Together, these results suggest the brain's network geometry is a computational substrate, not just wiring: low-dimensional connectivity compresses representations for generalization, while high-dimensional connectivity expands them for task-specific implementation 18/n