General discussion

We can compute reputations for pretty much any uniquely identifiable target. Back when we designed the system, we had a long discussion on what these targets should be. There were a couple of considerations to take into account:

1. It should be possible to point out a single entity that is responsible for a target. This means that we want targets that someone owns.

2. There shouldn't be too many, or too little targets. We want to divide the web in manageable, yet representative chunks.

3. The name of the target should convey us as little information about the user and her activities as possible.

Our decision was to use DNS names, because they naturally divide the web in administrative domains. Each domain name is controlled by its registered owner. If the owner wishes to share control with others, it can assign subdomains. We also believe domain names provide a good compromise between granularity and privacy.

Now, while most of the web uses domain names to manage ownership and control, there are exceptions as you pointed out. Yahoo's Geocities, for example, allows anyone to create content under their domain and uses directories to divide control between content owners. This results in the reputation for the site to be the aggregate of all the content under the domain.

To me, this is what I would expect. When I come across a random page, the first thing I look at is the domain name. If the page is hosted under geocities.com, my earlier experiences tell me that I shouldn't trust the content. As in real life, surroundings matter when it comes to trust and Geocities tends to be suspicious overall.

Of course, you have a point in wanting more of the URL to be used for identifying targets. We haven't ruled out expanding the name space in future, but at the moment, we feel domain names are the best choice all things considered.

You're absolutely correct. It's not possible to measure trust with a single value, because there are different types of trust. For example, you may be confident that someone won't steal your identity, but you may still suspect they might cheat you in a card game. This is the reason we ask users more than one question. Each question is there to measure a different kind of trust in the website.

Sure, users may still interpret the questions slightly differently and they probably have a variety of reasons for their opinions. And that's ok, because we don't aim to measure trust in surgically precise categories. While users may disagree with the exact definition of the term "business partner", the question points them all to the right overall direction.