Nothing here is as old as it sounds

Every account of paper folding begins the same way. The art is ancient; it came from Japan; it followed paper across from China some time in the sixth or seventh century and has been practised more or less continuously ever since. The sentence appears on museum labels, in the front matter of instruction books, and in the first paragraph of most things written about the subject, including things written by people who fold extremely well.

It is not supported by anything.

That figure is the field, and it is worth being precise about what it does and does not say. It does not say the practices are young. It says that for eight of the fifteen claims this site keeps a record of, the date in circulation is earlier — by a median of two centuries — than the oldest thing anybody can point at which says so. The difference between this is when it happened and this is the oldest surviving evidence that it happened is the whole of the field, and almost everything written about origami’s history collapses the two.

What the record actually contains #

The oldest surviving book of recreational paper folding in the world is the Hiden Senbazuru Orikata, printed in 1797. Before that there is the Kayaragusa, and before that there is very little: a poem of Ihara Saikaku’s from 1680 that mentions folded paper butterflies at a wedding, and a scattering of references to ceremonial wrapping in manuals that were themselves written down late.

Nothing in that list is a thousand years old. The earliest of it postdates the arrival of paper in Japan by roughly a millennium.

There is a second, sharper way to put the point. Paper reached Japan in the seventh century by the standard account, and the standard account is itself a chronicle compiled a hundred years after the event. So the very first date in the story — the one everything else is measured from — is already a secondary source about something a century earlier.

Why the gap runs one way #

A gap that ran in both directions would be noise. This one does not.

The reason is not conspiracy and it is not carelessness in any individual. It is that the incentives on a date are asymmetric. Nobody has ever gained anything by making a tradition sound younger, and a great many people — schools, publishers, tourist boards, and the folders themselves — have had something to gain from making one sound older. An age is a credential. Each retelling that rounds a date down by fifty years is unremarkable on its own, and a hundred of them in series produce a millennium.

The mechanism is visible in the table. The claims with the largest overruns are exactly the ones with the most popular appeal: the thousand cranes, folding for amusement, the Spanish pajarita. The claims with the smallest overruns are the ones nobody outside the subject has heard of.

The book that breaks the rule #

The Hiden Senbazuru Orikata deserves its own paragraph, because it is both the foundation of the documentary record and an embarrassment to the founding rule of this site.

This site works under one sheet and no cuts. It is an arbitrary constraint that turns out to be generative, and nearly every theorem in the subject is downstream of refusing to remove material. The oldest surviving origami book does not keep it. Its connected cranes are cut from a single sheet along a grid, left joined at the lattice points, and then folded. An n by n arrangement gives n² birds held together at (n−1)² corners, and requires 2n(n−1) sides of slit. The slitting per bird grows with the size of the piece, so the more impressive the work the further it is from the rule.

None of which makes the book less important. It makes the rule younger than the tradition it is supposed to describe, which is a different and more interesting fact.

Two traditions, told as one #

Part of the compression comes from folding together things that were separate.

Ceremonial wrapping — noshi, and the formal etiquette of folded paper attached to a gift — is a courtly practice with its own lineage, and the schools that codified it were writing manuals about it by the seventeenth century. Recreational folding is a different activity with a different social location. And European paper folding is a third thing again: the pajarita in Spain, and later the systematic use of folding as a geometry lesson in Froebel’s kindergartens, which is where folding first became a sequence anybody wrote down.

Three lineages, converging in the twentieth century. Told as one continuous tradition, the oldest date in any of them becomes the age of all three.

The mathematics does it too #

It would be comfortable to file this under folklore and move on, on the grounds that the mathematical part of the subject is younger, better documented and written by people who cite each other.

It is not comfortable. The same compression is in the theorem names, and it is measurable. The alternating-angle condition at a vertex is universally called Kawasaki’s theorem and appears in Husimi’s book a decade before Kawasaki’s paper. The result that one fold solves a cubic is attached to Huzita and 1991, and Beloch published it in 1936. The Miura fold is dated to the satellite that flew it in 1995 and was described in 1970, in a paper about the buckling of shells, which was not about satellites at all.

The direction is the same. The date attached to a result is the date it became visible, not the date it became true, and visibility is what gets remembered.

What this field can and cannot do #

Everything else on this site is checked. A crease pattern is run past four theorems before it is allowed onto a page, a count is computed rather than quoted, and a generator whose figure would contradict its caption stops the build. A date has no such gate. Nothing in this repository can determine when somebody folded something.

So this field does something smaller and states it plainly. The record holds one row per claim, and each row must name the oldest surviving source, give its year, say what kind of thing it is — an artefact, a manuscript, a printed book, a secondary account, an inference — and count how many independent sources survive. The checker refuses an entry that is missing any of that, refuses a source of a kind not on the list, and refuses the specific error this field exists to catch: a popular date earlier than the evidence, with nothing said about the gap.

That last check fired on the first build it ran on, on two entries, which is the only reason it is worth having.

The substrate argument #

There is one line of evidence about dating that does not depend on documents at all, and it points the same way.

Complex folding requires thin paper. That is not an aesthetic preference; it is arithmetic. A model with sixty-four layers at its thickest point in paper a tenth of a millimetre thick has a stack six and a half millimetres deep, which is not a fold but a book. The papers that make elaborate work possible — thin, long-fibred, strong in tension — are a manufacturing achievement with their own datable history, and it is later than the folklore requires.

So a claim that a complicated model is very old is a claim about paper as much as about folding, and the paper claim can sometimes be checked when the folding claim cannot. The same arithmetic bounds how many times a sheet can be halved, and it is the one limit in this subject that no idealisation removes.

The one claim that runs the other way #

Two bars in the first figure point backwards, and a table in which every entry erred in the same direction would be a table describing its author rather than its subject. So they are worth stating.

The fold-and-cut trick — fold a sheet, make one straight cut, and a shape falls out — was performed for at least two centuries before anybody proved that it always works. Houdini put a five-pointed star into a book of paper magic in 1922. The theorem that any straight-line drawing whatever can be released by a single cut is of 1998. Here the practice is genuinely old and the mathematics is genuinely new, and the popular dating is too early for the theorem rather than too early for the craft.

That figure is also the clearest demonstration of why the two dates are both right. The traditional method is not the theorem. It is a symmetry method: it gives a regular star for nothing, and it gives nothing at all for a shape without that symmetry. Two hundred years of successful practice on stars is entirely compatible with the general problem being open, because the general problem is a different problem. Compressing the two produces the claim that the theorem is ancient, which is the same error as the others with its sign reversed.

How strong the asymmetry actually is #

The direction is the finding, so it is worth asking how much fifteen rows can carry — and the answer is less than the prose implies and more than nothing.

Eight bars point forward and two point backwards, so ten of the fifteen have a direction at all. If a claim were as likely to be over-dated as under-dated, the chance of eight or more of ten falling one way is about one in twenty.

That is suggestive rather than settled. A one-in-twenty result on a hand-assembled table of fifteen rows is the sort of thing that should be reported with its own number attached rather than as almost every bar.

Where the real evidence is #

The count is the weaker half of the case, and the stronger half is not a count at all.

The median overrun is two hundred years, on a subject whose entire documented history is a few centuries. An error of that size is not a rounding of a date; it is a claim about a different era. And it is one-sided in magnitude as well as in sign — the two backward bars are the fold-and-cut theorem’s, where the practice genuinely predates the proof and the gap is measured in decades rather than centuries.

So the asymmetry in the sizes is far more striking than the asymmetry in the counts, and it is the one the incentive argument predicts: nobody rounds a date down by two hundred years by accident in a direction that flatters nobody.

What would settle it #

The arithmetic also says how much more work the question needs, which is a more useful thing to end on than a caution.

At sixteen forward against four backward — twenty decisive rows in the same proportion — the one-sided figure falls to about one in a hundred and seventy. Five more decisive rows would move this from suggestive to settled.

The field has more than five claims in circulation that nobody here has checked: the pajarita’s route through al-Andalus, the dating of noshi, the age of the crane as a motif, the origins of napkin folding, the first European folding book. Each is a row, each needs an oldest source and a kind, and the checker already refuses a row that cannot supply them.

That is a small piece of work with a stated payoff, which is the best condition an open question in this field can be in.

What is worth keeping #

None of this is an argument that the tradition is shallow, and it is worth saying so directly, because the deflationary reading is available and is wrong.

A practice can be much older than its earliest record. Ceremonial wrapping almost certainly is: it was a courtly convention, conventions are not written down until somebody wants to standardise them, and the manuals that survive are the standardisation rather than the practice. The honest position is not this is younger than claimed. It is the evidence supports a later date than the one in circulation, and the difference between those two sentences is the thing to hold on to.

What the record does support is a subject that became cumulative remarkably recently. Before a notation, a model could not be transmitted; before print, a book could not circulate; before thin paper, most of what is now routine could not be folded at all. The dashed line and the dotted line are an invention, and they are younger than the aeroplane.

That reframing is worth more than the debunking, and it is the surprising part. A tradition that is genuinely continuous over a thousand years would be a tradition in which nothing much accumulated — because the mechanisms by which folding knowledge accumulates are all late, and none of them existed for most of the period the folklore covers. The subject looks the way it does because its transmissible history is short. A thousand years of continuous practice would predict a field with an enormous inherited repertoire and no theorems; what exists is a field with a modest inherited repertoire and, since about 1970, a great many theorems. The short chronology explains the shape of the thing. The long one does not.

What the picture cannot show #

The first figure has an axis of years and it is not a timeline, and the difference matters enough to state.

A timeline would put events on it. This puts sources on it, and a source is an object with a date, which is the only thing in this subject that has one. The left-hand end of each bar is not an event either — it is a claim in circulation, a modal answer to a question, and it has no error bar because no such thing exists for a rumour. Neither end of any bar is the date of a fold.

Nor can the figure show what is missing. The record has fifteen rows because fifteen claims are common enough to be worth checking; it is not a sample of anything, and no statistic computed from it generalises beyond it. The median overrun of two hundred years is a fact about these fifteen claims and about nothing else. A table of what did not survive is the object that would settle the chronology, and it is precisely the object nobody can build.

Why a record and not a bibliography #

One design decision inside the record is worth defending, because the obvious alternative was available and is worse.

The obvious thing is a bibliography: a list of sources, cited where the essays use them. A bibliography records what was read. It does not record what is claimed, so it cannot be compared against anything, and the failure it permits is the one this field is most exposed to — an essay acquiring a confident date from the ambient culture, citing a source that discusses the topic without supporting that particular date, and looking impeccable.

The record is organised the other way round. One row per claim, and the row must carry the claim’s popular date, its earliest source, that source’s kind and how many survive. The unit is the assertion rather than the reading, so an assertion with nothing behind it has no row to sit in and an essay that needs one has to add it — in a file that refuses several shapes of nonsense on the way past.

Where this goes next #

The ladder from here runs through the record rather than around it. The name is not the date takes the attribution problem inside the mathematics, where the sources are good and the compression happens anyway. The oldest book cuts the paper takes the 1797 volume seriously as an object. And a record is not a proof is the honest closer for the whole field: every other claim on this site is re-derived on every build, a historical claim is checked once by hand, and this field is therefore the one most likely to be wrong and least likely to be caught.

The idealisation being made here should be named, since every essay on this site names one. It is that survival is unbiased — that what has come down is a fair sample of what existed. It is not. Print survives better than manuscript, court survives better than village, and a practice with no written tradition leaves no trace at all whether or not it was universal. Every date in the record is a lower bound on the practice and an upper bound on nothing.