If I understand this right, they can guarantee fresh point reads. But an actual query needs a consistent snapshot of all entities (rows) under consideration for

If I understand this right, they can guarantee fresh point reads. But an actual query needs a consistent snapshot of all entities (rows) under consideration for the query. ... for all of the "hundreds of queries a mature application might execute to fulfill a single request". You need a snapshot of the entire database (like git) FTA "When mutating a resource in the system we’ll store the last committed log-sequence-number for the entity making the request" This is what Datomic does by design, if you need infinity read scaling and perfect consistency just use that, and you get a host of other benefits (the replica is inside your application process, like a data structure, which has profound architectural consequences, for example you can query in a loop so N+1 problem is solved).