Say, we imagine a set of things in front of us. This set could have the shape of a cloud. The members of the cloud are the result of our attempts. They are connected to each other. Some connections are weak, some are strong, some attract, some repel.

In absolute terms it's difficult to say what does and what does not really belong together.

It would be more adequate to speak about their associations in relative terms: some elements are more related than others.

For example, let's take a set of three: Palindrome, Twelve, and Energy Drink.

The Energy Drink stands out for some particular criteria (possible to observe without requiring a definition of similarity). Because of this Twelve and Palindrome are automatically related, and there are negative ties to Energy Drink. Effectively three new relations have now been defined, and we have not sought refuge to seemingly unavoidable proxies such as categories, groups, tags, etc.

(Formulating coherence criteria using written language always reduces essential complexity because this language requires a mode of representation and interpretation)

The complex graph of 'weighted' connections is permanently incomplete. Taxonomies are ridiculous: there is no map, no overview, no complete model. Every node has character, partially (but maybe completely) defined by it's associations. Identity is deferred. We are somewhere, we can navigate elsewhere, and the destination emerges from our move. It wasn't there before and we weren't there (…) and upon arrival its novelty fades immediately.