More exact tests

[fedarko]

Now that we can set random seeds, it'd be good to test the output values more precisely if possible. See relevant conversation in this thread.

... This isn't very high priority at all, but might as well keep track of it