Nice article, I don't necessarily agree it has so many 'disadvantages' as you list as they are mostly an artefact of complex data or the underlying model not properly built. One should be able to build very precise intervals with the proper underlying model and properly selected non conformity measure. Also the width of intervals is a tradeoff with lack of bias. Any alternative model that claim they can build narrow intervals are usually hiding the fact that they are producing bias as a result :) wider intervals are harmless (narrow intervals are a nice thing to have, not a must to have) whilst bias is very harmful

https://medium.com/@valeman/does-ngboost-work-evaluating-ngboost-against-critical-criteria-for-good-probabilistic-prediction-28c4871c1bab

https://github.com/valeman/awesome-conformal-prediction