Hmm. Interesting. I just tested out your script with a number of different methods of converting uint64s to doubles. The method used in the beta was recommended in a number of places and is even used in a well-known RNG. However, it turns out that a different method actually resolves the issue you are reporting. The different method results in an equal distribution of odd and even numbers and a good distribution overall. I will be changing to this method in the next beta, so let's see how that works out. Thanks for the test script!
