Sort:  

hive_follows contains the custom_json data that is "understood" by hivemind. For arbitrary custom_json, we'll likely create one or more tables specifically for it, along with appropriate API methods to access it.

One of the ideas that's been tossed around is making hivemind more modular. For example, it'd be nice to be able to run a hivemind that just has accounts, transaction data, and custom_json, without all the heavy "social media" tables.

This could be extended further to where the hivemind pre-filters custom_json it's not interested in storing. This would make it easy for people to run very lightweight API servers for one or more specific apps.

Probably obvious, but this is all driving towards creating a simpler architecture for creating 2nd layer apps.