Mapping the landscape of frustration-free models