The Boundary Residual
A candidate boundary is measured by how much conditional mutual information leaks between the deep inside and the deep outside once the interface is known.
What decision changes?
When auditing a candidate boundary, measure the leakage directly rather than accepting a boundary defined by convenience — a low residual means the interface actually screens inside from outside.
The boundary condition divides variables around a candidate system into internal, sensory, active, and external roles, then asks whether the deep inside and deep outside become approximately conditionally independent once the sensory-active interface is known. A sensory variable is anything through which the outside predicts the future inside; an active variable is anything through which the inside predicts the future outside — no eyes, muscles, or motors required.
A small residual means the system has a boundary, at least approximately: it depends on the world, senses it, and acts on it, but does not leak into or out of it except through the interface. This is not the claim that the system is isolated. It is the claim that its dependence on the world is organized.
This condition, not a fixed list of examples, is what lets the same measurement apply to a chatbot, a firm, or a biological organism.