Good reading on when "doubly robust" is worth the effort?

It does to me at least. I would like to always see a comparison with an optimal method that doesn’t use weights in any way. I find that weights are always ad hoc and there’s often a full modeling approach that relies on the log likelihood.