Article
Kalshi and Polymarket are legally forbidden from sharing a customer. No arbitrageur can trade both books. The one mechanism that textbooks say aligns prices across venues — capital flowing between them — is structurally unavailable. And yet, across six matched World Cup winner markets this week, the two venues agree within 0.5 percentage points. This is the story of the largest natural experiment in price discovery ever run — and what the Assymetrix cross-venue dataset shows it found.
A note on what's different about this Intelligence Brief. The previous eighteen installments in this series synthesized public reporting, institutional research, and platform documentation. This one is built primarily on the Assymetrix cross-venue dataset — live orderbooks, manually verified matched markets, and a Kalshi trade archive of 363 million records with millisecond timestamps. Every observation below is reproducible against the Data API, and we include the code. This is what the intelligence layer is for.
The Wall
Start with a fact most casual observers of prediction markets don't fully register: Kalshi and Polymarket do not share a single trader.
Not "mostly don't." Not "rarely." Structurally, legally, by design — zero overlap.
Kalshi is a CFTC-regulated Designated Contract Market. Its order book is open to verified US persons who have passed KYC. Polymarket's international exchange — where the World Cup volume lives — is closed to US persons entirely, a condition of its 2022 CFTC settlement. A trader who can legally post an order on one book cannot legally post an order on the other.
This matters because of what it removes: arbitrage.
In every market structure textbook, arbitrage is the mechanism that aligns prices across venues. When Microsoft trades at $432.10 on NYSE and $432.25 on NASDAQ, someone buys one and sells the other, and the gap closes in milliseconds. The arbitrageur is the courier that carries price information between order books. Fragmented markets stay coherent because capital can flow across the fragments.
Between Kalshi and Polymarket, the courier is banned. If France-to-win trades at meaningfully different prices on the two venues, nobody on earth can legally buy the cheap one and sell the expensive one. The regulatory perimeter didn't just segment the customer bases — it severed the price-transmission mechanism itself.
Which sets up a clean experiment. Two capital pools. Same real-world event. No bridge between them. Economics gives them permission to drift apart freely.
The 2026 World Cup — the largest event in prediction market history, with Kalshi clearing $31 billion in June volume and Polymarket setting a $10.8 billion monthly record — is running this experiment at maximum scale, right now.
So: did the prices drift?
The Data
We matched the outright World Cup winner markets across both venues — Kalshi's per-country binary contracts against Polymarket's per-country binary contracts — and verified resolution compatibility manually (more on that verification below, because it turned up its own finding). Six countries had live, liquid, structurally comparable markets on both books this week.
Here's the comparison — one API call per matched market:
The output, from live snapshots taken July 9:
Team | Kalshi | Polymarket | Spread (K − P) |
France | 32.5% | 31.95% | +0.55 pts |
Belgium | 3.0% | 2.45% | +0.55 pts |
Norway | 6.5% | 6.05% | +0.45 pts |
Spain | 19.0% | 18.75% | +0.25 pts |
Morocco | 3.0% | 3.05% | −0.05 pts |
Argentina | 19.0% | 20.05% | −1.05 pts |
Update — July 10: One day after publication, the table began resolving. Morocco was eliminated in last night's quarterfinal; both venues' Morocco markets converged toward zero — the terminal-price version of the convergence this analysis measured. France repriced upward on both books after the win. The July 9 snapshot above stands as published; we're tracking the live readings through the final on July 19.
Median absolute spread: 0.50 percentage points. Maximum: 1.05. Mean: 0.48.
Two venues that cannot share a customer, pricing the biggest sporting event on the planet, agree on the probability of every major outcome to within about half a point.
Sit with how strange that is. There is no arbitrageur closing these gaps. Nobody is buying France on Polymarket at 31.95 and selling it on Kalshi at 33. The gap could be ten points and no mechanical force would close it. It's half a point anyway.
Convergence Without Arbitrage
When two prices align, there are only two possible explanations: capital is flowing between them, or both are independently processing the same information and arriving at the same answer.
Capital flow is ruled out by law. What remains is the more interesting explanation: two completely segregated trader populations — US retail and institutional flow on Kalshi, international and crypto-native flow on Polymarket — are each aggregating the same public information about the same 22 players on the same pitch, and independently converging on the same probability.
This is worth stating carefully, because it's the strongest empirical validation of the "prediction markets as information aggregators" thesis that this natural experiment could have produced.
The institutional adoption arc this series has documented all year — Goldman analysts citing prediction market probabilities, the Evercore ISI framework for when to trust market signals, the prop firm wave — rests on one foundational assumption: that these prices reflect information, not just the idiosyncratic sentiment of whoever happens to be on the platform. The skeptic's version has always been: "Polymarket's price is just what crypto Twitter thinks; Kalshi's price is just what US retail thinks. Neither is a real probability."
The World Cup just ran the test at unprecedented scale. If prediction market prices were platform-specific sentiment artifacts, two hermetically sealed populations would produce visibly different prices — and nothing would correct them. Instead, the median disagreement across six matched markets is 0.5 points.
The prices converge because the information converges. The crowd isn't the mechanism. The world is.
The Fingerprint
The spreads are small. Their direction is where it gets genuinely fun.
Kalshi prices the European favorites — France, Spain, Belgium, Norway — consistently above Polymarket. Polymarket prices exactly one team above Kalshi by a meaningful margin: Argentina.
The largest spread in the dataset — a full point — belongs to Argentina, the defending champions, and it points in the opposite direction from every European favorite. Kalshi's US-based population leans toward the European powers. Polymarket's international, Latin-American-heavy, crypto-native population leans toward Argentina.
The venues agree on probability. They disagree on who they like. And that disagreement is only visible when you can lay the two books side by side in one normalized frame — from inside either venue, the fingerprint doesn't exist.
We want to be careful not to overclaim here: six markets is a snapshot, not a study, and a 1-point lean is sentiment at the margin, not mispricing. But the pattern is exactly what you'd predict if two segregated populations shared information and differed only in affinity — and it's the kind of structural observation that cross-venue data exists to surface. We'll keep publishing the fingerprint as the tournament resolves.
Where They Don't Agree: The Structure of the Bet Itself
The comparison above required a verification step that produced its own finding — one that validates something we published in the very first Builder Brief.
Before comparing prices, we manually verified that the matched markets resolve identically. For the outright winner markets, they do: Kalshi asks "Will France win the 2026 Men's World Cup?", Polymarket asks "Will France win the 2026 FIFA World Cup?" — same binary event, same trophy on July 19. Different settlement mechanisms (Kalshi settles centrally as a regulated DCM; Polymarket settles through UMA's Optimistic Oracle with a bonded challenge window), but the same question. resolution_compatible = true.¹
The match-level markets are a different story.
Kalshi runs group-stage match winners as three-way markets: Team A / Team B / Tie, with the Tie outcome typically pricing around 4–5%. Polymarket runs match winners as binary markets where YES includes extra time and penalties.
Same headline — "Who wins France vs. Senegal?" — structurally different bets. A group-stage match that ends in a draw resolves Kalshi's market to "Tie" and would resolve differently on a binary market that has no such outcome. For group-stage matches, the two venues are explicitly not resolution-compatible — resolution_compatible = false — and no cross-venue price comparison is valid. Comparing those prices as if they were the same market is a category error that produces phantom spreads of several points — the Tie probability, misread as disagreement.
In the knockout rounds, the structures converge: with no draws possible, Kalshi's markets go binary too (confirmed on Paraguay vs. France, July 4, which settled with no Tie outcome), and cross-venue comparison becomes valid again.
This is the finding from Builder Brief #1 — most "matching" markets across platforms have at least one material difference in how they resolve — showing up on the biggest sporting event on earth. Anyone comparing group-stage match prices across these venues without checking resolution structure was comparing two different bets and calling the difference a signal.
The canonical schema catches this automatically. The naked price feed doesn't.
The Two Books Are Not the Same Shape
One more asymmetry the normalized data surfaces — and it is not subtle.
At the July 9 snapshot, Polymarket's France book showed $24,827,602 in bid depth resting near the mid (with $27.2M more on the ask side). Kalshi's verified near-mid depth on the same market: $23,370 — 73,030 contracts within reach of the 32-cent quote.²
That is not a 2x difference, or a 10x difference. It is a ratio of 1,062 to one, on two books whose mid-prices agree within 0.55 points.
The reconciliation is in the flow. Kalshi's France market has done $52.3M in lifetime volume through that thin resting book: its price is discovered by continuous, high-turnover trading, with market makers quoting tight — a one-cent bid-ask spread at snapshot — while resting minimal size. Polymarket's price is anchored the opposite way: $101M+ in lifetime volume with $50M+ in capital standing openly on the book. One is a flow market; the other is a depth market. Two completely different microstructures, independently arriving at the same probability.
For a researcher quoting "the prediction market probability of France winning," the venues are interchangeable. For anyone trying to move $500K, they are not comparable at all — one book absorbs the order near the quote; the other book's quote is real for a small fraction of that size. Price convergence and liquidity divergence, at the same moment, on the same market. Two different questions — what does the market believe? and what can the market absorb? — that a single price feed collapses into one number.
How Fast a $50M Market Thinks
Everything above is comparative statics — snapshots of where the books rest. The Assymetrix dataset also holds the dynamics: the full Kalshi trade archive — 363 million records with millisecond timestamps — including every trade on match markets that ran $30M–$72M per game (France vs. Senegal: $30M+; France vs. Iraq: $50M+; Norway vs. France: $72M).
That granularity makes it possible to reconstruct, second by second, how a market doing tens of millions in live volume absorbs a goal:
272,819 trades land in that five-hour match window — dense enough to resample locally into a one-second price series and read the repricing around any goal.
[CHART PLACEHOLDER — France vs. Senegal price series, ±5 minutes around goal, 1-second resolution. PENDING: team to run reconstruction script and confirm goal timestamp + repricing duration. Pull-stat format: "From the goal timestamp, the market moved from X% to Y% in Z seconds, on $N of volume."]
We'll publish the full reconstruction — and the tooling to run it on any match — as a companion Builder Brief. The headline observation belongs here: a live event market at this depth reprices in seconds, continuously, in public, with every trade timestamped and auditable. There is no equity market analog where the information event (a goal) and the repricing are both this observable. As a laboratory for studying price discovery, nothing else comes close.
What This Means
Pull the threads together and the World Cup experiment resolves three questions the institutional adoption arc has been circling all year.
Are prediction market prices information or sentiment? The strongest evidence yet for information. Two sealed populations, no arbitrage bridge, 0.5-point median agreement. The signal that Goldman's analysts and Evercore's framework treat as meaningful just survived its most demanding possible test.
Can you trust a single venue's price? Mostly — for probability. The 0.5-point convergence means either major venue's mid-price is a reasonable estimate of the consensus. But the fingerprint (systematic directional leans), the structural mismatches (three-way vs. binary markets sharing a headline), and the liquidity asymmetry (2x depth differences at identical prices) are all invisible from inside any single venue. The price travels; the context doesn't.
What happens when this converges less politely? The World Cup is the easy case: maximal public information, saturation coverage, deep books on both sides. The Evercore framework's conditions are all satisfied. The open question — the one we'd nominate as the most important research question in this category — is what the cross-venue spread does on markets where those conditions fail: thin books, ambiguous resolution criteria, information that reaches one population before the other. Segregated pools that agree when information is abundant can disagree sharply when it isn't — and with no arbitrage bridge, nothing forces the reconciliation. The wall that makes this week's convergence remarkable is the same wall that will make some future divergence durable.
When that divergence comes, it won't be visible on any platform's own screen. It will only be visible in the layer that reads all of them at once.
That layer is what we build.
Reproduce This Analysis
Every number in this post comes from the Assymetrix Data API and is reproducible with a free key:
The six-country comparison:
GET /api/v1/canonical/markets/compare— runnable in the API playground without writing codeResolution structure verification: the
resolution_compatibleflag and resolution metadata returned on every compare responseThe trade-level dataset:
GET /api/v1/kalshi/markets/{ticker}/trades— the 363M-record Kalshi archive with millisecond timestamps, Developer tier and above
Get a key at dashboard.assymetrix.com. If you extend this analysis — more matched markets, the knockout-round series, the post-final retrospective — we'd genuinely like to see it: dean@assymetrix.com.
This is the nineteenth installment in the Assymetrix Intelligence Brief series — and the first built primarily on the Assymetrix cross-venue dataset.
Previous: "Meta's Arena and the Question Nobody Is Asking: Who Resolves the Truth?"
Related: "We Indexed Every Prediction Market Into One Schema. Here's What We Found." — the Builder Brief that introduced the resolution-compatibility problem this post's verification step confirmed at World Cup scale. "Evercore ISI Has a Framework for When to Trust Prediction Markets. Here's What It's Missing." — the framework whose conditions this experiment satisfied.
Assymetrix is building the cross-venue, on-chain intelligence layer that turns public ledgers into readable, structured market data — independent of any single platform.
Methodology and footnotes
Snapshots. Prices in the six-country table are live orderbook snapshots from July 9, 2026 (Kalshi 12:07 UTC; Polymarket 09:39 UTC). Spot prices on liquid markets; both books were verified live and quoting at capture. [PRE-PUBLISH NOTE: if feasible, re-pull both venues simultaneously the morning of publication and refresh the table — removes the snapshot-gap caveat entirely.
¹ The UMA settlement wrinkle. Kalshi settles centrally and typically finalizes within hours of the final whistle. Polymarket's UMA Optimistic Oracle involves a bonded proposal (500 USDC) and a challenge window, which can extend settlement by days in the contested case. For pricing purposes the markets are identical; for capital efficiency purposes — when you get your money back — they are not. A structural difference that lives in the resolution metadata, not the price.
² On Kalshi depth figures. Kalshi's raw book totals for this market (73,030 contracts bid at 32¢ / 479K contracts at 66¢) mix two different things: the 66¢ figure is NO-side liquidity expressed in YES terms, not YES-side interest resting near the quote. The ~$23,370 figure is effective near-mid YES depth — the bid-side contracts actually within reach of the 32–33¢ market. Raw book totals and usable near-mid depth are different numbers, and the gap between them is itself a small argument for why liquidity quality scoring requires more than naive book sums.
What we deliberately did not use. An earlier draft of this analysis showed a 15.5-point cross-venue spread on France. It was an artifact, not an observation: a stale last_trade_price field in the Polymarket markets table, whose last captured trade for this market predates the tournament (May 1). Cross-referencing against the live orderbook mid (31.95%) eliminated the discrepancy entirely. The final comparison uses orderbook mid prices throughout, never last-trade fields — a data-quality distinction that is invisible in any single price feed and is, in miniature, the reason normalized cross-venue data has to be built carefully or not at all.
