One major goal of Mathematical Knowledge Management is building extensive repositories, in which the mathematical knowledge has been verified. It appears, however, that maintaining such a repository is as hard as building it – especially for an open collection with a large number of contributors. In this paper we argue that even careful reviewing of contributions cannot cope with the task of keeping a mathematical repository efficient and clearly arranged in the long term. We discuss reasons for revisions of mathematical repositories accomplished by the “core implementors” and illustrate our experiences with revisions of MML, the Mizar Mathematical Library.