Recent changes in rank exposed a new error: if a one hop has gone through AC, then one of the query graph nodes has been made a set. We were looking at the qg to decide if a node was a set. So this was leading to scores of zero, even for the original answers to the question before AC.
Now, we decide if something is a set by looking at the counts in the binding itself. I think that this is what it used to do, so added a comment to explain why it is implemented as such.
Recent changes in rank exposed a new error: if a one hop has gone through AC, then one of the query graph nodes has been made a set. We were looking at the qg to decide if a node was a set. So this was leading to scores of zero, even for the original answers to the question before AC.
Now, we decide if something is a set by looking at the counts in the binding itself. I think that this is what it used to do, so added a comment to explain why it is implemented as such.