I'd presume they could look at the difference between their confidence (relevance?) in the first and second results. If the difference is more than some thresho

I'd presume they could look at the difference between their confidence (relevance?) in the first and second results. If the difference is more than some threshold, bam preload the first.