I suppose, you had some "standard" terrain adjacency in mind when choosing '3 campuses by turn x', perhaps 2 or 3 points per campus. So in order to take care of weak campus spots, I'd suggest to adjust the number of permitted campuses [for 2*N*(N+1) Population] by the number of involved campus terrain adjacency points. Ie. a "blank" campus counts 1, a 2mountains campus counts 3. Yes, the predictability wouldn't be nice. But we have a more general issue here: modifying the rules (VALID FOR ALL PLAYERS) is one thing - and we humans love to add more and more complicated & complex rules ... look everywhere! The other thing is, that the procedures which control the behaviour of the AIplayers have no magic awareness helping them to adapt to changed rules. They obey of course to the existing rules, absolutely. But if they are somehow optimized (to the original rules) and perform comparably 'good', they probably become by tendency weaker under changed rules. The current example is quite harmless - I just write now, the issue is ubiquitous. (I'm tired of the argument, that all is fair, because the rules are the same for human & AI.) .