I have to admit that the text is a bit long! We sorta did say all of that you are saying, which means that the way I resumed the text here was a bit misleading.
There must be conditions when a heuristic like “follow the majority opinion” must be triggered in our heads: something is recognized maybe. There is selection pressure to find social exchange violation, but also to be ingenious in persuasion. Some of this already has experimental support. Anyway, we think that what we today call fallacies are not accidents—like the blind spot. They are good inference rules for a relatively stable environment, but cannot predict far into the future and cannot judge new complex problems. That may be why we don’t spot the fallacies of small talk, of experts in domains with expertise, or in domains for which we already have intuitions.
That would imply that a bad decision today is not necessarily the product of a cognitive illusion, but that we build a bad interface for the actual human mind in the modern world (a car will be lighter and faster if it shouldn’t accommodate humans). Reference class forecasting or presenting probabilities as frequencies are just technologies, interfaces. The science is about the function and the fallacies are interesting precisely because, presumably, they are a repetitive behavior. They may help in our effort to reverse engineer ourselves.