If situationism is true, why do the folk have such a robust theory of character traits? Can we provide an error theory for why people have such a theory?

Jones and Nisbett attempted to answer this question in their classic paper on actor-observer bias. It's an interesting read.

However, beware of falling into an overly strict interpretation of situationism (as I think Jones and Nisbett did) which amounts to little more than behaviorism in new clothes. People do tend to underestimate the extent to which their behavior and the behavior of others is driven by the environment, but there is nevertheless still good evidence that stable dispositions predict a respectable chunk of the variance in a person's behavior (where "respectable" means "similar in size to that for the situational data"). One of the errors of the strict situationist movement that arose in the late 1960s and 1970s is that it relied on an implicit endorsement of Kahneman and Tversky's "law of small numbers": situationist researchers erroneously expected a very small sample of a person's behavior (such as a single encounter with a person in need of help) to be representative of the general population of behaviors from which it was drawn. Not surprisingly, they found that stable dispositions are a modest predictor of these small samples of behavior. However, we have since learned that when behavior is properly aggregated across time, a robust effect of stable dispositions reliably emerges.

So in short, part of the reason that people have robust theories of character traits is almost surely that they actually map pretty well onto the territory. Situationism remains a useful paradigm, but it can easily be taken too far.