High-Stakes Testing and Second Chances: From Data to Models