The Sensory-Motor Coupling Wall

Three experiments returned null (5, 6, 9). All hit the same limitation: $\rho_{\text{sync}} \approx 0$. The FFT convolution kernel integrates over the full grid — patterns are inherently internally driven. V15 motor channels, V17 signaling, V18 boundary gating all failed to break it. The wall is about agency: action-observation causal loops, not signal routing.

It helps to read $\rho_{\text{sync}}$ not as a measure of internal autonomy but as loop-closure: the degree to which a system's own state, having passed through some state-preserving medium, returns to influence its later observations. The medium can be internal — a recurrent hidden state, a context window — or external — a depletable resource patch, a persistent signal trace in the world. What it cannot be is a substrate that mixes the state away before it can return. And the loop must be fenced by viability-relevance: only those returns that bear on the system's persistence count, so that idle echoes do not inflate the measure. This reframing dissolves the puzzle of why Lenia scores near zero. Lenia has no protected store: FFT convolution mixes every cell's contribution into a global field each step, and the patterns do not act on the world in any way that survives to be re-sensed. There is no medium to close the loop through, internal or external. Protocells, by contrast, deplete patches and lay traces — an external state-preserving medium — and so close the loop and read $\rho_{\text{sync}} \approx 0.21$. Recurrent agents and LLMs close it internally, through hidden state or context. The earlier “self-effect ratio” was the same quantity seen narrowly; loop-closure through any state-preserving medium is its general form.

Within Lenia the architectural space was exhausted. Breaking the wall required a substrate with a medium the loop could close through.