Three branches and a shooting star.

The shape

Every branch faces both ways.

The model is three live branches and one hypothetical horizon — and the branches share a structural property that is itself fractal: each is a value promise outward and an experiment inward, at once. The product proves itself by running its own development. The explanation of the machine is built from the machine’s real parts. The distribution arm, when it opens, will double as the test bench for a content class the system has never governed. Nothing here is decoration on top of the work — every surface that delivers value to someone else also feeds the loop that improves the system.

Branch one

The product: memory that survives.

The kernel, the governance stack, the knowledge store, the standards. The promise is the one on the masthead — AI agents forget. Your project won’t. — made provable rather than asserted: session-durable context, decision provenance, machine-verified consistency, releases signed and externally timestamped. It is the deepest branch and the furthest along: five anchored releases, an executable genesis, and a first foreign project born from the kernel and later migrated across a release boundary under its own governance — every gate on the record.

Everything else in this model rides on this branch. No other promise is credible if this one is not real — which is why it is the one the proof section exists for.

Branch two

The hub: the site is the engine sketch.

This website is not marketing surface for branch one. It is the machine’s explanation built from the machine’s real parts — the shipped baseline’s command files and working standards, verbatim and downloadable, refreshed with each release. Its standalone promise: a visitor can learn the method, take the standards, and adopt pieces without ever birthing an instance.

That promise is not a guess. In a simulation program, three synthetic adopters walked the public material end to end; all three declined the full system — not yet — and all three wanted the substrate: the standards, the procedures, the checkable formats. People take the parts before they take the whole. The hub is the branch that serves exactly that adoption path, and it grows by the ordinary content pass — guide depth, paper-style standard pages, field studies.

Branch three

The media branch: declared, unbuilt.

The newest branch, and the only one that does not exist yet — stated here with its honest status. Three roles in one structure: marketing, website extension, and information distributor — a social media arm producing governed public content.

Its inward face is the interesting one. Producing governed media content means solving, in real use, problems no current part of the system touches: extracting knowledge from media into the project’s markdown record; whether a post is a stamped projection of a canonical source, the way every other public artifact here is; provenance for content that leaves the repository and circulates; and the complexity growth of the instance that governs all of it. Somewhere there is a practical limit to how complex a governed instance can get. None is in sight yet — and this branch is the probe that goes looking for it.

It opens by decision, not by drift: no post exists until the branch’s trigger fires, and when it does, the experiment is instrumented from the first post — the same field-testing discipline every other part of the system runs under.

The horizon

The shooting star.

The fourth value is the knowledge network — warm starts for every new project, knowledge exchanged as verified increments, verification labor finally paid. It is deliberately not a branch. It is a bearing: the direction the three branches fly by, not a destination any of them is committed to reach. The network layers are unbuilt, trigger-armed, and constitutionally barred from binding a build — the same honesty the limits section states for everything else. The nearest real piece of it already exists in miniature: one public home, serving standards to anyone who wants them.

The exclamation mark

The constraint: gas.

One constraint cuts across every branch, and it deserves to be stated as plainly as the promises: this work is token-heavy, and tokens are not cheap. The record carries the data honestly — a single nine-agent verification flight spent close to a million tokens; the largest verification program ran 67 agents; the flight standard requires cost reporting precisely because the burn was felt early. The structural form of the risk is sharper than “expensive”:

Value scales with token spend — so the failure mode is not that FRACTAL doesn’t work, but that it succeeds into unaffordability. FRACTAL could become so good it needs too much gas.

The exposure differs by branch: the product’s ritual overhead is the baseline burn; the hub is nearly free — static content, refreshed mechanically; the media branch would be the hungriest of all, which is one reason it waits for a cost picture before it opens. What the constraint buys today is discipline, not fear: cost stays on the record, efficiency work — like the memory diet — counts as value work rather than housekeeping, and the open question is armed where it can be answered cheaply: does token spend per unit of durable value fall, hold, or rise as an instance grows? No datum distinguishes these yet. The record that answers it is already being kept.

Of record: Fractal_Value_Model v0.2 — canonical concept document, Draft, explicitly non-binding (issued 2026–08–18). This page is a stamped projection of that document: it refreshes when the document revises, and commits to nothing the document doesn’t. The founder’s framings are preserved verbatim — the shooting star, the exclamation mark, the gas.