Obviously, networking is highly technical, and what you can do within the context of networking is extremely bounded by the limitations of both what kind of connections people have and potentially by the Xbox Live service itself in some ways. How does that push-pull work on deciding those features and implementing them?

CB: It is a challenge. Online is a really multidisciplinary thing, in terms of the engineering technology that's needed to go into it, and the overall vision for what the community is that you want to create.

And also, really, the artistic ability to create an interface that can entertain players as they're using it, but it also feels natural and transparent. It doesn't feel like a burden to reveal the features that are hidden within it.

It's a really, really hard problem. It's something we have quite a lot of experience doing, because we had a chance to do it on both Myth games and on Halo 2 and Halo 3. So we understand, I think, a lot of the hard problems. It doesn't make them any less hard.

When you're coming up with ideas for the networking and things you can do with the multiplayer which push things forward, at what point do you, as a technical person, have to say, "Well, that sounds great, but that is not going to happen?" Is that something that happens a lot, or is it so collaborative that it doesn't happen?

CB: A little bit. I think there are some things that you might want to do. Let's say we want to do 16-player co-op, for example. That's not something we came up with Halo 3, it doesn't really make sense with Halo 3's design.

You have to look at both, "Why are you trying to achieve this in the first place?" and "Is this a core part of what your vision is for the game?" If it is a core part of the vision, then we get to consider it on the same level as all the other core features, so it's something that might require significant investment. But sometimes you do have to say, "We're going to have to find a different way of doing that."

Almost always, if you take a specific request from designers that seems crazy, you can drill back down to them and say, "Well, what we're trying to go for is this subjective feeling in the player, the feeling of epic scope," or so on. It doesn't actually mean we need 500 AI-controlled Warthogs.

Maybe we can just have something that looks like a whole lot of Warthogs off in the distance that makes the player feel like they're part of this big battle and conveys the same emotional response without necessarily being that specific technical implementation.

Halo 3 had ambitious four-player co-op. That must require a lot of testing, and if the levels aren't really basically ship-quality, then you can't test them accurately.

CB: Yeah, Halo 3 is a very difficult game to test, because it has so many features and so many challenges. Like, what if I'm playing Forge with 16 people and the host leaves and then the host migrates twice, and then we save a film of that whole experience? And if I'm watching that film back at a party with three other people... how do you test that? That's like a ten-stage process.

There's just a tremendous ball of all of these features that interact with each other, which is one of the things that's great about the game, but also makes it very difficult to test. The only reason it was really possible for us was... I guess there were two reasons.

One was that we have a really great test team that understands from a very early stage all of the features that go into the game and all the challenges they will be for testing. They're able to help us understand what would be necessary in order to test the game.

The other thing is that we have a pretty strong focus on automated testing of the game. We have stress farms of 400 Xboxes that just run the build overnight. I like to tell people that the E3 announcement build of Halo 3 that was in 2006, we finalized the build about five days before E3.

In those five days, it had two years of testing on it, just because of the number of Xboxes we had running continuously for 100 hours before we showed it. So we were pretty confident in that build before we showed it. It was going to be solid.