Ain't that what I beat you up on a while back? Extrapolating is just a guess and what to me chooses the degree is a low r squared. Almost always the higher the degree polynomial the closer r^2 is to 1.
I think it would be the handiest in 2 bar cars where you can touch on the whole graph but want something reasonable in there.
I was going to use some trickery with linest to do the table by rows then back out and do them by columns and average the difference in the untouchable zones. The big key on what I wanted to do is only use the data I knew was good data as opposed to smoothing everything