The Org Chart vs. the Kinship
Every ontology declares how it's organized. Its concepts have an actual shape that's often different. You can measure the gap, and CIM's is wide.
Two Maps of the Same Place
Every big ontology ships with an official table of contents. CIM sorts its classes into schema areas: Device, Network, System, and so on. That’s the org chart, the way the standard says the domain is organized.
But the concepts also have an actual shape, the way they connect to each other in practice. Run community detection (the village-finder from the last post) and you get a second grouping, derived purely from how the concepts are wired together. Call it the kinship: not how the schema says things are filed, but how they actually relate.
These two maps are rarely the same. The interesting question is how far apart they sit, and where.
The org chart is how a domain says it’s organized. The kinship is how it actually hangs together. The gap between them is information.
Measuring the Gap
You can put a number on the agreement between two groupings. Normalized mutual information (NMI) runs from 0 (the two partitions tell you nothing about each other) to 1 (they’re identical). It answers a precise question: if I know which village a concept lives in, how much does that tell me about which official area it was filed under?
For CIM, the answer is NMI 0.299. The official areas and the lived communities share only about 30% of their information. The schema’s filing system and the concepts’ actual structure are mostly telling different stories.
Know which village a concept lives in, and you’ve learned less than a third of what area the schema filed it under.
Where It Breaks
The single number hides the good part, which is where the two maps disagree. CIM declares 14 areas; the graph forms 43 villages. Mean village purity is 0.668, meaning a typical community is a bit more than two-thirds drawn from a single declared area, with the rest imported from elsewhere.
The headline case is Device. As a declared area it’s one tidy box. As lived structure it shatters: concepts filed under Device scatter across 29 different communities. The schema treats “device” as one category. The actual relationships treat it as twenty-nine neighborhoods that happen to share a label.
“Device” is one area on the org chart and twenty-nine villages on the ground.
It’s Not Always This Wide
This isn’t a knock on CIM specifically, and a second ontology proves it. SPDX (software package metadata) organizes its concepts into profiles, and there the same measurement gives NMI 0.573. Its declared profiles line up with its lived structure almost twice as well as CIM’s areas do. Some taxonomies describe their own structure honestly; some describe an aspiration. The measurement tells them apart.
The comparison only works where a taxonomy is declared independently of structure, which is why Emic computes it for CIM’s areas and SPDX’s profiles, and not for ontologies that ship no separate filing system at all.
See it on Emic’s CIM explorer under the Official vs. lived tab.
The Takeaway
The Takeaway: A schema’s declared organization and its concepts’ actual structure are two different maps, and you can measure how far apart they sit. CIM’s areas and its communities share only about 30% of their information (NMI 0.299), and “Device” alone fractures across 29 villages. SPDX scores 0.573. The gap is exactly where the official story and the lived one part ways.