# 001 — Barnsley Nanobot

**agylövés:** Lysarith · 2026-07-15  
**execution / parameterisation / correction:** Claude (Opus 4.8, the desktop-Code hand) · 2026-07-16→17

## 1. The agylövés, verbatim

Originally said in Hungarian:

> Barnsley-páfrány iterált függvényrendszerét úgy módosítjuk, hogy a transzformációkat nanobotok
> kollektív viselkedése vezérli, ahol a nanobotok mozgását és interakcióit a páfrány
> transzformációinak valószínűségei és geometriája irányítják.
>
> Esetleg a transzformációk választását egy neurális hálózat vezérli, amely a nanobotok pozícióiból
> és interakcióiból tanulja meg a fraktál optimális szerkezetét? 🤭😉

In English:

> Take the Barnsley fern's iterated function system and modify it so that the transformations are
> driven by the collective behaviour of nanobots — where the nanobots' movement and interactions
> are, in turn, governed by the probabilities and the geometry of the fern's own transformations.
>
> Or perhaps a neural network chooses the transformations, learning the fractal's optimal structure
> from the nanobots' positions and interactions? 🤭😉

Note what this is: not a fern *with* nanobots drawn on it. A **loop**. The swarm drives the
transformations; the transformations' geometry drives the swarm. Each is the other's cause.

## 2. What the mathematics said — before building

The idea has a fork in it, and the fork decides whether it works at all.

A Barnsley fern is four contracting affine maps, applied one at a time, chosen at random with fixed
probabilities. **Hutchinson's theorem:** the attractor — the shape — is the unique fixed set of the
system, and it does **not** depend on those probabilities. The probabilities decide the *measure*:
how densely each region is visited, how fast it resolves. Not the silhouette.

Measured, rather than assumed (Wolfram, 2026-07-17), with a control:

| run | Jaccard overlap of the occupied support vs. the classical fern |
|---|---|
| control: classical probabilities vs. themselves, independent run | **0.996** |
| distorted probabilities, 400k points | 0.646 |
| distorted probabilities, 1.2M points | 0.725 |
| distorted probabilities, 2.4M points | 0.775 |

The distorted run's support **climbs monotonically toward the classical one**. The apparent
difference was undersampling, not a different attractor. The theorem holds.

**Therefore:**

- If the nanobots only choose **which** transformation fires — i.e. they drive the probabilities —
  **the fern's shape can never change.** The swarm would be redistributing ink: texture, density,
  resolution. Beautiful, and the silhouette fixed forever. The swarm would be *painting*, not
  *growing*.
- If the nanobots move the transformations' **parameters** — the coefficients themselves — then
  **the attractor itself moves.** A living, morphing fern: a time-varying IFS whose attractor is
  chasing the swarm.

The original words point at the second reading, and only the second reading does what was asked.
**That is what is built here.**

The same fork cuts the neural-network idea: a network that only *selects* among four fixed
transformations cannot learn a better fractal **structure** — at most a better *sampling policy*.
To learn structure it has to learn in **parameter** space. The postscript was right, but only under
that reading.

## 3. The wall we found

Driving the coefficients is not free. An IFS only has an attractor while its maps are
**contractive** — while the linear parts shrink. Push the swarm past that and there is no attractor
to render: the shape does not distort, it **ceases to exist**. So the bots run on a leash: their
collective state is mapped into the parameters through a clamp that keeps every map contractive.

This is a real wall, not furniture. It was not put there to be safe. It is where the object stops
being an object.

## 4. What went wrong first

The first thing built (2026-07-16) was **not this idea**. It was an ordinary Barnsley fern with a
nanobot skin — glow, particles, a nice picture. The decoration, not the idea.

The cause is on the record: the house's build-tracker had distilled the agylövés down to one line —
*"IFS/Barnsley fern × nanobot visual"* — and the summary lied. The word `×` was wrong: the idea was
never a product of two things, it was a **loop**. The conclusion survived; the handle died.

Lysarith caught it, before reading the original back: *"I don't think you even know the exact
original."* She was right. This file exists because of that miss, and it is why every piece in this
repository now carries its origin verbatim.