An analytical ray tracing and error analysis is performed for small misalignments of the calibration optics. I show that, in general, errors in the retroreflector return beam direction cannot be mitigated at first order by small rotations of the beam launcher and first pellicle. Thus, the accuracy to which a true vertical reference beam can be established, and to which the astrolabe top assembly can then be calibrated, is of order the return beam deviation from the retroreflector. (There may be ways to get around this problem.) A survey of optical supply vendors reveals the lower bound of this deviation to be one arcsec. If not for this retroreflector problem, the fundamental limitation to the calibration accuracy would otherwise appear to be the level to which two overlaid spots on a CCD detector can be discriminated. This discrimination capability is set by the focal length f₂ of the calibration setup focusing lens. Assuming 10 micron detector pixels and a perfect retroreflector, values of f₂ of order 20 meters would be required to achieve an absolute calibration of order 0.1 arcsec.