The problem with just tracking ordinals is that a judge might have two teams 2/3, but seven points apart, while another judge might have 2 through 7 or 10 within four points of each other. Even knowing their score will be tossed, they can also low-ball by aiming to be the lowest within the corridor, leaving another low score in the calculation, or the opposite, stretching to the limit, so that another high score is in the calculation. Those are only two considerations, especially when none of this is statistically significant, given the small number of inputs. There are moving parts that need to be coordinated not only for collusion, but for the poker aspects of knowing your colleagues making good predictions based on prior scoring and prior interactions with similar panel composition.