(This, by the way, is why I prefer Eliezer’s method of starting from the dependencies…)
I wanted to note that if dependencies are randomly already present in some fraction of the population, the ‘reverse order’ lets you convey your point to growing fractions of the population (as you go back and fill in more and more dependencies), whereas the ‘linear order’ doesn’t let you convey your point until the end (when everyone is able to get it at once).
Yes, this is a fair point. (There does remain the fact that the dependencies ought to be marked, so that those who lack them can clearly see that they lack a specific, recognized dependency, and so that they may be able to trust that the author will later fill them in, to fulfill the ‘IOU’, as you say. But that aside, I agree that your point does make the case for the “dependencies first” order less clear.)
I wanted to note that if dependencies are randomly already present in some fraction of the population, the ‘reverse order’ lets you convey your point to growing fractions of the population (as you go back and fill in more and more dependencies), whereas the ‘linear order’ doesn’t let you convey your point until the end (when everyone is able to get it at once).
Yes, this is a fair point. (There does remain the fact that the dependencies ought to be marked, so that those who lack them can clearly see that they lack a specific, recognized dependency, and so that they may be able to trust that the author will later fill them in, to fulfill the ‘IOU’, as you say. But that aside, I agree that your point does make the case for the “dependencies first” order less clear.)