According to DeWitt himself, the DeWitt clause came about because he figured out that Oracle had no query optimizer. However, the response of a lot of database

According to DeWitt himself, the DeWitt clause came about because he figured out that Oracle had no query optimizer. However, the response of a lot of database vendors was to then include precomputed results for entire benchmarks, and also to have code that tried to figure out if you were running one of them. This obviously adds no value to the user, but is great for marketing. Arthur is a non-nonsense guy, and, according to him, there's a ton of hand-tuned assembly language code in the intepreter. To make the most of it, however, you need to know how to set up the machine properly. Most people don't, and First Derivatives and Shakti don't have enough marketing people or consultants to set up machines for test runs or to explain away bad results. In any event, k/kdb/q are wickedly difficult languages to program, but compared to modern C++ they are far faster to program and more flexible, and compared to pandas, they are faster and more likely to not have some massive performance hit due to a scalar function. I'm sure that they miss some optimizations that a good C++ compiler targeting a processor with some of the newer instructions can do, but they allow you to write fairly performant code that operates on vectors relatively easily.