I'll bet he want to complificate the formula to look something more like

"P(H|Q)=P(Q|H)P(H)/P(Q)" where

H = "User upvoted a honeypot article" and

Q = "User's votes are not a good signal for article quality"

And a similar adjustment for flagging.