You're right, their description should be clearer, or have an elaboration or something. But for onlookers:

> flexible

Minimal, non-tabular schema

> time-based

All historical values are preserved.

> facts

Triple store: Entity / Attribute / Value

with time:

Quad store: Entity / Attribute / Value / TIME

> supporting queries and joins

Datalog query language

> with elastic scalability

Reads are uncoordinated, so they scale horizontally

> ACID transactions.

You can trust the database won't lose stuff you write to it.

> leverage highly-available, distributed storage services

Storage engines are pluggable. You can choose backends such as filesystems, Riak, Dynamo, traditional SQL databases, etc.

> puts declarative power into the hands of application developers.

Reads occur in-process, like an embedded database. Queries, traversal, and other reads occurs locally, without network round trips.