I've read it, and that's some effort by You make to sampling from each attack type! To be truth, I also find its very difficult to create an ideal modelling / formulation, especially due to abilities could have Butterfly Effect if combined in certain formation. Comparing raw cards (cards without abilities at all), might be a simpler approach. The other constraint, as I mentioned above, there's just not enough cards available for sampling, which hinder the accuracy of the model.

I think its good for you compare cards with equal mana cost. I've considering the effect of Diminishing Return (see my post here @kid.miniatures/the-law-of-diminishing-return-in-cards) where higher mana cards tend to have less value than smaller mana cards combined for the same mana.