We present a generalisation of CPCES, a conformant planner that uses two procedures: candidate plan generation and sampling of the initial belief state. The new CPCES better distinguishes these two procedures and therefore provides a clearer framework for the resolution of conformant planning problems. We study CPCES theoretically by analysing the sampling phase through the lens of tags, width and basis. The benefit of this new interpretation is twofold: firstly it allows us to bound the maximum number of iterations required by CPCES, and second it allows us to individuate sampling strategies that guarantee the discovery of subsets of minimal bases. An experimental analysis reported in the paper shows that the greedy sampling (the original version of CPCES) is the more effective strategy, coverage wise. However, when either the quality of the plans or the size of the resulting samples is important a more sophisticated sampling is more effective.