Yeah, the FoundationDB row is kind of botched.

They also say that "ranges need to be manually locked to avoid phantoms" which is not true. Range reads, deletes, etc. in FDB are all strictly serializable with respect to other operations.

Finally, they describe FoundationDB as "modified percolator". There may be some similarities I guess, but FoundationDB was architected without knowledge Percolator and they are very different. In particular, Percolator is a batch system and FoundationDB is designed for OLTP workloads.