What Not to Compute: Edge-Based World Modeling as a Stopping Condition

World Models and JEPA face an underspecified problem: when should prediction and updating stop? This is the stopping condition problem. We propose edge-based world modeling, where the interior is excluded from computation and only the boundary (edge) is monitored. An edge is preserved if d(B_now, B_sig) ≤ κ. Computation triggers only when this condition fails. This provides a constitutive stopping condition reducing complexity from O(|World|) to O(|Boundary|), complementing learned world models without relying on reward or error minimization.