A Brier score grades a set of probability forecasts by squaring the gap between the number you gave and what actually happened, then averaging those squares across every call you made. Zero is perfect, 0.25 is what you score for shrugging and saying 50% every time, and lower is better. It answers the question a win rate cannot touch: when you say a team is 65% to win, does that number mean anything, or is it just a feeling with a percent sign on it?
How a Brier score is calculated
Meteorologist Glenn Brier published the measure in 1950 to grade rain forecasts, and the arithmetic has not changed since. Write each outcome as 1 if the thing happened and 0 if it did not. Subtract the outcome from the probability you gave, square the result, and average across every forecast in the set.
Say 80% and the team wins, and the squared error is 0.04. Say 80% and the team loses, and it is 0.64. The squaring is the entire design. A confident call that misses costs sixteen times what the same call costs when it lands, so bluster gets punished far harder than caution. That is also why the measure only works on a body of forecasts. One bet produces a number, but it tells you nothing.
What counts as a good score
Two reference points anchor the scale. Saying 50% on everything scores exactly 0.25 no matter how the games break, so 0.25 is the line between a forecast worth reading and noise. A score of 0 means every call was made at 100% and every one of them landed, which nobody manages in sports.
Here is the part most explanations skip. The best score available to you depends on which games you rate. An honest forecaster facing a genuine coin flip cannot do better than 0.25 in the long run. On a true 60% game the floor is 0.24, and on a true 80/20 mismatch it is 0.16. Someone grading blowouts will always look sharper than someone grading toss-ups, even when both are perfectly calibrated. So a Brier score is only ever a comparison, never a grade on its own. Score two forecasters on the same slate of games or you are measuring the schedule instead of the skill.
Run your squares on BettorEdge.
Skip the paper grid and the group chat math. Set up your Super Bowl squares with friends on BettorEdge and let it settle up automatically.
A worked example: five calls, one number
Take five sides you rated last week, with the result of each. Square the gap, then take the average.
Shorter bar, better forecast. Three of five landed, but the score comes from how far off each number was, not from the record.
The five squared errors are 0.063, 0.360, 0.203, 0.160 and 0.090. They add to 0.875, and dividing by five gives a Brier score of 0.175. Against the 0.25 baseline that is a skill score of 0.30, which is the standard way to express the improvement: 1 minus your score divided by the baseline score. Notice that the single 60% loss contributed more than the other four calls combined. Five forecasts is far too small a sample to trust, which is the honest reason most bettors should score a season rather than a week.
Score yourself against the market, not against a coin flip
The 50% baseline is a low bar. The benchmark that decides whether your opinion is worth acting on is the price, so rate the same games the market rates and score both sets of numbers side by side. Convert every posted price to a percentage first with the betting odds converter, then score the market's number exactly as you scored your own.
One correction has to happen before that comparison is fair. A sportsbook's two prices do not add up to 100%. Both sides at -110 imply 52.4% each, which totals about 104.8%, and that extra 4.8% is the fee rather than a claim about the game. Scoring those inflated numbers is scoring the margin along with the forecast. Strip it out with the no vig calculator and score the fair prices underneath. If you want the arithmetic behind that step, implied probability covers it.
Why a good Brier score is not the same as a profit
The measure is what statisticians call a proper scoring rule, which means your best expected score comes from reporting exactly what you believe. Suppose a game is truly 60%. Saying 60% gives you a long-run score of 0.24. Talking yourself into 70% on the same game pushes it to 0.25. Inflating your confidence always costs you, which is precisely why the measure is worth keeping.
What it will not tell you is whether you made money. Profit comes from being right where the price is wrong, so a forecaster with a mediocre overall score who finds three badly mispriced games a month can beat a beautifully calibrated one who agrees with the market on everything. Track both. The Brier score tells you whether your probabilities are honest, and the closing price tells you whether they were worth betting.
The short version
Square the distance between each probability you publish and the outcome, average them, and compare the result to 0.25 and to the market's own number over the same games. Below 0.25 means your percentages carry information. Above it means they do not, whatever your record says.
Brier score FAQ
What is a Brier score in betting?
A Brier score measures how accurate a bettor's probability forecasts are by averaging the squared difference between each forecast and the actual outcome, scored as 1 for a win and 0 for a loss. It runs from 0 for a perfect set of forecasts to 1 for a maximally wrong one, and lower is better. Unlike a win rate it grades the number attached to a pick rather than the pick itself, so a 60% call that loses is penalized less than a 90% call that loses.
What is a good Brier score?
Anything below 0.25 beats forecasting 50% on everything, which is the standard reference point, and the benchmark that actually matters is the score the market's own prices post over the same games. There is no universal pass mark, because the best score available depends on how lopsided the games being rated are: the floor on a true coin flip is 0.25, while on a true 80/20 matchup it is 0.16. Only compare scores calculated over the same set of games.
How is a Brier score different from accuracy or a win rate?
A win rate counts how often the side you picked came in, while a Brier score measures how well your stated probabilities matched reality, so it rewards saying 55% on a close game instead of dressing it up as 80%. A bettor can post a strong win rate on heavy favorites and still score poorly by being overconfident on every one of them. The two answer different questions, and neither one on its own tells you whether the bets were priced well enough to profit.
Scoring yourself is the easy half. The harder half is getting a price that does not have a fee baked into it before your forecast even starts. On BettorEdge community odds you name your own number and another real bettor takes the other side, so the price you are graded against is the one you set. Check your calibration, then open BettorEdge and get this price for real, peer-to-peer.
Run your squares on BettorEdge.
Skip the paper grid and the group chat math. Set up your Super Bowl squares with friends on BettorEdge and let it settle up automatically.