Without our recommendations those "weird" emails are what everyone would see. There is a lot of noise in contacts. We have a pretty good blacklist, and it's get

Without our recommendations those "weird" emails are what everyone would see. There is a lot of noise in contacts. We have a pretty good blacklist, and it's getting better all the time. We'll soon have enough data to make the blacklist more automated -- essentially classifying patterns of email automatically. Right now, it's a whack-a-mole. This is a big focus of ours. The ranking for this demo example is only pretty good because not many people have used this flow. We just launched today. Normally we have a lot more data to train the ranking. We do extract patterns across thousands of users. For example, do people tend to invite those with the same last name? To enable this learning, we do keep the data around, but we don't share contact data across apps. We aren't a data broker. The basic model is that we won't send more contact data than a user already shared with you. But we do try to find more signal to help with the ranking, and that can include, for example, counting mutual contacts that we learn from any source. Again, this data isn't shared back, it only informs our ranking. As far as pricing, it's based on the number of address books -- not your user count. So the free tier actually covers a social graph with around 1M edges -- which is a huge value. No, the service doesn't automatically shut off or anything like that. If the pricing doesn't fit your app (e.g. you're very high scale and a free app), just email me: ivan@yesgraph.com We already priced things to match a high scale B2C app, so this pricing is a steal for most users.