What you say about code being changed later and using lazy programming to meet a deadline could very well be the case. Trust me I know all about how that goes down.
One point of clarification though is that I would call it good/bad programming in relation to how the stats come out in terms of realism. The programing could be good and have unrealistic results as well could it be bad and have realistic results. So I wouldn't jump to wonky stats being the result of some major programming flaw such as using OVR ratings where they really shouldn't be used.