Learning when 'good' crosses over to 'bad'