Your bet met is running the algorithm ~20-100 times and comparing the outputs, then averaging.
It's pretty horrible honestly.
http://en.wikipedia.org/wiki/ATLAS-I
I found it while searching for large wooden buildings a while back:
"Trestle is the world's largest structure composed entirely of wood and glue laminate"
Hardware hardeneing is easier, just make large band gap's in your transistors.
My understanding is that the real threat from EMP is physical destruction of the circuits by inducing voltages beyond what the wiring can handle. Any bit flips would be transient, since the EMP doesn't last very long (thus the P part).
Trying to use software to protect your chips from being fried by hundreds or thousands of volts being shoved through parts that weren't designed to handle it is clearly not viable.
>I'd wager [..] that resilience in the face of EMP was probably a big factor too
The poster was talking about software reliability, which the next poster asked if software reliability was related to EMP hardening.
The only mention of software in the whole comment is a brief comment about "upgrades to both the system and the software".