It seems very clear to me that the methods ought to remain divided and have their distinct uses. I would define attributional as using partitioning and consequential as using system expansion and substition. I would not take a dogmatic stance on average vs marginal. It is a headache to define the use cases for each as I see too many variabilites to consider. The distinction of the two methods can help by making the issues clearer and the conductor of the LCA study would have to justify how the method matches the purpose of the LCA.