The NA48/2 collaboration has shown clear experimental evidence for a cusp in the data for K pi pi0pi0. We address this issue using a two-loop dispersive construction of pipi pipi and K pipipi amplitudes in the presence of isospin breaking.