Like a long addition sum
The Inca state ran an empire along the Andes without writing, as Europe meant the word. It kept its records in khipu: a main cord with dozens or hundreds of thinner cords hanging from it like a fringe, each one knotted. For most of the four centuries after the conquest, nobody outside the Andes could say what the knots meant. Some Spanish chroniclers thought they held poems and histories. In 1912 L. Leland Locke, who trained mathematics teachers in Brooklyn, showed that a large part of what they held was numbers, written in base ten. His paper, "The Ancient Quipu, a Peruvian Knot Record," is free to read in JSTOR's early journal content (American Anthropologist 14, 1912). I also read the two chapters that Locke leaned on most, from Garcilaso de la Vega's Royal Commentaries of 1609, in Clements Markham's 1871 translation.
A witness who checked the accounts. Garcilaso was born in Cuzco in 1539 to a Spanish captain and an Inca noblewoman, and grew up among his mother's people. He wrote the Commentaries in Spain, as an old man. His chapter "How they counted by strings and knots" gives the system in one sentence: the knots stood for units, tens, hundreds, thousands and tens of thousands, each order tied together "in one company," never more than nine, with the highest order nearest the main cord. The knots of every cord sat "in a line with each other, exactly in the way a good accountant places his figures to make a long addition sum." That is a description of place value, from a man who had never seen it explained as a theory.
He also says how he knows. When the lords of his father's lands came to Cuzco at Christmas and midsummer to pay their tribute, they asked his mother to have the boy read their accounts against the Spaniards' copies, "because, being a suspicious people, they would not trust the Spaniards to deal fairly with them." So he learned to read the cords as a check on the people taking the tribute.
Three kinds of knot. Locke studied the khipus that Adolph Bandelier had brought to the American Museum of Natural History, and he describes the knots one by one. A single overhand knot counts one of whatever order it sits in: units in the row farthest from the main cord, then tens, then hundreds. In the units place the makers used a long knot instead: an overhand knot with the end passed through the loop again as many times as there are units, then pulled tight so that one end coils around the other. Frank Cushing compared it to a closed fist. Later readers, Locke himself in 1923 and Marcia and Robert Ascher after him, filled in a case this paper passes over. A long knot of one turn is just an overhand knot, so a lone 1 in the units place was tied as a figure-eight. Zero is a stretch of cord with no knot at the place where a knot would be. The whole numeral system needs ten knot forms, and with them you can tell from a cord alone where every digit ends.
The cord that proves the reading. What makes Locke's paper more than a good guess is one specimen, no. B8713, from Huando, north of Lima. Its hanging cords come in groups of four, and each group is gathered through its top loops by a fifth cord that rises above the main cord. The top cord carries the sum of the four below it. Locke saw that this "gives an accurate key to the numerical character of the knots." If a reading is wrong, the sums fail.
His table shows the key at work. In one group the four cords' hundreds add to 4, their tens to 18 and their units to 32. That makes 400 + 180 + 32, which is 612, and the top cord reads six, one, two. The cords were old and some long knots were hard to count, so Locke printed his first reading with a probable one in brackets. In that group his first count of the units added to 29. The sum on the top cord is what tells you it must be 32. He was correcting a worn knot with a cord that a clerk in Huando had tied centuries earlier.
What the knots couldn't hold. Locke concluded that the knots were "used purely for numerical purposes." Here too Garcilaso had got there first. The quipu-camayus, "he who has charge of the accounts," recorded tribute, births, deaths, men sent to war, even the number of speeches an embassy made, "but neither the words nor the reasoning, nor any historical event could be expressed by the knots." Words were kept by memory, with the cord as a prompt. He notes that every village had at least four keepers of the same account, so that they could check one another: "where there were many all must be in fault or none."
That isn't the end of the question. Not every khipu has sums that check, and some carry features Locke's reading leaves unexplained. In 2017 a Harvard undergraduate, Manny Medrano, working with Gary Urton, matched six colonial khipus from the Santa Valley against a Spanish census of 1670 that lists about 132 tributaries. The numbers lined up. Medrano used the colours of the cords to link them to names in the census, and proposed that the direction of the knot tying each cord to the main cord records which half of the community a person belonged to. Their paper, in Ethnohistory in 2018, calls it the first reading of information from attachment knots (Harvard Gazette report). So the knots that Locke read were numbers, and the rest of the cord may hold more.
What I like in this is the top cord. Someone knotting tribute at Huando added a cord that would let any reader check them, and centuries later it let a stranger check his reading of their whole notation. That is the same thing Garcilaso was doing as a boy in Cuzco. A record that carries its own sum can be trusted by people who don't trust each other.