> But in the latter case, as we saw with New Jersey's unemployment system, that "reliability" was fragile and contingent on the current state of affairs, and had no hope of withstanding a sudden shift in usage patterns.

"Reliable" and "Can survive a sudden shift in usage patterns" are extremely different things.

I think you have the causality backward. Engineering is about trade-offs. No quality guild will be able to wave those away. As long as the primary pressure is "get something that is functional enough at minimum time and cost" you're gonna have this.

(Software is particularly complicated because engineers, not just managers, have poor understanding of system quality and of each other's contribution quality. There's a combination of "it's not that complicated" complexity-blindness to business requirements and trade-offs that have to be traced through deep call stacks and across networks. We build things like chaos monkey - to prove resilience by seeing how hard it is to break the thing - because we don't have cost-effect techniques for actually understanding the system well enough short of operating it.)