I realize this probably doesn't address the premise of your post, but Datomic is, by way of my personal understanding/analogy, a giant, multi-indexed SQL table which has 4 columns - a subject column, an attribute column, a value column for the attribute, and a transaction identifier column. The table is streamed to consumers into an LRU cache.
The transaction identifier column represents time which enables you to query the state of the database as of a point in time (e.g. yesterday, last month, etc.) for reporting. The subject-relation-object structure supports adhoc NoSQL-style changes to the schema. Transactions themselves are first-class concepts, so you can add relation-object pairs to them as well.
Their hypothesis seems to be that querying client-side data is faster than server-side data (obvious), and that updating data can be effectively synchronized by a dedicated transaction coordinator (potentially). Although, this is a single point of failure, the claim is that this code can be audited and perfected.
The transaction identifier column represents time which enables you to query the state of the database as of a point in time (e.g. yesterday, last month, etc.) for reporting. The subject-relation-object structure supports adhoc NoSQL-style changes to the schema. Transactions themselves are first-class concepts, so you can add relation-object pairs to them as well.
Their hypothesis seems to be that querying client-side data is faster than server-side data (obvious), and that updating data can be effectively synchronized by a dedicated transaction coordinator (potentially). Although, this is a single point of failure, the claim is that this code can be audited and perfected.