Assessing extreme outcomes: the strategic treatment of low probability impacts in scientific assessment