Clarification regarding Calibrate Fiducials

[radar_boy]

Am I correct in guessing that Calibrate Fiducials assumes the fiducial marks are symmetrical (equidistant from the principle point) and uses the given scan resolution to calculate their approximate coordinates? Is this done on a per-image basis, or use means (or some other statistic) if multiple images are selected together (therefore more representative coordinates)? I'm just seeking some clarification please. 

If this is correct it'd be great to add such to the manual for future reference.

Many thanks

Andy