No, complex (ie lots of knobs) models shouldn’t be used because they are, best case, useless, worst case they are self confirmatory Rorschach Tests. Methods to fool yourself. Examples abound, but I think almost everyone has to experience fooling themselves with this methodology at least once or twice - build big model, get burned, reverse engineer why it didn’t work correctly, notice 20 random things you picked that were highly uncertain but for prior-belief or motivation reasons…