Research › Range tests

Did the primer matter? A real comparison

GroupLab project, tested by Alan Hayes, researched and written with Claude. Written 2026-09-22; the data is from 2026-09-20.

The test

Alan loaded ten rounds of 6mm Creedmoor identically except for the primer: five with Federal GM205MAR, five with CCI BR-4. He fired the GM205MAR rounds at bulls 1 to 5 and the BR-4 rounds at bulls 6 to 10, one shot per bull, at 100 yards.

This is the comparison every reloader makes, and it is the one where it is easiest to fool yourself.

What happened on the paper

The BR-4 shots did not land on bulls 6 to 10. They landed a whole row low, on row 3, and one of them landed a long way from everything else, left of bull 21. That shot is not a flyer to be deleted: Alan fired it, it went where it went, and it belongs to the BR-4 group.

Nine of the ten shots are on the sheet. One is not, which is itself worth saying rather than quietly analysing nine and calling it ten.

What GroupLab says

Three things in that table are worth more than the numbers.

The BR-4 mean radius is withheld. Four shots is not enough to state a group size, so GroupLab does not state one. It would be easy to print 0.2 something and let you read it as a measurement. A figure with four shots behind it and no warning is worse than no figure, because you will compare it with the five shot number above it as though the two meant the same thing.

The centres differ and that is a real result. p = 0.001 means a shift this large would almost never happen by chance if the two primers put their shots in the same place. It matches what Alan could see standing at the bench: the BR-4 shots were a row low.

The dispersions do not differ, and that is not a result at all. p = 0.434 means the data cannot tell the two apart. It does not mean the primers group the same. Five shots against four can only detect an enormous difference, so failing to detect one tells you almost nothing. GroupLab says "no evidence either way" rather than "no difference", and those are different sentences.

The caveat that makes the number smaller than the truth

The BR-4 shots are measured from the bulls they landed on, row 3, not the bulls they were aimed at, 6 to 10. A shot's offset is the distance from its own bull, so measuring from a nearer bull makes the shift look smaller than it was.

So the p = 0.001 shift is real and it understates. The true shift is about a whole row of bulls, and to get that number out of GroupLab somebody has to move those four shots onto the bulls they were aimed at, by hand, one at a time.

That is a gap in GroupLab, not in the ammunition, and it is one we are fixing: telling the software which bulls you aimed at, so a whole row of shots can be assigned at once.

What you should take from this

If you change one component and the point of impact moves, you can often see that with very few shots, because a shift is a difference of means and means settle quickly.

If you change one component and want to know whether it groups better, you need far more shots than you think, and nine is not close. The honest answer from this test is: the primer moved the zero, and we do not know what it did to the group.

Do not let that disappoint you into reading the p = 0.434 as "they are the same". It says "we looked, with nine shots, and nine shots cannot see it".

Sources

1. The sheet and its ground truth: docs/NOTES-FROM-PLANNING.md, entry 120.
2. The analysis as GroupLab produced it: docs/PHASE1-RESULTS.md, \"Entry 120\", section 7.
3. The statistics behind the two tests: docs/STATISTICS.md sections 9.2 and 10.