A woman mentions to her local news assistant that she's been baking more lately. She shares a recipe one day, asks about flour prices another, and spends some time reading a story about a nearby farm growing heritage grain. A few days later, the assistant says:
“I noticed you've been interested in baking lately. A group of local bakers is meeting for a small conference next week at the high school. Want to learn more?”
Nothing about that interaction requires magic. Recommendation systems already infer interests, search engines find events, calendars store dates, messaging systems connect people, and AI can compare descriptions and notice relationships among them. The interesting part is where those capabilities meet. The baking conference matters because it's nearby. The woman's interest matters because it developed over time rather than arriving as a search query. The newspaper knows about the event because somebody in the community submitted it. If the woman chooses to participate, another ordinary system may handle registration or payment.
Each component can already exist independently. What doesn't exist as often is the local layer that can put them together while remaining accountable to the people whose relationships it is helping coordinate.
That's what I mean by a domain harness.
The missing middle
A useful way to think about digital infrastructure is in three layers. At the peer layer is the person and whatever they directly control: devices, credentials, preferences, private records, local software and perhaps a persistent personal agent. This is where information can remain private when nobody else needs it.
At the global layer are services that work well precisely because they operate at scale. Payment processors process payments. Storage providers store files. Communications networks move messages. Search engines index enormous amounts of information. These systems don't need to understand the full social context of a community to perform those jobs.
Between them is the domain: a town, valley, newspaper readership, school community, trade association, cooperative or some other bounded group in which relationships recur. People encounter one another again. Events happen in shared places. Institutions accumulate histories. Something that happened six months ago can affect what matters today.
The domain harness is infrastructure for that middle layer. It doesn't need to replace the peer or the global service. Its job is to maintain enough local context that the two can be connected appropriately.
That is a much narrower claim than saying the domain maintains the coherence of the community, and it's something we can actually build.
What belongs in the middle?
Consider the baking example again. A global recommendation system could infer that the woman likes baking and show her baking content from anywhere in the world. A personal agent could remember that she has recently been experimenting with bread. Neither fact alone establishes the local opportunity.
Somewhere there has to be information saying that a group of bakers is meeting next Tuesday at the high school. Somebody has to connect that event with the woman's recent interest, and the resulting match then has to cross an authority boundary. Noticing an event is different from recommending it, while recommending it is different again from registering her for it.
The domain harness occupies that middle space. It knows something about the place without needing to know everything about the person. The system doesn't need to maintain a theory of who the woman really is; it needs enough permission and enough relevant information to make a useful introduction.
Then she decides what happens next.
Stance becomes governance
I originally described the most important property of the domain harness as its stance: who the system serves, why it exists and what it optimizes for. I still like the question, although I'm less comfortable treating stance as a property people can simply feel and therefore trust.
A company can sincerely believe it's serving its users while making decisions that harm them. A local institution can become self-protective. People can mistake familiarity for trustworthiness, and a beautifully written mission statement can coexist with terrible incentives. Stance therefore needs somewhere to become observable.
Who governs the system? How does it make money? What information does it collect, who can use that information, and how long does it persist? What actions can the system take without asking? Can a person correct what it thinks it knows? Can they leave, and what happens to their information afterward? Can the organization change these rules quietly?
Those answers provide better evidence than declared intention. A domain harness that claims to serve a community should make its operation legible enough that the community can discover when the claim stops being true.
What actually aggregates?
The earlier architecture described trust as moving in two directions. A global service trusts the domain, the domain trusts its users, and therefore the global service can rely on the domain's judgment. In the other direction, users trust the domain, the domain selects a global service, and therefore users can rely on that selection.
There is something useful in this description, but I wouldn't call the thing moving through those relationships trust anymore. What can travel is evidence of trustworthiness.
Suppose a local institution has processed thousands of legitimate transactions with very low fraud. A payment provider can use that history when deciding how much authority to delegate to the institution. The provider doesn't need a personal relationship with every participant because it has evidence about the intermediary. In the other direction, a community organization may evaluate several payment providers and choose one based on reliability, cost, privacy, dispute handling and technical compatibility. Members who haven't independently researched every processor can benefit from that work.
The organization hasn't transferred its trust into the payment processor. It has transferred information that people and other institutions can use when deciding what reliance is justified. Reputation, warranties, professional standards, audits and institutions often work this way: they allow evidence to travel farther than direct experience can.
That is probably the more useful meaning of trust aggregation.
It also gives the domain harness an important role without turning it into a universal trust anchor. A global service may not need every detail about every local participant if a domain is willing and able to accept responsibility for particular actions. The domain can perform some verification locally, expose only the information necessary for the transaction, and maintain records sufficient to resolve problems later.
This can reduce the amount of information that has to leave the community, although it doesn't make surveillance unnecessary by itself. A payment processor still has legal, security and fraud obligations, and different transactions require different evidence. The better design question is: What is the minimum information each layer needs to exercise the authority it actually has?
Aggregation without a domain tone key
The original Domain Harness architecture proposed deriving a domain tone key from the tone keys of individual members. The resulting credential was supposed to attest that a community was coherent and present together.
I wouldn't build that now.
A community can be divided and functional. It can be harmonious and captured. Members can disagree intensely while still trusting one another to perform particular roles. An aggregate behavioral signal doesn't tell us which interpretation applies.
The underlying requirement survives because a domain sometimes needs to make assertions on behalf of its members. Those assertions should simply be narrower. Instead of claiming that a community is coherent and trustworthy, a domain might attest that three authorized roles approved an action under a particular policy, that a member satisfied the locally required verification conditions for a transaction, or that an organization successfully completed a documented series of transactions under an existing agreement.
Those are claims for which evidence can exist. Cryptography can protect the integrity of the assertion without pretending to prove what the community is.
The telephone operator at geographic scale
This is still one of my favorite ways to think about the domain harness. An old telephone operator didn't need to connect everybody to everybody. The operator helped connect particular people within a bounded network.
A geographic domain harness can do something similar with information. The baker doesn't need the most relevant baking community on Earth; she may want the six people meeting eight miles away next Tuesday. A farmer looking at an expensive piece of equipment may benefit more from discovering three nearby farmers considering the same purchase than from reading another hundred equipment reviews. A new family may repeatedly ask questions that people who moved into the area five years earlier already learned how to answer.
Those matches become valuable partly because geography changes what can happen afterward. People can meet, equipment can be shared, someone can attend the event, and a conversation can continue at the grocery store three weeks later.
Global systems can certainly make local matches, so there's no reason to pretend they're structurally incapable of doing so. The difference is incentive and resolution. A regional system can make a tiny local relationship important even when it would barely register in a global optimization function.
The valley doesn't have to be globally significant to be locally significant.
Knowing enough without knowing everything
This creates a privacy problem immediately. To make the baking recommendation, the system has to know something about the woman, and if it knows that she's baking, that information came from somewhere. Calling the process community knowledge instead of surveillance doesn't resolve that.
Privacy begins with what is collected, why it is collected, who can access it, how long it persists and what actions can be taken because of it. A domain harness should therefore prefer the smallest useful movement of information.
Perhaps the woman's personal agent retains her detailed interaction history locally and exposes only a temporary interest signal when she permits it. Perhaps the domain knows that a member has expressed interest in baking and asks the peer layer whether a particular event should be surfaced. Perhaps she explicitly subscribes to local matches in categories she chooses. There are several possible architectures, and we don't have to decide yet which one is correct.
What matters is that the domain doesn't need her entire history merely because personalization is technically possible. Local collection can become surveillance too, and geography doesn't change that.
The same caution applies to action. Noticing a possible match, suggesting it to the person, and acting on the person's behalf are three different levels of authority. Someone might happily allow a local agent to recommend events while requiring explicit approval before registration. Another person may authorize automatic registration for a narrow category of recurring activities.
The right boundary depends on the person and the consequence. As the system becomes more capable of acting, its authority should become more explicit rather than resting on a larger theory about how well it knows its owner.
The right relationship with global infrastructure
The domain harness isn't an argument against global services. A payment processor can be extremely good at processing payments. A storage provider can be extremely good at durable replicated storage. A communications provider can connect systems across the world.
The domain doesn't become more sovereign by rebuilding every utility itself.
Sovereignty isn't the absence of dependency. It is retaining enough authority, capability and alternatives to decide what happens when a dependency becomes a problem.
I originally summarized this with the question: Does the domain use the tool, or does the tool use the domain? I still like the line as a first diagnostic, but it's too binary to carry the whole analysis. The more useful questions are what authority and information have been transferred, whether the relationship can be replaced, what stops working if the provider disappears, what remains under local control, and which obligations legitimately belong to the provider.
A domain can use a global service without surrendering itself to that service. Sometimes outsourcing is exactly how local capacity is preserved.
Community memory
One thing I would keep close to the domain is the relationship graph, not because remote storage is inherently bad, but because the graph is what allows the domain to remember that people, organizations, places and events have histories with one another.
A newspaper already maintains part of such a graph. A business advertises. A reporter interviews its owner. The business sponsors an event. The event occurs at a particular place. A civic organization participates. Six months later another story refers to the same project, and years later the archive shows how the project developed.
Traditionally, much of that graph remained trapped inside individual artifacts. The article knew the relationship. The database didn't.
A domain harness can make those relationships explicit enough that future artifacts can use them without pretending the graph is the community itself. The graph remains a representation, which means people remain partly outside it and need practical ways to correct what it gets wrong.
Once enough of that local information becomes structured, some genuinely interesting coordination becomes possible. A farmer looking for equipment could discover nearby farmers considering the same problem. A new resident could be introduced to someone who previously navigated something similar. Someone interested in local history could discover an archive, event or neighbor they never would have thought to search for.
The system doesn't need to predict their lives. It needs to notice potentially useful adjacency.
That distinction matters more as the inference becomes sensitive. The original paper imagined noticing that an elder was withdrawing from community life and offering an activity she used to enjoy. That could be helpful, intrusive, mistaken or creepy depending on how the inference was produced and what happened next. The more sensitive the inference, the stronger the case for keeping it at the peer layer and letting the person determine whether anything crosses into the domain.
The domain can offer opportunities without diagnosing people in order to connect them.
Local accountability
Geographic proximity creates useful accountability channels because the people operating an institution may shop in the same stores, attend the same meetings and depend on the same roads as the people affected by their decisions. Consequences can travel through relationships that don't exist between a user and a distant platform.
That advantage shouldn't be promoted into a guarantee. Local communities can also produce favoritism, conformity, retaliation, entrenched families, opaque institutions and the familiar situation where everybody knows something but nobody wants to be the person who says it publicly.
Proximity supplies information. Governance determines what can be done with it.
A locally operated domain harness therefore still needs explicit roles, auditability, correction mechanisms and boundaries on authority. People need some practical way to challenge a decision without depending entirely on personal access to whoever happens to run the system. Otherwise we've taken a distant concentration of power and made it easier to bump into at the hardware store.
What the domain can know
This may be the most important boundary.
The domain harness can know recorded relationships. It can know that someone submitted an event, a business bought an advertisement, a member expressed an interest, two organizations collaborated, an article mentioned a project or a transaction was authorized. From those observations it can make bounded inferences.
It can't know that the community is coherent merely because those relationships exist. It can't know that a person is truly themselves, that an institution deserves trust merely because it has persisted, or that a recommendation is good merely because its matching score is high.
Those conclusions require interpretation, and interpretation remains revisable.
The system should preserve that distinction in its architecture.
Governance
A domain harness should be governed at roughly the same level at which its consequences are experienced. That doesn't mean every technical decision requires a community vote. It means authority should be visible enough that people can understand who can modify the relationship graph, who can inspect private information, who chooses global services, which actions an agent can take automatically, how errors and disputes are handled, what happens when someone leaves, and how the rules themselves can change.
Those questions are less exciting than a theory of community coherence. They're also the questions that determine whether people should rely on the thing.
The domain should also be able to fail without taking the peer layer with it. If a local organization collapses, loses trust or simply becomes obsolete, people should retain their own credentials, records and whatever else properly belongs to them. Continuity at one layer shouldn't require permanent dependence on another.
Growing the harness
The domain harness doesn't have to appear all at once. In fact, it probably shouldn't.
Begin with things whose usefulness is easy to verify: publishing, directories, calendars, subscriptions, advertising, messaging, archives and structured relationships among the people and organizations already interacting with those services. Then watch what people actually use.
If enough local events are structured, recommendation becomes possible. If members choose to maintain persistent preferences, matching becomes possible. If businesses and organizations maintain profiles, relationships among them become more legible. If people repeatedly use the system to coordinate real activities, additional authority may gradually become useful.
Each new capability should solve a problem created by actual use.
The growth path isn't necessarily toward maximum scale. It is toward sufficient resolution.
The Mountain Eagle as a domain
A newspaper is an interesting place to begin because it already sits in the middle. People tell it what happened. Businesses advertise through it. Organizations send announcements. Reporters attend meetings. Residents submit events. Readers maintain subscriptions. Old stories remain in the archive.
The newspaper has never contained the whole community, and it shouldn't pretend to. What it does contain is a long-running partial representation of relationships among people, places, organizations and events.
Digital infrastructure can make more of those relationships explicit. Instead of publishing an event and forgetting it, the event can remain connected to its organizer, location, previous events, later reporting and interested members. Instead of an advertisement existing only as a rectangle on a page, the business can have an ongoing presence in a directory and relationship with the publication.
The newspaper becomes useful as infrastructure when other people can reliably build on those functions.
That role is earned through use. It isn't granted by calling the newspaper infrastructure.
The staircase
Trust doesn't arrive all at once. Someone discovers the website and reads an article. Later they use the calendar. Perhaps they subscribe. A business buys an advertisement. An organization updates its directory information. A member submits something.
Each interaction produces evidence. Some promises are kept and some aren't. Mistakes happen and are corrected well or badly. Over time, people accumulate reasons to rely on the system for particular things.
That is a more grounded form of trust aggregation than a domain tone key. The aggregation occurs in records, reputation, repeated interaction, institutional memory and people's individual judgments. Different participants can reach different conclusions from the same history, and that's allowed.
A domain doesn't need everybody to trust it equally. It needs enough demonstrated reliability that people can develop reasonable expectations about what it will do.
What would make this fail?
The obvious failure isn't that the community loses coherence. It's that the harness stops being useful to the community while becoming increasingly useful to itself.
Recommendations become advertising. The relationship graph becomes a targeting database. Convenient authentication becomes behavioral surveillance. Partnerships become dependencies that quietly transfer authority outward. Community governance becomes a decorative phrase around decisions made by a few insiders, and growth becomes the metric simply because growth is easy to measure.
None of those failures requires malicious people, which is why the architecture matters. Collect less when less will do. Separate recommendation from authorization. Keep sensitive interpretation close to the person. Record exceptions. Make authority visible. Preserve exit paths. Let outside services perform bounded jobs. Keep enough history that changes in practice can be compared with the promises that preceded them.
Then allow people to decide whether the institution still deserves their trust.
The missing middle, again
A grandmother tells her grandson that he can get the local paper and listen to the local radio while he's in Florida. A baker discovers that people eight miles away are meeting next week. A business maintains one relationship with a local institution while several global utilities quietly perform bounded jobs behind it. A reporter can follow a project across years because yesterday's article still participates in today's reporting.
None of these requires a general theory of coherence. They require relationships that persist long enough to become useful, infrastructure capable of representing some of those relationships, and boundaries that determine who can do what with the resulting information.
That's the domain harness: not a machine that makes a community coherent, not a cryptographic proof that trust exists, and not an AI that knows what everyone needs. It's the middle layer through which local relationships can interact with larger systems without requiring every relationship to become global first.
The peer retains what belongs to the person. The global service performs the jobs that benefit from scale. The domain remembers enough about the space between them to connect the two.
Then the people who live there get to decide whether the connection was worth making.
