[Home]

After learning [here] that 2,933 November 2011 users at ArrowOutlet.com are robots, the natural next question is, "Who are these robots?". The data indicates that ArrowOutlet.com is using fake bidding robots to manipulate and/or win their own auctions. In November 2011, the data shows ArrowOutlet won 11,529 of their own 16,797 auctions and manipulated the other 5,268. The robots caused 905 legitimate users to lose a combined $122,501.22 in one month.

If you visit ArrowOutlet.com at anytime, you will typically see 10 to 25 auctions occurring simultaneously for items like iPads, laptops, and gift cards. The site looks popular with an average of 2,700 users each day, but our data shows 2,600 users each day are robots. Once you recognize the robots, the situation is ridiculous. Either ArrowOutlet is being seriously hacked or they are using robots to make money.

There is such a small percentage of humans that 5 to 10 auctions are 100% robots fighting themselves. After a while, a human will join in, spend $100, and wonder why they can't win. The robots all behave exactly the same, so maybe they all belong to one legitimate user. But why would they fight themselves and lose 1.5 million dollars each month? Common sense tells us that the robots belong to ArrowOutlet.

Below are specific arguments. From our previous discussion, we know that these robots are bidding without using ArrowOutlet's webpage interfaces. These robots are either computer scripts running on computer(s) that communicate directly with ArrowOutlet's server or they are fake bidders built into ArrowOutlet's auction software. How does the data indicate the latter?

Argument 1: All the robots work together and fight each other. When a human is bidding in an auction against 2 or more robots and the human is about to win, one of the robots will place a bid when precisely zero seconds remain and prevent the auction from ending. Even though each robot places 25% of their bids at zero seconds and all robots have collectively placed 684,003 bids in this fashion and 1 user can not know if another user is bidding at zero seconds, 2 robots have never simultaneously bid at zero seconds. [more details here] It is impossible for these 2,933 robots to be independently working scripts. The robots must be working together, but, in thousands of auctions, the robots fight each other and prevent each other from winning. The robots lost $1,548,709.42 when they didn't need to. This makes no sense unless the robots belong to ArrowOutlet.

Argument 2: The robots consistently place bids when precisely zero seconds remain. By analyzing the computer to computer communication times needed to accomplish this, the only way to achieve this type of accuracy is for the robots' computer(s) to be in the same Chicago building as ArrowOutlet's server. Knowing precisely when an auction will end requires extremely fast communication with ArrowOutlet's server because of clock synchronization [more info here]. Consistently bidding at the same time (whether it be at 7 or 0 seconds) requires immunity from variable internet lag. The robots placed half a million bids at precisely zero seconds with a standard deviation of 0.009 seconds. This accuracy is unachievable for a source outside the building. [more science here] Amazingly, the robots have slightly better accuracy than the Bid-O-Matic which is a script that runs on ArrowOutlet's own server and places bids for users when precisely the integers 5 through 15 seconds remain. The robots' integer placing behavior is strangely similar to the Bid-O-Matic also. The Bid-O-Matic placed 5,173 bids for user mflatt at precisely 7 seconds and suffered a standard deviation of 0.012 seconds. (This is worse than the robots.) ArrowOutlet takes an average of 0.01 seconds to process a new bid before sharing the information. The robots bid so precisely that the standard deviation we are observing is actually ArrowOutlet's fluctuating processing times and not even the robots' fluctuating bidding times. All 2,933 robots share the same histogram and behave the same. All these robots must be running the same computer script, but, in thousands of auctions, the robots fight each other and prevent each other from winning. Therefore if the robots do not belong to ArrowOutlet, they must either belong to 2,933 different people running the same script or 1 person who has strategically chosen to lose $1,548,709.42 each month fighting himself. What are the odds that 2,933 different people each rented a computer in the same Chicago building as ArrowOutlet and run the same script? And why go through so much trouble to lose money?

Argument 3: [here]