For each agent x_i, build trial = x_a + F·(x_b − x_c) (a,b,c distinct random peers). Crossover into x_i with probability CR. Greedy replace if trial fitness is better. Cluster spreads, then converges to optima.