Polymarket does not decide who wins its own markets. The Polymarket UMA oracle does: an outside, token-based voting system where anyone can propose an outcome by posting a $750 bond, anyone can challenge it within a 2-hour window, and a disputed outcome goes to a vote of UMA token holders whose ballots are weighted by how many tokens they stake. That design settles the vast majority of markets without incident. It also means that in a dispute, the final word belongs to whoever holds the most tokens, and in March 2025 a single holder controlling roughly 5 million UMA across three wallets cast about 25% of the vote on a $7 million market that resolved to an outcome the real world had not yet delivered, a concentration users allege swung the result.
That case is the story of this article, because it exposes the one question every prediction market has to outsource somewhere: who decides what actually happened? Polymarket's answer is a crowd of bonded token holders. Kalshi's answer is an in-house committee operating under a federally filed rulebook. Neither answer is free, and the difference between them is the single biggest trust decision a trader makes without realizing it. Later, we will price out exactly what that one vote cost the losing side of the trade, down to the cent.
The Quick Answer
Polymarket markets are settled by UMA's optimistic oracle: a proposer posts a bonded outcome, a 2-hour challenge period runs, and if someone disputes it, UMA token holders vote, with each staked token counting as voting weight. In March 2025, users allege that one large holder's voting weight forced a "Yes" resolution on the Ukraine minerals market before the underlying deal existed, and Polymarket, calling the situation unprecedented, declined to issue refunds. The full mechanism, the cent-by-cent damage to No holders, and how Kalshi settles the same question in-house are all below.
How The UMA Optimistic Oracle Actually Works
If you are new to event contracts entirely, start with our plain-English explainer on how prediction markets work; this section assumes you know that a Yes share trading at 60¢ implies roughly a 60% chance and its No side at 40¢ implies the remaining 40%. What it covers is the part most traders never read: what happens after the event, when someone has to turn "what happened" into a payout.
UMA calls its system an optimistic oracle because it assumes honesty first and only escalates on conflict. Per Polymarket's own resolution documentation, anyone can propose a market's outcome by "putting up a bond in USDC.e which will be forfeited if the proposal is unsuccessful"; that bond is $750. Then, in the platform's words, "once a market is proposed for resolution it goes into a challenge period of 2 hours." No challenge means the proposal stands and the proposer collects a reward.
A challenge changes everything. Per UMA's protocol documentation, disputes are forwarded to UMA's Data Verification Mechanism, where "stakers commit secret votes during a 24-hour commit period" and reveal them in "the following 24-hour reveal period." A dispute resolves "when a minimum 65% majority of staked UMA is cast in favor of a single outcome," which leaves room for up to 35% of staked tokens to disagree and still lose. Stakers who skip the vote or land against the majority are slashed, with the slashed amount redistributed to the winners.
Read that incentive structure closely, because it is the hinge of the whole case study. The system does not reward voting for the truth. It rewards voting with the majority, and the majority is counted in tokens, not people. For everyday markets those two things are the same. The March 2025 dispute is what it looks like when they separate.
The habit worth stealing: before you trade any event contract, on any venue, find three names in the rules: who proposes the outcome, who can challenge it, and who has the final word. If you cannot answer all three, you do not fully know what you are holding.
The Case Study: A $7 Million Market Flips From 9% To 100%
The market was "Ukraine agrees to Trump mineral deal before April?", which drew about $7 million in volume. Through late March 2025 it behaved like a market pricing an unlikely event: Yes traded near 9¢, meaning holders of No, at roughly 91¢, were being paid to underwrite the 91% consensus that no deal would be signed in time. Then, between March 24 and 25, the price surged from 9% to effectively 100%, per The Block's reporting, and the market resolved Yes even though no signed agreement existed — a resolution that set the Polymarket and UMA communities publicly at odds. The two countries did not actually sign a minerals deal until the end of April, more than a month after the market had already paid out.
What happened in between was a UMA dispute vote. On-chain observers, including the threat researcher who first traced the wallets, reported that a single holder controlling roughly 5 million UMA tokens spread across three accounts cast about 25% of the votes on the resolution. Risk firm Veritas Protocol put the concern plainly: "Whether it's a timing issue or a deliberate attack, 25% of votes coming from just three accounts is a red flag." Users alleged orchestrated manipulation, a governance attack in which a voter with a position in the market used token weight to resolve it in their own favor. One user quoted in The Block's coverage of the episode framed the allegation in general terms: "The event outcomes are determined by a group of influential users who secure positions and then vote in their favor."
To be precise about what is established and what is alleged: the price move, the resolution, and the concentration of voting power are on-chain facts. The intent behind the votes is an allegation by users and outside analysts, not an adjudicated finding, and no individual has been publicly identified or charged. Polymarket itself called it "an unprecedented situation" and said its team had "been in war rooms all day internally and with the UMA team to make sure this won't happen again," statements reported by The Block; the episode was widely covered as a governance attack. The platform promised improved monitoring and clearer resolution processes. But on the question traders cared about most, its Discord statement was blunt: "Unfortunately, because this wasn't a market failure, we are not able to issue refunds."
That sentence deserves a beat of attention. Within the rules as written, it is accurate: the oracle ran exactly as designed, bonds were posted, a vote was held, a 65% majority formed. Nothing malfunctioned. The system produced a settlement most observers believe was wrong about the real world, and it did so while working perfectly. That is the difference between operational risk and design risk, and only one of them shows up in an FAQ.
A Worked Example: What The Flip Cost A No Holder
Here is the promised arithmetic, using the market's own prices. Take a trader who liked the No side at the consensus price shortly before the swing:
| The Position | The math |
|---|---|
| 500 No Shares Bought At 91¢ | $455 at risk |
| Payout If No Settles Correctly | $500, a $45 profit (about a 9.9% return) |
| Actual Payout After The Yes Resolution | $0 |
The last row is the entire lesson. The vote meant Yes shares settled at 100¢ and No shares at 0¢, so this trader was not wrong about the world; no deal was signed by the deadline. They were wrong about the oracle. Meanwhile, anyone positioned on Yes at 9¢ turned $45 into $500 on the same vote. When settlement itself is the contested variable, the trade stops being "will the event happen?" and becomes "who counts the votes, and what do they hold?" — a question no amount of geopolitical research answers.
"Is This Rigged?" What The Design Does And Does Not Promise
The question traders keep typing, in forums, app reviews, and complaint threads, is three words: "is this rigged?" — usually aimed at whales, insiders, and manipulation. The honest answer for UMA-resolved markets is that "rigged" is the wrong frame, and the right frame is not entirely comforting. Token-weighted voting is not secretly biased; it is openly plutocratic. The design document tells you in advance that influence scales with stake. The $750 proposal bond filters spam, the 2-hour challenge window invites scrutiny, secret commits stop lazy vote-copying, and slashing punishes the minority side. Those are real safeguards against carelessness. None of them is a safeguard against concentration, because concentration is the pricing model of token-weighted voting, not a bug in it.
That is why the incident produced allegations rather than a clean verdict. If a large holder votes their honest reading of ambiguous evidence, the system worked. If they vote their trading book, the system was exploited. From the outside the two are indistinguishable, and that ambiguity, more than any single flipped market, is the trust cost. We took the same skeptical question to a different corner of this industry in our look at whether Kalshi's 15-minute bitcoin markets are rigged, and the pattern repeats: the answer usually lives in the settlement design, not in a smoking gun.
How Kalshi Settles The Same Question In-House
The contrast case is worth spelling out, because Kalshi's settlement process outsources nothing. Kalshi is a CFTC-regulated exchange, and its rulebook is filed with the federal regulator; the current version is public on the CFTC's site, and the quotes below come from that rulebook text, fetched directly from the CFTC filing in August 2026.
Under Rule 6.3, a binary contract pays its settlement value based on whether the payout criterion captures the expiration value, and where terms are unclear, "Kalshi has sole discretion to interpret a Contract's Terms and Conditions." When an outcome is truly contested, Rule 7.1 provides the Market Outcome Review process: an Outcome Review Committee "shall review all relevant evidence and determine a final Market Outcome within a 24 hour period," and "the determinations made by the Outcome Review Committee are final."
Set the two systems side by side and the trade-off gets sharp:
| Question | Polymarket (UMA oracle) | Kalshi (in-house) |
|---|---|---|
| Who Decides A Disputed Outcome? | UMA token holders, weighted by stake | Kalshi's Outcome Review Committee |
| How Long Does A Dispute Take? | 2-hour challenge, then 24-hour commit + 24-hour reveal | Committee determination within a 24 hour period |
| What Binds The Decider? | Slashing and majority incentives | A CFTC-filed rulebook and federal oversight |
| Can A Wealthy Outsider Buy The Vote? | Influence scales with tokens staked | No vote exists to buy |
The row to sit with is the last one. Kalshi's design makes the minerals-market scenario structurally impossible, since there is no token to accumulate. What you accept in exchange is discretion: one company, not a crowd, interprets the terms, and its review decisions are final. Traders have argued with those judgment calls too — we walked through how that plays out in what happens when a Kalshi market is disputed — and an exchange being its own referee is precisely the kind of arrangement that only works under real regulatory supervision. Neither venue offers a settlement you can appeal to a neutral third party after the fact. The difference is which failure mode you are underwriting: a vote that money can weight, or a referee that answers to a regulator instead of to you.
What A Trader Can Actually Do About Oracle Risk
You cannot vote-proof a token oracle or veto a review committee, but you can price both, and the callback to our worked example tells you how. The No holder's 9.9% expected return assumed settlement risk was zero. It was not zero, and on that market the difference was everything.
Practically, that means reading resolution criteria the way sharp traders read them before taking a position, on any venue: identify who proposes, who can challenge, who decides, and on what timeline. On UMA-resolved markets specifically, ambiguity is the raw material of disputes, so the fuzzier the wording ("agrees to," "officially announces"), the bigger the discount your entry price should carry, especially near a deadline, when a resolution proposal can land while the facts are still moving. And concentration risk cuts both ways in time: the same vote that settles a market against you has, in this case, also drawn public commitments from Polymarket to add monitoring and clarify rules, so the process you are pricing is at least a moving target rather than a frozen one.
None of this is a reason to avoid event markets. It is a reason to treat settlement design as part of the odds. If you want to sharpen the probability-reading reflex that makes all of this second nature, our free expert picks show probability-first handicapping applied daily, at no cost, on markets where the scoreboard does the settling.
FAQ: The Polymarket UMA Oracle And Disputes
How does the Polymarket UMA oracle settle markets? A proposer posts a $750 bond with an outcome; a 2-hour challenge period follows. Unchallenged proposals settle as proposed. Challenged ones go to a token-weighted vote of UMA stakers, with a 24-hour commit phase, a 24-hour reveal phase, and a 65% majority threshold.
Was the whale vote illegal or proven to be manipulation? No finding of wrongdoing has been adjudicated, and no individual has been publicly identified. The 5 million UMA across three wallets and the 25% voting share are on-chain observations; the claim of orchestrated manipulation is an allegation made by users and outside analysts. Polymarket declined refunds on the grounds that the oracle itself did not malfunction.
What should I check before trading a UMA-resolved market? Three things the case study rewards: how precisely the resolution wording is drafted (vague verbs like "agrees to" are where disputes live), how close the market is to its deadline (a bonded proposal can land while facts are still moving), and whether you would still like your entry price if settlement went to a token vote rather than to the newspaper record.
This article explains settlement mechanics and market structure for educational purposes. It does not recommend any trade or position, and descriptions of disputes reflect public reporting and user allegations, not adjudicated findings.
