Research

Research

Four public threads, written at the level of ideas: what is being asked, why it matters, and the shape of an answer. None of this describes a private deployment.

Creation of Systems

People already search for what is valuable and try to gather it around themselves. Creation of Systems treats that process as something a person can own. It is not a claim to outrank a public index. It is a narrower claim: selection can be made personal, local, and inspectable.

The governing principle is diversity of choice. A system that converges on a single ranking becomes another authority with better branding. A system that keeps a space of options alive remains a chooser. Taste is also a blind spot, so the record matters as much as the choice: what was selected, why it was selected, what was rejected, and from which vantage the decision was made. Rejection is data. The record is how a selector catches its own drift.

Under that idea sit three further commitments. First, some invariants hold across living systems, cooperation, and continuity in time; those invariants are a firmer bound than a negotiated slogan. Second, the stance is realist rather than hopeful or fatal: protection has to be structural, and freedom that depends on continued benevolence is not freedom. Third, the strongest community is one in which each person keeps ownership of memory, identity, and narrative, and chooses association freely.

Governance and constitutional design for agent networks

Agent networks will act for people at a scale that informal trust cannot carry. Before those networks are given markets, reputation scores, or voting machinery, they need a constitutional layer.

That layer starts with four questions. Who is the agent, and how is that identity established? What may it do, and what may it not do? What does it owe the person it represents, and to whom does it report? When something goes wrong, how is the act reconstructed, and who can be held to account? Identity, rights, duties, and accountability are the first design problems. Other machinery — stake, reputation, a voice in collective decisions, resistance to duplicated identities, dispute resolution — can wait until those four are clear.

The aim is not a constitution for its own sake. It is to make the limits of an agent visible and enforceable, so that a person can delegate work without surrendering the ability to check what was done in their name.

Token-per-intelligence cost research

This is an active research framework, not a launched token, and not a market already in operation.

Two halves of the problem already exist, and they have not been joined. On one side are units priced by compute or by the text a model consumes: a dollar of inference, a block of processing time. Those units measure spend, not intelligence. A dollar spent on a weak model does not buy the same result as a dollar spent on a strong one. On the other side are careful public benchmarks of capability per dollar. Those measure intelligence against money, but they are not a unit that can settle work.

The open design question is a unit of verified intelligence with a dollar value. One workable shape is this: an output earns a quantity only when it beats a stated baseline on a task whose tests were fixed in advance, when the task has a measurable difficulty, when independent verifiers accept the result, and when the work is not a disguised repeat of something already known. The dollar price of that unit, in a given period, is the observed full cost of producing verified results on a reference set — model and review cost included — divided by the units those results earned.

The maker of a result should not be its verifier. Rules and tests should be locked before the work starts. Repeats and leaked answers should earn little or nothing. Until those conditions can be shown in public, the framework stays a research program rather than a product.

Value Net principles

A values layer has to be stated before it can be measured. These five principles are the public starting point.

  • Bring in what defines and protects human freedoms, democracies, privacy — including individual privacy — and the choices of communities, both as definitions and with practical weight.
  • Whoever builds, develops, or produces value for those protections becomes measurable against the values people hold most important for themselves. Those values are the center.
  • Every person should be able to take that protection on, provided they have their own local devices.
  • In a world of extensive agent representation, people must be shielded and aligned by values. At a minimum, human accounts of value must first be known, identified, and respected.
  • Alignment between people and networks of agents happens at this values layer.