Three Numbers on Every Game: How CrystlQuant Works
Every game on CrystlQuant carries three numbers: the market's price, our forecast, and a blind model that never sees the price. All three are published before the game and graded after it, misses included. Here is how to read them, using this Sunday's NFL games.
CrystlQuant is the sports forecasting site we build at Crystl Labs. It covers the NFL, NBA, NHL, MLB, college sports, the Premier League and the big European leagues, UFC, tennis, cricket and more. It is free. Every page is built around one idea: a forecast only counts if it is written down before the game and checked after it.
Three numbers, not one
Most sites give you one number per game. CrystlQuant gives you three, side by side.
- The market. What the price says, turned into a plain chance to win. Thousands of people with money on the line set it, and it is hard to beat. We show it so you can see where we agree and where we do not.
- CrystlQuant. Our main forecast. It starts from the market's price and moves it toward what our model sees. Most weeks it backs the same team as the market; its job is to get the percentage right.
- The blind model. It never sees the price at all. It works only from the teams and players themselves. It disagrees with the market more often, so it wins bigger and misses more.
When all three agree, there is no story. When they split, that is the game worth reading about.
Reading this Sunday
Three NFL games this week show what the numbers say (as published on Thursday 29 October):
| Game | Market | CrystlQuant | Blind model |
|---|---|---|---|
| Ravens at Bills | Bills 61% | Bills 61% | Bills 79% |
| Vikings at Lions | Lions 63% | Lions 62% | Vikings 50.1% |
| Raiders at Jets | Jets 53% | Jets 53% | Raiders 54% |
In Buffalo all three take the Bills, but the blind model is 18 points more sure than the market. In Detroit the blind model goes the other way: the Vikings are 6 and 0 and still the underdog, and it takes them. In New York it takes the Raiders outright. Every game gets a side. When a forecast sits near 50, the site shows the decimal and names the team it points to.
Published first, graded after
Every forecast goes up before the game starts and cannot be changed once it is there. After the game it is graded in public on the Record page, with the wins and the misses. You can check how each of the three numbers has done, sport by sport, this season and all time. As of 29 October, the blind model has picked the winner in 66 of 105 graded NFL games this season.
Keeping the misses is the point. A site that only shows its wins is telling you a story. A record with the misses in it is telling you what to expect.
The Lab
The Lab is the part of the site for questions that are not on the schedule. "What if they fought" prices any two active UFC fighters, booked or not. Eras pits team seasons from different decades against each other: the 2024-25 Thunder against the '96 Bulls comes out 50.04% to 49.96%, Bulls by a hair. None of it is graded, because none of it will be played. It is there to argue with.
Why we built it this way
Forecasts are easy to make and easy to forget. Writing them down first, next to the market's own number, and grading them in public is what makes them worth reading. That is the whole site.