I agree with a lot of this. I don't disagree with the fact that it's not necessarily super effective because the distributions of personalities of people are n

I agree with a lot of this. I don't disagree with the fact that it's not necessarily super effective because the distributions of personalities of people are not bimodal, as would be ideal for this simplistic sort of classification. It's also unlikely to be supremely effective at clustering in that way because the design of the metric wasn't really data driven in any systemic way. It's not even like someone used principal component analysis to really establish that the four classifiers are reasonably orthogonal. It's probably not super obvious/trivial to find human language-friendly orthogonal aspects of personality that are each bimodal, but I imagine the surveys could be improved informed by better data science. It's certainly a compression algorithm, as are all classification based metrics based on answers to multiple questions, and I'm not arguing it's the most effective one around. However I think before making claims like it has no "external validity" the degree to which it might have it should actually be tested with reasonable methodology and I'm offering something that can absolutely be tested scientifically. There's nothing impossible about having some people take the test and having others who know them try to classify them without the results and see how well the two coincide. It's also possible to evaluate the degree to which people can decompress the data. You can set it up where people have equal time spent with two people of each MBTI and see how well they can guess their answers to the original survey questions with additional knowledge of the MBTI as opposed to without it. It won't establish an upperbound for how useful an MBTI is but it will help gauge a lower bound and I'm pretty confident the lower bound will be higher than that of a zodiac. I imagine we can design a better classification, probably with even the same survey questions, but just because a metric wasn't well-designed doesn't mean the information it provides is meaningless.