When should we conduct large-scale evaluations?