Models included span two popular datasets (MS MARCO and Natural Questions) and utilize different docid strategies (PQ and TU).