SportsAdvanced· 6 min read· 4 steps
Calibrate a sport adapter
Run any sport adapter and confirm its observed win rates land inside the Wilson interval around the model probabilities.
- 1
Run the adapter
Each sport uses a scoring model tuned to its shape — Poisson for low-scoring count sports, Gaussian for high-scoring ones. Run it to produce the four-probability table.
- 2
Open the statistics page
The sport's statistics page adds a Wilson 95% interval, a proportion z-test, and a chi-square goodness-of-fit across participants.
- 3
Judge significance, not size
A raw gap is not enough. A deviation only counts when it clears both statistical significance and a minimum sample.
- 4
Re-seed to confirm
Reload with a different batch seed. One-off flags are usually noise; deviations that survive re-seeding are the interesting calibration story.
Key takeaways
- Low-scoring sports use Poisson; high-scoring use Gaussian.
- Significance needs both a p-value and a sample floor.
- Real signals survive re-seeding.