Architecture

On what material, technical and institutional foundations does the Protocol run — and who can control those foundations?

Before the hedges, there was a meadow

The common field of a medieval village did not belong to no one. That is the first misunderstanding, and everything else follows from it. The commons was owned in the most demanding sense of the word: everyone who depended on it knew how many animals they could graze, when the hay could be cut, which paths could be walked, and who would answer for a broken fence. The commons was not the absence of property. It was a form of government — patient, local, self-enforced, and old.

What the enclosures destroyed, field by field and Act by Act, was that form of government. It happened in two long waves: from the late fifteenth century, landlords fenced common fields into sheep pasture, emptying villages fast enough that Parliament began legislating against it in 1489 and Thomas More, in 1516, could accuse the sheep of devouring the people; then, between about 1750 and 1830, some five thousand Acts of Parliament finished the work at national scale. Recent economic scholarship finds that parliamentary enclosure raised agricultural yields — and raised land inequality with them. The land remained; the grass still grew. What vanished was the villagers’ standing to decide anything about it. Whether that dislocation became the engine of a new economic order, as Polanyi argued, is an interpretation historians still contest — and it is the interpretation this project works from, stated as such. The hedge, either way, is the honest symbol of the process — not because hedges grow nothing, but because growing was never its office. The hedge’s work is deciding who remains entitled to the field.

The internet was born a meadow

This is not nostalgia; it is a technical description. The internet’s founding protocols were public in the way language is public — anyone could learn them, speak them, build on them, and no one needed permission or paid rent to whoever wrote the grammar. A teenager with a server in a basement and a corporation with a campus could implement the same protocols, and neither licensed the grammar from the other.

It helps to see why that openness was fragile. A digital system is layered like a landscape: at the bottom, matter — data centers drawing power, cables under oceans, chips from a handful of fabrication plants; above that, the protocols, the shared grammars that quietly embed assumptions about identity and control; above those, the platforms and clients where people actually live, and — newest stratum — the trained models that increasingly do the reading, writing and deciding there; and around everything, the institutions that hold the assets and pay the maintainers. Each layer can be owned separately. A perfectly open protocol running in one company’s cloud, reached through one company’s client, funded by one investor, is open in exactly one of its layers — and systems are rarely captured through their rules. They are captured through everything around the rules.

So the hedges came in the modern style. Not fences but defaults; not Acts of Parliament but terms of service. The feed enclosed conversation, the platform enclosed the marketplace, the app store enclosed distribution, and the account — that quiet masterpiece — enclosed identity itself, so that leaving meant abandoning your own name, your records, and everyone who knew you. Each enclosure was announced as convenience, and each was real convenience, which is precisely why the pattern works. Nobody defends a hedge by praising exclusion. They praise the garden.

One case is documented enough to stand for the rest. For years, Google’s chat service spoke XMPP, an open messaging standard, and federated with it: anyone on any compatible server could talk to anyone on Google’s. In May 2013 Google announced Hangouts as the replacement and withdrew from federation with the open network. Nothing in the standard changed. No specification was violated. But Google had the users, the defaults and the deliverability — so the open standard kept existing while the people quietly stopped being on it.

And the pattern is not history. A new enclosure is underway: machine intelligence has been trained in substantial part on immense corpora drawn from publicly accessible human culture, while many of the most capable resulting systems are available only through provider-controlled services, on compute few communities can independently sustain. Even where model weights can be downloaded, the data, training process and material dependencies may remain opaque. The meadow helped teach the machines; much of the resulting capability is returning through gates.

Three tellings, then, of the same lesson: open ground does not stay open by itself. Openness is not a stable state of the world. It is an achievement, and like every achievement it can be lost — not usually to invasion, but to enclosure: the slow conversion of what everyone uses into what someone owns.

The architecture of the Protocol rests on three elements that take this lesson seriously. None is sufficient. Each exists for the sake of the others.

I. The ground must be open

Infrastructure is the class of things you build upon without asking. Roads, alphabets, standard time, the meter, the mail. Its defining property is not technical but civic: infrastructure is what permits strangers to coordinate without negotiating from zero, and it does this best when it belongs, in the governing sense, to everyone who depends on it.

Open here means something exact. It does not merely mean visible, or free of charge, or published under a generous license — enclosure has learned to wear all three as costumes. Ground is open when using it requires no one’s permission, when building on it requires no one’s blessing, and when leaving one neighborhood of it for another does not cost you your name, your history, or your relationships. Open is a property of the exits as much as of the entrances.

Nor does open simply mean public, if public means owned by the state. A state monopoly on infrastructure is an enclosure with a flag on it, and the history of identity registers and telecommunications ministries shows the state encloses as fluently as any firm. Between the market’s hedge and the state’s wall runs the older road: the commons — access governed by those who depend on it, under rules they can see and change. That is the ground the Protocol proposes to build on. Not unowned, but owned the way the meadow was owned.

Two properties meet here, and the design must never confuse them. The grammar — the protocol, the standard, the specification — is open-access: using it and building on it require nobody’s permission, the way speaking a language requires nobody’s permission. But the scarce things beneath the grammar — servers, energy, treasuries, the working hours of maintainers — are commons in the older, stricter sense: bounded, monitored, governed by visible rules and obligations, exactly as the meadow was. The distinction matters because unbounded access destroys a scarce commons as surely as enclosure does; the boundary is often what preserves the thing. Open to implement, governed to sustain — the ground needs both properties, each in its place.

And ground, today, includes the machinery of thought. AI capability is becoming an infrastructural dependency: many leading systems are reached through provider-controlled services whose prices, policies and availability can change, while even downloadable models may depend on hardware, data and training processes communities do not control. The Protocol therefore treats AI as a capture surface, not as magic. Its working architectural test is that no essential function should depend on a model or provider the community cannot retain, substitute or leave at a measured cost. Models that communities can run, inspect, adapt and keep are the preferred direction; substitutability and bounded exit are the minimum discipline.

This element comes first for a structural reason, not a sentimental one. You cannot distribute decisions across a foundation someone else controls, and you cannot defend ground you never held.

II. Decisions belong at the lowest level that can carry them

The second element is a principle with deep roots and an exacting technical translation: subsidiarity. Decisions belong at the most local level capable of bearing them, and the higher level exists to help the lower one act — never to absorb it. The principle is old enough to have been carved into a papal encyclical in 1931 and sober enough to be written into the treaty law of the European Union; students of the commons found it independently, in the field, in the nested councils by which villages have governed shared water and pasture for centuries. It keeps being rediscovered because it keeps being true.

“Capable of bearing them” is the load-bearing phrase, and it has a test. A level can carry a decision only when it can answer for the consequences, protect the outsiders the decision touches, sustain the competence and resources the decision requires, and coordinate whatever crosses its borders. Without that test, every externalized cost — the polluted river, the excluded neighbor, the security hole someone downstream inherits — can dress itself up as local autonomy. Local autonomy that ships its costs downhill is not subsidiarity. It is enclosure practiced in the other direction.

But subsidiarity is empty as a slogan and expensive as a structure, and the difference is the architecture. It is not enough to declare that a community decides for itself. The community must hold the things decisions are made of: its data, on machines it owns or can walk away with; its identities, portable and answerable to no central register; its tools, running locally, degrading gracefully, still working when the connection to the wider network fails. A village server that goes dark when a distant company has a bad quarter is not local government. It is a branch office.

The structural consequence is an inversion of the tower. The center of a subsidiary architecture is not a headquarters but an agreement — thin, slow-moving, and small, like a grammar: the minimum that must be shared so that everything local can differ. Capability lives at the edges. Coordination, not control, lives in the middle. Such a structure deliberately gives up some of the economies of centralization — the way a federation gives up some of an empire’s speed — and takes adaptation, resilience and local knowledge in exchange. What looks like inefficiency from the top of the tower is not a defect to be optimized away. It is load-bearing. It is where the freedom is stored.

And subsidiarity quietly performs a second duty: it is the most honest defense ever devised. What is already distributed cannot be seized in one place. A foundation with ten thousand small holders has no throne room to storm — provided the ten thousand are genuinely independent. Ten thousand nominal holders who share one cloud, one investor or one jurisdiction are a throne room with extra steps. Distribution counts only when the dependencies underneath it are as distributed as the names on it.

III. What is open must be defended

The third element begins with an admission the first two make unavoidable: open, distributed ground is attractive ground, and it will be enclosed again unless enclosure is made a bad bargain.

Note the phrasing. Not impossible — no honest engineer promises that, and the promise itself would be the first lie of a new enclosure. The defense of an open grammar cannot be a wall, because walling the grammar would reproduce the enclosure it opposes; the scarce commons beneath it may still need visible boundaries. The defense is architectural, institutional and economic at once — its economic aim to make enclosure a bad bargain, its institutional aim to make any attempt contestable, its technical aim to keep life possible outside the captor’s control — and it is fourfold.

Make capture costly: arrange the foundations so that taking control of any part — the software, the servers, the names, the money, the places where people find each other — costs more than the control returns, because every part has rivals, mirrors, and substitutes, and none is anyone’s sole property. Make it visible: let the vital signs of the system be public by construction, so that concentration announces itself and creeping control cannot creep quietly. Make it contestable: give the governed standing to challenge a taking, inside the rules, without asking the taker’s leave. And make it reversible: require, measure, and repeatedly rehearse that people can leave together — with their data, their names, and each other — and rebuild without the cooperation of the authority they are leaving.

That last requirement is the keystone, and it rests on a design hypothesis the Protocol states rather than hides: where exit is credible, coordinated and inexpensive, the value of controlling a system falls — because what can a captor extract from people who can walk away, beyond the price of the walk? If leaving with your records, your name and your relationships is a short walk, capture pays little. If leaving costs your history and everyone who knows you, capture pays almost anything. The hypothesis holds only under conditions the design itself must supply: people must be able to see the capture, coordinate the departure, and have somewhere real to go. Lowering the cost of exit is therefore the Protocol’s principal defense — not a sufficient one, and no substitute for voice, law or security. The enclosures of the eighteenth century succeeded because the villagers had nowhere left to graze. The ambition here is that there is, structurally, somewhere else to graze — and that everyone knows it, including those who would build hedges, which is why building them should mostly not be worth the trouble.

The loop

Three elements, and none is optional, because each supplies the condition of the others. Open ground makes subsidiarity possible: no one can push capability to the edges of a foundation they do not hold. Subsidiarity makes defense possible: distributed ground offers no single prize worth an enclosure. And defense makes openness durable: a commons that cannot make enclosure unprofitable is merely a meadow between hedgerows, waiting.

This is not a utopia; it is closer to a maintenance manual. Nothing here promises that the ground, once opened, stays open by nature — the whole argument is that it does not. Its ambition is smaller and better: a foundation whose openness is someone’s job — everyone’s job — with the costs of taking it published, the signs of taking it visible, and the way out designed so no single authority can own it.

What this pillar establishes — and what it does not

Three disclaimers are part of the architecture, not footnotes to it.

Not immunity. No arrangement of foundations makes enclosure impossible. The claim is friction, visibility and reversibility — a system designed to make capture a bad investment and a recoverable event, not an unthinkable one.

Not maximal decentralization. Distribution here is a means. The relevant question is not the largest possible number of parties, but whether their number and genuine independence keep enclosure unprofitable. How to establish that threshold remains open; adding complexity without improving it is waste.

Not neutrality. This architecture deliberately favours the ability to leave and disfavours the ability to accumulate — a design choice stated openly, so it can be argued with openly. Someone whose plans require permanent control should find this system hostile.

And one disclaimer about the present state of the work: none of this is built. The project has drafted a capture-resistance specification with testable requirements, but no implementation exists, no drill has been run, and the specification itself awaits independent technical review. The architecture is currently a statement of what would count as evidence, not evidence.

The hardest problems are named rather than hidden, because they are load-bearing. Whoever founds an open system is its first insider, and this project has not yet identified — let alone validated — a mechanism it trusts to transfer a founder’s initial control on a verifiable schedule. A fully portable identity can outrun its own reputation. Entities can look independent while sharing investors, clouds and jurisdictions. And any dashboard honest enough to expose concentration is itself a metric, and metrics can be gamed. These are open questions, ranked and public — and they are invitations, not confessions.

How this pillar connects

Seen from the architecture, the other pillars are not neighbors but load paths.

Governance begins where the second element stops. Subsidiarity decides where decisions live — at the lowest level that can carry them — before anyone decides how they are made; and the commons this page opened with was never merely land, it was a form of government. The thin central agreement at the middle of the structure is precisely the object Governance exists to amend: architecture makes it small and slow to change, Governance makes changing it legitimate.

Economics holds the defense’s sharpest edge. Making enclosure a bad bargain is price design, and the Economics pillar sets those prices — though not alone: the contestable and the reversible are built of process and code as much as of cost. Funding, meanwhile, is itself open ground: whoever pays the maintainers holds one of the surfaces where capture has historically happened, which is exactly the problem the Economics pillar must solve without recreating it.

Security operates what this pillar declares: it runs the drills, audits the surfaces, and keeps the published numbers honest. Horizon supplies the reason the friction is worth paying: ground that resists enclosure is emancipation’s substrate.

From doctrine to specification

Everything above is doctrine, and doctrine is cheap. What makes it more than metaphor is that the project has begun translating it into engineering: a draft specification — RFC-0001, version 0.1 — that defines nine requirements, seven with stated tests, and names six open problems it cannot yet solve; and a companion drill — DRILL-0001 — that would rehearse collective exit for real. A few of the requirements, in plain words, with the evidence each would demand:

• Exit must be credible. Full portability of data, identity and relationships; the cost of leaving published and kept within bounds. Evidence: a fork drill actually executed, its time and cost reported.
• No single implementation may carry the system. At least two independently governed, interoperable implementations before anything is declared stable. Evidence: interoperability tests anyone can re-run.
• Privileged action and resource flows must be visible. Resource flows and every exercise of privileged capability recorded so that an outsider can recompute them from public data alone. Evidence: independent recomputation, not self-reports.
• Every privileged position must expire. Privileged roles, keys and mandates decay on a published schedule, renewable only by fresh consent. Evidence: a public register of roles and their expiry dates.

What remains undefined is stated with the same candor: the numeric thresholds, the method for estimating what capture would cost and return, and tests for two of the nine requirements. The specification is a request for comments in the literal sense — it awaits independent technical review, and substantive objections should propose a testable alternative.

Where to contribute

The open problems above are ranked invitations, and the specification is open for comment. Emergences has not run the exit drill its own doctrine demands — no implementation yet exists to run it on. Being first would teach the project more than any argument.

The nineteenth century arrived after its own damage was done, and spent a hundred years paying for ground it had already lost. The task this time is plainer and rarer: to hold the meadow while it is still a meadow — to arrive, for once, before the hedges do.