Our experience with data compression so far has been of the following type: We are given the statistical description of an information source, we then try to design a system which will represent the data produced by this source efficiently. In this note we depart from this model, and consider a method which will represent a sequence efficiently without knowing by which means the sequence was pr...