We've learned a lot about cernan's internal model and how it ought to be changed to improve aggregation performance and support sources/sinks with distinct ideas about things.

This milestone pays down a lot of domain mismatch and should simplify cernan's internal model.