Kind of interested. What happens if you do an XY plot?
Basically in the X-axis you have the accuracy of Tournament elo/ATP, on the Y-axis the accuracy of Ladder elo. You calculate the accuracy on a per-player basis, basically, on an individual level, which players in your dataset are the most and least consistently predicted by these metrics, and are there any outliers who are well predicted by one of the metrics but not the others.
Basically in the X-axis you have the accuracy of Tournament elo/ATP, on the Y-axis the accuracy of Ladder elo. You calculate the accuracy on a per-player basis, basically, on an individual level, which players in your dataset are the most and least consistently predicted by these metrics, and are there any outliers who are well predicted by one of the metrics but not the others.