Since the weights in this example are observation weights, they are not linked to any specific observation variable. Rather, they are linked to the individual observations. Since they are linked to the observations, there is only one weight for each row.
I've included a short script that computes the 'Goods' and 'Bads' for different variables without weights and with weights. It shows that the 'Good' and 'Bad' counts when aggregating different predictor bins are the same but that the 'Good' and 'Bad' counts change when you use weights.
