Agreed. Is the expectation that the numbers are actually integer values and double is being used only to allow really big numbers? I tried truncating each sum to remove the fractional part and got a failure where the expected value was 0.5 and the actual value was 0.0, so it would seem the numbers are not always expected to be integer values. However, it seems like the random tests will succeed sometimes and fail other times because of rounding differences. Is there some level of this problem we're not seeing, or are the rounding errors an unexpected issue in the problem?

Yeah, not similar code variations. Code exact-same-ations.