Abstract
Symbolisation is tens of thousands of years older than writing, so the question worth asking is not how human beings became capable of notation but why, after so long a period of capability, durable notation became worth its considerable cost. This article answers an architectural question rather than a causal one. It sets out four conditions any arrangement must satisfy if it is to keep a distinction actionable across absence: a carrier that holds its form, an invariance a later party can recover without having witnessed anything, a route by which the deposit is encountered again, and an institution that reproduces competent uptake. Two arrangements satisfy these conditions, maintained marks and maintained interpreters, and the choice between them is a cost comparison rather than a stage of development, which is why complex societies have repeatedly declined durable notation, adopted it for some domains and not others, and abandoned it. The article distinguishes address from retrieval as the two routes back to a deposit, argues that what arrives late in history is order-independent retrieval rather than retrieval as such, and traces the manufacture of invariance through standardised containers, measures, calendars, offices, and names. Cases are drawn from the archaic Uruk archives, Abydos and Wadi al-Jarf, Anyang, the Linear B palaces, the Indus valley, the Andean khipu, and documentary England. A closing section positions recoverable invariance against immutable mobiles, media bias, commensuration, classification, and phenomenological sedimentation.
I. The Wrong Question
The origins of writing are usually posed as a question about human capacities. At some point, the story goes, human beings began to make marks that stood for things, developed notation, signs, accounting devices, and eventually scripts, and the explanation for this development is sought in the mind: in a growing capacity for abstraction, representation, symbolic thought, or increasingly complex communication. This begins in the wrong place, because multisymbolization was already ancient by the time anything recognisable as writing appeared. Perforated marine shells worn as beads are attested well over one hundred thousand years ago. The engraved ochre plaques from Blombos Cave carry deliberate cross-hatched designs some seventy-five thousand years old, and the same site has yielded a cross-hatch drawn in ochre on a silcrete flake at around seventy-three thousand. The engraved ostrich eggshell containers from Diepkloof preserve a repertoire of hatched band motifs sustained across a long stretch of the Howiesons Poort sequence, the absolute duration of which depends on which of two competing chronologies for the site one accepts.
Human beings painted, marked bodies, made ornaments, buried their dead with grave goods, maintained ritual traditions, told stories, named places, and coordinated through elaborate symbolic systems for tens of millennia before any of them kept a written account of anything. The interesting question is therefore not why humans became capable of writing, but why, after so extraordinarily long a period of symbolic capability, durable notation became worth its cost.
An earlier article in this series distinguished the two poietic mediations, multimateriality and multisymbolization, from the three zoetic ones, and defined civilisation as the cumulative, intergenerational sedimentation of recursive coordination into multimaterial and multisymbolic anchors that restructure the conditions of later life. Writing belongs squarely inside that definition as one especially consequential instance of it, and is in no sense the beginning of multisymbolization. What increases with writing is one specific property: sedimentability, the capacity of a formed state to persist beyond the activity that produced it and to be taken up again by people who took no part in producing it.
II. What Kind of Claim This Is
Three distinct questions travel under the heading of the origins of writing, and confusing them produces most of the bad arguments in the field. The first is causal and local: why did proto-cuneiform appear in southern Mesopotamia in the late fourth millennium rather than somewhere else or at another time. Answering it requires evidence about that region in that period, and no general theory can substitute for it. The second is comparative and statistical: why do durable notational systems appear repeatedly in societies of a certain kind, and fail to appear in others of apparently similar scale. Answering it requires a sample and an account of variation. The third question is architectural: what must be true of any arrangement whatsoever that succeeds in keeping a distinction actionable across absence.
This article answers the third question and uses the first two as material. The architectural claim is that four conditions must all be satisfied, and that failure of any one destroys the arrangement regardless of how well the others are met. A carrier must hold its form. Something must retain a formed state without continuous attention, whether that something is clay, stone, knotted cotton, a trained body, or a landscape feature.
An invariance must be recoverable. A later party who witnessed nothing must be able to re-identify what the deposit is about, which requires that the referent, or some abstracted dimension of it, remain sufficiently the same to be picked out again. A route back must exist. The deposit must be encountered again, either because it stands where the relevant people must pass, which this article calls address, or because someone who knows it exists goes and finds it, which this article calls retrieval. Competent uptake must be reproduced. Somebody must be able to read the deposit and act on it, which requires an institution that goes on producing such people after the original participants are gone.
These four conditions are the analytic spine of everything that follows, and they generate the article’s negative predictions as directly as its positive ones. An arrangement that satisfies all four should be durable even when it looks primitive, which is why a notched stick and a knotted cord can administer a debt and an empire respectively. An arrangement that fails the fourth condition should die while its carriers survive intact, which is what an undeciphered script is. An arrangement that fails the third should sit inert while remaining perfectly legible, which is what an uncatalogued archive is.
Nothing in this framework says that thickening coordination causes writing to appear. The claim is that as persistent recursive relevance accumulates, satisfying the four conditions through living memory alone becomes progressively more expensive, and made anchors become comparatively cheaper. Cost comparisons do not determine outcomes, and the following sections take non-adoption, partial adoption, and abandonment as seriously as invention.
III. Notation Before Writing
The claim that notation is older than writing is safe. The claim that any particular Palaeolithic object is a notation is not, and the difference between these two claims deserves more care than it usually receives.
Three levels of inference should be kept apart. That a mark was deliberately produced is often securely established by microscopy, by the regularity of the incisions, and by the labour the object represents. That a set of marks is structured, in the sense of exhibiting recurrence, grouping, or convention, can sometimes be established statistically. That a structured set of marks had a particular function, numerical, calendrical, or linguistic, is a further claim of a different order, and it is where most of the argument in this field actually lives.
The Lebombo bone, a baboon fibula bearing twenty-nine incisions, is around forty thousand years old and secure at the first level. The Ishango bone carries three columns of grouped notches whose totals have supported arithmetical and lunar readings for seventy years, and is arguable at the second. Alexander Marshack’s proposal that many Upper Palaeolithic marked bones record lunar observation rests on microscopic evidence that marks within a sequence were made by different tools at different times, which supports accumulation over time without settling what was being accumulated. The recent proposal by Bacon and colleagues that dots, lines, and Y signs accompanying animal depictions in European cave art constitute a phenological calendar sits at the third level, and the assessment by Nowell, Bahn, and Le Quellec finds the theoretical and methodological grounds insufficient. The present argument requires none of the third-level claims and would be weakened by resting on them.
What the marked bones do establish is that mark-making capable of being returned to, inspected again, extended, and compared with itself is very old, and long precedes any of the social conditions this article treats as decisive. A tally is already a deposit in the strict sense, and it satisfies the first of the four conditions comfortably.
It fails, or barely meets, the other three. A private tally is legible to its maker and to a living interpreter who can be asked, so the recursive labour of remembering stays where it was and the mark merely assists it. It is unstandardised, so its invariance is recoverable by its maker rather than by a stranger, and it cannot be compared with another maker’s tally or audited against it. And no institution guarantees its interpretation once the maker is gone. Durability alone accomplishes very little, which is the reason this article treats sedimentation as a four-part achievement rather than a property of materials.
One further distinction is required. In coordination mode, a symbol operates on a counterpart who is present: the alarm call, the pointing gesture, the shout of warning. In reference-to-absence mode, a symbol stands for what is not there, whether a technique not currently performed, a maker not currently present, an obligation not currently being discharged, or a quantity not currently in view. The coupling of the poietic mediations depends on multisymbolization operating in the second mode, and notation is that mode given a material carrier. The tally already occupies it. What the tally lacks is a social world in which enough turns on absence to justify building institutions around marks.
IV. Two Solutions to Absence
Coordination across absence has two general solutions, and the comparative history of notation is better read as a running cost comparison between them than as a sequence of stages.
The first solution maintains interpreters. A body of trained people holds what must persist, and the persistence is secured by initiation, apprenticeship, repetition, performance, succession, and sanction against error. Genealogies, ritual sequences, legal precedent, navigational knowledge, technical procedure, and territorial rights have all been held this way at very large scale and with very high precision. The carrier is a trained body operating within a landscape whose features supply the fixed points, so the first condition is met by embodiment and dwelling working together rather than by clay.
The second solution maintains marks. A deposit is placed in a durable carrier, and persistence is secured by materials, custody, authentication, and retrieval.
The two solutions have opposite cost profiles, and neither dominates. Interpreter-borne anchoring requires continuous recursive labour from living people, which is expensive every year without exception, and it fails catastrophically at a generational break, since a single interrupted transmission destroys what centuries deposited. It is nonetheless richly mediated, since a trained interpreter carries the qualitative detail no notation preserves; it is difficult for outsiders to forge, since one cannot counterfeit a song one has never been taught; and it cannot be consulted by strangers, which is frequently the point rather than a defect. Mark-borne anchoring requires only intermittent maintenance, survives interruptions in transmission for long periods, and can be consulted by people the maker never met. It is forgeable by anyone who can hold a stylus, it demediates savagely, and it carries a standing retrieval cost that interpreter-borne systems do not, since a person who knows something can be asked, whereas a tablet in a basket must be found.
This comparison explains what a developmental story cannot. Non-adoption becomes intelligible: where an interpreter cadre is already being reproduced reliably for other reasons, whether a priesthood, a lineage office, a guild, or a court, the marginal gain from marks is small, and the investment in durable notation may never repay itself. Where a polity actively prefers its holdings not to be consultable by strangers, marks are a liability rather than an asset. Selective adoption becomes intelligible too: societies take up notation first in the domains where the interpreter solution is most strained, which are large numbers, long delays, distant places, and disputes among people who do not know one another. This is why the domains that fall first are accounting, tribute, calendar, and boundary, and why myth, law, and story often follow centuries later or never. Abandonment becomes intelligible: a mark system whose maintaining institution collapses loses the fourth condition immediately, and the deposits remain while the capacity to use them disappears.
Address and retrieval belong here rather than as a late refinement, because they determine the cost side of the mark solution. In address, the deposit stands where the relevant people must pass, and reaches them without anyone having to look: the boundary stone, the inscribed lintel, the monumental façade, the milestone, the posted edict, the coin passing hand to hand, the sign on a door. Address is cheap, and its reach is broad and shallow. In retrieval, the deposit sits inert until someone who knows it exists goes and finds it, which requires a place that persists, an ordering principle, a labelling practice, and a person trained to know the record is there. Retrieval is expensive, and its reach is narrow and deep. A society that wants the effects of address can have them with a stone and a chisel. A society that wants the effects of retrieval has to build a bureaucracy, which is a large part of why archives and administrations appear together.
V. The Thickening of the Past
Sedentism did considerably more than place people in villages instead of moving landscapes: it changed the temporal architecture of human existence. Fields remained where they were. Houses persisted and were repaired. Grain accumulated in structures built before the scarcity they anticipated. Animals belonged to particular households and their reproductive histories acquired continuity. Storage vessels contained measurable quantities. Paths became recurrent, boundaries mattered repeatedly, and debts survived the encounters in which they were incurred. Obligations extended across seasons. Property remained attached to persons who were absent. Ancestors became attached to places. Institutions had to coordinate people who could no longer resolve every conflict by separating.
The Levantine sequence shows how early this begins and how loosely it is tied to agriculture. Natufian communities were building substantial semi-subterranean dwellings and burying their dead within settlements well before cereal domestication. The granaries at Dhra' in the Jordan valley, with raised floors for air circulation and capacity well beyond household consumption, date to around eleven thousand three hundred years ago on the excavators' chronology, which places purpose-built storage before domesticated crops rather than after them. The monumental enclosures at Göbekli Tepe, with their carved T-shaped pillars, were raised by populations whose subsistence remained substantially foraging. At Çatalhöyük, houses were rebuilt repeatedly on the footprint of their predecessors, the dead were buried beneath the floors, and skulls were plastered, painted, and circulated among the living, so that a household’s material fabric was itself an accumulation of its own past.
Agriculture is therefore not the operative variable, and neither is population size, urbanisation, hierarchy, or cognitive complexity taken separately. James Woodburn’s distinction between immediate-return and delayed-return systems names the relevant threshold more precisely than any subsistence category. Where yields are consumed close to the point of acquisition and rights over them are not held across long intervals, very little needs to be stabilised symbolically. Where labour is invested now against a return that arrives much later, where the return is held in a store, and where claims on that store are distributed among people who will not all be present when it is opened, coordination acquires a temporal depth that living presence alone struggles to cover.
The relevant transformation is accordingly an increase in the density of persistent recursive relevance. A field is not recursively alive, and does not care who owns it. A wall does not recognise the boundary it marks. A grain store has no interest in its contents. These nonrecursive things nonetheless become extraordinarily important within recursive human relations. The field matters because several people may remember who cleared it, planted it, inherited it, or claimed it. The wall matters because it stabilises a distinction between inside and outside that would otherwise require continuous enforcement. The grain matters because its present distribution determines future survival, and access to it may depend on previous contributions, kin relations, status, or obligation. Nonrecursive material things become the anchoring points around which interrecursive relationships accumulate.
This produces a specific and productive asymmetry. Human relationships are responsive and therefore revisable at every moment, while fields, walls, and granaries are not. A person can change their mind about who owes what; a field remains where it is. Someone can deny that ten measures of barley were delivered, and the barley may already have been eaten. Someone can die leaving unresolved claims behind. A herd increases, divides, and is inherited. The growing persistence of the material world generates increasing pressure to stabilise relationships that possess no comparable persistence of their own, and both solutions to absence become more expensive at once, which is what makes the choice between them consequential.
VI. Recoverable Invariance, and What Would Refute It
Notation selects for stability rather than representing the world indiscriminately, and this may be its deepest property. What is easiest to write is whatever can be made to remain the same long enough for the mark to go on referring to it. Numbers are powerful because they discard nearly all qualitative variation. Names stabilise persons across bodily and situational change. Property categories stabilise relations between persons and things. Calendrical notation stabilises recurring temporal differences. Measures stabilise comparisons between otherwise heterogeneous substances. Notation flourishes wherever differences can be repeatedly recovered.
Grain is almost ideally suited to this operation. Individual grains are extraordinarily similar, so they can be measured without describing each one. Grain persists, divides, transfers, and can be owed. It carries immense metabolic significance while being itself entirely nonrecursive. It therefore combines material persistence, standardisability, quantifiability, transferability, and future relevance in a single substance. A notation associating a quantity of grain with a particular person, household, or institution allows a relation between people and resources to survive the encounter in which that relation was established.
Livestock work almost as well, at the cost of a more violent reduction. Animals are alive, mobile, individual, and variable, and yet they possess dimensions that abstract into stable categories: three sheep, five cattle, a number of goats of a given age and sex. The notation captures nothing of the animal as a lived counterpart. Its condition, temperament, smell, history, and relations with the herdsman disappear entirely. What survives is what the coordination problem requires, and this selection is the achievement rather than the defect, since only what has been stripped down in this way can travel to a reader who was never present.
The concept needs a criterion, or it becomes unfalsifiable decoration. An invariance is recoverable when a competent later party who witnessed nothing can re-identify the referent well enough for the coordination at issue. Two features of this criterion do real work. It is relative to the stake rather than absolute, so the invariance required to allocate a ration is far less than the invariance required to transfer title to land, and a notation adequate for the first will fail for the second. And it is relative to a competent party rather than to any reader, so recovery is a property of the arrangement as a whole rather than of the mark.
The criterion supplies failure conditions, and history supplies the failures. Boundaries are disputed because two parties re-identify a line differently. Measures are contested because a container in one town holds more than a container of the same name in another. Names collide, and two men called the same thing appear in the same register. Seals are forged, and the party who authenticated a document turns out not to have. If recovery were incidental to what notation is for, these failures would be tolerated as noise. What the record instead shows is enormous and continuous investment in repairing exactly these failures, through patronymics and professional epithets, witness lists, duplicate texts, sealed envelopes, metrological standardisation, physical boundary markers, and cadastral survey. The repairs are the evidence.
They also point to the argument’s most important addition. Invariance is very often manufactured rather than found. The archaic Mesopotamian record makes this visible with unusual clarity. Bevelled-rim bowls, crudely moulded, mass produced in enormous quantities, and approximately standardised in capacity, are among the most abundant artefacts of the Uruk period and are widely interpreted as ration containers. The archaic sign for a ration of bread appears to derive from the profile of such a bowl. A multimaterial anchor, the standardised container, thereby underwrites a multisymbolic anchor, the sign that counts rations, and neither would function as it does without the other. The same coupling recurs wherever measurement is institutionalised: in the sila and the gur of Mesopotamian capacity measurement, in the cubical chert weights of the Indus cities with their consistent ratios, in the shekel, in the standardised brick, in the surveyor’s rope.
Persons receive the same treatment. A name is a manufactured invariance imposed on a counterpart who changes continuously, and administrative systems reliably supplement it when it proves insufficient, with patronymics, professions, places of origin, seal impressions, and later with registers, numbers, and biometrics. What the record needs is a recoverable handle on the person rather than the person, and the history of administration is in large part the history of manufacturing better handles.
VII. The Archaic Sequence as One Trajectory
The Mesopotamian material is the best documented case available, and it is treated here as one regionally specific trajectory rather than as the sequence through which writing must pass. Its interest lies in showing the four conditions being satisfied step by step, and in the fact that its own specialists disagree about the steps.
Sealing comes first and is instructive in its own right. Stamp seals appear in the Neolithic Near East long before any script, and cylinder seals appear in the Uruk period around the middle of the fourth millennium. Wet clay pressed over the cord of a jar stopper, a bundle, a basket, or a door-lock, and impressed with a distinctive design, makes access visible and non-deniable rather than preventing it, since the sealing must be broken to reach the contents and cannot be reconstituted without the seal. The sealing is a nonrecursive witness posted at a point where recursive negotiation would otherwise be unconstrained. It is also a pure case of address, since it stands exactly where anyone who matters must pass, and it therefore requires no retrieval infrastructure at all, which is part of why sealing is so much older than archives.
Clay tokens of varied geometric shapes circulate across the Near East from the eighth millennium onward, and Denise Schmandt-Besserat’s influential reconstruction treats them as direct ancestors of the archaic signs, with token shapes carried over into impressed and incised marks. Robert Englund, Paul Zimansky, and others have raised serious objections concerning the chronological and typological continuity the thesis requires, and the strong version should be held loosely. What is not in dispute is the practice attested in the late fourth millennium, in which tokens were enclosed within hollow clay spheres whose exteriors were sealed and, in a number of cases, impressed with marks corresponding to the tokens inside. The arrangement contains its own resolution, since once the exterior carries the information the interior becomes redundant, and flattened tablets bearing impressed numerical signs and seal impressions follow. Numero-ideographic tablets add signs for the commodity counted. Proto-cuneiform tablets from Uruk IV, around 3300 BCE, add enough further signs to constitute a system.
The content of that system is famously disappointing to anyone expecting the externalisation of thought. Of the several thousand archaic tablets from Uruk and its successors, the overwhelming majority are administrative, recording quantities of grain, beer, textiles, livestock, land, labour, and rations against officials, institutions, and periods of account. The remainder are lexical lists, standardised inventories of signs and categories used in training and reference. There is no literature for several centuries. The history of writing begins with an accountant rather than a poet, which is disappointing only to an account that has misidentified the problem being solved.
Two peculiarities of the archaic material carry heavy theoretical load. The first is metrological. Archaic bookkeeping employs more than a dozen distinct counting and measuring systems, and the value of an impressed sign depends on which system is in play, with different systems for discrete objects, grain capacity, land area, particular commodity rations, and time. The same mark does not mean the same quantity across contexts, which is to say that there is as yet no abstract number in the record. System-independent number, and eventually the sexagesimal place-value notation that made Babylonian mathematics possible, emerge out of centuries of accounting and metrological practice rather than preceding them. Peter Damerow’s insistence that early representational systems be analysed as historically specific ways of producing knowledge, rather than as incomplete versions of later ones, is the right methodological setting for this. Notation comes first in a strong and specific sense: people do not need a philosophy of number in order to keep track of quantities, they need recurrent differences that matter, and the practice of stabilising such differences can itself generate the conditions for more abstract systems.
The second peculiarity concerns names, and here the argument should be stated more carefully than it usually is. An administrative system must identify persons, and a person’s name is not depictable in the way a sheep or a jar is. What can be recovered about a name is its sound, and the rebus principle, in which a sign for a depictable thing is used for the sound of its name, provides a route. Sumerian, with its short and largely monosyllabic roots, was unusually hospitable to that move. The defensible claim is that administrative identification of persons is one of the settings in which phonographic resources appear early and are put to sustained use, rather than that administration caused phonography. Even the weaker claim is instructive, since it locates the first sustained encoding of speech in the requirement to re-identify a particular someone across absence, which is exactly the second of the four conditions. A tablet from the archaic corpus carries the sign sequence conventionally read as Kushim, associated with quantities of barley in a brewing context, and has been described, with appropriate caution about whether it names an individual or an office, as the earliest personal name recoverable from any document.
The Egyptian evidence points the same way from a different direction. The small bone and ivory labels from Tomb U-j at Abydos, of around 3200 BCE, were attached to grave goods and carry signs indicating quantities, commodities, and apparently the estates that supplied them. The serekh, the panelled frame enclosing a royal name, appears on jars and sealings of the same horizon. Egyptian writing is phonographic very early because names had to be written, and it is administrative and display-oriented from the outset rather than passing through a purely administrative phase.
VIII. Why Some Archives Are Accounts and Others Are King Lists
The Aegean case tests the argument under near-laboratory conditions. Linear B, deciphered by Michael Ventris in 1952 and shown to record an early form of Greek, is attested at Knossos, Pylos, Thebes, and Mycenae from roughly the fifteenth to the thirteenth centuries. Its content is administrative without exception: personnel lists, ration allocations, land tenure holdings, quantities of wool, oil, and grain, inventories of chariot wheels and furniture, offerings to deities. There are no letters, no narratives, no laws, and no literature.
The material dimension of these archives makes an argument of its own. The tablets were sun-dried rather than fired, intended to serve the administrative year and then to be recycled as accounts were consolidated. They survive because the palaces burned, which baked what was meant to be temporary and preserved exactly the documents the system intended to destroy. A carrier’s durability is calibrated to the coordination problem, and this system required anchors that would last a season.
The collapse of the palace centres around 1200 terminated the script, and writing disappears from the Aegean for roughly four centuries. The fourth condition failed: the redistributive administration that trained scribes and required accounts ceased to exist, and with it the reproduction of competent uptake. When alphabetic writing returns in the eighth century, borrowed from Phoenician and adapted with vowel letters, the earliest Greek inscriptions are not accounts at all, but a hexameter joke on a drinking cup, a line of verse on a wine jug offered as a dancing prize, dedications, and graffiti. A script’s origination and its adoption have entirely different social profiles, and any general theory of writing that ignores the distinction will misread every case after the first.
The Chinese evidence yields a different content profile again. The oracle bone inscriptions from Anyang, from around 1250 onward, are divinatory records: the charge put to ancestors or to the high god, the date, the diviner, the interpretation of cracks produced by heat applied to prepared turtle plastrons and cattle scapulae, and in a proportion of cases a verification recording what subsequently happened. This is an institution auditing its own record against the world and archiving the result in pits. Contemporary and slightly later bronze inscriptions do something else again, recording royal awards and their occasion and closing with the formula that sons and grandsons for ten thousand years should treasure and use the vessel. The inscription addresses the unborn, which is as compact a statement of the sedimentary function as any theory could ask for, and it is a case of address rather than retrieval, since the vessel is handled and seen by exactly the descendants it names.
Mesoamerican writing is bound from its earliest appearances to calendrical notation and dynastic legitimation rather than to accounting. Monument 3 at San José Mogote carries a calendar name beside a sacrificed captive. Maya inscriptions concern accession, warfare, ritual obligation, and the placement of rulers within cycles of enormous length, with the Long Count fixing events against a fixed epoch in a way that permits coordination of dates across centuries. Comparative work on Mesoamerica has found no simple relation between polity scale and the elaboration of writing, which is exactly what the cost comparison of section IV predicts and what a developmental account cannot accommodate.
The accounting-first pattern is therefore a tendency generated by a particular strain on the interpreter solution rather than a law governing writing. Where a redistributive institution must track contributions and disbursements among people who cannot all be present, quantities and persons are what get written first. Where the pressing problem is the durability of authority, the legitimacy of succession, or the placement of the present within a ritual cycle, the same operation yields calendars, king lists, and monumental inscriptions. In both cases the principle holds: what gets deposited is what must remain recursively relevant when the people who established it are absent, and the difference lies in what a given institution cannot afford to have forgotten.
IX. Notation Without Writing, and the Coupling Under Strain
The Andean case forces a definitional decision that the Mesopotamian case allows one to postpone. The Inka administered several million people across thousands of kilometres of extreme terrain, with a road system, storehouses, labour levies, censuses, and tribute obligations, using knotted cords rather than any glottographic script. The khipu encodes quantity by knot type and position on a pendant cord in a decimal arrangement, with long knots, single knots, and figure-eight knots distinguishing values and units, and it carries further information in colour, in the direction of ply and attachment, and in the hierarchical organisation of pendant, subsidiary, and top cords. Gary Urton and Carrie Brezine showed for a set of khipus from Puruchuco that values on cords at one level sum to values recorded at another, which is to say that the corpus preserves an accounting hierarchy in the ordinary bureaucratic sense.
Reading the cords was an office. The khipukamayuq was trained, held responsible, and consulted, and Spanish colonial courts accepted testimony transcribed from khipus for decades before ecclesiastical authorities ordered their destruction. Later work complicates any assumption that the medium was confined to accounting: Sabine Hyland’s analysis of the Collata cords argues for a phonetic dimension, and Frank Salomon’s study of the patrimonial khipus of Tupicocha documents cords that served as insignia of communal office into the twentieth century, long after the capacity to read them had been lost. The Andean case is thus a system in which the mark solution and the interpreter solution were deliberately combined, with the cords carrying what could be carried and the trained officer carrying the rest.
The Indus valley presents a different problem. Some four thousand inscribed objects, mostly seals and sealings, carry short sign sequences averaging around five signs. Whether these encode language at all remains disputed, and the case that the system is nonlinguistic has been argued seriously. The dispute can remain open here, since on either reading the signs appear on the instruments by which goods were sealed, which is to say at exactly the point where a claim must survive an absence. Both the Indus corpus and the khipu are included as tests of the analytic category rather than as evidence that all such systems are versions of writing, and in both cases the honest position is that the fourth condition has failed for us: the deposits are intact and the competent readers are gone.
These cases also put pressure on the coupling thesis, which holds that multimateriality and multisymbolization only ever operate as a pair, and it is worth asking where that thesis could break. The khipu does not break it, since the material configuration of the cord is the notation, so that far from separating the mediations the artefact collapses them into a single object. Systems anchored in trained bodies and landscape features do not break it either, since embodiment supplies a carrier and dwelling supplies the fixed points, though that carrier is maintained continuously rather than intermittently, which is precisely the cost difference set out in section IV. What would break the thesis is a symbolic anchor with no material carrier of any kind, sustained across generations without bodies, objects, or places, and no documented case supplies one.
The definitional consequence is straightforward. Defining writing narrowly as the encoding of speech excludes the khipu, may exclude the Indus signs, and produces a category that cannot explain why all these systems occupy similar social positions and do similar work. Roy Harris’s case against a speech-transcription definition, and the argument of Elizabeth Boone and Walter Mignolo for taking writing without words seriously on its own terms, both point the same way. The category this theory requires is durable multisymbolic anchoring, of which the encoding of speech is one especially powerful instance, powerful because a phonographic system inherits the combinatorial resources of a language and can therefore record what its designers never anticipated. That versatility should not be mistaken for the origin.
X. Carriers, Authentication, and the Cost of Retrieval
No symbol persists without a material capable of holding a mark, and the properties of that material determine what survives to be studied. Clay is nearly indestructible once fired, and fires attend the destruction of cities, so Mesopotamia is over-represented in the record of early administration by an enormous margin. Papyrus survives only where it is dry, which is why the oldest inscribed papyri known, the logbooks from Wadi al-Jarf on the Red Sea coast, survived at all: they record the work of an inspector named Merer and his boat crew ferrying limestone from Tura to the construction site of Khufu’s pyramid, month by month, in what amounts to a timesheet from the building of the Great Pyramid. Bamboo slips, palm leaf, birch bark, and wooden tablets rot in humid climates, which has erased most of what South and Southeast Asia committed to writing before the common era and distorts every comparative account built on surviving documents. Wax tablets were designed to be erased and reused. The oracle bone survived because bone is durable and because the pits were sealed and forgotten.
Materiality also supplies what symbols cannot supply for themselves, which is authentication. A mark can be copied by anyone who can hold a stylus, and the value of a record depends on the reader’s confidence that it has not been. The split tally answers this elegantly. A stick notched across its width to record a debt is split lengthwise, creditor and debtor each holding one half, and the irregular grain of the split cannot be reproduced, so the halves authenticate one another by fitting. The English Exchequer used such tallies, in stock and foil, as instruments of royal finance for more than six centuries. Mesopotamian scribes solved the same problem differently, enclosing a contract tablet within a clay envelope on which the text was written again and sealed by witnesses, so that a disputed outer text could be checked against an inner one reachable only by destroying the seal. In both cases a multimaterial device secures a multisymbolic claim, and neither mediation would do the work alone.
Retrieval is where this argument must be stated carefully, because the obvious version of it is false. Ancient societies did not lack retrieval technologies. Mesopotamian archives used clay labels attached to tablet baskets, grouped records by type, sender, or transaction, stored them in dedicated rooms, and sealed the doors. Neo-Assyrian libraries used colophons and inventory texts. Callimachus compiled the Pinakes at Alexandria in the third century, which is a catalogue in any reasonable sense of the word. Retrieval was an institutional problem from very early, and institutions solved it.
What they solved it with was custodial knowledge. Early retrieval works because a person knows where things are, what the groupings mean, and which basket holds which year’s accounts, and the ordering principles are local, tacit, and unusable by anyone outside the establishment that produced them. What arrives late in history is order-independent retrieval: thoroughgoing alphabetisation, the subject index, pagination and running heads, the verbal concordance developed for preaching and legal argument in the thirteenth century, and eventually the catalogue that serves a reader the librarian has never met. The distinction is between retrieval that requires a custodian and retrieval that does not.
Stated that way the point becomes political rather than merely technical, and considerably stronger. For most of the history of the written record the deposit was durable and hard to find, and whoever controlled the place, the labels, the procedural knowledge, and the authorised reading controlled what the archive could be made to do. The cost of retrieval is a form of power, and reducing it redistributes power, which is why order-independent retrieval arrives in the company of universities, mendicant preaching, canon and common law, and later of printing.
XI. From Memory to Written Record
The transition is best documented not at any origin but in a case where an already literate society converted itself into a documentary one. Michael Clanchy’s study of England between the Norman Conquest and the early fourteenth century tracks that conversion in detail. Domesday Book in 1086 recorded holdings, values, and obligations across most of England with an ambition contemporaries found alarming, and over the following two centuries the volume of documents produced by royal government multiplied enormously, with the chancery’s consumption of sealing wax rising by roughly an order of magnitude across the middle decades of the thirteenth century.
The instructive part is the resistance and the supplementation. Documents were not initially trusted, since a parchment is silent and a witness can be questioned. So the document was surrounded with material and ritual guarantees. Charters were sealed, and the seal mattered more than the signature. Symbolic objects were attached to charters or handed over alongside them: a knife, a turf, a sod of earth from the land conveyed. Transfers of land were performed through livery of seisin, in which parties stood on the ground and a physical token changed hands before witnesses, with the charter recording what the ceremony accomplished rather than accomplishing it. Monastic houses forged charters in considerable numbers, often to document endowments they believed entirely real, because a new medium demanded a form that older transactions had never taken and the past had to be supplied with the evidence the present required.
The case is a textbook illustration of the two solutions running in parallel during a changeover. The witness is the interpreter solution, the charter is the mark solution, and for a century and a half English land law used both because neither alone commanded confidence. What eventually tipped the balance was not any improvement in parchment but the growth of an institution that could store, find, and authenticate documents on demand, which is the third and fourth conditions being satisfied at scale.
Literacy in this period is an institutional capacity rather than an individual accomplishment. Documents were read aloud, and a charter was published by being heard. The audit was a hearing, in the literal sense the word preserves. A magnate who could not read employed clerks who could, and this was the ordinary architecture of documentary power rather than a deficiency. The relevant question about any society is therefore not what proportion of its members can decipher marks, but how the capacity to produce, keep, find, read, and authenticate deposits is distributed.
This bears on a long argument in the anthropology of literacy. Jack Goody’s claim that writing, and particularly the list and the table, restructures thought by making comparison, classification, and contradiction visible identifies something real about what externalised marks make available for inspection. The stronger version, in which literacy as such carries uniform cognitive consequences, does not survive contact with the comparative record, as Brian Street’s critique established. Treating literacy as an institutional practice rather than as a technology with intrinsic effects makes both positions tractable. Marks outside the body can be inspected again, compared with other marks, extended beyond working memory, and applied recursively to one another, and records can become objects of further records. What determines whether any of this happens is the institution that maintains the deposits, trains the readers, and decides which comparisons are worth making.
XII. Recursive Escalation
Once records accumulate they can be compared. Once they can be compared, categories acquire consequence. Once categories are standardised, particular cases must be made to fit them. At this point the system begins to reshape its referents.
The Ur III administration of the twenty-first century provides the clearest early demonstration, and it should be stated with the precision the evidence supports. On Robert Englund’s analysis of the herding and labour accounts, the administration set expected outputs against recorded deliveries, using fixed daily labour norms for tasks such as earth-moving and expected rates of yield and reproduction for herds of given composition. The difference between expectation and delivery was recorded as a deficit and carried forward in the accounts from year to year, and there is evidence of deficits being charged against the estates of officials who had died. An accounting deficit is not automatically a legal debt, and the distinction between an entry, a liability, and an enforceable claim should be preserved rather than collapsed for rhetorical effect. The supportable statement is precise enough without exaggeration: within this apparatus an entry could render a shortfall visible, comparable, and actionable according to institutional rules, and the shortfall was computed from a norm rather than observed in the world. A shepherd could be held short of animals that had never been born, because the account specified how many should have been.
The same operation, at larger scale, is the history of administration. A field can be measured because it persists, and once the measurement becomes authoritative the field may be divided according to it. Grain can be counted because it is stored, and once accounting categories exist production may be organised around the categories. Persons can be named in registers because they persist socially across situations, and once registers exist, identities must increasingly correspond to the names the institution recognises. The Roman census, Chinese household registration, the Ottoman tahrir surveys, Domesday, the Napoleonic cadastre, and the imposition of permanent surnames for taxation and conscription all perform the same extraction of relatively invariant, comparable, recordable dimensions from complex living worlds. James Scott’s account of legibility gains rather than loses force when placed inside this trajectory, since the modern state inherited and amplified the operation rather than inventing it.
The reduction involved has a name in this theory. Notation is demediation, the stripping of a multisensorial, multispecies, dwelling-embedded, materially specific encounter to a category and a quantity that can travel. The term is used in the sense established across this series, for the removal of mediational richness from a counterpart, and it should not be confused with the media-theoretical use of remediation for the representation of one medium within another. To record three sheep is to forget almost everything about the sheep. To record a person’s name is to forget almost everything about the person. To mark a boundary is to suppress the ambiguous history of movements, uses, and relationships surrounding it. To write a law is to extract a generalisable distinction from the variation of particular cases. Writing is a technology of selective forgetting as much as of memory, and its portability is purchased precisely with what it discards.
Remediation is what a competent reader does with the result. The herdsman who reads three sheep knows which three, in what condition, in whose care, and on which pasture, and reconstitutes from the record a situation the record does not contain. The clerk who reads a name knows the household. Every functioning documentary system depends on readers who can put back what the notation removed, and the characteristic pathology of large administrative systems is the point at which readers can no longer do so, having never been present at anything the record demediated. At that point the categories confront everyone, including the administrators, as though they were the structure of the world.
This is where the history of notation becomes the deep prehistory of a characteristic error of social theory. Writing teaches its users to encounter stabilised distinctions as though the distinctions had been there all along. A field boundary comes to look like a property of the land, a recorded debt like a property of the debtor, a legal category like a property of the person, a classification like a property of reality, and eventually theoretical binaries like structures underlying human experience. The historical direction ran the other way. Living processes were continuous, multimediated, and recursively unstable; human beings extracted distinctions from them because distinctions were useful; they stabilised those distinctions materially because they needed them to persist; and only after sufficient sedimentation did the distinctions acquire the appearance of ontology.
XIII. The Dead Governing the Living
The asymmetry at the centre of the written record can be stated in a sentence: the inscription does not respond, and people respond to the inscription. A mark on clay holds the same shape regardless of who approaches it or how, which is exactly the property that lets it anchor, and everything politically consequential about writing follows from placing an unresponsive object in the middle of responsive human relations. A spoken claim depends on the person making it and can be met by addressing that person. A written claim detaches from its originator, travels where the speaker cannot, confronts someone years later, is copied, accumulates alongside other records, is consulted by strangers, and is incorporated into institutions. The appeal to what is written becomes a peculiar kind of argument because writing converts a historically situated act of distinction into something that confronts later actors as though it were part of the stable environment.
An anchor cannot enforce itself, and its makers have always known this. The most eloquent evidence is the curse formula. Mesopotamian boundary stones invoke the gods against anyone who removes, effaces, or alters them. Egyptian tomb inscriptions threaten those who would violate the burial. Assyrian foundation deposits address a future prince who will one day find the inscription during a restoration, instructing him to anoint it, make an offering, and return it to its place, promising divine favour if he does and calling down destruction if he does not. These formulae are attempts to attach an interrecursive enforcer to a nonrecursive deposit, because the maker can guarantee the mark’s persistence and cannot guarantee that any living person will care about it. A curse is what one writes when one needs an anchor to keep working after every institution that might defend it has gone.
Where the institution persists, the effects are correspondingly enormous. Laws survive legislators and are applied to circumstances their drafters never anticipated. Scriptures survive prophets and generate interpretive institutions whose entire purpose is the recursive uptake of a fixed deposit. Contracts survive signatories. Mathematical proofs are re-entered centuries later by readers who need nothing from their authors. Bureaucratic categories confront newborn persons long before those persons have any conception of them. The symbolic residues of previous recursion become part of the apparently nonrecursive world into which subsequent generations are born, and the dead govern the living through no supernatural agency whatever, but through the material fact that their coordination was deposited into the ground the living have no choice but to work within.
Lexical lists make the point with a purity no political example achieves. The Standard Professions List, an ordered inventory of official titles compiled at Uruk in the late fourth millennium, was copied by scribes for roughly a thousand years, long after many of the offices it names had ceased to exist in any recognisable form. Its persistence had nothing to do with its accuracy and everything to do with its position in training, since copying it was what made someone a scribe. A curriculum is a sedimented anchor that reproduces the readers who will go on reproducing it, which is the fourth condition maintaining itself by circular means.
XIV. What This Adds
Several established literatures approach this territory, and the contribution claimed here should be stated against them rather than in isolation.
The nearest neighbour is Bruno Latour’s immutable mobile. Both concepts concern the stabilisation of selected features so that they can be handled elsewhere, and both insist that the achievement is material and institutional rather than representational. The difference lies in the axis. Latour’s mobiles are optimised for movement across space and for combination at centres of calculation, so their defining virtues are portability, flatness, superimposability, and scale-independence. Recoverable invariance concerns immobility across time and the re-identification of a particular referent after absence. A boundary stone is an excellent anchor and a terrible mobile. A cadastral map is both, and the two properties can be prised apart, since what makes the map combinable with other maps is not what makes a parcel re-identifiable in twenty years. Latour’s account also supplies no criterion for the recursive status of a counterpart, and the asymmetry this article rests on, that the deposit does not answer while people answer to it, cannot be stated within a framework that distributes agency evenly across a network.
Harold Innis is the closest precedent for section IV. His distinction between time-biased and space-biased media, stone and clay against papyrus and paper, identified the durability and portability trade-off decades before it was rediscovered, and tied it to the character of empires. What is added here is that the trade-off runs not only between media but between media and persons, since maintained interpreters are the alternative a media-centred analysis leaves out, and non-adoption of writing is unintelligible without them.
The phenomenological precedent should be acknowledged plainly. Husserl’s account of the origin of geometry argues that writing secures ideal objectivity by making a sense-formation available in the absence of the originating subject, and calls the resulting layering sedimentation, a term this series uses in a related sense. Derrida’s commentary made the reactivation problem central, since a sediment that cannot be reactivated becomes a crust rather than a foundation. The present article runs the analysis in the opposite direction. Rather than asking how writing secures ideality, it asks what kind of world makes the securing worth its cost, and it treats reactivation as an institutional problem about trained readers, custody, and retrieval rather than as an epistemological one about recovering an original evidence.
Two sociological literatures are adjacent and distinct. Bowker and Star show that classification systems and standards institutionalise viewpoints, distribute burdens, and render some things invisible, and the argument here about congealed valuations agrees entirely. Espeland and Stevens analyse commensuration, the conversion of qualitative differences into common metrics. Both are synchronic, asking whether two cases can be treated as the same kind at a given moment. Recoverable invariance is diachronic, asking whether a case can be identified as the same one across an interval during which nobody was watching. Commensuration answers whether this is the same kind of thing. Recovery answers whether this is the same thing. Administrative systems need both and fail differently in each, since a system with poor commensuration produces incomparable records while a system with poor recovery produces records about nobody in particular.
Material engagement theory shares this article’s refusal to explain notation by prior cognitive achievement, and its insistence on mutual constitution among minds, materials, and practices is compatible with everything argued here. The difference is one of resolution. Flat co-constitution has no vocabulary for distinguishing a counterpart that answers back from one that does not, and that distinction generates the entire argument, since the point of a deposit is that it is nonrecursive and the danger of a deposit is that nonrecursive things enter interrecursive life and alter its trajectory without ever participating in it. A framework that levels the difference cannot say why an anchor anchors.
XV. Maintenance, Loss, and Ruin
Every deposit examined here requires someone to keep it in working order, and the labour involved is recursive labour devoted to preserving a counterpart’s nonrecursive affordances against the forces that would otherwise erode them. Clay chips, papyrus rots, ink fades, languages change, scripts fall out of use, and institutions lose the legitimacy that once made their rulings unquestionable. A deposit whose maintenance stops becomes inert rather than false.
Undeciphered scripts are ruins in the precise sense this theory gives the term, being sediment whose organising recursion has been lost, materially present and no longer capable of coordinating anyone’s life. Linear A survives in hundreds of documents that can be transliterated and not understood. Proto-Elamite, an administrative notation from Susa contemporary with archaic Mesopotamia, remains largely opaque despite a substantial corpus and a well-understood administrative context. The Indus corpus resists a century of effort. Rongorongo on Rapa Nui was still being produced within living memory of European contact and can no longer be read, because the people who could read it died within a generation of the collapse that removed the institutions sustaining them. In each case the first three conditions hold and the fourth has failed, which is the characteristic failure mode of a mark-borne system and one an interpreter-borne system never exhibits, since an interpreter-borne system loses everything at once rather than leaving legible wreckage.
Script death happens even to the most monumentally anchored systems. The last dated hieroglyphic inscription was carved at Philae at the end of the fourth century of the common era, the last demotic text a few decades later, and the Egyptian record then stood unreadable for fourteen centuries, immensely visible and entirely mute, until Champollion recovered its principles in 1822. The last dated cuneiform tablets are astronomical almanacs from around the end of the first century, written for a scholarly community in a script that had been in use for three thousand years and would shortly have no users at all. In each case the physical anchors were in excellent condition and the institution reproducing readers was not, which is why the scribal school, the copying exercise, and the curriculum belong to the technology of writing as fully as the stylus does.
The modern situation intensifies most of this. Digital records depend on continuously maintained machinery, on file formats whose interpreters must themselves be maintained, and on organisations willing to keep paying for both, and the resulting archive decays faster than parchment while presenting an appearance of permanence.
The Exchequer tallies furnish the appropriate closing image. Having served English royal finance for six centuries, they were abolished as an accounting instrument in 1826, and the accumulated sticks were kept in storage until, in October 1834, it was decided to dispose of them by burning them in the stoves beneath the House of Lords. The flues overheated, the fire spread, and the Palace of Westminster burned to the ground. The abandoned anchor consumed the institution that had abandoned it, which is a more literal demonstration of the costs of maintenance than any theory could reasonably request.
XVI. Conclusion: The Reader Removed
Writing is not a triumph of representation and should not be equated with civilisation, intelligence, or symbolic sophistication. Nonliterate societies sustain symbolic worlds of enormous complexity, and oral and performative traditions preserve exactly the mediation-rich detail that administrative systems are built to discard. What distinguishes durable notation is a particular arrangement of multisymbolization and multimateriality: the capacity to let selected distinctions remain externally recoverable beyond the living encounter, without requiring any living person to hold them in the meantime.
The real transition is therefore not from speech to writing, nor from concrete to abstract thought, nor from primitive to civilised cognition, but from a world in which recursive coordination could rely more extensively on living presence to one increasingly burdened by persistent absences: the absent creditor, the absent ancestor, the absent owner, the absent ruler, the absent witness, the harvest that has already happened, the payment that has not happened yet, the boundary established before anyone now alive was born, the obligation whose participants no longer remember its origin. Sedentism thickened the past. Notation gave absence a material presence, and once that became possible, human beings began to inhabit environments in which symbols made by people who are no longer present participate in the organisation of current life.
Across the whole of this history the deposit has never answered back, and it does not answer back now. A store that alters when it is written to is not thereby coordinating with its own ongoing life, and a system that returns results to a query anticipates nothing and addresses no one, so the most elaborate database remains nonrecursive multisymbolization of exactly the kind a notched stick already was. The categorical distinction holds where it always held. What has changed is the reader.
For five thousand years the activation of a deposit required a living being to retrieve it, interpret it, and remediate it back into a situation the record did not contain. Retrieval was the expensive step, and its expense was the reason custody conferred power. Order-independent retrieval reduced that expense enormously, and the reduction is now approaching completion, with selection among deposits increasingly performed by apparatus that has never been present at anything the records demediated. The result is demediation twice over: a record that already discarded almost everything about its referent, now selected and combined by something that cannot put any of it back. The question this raises is not whether the archive has begun to answer, since it has not and will not, but what becomes of anchoring when the recursive uptake that alone makes a deposit effective is itself deposited, and the remediating reader, who was always the point at which a mark rejoined a life, is no longer there to be trained.