Formative and summative evaluations

What are the differences between formative evaluations and summative evaluations? At which stage in product development do you believe that evaluations should use controlled settings instead of natural settings? Why?