What the Window Saw

A price is two numbers

Put a market price on a screen and it looks like one number. Forty-two cents on a team to win. It reads as a probability, and people treat it as one, and most of the time that is close enough to be useful and wrong in the way that matters.

It is really two numbers wearing one coat. There is what the crowd believes. And there is what it costs you to take that side, which is a separate quantity, set by the venue rather than the crowd, and free to move on its own.

A market price is a belief with a cost attached. The belief is what everyone quotes. The cost is what nobody shows you. The gap between them is where the interesting information lives.

You can watch the two come apart. In the middle of this tournament we found that Polymarket had raised the fee it charges on sports positions. No belief changed. Nobody revised their view of any team. The stated prices carried on exactly as before, and it cost measurably more to agree with them. If you were reading a price as a probability, you saw nothing happen. Something had happened.

It gets less obvious than that, because the two big families of venue do not even express the two numbers the same way. Prediction exchanges quote the belief and charge the cost separately, so the price on the screen is close to an honest opinion and the invoice arrives beside it. Sportsbooks quote a number with the cost already folded inside, so what you see is all-in and the opinion has to be recovered from it. Neither is dishonest. They are simply different objects, and putting them next to each other in a column and comparing them by eye produces a confident answer that is not about anything.

Which is the first argument for a window rather than a screen. The comparison is not something you can look at. It is something you have to compute, and to compute it you have to be holding every venue at the same instant, which no venue will ever do for you, because no venue has any reason to tell you that the same belief is cheaper somewhere else.

The board agrees. That is the finding.

The intuitive reason to watch many venues at once is that you expect them to disagree, and that disagreement to be worth something. We recorded every venue we could reach, continuously, for the whole tournament, specifically to find out how true that was.

It is not very true, and the size of the not-very is the finding. On the 38 settled matches where our stored pregame record carries a full three-way price from more than one venue, we took the widest gap between any two venues on any single outcome, which is the most generous possible way to ask how much they disagreed. The typical answer was about 1.5 percentage points. The widest gap we recorded anywhere in 100 matches was 2.7. On the final, across every venue we track, the blended read came in at a spread of 0.33 of a point.

Read that plainly. If you came to this looking for two venues that believed different things at kickoff, the 2026 World Cup had almost nothing for you. The pregame board is close to unanimous, and it is most unanimous exactly where the money is heaviest. Anyone selling a cross-venue edge on the biggest fixtures in world sport should be asked to show their spreads.

But that is a statement about belief, and belief is only the first of the two numbers. On the final the venues agreed on what would happen to within a third of a point and did not agree at all on what it cost to hold that opinion. Identical conviction, different invoice, decided entirely by where you clicked. That condition is not an anomaly we caught once. It is the ordinary state of this market, it persists precisely because it is invisible from any single screen, and it is the thing we built the window to show.

Divergence in belief is rare and closes fast. Dispersion in cost is structural and never closes at all.

What convergence leaves behind

So if the venues agree with each other, where is the disagreement worth watching? It is not across the board. It is in time. The interesting argument is never venue against venue. It is a market against itself, one minute against the last.

Convergence is the fastest and least glamorous process in this whole business. News arrives, the board reprices, the venues re-agree with each other, and it is finished long before anyone has written a sentence about it. Which produces a genuinely awkward property: if you check the market from time to time, you will only ever see agreement. You will conclude, honestly and wrongly, that nothing much happens. The move is not hidden. It is just over.

The only way to see it is to have been recording before you knew it was coming, and to still be recording afterwards. That is what the window is. Not a signal, not a prediction, not an edge. A recording device pointed at a crowd, kept running whether or not anything is happening, which is the entire cost of the exercise and the entire reason anyone else stops doing it.

What it leaves behind is a record of when a crowd changed its mind and how violently. Two of those are worth stating here, and both are quoted with full timestamps in the companion ledger rather than repeated at length.

In the second semifinal, England's leg reached 68.5 percent to reach a World Cup final. The crowd genuinely believed it. Minutes later, at 20:52 UTC, Argentina went from 14.5 percent to 81.5 percent in a single step of the record, a reversal of about sixty-seven points in about a minute. There is a box score for that match, and highlight reels, and a great deal of reaction. None of them contain those numbers, because none of them were pointed at the price.

The third-place match is the better example, because nobody cares about the third-place match. France against England reached 97.5 percent on England, fell back to a genuine three-way board with England at 40.5, and closed at 99.5. Anyone who saw only the final score will tell you the result was never in doubt. The record says the crowd was as certain as it ever gets, stopped being certain, and became certain again, all inside about seventy minutes.

A box score summarizes an outcome. A price record summarizes a belief. Only one of them can tell you what people thought while they still did not know.

This is also the honest answer to the question we get asked most, which is whether the window predicts anything. It does not. It observes. The market is doing the predicting, competently and in public and for free, and the useful thing nobody was doing was writing down what it said at the moment it said it, in a form that could later be checked.

A law, and the night it expired

A continuous record lets you notice things that no single match contains. The clearest one this tournament produced was a pattern in how the crowd was wrong.

When a market favorite failed, it usually did not lose. It got held. Through the quarterfinals the tournament produced 25 regulation draws against just 8 outright defeats of a favorite. When the chalk cracked in this World Cup it stalled, roughly three times more often than it broke. That shape held across the group stage and deep into the knockouts, and it held long enough that we published it, with a date on it, three days before the semifinals.

Then both semifinal favorites lost outright, back to back, in the two largest matches on the calendar. The law expired exactly when the stakes peaked, which is the way these things generally go. Across the full 100 matches the final tally reads: the market's favorite failed 35 times, and 24 of those were draws against 11 outright defeats. And in the last four matches of the tournament, the two semifinals, the third-place match and the final, the pregame favorite won 0 of 4.

We are not telling that story because the pattern was clever. We are telling it because of what happened next. We had written the law down, in public, before it broke. When it broke we said so, in the piece covering the matches that broke it, and then corrected our own phrasing of it again in the ledger when the full settled record showed we had overstated it. A claim that exists in a dated document can be wrong. A claim that lives in someone's recollection reshapes itself, silently, into whatever turned out to be true.

Why the misses are the product

Over 39 days the market was, on the whole, good. Favorites finished 65-35 at an average stated price of 60.9%. Its three-way score came out at 0.477 where somebody who knows nothing and splits every match evenly scores 0.667 and lower is better, which puts the crowd roughly 28% of the way from ignorance toward certainty. Informative. Nowhere near safe. Anyone selling you the second thing is selling you something.

The lowest price the market ever put on something that then happened was about 10.8%, and two matches tied there: Ecuador against Curacao and Argentina against Cape Verde. Both were heavy favorites held to a draw. A one-in-ten event is not a scandal. It is a Tuesday, and it happened often enough over 100 matches to bend the whole tournament.

Publishing that is the point. Any operation can assemble a reel of the calls that landed, and a reel of the calls that landed is evidence of nothing except editing. A record is only worth something if the misses are in it at the same resolution as the hits, sitting in the same table, findable by anyone who wants to make us look bad with them.

That includes ours. Over the tournament we published four corrections against ourselves. We carried a stale fee schedule for about a week, which meant the cost half of every affected price on our own surface was understated, and we did not catch it in a monitor. An exchange counterparty asked us about it in passing. One of our own automated briefs reported a divergence between two venues that did not exist, because a label in our mapping was wrong while the prices behind it were right. We overstated the draws pattern in print. And the first draft of the companion ledger carried a headline record that appears in no file we hold, which is why every scoreboard figure on both of these pages is now computed from the published dataset at build time rather than typed in by a human.

Coverage was uneven too, and we would rather say it than have it found. Our stored pregame record carries a full three-way price from one venue on 99 of the 100 matches and from the others on considerably fewer, so the headline score leans on one venue more than we would like. The live record is broader than the pregame snapshot. The full accounting, including the corrections in their own words, is in the ledger.

A record you can check is the only reason to believe the next one.

One hundred settled questions

There is a reason a World Cup is a better laboratory than it looks. It is not the football. It is the settlement.

For 39 days the same crowd was asked the same three-way question, 100 times, on a published schedule, and every single time the question was answered for good within about two hours. The wording never drifted. Nobody got to argue afterwards about what had really been meant. There was no version of the outcome in which the forecaster graded their own paper.

Almost nothing else in forecasting works like that. Most predictions are made about questions that never resolve cleanly, or resolve so far in the future that the person who made them has moved on, or resolve into an argument about definitions. A tournament hands you a hundred hard settlements in five weeks. That is why the numbers above are worth quoting and why a good month of pundit takes is not: not because the market is smarter, but because the market was graded and the takes were not.

Nothing in this article is a football technique. Recording every venue at once, separating what a crowd believes from what it charges to agree, catching the minute a belief collapses rather than reconstructing it from a result afterwards, and then grading the whole thing against a settlement nobody controls: those are properties of markets. Sport is simply where they happen fastest and most often, which made it the right place to prove them and the wrong place to stop.

Questions that settle are everywhere on the calendar. Elections settle. Policy decisions settle. Rate announcements, awards, launches, confirmations: anything with a date and a resolution and a crowd willing to put a number on it. Every one of them has the same two numbers inside it, the belief and its cost, and for almost all of them nobody is writing down the second one.

The window was never about football. Football just answered 100 times in 39 days, which made it the fastest way to find out whether the window worked.

The 100-match record stays online permanently, with the dataset attached, so it can be checked against whatever we claim next. The full ledger is at 39 Days, 100 Matches, One Ledger, every settled line is on the scoreboard, and the mechanics of reading a two-number price are in How to Read Tater. The window itself is still open, at taterit.com/pulse, pointed at the next hundred questions.