MYS0D4V7T824S7F·Mystery · script

Inka khipus: the knotted-cord records

Unknown author · CC0

Region: Andes, chiefly Peru, Bolivia and northern Chile⁠[2]
Main period: Inka empire, 15th to 16th century; some colonial and later khipus⁠[3]
Numbers deciphered: 1912, by L. Leland Locke⁠[4]
Open corpus: 702 khipus with public data and KH numbers (OKR v2.1.0, December 2025)⁠[2]
Largest holder: Ethnologisches Museum, Berlin (154 khipus in the OKR registry)⁠[2]
Largest computational study: 650 khipus; sum relations in at least 74 per cent⁠[8]
Status: Unexplained
Questions: 0 of 4 answered

People Gary UrtonAnthropologist; founded the Harvard Khipu Database Project Manuel MedranoSanta Valley decipherment; corpus-wide data science Sabine HylandAnthropologist; Collata and hair-khipu studies Ashok KhoslaIndependent researcher; Khipu Field Guide Jon ClindanielAuthor of a grammar of non-numerical khipu signs

All 7 people ›

L. Leland LockeDeciphered the numerical knot code Marcia AscherMathematician; recorded and analysed khipus with Robert Ascher

Catalog keyMYS0D4V7T824S7F

A khipu is a main cord with pendant cords hanging from it, often with subsidiary cords tied to the pendants. Knots on the cords record numbers, and the cords vary in colour, fibre (cotton or camelid wool, sometimes human hair), direction of spin and ply, and the way they are attached⁠[1]⁠[2]. Khipus were in use in the Andes for centuries, and most surviving examples come from the Inka empire and the decades after the Spanish conquest of the 1530s⁠[3].

The numerical part is solved. In 1912 L. Leland Locke showed that clusters of knots at fixed heights record units, tens, hundreds and higher powers of ten, with the type of knot marking the units⁠[4]. Marcia and Robert Ascher then showed that many khipus contain sums, in which one cord's value totals those of others⁠[1]. Studies of archives from Puruchuco and Inkawasi showed values being added and passed between levels of accounting⁠[5]⁠[6].

The rest is open. Colonial writers describe khipus that recorded histories, genealogies and messages, and khipu keepers who read them aloud⁠[7]. If some khipus record language, the code lies in features other than the knots. The Open Khipu Repository, released openly and now holding public data on 702 khipus, makes it possible to look for that code across the corpus rather than one khipu at a time⁠[2]⁠[8]. Of the 4 questions on this site, 0 are answered. The most likely answer to “Do any khipus record language, or only numbers and categories?”, not yet confirmed: khipus record numbers and the categories they belong to; colour, fibre and attachment label what is counted….

Key questionsset by the editors

Do any khipus record language, or only numbers and categories?Most likely: khipus record numbers and the categories they belong to; colour, fibre and attachment label what is counted…

Is the khipu number system positional with a zero, or non-positional?Most likely: khipu numbers are base-ten positional, with an empty position standing for zero

Who made and read khipus?Most likely: khipu making extended beyond the elite to commoners and local communities

All 4 questions

The object

Khipus survive mainly from coastal burials and storage sites where the dry climate preserved them, and are now scattered among museums in Peru, Europe and North America. In the OKR registry the largest holders are the Ethnologisches Museum in Berlin, the Museo Nacional de Arqueología, Antropología e Historia del Perú, the archaeological complex of Inkawasi and the American Museum of Natural History⁠[2]. Many have no recorded findspot. Archaeologically excavated groups include 34 khipus from the Inka storage site of Inkawasi in the Cañete valley and 21 from Puruchuco near Lima⁠[6]⁠[5].

First page of The khipus of the American Museum of Natural History read by Locke — The khipus of the American Museum of Natural History read by LockeL. Leland Locke, 1912. Public domainArtifact

The OKR's relational database records each khipu's primary cord, every pendant and subsidiary cord with its position, fibre, colour, twist, length and attachment, every knot with its type, direction and number of turns, and the grouping of cords into clusters⁠[2]. Its tables include the Aschers' colour codes and cord records, and it carries contributions from many recorders under an MIT licence⁠[2].

Map of the regionOpen the full map

+−

CuscoInka capital

Within 1 km

Central AndesRegion

A region

Inkawasi, Cañete valleyInka garrison and storage site

Within 10 km

PuruchucoInka administrative centre

Within 100 m

San Pedro de Corongo, Santa valleyColonial town

Within 1 km

MappedWithin 1 kmA regionNatural Earth

Timeline

Timeline

Squeezed where little happened. Each mark is a published claim, by how it stands:AcceptedMost likelyStill possibleDisputedRuled out

What is known

The knot code is a base-ten positional record. Single knots in a cluster mark tens, hundreds and higher places by height on the cord; long knots with two to nine turns, or a figure-eight knot for one, mark the units; an empty position stands for none⁠[4]⁠[1]. A 2025 preprint argues that the system is better described as non-positional and without a concept of zero, like other ancient numeral systems⁠[9].

Arithmetic structure is common. Medrano and Khosla's 2024 study of 650 khipus found that at least 74 per cent contain sum relationships of the kind the Aschers described, identified top cords as markers of low-level working khipus, rejoined two fragments of one broken khipu from their arithmetic, and proposed that white pendant cords mark the boundaries of groups of sum cords⁠[8]. At Inkawasi, paired khipus carried nearly identical records, consistent with checking accounts against each other⁠[6].

The six Santa Valley khipusTaphonomyMan, 1670. CC BY-SA 4.0Artifact

Non-numerical features carry information in at least some cases. In six colonial khipus from the Santa Valley, Medrano and Urton found 133 colour-coded groups of six cords whose first-cord values sum to the tribute in a 1670 census of 132 named tributaries, and argued that whether those cords are attached recto or verso divides the tributaries into the two halves (moieties) of the community⁠[10]. Hyland's study of two village khipus from Collata, described locally as letters, argued that their many distinct combinations of colour, fibre and ply could encode syllables⁠[7].

What has been tested

Most tests have been bounded checks. The Aschers' sums and Medrano and Khosla's corpus-wide search test the numerical reading, and they pass⁠[1]⁠[8]. The Santa Valley study tests a reading of non-numerical features against an independent written document from the same procedure, and the total matches⁠[10]. Jon Clindaniel's 2019 dissertation analysed non-numerical signs across the corpus and argued that they act as paired predicates arranged in binary hierarchies, conventional across the empire but varying by genre and region⁠[11]. Gary Urton had earlier proposed that each cord encodes a set of binary choices (fibre, spin, ply, attachment, knot direction) amounting to a large inventory of signs⁠[12].

Material tests have added a new line of evidence. In 2025 Hyland and colleagues radiocarbon dated and isotopically analysed the human-hair primary cord of one Inka khipu, KH0631, and found a diet typical of low-ranking commoners, which suggests that khipu making was not confined to the male bureaucratic elite the chronicles describe⁠[13].

The readings

Four positions are held. The first is that khipus were records of numbers and categories only: colours, fibres and attachments label what is counted and for whom, but do not record language⁠[1]. The second treats the non-numerical features as a system of labels for places, groups and social divisions, as in the Santa Valley moieties, which can be decoded as categories without being phonetic⁠[10]⁠[11]. The third is that some khipus, especially narrative ones, recorded words phonetically, for example lineage names, as Hyland proposed for Collata⁠[7]. The fourth is Urton's binary-coding model, which treats each cord as a packet of binary choices that could carry a large and structured code, numerical or linguistic⁠[12]. These readings are not exclusive: accounting khipus and narrative khipus may have used the same features differently.

Published claims

Published claims and their status

Year	Who	Claimed	Status	Reason
1912	L. Leland Locke⁠[4]	Khipu knots record numbers in base ten, by the knot's type and its position on the cord, with an empty position standing for zero.	Accepted	Sum relations hold under this reading across most of the 650 khipus studied in 2024⁠[8].
1981	Marcia and Robert Ascher⁠[1]	Khipus are a logical-numerical recording system: knots give the numbers, and colour, placement and construction label what is counted and how the totals relate.	Most likely	True of most of the corpus, which is arithmetic⁠[8], but it does not show that no khipu recorded words⁠[7].
1996	Laura Laurencich-Minelli⁠[14]	A seventeenth-century Jesuit manuscript in Naples shows that royal khipus encoded Quechua words by syllable.	Disputed	The manuscript's authenticity has not been established by independent tests, and specialists doubted both its date and its contents⁠[14].
2003	Gary Urton⁠[12]	Each khipu cord encodes a sequence of binary construction choices, which together form a structured sign system that could carry numbers or language.	Still possible	Attachment direction carries information in at least one decoded set⁠[10], but the model as a whole has not been tested across the corpus.
2005	Gary Urton and Carrie J. Brezine⁠[5]	The Puruchuco khipus are a hierarchical accounting archive in which census and tribute figures were combined and passed up between levels.	Accepted	The excavated Inkawasi archive shows the same kind of layered accounting⁠[6].
2015	Gary Urton and Alejandro Chu⁠[6]	Khipus found with produce at the Inka storehouse of Inkawasi recorded its stores, and pairs with near-identical records served as accounting checks.	Accepted	The khipus were excavated in context with the goods they record⁠[6].
2017	Sabine Hyland⁠[7]	Two khipus from Collata, called letters by their keepers, may record lineage names phonetically, with cord types standing for syllables.	Still possible	Proposed from two khipus and not yet tested on others⁠[7].
2018	Manuel Medrano and Gary Urton⁠[10]	Six Santa Valley khipus record the 1670 tribute census, with cord groups matching named tributaries and attachment direction marking the two moieties.	Most likely	133 cord groups against 132 named tributaries and matching tribute totals⁠[10]; the match has not been tested against chance, and no second pair has been found.
2019	Jon Clindaniel⁠[11]	Non-numerical khipu signs worked as predicates in paired, hierarchical sets, with conventions shared across the empire.	Still possible	A corpus analysis rather than a decipherment; its proposed readings have not been checked against a transcribed khipu⁠[11].
2021	Sabine Hyland, Sarah Bennison and William P. Hyland⁠[3]	Khipus were kept as texts for about a thousand years, and village communities still used khipu boards in ritual into the 1950s.	Accepted	Rests on colonial khipu boards and a village manuscript describing their use⁠[3].
2024	Manuel Medrano and Ashok Khosla⁠[8]	Sum relations characterise at least 74 per cent of 650 khipus, so most of the surviving corpus is arithmetic.	Accepted	Computed on the open Khipu Field Guide database, where it can be rerun⁠[8]⁠[15].
2025	Sabine Hyland and colleagues⁠[13]	The human hair in the primary cord of khipu KH0631 shows a commoner's diet, so a commoner took part in making it.	Most likely	One khipu, and it assumes the hair was the maker's own⁠[13].
2025	Cinzia Florio and Karenleigh A. Overmann⁠[9]	Inka numbers, as shown on Guamán Poma's yupana, were non-positional and had no concept of zero.	Disputed	The knot data's sum relations work under the positional reading⁠[8], and the preprint does not test its reading against them⁠[9].

What would settle it

A khipu paired with a Spanish transcription of what its keeper read from it, of which the Santa Valley case is the closest so far, would allow a direct test; colonial archives may hold more such pairs. Corpus-wide statistics can now test whether colour, ply and attachment sequences behave like language (with word-like repetition and Zipf-type frequency profiles) or like category labels (with small inventories tied to positions), using the OKR with proper controls. Material tests on more khipus would show who made them and when.