Yeah, I see what you mean. That feeling crops up with me at some times as well. I dont know what kind of strategy, planning and workflow methods they use. But from the outside they seem to have started many different functionalities at once, working to a convergence point. This has the advantage of being able to show the public (and stakeholders) something early on. Progress (in the early stages) looks great this way. But as the devil is in the detail this progress will slow down over time. A good thing with this way of working is that it allows developers to gain and apply empirical knowledge.
I think they could improve on some error detection by using more predictive methods. Even simple time-outs for states dont seem to be implemented yet. This bewildered me too as they just finished their engine. Without those safe guards errors can spread through a system like wild fire as errors go unnoticed. But without insight in a more detailed road map it is hard to come to conclusions.
Maybe they have those safeguards in place but do not enable them in public releases. That is why I suggested simple animations for when the in-game detects a fault state. I rather have object disappear with a visible clue and maintain engine stability then waiting for the complete crash as what is the current state of the game.