Probability is not certainty
If PTN gives a team an 80% win probability, the correct interpretation is not that the team will win. It means the model believes that team would win roughly eight times out of ten if comparable conditions could be repeated many times.
Upsets are therefore not proof that a probability was useless. The real test is calibration across many predictions.
Expected margin adds context
The probability answers who is more likely to win. Expected margin estimates the size of the underlying advantage. Looking at both is more informative than either number alone.
PTN also displays the teams’ Power, Attack, Defence and Form ratings so users can see where the predicted edge is coming from.
How PTN will judge itself
The model should be evaluated over time rather than by cherry-picking individual results.
- Winner accuracy
- Mean absolute margin error
- Brier score
- Probability calibration by confidence bucket
Where PTN uses official NRL statistics, the raw inputs are identified as source data. Ratings, percentiles, model probabilities and interpretation are Play The Numbers calculations.