False positives are the difficulty.
yes.
The essense of the original proposal would be this wording:
If a MAIL FROM address is known to be forged, a bounce MUST
NOT be sent to that forged address.
well, I'd probably say SHOULD NOT, but I think that's the basic idea.
I suppose we could leave "is known to be forged" as an exercise
for the reader.
I have doubts about that. I've seen too many bogus spam detection
algorithms. I don't want to encourage silent discarding of mail based
on bogus criteria. and it's very difficult to define what reasonable
criteria would be.