FWIW, thought experiment, here are some results of changing the case capacity in tenths of a grain under simulation while keeping everything else the same. In this case, H4350 as the powder and we change the case capacity (not case weight) for 6.5 creedmoor example round:
50.230 Capacity Velocity 2702
50.330 Capacity Velocity 2699
50.430 Capacity Velocity 2696
50.530 Capacity Velocity 2694
50.630 Capacity Velocity 2691
50.730 Capacity Velocity 2689 (load 40.3 grains powder)
50.830 Capacity Velocity 2686
50.930 Capacity Velocity 2684
51.030 Capacity Velocity 2681
51.130 Capacity Velocity 2679
51.130 Capacity Velocity 2676
With everything else being the same, a case capacity variance that has even distribution across 1 grain "could" introduce ~8 fps spread within the first std. deviation and extreme spread of 26. However, there are many other variables which will contribute (not in a non-linear/additive fashion) to the actual std. dev. so whether such a distribution is relevant is hard to know. Some of the variables such as neck trim, tension, eccentricity may have a lessor or greater effect for which there are an equal number of theories and opinions. Add to that, the error in simply measuring the h2o case capacity is probably in the range of several tenths.
I'd also point out "could" since the distribution is not likely to be even and you'll end up with a better std. deviation than this example illustrates. For instance, if we have a lumpy distribution around the center,
2702 |
2699 |
2696 |
2694 |
2691 |
2691 |
2689 |
2689 |
2689 |
2689 |
2686 |
2686 |
2684 |
2681 |
2679 |
2676 |
The std. dev. in this population is closer to ~6.7. Again, keeping in mind, that this is not additive to other sources of error.
Personally, I have yet to cull brass based on either weight or capacity but will likely experiment with this in the future (am using decent brass). The items I have played with include consistent trim length, flash hole consistency, annealing the necks, neck thickness, seating depth, and runout within ~0.001. Then add consistent charge. This gets this particular load at or below ~10fps std.dev. (measured) over several dozens test rounds.