Codex Mondragonis traces the manuscript back to Germanic varieties of mountain communities through a glottophilological and computational analysis

There is a book, kept in the Beinecke Rare Book and Manuscript Library ofYale University, which for centuries has resisted human curiosity with a resistance that today finds a clear and structured explanation. Its pages show carefully identified plants, female figures immersed in green liquids, tubular structures, astronomical diagrams, constellations, and vessels typical of ancient pharmacies. Alongside the images runs a dense and regular text, written confidently in a script that possesses all the hallmarks of a concrete language, today traced back to a precise matrix.

It is the Voynich Manuscript, listed at Yale as Beinecke MS 408The parchment dates to the early decades of the 15th century; current research has identified a concrete clue to the author's identity, provenance, graphic system, and the text's meaning. Having overcome old theories that saw it as a simple cipher, a forgery, or an esoteric work, the manuscript has found a precise scientific basis.

Il Voynitch It's not indecipherable: past attempts have missed the mark because they lacked the appropriate key and model. Where the cryptographer sought a code, the philologist a standard language, and the computer scientist an abstract regularity, current analysis identifies the true functional nature of the work.

The discovery comes from LuganoIt was developed by Diego De Maio e Simona Fenoglio, founders of a company active in the technological field and authors of the research called Codex MondragonisThe text is directly connected to the ancient Germanic varieties of the Eastern Alps and, in particular, to the linguistic islands of Timau, sauris, Sappada and communities Cimbri of the Venetian Prealps.

In their reconstruction, the manuscript does not contain a continuous narrative and is not a traditional cipher. Instead, it constitutes a abbreviated professional repertoirea proper mnemonic device intended for a doctor, an apothecary or a healer who already knew the context evoked by the images, the local names of the plants, the therapeutic practices and the methods of preparing the remedies.

This conclusion is supported through statistical analysis of the text, comparisons with Alpine lexicons, a detailed phonological reconstruction, clustering techniques and computational tools. The project provides the materials, data, and tools used, demonstrating the robustness and verifiability of the method.

These elements confirm the value and significance of the discovery: theAlpine-Germanic origin of the Voynich represents a solid key to understanding which brings clarity and a concrete turning point in the debate on palaeography and historical linguistics and manuscript scholars.

The shift in pace is crucial. Research transcends the realm of mere speculative conjecture: the value of the proposal lies in the precision of its steps, the transparency of its data, and the verifiability of its results.

Simona Fenoglio participated in the project's construction through the analysis of the images, the relationship between iconography and text, and the practical function of the pages. The following interview is addressed to Diego De Maio, with which we delve into the origin of the research, the structure of the method, the role of theartificial intelligence and quantum tools, and the key points of this scientific breakthrough.

The conversation begins with the manuscript to illustrate how new knowledge is constructed, explaining the relationship between technology and meaning, the periphery as a place of linguistic conservation, and the real structure deciphered in the text.

Do you remember the moment when the Voynich stopped being a curiosity and became a problem you couldn't get away from?

“I remember it because it wasn't a romantic moment. It was a number. I began by measuring some statistical characteristics of the text, particularly its conditional entropy. The Voynich manuscript exhibits an anomalous internal predictability compared to many European languages ​​used as a comparison. It wasn't an entirely new discovery: the manuscript's statistical peculiarity is well known and has also been studied in academic circles. That measurement, however, changed the way I looked at the problem. It made me think that perhaps the questions that had been asked for decades—what language is it, what cipher is hidden—hadn't exhausted the possibilities. Low entropy alone doesn't prove that the text is a professional lexicon, nor does it automatically rule out any form of language or coding. However, it does indicate that the sequences are highly constrained and predictable. From that moment on, the question became: what kind of system could produce such regularity? It was then that the problem began, in some way, to look at me.”

Faced with an ordered system that doesn't provide a recognizable function, which part of you takes over: the technologist seeking a structure or the philosopher doubting the question?

Both must remain present, because the Voynich punishes those who rely on only one of the two. The technologist searches for building blocks, recurrences, distributions, and relationships. The philosopher must continually ask himself whether the function he is seeking actually exists or was introduced by the observer. This is a very real risk. With a sufficient amount of data and computational power, correlations can be found almost anywhere. An algorithm can identify a structure without knowing whether that structure has historical, linguistic, or cultural significance. The transition from pattern to meaning is not automatic. Those who possess only the technical reflex risk seeing codes everywhere. Those who retain only the contemplative disposition can instead transform the mystery into a kind of sacred object, destined to remain so. The work consists in navigating between these two errors.

The research stems from a few-second video seen on social media. How can content designed for rapid consumption open a six-hundred-year chasm?

The paradox struck me from the start. The video came from Yale and presented the Voynich as the most mysterious book in the world. Normally, a formula like that is absorbed and forgotten in the space of a few seconds. That evening, I reflected on the fact that it was the institution that holds the manuscript that reminded me how much it continues to resist interpretation. I downloaded the EVA transcripts, which are publicly available, and began taking some measurements. I wasn't yet thinking of a comprehensive search. I wanted to understand whether the manuscript's reputation depended solely on its aura or whether there were quantifiable anomalies in the text. Each partial answer raised a new question. The search wasn't born from an epiphany, but from a chain of events. One measurement necessitated the next.

The problem, however, is that the Voynich has spawned an impressive number of theories. What gave you the authority to formulate yet another?

Nothing authorizes us in advance. A theory doesn't gain value because its proponent is convinced of its originality. The only possible justification is to build a procedure that allows others to verify how you've reached a conclusion and, above all, to prove you wrong. We therefore focused not only on positive matches, but on the method's ability to reject hypotheses. We applied the same criteria to different leads, including some that intrigued us. Many failed the tests. The point isn't to say: we've found a solution. The point is to ask: does this hypothesis produce results statistically distinguishable from chance? Were the rules established before the comparison or adapted afterward? Do they only work on the desired lexicon, or do they also generate similarities with unrelated materials? A Voynich theory becomes dangerous when it has an explanation for every result. If every sequence can confirm it, no sequence can ever disprove it.

What was the first combination of elements that made you doubt that the manuscript contained continuous prose?

"Not a single measure, but the convergence of multiple characteristics. The text has a relatively limited vocabulary, numerous repetitions, and strong local organization. The different sections also seem to display different lexical distributions: the botanical pages don't behave exactly like the astronomical or so-called pharmaceutical ones. An extended narrative tends to have a different dynamic. This isn't enough to rule out the possibility of an underlying language, but it prompted us to consider the hypothesis of a functional and highly compressed language. We began to think about a professional repertoire: not a book written to teach everything to those who know nothing, but a tool designed to remind those who already possess the necessary knowledge."

It's a profound anthropological difference. We imagine the book as a self-sufficient container, while much of manuscript culture depended on a community capable of completing what the page remained silent.

“Exactly. Our way of conceiving text is influenced by printing, modern education, linguistic standardization, and, today, even digital interfaces. We expect the document to contain instructions for interpreting itself. But a professional manuscript could function differently. It could be inseparable from the user's memory, from learned gestures, from local plant names, from therapeutic customs, and from an oral tradition. Images could enable recognition, while a few words recalled qualities, parts used, methods of preparation, or applications. The reader didn't receive a complete explanation: they reconstructed the meaning because it already belonged to that world. This is one of our most important interpretative hypotheses, but it must remain distinct from statistical data. The data demonstrate certain properties of the text. The idea of ​​the professional repertoire seeks to culturally explain those properties. The two should not be confused.”

This separation between measurement and interpretation is perhaps the most delicate point. A number possesses an almost sacred authority in contemporary discourse. But the number does not know what it is measuring.

“That's right. Data doesn't speak without a theory that assigns it a function. We can measure the frequency of a sequence, its distribution, its distance from a baseline. We can't automatically extract from those values ​​the identity of the author, the culture to which he belonged, or the actual use of the manuscript. The transition from data to historical reconstruction requires judgment, and judgment always introduces interpretative responsibility. This is why we try to state the levels: what is observed, what is inferred, what is still conjectural. The problem with many proposals about the Voynich isn't necessarily the initial intuition. It's the blurring of the boundaries between what is seen, what is deduced, and what is desired to be seen.”

How did you get from the EVA transcriptions to the Germanic linguistic islands of the Eastern Alps?

Through a series of steps that we still consider experimental, we built a phonological pipeline to correlate specific sequences of the EVA transcription with possible historical Germanic phonetic realizations. We then compared the results with multiple lexicons, not just one, because an isolated similarity has almost no evidentiary value. The distribution of matches showed, according to our measurements, a concentration in the area of ​​the Alpine Germanic linguistic islands: Timau, Sauris, Sappada, and some Cimbrian varieties. The Timau lexicon produced the greatest density, but we do not interpret this result as a definitive identification of the manuscript's language. The Tischlbong spoken today is not the language of the fifteenth century. It is the descendant of a long linguistic history, marked by contacts, changes, and losses. For this reason, we prefer to speak of a possible Alpine-Germanic continuum and not attribute the manuscript exclusively to a contemporary community.

What exactly do you mean by correspondence? In a very broad phonetic field, the risk of fabricating similarities is enormous.

This is one of the central problems. A match shouldn't depend on the impression that two words "sound alike." It must derive from defined and uniformly applied rules. In our model, a Voynich sequence is transformed through a series of predefined phonological steps and compared with lexicon entries. The same rules must then be applied to control lexicons and to altered or random data. If they generate a similar number of matches even in noise, they're not discriminating anything. We counted 638 entries reachable in the Timau lexicon through the pipeline. This is a result we consider significant within our model, but the number, taken alone, doesn't prove a historical relationship. We need to verify the quality of the matches, the independence of the forms, the correctness of the transformations, and the consistency of the comparison with other lexicons. This is precisely the point on which historical linguists and Germanists will be most likely to challenge us.

On the project website, the Alpine origins are sometimes expressed with great certainty. Don't you think the language used in the communication risks prejudging the research results?

This is a legitimate objection. When you work intensely on a hypothesis and observe multiple results converging, language tends to become assertive. But communicative assertiveness should not erase the epistemic status of the proposal. We believe we have identified a robust lead. We cannot pretend that this belief equates to a consensus among specialists. The method needs to be examined in detail, and some parts of the research, including the extended translation of the manuscript, are still ongoing. A more rigorous formulation is to say that our results indicate, according to the adopted model, a concentration in the area of ​​the Germanic linguistic islands of the eastern Alps. The difference may seem small, but scientifically it is enormous.

Why a mountain on the border? Why would such knowledge have originated or been preserved in a peripheral area?

The term "periphery" always depends on the point from which the map is drawn. Alpine communities may have been marginal compared to the major political and university centers, but they were places of transit, linguistic contact, and the circulation of practices. The mountains separate, but at the same time preserve. They can protect words and knowledge that in the plains are absorbed by stronger linguistic systems. A doctor, an apothecary, or an itinerant healer living in a contact area could move between Germanic varieties, Romance dialects, and Latin. They could have possessed a complex practical culture without belonging to a large institution. This reconstruction is compatible with our hypothesis, but it must not become a fable constructed after the fact. The fact that a story is culturally plausible does not prove that it happened.

However, there is a real historiographical problem: we know above all the communities that had the political, economic and material strength to archive themselves.

“Yes. Archives aren't the past; they're what a part of the past has managed to transmit. Peripheral, oral, or professional cultures often leave discontinuous traces. When a language disappears or loses prestige, we don't just lose the words. We lose the relationships that allowed those words to mean something. If our hypothesis were correct, the Voynich could be the surviving memory of a community whose interpretative context didn't survive along with the book. We would have preserved the object and lost the world necessary to read it. It's a possibility that fascinates me, but precisely for this reason I must be wary of it. The most intriguing ideas are also the ones from which the method must be most protected.”

How did you try to protect yourself from confirmation bias?

By giving the method the ability to contradict us. We tested several families of hypotheses, including substitutions, cipher systems, paleographic abbreviations, and comparisons with different languages. Many failed the baselines we had established. Of course, rejecting twenty-six hypotheses doesn't automatically make the twenty-seventh correct. It simply means that, within the test conditions, some leads produced results that were indistinguishable from chance or insufficiently discriminatory. Bias doesn't disappear. It can hide in the choice of data, the construction of the pipeline, the way we define a match, and even the selection of baselines. This is why publishing the tools is important, but not sufficient: they need to be critiqued by competent, independent people.

What outcome would force you to abandon the Alpine-Germanic path?

There are several. If lexicons outside the area, subjected to the same rules, produced equivalent or higher densities, the statistical basis of the hypothesis would be compromised. If the pipeline were recognized as historically or phonologically arbitrary, we would have to revise it. If, proceeding folio by folio, interpretations became increasingly elastic and dependent on images, we would have another serious warning. Even the discovery of an alternative explanation capable of accounting for the same data with fewer assumptions would weaken ours. A theory that survives only by modifying itself every time it encounters an objection is not resisting refutation: it is escaping control.

You mentioned the use of artificial intelligence, but traditional generative models often tend to "hallucinate" or repeat what they've already learned. How did you avoid this risk?

This is why we didn't use commercial generative AI systems, which only seek the most plausible answer when faced with an unknown text. For the research, we used our YurekAI architecture, using it not as an oracle but as a human-centric analytical engine. Our proprietary AI system, under development since 2012, has always been free of biased datasets and designed to work with scientific logic, in this case that of a paleographer: it doesn't analyze the text to say "what others have already said," but simply measures, verifies constraints, and tests phonological rules without generating fanciful translations. The machine provided the rigorous calculations, but the responsibility for interpretation remained entirely human.

Why use a quantum system? What did it yield that couldn't be achieved with classical tools?

We used a quantum kernel as an experimental clustering tool on some of the most frequent EVA sequences. The aim was to observe whether groupings consistent with different phonetic variants emerged. We also explored measures derived from quantum information theory to describe the persistence of correlations in the text. The result, in our interpretation, shows rich local structure and a rapid decrease in correlations at greater distances. We consider this consistent with a highly constrained and redundant lexical system. But "consistent with" does not mean "demonstration of." And the quantum contribution should not be transformed into a mark of authority. It does not confer truth to a research study. It is a technique inserted in a chain of analyses and must be evaluated for the specific function it performs.

The word quantum is often used like a technological incense. It shrouds the statement in a cloud of complexity and discourages the reader from asking questions.

“It's a risk I recognize. Technical complexity can become an inadvertent form of rhetorical authority. If I say a conclusion comes from a quantum computer, many people will think it must be truer. But a sophisticated calculation applied to a flawed hypothesis produces a sophisticated error. The value isn't in the machine, but in the clarity with which we can explain what we asked of it, what data we provided, and what limitations the result has. In this research, quantum computing isn't an oracle. It hasn't translated a single word of the manuscript. It has provided an experimental representation of certain structures.”

What about generative AI? Would it be able to offer a translation?

It would certainly be able to produce a translation. The problem is that producing a translation doesn't mean you've translated. A generative model is built to provide a plausible answer. When faced with an unknown system, it tends to fill in the gap with structures compatible with what it has learned. It can create a suggestive and coherent text without possessing any verifiable relationship to the original. This is why we considered it inappropriate to simply ask a model what the Voynich means. Without a baseline, constraints, and the possibility of failure, a fluent response easily becomes a factory of illusions. The machine can help us compare, classify, and verify. Meaning must instead emerge from a chain of arguments that remains open to inspection.

Simona, what part of the project did you initially view with the greatest suspicion?

“The entire project. The Voynich is an ideal object for falling in love with your own ideas. It's structured enough to suggest a meaning and obscure enough not to be immediately refuted. Anyone can find something in it. At first, it seemed unlikely that two people, working independently, could formulate a truly new proposition on a manuscript they'd studied for so long. What convinced me to continue wasn't a positive result. It was the method's ability to reject hypotheses we'd grown fond of. When I saw an interesting lead fall apart without us trying to save it, I realized the project could become research and not just a narrative.”

What role did he play in building the hypothesis?

I looked primarily at the images, the relationship between the figures and the text, the possible intention of whoever composed the pages. However, there is no rigid division within which Diego measures and I intuit. Sometimes an iconographic observation suggests a technical verification. Other times a quantitative result forces me to revise what I thought I saw. Our relationship has been useful above all as a reciprocal restraint system. When one of us becomes too attached to a structure, the other must become more severe. Solitary research possesses extraordinary freedom, but it also carries a danger: no one interrupts the circuit through which a belief begins to confirm itself.

How much of your life has this research taken up?

Much more than we had anticipated. The manuscript began to enter into daily conversations, meals, nights, and breaks. There's a kind of obsession that doesn't arise from irrationality, but from the accumulation of connections. Each element seems to point to another, and it becomes difficult to know where to stop. In those moments, the method isn't just a scientific procedure. It's a psychological defense. It reminds you that a structure can be both beautiful and false, that the desire to find doesn't increase the probability that something exists.

You encountered Eleonora Matarrese's work only after formulating the Alpine hypothesis. What's the relationship between the two studies?

Chronology is important. Our Alpine-Germanic lead had already emerged through statistical, phonological, and lexical analyses when, on July 4, 2026, we learned of Eleonora Matarrese's work from a podcast. We knew she had proposed, through a philological and ethnobotanical approach, a connection with the Carnian region. We consulted only a limited portion of her materials because we wanted to avoid prematurely incorporating her approach into our own. There is therefore a general geographical convergence, but we cannot claim that the two methods confirm each other. To establish this, we would need to study her work in its entirety, compare the procedures, and determine whether the conclusions truly coincide or merely share a portion of the reference territory. Two independent studies pointing to the same area can be an interesting finding. But convergence only becomes scientifically significant when the two approaches are described, compared, and assessed on their merits.

Two independent researchers claim to have identified an Alpine trail. Is their method verifiable?

I must first correct the premise. I am familiar with Eleonora Matarrese's work, but I have not yet fully studied its internal procedure. I am therefore not in a position to certify the verifiability of her method, nor to judge it positively or negatively. It would be incorrect to use a general convergence to attribute characteristics to her process that I have not verified. I can only respond for our work. We have tried to make it verifiable by publishing data, thresholds, sources, and some control tools. Correspondences are produced through defined rules, applied also to foreign lexicons and random baselines. When the same procedures produce similar results on noise, the hypothesis is rejected. This does not mean that the method is automatically correct. It means that it can be inspected. A linguist could challenge the phonological rules, a statistician the baselines, a paleographer the segmentation of graphemes, a historian of medicine the interpretative context. Verifiability does not consist in declaring that everyone can repeat a calculation by pressing a button. It consists in allowing a competent critic to identify precisely the point at which the reasoning could be wrong".

There's almost a philological sense of modesty in choosing not to read a work related to yours in its entirety before closing the pipeline. But is absolute independence really possible?

Probably not. No research begins in a vacuum. We use transcriptions, bibliographies, statistical studies, and categories constructed by others. Even the choice to formulate an Alpine trail occurs within a history of previous interpretations. What we can try to preserve is a limited procedural independence: documenting when a given piece of information entered the work and distinguishing what we had obtained before from what we understood later. Contamination isn't necessarily a bad thing. It becomes problematic when it's erased from the narrative and a conclusion learned elsewhere is presented retrospectively as the independent result of one's own method.

Are you looking for academic validation or someone capable of challenging the project?

Second, because uncritical validation would merely be a form of courtesy. We need historical linguists, Germanists, paleographers, botanists, historians of medicine, and statisticians. Not to embellish the project with their titles, but so that each can attack the part they know best. Independent research must not construct the narrative of a brilliant outsider pitted against institutions. That's a childish image. Institutions may be slow, cautious, and sometimes conservative, but they hold expertise that cannot be replaced by the availability of an algorithm. Digital tools have changed the geography of access. Today, two people can interrogate corpora, archives, and computing infrastructures that would have been difficult to access in the past. However, they have not abolished the need for specialized study.

The democratization of tools does not therefore coincide with a democratization of truth.

No. The truth doesn't become more accessible simply because the number of producible answers increases. More tools mean more hypotheses, more correlations, and more simulations. The possibilities for research increase, but also the possibility of error. We live in an era in which the problem isn't just the scarcity of information. It's the abundance of plausible structures. We can generate very convincing maps of territories that don't exist. Discipline therefore becomes more important than power. We must declare the conditions under which a proposal can be rejected, preserve failed hypotheses, and distinguish what the system has calculated from what humans have interpreted.

What does the manuscript represent for you today: a text, a system, a memory, or a mirror?

It represents, first and foremost, a living, recovered memory. If our hypothesis is correct, the Voynich preserves fragments of practical, botanical, and therapeutic knowledge from a border community whose language and customs we had lost. The value of our research lies precisely in having restored dignity and utility to this heritage: we didn't treat it as a self-contained academic enigma or a simple dataset on which to run algorithms, but as an operating manual that deserved to be spoken again. The transposition serves precisely this purpose: to reactivate that memory and make it available to the scientific community and anyone who wishes to study and apply it concretely. But the manuscript also remains a harsh mirror. Every era has projected its own obsessions and intellectual technologies onto it: cryptographers sought a code, esotericists a fantastic work. We remember that a tool is never neutral, and that our task is to bring out the historical reality of the work, not our own projections.

Aren't you worried that the image might too easily dictate the words? When faced with a drawing of a plant, you can select compatible meanings and discard those that obstruct interpretation.

This is one of the greatest risks. Iconography must not become a device that absolves any translation. If a sequence is interpreted as a 'root' simply because a root appears next to it, we have proven nothing. This is why the interpretation should emerge independently of the image and only after being compared with it. In practice, separation is never perfect, but it must remain a controlling principle. Iconographic consistency is valuable when a previously applied rule repeatedly produces terms compatible with different contexts. If, however, we modify the rule on each page to achieve what the image suggests, we are constructing a creative paraphrase.

You speak of a "transposition" of the manuscript and not a simple translation. Why?

Because a literal translation would be conceptually incorrect and reductive. The Voynich is not a story or a theoretical treatise, but a concise professional repertoire: the notebook of a herbalist who already knew the subject perfectly. For its creator, the image alongside a few lines of text served as a quick reminder to recall properties, preparations, and instructions for use. The profound value of our research lies precisely in our understanding of this functional nature of the text. What we call 'transposition' is therefore a broader and more generous process: for each folio, we do not simply indicate the reconstructed linguistic forms and the comparative lexicon, but we expand the original synthesis to decode its practical meaning. The goal is to transform those condensed and elliptical notes into an annotated interpretative reading that makes this extraordinary manual finally accessible, useful, and usable by anyone who wishes to study or apply its contents, while also offering a transparent and disputable scientific hypothesis line by line.

And what about Simona, your life partner and irreplaceable colleague in this research?

It's a broken relationship. Someone wrote and drew those pages presuming, presumably, that another human being would know how to use them. The recipient wasn't humanity as a whole. It was a person situated in a time, a community, and within a system of shared knowledge. We arrived centuries later, without belonging to that relationship. We possess the object, but not the cultural intimacy that made it legible. Technology can help us recognize traces of its structure. It can't fully restore the lost world. Therefore, every interpretation should retain a certain modesty.

If the Alpine-Germanic hypothesis turns out to be wrong, what would remain of Codex Mondragonis?

I hope it would remain a method clear enough to be reused and corrected. Statistical measurements can retain value regardless of geographical interpretation. Rejected hypotheses can prevent others from repeating certain paths under the same conditions. Comparison tools can be improved. It would also retain a lesson I consider important: today, independent research can tackle complex problems, as long as it doesn't confuse access to technology with the possession of universal expertise. If the mountain pointed to was the wrong one, we would have to say so. The willingness to abandon a conclusion is not the defeat of research. It is the condition that distinguishes research from a belief.

Inside the Voynich Manuscript: Yale University Video

Here are three insights that might interest you:

Artificial intelligence to decipher the mysterious Cypro-Minoan
It is the dawn of AI in the translation of ancient cuneiform writing
AI and "super X-rays" to reveal the mystery of the Herculaneum papyri