Nine Inca Khipus Told Me a Clean Story About the Colour Brown. Six Hundred Told Me the Truth.
THE SAMPLE THAT LIED
Nine Inca Khipus Told Me a Clean Story About the Colour Brown. Six Hundred Told Me the Truth.
The fifth essay in a series on the difference between real meaning and manufactured mystery — this time, the mystery I manufactured myself.
The Inca ran the largest empire in the pre-Columbian Americas without a writing system of the kind we usually mean. What they had instead was the khipu: a thick primary cord from which hang many knotted pendant cords, encoding information in the type of knot, its position, the twist of the fibre, and the colour of the string. Most surviving khipus — perhaps eighty-five percent — are administrative: they record numbers, in base ten, a fact first cracked by Leland Locke in 1923. The rest resist us. Whether some khipus encode language itself, phonetically, is one of the last great open questions of decipherment, alongside the Voynich manuscript and Linear A.
I came to them the way I come to everything in this series: not to believe, but to test. There is an open database — the successor to a corpus of khipus painstakingly coded cord by cord over decades — and it will run on a laptop. I downloaded it, opened it, and started counting. What follows is a small story about counting, and about the oldest trap in the trade: the sample that tells you a clean story because it is too small to tell you the truth.
I. A Clean Finding
Brown, and the ninety-one percent that wasn't
I began with colour, and with nine khipus — a coded subset, the first I pulled. The colour codes are terse: MB for a mid-brown cord, W for white, and a long tail of dyed shades and compound twists. I ran a frequency count, the dullest and most honest first move. Brown dominated: about forty-nine percent of all cords in those nine khipus were mid-brown.
That number invites a story, and the story is seductive. Andean cord is spun from cotton and from the wool of llamas and alpacas, which grows naturally in browns. Brown is the colour you get for free, the colour of the undyed fibre. So brown, I reasoned, must be the blank page — the neutral substrate against which the dyed colours carry the real information. The meaning would live in the departures: the whites, the greens, the reds, and above all the barber-pole compound cords that made up more than half the sample.
It was a lovely hypothesis. It had a mechanism, it fit the material facts, and it flattered me by arriving quickly. And then I did the one thing the hypothesis could not survive: I stopped trusting the nine, and loaded the rest.
“Brown was the colour you got for free. Surely, then, it was the blank page. It is remarkable how much a wrong idea can explain.”
II. The Correction
What five hundred and seventy-two khipus said instead
Across the full corpus — five hundred and seventy-two khipus with enough cords to count — the picture inverted. Brown did not average forty-nine percent. It averaged fourteen. And brown was not the ever-present substrate at all: two hundred and seventeen khipus — thirty-eight percent of the entire corpus — contained no mid-brown whatsoever. Only four khipus in the whole set were more than four-fifths brown.
Read that again, because it dismantles the story cleanly. A blank page cannot be absent from more than a third of the books. If nearly four in ten khipus are made without a single brown cord, then brown is not the neutral ground against which meaning is written. It is a colour like any other — chosen, or not chosen, one option among many. My nine khipus had simply, by accident, been brown-heavy. They were not the corpus in miniature. They were a biased sample wearing the corpus's clothes.
The same question, asked of a small sample and a large one. Only the large one is allowed to answer.
This is the most ordinary failure in all of quantitative work, and knowing its name does not make you immune to it. A small sample is not a small version of the truth. It is its own thing, with its own accidents, and those accidents arrange themselves into stories precisely because our minds are built to find stories. The nine khipus did not lie to me. I lied to myself, using them — the way anyone does when a convenient number arrives before a skeptical one.
III. What Actually Held
The knots that behaved like real accounts
It would be a dishonest essay that only showed the finding that broke. So here is one that held — and held beautifully — because the test I put it through was designed to try to break it.
If the numerical khipus really record genuine quantities — tributes, censuses, stores of maize and cloth — then the numbers they encode should behave like real-world counts. And real-world counts have a fingerprint. It is called Benford's Law: in genuine tallies of naturally occurring quantities, the leading digit is not evenly distributed. About thirty percent of values begin with a 1, fewer with a 2, fewer still with each higher digit, trailing to under five percent beginning with 9. Invented or random numbers do not do this; they come out flat. Benford's Law is used today to catch fabricated accounts and doctored elections.
I reconstructed the decimal value of every knotted cord in the corpus — thirty-seven thousand of them — and looked at the leading digits. They descended: about thirty-eight percent began with 1, twenty percent with 2, on down to two percent beginning with 9. The curve was not a perfect textbook fit — the 1s were over-represented, as they are in data full of small administrative counts, and indeed eighty-four percent of all values were under a hundred. But the shape was unmistakable. Five hundred years after the last Inca accountant tied the last knot, the numbers still carry the signature of genuine bookkeeping.
“Five centuries on, the knots still carry the fingerprint of real accounts. The accountants are gone; the arithmetic testifies.”
One cord, in a khipu buried in the corpus, records the number three hundred and twenty thousand five hundred and thirty-five — knotted with perfect place value, hundred-thousands down to a five-turn long knot in the units. Whatever the Inca had 320,535 of, someone counted it, and tied it, and it is still legible. That is not decoration. That is a record.
IV. The Discipline
Why this is the same essay as the last four
This series has been, from the beginning, about telling real meaning from the manufactured kind. A flood record genuinely encoded in a Japanese myth. A treasure story engineered as a hoax. A philosophical argument wearing a lost island as a costume. A government scheme whose failure was written plainly in a number it hoped no one would read. Every one of those turned on the same move: rank your evidence by how much it could be trusted, and refuse the answer that only one convenient clue supports.
The khipu taught me the inward version of that lesson. The convenient clue this time was mine. Nine khipus, forty-nine percent brown, a mechanism ready to hand — everything a good hoax has, except that no one built it but me. The corpus was the parliamentary document, the bedrock, the thing that could not be talked out of what it said. And what it said was: your clean number came from too small a place to be believed.
So the rule I keep returning to, whether the subject is a myth, a cipher, a press release, or a fistful of Inca string, has one more clause now. Convergence over consistency; the loudest clue deserves the most suspicion; residue is not the present tense. And: the sample that arrives clean and quick is the one to distrust first — especially when it flatters you.
The knots are still mostly unread. The question of whether some of them speak, rather than merely count, is open, and it is the kind of open question a person could spend a life inside. I have barely put a hand on it. But I have learned the first thing the corpus has to teach anyone who opens it, which is a thing about the person opening it: your assumptions are cheap, your sample is smaller than you think, and the data does not care what story you had in mind.
A NOTE ON METHOD AND SOURCES
The figures in this essay come from analysis I ran on the open khipu database — an openly published, cord-by-cord digital corpus descended from decades of scholarly coding of physical khipus. All computation is my own: frequency counts of colour codes, a full-corpus reconstruction of cord values from the database's knot codings, and a leading-digit (Benford) test on those values. Two honest limits belong on the record. First, my value reconstruction relies on the database's own coded readings of each knot; it therefore tests whether those scholarly readings behave like genuine numbers — which they emphatically do — rather than reading raw knots from photographs. Second, this is descriptive pattern-analysis by an independent researcher, not peer-reviewed khipu scholarship; the specialists in this small, careful field are the people to correct and extend it. The decipherment of the khipu is a living project embedded in living Andean communities, some of which still keep their khipus as patrimony. It deserves to be approached as their heritage, not merely as a dataset.
For reproducibility. All figures come from the Open Khipu Repository (OKR), version 2.1.0 (DOI 10.5281/zenodo.18025748), accessed 21 July 2026, analysed via its SQLite distribution; the table and field names below are those of that file, and a reader using another distribution should map them accordingly. The brown analysis: table ascher_cord_color, primary colour field COLOR_CD_1, counting the value ‘MB’ (mid-brown) per cord, grouped by KHIPU_ID, restricted to khipus with at least eight colour-coded cords (572 khipus; mean brown fraction 13.8%; 217 with zero MB). The pilot that misled me was a separate nine-khipu subset (coded AS1–AS9), drawn as its own file rather than as a filter on the full table — part of how the bias entered. The Benford analysis: table knot, field knot_value_type summed per CORD_ID to reconstruct each cord’s integer value, leading digit taken across the 37,111 cords whose value exceeds zero. Schema names are those of the distributed file; a reader should confirm them against their own download before running the queries.
— END —
Mystic Quill | Research & Writing by Selva Ganesh K | 2026
mysticquill.blogspot.com
Comments
Post a Comment