It is accepted that the alignment of a return-address-stack can be preserved by checkpointing on branches and recovering on branch mispredictions the top-of-stack pointer. However, virtually all published literature, is vague or proposes sub-optimal algorithms on how to repair the alignment of a return-address-stack on call and return mispredictions. Possible manifestations of this ambiguity are lower performing implementations of return-address-stack mechanisms in processors and simulators. This paper suggests that on call and return mispredictions the recovered top-of-stack pointer should point to the