You definitely do not want a simple system if your singular goal is resiliency. The most resilient systems we know about are all complex, from ecosystems to organisms to cities to the internet itself. The main problem is that our current toolkit is so primitive that understanding anything about them, from how to intelligently intervene to how resilient any particular system is at any particular point in time is damn near impossible. There's a lot of research trying to change that, but it's obviously a difficult problem.
Complex systems are always in a failed state.
But "failed state" simply means that some of the controls have been disabled, at least according to John Gall. That just translates into working without quite knowing how they work.
Complex systems are always in a state of maintenance. Some part is
always failing, but we can repair it. Systemic failure happens when we
can't keep up with the necessary maintenance. It's very easy to look a
a big complex system, see lots of failure, and conclude the whole
ship is going down, but that may not be the case.
Funny thought: then we may characterize a complex system as a system that's (most of the time) too big to fail all at once, and whose resilience to failure simply arises from continued previous failure. Which sounds like another way of saying "I don't know what I'm doing, but it kinda works"... until it doesn't. Maybe when it has inevitably grown too big to be successfully maintained anymore.