The point here is: 'SHOULD work on "3"' has to be proved.
Maybe, there are some negative numbers c so that (a, c') validates ALTHOUGH a is invalid. (where c' is the positive equivalent of c)
Maybe not.
So, proof of the algo as well as the code is needed.
I'm done advocating this here for a while, I can better spend my time.
I've stated it number of times, it's the implementation that's wrong, not the math. Math stays the same.
If you're interested you can go through my earlier posts.
Some interesting links:
I'm going to run NRS with patched Curve from now on.
EOT for me
Please provide the source code so we can compile and swap. By We I mean the ones who agree with you.