The machine learning is saying, based on all the programs you've run before, if we try these thousand edits, the r o i on that portfolio will be higher. And again, this is not hypothesis driven. It's a litle bemore yo ws that sometimes it is, but mostly it's not hypothesis driven. You have to grow the strains up in these little ninety six well plates, about the size like the old four by six index card. Those get put into fermentation tanks and then we'll actually use another set of machine learning to figure out which one's to valuate performance because performanceis super noisy....