Great stuff. A couple of questions:

1. Doesn’t projecting through Kafka look like overkill over catch-up subscriptions, skipping the fact that subscriptions tend to fail on overflow? I think this issue can be fixed, then there will be no need for Kafka there?
2. Greg was telling that they want to work (or already working) on weaker consistency replication model, specifically targeting cross-datacentre replication scenario. Have you talked to them before jumping to build your own store?
3. Do you have plans to open-source your replication component? It would be really cool to look at what you guys have done.