The rankings on The Record now run on Record Rating, a skill model built from every stage you shoot instead of from match percentages. The full write-up, with live numbers, is at zumbroriver.com/method
How it works
- Reads every stage. It uses the hit factor on each stage, and measures how hard that stage was from everyone who ran it. One dominant winner no longer makes a field look strong.
- Forgives blow-ups. A miss or a jam counts for less than a bell curve would give it. A stage zeroed outright is counted separately and subtracted, because those points really are lost.
- Follows your form. This year's results outweigh last year's.
- 100 means a winner. A rating of 100 is the skill expected to win an average field in your division.
- Shows its uncertainty. Each rating has an interval of about 4 to 6 percent either way, and the ranking uses the low end, so a short record has to be exceptional to outrank a long one.
Does it predict better?
Every night the model is fit on data up to a year ago, then asked who finishes ahead in every head-to-head since. Share of those calls it got right:
| Division | Plain average | Previous rating | Record Rating |
|---|---|---|---|
| Carry Optics | 80.7% | 81.3% | 83.4% |
| Limited Optics | 81.7% | 82.7% | 84.9% |
| Open | 82.9% | 84.5% | 86.8% |
| PCC | 82.3% | 83.9% | 85.4% |
| Production | 83.7% | 84.0% | 84.5% |
In Carry Optics that is about 8,000 more correct calls out of 386,761, and resampling whole matches puts the gain at 1.9 to 2.3 points: well clear of luck.
Caveats
- It knows nothing about equipment, weather or squads.
- A mistyped member number turns one shooter into two.
- About 96,000 entries that clubs filed under abbreviations like LTD, PROD and CO are being folded in now, so some ranks will move.
Tell us
- What should the rating weigh that it doesn't?
- Should ranks follow the low end of the interval, or the rating itself?
- Find your division at zumbroriver.com/rankings: where does it look wrong?