Research
Research
What we have measured, on real targets, with the numbers and the data behind every claim.
Every figure says how many shots it rests on. Nothing here is called proven that the shots cannot prove, and where we do not know, it says so.
On this page
Reading targets#
Why a photo cannot tell you your bullet's size
The same holes measure 0.90 to 1.45 times the bullet depending on the light they were photographed in. Scans of the same sheets agree within three percent.
One hole or two?
Two shots through nearly the same spot leave one ragged hole. Deciding whether a mark is one shot or two is the single measurement that decides whether your group is real, and for a long time GroupLab got it wrong on photographs.


Curled, angled and wrinkled paper
A target you have laid flat on a bench is not flat. Measured against thirty known points, a sheet that looked flat was bent by two thirds of an inch, and that bend is larger than the groups people argue about.
A bullet hole is not the bullet
A .308 bullet does not leave a .308 hole. Paper stretches and springs back, so the hole measures smaller than the bullet, and how much smaller depends on things nobody records.
Zeroing on a blank sheet with a hand-drawn cross
You do not need a printed target to zero a rifle. You need an aim point and a known distance on the paper, and a flatbed scanner already knows the second one.

A printed number is not a bullet hole
On a clean store-bought target, GroupLab's hole finder marked the printed ring numbers as shots. Telling a bold 6 from a torn hole turned out to need two measures, because the obvious one also throws away real holes.
Scale markers beside a target you did not print
A store-bought target has no printed codes for GroupLab to measure against. Four things placed beside it can give the scale instead: a measured board, a scale bar, corner brackets and a bank card. What each gave on computer-made photographs, and why none of it counts as proven until it has been photographed on real paper.
How far off square can a photograph be?
GroupLab corrects for a photograph taken at an angle, and refuses one past 37 degrees. What the measurements say about angle, what a wide phone lens adds on top, and what is still unmeasured: whether a narrower lens from farther away does better.
Recognizing a store-bought target, and the scale you still check
A store-bought target can be recognized from a photograph without naming the wrong one, but only once its colors are checked as well as its shapes. Its scale is good on the sheet it was learned from. Whether your sheet was printed the same was never measured, so GroupLab says so.
How GroupLab reads a target
From a photograph to a group size, in six steps, with what each one can get wrong and how you would know.
When shots land on the wrong bull
On a 25 bull sheet, a rifle that is not zeroed puts every shot nearer a bull it was not aimed at. The group you get is then real, precise, and about nothing.
Scans against phone photos: how close is close enough?
We paired 59 range photographs with their flatbed scans and measured the gap. A hole in a photograph sits about 0.03 in from where the scan puts it, which is a fifth of a decent group.
Measuring groups#

How many shots do you need?
A five-shot group can be almost half or nearly double your rifle's real dispersion. How uncertainty shrinks as you add shots, and how many rounds it really takes to show that one load beats another.
Pooling groups: when two sheets are one load
Forty shots tell you far more than twenty. But two sheets shot an hour apart have two centers, and which center you pool them around decides whether you are measuring the ammunition or the afternoon.

CEP 50 and 90 explained
Circular error probable is the radius of the circle that holds half, or nine in ten, of your shots. What CEP means, how it relates to mean radius, and what happens when your group is not round.

Velocity SD from 5, 10 and 20 shots
A five-shot chronograph string that reads SD 10 is consistent with a true SD anywhere from 6 to 29 ft/s. How much a velocity SD can tell you at each string length, and why extreme spread in velocity grows with every shot.

MOA, mils and inches: one group four ways
Inches and centimeters measure the paper; MOA and mils measure the angle, so they mean the same thing at any distance. What each unit is, how they convert, and why GroupLab will not show an angle until it knows the distance.

Zeroing: when to adjust and when to leave it
Adjusting your scope after every small group can make your zero worse, not better. How to tell a real zero error from ordinary shot-to-shot noise, and how many shots the decision needs.
How many shots to zero?
Landing within one click of a perfect zero takes a handful of shots. Landing on exactly the right click can take hundreds, because a true zero near the line between two clicks is very hard to resolve. The numbers, and what a one-shot zero can honestly say.
Point blank range, honestly
The classic maximum point blank range keeps the bullet's path inside the animal's vital zone, not your shots. At the top of the band and at its far end, about half the shots miss it. A band built on the chance of a hit instead, how much a field position shortens it, and the case where no single zero is enough.

Mean radius or extreme spread?
Extreme spread is the number everyone quotes and it is measured from only two shots. Why GroupLab leads with mean radius, and when the difference starts to matter.
Does your rifle group as well as Top Gun says it should?
Top Gun is a trend of how small a rifle and load should group, from the recoil it makes and the rifle's weight. One or two groups cannot tell you whether you beat it or fall short. Here is how many shots can, and what to check before you blame the rifle.
What a canted scope costs at distance
Tilt the rifle a few degrees and the elevation you dialed pushes the shot sideways and a little low. At 300 yards it hardly matters; at 1000 it is a miss. And a scope's own travel decides how far you can dial at all, which a canted rail changes.
When a click is not what it says: the sight scale factor
A scope that moves 0.98 MOA when it says 1 is off by 2 percent. That is normal for many scopes, harmless at 300 yards and a miss at 1000. How a tall target test measures it, and what GroupLab does with the number.
How much of your vertical is your velocity?
A velocity SD of 10 ft/s adds two hundredths of an inch to a group at 100 yards and three inches at 1000. So a 100 yard group cannot show you your velocity, and at long range a good rifle's vertical can be half velocity. What that means for chasing a low SD, and how to measure it.
Where a first-round miss comes from
At 300 yards a first-round miss is mostly the rifle. At 600 and 1000 it is mostly the wind call, and with a guessed distance the range estimate too, while the velocity spread costs under one hit in a hundred and the air about one at most. What that says about where to spend practice, and when to correct off a single shot.
Range tests#
Wind or rifle?
A group wider than it is tall looks like wind. Most of the time it is not wind, it is what a handful of shots from a perfectly round rifle looks like, and telling those apart needs a test rather than an opinion.
Did the primer matter? A real comparison
Five shots against four, two primers, one rifle. The point of impact moved and the group size did not, and only one of those two statements is worth anything.

The range test log
Every range and bench test behind GroupLab in one place: what was shot, with what, what it was meant to find out, and where the results are written up. Updated after every trip.


Did the suppressor move the point of impact?
Two 25 shot sheets from one rifle and one load, one with each suppressor. The second suppressor's shots centered about a quarter of an inch lower at 100 yards, and the test just sees it.

Can you see the bull? Aim points and optics at 100 yards
Through a top-tier 36x scope the center of GroupLab's original bull was visible at 100 yards; at 10x nobody could center on it. The arithmetic of why, nine candidate aim points, and what two shooters scored through four scopes.
Guides#

How to photograph a target so it measures well
GroupLab places a photographed sheet to within a few thousandths of an inch, but light can still fool it about hole size. What 176 holes on real range photos showed, and a short checklist for taking a photo that measures like a scan.

Does your printer print at true size?
One wrong setting in the print dialog shrinks a target by about 6 percent, and every group on it with it. How print scale affects your numbers, how GroupLab measures it from a scan or a one-time printer check, what one thermal printer did across and along its feed, and a thirty-second check you can do with a ruler.

Scanner traps: cropping, DPI and color
A flatbed scan is the most accurate way to get a target into GroupLab, but a few default settings can quietly spoil it. Cropped edges, black-and-white document mode, heavy compression and made-up resolutions, and how to avoid each.
How GroupLab is built#
How GroupLab updates itself safely
An updater that can be tricked into installing someone else's file is worse than no updater. Here is what GroupLab checks, what it cannot promise, and the bug that stopped every installed build from updating at all.
Choosing the markers
The little black squares on a GroupLab target are AprilTag 36h11. The choice came down to one number, and the weaker options were measured failing on real target artwork rather than rejected on theory.
Designing a target GroupLab can read
Everything printed on a target sheet is something a hole detector might mistake for a hole. The design is mostly about telling the software what is ink before it has to guess.
What GroupLab sends from your computer or phone
Five things leave your computer or phone, all of them only when you ask. Here is the complete list, what is in each, and the rule that stops a sixth appearing by accident.
Nightly builds, from commit to installer
Every change that passes its tests becomes an installer you can download within about half an hour. Here is what happens in between, and the three mistakes that shaped it.
Cutting the installer from 97 MB to 81 MB
A third of what GroupLab shipped was debug symbols for two libraries that nothing would ever read. Finding it was an accident, and the accident is the point.
What a thermal printer is sent: the Phomemo M834, decoded
The Phomemo app printed a Letter sheet 5.3 percent small and cut its last line in half. How GroupLab recorded what the app sends to the M834, rebuilt it, found that the printer feeds about 0.7 percent short, and stretched its pages to match. Printing from a computer is built but has not been tried on a real printer.
Registering a sheet whose far margin has lifted
A lens fit assumes the paper is flat. When one margin lifts toward the camera it cannot follow, and the holes beside it read up to 0.08 inch out. How GroupLab lays a smooth correction over the lens fit, when it takes it, what it fixed on two pictures, and the three places it cannot reach.
OpenCV's marker and QR detectors on Android, from .NET
In September 2026 no published runtime gave a .NET application on Android OpenCV's ArUco markers and WeChat QR detector. Building OpenCvSharp's native half yourself is one script, and two of its traps cost a phone run each.
Every upload is rebuilt from pixels
A photograph you send is stripped of everything except the camera and the exposure. Not edited, not blurred: the metadata around the image is thrown away and written fresh, and the pixels are copied untouched.