Proposed change
The historical model is waaay too slow. @newtectonics has some ideas about how to optimize it. He should document here.
Why this is important
We are starting to run into use cases that can't be addressed until this model is significantly faster.