The model's scoring was done by another model though no? That was the source of the answer being mislabed as correct. So a different model thought that 45+8=63.