Hatcher reported developing a National Match load one year between the WW's at Frankford Arsenal for which he had two powders that had burn rates similar to IMR 4320. One was a short grain stick the arsenal's powder measures could throw ±0.3 grains. The other was a coarse stick, like 4064, which the same powder measures could only dispense ±0.85 grains. Nonetheless, the latter powder thrown with that wide 1.7 grain charge weight spread proved to produce consistently more accurate cartridges in machine rest firing tests and wound up being that year's NM load and several records were set with it at the National Matches.
Some powders, and 4064 seems to be one in .308 and 30-06 match loads, tend to compensate for charge error when thrown by volume. A heavier charge with closer together grains burns just enough more slowly to produce little net difference in pressure as compared to a lighter charge in the same volume.
A few years back I bought a JDS Quick Measure, which boasts it will never throw stick powder loads off by more than 0.2 grains. I have experimented with it enough to find the claim true. It uses an unusual metering chamber geometry that cannot cut grains. Grain cutting has to be avoided for best charge weight consistency not because cut grains themselves have any appreciable effect on how fast a charge burns, but because the act of cutting produces a sharp knock in the measure that tends to settle the powder reservoir near the metering cavity, causing the next couple of throws to be heavier than average. By eliminating the cutting, that settling effect is avoided.
One method that measure has allowed me to pursue is to tare a scale with a primed case on it, dispense the powder and set the charged case back on the scale to check the weight of the finished throw, and then sort the charged cases by throw weight for subsequent testing. This gives me weight and volume equality simultaneously. Depending on the powder, I will end up with three to five weights in 0.1 grain increments. I can then keep them segregated for testing or dump some back in the hopper and keep going until they are all the middle value. If I have a good load I can't really tell the difference, but when I am doing load workups, I like to eliminate all the variables I can until I settle on the load. Then I can start dropping controls to see if it makes any difference.