Tuesday, November 08, 2005

GR Mapping

For you implementers out there here is an excellent article written by Steve Churchill, ooTao’s CTO, that gives a concrete example of how we executed our first successful Graph/Relational mapping implementation. This implementation utilizes the plugin architecture that our XDI Like (no spec) server supports.

Friday, November 04, 2005

I fought the law...

So, I went out on a limb and coined a law a while back:

The value of a transaction between 2 parties should never be greater than the reputational collateral exposed by either party.

While at IIW2005 I had the opportunity to chat with Allan Schiffman of CommerceNet we got to talking about reputation and so I rolled out my law… he very nicely pointed out the flaw in my thinking; My law assumes that, like in the real world, people can only be in one place at a time, doing one thing at a time. My law breaks down if a user can build a $10K reputation and then enter into fraudulent $5K transactions with 100 people at the same time such that the latency of the reputation system is greater than the time needed to complete the transaction. It’s a great point.

I think this drives a change to my law but doesn’t completely invalidate it. There are 2 ways that I need to mentally adjust my thinking to accommodate this new reality;

  1. Think Small. Reputation systems that are designed to operate within the size limits of workable human groups will be much more effective. For social accountability to be meaningful you need social collateral not just reputational collateral.

  2. Add latency to reputational transactions that mirrors the latency of the reputational feedback mechanism. For example; when I agree to buy a laptop from you for $500 bucks you put your ‘seller’s reputation’ on the line, I get the right to effect your seller’s reputation by 100 points. That 100 points should be put ‘on-hold’ the moment that we agree to consummate the transaction and should only be freed once the transaction is completed to the satisfaction of the buyer. Another would-be buyer might now see the seller’s reputation as 400, with 100 on hold. This would reduce the latency of the system to almost 0 and limit ability for a seller to over-use their reputation. I think. This is just one way to negate the latency effect, I can think of others, it’s basically just having awareness of this problem and implementing _some_ solution to mitigate or remove the risk.

So maybe I can simply change the law to:

The value of a transaction between 2 parties should never be greater than the available reputational collateral exposed by either party taking into account the latency of the reputational feedback mechanism.

I think I need to strengthen that; I’ll work on it.