Prediction market resolution disputes: who decides and how to challenge

The short version

• A prediction market “resolves” when its result is fixed, and “settles” when the winning shares are paid: $1 each, losing shares nothing.
• On a CFTC-regulated exchange the exchange’s markets team decides the outcome against the sources named in the contract’s rules, usually within hours.
• On Polymarket’s non-US markets anyone can propose the outcome by posting a bond; it stands unless disputed inside a two-hour window, after which UMA token holders vote.
• To challenge an outcome you go to the exchange and, failing that, to binding arbitration or the CFTC; or you dispute on the UMA app inside the window and post a matching bond.
• Once the exchange’s determination is final, or the oracle vote closes, the result cannot be appealed or re-settled. Disputed outcomes are rare but carry a large share of the money.^1,2,3,5,6,8

Who decides how a prediction market resolves?

It depends on the venue. On a CFTC-regulated exchange the exchange itself decides: its rulebook names it as the body that determines the outcome against external sources, and its markets team finalises the result. On Polymarket’s non-US markets no one at Polymarket decides. An outcome is proposed to the UMA Optimistic Oracle and becomes final unless it is disputed, and a dispute is settled by a vote of UMA token holders.^1,3,5

A few definitions help. An event contract is a yes or no contract that pays a fixed amount, usually $1, on a stated outcome and nothing otherwise. A designated contract market (DCM) is the label the Commodity Futures Trading Commission (CFTC) gives a registered exchange allowed to list these contracts. An oracle is the software that reports a real-world fact onto a blockchain, so a smart contract, code that runs automatically once its conditions are met, can act on it. Kalshi, Polymarket US and ForecastEx are DCMs; Polymarket’s non-US venue settles through the UMA oracle instead.^5,9,10

Three moments that sound alike differ. A market resolves when its result is fixed, settles when the payout is made, and closes when trading stops. Kalshi warns that a market staying open is “not an indication that payout criteria have not been met,” and that markets “are not re-settled because an event appears to have concluded while trading was still available.”³

Resolution is closer to a decision than to a data lookup, behaving, in the words of a Harvard Journal of Law & Technology digest, like an “adjudicative” layer. The wider shape of these disputes is in our guide to prediction market disputes; where the regulation itself is being fought out is in prediction market lawsuits.⁹

How does a regulated exchange decide and settle a market?

The exchange confirms the result against the source named in the contract rules and finalises it. Kalshi says a market “settles when the official outcome is confirmed and our markets team finalizes the result,” most markets “often within about 3 hours,” longer when it is waiting on official data.³

The named source is the “Source Agency.” On ForecastEx that is “any agency, whether public or private, that publishes data used to determine the Outcome of an Event Question” (definition 71). On Polymarket US the Source Agency is the exchange itself, referencing external Sources set in each contract as Primary, Secondary and Tertiary Sources. A contract resolves against those named sources, not a general sense of what happened.^1,2

The close time is not the determination time. Trading can end before or after the event, and settlement happens only once the official result is confirmed, with some contracts setting a determination time later than the apparent end. Kalshi tells users to check a market’s Rules for the official source and the determination criteria.³

Before it settles a hard case, the exchange can open a formal review. Polymarket US calls it the Contract Outcome Review Process. A named officer, the Chief Executive Officer, Chief Compliance Officer, Chief Operating Officer or Head of Markets, may review “circumstances that may have a material impact on reliability or transparency” of the event, determine the final outcome, and reverse an outcome “in the case of obvious error.” The rulebook adds that “Determinations made by the Company are final” (Rule 10.4(a) to (c)).¹

ForecastEx runs the same idea through a standing body. Its Event Review Process (Rule 415) pauses settlement until the review ends. A suspected error in the data from the Source Agency can be corrected by unanimous consent of its Change Control Oversight Board; failing that, the Event Review Committee, a majority of whose members must be Public Directors (Rule 203(e)), determines the resolution with “full discretion,” and “the determinations by the Event Review Committee are final” (Rule 415(f)).²

Kalshi’s rulebook sets out the same structure. Before settlement it can open a Market Outcome Review Process (Rule 7.1), sending a contract whose underlying measure may be unreliable to a standing Outcome Review Committee of three, at least two of them Public Directors (Rule 2.7(e)). The committee must “determine a final Market Outcome within a 24 hour period,” has “full discretion,” and “the determinations made by the Outcome Review Committee are final” (Rule 7.1(a) to (c)).⁴

There is a carve-out for contracts about a person. If the natural person a contract is about dies or is incapacitated before expiry so the outcome can no longer occur, Polymarket US settles the affected contracts “at last traded prices prior to such an event,” and those determinations “are final and not subject to review” (Rule 10.5(a), (f)).¹

How does Polymarket’s optimistic oracle decide an outcome?

On Polymarket’s non-US markets the result is set by the UMA Optimistic Oracle, not by Polymarket. When the result is clear, someone proposes the outcome against the market’s pre-defined rules, shown under the order book, and posts a bond in USDC.e, a dollar-pegged token, “which will be forfeited if the proposal is unsuccessful.” An optimistic oracle treats a proposed answer as correct unless challenged inside a set window, so most settle with no vote.⁵

Polymarket’s help centre, last updated 11 January 2026, states a $750 USDC.e proposer bond, forfeited in full “if you propose a market too early, or are unsuccessful,” and a “challenge period of 2 hours.” As of 30 October 2026 those are the figures on Polymarket’s own page, which can change. An unchallenged proposal is accepted and the proposer gets the bond back plus a reward.⁵

A disputed proposal goes to UMA’s Data Verification Mechanism (DVM), a vote by disinterested token holders aiming at a “Schelling Point,” the answer “where everyone arrives at a reasonable and fair conclusion.” UMA describes the voters as “a group of reasonable strangers,” individuals without a stake in the result. The DVM vote is the panel in this model, an adjudicator made of many holders rather than one named person.⁷

This design struggles with what a human panel does: judging a natural-language statement open to more than one reading. UMA calls these disputes “intersubjective,” based on interpretation rather than a fact that can be “mathematically proven,” and notes that because a market is zero-sum, a losing side has “a strong incentive to dispute the outcome.” The oracle and code-level mechanics sit alongside smart contract dispute resolution.⁷

How do you dispute a prediction market outcome?

The route depends on the venue. On a regulated exchange you raise it with the exchange first. On Polymarket’s non-US (crypto) markets you dispute on the UMA app inside the window.^3,6

On the regulated path, Kalshi tells users to check the market’s Rules and determination time, then, if the result still looks wrong, to use in-app chat with the “market title and any details (e.g., source) so we can escalate to the markets team.” The exchange then runs its review process, the Market Outcome Review Process (Kalshi Rule 7.1), the Contract Outcome Review Process (Polymarket US Rule 10.4) or the Event Review Process (ForecastEx Rule 415). Your objection is heard first by the body that made the call.^1,2,3,4

If you still think the exchange is wrong, the recourse is a claim against it, not a re-run of the outcome. Polymarket US routes “any dispute, controversy, or claim against the Exchange” to “binding arbitration in New York, New York” (Rule 9.2), with the fees in arbitration among participants “borne by the non-prevailing party” (Rule 9.1(d)). ForecastEx routes any claim brought against it to arbitration under the NFA’s rules (Rules 801 to 802), “voluntary” for participants other than eligible contract participants. The CFTC’s own page sets out the complaint, exchange, NFA and reparations routes for a prediction market dispute.^1,2,10

On the crypto path, you dispute on the UMA app at oracle.uma.xyz, linked from the market’s Rules section, and post a bond “the same amount as the proposer bond (usually $750).” A debate period of 24 to 48 hours follows, with evidence in UMA’s Discord, then UMA token holders vote over about 48 hours. Four outcomes are possible: the proposer wins (bond back plus half the disputer’s), the disputer wins the same way, the proposal was “too early,” or the result is unknown and the market settles 50-50.⁶

The detail on the NFA and CFTC routes is in prediction market arbitration and NFA arbitration; market abuse and insider cases are in prediction market insider trading.²

Can you appeal a resolution once it is final?

Largely no. On both models the result is built to be final. On a regulated exchange the review determinations and the natural-person carve-out are stated to be final (Kalshi Rule 7.1(c); Polymarket US Rules 10.4(c) and 10.5(f); ForecastEx Rule 415(f)), and Kalshi says a market is not re-settled merely because an event appears to have concluded while trading was still available.^1,2,3,4

On Polymarket’s non-US markets the platform says it is non-custodial and “cannot alter or reverse market resolutions,” and that “once finalized by UMA, outcomes are immutable.” The token-holder vote is the last word inside the platform. “Immutable” means the recorded result cannot be changed, not that it was proven right.⁶

That settles the money question too. A contested or corrected outcome is not a refund of your stake. Winning shares still pay $1 and losing shares nothing, against the final result. A market is unwound, settled 50-50 or at the last traded prices before the event, only in the narrow cases the rules name: a UMA vote that returns “unknown,” or a contract whose subject can no longer meet the condition (Polymarket US Rule 10.5).^1,5,6

So “appeal” here means challenging the process, through arbitration or the CFTC, not asking for a different answer to the same question. Where that process goes is in arbitration in New York and prediction market arbitration.¹⁰

Why do two platforms settle the same event differently?

Because each runs its own rulebook, names its own sources and sets its own cut-off, two venues can reach different results on what looks to a trader like the same event.^1,2,3

The deeper split is in who holds the pen. One model vests the final decision in a single exchange applying its written rulebook, Polymarket US, ForecastEx and Kalshi; the other distributes it across UMA token holders in a stake-weighted vote. A Harvard Journal of Law & Technology digest argues the same: a centralised resolution “vests final adjudicative authority in a single entity,” while a decentralised one “distributes adjudicative authority across token holders, stakers, or a decentralized oracle network,” and resolutions “can be credible, auditable, and reliable at multiple venues,” “where outcomes are verifiable from public sources and rules are transparent.” Whether these markets are even lawful is fought out in prediction market lawsuits.^1,2,4,9

The data points the same way. A preprint study of Polymarket markets finds dispute rates swing sharply with the kind of source a contract names, the same ambiguity that makes two rulebooks diverge.⁸

How often are outcomes disputed, and which are hardest to resolve?

Formal disputes are rare but concentrated in the money. A preprint study of Polymarket, posted to arXiv in October 2026 and not yet peer-reviewed, covers almost every resolved market on the platform, 802,863 listed between January 2023 and May 2026. Of those whose dispute history is visible, 0.31%, about one market in 320, were formally challenged, yet disputed markets carried 5.8% of all trading volume. The dispute history is visible for about 66% of markets, so the rate is measured on that subset.⁸

Which are hardest follows the source. The dispute rate runs from 0.08% for markets resolved from an on-chain price feed, through 0.66% from the description alone and 1.01% where no source is identifiable, up to 3.79% for markets resolved from entertainment-industry references such as box-office trackers, a span of almost fifty times, though that top rate rests on only 18 disputes in 475 markets. Harder-to-read rules carry more risk too: the study puts each extra grade level of reading difficulty at roughly 23% higher dispute odds.⁸

The authors are careful here. At a 0.31% base rate the features rank risk “far better than chance” but cannot predict any single dispute, so they suit triage, grouping markets by risk, not an automatic accept or reject.⁸

Disputed outcomes are the hardest category. They are the close calls where the wording left room for argument, the rule did not anticipate reality, and both sides are paid to fight. On UMA’s own account of intersubjective disputes and a losing side’s incentive to dispute, that is exactly where a clear record and a named decider matter most.⁷

How can you avoid a resolution dispute before you trade?

Read the rules under the order book before you trade, and prefer markets whose outcome comes from a single, named, verifiable source with a stated determination time. In the study the markets disputed least resolve against a mechanical or official source, and the markets disputed most against free text or no identifiable source.^3,5,8

• Check the named Source and the determination time, not just the close time.
• Prefer a mechanical or official source over a free-text or consensus-based description.
• Know who decides before you trade: the exchange’s markets team, or an oracle vote.
• Learn the challenge window and the bond in advance, not after a result you dislike.
• Keep your own evidence of the result, with its date and source, in case you object.^1,3,5,8

How AI and dispute processes are handled across resolution bodies is in our guide to AI in dispute resolution.¹⁰

What would make prediction market resolution disputes fairer and faster?

In our view, these disputes are already settled under a written rulebook: a contract’s resolution criteria, a named source, an exchange’s review process, a code of arbitration. That is the strength to build on, not to work around. The numbers point the same way. Formal disputes are rare, about one Polymarket market in 320, but they cluster where the stakes are high and the wording leaves room for argument, so disputed markets hold a share of the money far above their count. The fix for those cases is procedural. Run each one the same way under the rules that were published before trading, with a named person or panel applying the stated source inside the stated window. And leave a record each side can inspect and a court or regulator can later rely on: the two-hour window, the determination time, the final determination, the arbitration that follows. Finality is fair only when the path to it was legible.^1,2,3,6,8