It's parsed as - 0 .0 ÷ 2.0. The leading - becomes the chain's argument and we evaluate 0 .0 ÷ 2.0 monadically, ignoring the argument of -1
So it computes 0 ÷ .0, slightly less crazy :P
2.0÷-0.0 is parsed as 2.0 ÷ - 0 .0, so we calculate 2.0 ÷ - (-2.0), then output it when we encounter the unparseable nilad 0. We repeat that again with .0, then output 0.0 at the end, thus outputting [-2.0][0][0.0] (brackets added for visualisation)
CMC: Without running them, what do -0.02.0 and 2.0-0.0 output?