@michael.williams pointed out that the data consistency test is pretty unclear. It looks like there's an argument we could be using instead of the combined statement.
I also added a test for the case where one result has a nan log noise evidence but the others don't as that is a case that should fail.