Hidden likelihood support in genomic data: can forty-five wrongs make a right?