Fractal composition

Frames compose the same way at every size. Structurally, a paragraph’s frames make up a section, sections a document, documents a corpus; semantically, a claim with its evidence and sources makes up an argument, arguments an argumentation. A section is built like a document, an argument like a single supported claim — so a larger stretch of the graph tends to look like a smaller one.

The recursion is not only structural. A membrane — a boundary drawn around a co-active region — is itself a frame: collapse it to a point and it behaves as one. An agent is a membrane, the bounded bundle of its current context, so an agent is a frame one scale up. Frame, membrane, and agent are the same kind of object at different grain. Because they are, the same mechanics run at every level — activation, coupling, resonance, decay — and the same metrics can be read over a membrane as over a leaf frame.

We are not looking at a fractal in the strict sense: the levels are not identical copies, the recursion bottoms out at primitive frames, and a membrane is a dynamic boundary over what is co-activated at a step, not a fixed tier. The point is only that the same compositional method, and the same dynamics, recur — which is what lets the graph be read across scales, and retrieval set its depth by how far down into the nesting it reaches (see resonance).