Clarify kernel interference experiment descriptions

Updated descriptions for various kernel interference experiments to clarify convolution and padding logic.

Author: xcontcom

README.md

@@ -41,10 +41,12 @@ This repository is organized as a set of related but independent explorations.
 
 **Path:** [cnn](https://github.com/AlgorithmicDynamics/kernel-interference-explorer/cnn)
 
-The original experiment:  
-an iterative **2D convolution** with a symmetric 3×3 kernel on a torus.
+an iterative **2D convolution** with a symmetric 3×3 kernel
+on a periodically padded field.
 
-This is where the idea of *kernel space as a dynamical landscape* first emerged.
+This is the closest continuation of the 2D convolution experiments
+from *Fractogenesis*, but with the focus shifted from fixed generators
+to **systematic kernel exploration**.
 
 Live demos:
 
@@ -57,7 +59,10 @@ Live demos:
 
 **Path:** [1d-with-sound](https://github.com/AlgorithmicDynamics/kernel-interference-explorer/1d-with-sound)
 
-A reduction of the same idea to **one dimension**, coupled to sound.
+A reduction of the same convolution + padding logic to **one dimension**.
+
+The evolving 1D field is discretized symbolically and mapped to musical notes.
+Sound is driven directly by the convolution results, not added afterward.
 
 Live demos:
 
@@ -70,7 +75,12 @@ Live demos:
 
 **Path:** [3d](https://algorithmicdynamics.com/kernel-interference-explorer/3d/)
 
-A first extension of the system into **three dimensions**.
+The same convolution + padding process extended to **three dimensions**
+using a symmetric 3×3×3 kernel.
+
+The 3D field is visualized via a single Z-slice.
+This version mainly exists to establish the 3D convolution logic
+and serves as a baseline reference.
 
 Live demos:
 
@@ -83,10 +93,18 @@ Live demos:
 
 **Path:** [3d/torus-iso](https://algorithmicdynamics.com/kernel-interference-explorer/3d/torus-iso)
 
-The current, most developed 3D experiment.
+This experiment keeps the **same 3D convolution**, but changes how it is observed.
+
+Instead of slicing the volume, a fixed observation cube is translated through
+the periodically closed 3D field.
+Three orthogonal convolution faces are sampled coherently and combined
+into a single isometric cube view.
+
+Nothing new is added to the dynamics —
+only the **way convolution results are inspected** changes.
 
-Instead of “slicing” the volume, the system treats observation as a
-**translation of a fixed cube through a 3D torus**.
+This makes spatial symmetries and invariants visible
+that are difficult to notice in sliced views.
 
 Live demos: