I agree we should abandon this idea I suggested, because the owner can not
probably estimate the amount of forgery that will occur on his domain in
*future*, even if s/he can estimate it now.
I think the only determinant SPF record is "-all". All other records types are
indeterminant of a balance between your false positives and your domain
reputation (false negatives), depending on recipient interpretation (unless
recipient has the breath of data to actually measure the probability).
It really depends on what recipients do with SPF. It will be an interesting
experiment.