The default rounding method is round-to-even, so e.g. 0.9999999 would be expected to come up slightly more often than 0.99999994 (the last two floats smaller than 1).

Being more precise would probably require thinking about all the possible int values.