Stratigraphic architectures exert a strong control on subsurface heterogeneity and on physical processes in the subsurface, but are challenging to characterize from well data. For this, we propose a new multi-well correlation approach, which allows for defining several correlation principles and aggregate them in an automatic way. The proposed algorithm uses a modified version of dynamic time warping, which is applied hierarchically on groups of wells to produce n-best solutions. The application of this method on a synthetic test case illustrates the danger of log-based correlation, and highlights the value of using dipmeter data and sedimentological concepts. More generally, the proposed methodology can be adapted to many types of data, empowering interpreters to generate interpretive scenarios is a reproducible way.