MCAO systems on ELTs should deliver the best available relative astrometric precision from the ground, which may be sufficient to detect exo-earths orbiting nearby brown and red dwarfs (< 30 μas). Yet despite the success of AO astrometry on large telescopes, we still do not understand the systematic errors that may dominate long-term astrometric stability with MCAO, such as dynamic optical distortion and differential atmospheric refraction (DAR). We present results from two on-sky pathfinders intended to characterize longterm systematic errors in MCAO systems. The first is an astrometric monitoring program of bright stars in 2013A on the GEMS LGS MCAO system. We isolate the effects of dynamic optical distortion by using narrow filters to counter atmospheric refraction and observing sparse fields to avoid crowding errors. Initial results on stable astrometric calibrator stars in sparse fields (less than 10 stars per square arcminute) suggest the presence of systematic errors of at most 0.4 milliarcseconds on short time baselines of a few days. The second pathfinder is an on-sky test of the diffractive pupil concept on a 1-meter telescope at Lick Observatory, intended to calibrate changing optical distortion with diffracted light from the target star (Guyon et al. 2012). We show that the diffractive pupil stabilizes the instrumental distortion to at least the precision of the experiment (~1 mas over 3').