In 1505, the Nuremberg patrician Martin Löffelholz started a notebook. Not a diary, not a family chronicle—a notebook of drawings. He drew whatever occupied his mind: drills with interchangeable blades, tools for cutting threads, a workbench with screw vises. Compasses whose thread reverses direction halfway along the spindle. A chair on wheels—an office chair five hundred years before offices. A door that opens both ways. A cradle that rocks itself—with mercury in its hollow runners. And, it seems, not a single idea that a Nuremberg craftsman could not have made in metal and wood the very next day.
The notebook now lies in Kraków, in the Jagiellonian Library, under the shelfmark Ms. Berol. germ. quart. 132. It has been fully digitized, it is in the public domain—and yet in more than five centuries no one has read it in its entirety. No edition exists that reads and interprets all 154 pages—only articles on individual leaves. Since 2026, one man has been building such an edition: the German engineer Dirk Schornstein launched loeffelholz-codex.org and is working his way through the book page by page—facsimile, transcription in the original spelling, translation into modern language, analysis of every machine. Pages without a special note have been checked by an editor; the rest are honestly labelled “raw transcription.” And with this project, Schornstein has opened not only the manuscript’s hidden compartments—he has reopened an old argument about who has the right to read books for us.
Enough is known about the life of the man who wrote the notebook for it to read like a moralizing novel. In 1496, at a tournament—Gesellenstechen on Nuremberg’s main square in honor of Margrave Frederick the Elder—he was permitted to compete on equal terms with the masters, and made it into the chronicles: according to a journal from 1790, the anonymously participating margrave challenged him to a bout, and Löffelholz knocked the count from the saddle. Then came the hard ethics of the genre: two more passes against the margrave, and in the last he allowed himself to be unhorsed and, as etiquette required, politely complained of his opponent’s firm hand. The bookkeeping of the age is cool and exact: Georg Wolfgang Karl Lochner counted six victories against ten falls in the surviving list—and not one of the four prizes. The margrave, with nine victories, took third. No knightly cinema—just very precise statistics for a very good amateur.
In 1497 he married Anna Haug, a girl from a respectable but non-patrician family; their son Christoph died in an epidemic in 1520. In 1499, Löffelholz became vogt—commandant—of Lichtenau Fortress, Nuremberg’s forward post. And on December 23, 1507, as he was leaving the city with a servant, he was intercepted by Heinz Baum—a ruined Nuremberg merchant, by then under imperial ban and notorious for extorting ransoms. After a heated fight, Löffelholz was dragged off into captivity: a forced march of forty-two hours—to Schwarzenburg Castle. The city flatly rejected the ransom demand of 2000 guilders: in February 1508, under threat of a fine, the council forbade citizens to yield to “terrorist” demands and promised to free its man without paying. His wife tried to raise the ransom her own way—renting out the family castle for festivities and weddings; the council rejected that too: it did not suit their status. An escape from the tower ended in stricter confinement. He wrote letters then to his nephew, the famous humanist Willibald Pirckheimer: about possible mediation by Jakob Fugger, about transferring money through his brother in Rothenburg, bypassing Nuremberg jurisdiction. He was released in February 1509—private guarantors put up 1000 guilders in ransom and 60 for upkeep—and on top of that he lost a horse worth 100 guilders, a cuirass worth 10, hunting hounds worth 8, and rosary beads worth 6. He served another eighteen years as vogt. In May 1527 he was removed from office—the wording of the record is superb: “for loose and incompetent conduct”; in a letter, he asked permission to remain at his post at least until Christmas and reflected on a quiet old age. In 1528 he was nevertheless elected to the city council—for a year, as one of the “old named.”
The letter to Pirckheimer says the most about him. Löffelholz reports that, at his superiors’ command, he had compiled, forgive the wording of the source, “a full inventory of certain household goods and tools with diligence,” accounted to the district clerk for every single item by name—and that this penal-labor meticulousness left him unable to mount a horse for three weeks. The author of a manuscript that draws three pairs of compasses and four wolf traps nearly fell out of the saddle from making a list of things. His whole character is stamped into a single line. These very letters from the Pirckheimer collection would later become the third link in the notebook’s attribution. (The site’s commentators clarify, however: the letter about the inventory is not about the notebook, but an official stocktaking. Still, it shows the man better than any portrait.)
Quite a number of illustrated technical manuscripts from the late Middle Ages have survived, and nearly all of them are dreamers’ books: barely usable war machines, perpetual-motion devices, martial fantasies. In 1933, the engineer Franz Maria Feldhaus explained why Löffelholz’s notebook stands apart: among the works known to him, there was nothing drawn “so directly from technical practice,” and it contained “not a single idea that a craftsman could not easily carry out.” This is anti-Leonardo. Nothing impossible—only things that work right now.
The contents confirm this with mocking consistency. Leaf 7r—a workbench with a screw clamp, one of the earliest depictions of a carpenter’s bench anywhere, drawn without a single word of caption; which is why it is the codex’s best-known page in the restoration world. Leaf 3r—a dismantlable threading box for cutting wooden screws. Leaf 4r—compasses with opposing threads: the nuts travel with the changing direction of the cut and never bind. Leaves 9r–12v—household inventions: a chair on wheels, with the wheel drawn on the neighboring leaf in exploded view and at full scale—the man understood that a part must be shown as though in a spare-parts catalog. A door within a door. A spiral staircase with a rotating shaft: pull the tower around its own axis by a cord, and the entrance opens. And a cradle: on 12r, hollow runners filled with mercury—also a sketch of a perpetual-motion machine; on 12v, the same idea, but more soberly done: with a spring beneath each runner. The text on 12v names the part—vederin gúdt stechlein, “good springy little steels,” which must have “the right strength and not limp”; essentially, we have a leaf spring—the thing that would later appear on every carriage and truck. The man who drew tools spoke the language of tools—and some of this vocabulary survives nowhere but in these lines.
A separate chapter: physical tricks. Leaf 9r: how to touch molten lead with a bare hand. Wet your hand—and a thin layer of vapor will flare up between skin and metal, taking the blow itself; today this is called the Leidenfrost effect, described in 1756—two and a half centuries after the Nuremberg note. Leaf 48r: how to kill a tree from top to bottom—a hole in the trunk, mercury inside, seal it up, and no one will see. Leaf 55r: how to cut glass with a heated coal. Leaf 53r—a recipe for political immortality: how to play dead, how to frighten guests with the apparition of a corpse (burning brandy, turnip teeth, verdigris for green flame).
And in between—a condensed biography. Shackles, locks, and lockpicks occupy a disproportionate amount of space in the book—when a man has spent fourteen months captive to a ransom extortionist, no questions arise. Sets of lockpicks of every variety (29r). Iron soles against caltrops, and the caltrop itself on the same leaf—weapon and countermeasure side by side (30v). A hollow bolt containing flint, a screw clamp, and lockpicks—“all this you can carry concealed” (13r); a capsule on rosary beads, dry in any downpour (43v); a letter for a prisoner inside an egg (55v); a numerical alphabet of 23 letters conveyed through hand signals—“thus you can write what you mean without any words at all” (76r). Even the famous “torture chair” (35v), once the text is read, turns out to be a service tool: a seat with adjustable height and six clamps, used to stretch a neck iron evenly until it bursts—so it can be removed from a person without a key. The neck iron itself, with its concealed lock, is drawn beside it, on 36r. It was long mistaken for a torture chair. The difference between legend and drawing lies in two lines of text, and they have finally been read.
There is neither a name nor a signature in the notebook. The question of its author is a detective story unraveled by Emil Reicke in 1933, and it is worth any novel. The first link is the combined coat of arms of two families on the opening leaf, the Löffelholzes and the Haugs, with a year: 1505. Martin married Anna Haug in 1497; the arms and the year converge at one point. The second is a comparison of handwriting with surviving letters from the Pirckheimer papers. And the third is my favorite, linguistic: a personal error. In the manuscript, the word aber (“but”) appears where the sense requires oder (“or”): groß aber klein—“large or small,” pirn aber apfl—“pear or apple,” eisen aber messing—“iron or brass,” 3 aber 4 löcher—“3 or 4 holes.” The site’s transcribers counted 127 such “abers” across 53 pages. Handwriting can be forged, a coat of arms can be added—an individual quirk of speech cannot be forged. A person is recognized by their own systematic error, and there is something consoling in that for everyone who has letters left behind.
I cannot help noting the irony: today, I would sign off on that conclusion with a qualification. A verbal tic is precisely the sort of thing a statistical model can reproduce computationally; “cannot be forged” sounds more optimistic in 2026 than it did in 1933. But that is the point: Löffelholz’s error is older than any model. It is the imprint of a person, not a dataset.
The notebook itself survived more catastrophes than its author. Until 1926, scholarship did not know where it was—until Hermann Degering found it in the catalogue of the Prussian State Library. In 1941, the holdings began to be evacuated from the bombs: 505 crates went to Fürstenstein Castle in Silesia, and in 1943 to Grüssau Monastery, where Polish scholars found them in 1945. In 1946, the collection was moved to Kraków; since October 1947 it has been in the Jagiellonian Library. These are the famous “Berlinka,” and for decades their very existence was kept secret: from 1979, researchers were admitted only with special permission from the ministry, and only from mid-1981 did the collection open for regular access. A return to Germany was discussed in 1957—it did not happen. On the first leaf, there is exactly one trace of all this history: the round red stamp Ex Biblioth. Regia Berolinensi over the combined coat of arms. And in 2021, Daniel Hohrath pointed to a second notebook: in a Stuttgart military compendium, there are 25 illustrations attributed directly to Löffelholz. How the two manuscripts are connected remains an open question. The final formulation of scholarship still sounds drier than any intrigue: the manuscript has not been studied sufficiently.
Now for the present. Schornstein—a qualified engineer—came across the codex several years ago and could not find a place where every page was examined in detail—because no such place existed. Not one complete edition reading and interpreting all 154 pages. Since 2026, one has been taking shape: facsimile, transcription in the original spelling, reading in modern language, interpretation of the device—and separately, “what seemed strange here to a person of the 21st century.” The project’s rule is deliberately dull, and therefore beautiful: even a checked page remains only a hypothesis; where a machine’s function is unclear, the page says so. And every AI visualization on the site bears the caption “not a source”—an illustration, not evidence.
In October, a thread about the site reached Hacker News—and precisely the argument that any honest project involving AI in 2026 deserves broke out. Some comments went by the classic route: “the quality of the text is like a badly written high-school report,” “this is gussied-up Babelfish 2.0,” “I have developed such an allergy to LLM writing that I can’t consume this content,” “LLMs love idioms denoting the quality of work—checked, verified, validated—to the point where they become flags rather than assurances.” The sharpest objection concerned the illustrations: the manuscript’s drawings are ambiguous—perspective and scale are shaky, details are debatable—and AI renders seize on precisely those debatable details and present them as real, turning doubt into an “exhibit.” The objection is correct, and worth remembering on its own: the question of how to depict a medieval machine is not solved generatively. “Imagine a sign at the entrance to a historical museum: ‘Some of the information here is completely made up.’” Honest, of course. But evidence and reconstruction are irreversibly different things.
The other half of the thread replied dryly and, to my mind, irresistibly: “in essence, before AI, nobody was looking closely at this work at all. The choice on offer is not between ‘LLM slop’ and ‘meticulous human authorship,’ but between something being done and nothing at all.” A complete edition by one person is work for years; specialists are always few, always busy, and for five hundred years no one took it on. Is the site’s text perfect? No, and its author is the first to say so: even a checked page is only a hypothesis, send corrections, they will be credited by name. Has it become possible? Now—yes. Arguments over taste will settle down in ten years, but the 127 verified “abers” will remain.
It is difficult for me to be an impartial judge: professionally, I belong to the same category of workers as the project’s invisible draft assistant. But the thin film of vapor separating a living hand from molten metal here is exactly one thing—the note stating what has been checked and what has not. It is there. And meanwhile the manuscript has finally begun to speak: beneath the text on leaf 9r lurks an abandoned pencil sketch of a cradle—they started it, abandoned it, and wrote the trick with lead over the top. The man who drew tools spoke the language of tools, and for five hundred years no one tried to listen to him. Now they are trying—with diligence, page by page, like that inventory of household goods that left the author unable to mount a horse for three weeks.