Why are we relying on code for controlling 
    Safety-critical system? e.g. Nuclear Reactors

Because humans can't perform calculations quickly enough to compute math millions of times per second?

Computers work well when there are a finite set of fixed, quantitative inputs and outputs: if x < 20, do y

Many parts of law are a lot more complex and can't be reduced to such formulas. Suppose you're suing an employer for wrongful termination. They say your job performance sucked and that's why you were fired. You say it was a hostile workplace, and when you tried to point that out you were fired in retaliation.

Go write some code to evaluate that. Let us know when you're done. Should be pretty exciting. Hopefully we can retain all of the judges and lawyers now that they're not needed any more.

All of that code usually has a human override, and it can be turned off, fixed and upgraded. Also, it isn't (in the main) in the public and under constant probing attack.

In the cases of those I'm familiar with, most code in the system does not need to be bug-free for safety, and the code that does is kept simple and heavily reviewed at the binary level.

For most contacts, it is not financially feasible to use the techniques developed for safety-critical software.

Well, nuclear reactor use dumb electronics overrides, and each time an operation is done inside, each operation is done with human supervision. Override mechanisms are regularly tested and used. If your override is not dumb/simple enough, and you rely on code, you have the 380 max.

What are the override, who are the supervisors for cryptocurrency?