Hi c'
I don't know that it would be considered a bug or if it is a situation covered in the ISO standards for floating point binary numbers.
If your base data is precise (and accurate) to three decimal places, you can force all instances of n.xx5 to round up by adding a tiny 'fudge factor' to the raw data, as I've done below. This is the same table as the one above, with a small change in the formula in columns B, D, F and H. The new formula, as it appears in B2, is shown below the table.
B2: ROUND(A2+0.0000001,2)
Note that the negative values (in A2 - A8) are also rounded up. This may or may not be significant in your data set.
Regards,
Barry