I didn't realise this same question was posted in two different sections, I just thought it was a bug with the board software.
I had the same problem last night when trying to replicate it. At first I though it was 'cos I'd forgotten to reverse the normals on the bottom part which the inside of 1/2 a sphere. Now I'm not so sure if it's the order that the constraints and gravity are added, or if it's just a size/scale related thing.
In any matter, you'll find a working solution here