Every model is a reduction. The question is not whether to simplify but how — what to keep, what to discard, and whether those choices are made deliberately or by default. This pattern describes the discipline of finding the simplest model that generates a useful result, rather than the most complete model that data and computing power allow. It connects forward to The Single Future Pitfall and backward to Science Waits Its Turn.

The most sophisticated model that can be built is rarely the most useful one: complexity hides assumptions rather than testing them, and produces outputs that are difficult to interrogate and impossible to own.

There is a genuine pull toward comprehensiveness. Leaving things out feels like dishonesty. And elaborate models signal rigour — to funders, to decision-makers, to the institution commissioning the work. But a model is a story made precise, and a story that tries to say everything says nothing clearly. The elaborate model tends to produce confident-looking outputs — a projected flood level, a design standard, a number that can be written into a plan. That confidence is frequently false. The model has absorbed uncertainty rather than surfaced it. Assumptions are buried inside parameter choices and calibration decisions, invisible to everyone receiving the result.

The discipline starts from a different question: not what is the most complete model I can build? but what is the simplest set of assumptions needed to generate an interesting and useful result? A lean model makes its assumptions legible and invites challenge. It can be handed to a decision-maker or community representative and interrogated: here is what this rests on — do you believe it? That interrogation is not a weakness. It is the point.

One distinction is worth preserving. A model can claim to show what must happen, or it can show what can happen. The first claim is almost always too strong. A model that demonstrates the plausibility of a mechanism leaves room for judgment and local knowledge. It opens a conversation. A model that presents its output as prediction tends to close one.

Therefore: before adding complexity, ask what the simplest version capable of a useful result looks like, and build that first. Add only what demonstrably changes the answer. State assumptions explicitly so they can be examined. Present outputs as illustrating what can happen, not predicting what will. Treat simplicity not as a limitation but as the condition under which a model can be owned, tested, and trusted.

Connected patterns: Science Waits Its Turn — The Single Future Pitfall — The Mirror That Reflects the Expert — Governance Cannot Reflect on Itself