Sorting projectiles by weight......

[eric n]

You would be much better served taking a class and learning proper fundamentals. That would put you on your way toward sub moa much faster then sorting bullets.
After a lot of practice with proper fundamentals, you will know when you need to sort.

[Whiterabbit]

How are you shooters using this system interpreting the data?

FOR NOW, I (firstly) only interpret data within a single day, never across days. Then, I calculate group size and SDx and SDy, separating elevation and windage. For interest, I calculate the average, but there is no weight on it.

I put the most weight on SDx and SDy, less weight on group size. average SD is of interest.

Any conclusion has to be weighted on more esoteric terms:

Were the groups fired representative of the what was desired to be tested? How did I feel that day while shooting? can I be confident in the raw data? Is there enough raw data to not be concerned about shooter-induced flyers? Should I choose to include outliers in the data? Should I include them? Should I include everything including fliers? How many "repeat" datapoints did I take? If I am taking data across days, I need a "baseline", does the baseline match relatively from day to day? Did I fatigue through the strings, yielding worse performance as the day progressed?

I interpreted your question as a "what do you do with the numbers", but we probably both know that straight numbers can be misleading without a confidence. At this time, I do not calculate a numerical statistical confidence interval. I use the questions above to guide me to decide how much confidence to put in the numerical conclusion.

--------------

In the future I plan to try to input the data into JMP and model the real world performance based on inputs of powder charge and bullet seat depth, and see how many outputs (vertical variability, horizontal variability, overall group variability, velocity variability) I can accurately model in the profiler.

In that case, I will still not calculate a numerical Confidence Interval. In that case, I expect to generate a test model based on lots of quadtratics that will end up being linear. In every case where that happens, the test that defines the quadratic becomes a repeatability, which should make the resulting profile much stronger.

That's a *should*, but it's actually a "small should", not a big one.

OR, I find a ton of quadratics and I just have to "waste" another box of bullets repeating the test to get more data into the model

---------------

For me Lynn Jr, this is just for fun. if it ceases to be so, I missed the point!