Four years on the model. Every coefficient measured. The simulations produced rivers — plausible, statistically sound, wrong.
Not wrong in the physics. Wrong in the placement. Her rivers chose different valleys than the real ones.
Refine the initial conditions, her advisor said. Increase resolution, the committee said. She did both. The systematic offset persisted. Her simulated rivers carved their way to the wrong low points.
The insight came standing in a real valley, not looking at water but at rock. Fault lines. Bedding planes. Contact zones between formations. The geology had already written the drainage network before the first raindrop.
Her model computed the carving and called it creation. But the river didn't create the valley. The valley permitted the river.
She rebuilt from the geology up. Started with the fractures, the soft layers, the places where rock was already willing to be absent. Let water find what was already there.
It matched. Not because the hydrology improved. Because she stopped assuming the river was the author.