I think there's an economic angle here. "Bad" code (like a monolithic system - allow me to beg the question a little) can be cheaper to write, in many cases, t

I think there's an economic angle here. "Bad" code (like a monolithic system - allow me to beg the question a little) can be cheaper to write, in many cases, than good code. But it is more expensive to scale, maintain, extend and debug. But it means there's a lower investment of time to get a product out. So I'd expect monolithic systems to have a higher success rate being converted post-hoc, because the systems that get that far have proven their worth. Investing more to build a system 'right' isn't necessarily a good move, if you're not sure of the return. Or, put another way, investing 5X in 5 cheap-hacky products, and then spending Y >> X to make the one that works un-hacky is often a better strategy than spending that Y up front.