It seems all methods have flaws. Self-typing exists because it's convenient, people know more about themselves (especially the unverbalized/internalized components), and because it's relevant to decision-making. Then again, you lose perspective and might have blind spots to your own behavior, or false conceptions of self.As for the stats you mentioned, I've felt that way about them for years, if not decades.
1. All the stats are based on self-typing which is a method rife with error.
External typing offers the benefit of an outsider's view, potentially "more objective" assessment, seeing what people actually do (rather than imagine they do), it's more quantifiable. There are different types of external assessment (singular assessment by an "expert," collected data pools from a variety of observers about the person in question, etc.)
It seems to me some sort of method that combines approaches would be the most successful.
True. Less descriptions leaves more possibility you won't fit with any, more descriptions means there will likely be overlap and uncertainty.2. Many of the types in any given system are more or less similar to one another. It seems that the more types a system has, the more similarity there is between type-descriptions. So you might end up reading and studying a few pages on type "look-alikes" when trying to decide on your own place in the system.