The input is a number that can be converted to a length. The output is a number of ore pieces. These are not the same units. There is no reason to convert anything to "cubic blocks" as a starting point when you already have a perfectly usable number given by the size. You only need two things to determine the (average) output: the shape of the curve is that of size cubed, and the coefficient. With the added results, the current average coefficient is 2.51 and with 19 samples ranges from 2.03 through 2.88 (or 2.66 if you ignore the size 3 deposits.) Given the amount of deviation seen, if you use 2.5 you are going to be as accurate as it's possible/practical to be. It's not like we're going to home in on 2.512357325723 and notice the difference (or care) that we were getting 12 more or fewer ores out of a size 10 deposit. I mean, if you're cutting it that close, you're going to get burned by the random variation anyway.