<- all posts
Directive 2026-11-12

CrystlQuant Agrees With the Market 95% of the Time. The Other 85 Games Are the Business

In 1,741 games since August, CrystlQuant named a different winner from the market only 85 times, and won 40 to 32. Why the rare disagreement is the product.

/ 5 min read

This hockey season, CrystlQuant has named a different winner from the market 17 times. It was never sure of itself in any of them. Its highest number in those 17 games was 51.9%, and the gap between CrystlQuant and the market was never wider than 2.8 points. CrystlQuant won 11 of the 17. The market won 6.

That small run of hockey games is the whole business of a forecasting site in miniature, and it needs explaining, because from the outside it looks backwards. You would expect a forecaster to earn its keep by disagreeing loudly and often. Ours mostly says what the market says. What it is worth lives in the few games where it doesn't.

Ninety-five percent of the time, we say what the market says

On 10 November 2026 we counted every game CrystlQuant has graded this season in six sports where the market also had a closing price: the NFL, NBA, NHL, MLB, college football and football, from mid-August to 9 November. That is 1,741 games. In 1,656 of them, CrystlQuant named the same winner as the market. That is 95.1%.

Those games are not wasted. When the two agreed, the side they both named won 1,057 times, 63.8%. But they cannot tell us apart from anyone. If you and the market back the same team, you are right together or wrong together. A forecaster that only ever agrees is a mirror, and nobody needs a mirror of a number they can already see.

The other 85 games are where it gets interesting. In each one, CrystlQuant and the market named different winners, so one of them had to be wrong, unless the game ended level. Thirteen of them were football matches that finished as draws, so neither side got those. In the 72 that had a winner, CrystlQuant was right 40 times and the market 32. CrystlQuant took 55.6% of its arguments with the best forecaster in sports.

Why we disagree so rarely

The market's price on a game is the most accurate single forecast there is. Thousands of people with real money on the line set it, and it moves the moment news breaks. CrystlQuant starts from that price and moves it toward what our model sees. Most nights the move is small and the side does not change. That is the design, not a weakness.

We know what happens when a model disagrees freely, because we publish one. They say nobody beats the market. One model never sees the market price. The other fixes it. The CQ blind model works only from the teams and players, and it argues with the market far more often: 337 times in the 2,260 priced games it had a number for in the same six sports, about one game in seven. It loses those arguments. The market was right 183 times, the blind model 131 times, and 23 were draws.

So there are two ways to disagree with the market. Disagree often, on instinct, and lose more than you win. Or disagree rarely, only when the evidence moves the price far enough to change the side, and win more than you lose. The CQ blind model is the first kind, and we keep it on every page anyway, because it is the honest raw read and it is sometimes right when everyone else is not. CrystlQuant is the second kind.

Not every sport pulls its weight

The 40 to 32 is not spread evenly. Hockey and football carry it: 11 to 6 on the ice, 16 to 10 in football with those 13 draws. Basketball and baseball are level at 3 to 3 and 4 to 4.

The NFL barely registers. In 103 NFL games, CrystlQuant went against the market twice. On 27 September it had the Titans at 50.9% in New York, and the Giants won 12-7. On 5 October it had the Falcons at 51.8% in New Orleans, and Atlanta won 45-24. One each.

College football is the row we would rather not show. CrystlQuant disagreed 13 times in 395 games, and the market won 8 of them to our 5. That row stays in the count. A record that drops its worst sport is a sales pitch, and the overall lead holds with it in.

Why the five percent is the product

CrystlQuant is free. Nobody pays to read it, so the only reason anyone comes back is that what we said last week held up. The question for a free forecasting business is what to put in front of people, and the answer is in those numbers.

A forecast that agrees with the favorite is not news. Plenty of sites repeat the market's favorite and call it a prediction, and when it comes in, nobody learned anything. A disagreement is different. It is a claim that can be checked, in public, by anyone, and either way the result is a story: we were right when the crowd was not, or we were wrong and we say so.

So since 8 November, every call CrystlQuant puts out on social media and in its Weekend Calls newsletter is one of these games: CrystlQuant backing the side the market does not favor. Every one of them gets graded afterwards, misses included. That makes the calls rarer. It also makes each one mean something.

Our take

Forty to 32 over 72 games is a start, not a proof. An eight-game lead can vanish in a bad fortnight, and if they do, we will print that too. But the shape is right. The model that argues all the time is losing its arguments. The model that argues one game in twenty is winning them, and it is winning most in hockey, where it was never more than 51.9% sure.

That is the business we want to be in: agree with the market when it is right, which is most of the time, and say so plainly. Disagree rarely, name the side with its number before the game, and grade it in public afterwards. The 1,656 agreements keep us honest. The 85 disagreements are why anyone should read us. Every one of them, sport by sport, is on CrystlQuant's public record.