Sort algorithm for probablistic comparison function

From: Anthony Uk (chip_at_dataway.ch)
Date: 04/06/04 

Date: 6 Apr 2004 09:29:11 -0700

My apologies if this question is not posted to the right group.

I am looking for any pointers regarding sort algorithms where the
comparison function returns a probability value rather than a clear
cut < / = / > as in normal sorting. Consider the task of ordering n
different wines by quality. The sort algorithm gives a human taster
two glasses of wine and asks which is better. In some cases, the wines
will be very similar, so the human can only reply "I believe @25% that
wine A is better". What algorithms are there that can make sense of
this kind of data?

(As an aside, a high probability value would indicate that the
distance between the two items is large; there may be optimisation
potential here.)

Any pointers to relevant papers etc. would be much appreciated.

Thanks,
anthony