Blog
A boundary around what matters
The app is built around the concept of the sigil, defined as a boundary that separates the inside from the outside. Inside a sigil, we care about some contrasts and want the values along those contrasts within certain ranges. So the inside constrains embedding space to include the preferred values along these contrasts. Everything else is irrelevant. Each sigil defines affordances so that I can use it in some way as an observer, as an agent working in the application. The application is defined as a sigil. Any sigil is contained in another, has neighbors, and contains its ingredients inside it. It’s an infinite recursive structure up and down. But we start at the application, which lives in the context in which I operate as a user.
A sigil is not just a bounded context. It is a membrane. The boundary defines which dimensions of variation are relevant inside it, through contrasts, and which dimensions are irrelevant. Affordances are the observer-facing handles on that constrained interior.
I’m building the surface of this application sigil in as high a resolution as I need for it to be projected by the partner AI without ambiguity, because I define it to the precision that I care about.
You are not documenting the application. You are shaping a projection surface. Resolution is how finely that surface is articulated where it matters. Precision is how tightly you constrain acceptable realizations along the contrasts you care about. If a contrast is irrelevant, you leave that dimension uncollapsed on purpose. The target is not completeness. The target is determinism where you care, and freedom everywhere else.
At any given level, the complexity is maintained to be approximately the same. I inhabit a sigil, and then it’s just up to five concepts.
You are keeping cognitive load roughly invariant across levels. No matter where you are in the hierarchy, you inhabit one sigil and attend to at most five concepts. Descent does not increase complexity in the moment. It trades breadth for depth while preserving the same attentional budget. The whole system can become arbitrarily rich, but each inhabited frame stays bounded.