Please be aware that the cred.json file is currently filtered so that it only includes the 100 top scoring nodes of each type (across all time). This means if you e.g. filter down to a particular interval, you may get extremely unrepresentative results, because the top nodes from that interval might not have been top across all time.
I’ve created a branch which disables this behavior. I used this branch for the cred analysis I published last week. If you want to do an ad hoc analysis, I recommend using that notebook as a starting point, and I recommend re-computing the cred with the no-reduce-size commit added. (I’m planning to PR that commit to master for just this reason.)