Then what would you be measuring? My method measures expected vs. actual. Yours is simply measuring actual vs. actual, which will always be zero. My goal is to create a test for the specific claim of whether or not a mutation is random with respect to function. We can get a value for “random with respect to function” by actually sampling them and finding out what that is! Then we can see if biological mutations are in fact random with respect to function or not. You are trying to pre-eliminate the very question being asked.
There is error in everything, but if the direction of selection when growing your culture is not influenced by the selection you are testing against, I don’t see how it would be problematic. Additionally, as mentioned at TSZ, you could actually use a replica plating method to check for pre-existing mutations that occurred when doing pre-experimental replication.
It is indeed closest to the model when applied to a single generation, and that is the assumption of the experimental method. The “relative” one was applied over multiple generations, simply because that was the data available. However, also in the relative case, the I_Omega was also subject to selection as well.