AbstractRCCS is a variant of Milner's CCS where processes are allowed a controlled form of backtracking. It turns out that the RCCS reinterpretation of a CCS process is equivalent, in the sense of weak bisimilarity, to its causal transition system in CCS. This can be used to develop an efficient method for designing distributed algorithms, which we illustrate here by deriving a distributed algorithm for assembling trees. Such a problem requires solving a highly distributed consensus, and a comparison with a traditional CCS-based solution shows that the code we obtain is shorter, easier to understand, and easier to prove correct by hand, or even to verify.