The Max-SAT problem consists in finding an assignment maximizing the number of satisfied clauses. Complete methods for this problem include Branch and Bound (BnB) algorithms which use max-resolution, the inference rule for Max-SAT, to ensure that every computed Inconsistent Subset (IS) is counted only once in the lower bound estimation. However, learning max-resolution transformations can be detrimental to their performance so they are usually selectively learned if they respect certain patterns. In this paper, we focus on recently introduced patterns called Unit Clause Subsets (UCSs). We characterize the transformations of certain UCS patterns using the UP-resilience property. Finally, we explain how our result can help extend the current patterns.