What I wrote above was incorrect. Apparently InverseGaussian is not 1/Normal 
and unfortunately we don’t have one implemented. That just means we need to implement another building block (or just do everything together). Wikipedia gives a suggestion how to take a random draw from an InverseGaussian, so things shouldn’t be too bad. In that case the random method, should just be a draw from the correctly specified InverseGaussian + shift?