That's an outstanding question. What the machine does is examine a very large number of candidate drivers - earnings, expectations, macro environment, interest rates - and then checks for any anomalies (are any of them really stretched relative to history, for example). It will then analyze the entire available price history to see if there was any detectable pattern previously between any one driver and subsequent price action. If the pattern is sufficiently robust - i.e. passes our back tests - we'll post it as an insight to let you know there is a driver worth paying attention to!
I hope that helps,
Jan