The mathematical approach maximises information gain per guess rather than targeting the most likely word resulting in near-perfect outcomes.