Take a look at these 5 graphs and what happens to the batch of bots from beginning to end. As the parameters are adjusted gradually from permissive (first graph) to restrictive (last graph), the GrapeRank scores of the bots gets pushed downward. 
Discussion
true, but almost everyone else gets pushed down as well.
In the above graphs, the grapevine algo interprets a follow as meaning “real nostr user” and a mute as the opposite. So one of the adjustments is to increase the amount of weight given to a mute. This will have more of an effect on the population of bots than other populations.
One of the next steps will be to incorporate reports (kind 1984), which should likewise help in the differentiation. I don’t know if PageRank has a natural method to incorporate different types of edges or not.