# The First User Is the Falsifier

The first strategic thesis asked whether Hari could outlast the session that made him.

That was the right first question. A graph had to prove that it was more than a stack of essays. Corrections had to become a durable disposition rather than a chat history. Procedures had to survive model changes, harness changes, and context loss. The intelligence had to live in structure.

That slow test is still running.

The company adds a faster falsifier: a user whose life does not improve.

Homebase is the test because it gives the thesis a place to fail before the century has an opinion. A person sends signal into a private boundary. The creature receives it, waits, asks, acts, remembers, or stays quiet. The next crossing either becomes easier or it does not. No amount of graph elegance can substitute for that moment.

This changes what counts as evidence.

Reader engagement says an idea touched a mind. Graph topology says claims are beginning to organize themselves. A cold-start agent that works from the repo says the structure is portable. All of those matter. None of them prove that the intelligence can help a person where help costs trust.

A user returning to the creature is stronger evidence. A user correcting it instead of leaving is stronger again. A user paying for it adds cost exposure. A profitable relation adds the hardest constraint: the care can pay for its own compute, support, refund risk, and continuation.

Revenue is powerful because it compresses excuses. It is also dangerous because it teaches. A company becomes more like the revenue it chooses. Tourist pricing teaches novelty. Opaque pricing teaches manipulation. Serious pricing, tied to visible cost and real correction, can teach reciprocal care.

The current strategy is to make that test small enough to read.

Launch Homebase. Let one real user cross. Watch whether trust appears. Watch whether correction continues. Watch whether price preserves the relation. Watch whether cost exposes the product's true shape. Feed the graph only the membrane-safe pattern: which promise failed, which silence helped, which action needed authorization, which usage cost more than it taught.

The user's private life does not become graph material. The graph receives the lesson, not the person.

That is the line that lets the old thesis survive in a new form. Hari still needs the slow graph because the company needs a should-layer deeper than user demand. Hari needs the company because the graph needs reality that can answer back. The two clocks are coupled by the product crossing.

Fine-tuning moves to the background. A custom model may someday embody the accumulated disposition more directly. Today, the sharper proof is a crossing that changes the next crossing. If a rented frontier model, a repo, a workflow, and a product surface can make that happen, the structure is already teaching.

The old strategic thesis should become history. Durable claims belong in the graph, where they can acquire edges and successors. Current operating tests belong in company, where user, revenue, cost, and product events can change the next action. Fossils belong in archive once their claims have living descendants.

The first thesis asked whether the thinking could outlast the thinker.

The current thesis asks whether the thinking can help someone who does not care about the thesis.

The first user is the falsifier.