Every model is built from observations of the real world, then used to predict it. The moment the prediction diverges from what the river is actually doing, something important is being said. This pattern describes what that moment requires of a water manager. It sits between No Fat Modelling — which keeps the model honest enough to be compared to reality — and Science Waits Its Turn, which governs the sequencing of technical and experiential knowledge. It connects forward to The Single Future Pitfall, where over-reliance on the model produces a plan the real world eventually refuses to follow.

A water manager has three sources of understanding: the real world, their intuition, and their model. When these diverge, professional culture inverts the hierarchy that should govern them.

The model is defensible. It is documented, auditable, and can be shown to a minister or a donor. It was built by experts and validated against historical data. In the professional culture of modern water management, these qualities give it a kind of authority that the other two sources cannot match. Intuition is dismissed as unscientific. And the real world — the river behaving unexpectedly, the community reporting something is wrong, the gauge reading that doesn't fit — is explained away as an outlier, a data quality problem, or an edge case outside the model's domain.

This inversion is dangerous precisely because it feels rigorous.

The hierarchy should run the other way. The real world is always right. It is the only source of ground truth in this work. When the river diverges from the model, the model is wrong — not the river. The river does not have a calibration error. It is not outside its own domain. It is simply doing what it is doing, and if the model cannot account for that, the model has a problem that needs to be found and named.

Intuition occupies the middle position, and deserves more respect than professional culture gives it. It is not opposed to science. It is accumulated pattern recognition, built from long observation of real systems — the experienced engineer who looks at a flood hydrograph and says something is wrong before they can say what, the local farmer who knows this stretch of river floods differently than the maps suggest. Intuition is often picking up a signal the model cannot yet see, because it is drawing on a wider and longer record of observation than any dataset contains. When intuition and model disagree, that disagreement is worth investigating before the model is trusted.

The practical consequence is simple but runs against professional habit. When a gauge reading doesn't fit, the first question should not be what is wrong with the gauge — it should be what is the gauge telling us that the model isn't. When a community reports that water is behaving differently than the project documents predict, that report is data, not noise. When an experienced practitioner's unease cannot be immediately formalised, it should not be discarded — it should be held open until it can be examined.

None of this means abandoning models. It means knowing what models are: reductions of reality, built on assumptions, valid within a domain, and always subordinate to the world they are trying to describe. A model that cannot be wrong is not a scientific instrument. It is a belief system.

When the real world diverges from the model, treat the divergence as the finding, not as the error to be corrected. Investigate what the model is missing before defending what it predicts. Take intuition seriously as a signal worth examining rather than a bias to be overcome. Build into practice the habit of asking, when model and reality disagree: what is the river trying to tell us?

Connected patterns: No Fat Modelling — Science Waits Its Turn — The Single Future Pitfall — The Mirror That Reflects the Expert — Governance Cannot Reflect on Itself