That scientific rounding would have no significance whatsoever, as by the definition of rounding to the nearest half unit, a number like N.5 (N being a natural number) would stay as it is, it already is a multiple of .5.
You could apply scientific rounding on the double value (in a case like 47.25, scientific rounding applied to 2*47.25 = 94.5 would round down to 94, yielding 47 instead of 47.5 as the end result). Dave Higgins has to decide whether that should be handled this way. It's not hard to do also, just check bit 0 of Floor(n*2) and if 0, use Floor instead of Round. But I don't think this is asked for.
The resoning just is making the rounding errors point up and down evenly, not having a tendency for adding to values, especially if the numbers you round just have one significant digit, eg if you measure one digit only, as that makes +.5 in 10% of all cases. As soon as you have values with several digits, these special cases don't contribute as much to an overall tendency anyway. Eg rounding prices that way, having 2 digits in most cases, you'd only have a tendency of 1% more than averagely adding instead of subtracting, if your original numbers are having even more digits to them, the tendency effect is neglectible.
Bye, Olaf.
