That second point is very interesting. Beyond reading code / SLA for the lib, I'm not sure there's an easy (read: time efficient) way to understand what data points are used for what purposes currently. At least it seems that would hold for a lot of services.
Am I missing something here?