The first 48-team World Cup ended last night. If you trade or research prediction markets, the tournament left behind something almost as interesting as the football: a month of order-book history on a genuinely liquid, genuinely global event — the kind of market that only exists once every four years.
The catch is that Polymarket doesn't retain or publish its order-book history. Once a snapshot passes, the official record is gone. Everything below comes from an independent live capture that recorded the top 20 levels of every listed World Cup match book at a ~30-second median cadence, from the group stage on June 20 through last night's final. This post walks through what that data shows for the final itself — and what it teaches about how prediction markets price a one-off event.
The final, as the market saw it
Spain vs. Argentina, July 19, 19:00 UTC kickoff. In the hour before kickoff, the three-way match-winner market (win / draw / win, 90 minutes) averaged:
| Outcome | Pre-kickoff mid (18:00–19:00 UTC) |
|---|---|
| Spain | 42.6% |
| Draw | 32.0% |
| Argentina | 26.1% |
Two things stand out. First, a 32% draw leg is enormous by club-football standards — the market fully understood that a World Cup final between two evenly matched sides is a coin-flip that often refuses to land inside 90 minutes. Second, those three mids sum to 1.006: a total book overround of about 0.6%. For comparison, a typical sportsbook 1X2 market carries 4–7% vig. On the biggest match of the cycle, the order book priced the event more tightly than almost any retail bookmaker on earth.
In play: two favorites deflate, the draw inflates
Across the match window, Spain's leg traded as high as 47.5% and as low as 4.1%. Argentina's ranged from 25.6% down to 0.4%. And the draw did what draws do in a tight final: it ground upward as the clock burned, finishing the 90-minute window at 0.996 bid/ask mid — regulation ended level, and the market settled Draw.
Note what that market can't tell you: who lifted the trophy. The three-way market's story ends at full time; extra time and penalties belong to a different contract. That distinction — knowing exactly which question each market answers — is half of reading prediction-market data correctly, and it's invisible in candle or last-price data. You need the book.
Why depth matters more than price for research
A price series tells you what the market thought. The ladder tells you how hard it would defend that opinion. The final alone produced just over one million L2 depth rows in this capture — every ~30 seconds, both sides of the book for all three outcome tokens, price and size at each of the top 20 levels. That's the data you need to answer the questions a backtest actually runs into:
- Could I have filled 500 shares at the displayed price, or was the top of book three lots deep?
- How wide did spreads blow out in the seconds around a goal, and how fast did they re-tighten?
- Did size arrive ahead of the move or after it — is displayed depth information or decoration?
Every one of those questions dies on last-price data. If you've read our earlier guide on reading a Polymarket order book, this dataset is that skill applied historically, at scale.
What a month of World Cup capture contains — and honestly, what it doesn't
The full archive covers 47 of the 104 match markets Polymarket listed: 20.8 million depth rows plus 955,000 top-of-book snapshots — 38 group-stage matches (June 20–28) and the complete knockout run from the quarterfinals through the semifinals, the third-place match, and the final.
It does not cover June 11–19 (capture started June 20) or the round of 32, which fell to a market-discovery bug in the capture pipeline. The vendor lists every included and missing match in a coverage file that ships inside the archive — worth calling out because "dense but honest about its gaps" is rare in a corner of the data market where most sellers simply don't mention holes. We've written before about vetting data and edge vendors; a disclosed gap map is exactly the kind of signal to look for.
Where the data lives: the capture described here is the World Cup 2026 Polymarket Order-Book Archive from ZenHodl — $39, per-match parquet files, manifest, and the full coverage map. There's a free complete group match (872K depth rows) on the samples page if you want to inspect the schema before paying anything.
Three research questions this dataset can settle
1. How efficient is a prediction market at peak liquidity?
The 0.6% overround number above is one snapshot of one market. With 47 matches you can measure overround as a function of time-to-kickoff, liquidity, and stage of the tournament — a clean empirical answer to "when is the market beatable in principle?"
2. How do books absorb scheduled information?
A goal is an information shock with a timestamp. Depth around goals — how many levels vanish, how long the ladder takes to rebuild — is the soccer version of an earnings-release microstructure study, and nobody has published one on prediction markets yet.
3. Does displayed depth predict anything?
Order-book imbalance is a known short-horizon signal in equities and crypto. Whether it means anything on a 30-second sports book is an open question — and an answerable one with per-level size on both sides for a full month.
The honest caveat, as always
None of this is a trading signal. The tight overround is the market telling you it's efficient; the write-up of this same data on the vendor's own blog — their World Cup final post — says the same thing in their words. Historical depth data is for understanding execution, calibration, and microstructure. If someone sells you World Cup order-book data as a way to beat next season's markets, reread the vetting guide above.
Want to look at the actual data?
Free sample: one complete group match, full L2 depth, no signup.