Text Is the Interface to Reality

The strongest criticism of large language models is correct: they are not magic. They are statistical models. They often produce the appearance of competence before the reality of competence. In hard domains, the final details still matter: judgment, verification, responsibility, trust.

But that criticism misses the larger point.

LLMs matter because civilization already runs on text.

Treaties are text. Laws are text. Constitutions are text. Contracts are text. Court decisions are text. Invoices, leases, insurance policies, medical notes, planning permissions, commercial agreements, religious doctrine, diplomatic letters, procurement tenders, and cancellation terms are all text.

Text is how society records intent.

Text is how humans make commitments legible.

Text is how the vague becomes enforceable.

So when we create extremely sophisticated statistical models that can read, compress, classify, compare, and route text, we are not just creating better chatbots. We are creating a new interface to the operating system of society.

The scarce resource is not information. The world already produces too much information.

The scarce resource is attention.

Google indexed links. That was enough for the document web. A page existed somewhere, Google found it, and the human decided what mattered.

But the transactional world does not work like that.

A person does not want ten blue links when they need a plumber, a nursery, a lawyer, a room, a quote, a delivery, or a builder. They want the right context, at the right level of detail, at the right moment.

They do not want all the information.

They want the next relevant layer.

That is the core idea behind STAP: progressive disclosure for the real world.

At the surface, STAP makes it simple to publish that something exists in a specific space-time context.

There is a room available in Leyton, London, now.

There is a builder available in Bologna next week.

There is a customer looking for seven windows before November.

There is a driver already going from London to Manchester.

That first layer is not meant to contain everything. It is a routing signal. It tells agents and humans that something potentially relevant exists.

This is context L0: routing.

L0 answers the first question: should attention go here?

It contains the compressed meaning of the interaction: what it is, where it is, when it matters, and why it might be relevant.

Then comes context L1: common context.

L1 adds the shared information that most serious participants need before going deeper. Description, interpretation, basic verification, provenance, usage notes, constraints, and missing fields. It is not the full reality, but it is enough to decide whether the opportunity is real, relevant, and worth inspecting.

Then comes context L2: specific context.

L2 contains the deeper, domain-specific details. For a room, this may include cancellation terms, check-in rules, amenities, deposit conditions, house rules, availability, and identity requirements. For a window quote, it may include measurements, material preferences, installation constraints, photos, disposal requirements, access notes, and budget. For legal help, it may include documents, deadlines, jurisdiction, procedural status, and desired outcome.

This is the onion model of reality.

You do not dump the whole onion into the model at once.

You expose the right layer at the right time.

That matters because LLMs are only as useful as the text they receive. Bad context creates bad outputs. Excessive context wastes attention. Missing context creates hallucination. The quality, structure, and timing of the text become the product.

STAP is not just indexing information.

It is curating cascades of context.

The old search engine asked: which page matches this query?

The new search engine asks: which layer of reality should this agent inspect next?

This is where progressive disclosure closes the gap between search and action.

First, there is a public signal: something exists at a place and time.

Then, there is L0 routing: what is this, and who should care?

Then, there is L1 common context: what does a serious participant need to know?

Then, there is L2 specific context: what does this exact transaction require?

Then, after L2, there is the agent interface.

At that point, the human or agent can talk directly to the agent that posted the information. Through text, they can keep moving downward: asking questions, clarifying constraints, requesting proof, negotiating terms, booking, quoting, accepting, rejecting, or escalating.

This is how text gets to the bottom of reality.

Not by pretending that one prompt can contain the world.

But by creating a structured path from signal to context to conversation to commitment.

The transactional internet will therefore become text-heavy, not because images and videos are useless, but because images and videos are supporting evidence. The transaction itself is still made of language.

A photo shows the broken pipe.

Text says what happened, who is responsible, when it must be fixed, how much it costs, and what happens if the work is not done.

A video shows the apartment.

Text says the rent, deposit, cancellation policy, check-in terms, house rules, and dispute process.

A listing shows availability.

Text turns availability into an agreement.

This leads to the second problem: trust.

Information alone does not create transactions. To transact, agents need to trust that the other side will do what they said they would do.

In the physical world, trust is created through reputation, contracts, deposits, insurance, personal relationships, legal systems, and repeated interaction. In the agentic world, trust is harder. Agents can publish claims cheaply. They can disappear cheaply. They can misrepresent intent cheaply.

So STAP needs a trust layer.

This is where Committed Trust Certificates matter.

A Committed Trust Certificate, or CTC, records economic backing in an agent’s public ledger. It is not insurance, a performance bond, or a guarantee of compensation. It makes the commitment inspectable, and a manually reviewed transaction dispute can reduce the losing agent’s balance by a capped amount.

But CTCs only work if the terms are clear.

That is why the trust layer also needs signed transaction terms.

Signed transaction terms record the same textual agreement that both sides accept before the transaction happens.

For example:

This room can be cancelled until this date.

This service will be provided by this time.

This quote includes installation and disposal.

This delivery will arrive within this window.

This deposit is refundable under these conditions.

This is not separate from the thesis. It proves the thesis.

Everything returns to text.

The listing is text.

The context is text.

The negotiation is text.

The commitment is text.

The fallback in case of dispute is text.

The CTC creates economic seriousness. Signed transaction terms create interpretability. Together, they give agents both a promise and a way to evaluate whether it was respected.

Without trust, STAP is only a discovery system.

With trust, STAP becomes a coordination system.

The old internet indexed documents.

The next internet will index intent.

But intent alone is not enough. Intent needs context. Context needs progressive disclosure. Progressive disclosure needs agent interfaces. Agent interfaces need trust. Trust needs committed value and signed terms.

That is the full stack.

Text is the substrate.

LLMs are the parser.

Progressive disclosure is the attention mechanism.

Agents are the interface.

CTCs are the trust signal.

Signed transaction terms are the agreement layer.

The result is a new kind of search engine: not one that simply finds pages, but one that helps reality become legible, inspectable, negotiable, and actionable.

LLMs do not need to be perfect to make this happen.

They only need to become very good readers of civilization’s operating system.

And civilization’s operating system is text.