Is the four-letter type thing actually real, or is it basically astrology?

It is neither, and the honest answer is more interesting than either camp's version.

What holds up: three of the four axes correlate substantially with dimensions that appear in every well-replicated trait model. E–I tracks extraversion, S–N tracks openness, T–F tracks agreeableness. Those correlations are consistently found and reasonably strong. So the questions the axes ask are pointed at real, measurable variation — this is not astrology, which is keyed to birth date and correlates with nothing.

What does not hold up is the part people care about: the sixteen types. Turning each axis into a letter requires cutting a continuous distribution at its densest point, which produces categories that are unstable on retest for anyone near the middle — a substantial share of takers get a different code weeks later. Test–retest reliability for the whole four-letter code is poor for that reason. The type-level claims, the function stacks, and the predictive claims about careers and compatibility have weak or absent support. J–P also has the loosest correspondence to anything in the trait literature.

So the useful summary: the axes are pointing at something real, and the code is a lossy compression of it that throws away the one number you most need — how far from the middle you were. The reason the framework survives despite this is not fraud. It is that the code is a portable identity symbol and a percentile is not, which is a genuine advantage the trait models have never matched.

What to do with it: use the code as a starting vocabulary and treat any specific claim it makes as a hypothesis. Where it says something checkable about your behaviour, check it.

One thing that helps at work, since that is where you met it: the framework is much less harmful as a shared vocabulary than as a filter. A team using it to say "I need the agenda in advance" out loud is having a useful conversation that would otherwise not happen. A team using it to decide who gets client-facing work is making a hiring-grade decision on an instrument with poor retest reliability, and the published validity for that use is far weaker than most people running the session realise.

So the eye-rolling and the enthusiasm are both responding to something real. If your workplace stops at vocabulary, it is fine. If codes start appearing next to names in staffing decisions, that is the point to push back, and the retest reliability figure is the specific thing to push back with.