A 15th-century codex written in an unknown script — illustrated with botanical drawings of plants that match no known species, astronomical diagrams, and pages of naked women bathing in green liquid — has defeated World War II codebreakers, professional cryptanalysts, and modern machine-learning systems alike. It is roughly 240 pages of fluent, confident, beautifully penned text in a writing system that appears nowhere else on Earth, and after more than a century of serious attack, no one can read a single sentence of it with any consensus.

What we actually know is more than people assume, and that is what makes the mystery sharper rather than softer. The manuscript is real and physical, held since 1969 at Yale University's Beinecke Rare Book and Manuscript Library as MS 408, named for the antiquarian Wilfrid Voynich who acquired it in 1912. In 2009, the University of Arizona radiocarbon-dated the vellum to between roughly 1404 and 1438. This is hard data: the calfskin pages are genuinely early-15th-century. The iron-gall ink is consistent with the period. Whatever the Voynich is, it is not a modern hoax printed on aged paper — it was written when its parchment was fresh, six centuries ago.

The text itself is where the trouble begins, because it behaves like a real language and like no real language at once. The script — dubbed 'Voynichese' by researchers — has somewhere between 20 and 30 distinct characters. The crucial point, established by statistical analysis, is that it is not random gibberish: the word-length distribution, the way certain symbols cluster at the beginnings and ends of words, and the overall statistical structure obey what linguists call Zipf's law, the same frequency pattern that governs every natural human language. Strings repeat. Some 'words' appear over and over, others once. This is exactly what meaningful text looks like and exactly what randomly scribbled nonsense does not. And yet no scholar has matched it to any known language, cipher, or shorthand.

The roster of people who have failed is the real evidence of how hard this is. William Friedman — the most important American cryptologist of the 20th century, the man whose team broke the Japanese PURPLE machine — spent years on the Voynich and concluded only that it might be an artificial or invented language. NSA cryptanalysts took runs at it. Computational linguists have fed it through every statistical tool available. In 2018 there were splashy headlines claiming an AI had identified the underlying language as Hebrew; the claim did not survive scrutiny and is not accepted. Every few years someone announces a solution, and every time the field examines the work and finds it does not actually let anyone read the next page. That is the iron test the Voynich keeps passing: a real decipherment lets you read text you have never seen before. No proposed solution has ever done that.

The skeptical-but-fair reading splits into a few live possibilities, and intellectual honesty requires holding them all at once. One: it is an encoded real language, a cipher so clever or so idiosyncratic that no one has cracked the key. Two: it is a 'constructed' or artificial language, a private invented tongue, which would explain why it matches no known one. Three — and this is the explanation that has gained real ground — it is an elaborate, meaningless fabrication, possibly produced to defraud a buyer (the Holy Roman Emperor Rudolf II is documented as a possible early owner, and the period had a market for occult and alchemical curiosities). The hoax theory is strengthened by the fact that some scholars have shown the Zipf-like statistics can be reproduced by simple table-and-grille generation methods a medieval forger could have used. But the hoax theory has its own problem: faking that much internally consistent text, with stable grammar-like rules, by hand, six hundred years ago, without a computer, would itself be a remarkable and labor-intensive feat with no obvious payoff matching the effort.

What keeps the Voynich honest as a mystery — as opposed to the usual ancient-aliens nonsense — is that you can examine the primary evidence yourself. Yale digitized the entire manuscript at high resolution and put it online in its digital collections; full scans live on the Internet Archive. There is no gatekeeper, no missing original, no 'they won't let you see it.' Every researcher and every armchair codebreaker is staring at the exact same pixels, and the document has simply refused to yield to any of them. That is extraordinarily rare. Most 'unsolvable' artifacts are unsolvable because they are lost or guarded. This one is in front of everyone, fully lit, and still silent.

The unresolved question cuts in two directions, and both are uncomfortable. If the Voynich is meaningful, then a person in early-1400s Europe encoded knowledge so well that six centuries of the best codebreakers on the planet — wartime, governmental, and computational — have not recovered one verified word. If it is meaningless, then a forger built a hoax so structurally convincing that it has fooled experts into treating it as language for over a hundred years. Either answer describes a feat we cannot fully explain. The book is open on a table in New Haven. It has been read by no one, and it is not telling.