What is a good Brier score?
The Brier score is the single number we trust most to judge whether predictions are any good. Here is exactly what it means.
What it measures
The Brier score is the average squared difference between a predicted probability and what actually happened (1 for the event, 0 for not). Predict 70% and the player wins, you are penalised (1 − 0.70)² = 0.09; predict 70% and they lose, (0 − 0.70)² = 0.49. Lower is better.
It rewards being both accurate AND honest about uncertainty — confidently wrong predictions are punished hard, which is exactly what you want.
The 0.25 coin-flip line
Always predicting 50% gives a Brier score of 0.25. So 0.25 is the line a model must beat to add any value. A score meaningfully below 0.25 means the model genuinely separates likely winners from losers.
What counts as good
It depends entirely on how predictable the sport is. In tightly matched, round-the-clock table tennis, honest scores sit around 0.24 — only a little under the coin-flip, because the matches really are close. In lopsided fields they can drop further. A model claiming 0.15 in an even field would be over-confident, not brilliant.
What matters is that the score is stable over time and beaten honestly. You can see ours, month by month, on the track-record page.