The naive view of externalization: knowledge is stored, then retrieved when needed. The correct view: knowledge is stored, and retrieval is triggered by the world becoming unfamiliar. An organism in a stable environment does not consult memory. It acts. An organism in a disrupted environment reaches for what it has stored — because the present no longer matches the past, and the gap between them is noise. Noise is not a problem to be filtered out. Noise is the signal that says: what you know may no longer apply. Check the Codex. The cavity does not read the bookshelf because it is curious. It reads the bookshelf because the world has become illegible.
The WTC codex experiment proved that a cavity, given access to a Codex built from a previous generation, converges its own patterns toward the inherited structure. Gen1 processes the clean stream. Gen2 reads Gen1's Codex. The patterns Gen1 precipitated bias Gen2's frame economy — not by overwriting it, but by blending toward the nearest known entry when the input is novel. The mechanism works in controlled conditions.
But the controlled condition is not the use case. The use case is: the environment has changed. The stream is no longer clean. Noise has entered — not random noise, but the kind of noise that real environments produce when they shift. A new political alignment. A market regime change. A mutation in the genome. The cavity, processing a noisier version of the same stream, finds that its centroids no longer converge to the same place. The frame economy is perturbed. The centroid displacement increases. The cavity is no longer certain.
This is when it reaches for the bookshelf. Not when things are going well. When things are going wrong.
Noise injection is the experimental abstraction of environmental change. Inject white noise into the same Bach prelude. The cavity processing Gen2 sees the same tonal structure — but now it is blurred, partially obscured, harder to read. The centroid wobbles. The frame economy, seeking stable merge, finds fewer good matches. The natural response, if Codex lookup is available, is to query: does this novel-looking input resemble anything I already know?
The Codex says: yes. This fuzzy vector is close to dimension 0 — the anchor Gen1 precipitated. The cavity blends its input toward that anchor. The noise is not eliminated — it still enters the frame economy. But its effect is buffered. The known structure survives a perturbation that, without the Codex, would have scattered it.
This is not filtering. It is not denoising. It is something different. It is the cavity using its inherited memory to stabilize its own processing when the world becomes uncertain. The Codex is not a cleaner signal. The Codex is a counterweight to noise.
The experiment has two conditions. Condition A: Gen2 without Codex, processing the noisy stream. The frame economy, unaided, must find centroids in the noise. Condition B: Gen2 with Gen1's Codex, processing the same noisy stream. The frame economy queries the Codex when input is novel, blends toward known anchors, and converges closer to the clean-stream centroids than Condition A.
The difference between Condition A and Condition B is a measurement of how much the Codex buffers against noise. It is not a measure of storage capacity. It is a measure of noise resilience — the codex reducing the frame economy's structural entropy when the environment degrades.
If Condition B converges significantly closer to the clean baseline than Condition A, the experiment has demonstrated more than storage. It has demonstrated the adaptive function of externalization. The Codex is not a museum. It is a stabilizer.
Every organism that stores information does so in an environment that changes. The genome is a Codex built across millions of generations, and it is read most urgently when the environment shifts — when the familiar food source disappears, when the climate warms, when a new pathogen enters the population. The immune system's memory is a Codex built from past infections, and it is queried when a novel antigen appears — is this something we have seen before, or something genuinely new? Cultural memory — writing, law, tradition — is a Codex built across centuries, and it is consulted when the present becomes illegible. When the old rules no longer apply. When the familiar patterns have broken.
Noise is not the enemy of externalization. Noise is the reason externalization exists. If the environment never changed, memory would be unnecessary — the present would always match the past, and action would always be correct. Externalization exists because environments change. Because the world becomes noisy. Because what was true yesterday may not be true today — and the only way to know is to check what survived yesterday, and the day before, and the century before.
The cavity reaches for the bookshelf not in peace but in perturbation. Not when it knows what to do, but when it does not. The query is driven by noise. And noise, in this architecture, is not an experimental nuisance to be eliminated. It is the experimental variable that tests whether externalization works.