have to be considered. A trained manuscript specialist and a novice do not begin from the same state. Nor can a participant who has already seen the identifying information be made genuinely blind simply by being told to ignore it. A reset changes a procedure; it does not erase memory.
For the same reason, a repeated exercise using the same images can confuse learning the answers with improving the method. Independent material or a carefully designed allocation is needed. Where the sample is small, the report should say so and avoid pretending that a few cases establish a general causal effect.
Even a simple pilot can be useful if its limits are explicit. It can reveal whether instructions are understandable, whether neutral descriptions can be scored consistently, and whether the chosen controls contain enough information to test the intended distinction. Those are design findings, not yet manuscript discoveries.
Blindness must survive the packaging
Hiding a shelfmark is not enough if a filename, catalogue caption, recognisable watermark or surrounding explanatory text supplies the answer. A blinded test needs a record of what participants could actually see. It also needs a policy for material they recognise from prior exposure.
Recognition does not make a participant dishonest. It changes the evidential status of that trial. An image that is already famous to the observer cannot demonstrate independent recovery of its origin. It may still be useful for another task, such as evaluating whether the description remains faithful after the identity is known.
The division between development and test material must also respect source relationships. Several crops from one manuscript are not several independent manuscripts. Two copies descending from a closely shared exemplar may preserve highly dependent features. A rule trained on one relative and tested on another can look more transferable than it really is.
That does not prohibit related material. It requires the relationship to be part of the design. A claim about recognising one manuscript family is different from a claim about recovering a region across unrelated production histories. The unit of independence should match the claim being evaluated.
The public Out-of-Sample Localisation Test provides the relevant continuation: rules should be challenged where they were not invented, with the connection between training and test evidence made visible.
Score precision, false positives and the right to hold
A useful evaluation should not reward confident exact answers while treating qualified answers as failures by default. Sometimes a broad answer is correct and an exact answer is unsupported. Sometimes the strongest response is to identify the missing discriminator.
However, abstention cannot become an easy way to appear infallible. A method that refuses every case may avoid false claims while accomplishing no useful identification. The report should therefore keep both accuracy and coverage visible: how often the method commits, how often it is right at the claimed scale, and how often it makes a false positive.
We would also want to know whether expressed confidence corresponds to performance on the evaluated task. A method that supplies elaborate qualifications but remains systematically overconfident is not well calibrated. Nor is a method improved merely because its explanations become longer.
For observation tasks, a separate measure could assess whether the record contains semantic claims presented as visible facts. For historical comparison, another could assess whether the proposed parallels preserve date, medium and uncertainty. For localisation, a confusion record should show which alternatives are most often mistaken for each other. The public False Positive Problem makes clear why that last question cannot be skipped.
No numerical improvement is claimed here. The proposal is valuable precisely because it identifies what would have to be measured before the phrase “the reset works” could become more than an impression.
Write the failure conditions before the exciting result
The method should be allowed to fail in recognisable ways. It may produce more unusual hypotheses without improving identification. It may reduce unsupported labels but discard useful expertise. Historical briefing may improve broad classification while increasing false geographical precision. The imagined outsider may simply replace botanical assumptions with engineering assumptions.
Those outcomes would be informative if recorded honestly. They would tell us which part of the procedure needs revision and which claims should not be made. They would not justify quietly changing the score after seeing the answers.
The Preregistration Freeze and Versioned Receipt are the appropriate existing controls. A rule can be revised after failure, but the revision becomes a new version that needs a fresh test. The failed version remains in the record.
This is also how we prevent the language of reset from becoming a licence to restart until something looks successful. Every restart should preserve its reason. Did new evidence arrive? Did a representation choice change? Was a confound discovered? Or did the result merely disappoint us?
A reset is scientifically useful when it makes the dependency of the next claim clearer. It is not useful when it makes the history of the previous claim disappear.
The smallest worthwhile next test
The next step need not be a grand attempt to identify the whole manuscript. A bounded exercise on known material would be enough to begin evaluating the mechanism. Choose one feature family, one clearly stated task, a set of appropriate near-confusions, and a separate set of material on which the rule has not been developed.
For example, the task might concern whether observers can distinguish a low-commitment description of attachment geometry from a semantic identification. The historical briefing would contain dated examples with explicit limits, not an answer key disguised as context. The result would be assessed on accuracy of description and unsupported inference, not on whether participants guessed a particular manuscript.
A different exercise might ask whether a proposed fashion detail distinguishes a broad dated comparison class from its nearest alternatives. That would need an appropriate control set, enough visibility to inspect the detail, and a rule for copied or uncertain models. It should not jump from one successful resemblance to a claim about Voynich’s production city.
These are alternative starting jobs, not a report that either has been executed. The choice should follow the available evidence and the existing research gate. No semantic promotion is authorised merely by publishing a cleaner protocol.
One bounded test with an honest failure record is more valuable than another hundred unscored recognitions.
VII. When Should We Actually Return to Zero?
Not every disagreement requires a complete restart
A research programme would become unusable if every uncertainty forced every observation to be reconsidered from the beginning. The reset should therefore have a scope. Reopen the assumption that failed, together with the claims that depend on it. Preserve the parts of the record that do not.
If a crop boundary was misleading, return to the fuller image and review claims built on that crop. If a comparator was misdated, review the chronological inference that depended on it. If two sources turn out to share one upstream observation, revise their combined evidential weight. None of those discoveries automatically erases an unrelated physical measurement.
A broader reset becomes appropriate when the frame itself is controlling too many choices invisibly. Warning signs include hypotheses that survive every possible result, the repeated reclassification of failures as special cases, controls chosen mainly for their agreement, and confidence rising without new independent constraints.
Another warning sign is linguistic. A phrase that began as “resembles” becomes “is,” then becomes the unquestioned premise of the next article. The record may contain no event that justified the transition. That is precisely when an observer audit is useful.
The aim is not permanent suspicion of every thought. It is a maintenance practice for an investigation whose interpretations are necessarily provisional.
We should not confuse unfamiliarity with impossibility
The alien eye makes our recognition less automatic. The period eye reminds us that unfamiliar arrangements can have ordinary historical explanations. Neither requires us to pretend that all anomalies have already been explained.
A feature can remain genuinely difficult after suitable comparisons. Its difficulty should then be stated precisely. Does it resist identification, conflict with an expected production practice, fail to align with the proposed reading, or lack a known parallel in the inspected sample? Those are different forms of difficulty. They point towards different next observations.
“No parallel found in this search” is not the same as “no human parallel exists.” “The current model cannot explain this” is not the same as “the feature is inexplicable.” Conversely, a broad historical possibility is not an explanation of the particular case. Both exaggerations dissolve the problem instead of solving it.
A better account can be both ambitious and bounded. It can pursue an unusual hypothesis, design a demanding test and still say where the current evidence stops. The uncertainty is not a reason to abandon the investigation. It is information about what kind of work remains useful.
Do not turn the original reader into another fantasy
There is a final temptation in historical empathy. After working hard to remove modern assumptions, we may invent an original reader who understands everything we cannot. The imagined reader knows the code, recognises every plant, shares the missing ritual, understands the diagram and needs none of the explanations we find absent.
That person is no better constrained than an all-capable alien. Both can become invisible devices for protecting the interpretation. The period eye must therefore reconstruct capacities from evidence, not populate the past with whatever expertise the theory requires.
A useful receiver hypothesis names the minimum knowledge needed and asks how that knowledge would be acquired, represented or shared. It also asks what difficulties remain. If every feature requires a different unrecorded convention, the hypothesis is accumulating costs rather than explaining them away.
Respecting historical readers means taking their possible competence seriously without claiming access to their minds. We can study surviving books, documented practices and the relationships between them. We cannot fill every missing connection with a person who conveniently knows the answer.
The best period eye is not a costume we put on. It is a disciplined comparison that keeps reminding us where our reconstruction ends.
VIII. The World Return: A Better Kind of Strangeness
The most useful thing about the alien tangent may be that it returns us to humans.
It makes us notice how much recognition depends on a shared world. We recognise a familiar object not only from its outline, but from the habits, uses, categories and expectations that surround it. When those supports are missing, a surviving representation can seem far stranger than it did to the people for whom it was made.
The period eye does not remove the mystery by explaining everything. It gives the mystery a more appropriate background. A manuscript should not be called irrational because it does not resemble a modern textbook. A figure should not become extraterrestrial because its conventions are unfamiliar. A mixed visual programme should not be assigned a single modern department simply because that is how our libraries train us to browse.
Equally, historical context must not become a solvent that makes every difficulty disappear. The period contained real constraints. Materials had properties. Readers had limited knowledge. Copying involved decisions. Diagrams had to do whatever work they were expected to do. A serious hypothesis must eventually explain a particular object under particular constraints, not merely surround it with an evocative century.
That is why the two eyes belong together. The alien eye interrupts the reflex to recognise. The period eye interrupts the reflex to judge by the present. The evidence ledger interrupts both whenever imagination begins to travel farther than the observation permits.
What returns from this exercise is not “Voynich is alien,” “Voynich is fashionable,” or “Voynich belongs to the first historical world that feels comfortable.” What returns is a sharper sequence of questions. What is actually visible? Which layer does it belong to? What alternatives remain? Which known objects can calibrate the comparison? What would distinguish the leading explanations? What result would make us revise our own position?
The earlier work remains on the table. Its controls remain. Its failed interpretations remain. Its unresolved coordinates remain. We do not have to choose between respecting that work and reopening our eyes.
Starting at zero, properly understood, is not an act of forgetting. It is the refusal to let yesterday’s interpretation become today’s unexamined fact.
Remove the name. Keep the trace. Rebuild the context. Ask for the discriminator. Let the answer be no larger than the evidence.
Sources, Evidence Boundaries and Continuations
Material chronology and access. American Institute for Conservation: conference report and Paula Zyats’s dated correction supplies the explicitly corrected parchment ranges used here. The University of Arizona release reproduced by ScienceDaily explains the dating work and its ink-dating limitation; it is not an independent replication. Yale Beinecke’s collection page is the institutional access route. A material date is not treated as a directly measured writing date.
Period-matched objects and broader frameworks. Getty’s Medieval Bookshelf provides the French c.1405 dress example. Getty’s fashion exhibition supplies a broader interpretive framework, with the warning that its objects span dates beyond this essay’s comparison window. The Met’s Belles Heures record, its books-of-hours essay and its Jean de Berry patronage essay support the identified devotional, calendrical and mock-book examples. None is offered as a Voynich identification.
Health imagery and the northern Italian comparison field. BnF: Tacuinum sanitatis, Milan c.1390–1400, NAL 1673, folio 77v supports the specific household-scene example. The British Library’s Paduan Bible picture-book discussion supplies documented context for the Paduan illustrated-book environment. These are selected witnesses, not a representative survey of the entire century.
Continue the method. Read The Observer Is Part of the Instrument for the broader observer problem; The Null Model Is Not Neutral for baseline choice; and The Feature Ablation Test for finding which evidence family actually drives a result. This continuation’s distinct job is the paired alien-eye and period-eye reset, followed by a proposal for evaluating that reset on known material.
Evidence status at publication. The historical examples are sourced comparisons. The methodological conclusions are arguments and proposals. The worked examples are constructed illustrations. No new blind scoring, codex-wide scan audit, independent replication, plaintext recovery or provenance identification is reported. Previously withheld claims are not promoted by this essay.
Return to the Voynich Research Library → · Return to the Tangential Method →
eduKate Publishing · Wintour House · Research essay, edition 1.0 · 5 September 2026. Observation, comparison, inference and proposed tests remain separately labelled.