I can't say how this applies within the NSA, but (security) hackers tend to write better publicly released code than academic mathematicians, as the constraints imposed on them tend to encourage soundness of implementation over soundness of concept. I have no idea how applicable this trend is to NSA security engineers vs whoever inside the NSA might have been tasked with this. I just think it's possible that if this were an NSA project, that doesn't necessarily imply good code.
How much NSA code have you seen? SELinux is one of the few pieces available and seems reasonably competent.