The Tennessee STAR random-assignment class-size experiment of the mid-1980s—comparing classes of about 23-24 students to 15—found at best small, modest gains in kindergarten or first grade that did not compound in subsequent years, undermining class-size reduction as a reliable lever.