Beyond the Metric. Human performance measurement: how much confidence a number actually earns.

Goals saved above expected, season to season

r = 0.14

If this measurement repeated, the dots would sit on the diagonal. They are the cloud. We measure how much confidence a number actually earns before you make a decision on it.

The more isolated metric
repeated less.

In one of our analyses, the statistic built to replace save percentage turned out to be less consistent from one season to the next than the one it replaced. The more isolated metric repeated less. We test measurements on real data at real scale, publish what fails beside what holds, and build tools on what survives.

One question, every sport, every domain

Reliability

Does the
measurement repeat?

A goaltender's save percentage. A pitcher's spin rate. A striker's finishing rate. A combine vertical jump. If a number does not repeat well enough for the decision at hand, the decision needs more than the number.

Applies toNHLMLBSoccerCombineClinical

From the work: The .930 Mirage · The GSAx Paradox

Selection

Does the screening
decision predict?

An NHL draft board. The NFL combine. An NBA lottery pick. A tryout cut. Every selection is a screening test, and screening tests have a known mathematics of being wrong.

Applies toNHL DraftNFL CombineNBATryoutsOccupational

From the work: The Screening Test

Inference

Does the outcome
justify the verdict?

A playoff series. A World Cup shootout. A major final. A season award. Outcomes arrive one at a time, and one outcome is almost never enough to conclude anything.

Applies toNHLWorld CupTennisAny tournament

From the work: The Colorado Problem · The Momentum Illusion

Physiology

Does behaviour reveal
an internal state?

Shooting time under load. A driver holding 5g in traffic. A rider's third week in France. A HYROX pacing collapse. The body leaks information through behaviour, and most of it is fainter than it looks.

Applies toBiathlonF1Tour de FranceHYROXMilitary

From the work: The Autonomic Frontier · The Lost Step

Structure

Does the pattern
predict the outcome?

Shot entropy. Possession structure. A bullpen's usage pattern. Concentration of chances. Structure is real and measurable. Whether it predicts anything is a separate question, and we test it.

Applies toNHLSoccerMLBNFL633,318 shots

From the work: The Entropy Series · The Restoring Force

The evidence

NHL

The Lost Step

A veteran loses a step, scouts say. Five seasons of NHL tracking data show top speed holds and distance per game does not.
9 min
NHL

The Momentum Illusion

Carolina scored three goals in 39 seconds and the whole sport called it momentum. We checked 477 playoff goals to see if momentum is real, or just memory.
11 min
NHL

The GSAx Paradox

The analytics community's preferred goaltending metric is less reliable than the one it was designed to improve. 550 consecutive-season pairs across 18 years explain why.
7 min
NHL

The .930 Mirage

1,869 goalie-seasons. 18 years. The year-to-year correlation for save percentage is r = 0.33. Only 11% of the variance is real. The rest is a mirage.
9 min
All analyses →