Predicting fault surfaces from sparse evidence is key for capturing the layout of rock units in the subsurface. This can be consequential to optimize the trajectory of horizontal wells in layered reservoirs. Whereas 3D seismic images provide an excellent source of information to address this problem, limited bandwidth, noise and processing errors may lead to ambiguities in the prediction of faults. In some cases (or at some scales), fault surfaces must be inferred only from 2D geological sections fault intercepts identified in boreholes. Cherpeau et al (2010); Cherpeau & Caumon (2015) started to address this problem by proposing a sequential fault correlation algorithm, which iteratively simulates fault data anchored on observations or simulated fault points. More recently, we have realized that the fault association problem could be more efficiently addressed using graph theory (Godefroy et al., 2019). Several simple criteria can be used to describe the likelihood that any two fault observations belong to the same fault surface, which allows to efficiently rule out impossible configurations with the graph framework. Using these criteria, a large set of realizations can be produced. The use of far-field displacement or deformation data appears like an interesting avenue to improve this method and further reduce the set of possible models.