Not to revive a dead horse just to beat it to death again, but here I go...
I have read through the entire thread, and I think I have something to offer. Like many others I was very disturbed by Bryan Litz's chapters (Modern Advancements Vol 3) on rifle accuracy and small sample testing (which by the way came out well before Hornady's podcast). I was just about to begin development on my new barrel, a 26" Benchmark chambered in 284 win (Tikka platform), and this rocked my world.
I used to consider my load dev method a 5x5 on powder in 1% increments, then pick best powder based on group and poi trend, do a 5x5 on seating depth, pick best combo based on group/poi trend, then load up 25 for verification and MV.
Once I read the book, I was thrown into a world of doubt. So I decided to do some testing myself. I grabbed my T1x and went to 50yds and shot 19 10 shot groups over the course of a couple days, and began crunching numbers. This activity resulted in the video below, an interview with Keith of Winning in the Wind, who is a local F class champ and fellow aerospace engineer.
Now I understand not everyone wants to sit through over an hour of two engineers rambling (though I think it's worth your time, obviously biased). So I'll put some cliff notes here of the relevant findings from my testing and how it influenced my load development method.
Actually before I do that, let me say a little bit about load development. I am a structural dynamicist by education and profession, meaning I model and predict structural dynamic loads, deflections, and accelerations. ALL elastic structures have modes of vibration which are determined by their stiffness and mass characteristics. Elasticity means that the material stretches under load. Most materials we interact with on a daily basis are elastic, including the metal of your barrel. The material and contour of your barrel determine it's stiffness and mass distribution, and therefore it's vibrational frequencies and shapes (modes). For a free floated barrel, this can be considered a cantilevered beam in engineering terms, and the stiffness or rigidity of the barrel attachment to the action/stock/shooter/ground system also influence its modes. This page explains the math
https://endaq.com/pages/bernoulli-euler-beams, see animations of modes here
https://en.wikipedia.org/wiki/Euler–Bernoulli_beam_theory.
So when I talk about load dev, I'm talking about adjusting bullet time such that the bullet leaves at an extreme of barrel motion, where velocity is close to zero (at least for the largest amplitude modes), and small variations in bullet time don't produce large variations in bullet exit angle. I DO NOT believe that certain powder charges posses magical better MV properties - MV is linear with powder for the useful charge range. MV ES and SD are a function of your components and your reloading consistency.
OK now on to the critical findings:
- Impact radius is far more useful and meaningful than group size. You get more data for the same set of shots, converge to the "correct" answer faster, and characterize the system more accurately. Learning to work in terms of mean and SD radius is game changing.
- Using radius, you can continuously "overlay" more and more shots, so as you go and practice and shoot your baseline load, you can end up with a database of 100s of shots informing your mean/SD radius stats. This is like having a group size of 100s of shots, only much easier and more informing.
- Small samples (3-5 shots) CAN be useful for assessing differences between populations, provided the differences are large (50-100%). As the differences become smaller you need more samples to tell them apart. Due to the large samples required to fine tune a load, it's probably not worth doing. Assessing major deltas across component combinations and broad sweeps of powder is most worthwhile, and once you get below your required accuracy threshold, characterization is more valuable.
- The "T-Test" can be used to statistically assess whether populations of data are actually different. You need a mean and standard deviation of each population to do this.
My load dev process after all this is as follows, and keep in mind I'm always working in radius and logging every shot as I go:
- Load up 5 shots for each bullet/powder combo I want to test (I don't mess with primers until I look at MV stats). Shoot and look for best combo by a large margin. If all similar pick my preferred and move on.
- Load up a 5x5 broad powder sweep, 1-2% powder increments. Shoot and look for best combo by a large margin. If all similar pick my preferred and move on.
- Validate - load up 20-30 and shoot for groups and MV. Note that these groups will later be overlaid via radius methodology, but you get 4-6 5 shot group agg for comparison to that metric and other rifles. At the end of validation you will have between 20 and 40 shots informing your mean and SD radius stats, and your mean and SD MV stats.
- If at the end of this process (~80rds) you don't have a load you like, start over with new components. If you do like it, start shooting and practicing with it and log all your data as you go to increase your statistical basis.
It's worth noting this is about the same number of shots as before, but I am using different metrics and looking for different information. I'm also making much more significant changes in each set of testing.