Just from a quick glance at the tech paper, it looks like the runout issue is mostly about how to deal with the signal noise coming from prox sensors used to monitor shaft radial movement. If the OP is asking how to filter out signal noise versus actual shaft motions from prox sensor outputs, that is a very complex issue. The shaft dynamics existing in a high-performance gearbox can involve more than simple runout. There are also various combinations of other dynamic modes like bending, torsion, whirl, etc.
Do you have some, as detected with a slow roll? After the mechanical runout is thoroughly verified as well.
Can you shift the prox probes axially to a "better" surface?
It can be caused by actual magnetism, or varying material properties within the shaft, as might result from situations like uneven chrome thickness on a repaired shaft.
Some turbo machinery repair shops offer demagnetizing, burnishing, and probably other processes to to reduce it. Seems like modern software ought to be able to compensate for it, after detecting and mapping with a slow roll.