Optimizing the bridge only works because my method for solving the forces in each beam (least squares) happens to be differentiable.
Then the network is differentiable.