Two Instrument Classes, One Golf Shot
Which rows are observed, which are inferred, and why that ordering matches the band widths above.
Read the note →
Translation Log · Note 04
Tolerance Bands, Not Single Numbers
Carry 262. Spin 2,510. Launch 12.8. Three figures that look like facts, printed to a precision nobody has earned, with no indication of how much they would move if you hit eighteen more balls. This is the note the whole studio is built on.
Take any golfer, hand them one club, and have them hit twenty shots. The carry distances will not be twenty copies of one number. They will be a cloud, and the width of that cloud is a property of the golfer far more than of the club.
A scratch player's cloud with a seven iron might be eight yards wide. A fifteen handicap's might be twenty-two. Neither of them is doing anything wrong; that spread is simply what their delivery currently produces. And the width of that cloud sets a hard floor on what any fitting can detect. If your cloud is twenty-two yards wide, a component change that is genuinely worth four yards is invisible inside it, and any four-yard result you see is noise wearing a lab coat.
What Our Bands Actually Are
The band beside each column is not a manufacturer's specification and it is not an instrument accuracy claim. It is derived from your own sample, in that session, on that club.
The procedure is deliberately unglamorous. We take the accepted strikes, trim the top and bottom ten percent, compute the central value from what remains, and report the half-width that contains the bulk of the trimmed sample. Wider cloud, wider band. More accepted strikes, slightly narrower band, which is the entire reason the minimum is eighteen rather than eight.
Because the band is computed from your swings, it moves. A golfer who is tired at the end of a two-hour block will see the bands widen on the last candidate, and that widening is honest information about how much to trust the last comparison.
Instrument uncertainty is included but is rarely the driver. Both units carry a
published repeatability, and where they disagree the band inherits that disagreement. In practice
the instrument contribution is small compared to the human one, except on spin for slower swing
speeds where it can be a third of the total.
The Rule That Falls Out of It
Once every column carries a band, a decision rule appears, and it is the rule the acceptance stage of every block runs on.
A difference smaller than the band is not a finding. It is a coin landing the same way twice.
Here is the same comparison written two ways. The first is how it appears on a typical printout. The second is how it appears on ours.
| Reported as | Baseline | Candidate | Conclusion invited |
|---|---|---|---|
| Mean only | 158 yd | 161 yd | Candidate is three yards longer, buy it |
| Mean with band | 158 ± 6 yd | 161 ± 6 yd | No detectable difference, do not buy it |
Same session, same eighteen strikes, opposite recommendations. The first convention is not a lie. The candidate really did average 161 that afternoon. It is simply reporting a result at a precision the data does not support, and the reader has no way to know that.
Where Bands Change the Answer Entirely
The more interesting case is when the mean does not move and the band does.
Suppose the baseline carries 158 with a band of six, and the candidate carries 158 with a band of three. On a mean-only sheet those two clubs are identical and there is nothing to discuss. On a card with bands, the candidate is producing the same distance with half the scatter, which on a golf course means the difference between a green in regulation and a chip from behind.
That result is invisible under the first convention and it is the single most common genuine finding we produce. Consistency lives in the spread. Distance lives in the mean. Most golfers are sold on the mean and would have been better served by the spread.
Which Columns Carry the Widest Bands
Not all rows are equally noisy, and knowing the ranking helps you weight a sheet.
| Column | Typical band | Why |
|---|---|---|
| Ball speed | ± 0.8 mph | Observed by both units, tightest row on the card |
| Launch angle | ± 0.4 deg | Observed by the camera, mostly delivery variation |
| Face contact | ± 2 mm | Physically measured off tape with a scale |
| Carry | ± 6 yd | Modelled, so it inherits the noise of three inputs |
| Backspin | ± 210 rpm | Sensitive to strike height and to ball condition |
| Descent angle | ± 1.1 deg | Modelled from carry and spin, widest on the card |
Descent angle is a good example of why this matters. It is a genuinely useful column, it sits at the end of a chain of three modelled quantities, and its band is wide enough that a one-degree difference between candidates is meaningless. People make decisions on one-degree differences in that column constantly.
How to Use This Anywhere
You do not need our card to apply the idea. On any sheet from any studio:
- Ask how many shots each figure was computed from. Fewer than fifteen and the mean is unstable.
- Ask whether any shots were removed, and for what stated reason.
- Ask what the spread of your own baseline was. That is the floor on what can be detected.
- Compare every claimed improvement against that floor before agreeing that it is real.
None of these questions require any statistics beyond comparing two numbers. They will, however, end quite a few conversations about three-yard gains, which is the point.
Corrections: none since publication.
Related notes
Two That Build on This
When the Finding Is That Nothing Should Change
What happens when no candidate clears the band, and why that is a completed block rather than a failed one.
Read the note →See Your Own Band
The baseline stage of any block produces it inside twenty minutes, and it is usually the most surprising number of the afternoon.