Oh I see! And this implies as you would just use a "19-bit" type B in place of a "20-bit" type A, as it would cost 2x as much they are exactly equivalent with regards to cost & variance. QED. Adam On Fri, Aug 13, 2004 at 03:25:23PM +0000, Justin wrote: > On 2004-08-13T09:07:44-0400, Adam Back wrote: > > Yes. It's just the 21st bit of a 20 bit stamp with type B would > > _have_ to be 1 whereas now (type A) it can be either 0 or 1. > > Yes, but I think you miss my meaning. > > The new scheme (B) is indistinguishable computationally (or wrt entropy > or probability distributions) from a 21-bit stamp. > > You don't gain any advantage. You could specify any pattern you wanted > for collisions. For a 20-bit stamp, some could be: > > 01010101010101010101 > 10101010101010101010 > 00000000000000000000 (current) > 00010000000100000001 > > What you propose is > > 000000000000000000001, which is a 21 bit pattern, therefore it is the > equivalent of a 21-bit stamp under the current validation scheme.