Picks a chart by starting from the question, not the data: what kind of question it is, which visual encoding answers it most accurately, then the chart.