@dzamie @ipg I'm guessing it's because the gap between floats gets large enough or it gets past 0 in the relevant axis, so that whatever rounding error starts rounding towards the coordinate it's currently at instead of away