70% means it loses three times in ten
A 70% chance to win is not a prediction that the team wins. It is a claim about many games like this one: out of ten of them, the favorite should win about seven and lose about three. A single loss does not prove the number wrong, and a single win does not prove it right.
- A 70% call loses at least once in five games 83.2% of the time.
- Three 70% calls all lose 2.7% of the time: rare, but over a season of games it will happen.
How to tell whether the percentages are honest
Group the calls by the chance they were given and count. If the 70% calls won about 70% of the time and the 55% calls about 55%, the numbers are calibrated: they meant what they said. This needs dozens of calls in each group before it says much; ten games prove very little either way.
A win-loss record cannot show this on its own. A site that only calls heavy favorites will post a high win rate while telling you almost nothing you didn't already know from the odds.
Scoring a forecaster: the Brier score
The Brier score is the average squared gap between the stated chance and what happened (1 for a win, 0 for a loss). Lower is better. Take two forecasters who each call ten games and each get seven right:
- One said 70% every time. Seven of ten is exactly what 70% means, for a Brier score of 0.210.
- The other said 90% every time. Same seven wins, but the three losses were claimed to be near-impossible, so it scores 0.250.
0.250 is what calling every game a coin flip scores. The overconfident forecaster, with the same hit rate, did no better than that.
What CacheOut's percentages are
Each CacheOut call is the market's price with the bookmaker's margin taken out: what the betting market as a whole expects, not a secret model. So the record measures how good the market's read was, and a calibrated record is what you should expect, not a sign of an edge. The record page shows the win-loss count and the Brier score for each sport, misses included.
Not betting advice. 21+ where permitted.