

There is a Dutch saying — de schoorsteen moet roken, "the chimney must keep smoking." It is what you say about the unglamorous, essential work that keeps a household fed and a business running. Not the launch, not the keynote, not the clever idea on a whiteboard — the steady, faithful labour that means the lights are still on next winter.
This article is about the people who keep the chimney smoking for the information side of an organisation: the modelers and information architects who took FCO-IM not as a theory to admire, but as a practice to live by — for years, and sometimes for decades. They rarely get written about. So this is for them.
TL;DR — Metadata is descriptive and secondary — it points at data that already exists. Meaning is constitutive and primary — it precedes data and is what data should have been derived from. The semantic layer treats meaning as an annotation, when it is actually a foundation. You cannot retrofit a foundation. If you want AI to reason about your domain rather than hallucinate about it, you need a domain model constituted from meaning, not annotated toward it.
Some arguments need time to settle before their full weight becomes clear. Against the Stream made the case that most business-language modeling artifacts flow in the wrong direction — they enter the river midstream, assume vocabulary is already established, and call the resulting drift alignment. The argument was about directionality. But a question has been nagging since: why does the direction matter so fundamentally? What is it, precisely, that gets lost when you start from the wrong end?
The answer is deceptively simple, and it changes everything.
We ask AI to find meaning in our data — but the authentic story was gone before it ever looked
In What We Lost When Datum Became Data, I traced how a word that once meant "something given" shed its own etymology and became a synonym for "values in a system." Along the way, the who, the why, and the context quietly disappeared. I used the Groningen definition, the Employee-Person example, and the metadata stack to show where those losses happen.
That argument had its own reasons. It has sharper ones now.
Download 
Use Online