The documentation has focused (rightly) on fundamentals of knowledge graphs and ontologies; how to define them and use them for productive efforts. It is functional and commendable, though descriptions of what to do in certain edge-cases and the finer points of syntax could be expanded upon. Practice with the software meets a lot of those needs.
Where the documentation in its current state has fallen down for me is where it leads us immediately after we get excited about using it on our desktops. Having shown us all the cool stuff we can do, we are then confronted with the challenge of establishing lab systems or other persistent, production-like servers that can demonstrate this magic to colleagues, clients, and superiors.
Now I realize that, at a certain point, Grakn as a company must demonstrate the ability to generate revenue, and thus develop an unambiguous pathway from the world of free software into the licensed, paid, enterprise realm of the KGMS. However, the intermediate step here needs to be recognized and enabled so that we can point at our various knowledge-graphs, show the value they’re generating, and then ask for money so we can have more of the same, faster, with better control and assurance of security. At the moment, though, this transition from a development virtual-machine environment to something that lives in infrastructure is largely a guessing game.
It would be grand if there was more about what best practices and intended use-cases were for operating a performant medium-scale graph database. What do I need to do with Java when I want to store a million objects? What sort of replication strategy is most appropriate in a Cassandra cluster to optimize query performance? How do indexes in Grakn work, specifically? What are some gotchas to look out for when writing queries? What are some problems with scaling that people can expect to run into? What are the functions of the various tables in the Grakn cassandra data, and what can we expect if we run into problems?
The more useful you can make this system for real world applications the more likely it will be that people with money to spend will be directed your way.