Also to Dave: Have you coined “zero-knowledge auction” by analogy to zero-knowledge proof? Does it boil down to a commitment scheme (see link at the end of the post)? I tried having wolframalpha multiply some big random numbers, but for anything not totally unwieldy it seems to factor them as fast as it multiplies them. But you can do this:

http://www.wolframalpha.com/input/?i=md5+%22my+bid+is+123+and+my+nonce+is+somerandomstring%22

The hash is in theory unreversible so if you first tell each other the hashes per above, you can then reveal your bids sequentially. The second person has to stick with the bid they committed to because that’s the only thing that will generate the same hash. (Note the need for the nonce otherwise you could find out the bid from the hash just by trying all possible bids.)

It might make Dan G happy (and increase the security) to go further and bit.ly that URL, but I agree with Dan R that this seems like overkill.

As for this application, I’m not convinced it’s such an issue. Even in the famous example where subjects anchored on their own social security numbers they first wrote down a portion of their SSN that was within a plausible range for their bids. How do you get anchored by a totally random long-ish string of digits?

The effect of anchoring may cancel out over repeated plays, but why have it there when it makes it harder for the other person to think clearly? I mean, you wouldn’t sing the Star Spangled Banner at the top of your lungs when the other person is trying to come up with a bid, a distraction is a distraction.

While no representation is neutral, you can probably do better than a longish number.