If I understand the calculation process correctly, I think the computer converts each judge's GOE to its point value and then averages the values for the whole panel. It's possible it works the other way around. That would be a topic for discussion in the math thread in The Edge.
Actually that's a good point. It's probably using the scale of values of each individual judge's GOE, and then averaging the result. This is because averaging first then converting via scale of values leads to ambiguity if the average GOE were -2.42 for example. For quads it's -1.2 for -1 GOE, -2.4 for -2 GOE, and -4.0 for -3 GOE (not linear), and so the ambiguity is whether or not you linearly interpolate in between or whatever. So I don't know what the rules actually are, but I think it's more likely that it converts each judge's GOE to those values, then averages them.