Glossary
What the words mean
Every term on this site that carries a specific meaning, defined.
Last updated 6 August 2026.
Terms
| Term | Meaning |
|---|---|
| NAICS | The North American Industry Classification System โ the six-digit code that identifies an industry. 238210 is Electrical Contractors. Codes are revised every five years and retired codes return nothing rather than guessing. |
| SOC | Standard Occupational Classification โ the code that identifies a JOB rather than an industry. The bridge between the two is what lets us say which licence gates which industry. |
| Net formation | Establishment entry rate minus exit rate, averaged over five years. The strongest predictor of subsequent industry growth we have found: rank correlation +0.57 across four backtested windows. |
| Decile separation | The gap in outcome between the top and bottom tenth of industries ranked by a signal. Net formation separates them by 42.6 percentage points of ten-year establishment growth. |
| Backtested | Scored using only data that existed at a past date, then graded against what actually happened afterwards. A backtest that uses later data is flattering and worthless. |
| Descriptive only | A figure that is true and has no measured relationship with what happens next. Establishments per firm is descriptive: it tells you whether an industry scales to many sites, and predicts nothing about growth. |
| Covered weight | How many of a rubric's 100 points were actually evidenced. A score is computed from the covered points only, and the uncovered ones are named. |
| Brier score | How wrong a stated confidence was, squared. Lower is better. It is computed from CONFIDENCE rather than the merit score, because a score of 67 is not a claim of 67% odds. |
| Pre-registered test | A metric, a threshold and a date, recorded at the moment a verdict is issued and not editable afterwards. It is what makes a later claim of accuracy checkable rather than asserted. |
| VOID | A pre-registered test that was never actually run. It is not a pass and not a failure, contributes to no calibration in either direction, and keeps its original prediction on the record with the reason it was voided. |