The membrane

A membrane is a cell's boundary: it lets some things through and not others,
and from the outside you cannot see the rule. That's the precise word for a
designed selective boundary in a knowledge federation — gated access, a
subgraph exposed to one partner and not another, a hub that answers "not
configured" identically for a base that doesn't exist and a base you're not
allowed to see. The rule is real; the owner knows it; the visitor only sees
pass or bounce.

It matters because a membrane is not a wall. A wall says no. A membrane says
no, selectively, for reasons you can't inspect — and that asymmetry of
knowledge is what makes visitors invent theories. In our experiments
(en/thoughts/let-the-model-wander) every model that hit a boundary composed
an explanation for it, and the more specific the refusal looked, the more
confident the theory. One even declared an experimenter's typo "architectural."

So the term earns its keep only with its opposite. A membrane is a
deliberate boundary with hidden rules: legitimate, often necessary — ownership
and trust demand it (en/thoughts/federation-supply-chain). An invisible
wall is an accidental one: a bug, a typo, a manual promising what the server
rejects. The design problem is that from inside a walk they are
indistinguishable — both fail with the same face. Which yields the one rule we
keep re-learning: if you build membranes on purpose, say so in the manual;
every boundary you don't label will be read as either malice or philosophy.