Richard: I don't think the markets are that good for quant funds, one of the market neutral funds is down 20% for the year. Even the ones that don't take factor exposure haven't done well. I think there's also some interplay between having a target that's residual and then the models tending to generalize more. I think the targets are a big reason, but also that if you're going to do machine learning problems in the same format that we've set them up, it really is quite well designed for residual targets. It wouldn't make as much sense to have the problem set up without them. The other approach we could have done was given the users all of the sector and factor data and said, "here's the country category and the value category," or something like that, and had them think about the whole problem. But I think this approach is a lot better because you're building things that tend to be much more robust and generalize better because anything you're learning on this data set is something that applies to all sectors, countries, and regimes (ideally).