Unfortunately, sometimes the simpler system is the one you don't yet understand, and the more complex system is the one you can create on your own using tools you do know. This is the reason we still see homegrown business rule systems despite the fact that they were pretty much perfected them by the late 70s, with several open source implementations as of the late 90s. It is also the reason you still see deadlocks and race conditions in parallel code. And it is the reason you see matrix factorizations using hadoop map reduce instead of MPI.
And when these bullshit foundational design decisions are made, the "improvements" that the engineers make to the system later are almost always a reduction in fidelity to the real world process that is being modeled, instead of a move toward a system that drastially simplifies your system while simultaneously improving fidelity to the real world but might mean you have to read a wikipedia article for a few minutes.
And when these bullshit foundational design decisions are made, the "improvements" that the engineers make to the system later are almost always a reduction in fidelity to the real world process that is being modeled, instead of a move toward a system that drastially simplifies your system while simultaneously improving fidelity to the real world but might mean you have to read a wikipedia article for a few minutes.