If your method of protection against basic code is to use a private language, then (under the correct circumstances) the injected code behaves as random code when attempting to execute.
The hope is that random code will crash, and the sooner the better, but do we have some numbers for it?
It turns out that we can (sort of) calculate what is the probability that random code does not crash, but instead lands in some valid code which is bad as no alarm will be raised about the attempted attack.