Looks like A and B are swapped in the "errorB vs. errorA" plot's description: in other places (and on the plot itself, actually) A is the noisy sensor, B is the occasionally-faulty one.