Main idea

A set of notes taken on Pierre Bourdieu's sociology already yields a dense graph. Add the notes taken on Niklas Luhmann in order to compare the two authors, and the density becomes unreadable: every concept touches several others, and the number of edges grows faster than the number of notes.

The problem is not the density. It is that the edges are not of the same nature. A logical implication, an antagonism between two positions, a vague kinship, an indirect relationship passing through a third concept: in the graph, these are four identical lines. Grouping by broad concepts to simplify changes nothing, because the simplification bears on the number of nodes, not on the qualification of the links.

Such a network is therefore proof of complexity rather than a tool for handling it. It shows that everything hangs together; it does not say how, and the "how" is what you came for.

Why it matters

This separates two promises we conflate when talking about a knowledge graph: faithfully representing a domain, and being able to work in it. The first is kept, the second is not — and the first makes the second harder, since it adds links without adding discernment.

It also gives a criterion to demand of any tool that connects automatically: that it name the relationship, failing which it produces connection and not knowledge.

Layer added by "What Is an Atomic Note?" (2026-06-16). The same observation, taken from the side of the person writing their notes, turns into a working rule: the goal is not to have many links, the goal is to have links that explain a relationship — a cause, a consequence, an objection, an example, a tension, a neighboring idea. This layer adds that the qualification is not an optional typing bolted on afterwards, but what distinguishes a connected note from an isolated one: a link that names nothing leaves the note exactly as alone as before.

Nuances and limits

A graph whose every edge carries a type — depends on, contradicts, illustrates — escapes the criticism, but that typing has a price: it must be laid down by hand, or by a judgment that will have to be verified.

And the overall view has a value of its own when what you are after is precisely to take the measure of a subject's extent, before working on it.

Open questions

• Can a typing of relationships laid down by a machine be audited other than by rereading every edge?