I am surprised that the author left out another mitigation. To build solutions (models) that are constructed to be more transferable (amenable to out of domain data). For example, in machine learning, using physics informed models. Perhaps this is simply a sign that the author is a proponent of generic, deep-learning.