Totally ignoring the parts you put in there to make yourself look smart, the truth is that the functions which represent CF's behavior are tweaked too often to gain incredibly meaningful insight into how many things work and additionally are blackboxed (you don't know that any given action alters only one variable).

I tend to stick with the motto that unless its easily observable, its not worth data mining, in regards to CF. Here's an example, race and int/wis are related in a general sense, but that isn't to say some high int/wis races aren't given an additional boost over others. It wouldn't be worth it to verify all the variables. That's what Dwoggurd means. You can't possibly know the magic number is 21+/20+ and be certain that you're aware of the entire scenario.

You can't guarantee single variable alterations because you don't know what tweaks what variable, only the part of that process you can readily obvserve. Hence you're pulling those numbers out of your ass, since, while they weren't arbitrarily chosen, they aren't the result of any sort of conclusive analsysis. More like educated guesses, which are still guesses.