Chapter 13

How the Concept System Is Managed

Chapter 12 called a Concept a handle for pulling meaning out of a document again and again. How far should those handles go?

A graph needs Node kinds. But not every concept we use has to become one. “Search quality”, “cost efficiency”, “Korean-language performance” and “reproducibility” are important concepts, and that is no reason for each to become a new Node kind.

A Node kind is a role the system has to handle differently at the structural level. A concept is a range of meaning people work with inside the system. So one kind, Concept, holds a great many concepts. That is why a new technology or a new research topic can be expressed inside the existing Concept structure without touching the schema. Miss this distinction and the schema never stops growing.

1. The trouble with an open concept space

The same word can carry different meanings, and different names can point at the same thing. “Embedding”, “vector representation” and “semantic vector” are not always one concept. In one context they are; in another they mark different technical scopes. Comparing strings cannot settle whether two concepts are the same.

So the concept system has to sit on a different layer from the graph's schema. The schema fixes the structure the system must understand; the concept system widens the space of meaning the system handles.

The two influence each other. When a kind of information keeps appearing that the existing kinds cannot express, that is a signal to look at the schema again. When the existing schema is enough, there is no reason to build a new structure for a new concept.

2. Concept shows where the boundary is

Concept shows this boundary well. It is a kind the system recognizes structurally, and its title is bound to no fixed list. The structure of the graph stays put while the meaning inside it keeps widening.

Making concepts needs a boundary too. Turn every noun in every document into a Concept and the graph grows fast, but meaningful memory and passing phrasing become indistinguishable. Restrict it too tightly and nothing new can be expressed.

So extracting concepts is not collecting strings. It is judging whether an expression deserves to be referred to on its own in later work. The rules that automate that judgment are in Chapter 15.

3. The type vocabulary is authored

A Concept's title is one open space. There is another above it: the type names themselves.

The five kinds of Chapter 10 belong to the system. But a person structuring their own subject wants names the system has never heard. Vinc does not make them choose the nearest of the five. The node lands on the open carrier, Entity, under exactly the type name they chose.

The collection of those names is the vocabulary of a Theme. A theme is the frame of interpretation a person builds over a subject the world hands them. It holds guiding questions and a type vocabulary, and it is the unit of ownership, sharing and scope. The vocabulary lives outside the graph, in a file beside the database, because a vocabulary is the frame of an interpretation and not its content. Editing the frame never touches the data.

A new type name is not accepted as fact. It is folded in as proposed and reported back as such. A proposed type exists and takes part in every write. What it lacks is confirmation, and confirmation is a person's act: ratify the name, reject it, or rename it. The system proposes a starting vocabulary to spare the person a blank page, but it never quietly confirms its own proposal. Ratification is not a tool an AI assistant can call; it exists only on a screen a person opens, because it is a person's act. Only ratified vocabulary travels in a pack.

None of this loosens the rule that matters most. Two Nodes are the same only when their identifiers are the same. A shared type name, like a similar title, is never grounds for merging.

So the concept system is open in two places. A Concept's title can be anything, and a type name can be anything a person is willing to stand behind. What the system recognizes structurally stays five kinds wide.

4. Storing a name is not understanding a meaning

A string becoming a Node's title does not mean the system understands what that concept means. The real meaning has to be read together with the document's context, the connected relations, the evidence and the time.

The concept system is therefore wider than the graph. The graph provides the structure for storing and connecting concepts; it does not need to fix, as schema, the whole of a person's sense of which concepts exist. That distinction is what lets Vinc keep its structure stable while its meaning keeps widening.

In a space this open, people fill the gaps from context. A system cannot. For a system to write and read memory, there has to be an explicit contract saying what is allowed and what is not. That contract is the next chapter.

Continue with Chapter 14, The Contract for Writing and Reading Memory.