The first pointing update fails for wrong coordinates. Probably the simulated coordinates are not being updated early enough.
Found this today while looking at a different problem. Fold it into the 3.6b release of Scheduler.