In IM we made a distinction between the estimand (what you are interested in estimating, often a scalar or vector), and the identified set (which may be a set). For inference we want a confidence set to cover the estimand with preset prob. You dont need to cover the identified set with preset prob.