This wiki describes the zoology investigations of the false positives when looking at the images (not necessarily the flags).

Many of the stack skycells show a bias (more false detections on the top or bottom half of a cell). These biases tend to run horizontally on a chip and often do not continue onto neighboring chips.  

The stack that I investigated was stack 336346 (skycell.1316.089).  I chose a square that had a lot of false detections (green), and one that did not (red):

[[Image(falsedetbias.jpg)]]

I then chopped the cell that had a lot of false detections into two (top half = red, bottom half blue), and divided the pixel values by the sqrt of the variance.  A gaussian was fit to each of these parts, and it is in log space (y):

[[Image()]]

black = the cell with few false detections, 
blue = the bottom half of the cell with the false detections
red = the top half of the cell with the false detections