Any reason to choose this particular division between the 2nd and 3rd layer (we proposed k=15 before as lowest index for layer 2)? I suggest to explicitly define also other aspects of the base case: 100 additional nodes, BHP data with 0.1% error, minimum of 5 bar?, all instantaneous rates with 1% relative error and minimum error of ??, no resampling, ensemble size 1000 (or whatever we think we can afford), 10 iterations with weights 4,2,1,1,etc., schedule with constant rates, no gas injection, simulation period until ??, parameter type individual, no use of the region option, and some fixed initial distributions (use uniform permeability!), no aquifer. Since the distribution parameters are a bit difficult to list, we should store and distribute the base case config for reference. Do we use the "constraining" functionality at all? Also the number of additional nodes could be tested as part of the "geometry" experiments. Is there anything related to the treatment of faults and removal of connections that is relevant here?

No particular reason for the layering setting, it can be k = 15 (just changed it in the list above). Also added one experiment with variation of number of additional nodes (base case = 50, then 100). For the other settings, I think it is better to do as you say (distribute the base case config), my intention here was to focus on geometrical aspects for now. For the faults, I guess it would be interesting to see the impact of removing connections across sealing faults.