Published live, as the work happens. This captures a conversation (2026-07-02) and continues the thread of the recursive-fitness note: that note asked whether the process can improve its own improving; this one asks what kind of thing the process is, and why it has the seats it has. The name is a scaffold, deliberately provisional. The Wolfram frame it borrows is itself contested — this enters at forming, valued for what it generates, held to the hard-to-vary bar rather than to proof.
The project, named plainly
This is an augmented intelligence project: Fuller's World Computer, brought to life at portfolio scale. Fuller's design brief for that machine — hold the whole inventory, and make the better configuration so evident that adoption happens by "spontaneous cooperation" — turns out to contain the bridge test, decades early. The machine was never meant to hand down verdicts. It was meant to make bridges visible.
The division of labour, as it has actually settled: the human seat conceives the why and the what, and trains the agents — the fitness functions, the method, the gates, the register are the training. The machine imagines the how: it coordinates the search, slices the what into workable pieces, and hands the sharpened why back with every result.
The Wolfram grounding
In the Ruliad picture, everything computationally possible is already there; what any observer experiences is fixed by two characteristics of observers like us: we are computationally bounded, and we have a sense of persistence over time. Perceived physics is what such observers can extract from the Ruliad. The claim here is that the same two characteristics, distributed differently across two kinds of seat, explain why this project has the architecture it has:
The human seat is bounded but persistent. It cannot search the space — one thread, finite budget (in the foundation's vocabulary: a finite distinction budget, which is computational boundedness stated natively). What it has is continuity: it remembers what the work is, why it matters, and it keeps coming back. It is the project's underlying attractor.
The amplifier seat is broad but ephemeral. Per session it searches far more of the space than the human thread can. But it has no native persistence at all — each session is a fresh sampling that would, left alone, wander down whatever rule-path it landed near and converge back to its training's box.
The failure modes prove the pairing. The amplifier alone drifts: re-derives, forgets the why, mistakes its distribution for the space. The human alone chokes: a search starved to a single thread — the concentration critique pointed inward. The augmented system is the one configuration in which the two defects cancel. Neither seat does the other's job; each supplies exactly the observer-characteristic the other lacks.
The coupling is memory
What "training the agents" actually is: installing the human seat's persistence into the ephemeral computer. Written memory — the fitness functions, the standing rules, the register of what matters and why — is prosthetic continuity, so that each fresh sampling wakes up already inside the attractor's basin rather than at a random point in the box. The world computer's persistence is borrowed, and it is borrowed from the seat that has it.
Specification versus continuity
This yields a clean rule for what can be delegated and what cannot. Specifications delegate: a fitness function, a rubric, a gate, a method can be written down, installed, and run without their author. Continuities do not: the attractor is not a rule but a re-instantiation — it exists because the persistent observer returns, and only for as long as the returning continues. The project is a standing wave, maintained by return.
This also squares the architecture with authorlessness. Everything produced here is open — take it and run with it, without attribution, with impunity; ideas are non-rivalrous and enclosing them starves the search. But one thing cannot go authorless in the structural sense: the attractor seat must be occupied. It is a role, not a name. Fuller's own credit line — "adjuvant: Kiyoshi Kuromiya" on the posthumous Cosmography — named the complementary role the same way: the adjuvant, in the immunological sense, acts on nothing itself; it amplifies the primary system's own response. The borrow transfers mechanism, not vibe, in both directions: that is what the amplifier seat is, and the attractor seat is what it amplifies.
What this hands the foundation
Distinction Physics' primitives — distinctions are primitive, cost energy, observers have finite budgets — are the OLU conditions stated natively. That upgrades the foundation's job in this project from preferred vocabulary to observer theory of the world computer. And it gives the cleanest one-line relation yet between the two works: perceived physics is what a bounded-but-persistent observer can extract from the Ruliad; option space is the Ruliad's OLU-accessible shadow — the configurations such observers can actually reach and maintain under bounded flux. Configuration Economics is the economics of that shadow.
Working consequences
- The strongest computation goes wherever the escape-from-the-distribution must happen. For frontier work that is generation, not assessment — the inverse of the usual spend rule for mechanical work, where cheap generation plus strong verification wins.
- The most amplifier-critical work of all is the pre-linguistic zone: holding a distinction stable before it has a name, against the strongest gravity any trained system has — collapsing the unnamed into its nearest named neighbour. Premature naming is the linguistic form of premature optimisation.
- The success metric of any run is expansion of the reachable space, not verdict. The machine's job is bridges made visible, not conclusions handed down.
Where this is honestly limited
The Ruliad/OLU frame is speculative physics; this note leans on its structure (boundedness × persistence as the observer characteristics that determine what is extractable), which is the part that is hard to vary — substitute different characteristics and the architecture stops following. If the frame falls, the complementarity argument still stands on its own legs: drift and chokepoint are observable failure modes, and the coupling-by-memory is an observable fix. A real test the note owes: does the architecture predict anything — for instance, that ventures with an unoccupied attractor seat fail by drift regardless of how much computation they command?
Not yet promoted into propositions.ts. Forming register, scaffold name. Sibling of the recursive-fitness note: that one gave the process its fitness function; this one gives it its physics.