(ack! accidentally submitted early; major edits included with this header)
The hypothesis about how the "anti-" prefix is being parsed is interesting; deploying some type of updated (probably machine learning) ranking algorithm with wild and unpredictable consequences to random subjects of that algorithm is consistent with other environments like YouTube (random & unexplained demonetization, etc).
However, this doesn't change much; the results are the same regardless. Google is responsible for the decision to de-rank antipolygraph either deliberately or indirectly with their design of the ranking algorithm.
If it was due to an algorithm, this is another example of the need for algorithmic transparency. You don't get to abdicate responsibility for decisions simply because an algorithm was used; the choice of algorithm (and the data used as input) is itself a decision.
The hypothesis about how the "anti-" prefix is being parsed is interesting; deploying some type of updated (probably machine learning) ranking algorithm with wild and unpredictable consequences to random subjects of that algorithm is consistent with other environments like YouTube (random & unexplained demonetization, etc).
However, this doesn't change much; the results are the same regardless. Google is responsible for the decision to de-rank antipolygraph either deliberately or indirectly with their design of the ranking algorithm.
If it was due to an algorithm, this is another example of the need for algorithmic transparency. You don't get to abdicate responsibility for decisions simply because an algorithm was used; the choice of algorithm (and the data used as input) is itself a decision.