You're right, it's definitely one data point. If you look at your algorithm's ROC: you really want to avoid false negatives ("nope, not a real friend" when the

You're right, it's definitely one data point. If you look at your algorithm's ROC: you really want to avoid false negatives ("nope, not a real friend" when they actually could be). Whereas a false positive ("yep, could be a real friend" and they turn out not to be) is something you will notice & correct eventually.