1. What an ontology really is

This is chapter 1. Nothing is assumed. If you have never heard the word before, you are in the right place.

In a meeting, three people say three things.

The first says, "We built an ontology."

The second says, "That is just a taxonomy."

The third says, "An ontology is a formal, explicit specification of a shared conceptualization."

The room goes quiet. Nobody asks what the third sentence means. That sentence is the standard definition, it is quoted in almost every paper on the subject, and most people who repeat it could not explain it.

By the end of this chapter you will be able to.

Start with a warehouse, not a definition

Picture a warehouse.

Inside it there are freezers. Inside the freezers there are boxes of vaccines. Each box belongs to a shipment. Each shipment is going to one store.

That is the real world. You could walk in and touch it.

Now picture what is in your head when you think about that warehouse. You know a freezer is a kind of thing. You know a freezer holds boxes. You know a shipment goes to exactly one store.

That mental picture has a name.

Notice that your picture already left things out. It has no floor, no lighting, no forklifts, no staff. That is not a mistake. A picture that kept everything would be the warehouse itself, and would be no use.

Problem 1: it is stuck in your head

Your colleague has a mental picture too. Hers is slightly different. In her picture, a shipment can go to more than one store, because she once handled a split delivery.

Neither of you knows the other's picture is different. You both say the word "shipment" and you both nod.

Six months later a report is wrong and nobody can work out why.

The fix is obvious. Write the picture down.

Problem 2: written down is not the same as clear

You write: "A shipment goes to a store."

Does that mean exactly one store? At least one? Can a shipment exist before a store is chosen?

You wrote it down and it is still ambiguous. A human reading it fills the gaps with their own assumptions, which puts you back where you started.

And a machine cannot read it at all.

This is the step most projects skip. A slide deck describing your entities is a specification. It is not formal and it is not explicit. It cannot be checked and it cannot be run.

Problem 3: yours is not everyone's

Now you have a formal, explicit specification. It is precise. A machine can read it.

It is still only yours.

Sales has one. Billing has another. Both are precise. They disagree about what a customer is, and now they disagree precisely.

The definition, assembled

Now read it again:

An ontology is a formal, explicit specification of a shared conceptualization.

• conceptualization: the picture of what exists and how it connects
• specification: written down, outside somebody's head
• explicit: nothing left to guess
• formal: in a language a machine can check
• shared: agreed by the people who will use it

Four words, four repairs, four researchers, roughly five years.

Why this matters now, and not in 1998

For twenty years this was mostly an academic concern. The reason it suddenly matters is AI agents.

A person reading your ambiguous document fills the gaps using judgment. They have worked here for six years. They know a split delivery is rare and they know who to ask.

An AI agent has none of that. It fills gaps with whatever sounds likely. It does this confidently and it does not tell you it guessed.

So the gaps you could previously afford are now the places your agent invents things. Making the picture formal, explicit, and shared is how you stop that.

Where the written-down picture lives

In Microsoft Fabric, the place you write the picture down is called the ontology item. It is a thing you create in a workspace, like a report or a lakehouse.

Inside it you declare three kinds of part:

1. Entity types: the kinds of things. Freezer, Shipment, Store.
2. Properties: the facts each kind carries. A freezer has a maximum temperature.
3. Relationship types: how the kinds connect. A shipment goes to a store.

Then you do one more step that has no equivalent in the philosophy. You bind each part to real data: a table in OneLake, which is Fabric's single storage layer for the whole tenant. Binding is what connects the picture to actual rows.

So Fabric splits the job in two. The ontology item holds the shared picture. The bindings hold the connection to reality.

The one trap

Do not use the word "ontology" for a diagram. A boxes-and-arrows picture in a slide is a conceptualization that somebody drew. It is not formal, a machine cannot check it, and nothing stops two boxes meaning different things to two readers.

That is not a small distinction. It is the whole difference between a document people ignore and a model an agent can be held to.

Next, chapter 2. You now know what an ontology is. The trouble is that a lot of things get called one without qualifying. Chapter 2 gives you a ladder, from a simple list up to a full ontology, so you can place any model somebody shows you and say exactly which rung it sits on.