What you are looking at is a potential energy landscape. The curve is the terrain; the ball is a system—a learning model, a person, a mind—rolling through it under the influence of gravity and noise. The basins are attractors: stable configurations that pull the system in and hold it there. The ridges between them are barriers.

This is the geometry beneath The Basin. That essay argued that the gap between knowing and doing is not a distance to be crossed by effort but a topological feature of the solution space—a ridge between two valleys. The ball does not choose which valley to occupy. It falls. And what determines where it falls is not the ball’s intention but the shape of the landscape it moves through.

Try the presets. In Single Basin, the ball spirals inward and settles. This is convergent learning—one answer, one attractor, no ambiguity. In Double Well, two basins compete. At low temperature the ball is trapped; raise the temperature and watch for the moment it escapes. That moment—the sudden jump between basins—is a phase transition. It does not happen gradually. The ball is in one basin, and then it is in the other. The distribution is bimodal, not continuous. This is what Saphra found in language model training: capability acquisition that looks sudden is really a transition between coexisting solutions separated by a barrier. In Rugged, the landscape has many local minima. This is closer to reality—the loss landscapes described by Singular Learning Theory, where the geometry is complex, degenerate, and riddled with attractors of varying depth.

Now click to reshape the landscape. Add a basin. Deepen one. Raise a barrier. Watch how the ball’s behavior changes—not because the ball changed, but because the terrain did. This is the central insight: you do not fix the ball. You reshape the landscape. Eighteen copies of a rule are gradient pressure applied to a marble. Code that restructures the environment before the system wakes is a reshaped landscape. One is force. The other is topology. The landscape does not care about your intentions. It cares about its own geometry.