If it is that important to reliability why isn't all software failing around us?

It's important but it's not that important outside of specific applications/use cases.

> If it is that important to reliability why isn't all software failing around us?

They do fail. Lots and lots of times. And it gets worse over time, as chips degrade due to hot-electron effects or electromigration.

You forgot random bits flipping due to cosmic radiation.

Speaking with data regarding failure rates and its actual impact would be even more appreciated.

Yah that's not going to happen. Lot of that is Fab trade secret.

You'll have to use your best guess or intuition on this.

My personal experience tells me that a bit flip is very unlikely to cause a system to fail before a bug or normal termination turns it down. My intuition tells me this is due to shorter and shorter individual program life span due to expansion of client/server architecture, higher frequency of bugs due to expansion of untyped/unstructured data, and the fact that most data manipulated by a program is not initialized/used anyway. Only in very big clusters have I ever heard of memory failure being noticable, but even there I've never seen memory corruption cited in a post mortem. Having to deal with bugs and other hardware defects forces one to work around lack of data integrity anyway.