Druckenmiller uses small amounts of algorithmic money as a signal—when multiple machine systems detect something unusual, he monitors it closely for a year to determine if they're genuinely onto a pattern or just experiencing correlation noise.