# What We Keep Or Seal

A woman hands a shopkeeper in Asan a thousand-rupee note and walks away with a bag of groceries. A man hands the same shopkeeper a sheet of paper of the same size, printed with the same inks, and walks away with nothing. Under a microscope the two sheets look alike. The difference lives in the shopkeeper, the woman, the bank that issued the note, and everyone who will take it tomorrow because they expect others to take it too.

Philosophers call the study of this difference social ontology. Ontology asks what there is. Social ontology asks how there come to be things like money, marriages, borders, licenses and committees, which exist only because people take them to exist. The weight of the note is a brute fact, which would hold if everyone vanished. Its value is a social fact, which holds as long as enough of us keep treating the note as money. The usual name for that treating is collective acceptance.

For a while now I have been working on one question in this field, and I am writing it up as a paper called *What Absence Costs*. The question is which social facts need an authority behind them, someone with the standing to settle them, and which ones anybody holding the right records can settle alone. The answer I defend turns on whether a fact asks for something to exist or for something not to exist. The question comes out of [an essay on collaborative systems](https://risav.dev/socioware) I published this year, and the essay comes out of [a grid of research questions](https://risav.dev/matrix) I drew in 2014.

## Things that exist because we keep them

### A piece of wood that jumps

Two children are learning chess. One of them picks up the knight and moves it in a little L, over the heads of the pawns, and the other protests that wood cannot jump. Both children are right: the wood cannot jump, and the knight can.

Before the rules of chess there is no knight, only a carved horse's head on a base. There is no check and no checkmate, only a small wooden king with other pieces near it. The rules make the knight and checkmate possible in the first place. John Searle called rules of this kind constitutive rules, and he set them beside regulative rules, which govern something that already goes on without them. Drive on the left is a regulative rule: people drove before it, and it tells them how. Checkmate ends the game is a constitutive rule: there was no game for it to end until rules like it were written.

Searle's formula for a constitutive rule is *X counts as Y in context C*. Carved wood counts as a knight in chess, and printed paper counts as a thousand rupees in Nepal. The rule hands out a status, and the status lets its bearer do what its physical makeup never could, the way the knight jumps. Searle called these status functions, and each comes with powers and duties: the holder of the note may pay with it, and the bank must honor it.

In his later work, Searle argued that one kind of act assigns every status. We make something the case by declaring that it is the case: "I now pronounce you married", or "The meeting is open". He called these status function declarations. This post has a status of its own. Nothing in its words makes it part of a series called An Architecture; a setting on my blog declares it so.

### The wedding nobody wrote down

Searle's account leans on what people accept. The Italian philosopher Maurizio Ferraris leans on what people write down.

A couple marries with all the ceremony two families can manage, and nothing about it is ever written down. Once the guests who remember are gone, the wedding is hard to tell apart from one that never took place.

Ferraris builds his theory, documentality, on that feeling. A social object, for him, is an act that has been inscribed. A promise becomes a debt when it is written into an IOU, and a sale becomes a title when it is entered in a register. His formula is *Object = Inscribed Act*. He distinguishes documents that inscribe an act, like a will or a contract, from documents that merely register a fact, like a thermometer reading copied into a chart, and the first kind is where social objects live. Records let a social fact outlast the moment it was made, and they carry it to people who were never in the room. This prelude is a small inscribed act in Ferraris's sense: a promise to explain, written down where readers can hold me to it.

### Signed, kept, and checked

Most of the records that make our social facts now are digital, and the best of them are signed. A digital signature does two jobs at once. It names who stood behind a record, and it binds them to these exact words, so that changing a single character makes it a different record that they never signed. Anyone who holds the record and the signer's public key can check both without asking anyone. The history a record carries with it, who made it, what it answers, what it replaced, is its provenance.

The setting I care about takes signed records seriously. Each person or organization keeps its own log of signed records. Records only get added: a correction is a new record pointing at the old one, and a withdrawal is a new record too. Records travel between people as messages, at any speed and in any order. There is no shared ledger in the middle and no clerk whose copy is the copy.

This post, I should admit, is kept the old way. It carries no signature a reader could check. Its provenance rests on the domain it is served from and on the platform that hosts it, which keeps the copy and could change it without leaving a trace. Like most of the web, it runs through a clerk.

## Codes, weights and the words around them

### Two kinds of machinery

The clerk has new machinery now, and two kinds of it sit between most people and their own records. The first kind I call codes: the schemas, forms, diagnosis codes, units and protocols that decide in advance what can be recorded and in what shape. The second I call weights: the trained parameters of machine learning models, which can now produce almost anything that can be recorded, on demand, from a summary of a meeting to a photo of one. Both are useful, and both let a record travel far from the person who made it. [Collaborative Systems in the Age of Codes and Weights](https://risav.dev/socioware) follows what happens to meaning on the way, through an app imagined for the purpose, in which a person with diabetes, her family and her clinics keep records together and every idea has to answer to one glucose reading. Its roots go back to [Anarchitecture](https://risav.dev/anarchitecture) and the essays on [time](https://risav.dev/time-i) from 2020, and to [a proof](https://zenodo.org/records/24026645) of what parties who keep their own signed records can do together.

The essay draws on two traditions that rarely share a page. One is continental philosophy, from Husserl to Žižek, a line of thinkers who each worried about what writing, images, technology or institutions do to meaning. The other is the sociology of records and classification, which studies how forms, files and categories move through organizations.

### Recipes and the lifeworld

A grandmother, asked how much water goes with the rice, says "until it looks right". Write that recipe down for a grandchild abroad and the words survive the journey, while the looking-right stays in the grandmother's kitchen. The grandchild can follow the recipe for years without ever learning what right looks like, until one day she cooks beside her grandmother and sees it.

Phenomenology is the part of philosophy that studies experience as it is lived from the inside, and Edmund Husserl founded it. Late in life, in a book called *The Crisis of European Sciences*, he gave a name to the everyday world of lived experience that all our sciences and techniques start from: the *lifeworld*. In a short text from the same period he gave names to what happens to the rice recipe. Meanings that get passed on settle in layers, each generation standing on what the last laid down, and he called this *sedimentation*. Going back down through the layers to redo the original seeing, as the grandchild does in her grandmother's kitchen, he called *reactivation*.

Records are sediment too, and a signed one leaves a way back down to whoever signed it. An introduction like this one is sediment by design, handing on words whose seeing happened somewhere else.

### A note on a shop door

A shopkeeper tapes a note to his shutter that says "Back in ten minutes". It works only because it can be read when he is gone. The same absence makes it slippery: nobody passing by knows when the ten minutes started, and if he forgets to take it down, the note goes on promising his return long after he has gone home. He can tape up the same note tomorrow, and it will still work.

Jacques Derrida built a philosophy on observations like these. Writing, for Derrida, is defined by its capacity to function in the absence of whoever wrote it. What a piece of writing carries of its writer is a *trace*: a mark of someone who is no longer present. And every sign has to be repeatable to work at all, a property he called *iterability*. One of his best-known examples is the signature, which is supposed to tie a text to one person at one moment and can do so only because the person can make the same mark again.

A digital signature keeps that trace fastened to the exact words it signed. This page reaches its readers without me in the room, and the byline at the top is the only trace of me that came along.

### A courtyard more traditional than tradition

A new hotel goes up, built to look like an old Newar courtyard house, with carved windows, brick and a small shrine in the middle. It looks more like the postcards than most courtyards ever did, and visitors photograph it as the authentic thing. After a while, some new houses start copying the hotel.

Jean Baudrillard called a copy of this kind a *simulacrum*: a sign that no longer points back to an original, or that comes before any original and shapes what follows. *Simulation* is the production of such signs from a model, and *hyperreality* is his word for the condition in which the copies feel more convincing than anything they might once have copied. He described a history of three *orders of simulacra*: the counterfeit that imitates an original, the mass-produced series that has no original, and the simulation generated from a code.

A generative model is a machine of the third order, able to produce a record of an event that never took place. An introduction is also a map drawn before the territory.

### Glasses and a mountain

People look through their glasses all day without seeing them, and notice them only when they smudge or slip. Martin Heidegger used examples like this to describe how we live among things: absorbed in using them, so that they withdraw into the use, until something breaks and they stand out. He called equipment in use *ready-to-hand*, and the same thing stared at as an object *present-at-hand*. His word for the kind of being that lives this way, always already thrown into a world and busy in it, is *Dasein*.

Later Heidegger turned to modern technology and named its danger *Gestell*, usually translated as *enframing*. Enframing is a way of revealing the world in which everything shows up as *standing reserve*, stock on call for further use. Seen that way, a forest is timber volume and a mountain is a season of permit fees. The same frame can fall on people, who start to show up as records, profiles and scores.

Heidegger's first pair of words applies to this page too. While the vocabulary works, it disappears into the reading, and a term that stops a reader has turned from ready-to-hand into present-at-hand.

### A blind man's cane

One of Merleau-Ponty's best-known examples is a blind man walking with a cane. After a little practice he stops feeling the cane in his hand and starts feeling the pavement at its tip. The cane has become part of how he perceives.

Maurice Merleau-Ponty developed phenomenology around the *lived body*: the body as the place from which we perceive and act, which is different from the body as an object in a medical chart. Perception, in his account, always happens from somewhere, through a body that moves among the things it perceives. His last essay, *Eye and Mind*, makes the same case through the painter.

A glucose reading comes from such a body, on a particular morning, before or after breakfast, and the record keeps the number long after it has lost the morning.

### Rankings and the end of big stories

A university once justified itself by a big story: it advanced knowledge, freed minds, built the nation. Today it is more likely to justify itself by a ranking.

Jean-François Lyotard described this shift in *The Postmodern Condition*. A *metanarrative* is a grand story that legitimates everything inside it, such as the march of progress or the emancipation of humanity. He defined the postmodern as incredulity toward such stories. What replaces them, he argued, is *performativity*: legitimacy measured by efficiency, by the best ratio of output to input. Without one big story, knowledge also splits into many *language games*, each with its own rules, a term he borrowed from Wittgenstein.

Platforms took the prediction literally, and a recommendation now counts as good because engagement went up. This post will be measured the same way, in views and reading time, and it is written as a set of small stories, the form Lyotard expected to survive the big ones.

### Forms under the pipal tree

A village water committee used to meet under the pipal tree and settle who gets water first by arguing until everyone could live with the answer. Now a project brings in forms, quotas and a portal, and the arguing stops because the portal decides.

Max Weber called the long spread of calculation, rules and bureaucracy through modern life *rationalization*. He pictured where it leads as a casing as hard as steel, which became known in English as the *iron cage*. Jürgen Habermas gave the worry a more exact shape. He split society into the *lifeworld*, where people coordinate through *communicative action*, talking things through until they reach understanding, and the *system*, the economy and the state, which coordinate through the *steering media* of money and administrative power. When the steering media take over practices that people used to coordinate by talking, Habermas called it the *colonization of the lifeworld*.

When a platform becomes the place where parties meet and then sets the terms of their meeting, the lifeworld has been colonized at the scale of the web.

### A referee's whistle

A referee blows the whistle and the goal stands. Players argue and fans shout, but everyone acts as if the whistle settles it. Then replays arrive on every phone, everyone can see for themselves, and the whistle starts to carry less weight, even when it was right.

Jacques Lacan divided experience into three registers. The *Imaginary* is the register of images and identifications. The *Symbolic* is the register of language, law and institutions, and of what Lacan called the *big Other*: the shared order that everyone acts as if someone guarantees, which is what the whistle relies on. The third register, which Lacan called the *Real*, is whatever resists being put into words and keeps returning anyway. Slavoj Žižek describes our time as a *decline of symbolic efficiency*: the big Other no longer quite works, fictions multiply, nobody fully believes any of them, and the Real comes back as shock.

A shared dictionary of terms needs a guarantee of the same kind. For this post the guarantee is modest: trust that the name at the top wrote it rests on a domain and a hosting platform standing behind the page, a small big Other.

### A passport, a field guide and a disease code

A passport is a small booklet that crosses borders, gets stamped, copied and scanned, and arrives everywhere unchanged. Bruno Latour called inscriptions like this *immutable mobiles*: they move, they combine with other inscriptions, and they keep their form along the way, which is why so much of science and administration is built on them.

Susan Leigh Star and James Griesemer studied a natural history museum where amateur collectors, professional biologists and administrators all used the same specimens and field notes for different purposes. They called such shared things *boundary objects*: flexible enough for each group to use in its own way, stable enough to keep one identity across all of them. Geoffrey Bowker and Star went on to study the International Classification of Diseases and showed how much politics a classification system hides, until a case fits none of its boxes. And Theodore Porter, in *Trust in Numbers*, showed how numbers let strangers trust each other's records at a distance.

A signed entry is an immutable mobile, and a shared dictionary of terms is a boundary object. This prelude tries to be a boundary object of its own, one text that a philosopher and a programmer can each use for different purposes and still recognize as the same thing.

### Passbooks, diagrams and conductors

An old bank passbook only grows. Each deposit and withdrawal is a new line, and a mistake is fixed by a new line that reverses it. Programmers call a record kept this way an *append-only log*. When each line is signed with a cryptographic key, anyone holding the log can check who wrote every entry.

An *ontology*, in computing, is a formal account of the kinds of things a system deals with and how they relate. David Spivak and Robert Kent proposed a version drawn as a small diagram of boxes and arrows from category theory, and called it an *olog*, short for ontology log. An *ontolog* is a shared dictionary where several ologs meet.

An *orchestrator* is the program that keeps the others in step, like a conductor in front of an orchestra. A collaboration is *monotone* when adding information never takes back a conclusion already drawn. Finally, *Conway's law* is the old observation that organizations design systems shaped like their own communication, or as the joke goes, four teams building a compiler will produce a compiler in four passes.

The passbook gives a fair test for this post too. The monotone way to correct it would be a dated note added at the end, leaving the original sentence visible above it.

Passbooks kept by many parties, ologs that translate between them and no conductor at the front add up to what the essay names in its last word: *socioware*, software whose units are parties and whose wiring is their agreements.

## Who seals the books

Without a clerk, the facts we rely on split in two: the ones anyone holding the records can still settle, and the ones the clerk was settling for us all along. The second kind is where *What Absence Costs* begins.

### Three endorsements

A neighborhood committee decides things by written proposal. At its founding meeting, its members agreed on a rule: a proposal goes to a vote once three members have endorsed it, and an endorsement is a signed record saying so.

Months later, Sita files a proposal to fix the streetlight by the temple. Hari signs an endorsement, then Maya, then Bishnu. The proposal is ready for a vote.

Two layers of fact are at work here, and the philosopher Brian Epstein gives them names. Grounding is what makes a particular fact hold: these three signed endorsements make this proposal ready for a vote. Anchoring is what put the rule in place to begin with: the founding meeting where the members accepted that three endorsements would be enough. Collective acceptance sets up the frame, and records fix the facts inside it.

Under a frame like this, every fact has what I will call a grounding condition: a condition on the records that says when the fact holds. *The proposal has three endorsements* holds when there are three endorsement records from three members. *Hari endorsed it* holds when Hari's signed endorsement exists. On this view, a question about authority becomes a question about grounding conditions: who can check a condition, and on whose records.

### What the night nurse cannot know

In the imagined clinic, a family and its clinics keep a diabetic patient's records together. The patient's meter signs each glucose reading. The laboratory signs each result. Doctors sign prescriptions. A family member signs when a delivery of insulin arrives. Each of them keeps their own log, and records flow between them.

It is two in the morning. The night nurse looks at what has reached her. A reading of 7.2 was taken at ten past ten last night, signed by the meter. The lab result from Tuesday is in. The last prescription she can see says ten units.

Some of what she knows will hold up whatever arrives next. The reading was taken; nothing that arrives later can untake it. The lab reported its result; a correction would be a new record, and the old one would still have been reported. Facts like these only ask that certain records exist, and adding records can never make them false. Logicians call a condition like that monotone: more input never takes away an output already drawn.

One thing she knows is different. *The current dose is ten units* asks for something more than the prescription she holds. It asks that there be no later prescription. And the day doctor may have signed one at nine in the evening that is still sitting on a laptop that has not synced. The nurse can be careful and competent and still give the wrong dose, because the fact she needs depends on the absence of a record, and no amount of looking at her own records can establish an absence.

Absence facts turn up everywhere once they have a name. *No objection remains open.* *This is the latest version.* *This license has not been revoked.* Each one holds of some set of records and fails of a bigger one. That makes them non-monotone, and it is why they are hard. This page carries one too. Whether a reader's copy is the latest version of this post is something the copy itself cannot settle, since an edit made after it loaded is the kind of record it lacks.

### A theorem from the machine room

Computer scientists ran into the same wall while building programs that run on many machines at once. Messages between machines can be delayed, reordered, or duplicated. Waiting for every machine to agree before doing anything makes a program slow and fragile. Acting on whatever has arrived makes it fast and sturdy, and leaves open whether every machine will reach the same answer.

In 2010 Joe Hellerstein conjectured an exact answer, and Tom Ameloot, Frank Neven and Jan Van den Bussche later proved it. It is called the CALM theorem, for Consistency As Logical Monotonicity. A distributed program can give consistent answers without its machines stopping to coordinate if, and only if, the program is monotone. Hellerstein and Peter Alvaro's [own explanation](https://cacm.acm.org/research/keeping-calm/) uses a cleanup example. A machine may throw away an object only when no other machine still points to it, and the only way to know it has heard from everyone is to ask them all and wait.

The theorem was written about programs. My claim is that it draws the same line through social facts the moment those facts are grounded in records kept in many hands. A fact with a monotone grounding condition can be settled by anyone who holds the records, in any order, and never has to be taken back. A fact whose grounding condition asks for an absence cannot be settled that way by anyone, however diligent, until some coordination has happened.

### "Any objections?"

Back at the committee, Sita's streetlight proposal has its endorsements, and the vote passes. But the committee's rule says a decision is final only when no objection remains open. The chair looks around the room and asks, "Any objections?" There is a pause. "Hearing none, the motion carries."

The ritual is older than any computer, and it does what CALM says must be done. The chair cannot know that no member, anywhere, has an objection in their pocket. What the chair can do is declare that the set of objections raised by now is the set that counts. After that, anyone who objects next week starts a new proposal, and this one stays closed.

Barry Smith once pressed Searle with the case of a painting that had gone unclaimed for decades, and asked who owns it. Smith pointed out that ownership in such a case rests on what he called a negative basis: the absence of any competing claim. Searle replied that disputes like that get settled by judges and lawyers. Both were right, and the theorem says why: a court closes the record, declaring that the claims before it are all the claims that count.

I call such a declaration a seal. A land registry seals the titles to a plot, and an editor closes an issue. Each of them is doing a status function declaration in Searle's sense: making it the case that a set of records counts as complete for a question. Once the seal exists, the absence fact can be grounded in the seal itself, which is a record like any other. *The decision of the meeting on the 12th* holds as long as that seal exists, and adding records can never remove it. The fact has become monotone again. Coordination was spent once, at the moment of sealing, and after that the fact travels freely.

Seen this way, an authority may know no more than the rest of us. In a world of records kept in many hands, no one can know that nothing is missing. What an authority has is the standing to make what is missing stop counting.

### Walking together

A seal can also come from somewhere other than a chair, a court or a registry, and Margaret Gilbert's best-known example, from her essay "Walking Together", shows where.

Two friends agree to walk to the market together. Halfway there, one of them speeds up and leaves the other behind without a word. Something has gone wrong, and both of them know it. The one left behind has standing to object: "I thought we were walking together." And neither of them, Gilbert points out, can call the walk off alone. It took both of them to begin it, and it takes both of them to end it. Gilbert calls what holds them a joint commitment. Through it two or more people become what she calls a plural subject, a "we" that has commitments of its own, which no one of its members can undo unilaterally.

A seal can be made this way. Instead of one chair, every party the question names signs a record naming the same set of records. The seal exists when all the signatures do. Each signature is an open, checkable expression of readiness. Every signer can see who else signed. And none of them can unseal it alone: reopening takes another record signed by the same people, so even a change of mind becomes one more record in the history.

At the clinic, this is a handover. Before the day doctor leaves, she and the night nurse co-sign the set of prescriptions the current dose is computed from, and the patient can sign it too. The nurse no longer has to guess whether a newer prescription is hiding somewhere. She is working from a sealed set that the people responsible for it committed to together. The authority to seal is still there, shared now among the people the question concerns.

### A picture the camera never took

Cameras have started signing their photos. The idea is that a picture carries a credential saying which device made it and that nothing has been altered since. In September 2025 a photographer named Adam Horshack showed that one camera with this feature [could be made to sign an image](https://petapixel.com/2025/09/04/nikon-z6-iiis-c2pa-functionality-has-a-significant-security-vulnerability) it never took: a graphic he had made on a computer, which then passed the official verification tool. The manufacturer [suspended the feature](https://petapixel.com/2025/09/05/nikon-suspends-c2pa-functionality-on-the-z6-iii-due-to-authentication-issue) and later [announced](https://nikonrumors.com/?p=209931) that it would revoke every certificate it had issued.

Two lessons sit in that story. The first is about what a signature can and cannot vouch for, which takes Baudrillard's map one step further. The [standard behind these credentials](https://spec.c2pa.org/specifications/specifications/2.4/explainer/Explainer.html) says what a credential checks: that the record is well formed, that it is untouched, and that a signer on a trusted list signed it. Those checks establish the kind of fact a signature can make: *this signer vouched for this file*. Whether the scene in the file ever happened is a different kind of fact, and no signature can settle it. As generative models write more records of events that never happened, the question to ask of a record moves from what it shows to who stands behind it.

The second lesson is about absence. A credential stays valid only while its certificate has not been revoked. Lawyers call statuses like this defeasible: they hold unless something defeats them, a revocation, an appeal, an exception that comes to light later. Every defeasible status, from a valid license to a current dose, hides an absence. A verifier that never received the revocation will go on calling the credential valid, without anyone doing anything wrong. Its grounding condition asked for an absence, and nobody had sealed the list of revocations it consulted. A revocation list that is published, signed and dated is such a seal, and a verifier that checks one can say *valid as of that list*. Preprint servers work the same way. Zenodo, where [Collaboration Is Composition](https://zenodo.org/records/24026645) lives, gives every version its own identifier and keeps the old ones, so each version is sealed as of its date. A blog post like this one has no such seal, which is one more reason the paper itself will go to a preprint server.

## A grid from 2014

In August 2014 I drew a grid for a group of thinkers, system architects and computer science students, as a roadmap for research we wanted to do together. It has twelve rows and three columns.

Going down, the rows climb. They start with structure: parties, and the networks between them. They rise to meaning: records, and what they stand for. Then they reach the collective: clusters of context that form, synchronize and gather. I pictured those clusters as planets pulling on one another and keeping in sync the way bodies in orbit do, and that picture is half the reason for the name we chose, Co-Planetary Systems. The first seven rows are the fundamentals. The last five are built from them: models and simulations, actions in the world, collaborations near and far, and finally the discipline that keeps theory and practice equivalent.

Going across, every row has to make sense three times. The language column asks what it means in words. The computation column asks what it costs when many machines compute as one. The mind column asks what a person can hold of it. At the bottom of the mind column sit two cells I only half understood when I wrote them. Row 11 says *SocioWare*: software whose units are parties and whose wiring is their agreements. Row 12 says *socio-aware software*: software that knows whose it is, who signed each part of it, and what each of them agreed to.

I have since walked the grid one column at a time, through [language](https://risav.dev/matrix-language), [computation](https://risav.dev/matrix-computation) and [mind](https://risav.dev/matrix-minds), and going back to it twelve years later felt like the reactivation Husserl described, digging down through a layer I had laid myself and redoing the seeing. The socioware essay is what I found at the bottom. *What Absence Costs* came from taking row 11 through all three columns at once. The computation column asked which parts of a collaboration cannot grow by addition, and CALM answered. The language column asked what kind of fact a decision or a dose is, and the theory of social facts answered. The mind column asked what people must do where records cannot settle things, and Gilbert's two friends on their walk to the market answered.

## Where authority goes

We are moving the records that make our social facts out of drawers and registries and into systems that hold them for us: platforms, clouds, chains, and soon agents that write records on our behalf. Each of those systems makes a choice, usually without saying so, about who gets to settle which facts. When one platform holds everyone's records, it settles all of them, monotone or not, and becomes the authority over facts that never needed one. A global ledger makes every participant agree on every fact, including the many that only ask for a record to exist.

The line I have been tracing points to a different design. Records stay with the people who signed them. Every monotone fact, from a reading to a declared agreement, is settled by whoever holds the records. Coordination is spent only where a fact asks for an absence, and only among the people that fact concerns, who seal it together. Authority is needed where absence matters, and where it is needed, it can be shared.

*What Absence Costs* makes this argument in full, with the objections a philosopher would raise, and will appear as a preprint on PhilArchive, its mathematics resting on *Collaboration Is Composition*. Until then, any objection that reaches me in writing goes into the log as one more record.

