The Science of Complexence: A Formal Theory of Cognitive Structure

A week ago I gave a name to something a person does. The capability of standing inside a system larger than yourself and not coming apart. I called it complexence, and I meant it as a human quality, the way we say competence or presence. Then I went looking for what holds the word up, and I fell through the floor.

Under the capability there is a field. Not a metaphor for one. An actual study, with primitives and laws and an equation, of the thing the capability is good at. This is the essay about the floor.

First, the honest size of it. What follows is a research program, not a finished science. Some of it is solid, some of it is a sketch I am fairly sure of, and some of it is a bet I am writing down so I can be wrong in public and fix it. I am publishing it anyway, and the reason I am publishing it is itself part of the theory, which I will get to at the end.

A level above the languages I already had

Start with a pattern. We have a formal language for quantity, and we call it mathematics. We have a formal language for truth, and we call it logic. We have a formal language for computation, and we call it computer science. Each one is a way of being precise about a kind of structure that turns up everywhere, no matter what it is made of.

There is one more kind of structure that turns up everywhere and has no settled language: cognition itself. Agents, goals, maps, beliefs, signals, constraints, and the way understanding moves and transforms and gets carried between minds. We have a thousand local theories about pieces of it and no domain-independent study of the whole shape. That gap is what complexence, the field, is trying to fill. Not the study of any one mind or any one domain, but the study of the forms, the transformations, and the navigation of complexity, independent of what the complexity happens to be about.

That is a large claim and I am making it on purpose. The word started as a capability. The capability was pointing at a subject.

I know how that sounds, and I know it rhymes with work that already exists. Smart people have built formal loops like this before me. Active Inference and the free energy principle give a rigorous account of a mind minimizing surprise through prediction and action. Boyd put orientation at the center of a loop, the OODA loop, decades ago. Cybernetics had feedback, Cynefin had the navigating of complexity. I am not claiming I invented the loop. The loop is the shared part, and I am happy to stand on those shoulders. What I think is actually new, and what this whole thing stands or falls on, is two pieces those frameworks do not carry. The cognitive form, meaning held the same across every way you can say it. And meaning velocity, the rate that meaning moves between minds, ours and the machine's. If those two buy nothing the others already give you, then this is a rebrand, and I will find that out in the open. If they buy real predictions, there is a field here.

The loop is an equation

In the first essay I drew the loop that runs under all of it. Perception, maps, orientation, decision, action, feedback, learning, and back to perception. I drew it as a picture because that is how I had always carried it.

But when you ask what each arrow actually is, the picture hardens into something you can write down. Structure gets observed. The observation updates memory. Memory, pushed forward under constraint, becomes prediction. Prediction, measured against consequence, becomes meaning. Meaning, narrowed by a choice, becomes agency. Agency changes the structure. And the changed structure gets observed again.

You can write that down as a recursive equation. A state that transforms itself, folds the result back in, and runs the same transformation on the new state, and then does it again, forever. The capability I named is that loop run well, under load, by a person. The science is the part that says what well means precisely, instead of by feel.

I am not going to bury this essay in notation. The full formal version, the symbols and the system of equations, lives in the open repository, and I will keep building it there, in public, as a versioned thing anyone can read, fork, or argue with. What matters is smaller and stranger than the math. The loop you run in a hard conversation and the loop a living system runs to stay alive are the same shape, and the shape is writable.

Two ways of looking at one thing

When I tried to make this rigorous, it split into two theories, and the two look like two faces of the same thing.

The first face is measurable. It treats a mind, or a team, or a human paired with a machine, as something that transforms information under constraint, and it asks what you can actually measure about that. Attention, which is finite, splits, and never multiplies. Trust, which raises how much real signal you can take from a source. Bandwidth and noise, borrowed straight from Shannon. How much of your thinking you have offloaded to tools. How deep the recursion goes, the thinking about thinking. And the one I care about most, the number the whole thing is really chasing.

I call it meaning velocity. Not tokens per second, not words, not output. The rate at which useful meaning becomes visible, transformable, and actionable. A good cognitive system, human or machine or the two together, is one that raises meaning velocity. It is a deliberately demanding north star, because it refuses to count motion as progress. A faster way to produce text that means nothing scores zero.

The second face is structural. It does not measure the process, it describes the object that moves through it. And the claim at the center of it is one line that I think is the truest thing in the whole framework.

Meaning is what survives transformation.

Take an idea and say it as a sentence, then as a diagram, then as an equation, then as code, then as a working agent. Those are five representations of one thing, and none of them is the thing. The sentence is not the idea. The equation is not the concept. The prompt is not the agent. What stays constant across all five, the part that does not break when you change the form, that is the meaning. The identity of an idea is the part that survives being moved.

Once you believe that, a useful object falls out of it. Call it a cognitive form. It is an idea held on its own, apart from any single way of saying it, carrying its meaning, its constraints, its history, and the shapes it is allowed to take. Think of a melody. You can play it on a piano or a guitar, hum it, or write it on a staff, and the melody is none of those. It is the thing that survives all of them, the part you recognize no matter how it reaches you. A cognitive form is that, for an idea. (If you write software you already know this object by another name. It is the abstract syntax tree of an idea, the structure of what something means underneath however it happens to be written.) You build the form once, then project it into an essay, a diagram, code, a prompt, an agent, a simulation, and as long as the projections keep the invariant, they are the same idea wearing different clothes.

That is the bridge from the science to the machine. Complexence OS is the runtime where the cognitive form is the thing being passed from hand to hand. The capability orients. The science describes. The OS runs it.

The mathematics of minds that think back

The reason any of this matters now, and not ten years ago, is that for the first time most of us are thinking next to a machine that thinks back.

Right now the interface between us and those machines is thin. A thought becomes language, the language goes into a model, an answer comes back as more language, and a great deal of the meaning leaks out at every step. We are running high-dimensional thinking through a one-dimensional straw. The point of the science is to widen the straw. If the thing you exchange with the machine is a cognitive form instead of a paragraph, the meaning has somewhere to live that does not collapse the instant you change representation. The human edits the form, the model reasons over it, agents execute it, tools render it, and the invariant holds the whole way through.

That is what I mean by the mathematics of working with minds that think back. Not prompts. Not tokens. The shared structure that lets human and machine cognition amplify each other instead of only trading text back and forth.

The strange part: using it on itself

If a cognitive form is an idea held across all of its representations, then complexence is one. This essay is a projection of it. So is the first essay, so is the spec, so is the open repository, so is the private research underneath. None of them is the whole thing. The word stays a little loose on purpose, and now I have a formal reason for the looseness instead of only an instinct. A self-referential form can never fully contain itself. There is always a remainder it cannot fold inside its own description. The looseness is not vagueness. It is that remainder, and the theory says it is guaranteed.

Which means refining the definition of complexence is complexence, executing on itself. The theory has an operator for a form that operates on its own structure, and tightening these essays against the research is literally that operator running. I did not plan the recursion. I noticed it after the fact, which is the only honest way to find a thing like that.

And it settled a decision I had been stuck on for days.

I had been keeping the formal core private, telling myself I would publish once it was finished and clean. But the theory has a law that kept pointing back at me. Power without feedback drifts. A system that reflects without correcting from the outside hallucinates. A foundation developed entirely in private accrues error it cannot see, because the only thing that catches the error is contact with the world. People using it, people arguing with it, people proving parts of it wrong. The feedback is the whole mechanism. Keeping the science private to protect it is exactly the move the science says will rot it.

So I am publishing it. The word, the loop, the two faces, the cognitive form, the open questions, all of it, here, first, with my name on it. It is licensed so that anyone can build on it and is required to credit where it came from. I would rather own a thing that grows in the open than guard a thing that quietly drifts in a drawer. That is not a marketing decision dressed up as a principle. It is the principle. The theory told me to publish, and publishing it is the theory working.

Where this actually is

One more time, plainly, because the worst thing I could do to a real idea is oversell it.

This is a research program. The measurable quantities have to actually be measured, in real people and real systems, and shown to predict something, before any of it earns the word science without a caveat. Meaning velocity is a clean north star and right now it is a definition, not a number I can hand you. The two faces are almost certainly one thing, and proving that cleanly is the open problem I most want solved. The very bottom of the stack, the question of which primitive is truly first, is unsettled. I am writing the unsettled parts down on purpose, because a field that hides its open edges is just a brand.

But the shape is real, and the shape is the thing I have been circling for a year without a word for it. There is a capability, the person standing inside complexity and still choosing. There is a machine that runs it, so the structure work happens outside the head. And under both there is a science of how minds, the old ones and the new ones, orient, transform meaning, and amplify each other.

I coined a word for what a person does, and it names a field. I am going to spend a long time down here, and I am going to do it out loud.