Blockchain prediction markets: what the network settles and what it does not
What is on-chain at Polymarket, how a market is resolved with UMA and why legal classification decides where to operate: status as of November 2026.

The CFTC, the US derivatives regulator, estimates that trading volume on the prediction markets registered with it exceeded $25 billion in 2025, according to the proposed rule it published in June 2026. That growth has led fintechs, media groups, gaming operators and exchanges to consider launching their own market or integrating an existing one, and the model they usually have in mind is Polymarket, which settles its positions on a blockchain network. The question to answer before choosing technology is which part of the product that network solves, because a prediction market combines three distinct things: a system that holds the money and the positions, a procedure that decides what happened and a legal classification that says who may offer it and to whom.
Blockchain handles the first well: custody, positions and payouts are recorded in contracts anyone can inspect, without relying on the operator's internal books. The other two sit off the network. The outcome is declared by an oracle, a mechanism with its own incentives and failure modes, and whether the contract is a supervised derivative or gambling is decided by regulators and courts, which in 2025 and 2026 reached opposite conclusions depending on the country and, within the United States, on the state. That second piece has carried the most weight over the past two years, and it is the one to study first.
What happens on-chain in a market like Polymarket
A binary prediction market issues two kinds of shares for each question: one that pays one dollar if the outcome is "yes" and another that pays one dollar if it is "no". Their price, between zero and one dollar, reflects what participants are willing to pay for each outcome given the liquidity, fees and access restrictions at that moment, which is why it reads as a market estimate rather than an objective probability. On Polymarket, those shares are ERC-1155 tokens created with the Conditional Tokens Framework, and they are exchanged against dollar collateral through an exchange contract.
The architecture is hybrid. According to the repository of its exchange contract, an operator provides off-chain order matching, while settlement happens on-chain and non-custodially: each user signs orders with their own key and the contract executes the swap atomically, so the operator matches orders but cannot move funds on its own. The network is Polygon. What a user can check on-chain is their balance, their positions and each settlement; what they cannot check is how orders were matched before reaching the contract, which depends on the operator.
The outcome is declared by an optimistic oracle, with bonds and a last-resort vote
No contract knows who won an election or whether a central bank cut rates. That fact enters the network through an oracle, and Polymarket uses two. Up-or-down markets on an asset's price are resolved with Chainlink's time-weighted average price, with no human involvement, because the comparison is mechanical. The rest are resolved with UMA's optimistic oracle, whose operation the Polymarket documentation describes: anyone can propose the outcome by posting a bond, typically $750, which opens a two-hour challenge window. If nobody disputes, the market resolves.
If someone disputes, they post a matching bond and a second proposal round opens. If that is disputed too, the case moves to a 24-48 hour debate period and then to a vote by UMA token holders lasting about 48 hours; a disputed resolution takes four to six days. Whoever is right gets their bond back plus half of the other side's, and anyone who proposes an outcome too early loses theirs. The design works because a wrong proposal costs money, but its last resort is a token-weighted vote, and its reliability depends on how those tokens are distributed and on voters reading the market's rules.
The March 2025 dispute and what changed afterwards
That dependency became visible in March 2025 with the market "Ukraine agrees to Trump mineral deal before April?", which had traded about $7 million. Between 24 and 25 March, the "yes" price rose from around 9% to 100% and the UMA vote resolved it as "yes", even though Ukraine had not officially accepted the deal and Polymarket itself had said it was too early to resolve. Platform users attributed the result to a large UMA token holder. According to The Block, Polymarket acknowledged the market had resolved against its users' expectations and its own clarification, but did not refund bets because it did not consider it a market failure.
In August 2025, UMA changed the procedure for Polymarket: since then only a list of proposers with a track record may propose outcomes, 37 addresses at the start, selected for having made more than 20 proposals in three months with over 95% accuracy, while disputes remain open to anyone. According to UMA, proposals from inexperienced users caused unnecessary disputes; the change reduces them at the cost of restricting who can propose. It does not alter the last resort, the token holders' vote. For anyone designing a product, the case shows that the oracle problem is managed through rules, incentives and governance, and that those decisions weigh as much as the contract.
Derivative or gambling: classification decides where you can operate
In the United States, a contract on an event can trade as a derivative on an exchange designated by the CFTC (a designated contract market, or DCM). Kalshi operates that way, and Polymarket returned to the US market by the same route after buying an already registered exchange and clearinghouse: on 24 November 2025, the CFTC approved an amended order of designation allowing it to operate through intermediaries. That entity, Polymarket US, is supervised as a derivatives exchange and runs separately from the international platform that settles on Polygon.
Being a DCM has not settled the matter. Several states argue that contracts on sports outcomes are bets subject to their gambling laws, and the federal appeals courts have split: in April 2026, the Third Circuit sided with Kalshi against New Jersey, treating those contracts as swaps under federal law, and on 28 August 2026 the Ninth Circuit reached the opposite conclusion against Nevada, finding that Kalshi had not shown federal law displaced state law. Both rulings concern preliminary injunctions. Meanwhile, in June 2026 the CFTC proposed a procedure to decide, contract by contract, when a contract on gaming, war or unlawful activity is contrary to the public interest; the comment period closed on 27 July 2026.
In Spain, the answer has been gambling. On 26 May 2026, the Ministry of Social Rights, Consumer Affairs and 2030 Agenda opened sanctioning proceedings against Polymarket and Kalshi for operating without the required administrative authorization and ordered their websites blocked as a precautionary measure until the proceedings conclude. Spain's gambling regulator, the DGOJ, considers prediction markets to be games of chance when people bet on uncertain future outcomes, and holds that unlicensed operators lack safeguards such as identity verification or controls on minors and self-excluded people. The CFTC, by contrast, lists price discovery and information aggregation among the criteria in its proposal: the same contract is judged by its informational value in one case and by the risk to the player in the other. For anyone launching a product, the consequence is practical: the licence required depends on the country, and the network does not replace it.
Writing a contract on the ECB: where disputes begin
Imagine, as a hypothetical case, that a European fintech wants to offer a binary contract on whether the European Central Bank will cut the deposit facility rate at a specific meeting, and considers settling it on a public network with an optimistic oracle. The first thing it has to write is the question, and that is where the decisions that prevent the most disputes appear: which rate exactly, which source (the ECB's official statement), when the outcome counts as known, what happens if the meeting is postponed and how it resolves if the ECB cuts a different rate from the one asked about. Every ambiguity left in the text is a possible dispute.
With a mechanism like Polymarket's, on the day of the meeting a proposer declares the outcome after the statement; proposing earlier would forfeit the bond, because the documentation treats a premature proposal as incorrect. If nobody disputes within two hours, holders of the winning outcome receive one dollar per share. If there is a dispute, payout is delayed by days and the final decision rests with the oracle's voters, who will read the question exactly as written. Meanwhile, the fintech has to show users the difference between the price they see, the quantity they actually bought and the state of the resolution.
The decision that conditions everything above is the legal one. If the fintech wants to offer the contract to customers in Spain, the DGOJ will treat it as gambling, so it will need a gambling licence or will have to leave that market; if it frames it as a financial instrument, it falls under investment services rules, with different requirements. In the United States, the route would be a CFTC-designated exchange. The chosen network changes none of those answers, which is why the legal analysis comes before the architecture.
The network settles prediction markets; outcome and licence are decided elsewhere
What blockchain adds to a prediction market is concrete: custody in which the operator cannot dispose of funds, positions and payouts anyone can verify, and a public record of each resolution, with the proposal, disputes and vote. That transparency lets a user check why an outcome was paid, something a centralized operator offers only if it chooses to. From the sector's perspective, centred on users' control of their assets, non-delegated custody is a real advantage over a model in which the balance is an entry in the platform's books.
What it does not add is equally concrete. The outcome still depends on an oracle whose last resort is a token-weighted vote, as the March 2025 dispute showed, and the ability to operate depends on a legal classification that currently divides US courts and that Spain has settled as gambling. A project that starts with the network and leaves the question, the source and the licence for the end has the order reversed: the two pieces that decide whether the market works are off the chain.
Metlabs builds the settlement smart contracts and oracle integration for products of this kind, and we start with how the question is defined and how it fits the regulation of each market. The general US framework is covered in our guide to blockchain regulation in the United States.
- CFTC: Prediction Markets; Public Interest Determinations, 91 FR 35806 (June 12, 2026)
- US Court of Appeals for the Ninth Circuit: KalshiEX, LLC v. Assad, No. 25-7516 (Aug. 28, 2026)
- CFTC: Polymarket US amended order of designation (Nov. 24, 2025)
- Polymarket: CTF Exchange repository
- Polymarket documentation: UMA resolution
- UMA: Managed proposers (Aug. 12, 2025)
- The Block: Polymarket on the UMA vote in the Ukraine mineral deal market (Mar. 26, 2025)
- DGOJ (Spain): sanctioning proceedings and blocking of Polymarket and Kalshi (May 26, 2026)
Frequently asked questions
Got a blockchain project in mind?
From idea to product, strategy and technology must move forward together.







