Who decides how a prediction market resolves

Four different things get called "resolution", and in every support conversation that starts with "I was right, why have I not been paid", at least two of them have been mixed up. They are: who declares the outcome, which document they read it from, how long the declaration takes to become money, and what happens when the question has no clean answer at all. A venue can be fast on one and slow on another, and the four answers vary more between venues than the fees do.

This is the one question in the sector that does not depend on where you live. The rest of a venue's page changes by jurisdiction — who may open an account, which contracts are listed, what a regulator has allowed. Resolution is the same mechanism for everyone holding the same contract.

How it works

Between a market closing and money moving, four separate decisions happen, usually made by different parties on different clocks:

1. Somebody declares the outcome. An exchange employee, a committee, a price feed, a bonded proposer, or the person who wrote the question.
2. They read it from a named document. Not the title of the market — a source agency, a statistical release, an official classification, an oracle report.
3. The declaration becomes a payout on a timetable that is written down in some places and not in others.
4. If the question cannot be answered as written, somebody decides what a position is worth instead. This is the part nobody advertises, and it is the part that pays you your cost basis rather than a dollar.

Who declares the outcome

The exchange, by its own determination. This is how the CFTC-regulated venues work and it is the shortest chain: the exchange reads the source, states the outcome, and its statement is the outcome. Polymarket US does not merely resolve markets — its rulebook defines the Source Agency as the Exchange itself, which then references the external sources designated in each contract's terms, organised as primary, secondary and tertiary. Rule 10.4 lets the CEO, the Chief Compliance Officer, the COO, the Head of Markets or their designate open a review at sole discretion before settlement, determine the final outcome, and reverse a final outcome in the case of obvious error. Rule 10.4(c) is one sentence long: the Company has full discretion in reviewing markets, and determinations made by the Company are final. The rulebook is published and is dated on its cover; the version behind this page is 14 September 2026.

Kalshi and ForecastEx work the same way with different furniture. Kalshi's contract terms, as filed with the CFTC, say that before settlement Kalshi may at its sole discretion initiate the Market Outcome Review Process under Rule 7.1 of its rulebook, and that determinations of its Outcome Review Committee are final and not subject to review. ForecastEx Rule 415 gives the same power to an Event Review Process, with a twist worth knowing: where the review was opened because of a potential error in the data retrieved from the Source Agency, a Change Control Oversight Board may determine the correct resolution by unanimous consent, and anything else goes to the Event Review Committee, whose determinations are final.

What this buys you is speed and a named accountable entity operating under a filed rulebook. What it costs you is that there is no appeal. "Who do I complain to" has an answer — the exchange's own compliance function, and behind it the CFTC — but it is not an answer that re-opens your contract.

An automated feed, with nobody reading anything. Limitless resolves the majority of its markets from Pyth Network at the deadline, and its short-cadence crypto up-or-down markets from a Chainlink time-weighted average. The documentation is specific about the edge that matters: the report at the exact resolution time is used first, and if it is unavailable, the first Chainlink observation inside the market's stated tolerance window is used instead. Everything a feed cannot settle — sports, politics — is resolved by the Limitless team.

A feed removes discretion and replaces it with plumbing risk. There is no committee to have an opinion, and equally no committee to notice that the feed printed something absurd for one second.

A bonded proposal with an escalation path. Polymarket's international platform resolves nothing itself. An outcome is proposed to the UMA optimistic oracle with a bond behind it, and for two hours anyone may dispute that proposal by posting a challenge bond of the same size — Polymarket's own documentation puts the usual figure at $750. A dispute opens a debate period of 24 to 48 hours, after which UMA token holders vote, a process the documentation puts at roughly 48 hours. The venue's statement of its own position is unambiguous: Polymarket is non-custodial and cannot alter or reverse market resolutions, and once finalised by UMA, outcomes are immutable.

This is the only process in this catalogue a third party can actually enter. It is also the only one where the party that decides has no relationship with you, no support inbox, and no obligation under a filed rulebook. Both halves of that are the same fact.

Zeitgeist built the fullest version of this shape — jurors and delegators locking the chain's own token, votes cast as sealed hashes and then revealed, plurality winning, up to three appeals and then a vote of every token holder — and it is worth knowing as a construction, because it is the design the rest of the category gets compared against. Its Polkadot slot lease expired in August 2026, its published endpoints no longer answer, and none of it is available to trade against today.

The one time a token-holder vote was actually run to the end. Everything above describes what these mechanisms are designed to do. Augur is the only case where the hardest one ran to completion on a public chain: a disputed market on a rocket launch triggered a fork in April 2026, the escalation game ran for two months, and holders then had until 3 August to move their tokens into whichever of two universes they believed. Fifty-nine and a half percent of the supply migrated, the deadline was enforced by the contract rather than by a person, and what did not move is stranded where it was. Read it before deciding that a vote of token holders is a milder outcome than an exchange saying no: the exchange has a support inbox, and a fork has a countdown.

The person who wrote the question. On Manifold the creator of a market resolves it, which the documentation states plainly and defends as trust placed in users. Moderators overturn resolutions in exceptional circumstances, and only moderators and staff can resolve a question to N/A. This is a real mechanism rather than an absence of one, and it scales the way you would expect: excellent on a question whose author cares, poor on a question whose author has lost interest.

Which document the outcome is read from

Never the title of the market. Every venue here resolves against a named source and a written payout condition, and the gap between the two is where most disputes live.

Kalshi publishes this as data, which is the most checkable form of it in the sector. Each series in the public API carries a settlement_sources array — each entry a name and a URL — plus a contract_url and a contract_terms_url pointing at the filed PDF. The inflation series KXCPIYOY names the Bureau of Labor Statistics and its CPI page; the presidential party series names the Office of the Presidency. You can read this before you take a position, from a script, without an account: it is a field on the public market data API, covered by the Kalshi API card.

Sources are frequently a hierarchy rather than one document. The motorsport head-to-head contract Kalshi self-certified in March 2026 names nine source agencies in order: the governing body, then the official website or app of the series, then ESPN, Sky Sports, Motorsport.com, Autosport, the Associated Press, BBC Sport and Reuters. If the first one is silent, the second one decides.

Two more clauses in that document are worth reading twice. Revisions to the underlying made after expiration will not be accounted for in determining the expiration value — so a corrected figure published next week does not move a settled contract. And all instructions on how to access the underlying are non-binding, provided for convenience only, and not part of the binding terms. The link is not the rule. The rule is the rule.

ForecastEx publishes a terms-and-conditions PDF per product code, all of them in one regulatory directory, and each one names the source agency, the exact publication the number is read from, and the release time. The US unemployment contract names the Bureau of Labor Statistics, the U-3 rate in the monthly Employment Situation release, and 7:30 AM CT as both the resolution time and the last trading time — and then says, as Kalshi does, that its links are informational and the source agency may publish the data somewhere else at any time. The CPI contract carries the sentence that decides a whole class of arguments in one line: ForecastEx will only consider the initial CPI value from the official release.

That is worth stating as a general rule, because it is one: on these venues the first print wins. Revisions, benchmark rebasings and corrections to the same statistic are real events in the economy and non-events for your contract.

A source can be replaced mid-life, and each venue says so. ForecastEx Rule 413(a) lets the exchange designate a new source agency and underlying after the first day of trading where the existing one becomes unreliable or unavailable, requiring the replacement to be objective and verifiable and posting the change to its website. Polymarket US Rule 10.3 is the same power, and its rulebook defines a "Modification" as a binding change to a contract's terms, specifications or resolution criteria — expressly including the interpretation of resolution criteria — that becomes part of the contract terms upon issuance. Polymarket's international platform does this through what it calls an "Additional context" clarification, published on chain via a bulletin board contract for UMA voters to consider, and it states the limit: a clarification cannot change the fundamental intent of the question.

So the rules you read when you opened a position are not necessarily the rules that settle it. They are, however, always published.

What it costs

The cost here is not a fee. It is capital locked between the moment a market stops trading and the moment the money is yours, and it is longer than most people expect and much longer when anything goes wrong. A market that has stopped trading is not a market that has paid.

• Polymarket, undisputed: the challenge period is 2 hours, and the documentation gives resolution as roughly 2 hours after the proposal is made. Redemption is then a transaction you send yourself.
• Polymarket, disputed: 24 to 48 hours of debate plus about 48 hours of voting, which the documentation totals as 4 to 6 days.
• Kalshi: the settlement date is no later than the day after the expiration date — unless the market outcome is under review under Rule 7.1, in which case the terms state no outer bound.
• ForecastEx: settlement no later than the day following expiration, unless the outcome is under review under Rule 415, in which case settlement occurs at the settlement cycle immediately following the conclusion of the review.
• ForecastEx, source delayed: if the source agency delays its release, the exchange delays the resolution time until the data is published, and posts one notice when that happens and a second when the contract resolves. An agency that misses a print by a week locks your capital for a week.
• Limitless, feed-resolved: at the deadline. Team-resolved: typically 24 to 72 hours after the deadline, depending on the complexity of the event.
• Polymarket US, under review: Rule 10.4 says settlement occurs on the date the review concludes and reaches a determination. The rulebook sets no maximum length for that review.

Two of these deserve the emphasis. The first is that every regulated venue's "next day" promise carries the same escape clause, and it is the clause that applies precisely on the contracts you will care most about — the close ones, the ambiguous ones, the ones where somebody has complained. The second is that a disputed Polymarket market is not a slow version of an undisputed one: it is a different process, run by different people, on a multi-day clock.

When the question has no clean answer

This is the section with no marketing page behind it on any venue, and it is where the difference between a good trade and a returned deposit actually lives. The pattern across all four mechanisms is the same: when a question cannot be answered as written, the fallback is close to your entry price rather than the dollar you were owed.

• Polymarket has a fourth voting outcome, Unknown, described in its own documentation as rare and reserved for events where none of the other options are appropriate. The market resolves 50/50 and each token redeems for $0.50. If you bought Yes at 12 cents and you were right, an Unknown outcome is a windfall; if you bought at 88 cents and you were right, it is a loss.
• Polymarket US Rule 10.5 covers contracts about a named person who dies or is incapacitated before expiration such that the outcome can no longer occur or be objectively determined. Those contracts settle at last traded prices before the event — and if trading was materially affected, at last traded prices before the circumstances became known or could reasonably have been anticipated, with the exchange resolving conflicting evidence in favour of the earliest time. If last traded prices are not a fair settlement, it goes to a Rule 10.4 review, and determinations under the rule are final and not subject to review. Kalshi's filed terms carry a materially identical provision, with its Outcome Review Committee determining a fair settlement price where no last traded price represents one.
• ForecastEx Rule 414 governs accelerated settlement where a circumstance prevents the event question being answered accurately. The exchange uses the most recent last prices where they are available; where they are not, or where it judges them an unfair allocation, the Event Review Committee makes a binding determination, and in no event shall the combined payout for a single Yes position and a single No position exceed $1.00. That last clause is the whole design in one sentence: the pot is not topped up.
• Limitless treats a market whose published criteria cannot be applied as misresolved and refunds the cost basis of the shares held on the side that should have won. Holders on the wrongly-paid side keep their payout, because payouts settle on chain and cannot be clawed back. Being right and being misresolved still costs you the trade, and the venue publishes that plainly rather than burying it.
• Manifold annuls to N/A, which returns trader mana and is restricted to moderators and staff rather than the creator.

Read these before you need them, because the design decision they share is invisible from the market page: you are not insured against a badly written question, you are refunded out of it.

What you can do about it

Two neighbouring questions have their own pages: what the position costs once you know it will be paid is in what a trade actually costs, and why the price you are reading may not be a probability at all is in where the liquidity comes from.

Everything above assumes the answer decides whether you are paid. Where it decides whether you are scored instead, the same question has different stakes and a different failure mode — the author of a question often resolves it themselves, which what a forecasting score measures takes apart.

Read three things before you take a position, in this order. The market's own rules — the payout condition, not the headline. The named source, and whether it is one document or a hierarchy. And the settlement timetable, including what happens if the source publishes late. On every venue in this catalogue, all three are published before you can trade.

Here is where the named source actually lives.

• Kalshi — the settlement_sources array on the series object in the public API, alongside contract_url and contract_terms_url, which point at the terms as filed. No account needed; see the Kalshi API card for the endpoints. The same rules are printed on the market page.
• ForecastEx — one terms-and-conditions PDF per product code, all in the regulatory directory, alongside the rulebook. Find the product code first; the file is named after it.
• Polymarket US — the contract's terms and conditions, read against the rulebook's definitions of Source Agency, Expiration Value and Contract Outcome. The rulebook is on the exchange's own legal path and is dated on its cover page.
• Polymarket — the Rules section of the market page, which states the resolution source and carries the link into the UMA oracle portal where a proposal can be disputed. Any clarification is published on chain rather than edited into the page.
• Limitless — the market page, which the documentation names as the authority on which real-world results resolve Yes, which resolve No, and how ties, cancellations and unlisted outcomes are handled.
• Manifold — the question's description, and the profile of whoever wrote it. There is no external source to check, which is the honest summary of the mechanism.

Ask these five questions when the rule does not obviously cover what happened. They are phrased so that a support desk can answer them, and each one maps to a clause above.

1. Which named source governs this contract, and if it is a hierarchy, which level are you reading? A silent primary source is the most common reason two people disagree about a settled contract.
2. Which publication of the figure governs — the first print or the latest revision? On the regulated venues the answer is usually the first, and it is usually written down.
3. If the source has not published, is resolution delayed or is the outcome determined another way? ForecastEx delays and posts a notice; a feed-resolved market falls through to a tolerance window instead.
4. Has a review or a clarification been opened on this market, and where is it posted? Every venue commits to publishing that somewhere — a website notice, a system notification, an on-chain clarification — and none of them will email you unprompted about a market you do not hold.
5. If this question is voided, do I receive $1.00 or what I paid? You now know the answer is almost always the second one. Ask anyway, in writing, before you size the position.

Size the position for the worst branch, not the modal one. The modal outcome is that the source publishes on time, the exchange settles the next day and none of this matters. The branch that should set your size is the one where the question is ambiguous, the review takes as long as it takes, and the contract returns your cost basis. If that branch is survivable, the mechanism is a detail. If it is not, the mechanism was the trade.

And check which entity you are actually trading with. The clearest illustration in this catalogue is one brand with two answers: Polymarket US determines outcomes itself under a filed rulebook and its determinations are final, while Polymarket's international platform determines nothing and cannot reverse what UMA finalises. Same name, same look, opposite answers to "who do I complain to". The prediction-market venues listing carries the resolution mechanism on every card, which is the fastest way to see how far apart they really are.