Most importantly, the analysts using the satellite images can look look at historical earnings reports to see whether the tool has any utility before they use i

Most importantly, the analysts using the satellite images can look look at historical earnings reports to see whether the tool has any utility before they use it to make a prediction. In fact, I'm almost positive they did (in part because the article says so, more or less). EDIT: In particular, the regression mentioned was no doubt based on the relationship between historical earning reports and the sat imagery (once controlled for daily, weekly, monthly, and seasonal variation, etc.)