Asymptotically most accurate double-side confidence intervals in normal-normal model

This article presents a form of the most precise class of confidence intervals for the normal-normal model in $d$-posterior approach. It is possible to state its asymptotical behavior with the increase of the sample size. A way of the most asymptotically accurate confidence family construction is also suggested in this article.