§ 1 — DefinitionThe useful simplification

A model is an intentional simplification of reality. Conceptual, logical, or mathematical, it is a representation that captures the essential structure of a system while deliberately discarding what is irrelevant to the question at hand. A map is a model. So is an equation, a flowchart, a mental framework, an economic theory, a clinical heuristic.

Models are how the mind makes the world tractable. Reality is infinitely detailed; cognition is bounded. To think at all, we must reduce — selecting which variables matter, which forces dominate, which couplings can be ignored. A model is the formal record of those choices.

The purpose of a model is not to reproduce reality but to illuminate it: to let us understand a system, predict its behaviour, or optimise its outcomes. By those criteria — not by its completeness or beauty — every model is judged.

§ 2 — RelationshipsThe bridge between fact and meaning

A model sits in the middle of the knowledge triptych for a reason. Without it, the other two cannot speak to each other. Data on its own is noise — an unstructured cascade of observations. Story on its own is unmoored — a compelling shape with nothing inside. The model is what converts the first into the second.

A model interprets data: it tells us which observations are signal and which are noise, which variables are causes and which are mere correlations, what the missing data points are likely to be. It is the lens through which evidence becomes information.

A model scaffolds story: it supplies the logical structure on which a narrative can be hung. When a doctor explains a diagnosis, a strategist defends a forecast, a teacher introduces a concept — each is voicing a model. The story is the model wearing language.

Models, then, are bridges. They connect the raw and the rendered, the empirical and the meaningful. Lose them, and the two banks of the river cease to communicate.

§ 3 — CraftHow to build a model that earns its keep

Modelling is a discipline of restraint. The temptation is always to add more — more variables, more terms, more fidelity. The art is to resist that temptation in service of clarity.

• Begin with a question, not a system

  Define the scope and objective before any structure is sketched. What decision will this model inform? What behaviour will it predict? A model built without a purpose tends to grow until it explains nothing useful about anything in particular.

• Identify the variables — and how they relate

  List the inputs, the outputs, and the couplings between them. Distinguish causes from correlates, stocks from flows, signals from noise. A model is, in the end, a statement about which things move together and which move alone.

• Prefer parsimony — the principle of "as simple as possible"

  Einstein's reformulation of Occam's razor: as simple as possible, but no simpler. Each added term must justify its place by improving prediction or insight. Complexity that does neither is a liability dressed as rigor.

• Beware over-fitting

  A model that explains every past observation perfectly almost always predicts the future poorly. It has memorised the noise. Build for the regularities, not the residuals. A good model leaves something unexplained — and is honest about it.

• Validate against the world, and update when it disagrees

  A model is a conjecture about reality, not a substitute for it. Hold it to the standard of out-of-sample data, of new conditions, of inconvenient evidence. When the world says the model is wrong, the world is right.