Resolution Is the Product: How Prediction Markets Settle (2026)
Resolution, not pricing, is the hidden cost centre of prediction markets. How Polymarket and Kalshi settle, why 2026 disputes happened, what a pipeline needs.
TL;DR
A pricing error costs a few points of margin. A resolution error costs the whole stake on one side, a dispute and the trust of every customer who read the title rather than the fine print. The largest public disputes of 2026 were not about who won but about what the question meant: whether a bitcoin sale confirmed one day after a deadline counts, or whether a contract titled "out as Supreme Leader" pays on death. Polymarket resolves through UMA's optimistic oracle; Kalshi's rulebook names a Source Agency for every contract and a committee for when the source fails. An operator running event contracts as fixed odds inherits all of this, and needs a source hierarchy, timeouts, a human fallback and a public rationale before the first market opens.
Key takeaways
- Every dispute in this post traces back to the question, not the event: an unwritten evidence-timing rule, or a title that promised more than its rule.
- Polymarket's documentation: a 2-hour challenge period after a proposal; a dispute bond equal to the proposer's, "usually $750"; a 24–48 hour debate, then a UMA token-holder vote of about 48 hours; outcomes include "Too Early" and a 50/50 "Unknown".
- Kalshi's rulebook: a "Source Agency" publishes the value that decides each contract; when the outcome cannot be determined, Kalshi may pay at the last traded price or its Outcome Review Committee decides within 24 hours; Rule 7.2 lets Kalshi replace a Source Agency after listing.
- The CFTC staff advisory of 12 March 2026 expects a US exchange to identify "the specific data source(s) on which settlement will be based" and to assess their "reliability, objectivity, and manipulation resistance"; it warns that "a consensus of yet-to-be-determined sources may not be sufficient".
- Adkuu Pulse resolves markets with an automated multi-tier process (data API first, then web search, then consensus) with human review as the fallback.
Why resolution, not pricing, is the cost centre
Two 2026 cases set the scale. On Polymarket, a market asking whether Michael Saylor's Strategy would sell any bitcoin by 31 May 2026 traded around 10% YES for most of its life. On 1 June the company disclosed in an 8-K filing that it had sold 32 BTC, about $2.5 million, between 26 and 31 May. YES rose from 10% to 80% on the filing, then fell below a penny when the market resolved NO after two disputed rounds, with about 98.6% of UMA voting power behind NO. Two traders sued in New York Supreme Court on 3 July over at least $797,198 in refused redemptions. Blockhead's report of 8 July 2026 says Polymarket had passed 1,150 contested markets in 2026 by then, more than in all of 2025, a figure Blockhead attributes to the lawsuit.
On Kalshi, a contract titled "Ali Khamenei out as Supreme Leader?" was settled, after two clarifications and roughly 19 hours between halt and formal close, at the price before air strikes on Iran began. Kalshi reimbursed every contract bought after the strikes and the fees on them; its chief executive, Tarek Mansour, said the business took a $2.2 million loss on a market InGame estimated would otherwise have earned around $500,000 in fees. Neither case involved a mispriced probability. Both were failures of the question.
How Polymarket resolves: an optimistic oracle
Polymarket's documentation describes resolution by the UMA Optimistic Oracle.
| Stage | What the documentation says |
|---|---|
| Challenge period | "Once a market is proposed for resolution it goes into a challenge period of 2 hours." Undisputed, the proposer receives the bond back plus a reward |
| Dispute | A challenger posts a bond equal to the proposer's, "usually $750", on the UMA dapp; "This begins the debate period of 24-48 hours" |
| Vote | UMA token holders vote; the vote takes about 48 hours |
| Outcomes | Proposer wins; disputer wins; "Too Early", for "proposals for which the underlying event has not yet happened"; or Unknown, where "the market price resolves to 50 yes and 50 no" |
| Finality | Polymarket is non-custodial and cannot change or reverse a resolution; once UMA finalizes an outcome it cannot be altered |
UMA's own documentation adds the vote mechanics: "stakers commit secret votes during a 24-hour commit period", then reveal them over 24 hours; "Disputes resolve when a minimum 65% majority of staked UMA is cast in favor of a single outcome"; and stakers who do not vote, or vote with the losing side, are slashed. UMA says the optimistic path "resolves the vast majority of requests, currently 99.8%".
Two further rules matter to anyone copying the model. "The market title describes the market, but the rules define how it should be resolved." And a fix after listing is expensive: "When a market is clarified, the order book is cleared and all resting orders cancelled at the time of clarification." Adkuu's reading: the design buys speed on the 99.8% path and hands the remainder to a token-weighted vote on what the rule means, which makes rule wording the whole game. The answer page on how prediction markets resolve disputes covers the dispute side in more detail.
How Kalshi settles: a Source Agency and a committee
Kalshi describes itself as regulated by the Commodity Futures Trading Commission as a Designated Contract Market, and its rulebook (KalshiEX LLC Rulebook, version 1.15, dated 17 January 2025, in the copy published by Robinhood) reads like a settlement specification. A "Source Agency" is "the agency that publishes the Underlying and/or Expiration Value for any Contract"; the Market Outcome is YES "If the Expiration Value is encompassed within the Payout Criterion" and otherwise NO. The help centre states the operational rule: "A market settles when the official outcome is confirmed and our markets team finalizes the result," most markets "within about 3 hours", and "If their data is delayed or revised, we may wait before settling, depending on the market rules."
The rulebook plans for failure. Under Rule 6.3(b), if at expiry "it cannot be determined whether the Payout Criterion encompasses the Expiration Value of the Underlying", Kalshi determines the payouts: "If available, Kalshi may use the last traded price of the Contract to determine the payout", and otherwise "the Outcome Review Committee will be responsible for making a binding determination of fair allocation." Under Rule 7.1 that committee must "determine a final Market Outcome within a 24 hour period". Under Rule 7.2, if a Source Agency "is unavailable or undergoes significant modifications, Kalshi retains the authority to designate a new Source Agency and Underlying for that Contract", one that "would be objective and verifiable", and may move the expiration date for "the rescheduling or cancellation of an event whose outcome governs a Contract's Underlying, or delayed data from a source".
| Design question | Polymarket (per its documentation) | Kalshi (per its rulebook and help centre) |
|---|---|---|
| Who decides | A proposer, then anyone who disputes, then UMA token holders | The markets team; the Outcome Review Committee when the outcome cannot be determined |
| Source | Named in each market's rules | A Source Agency per contract, replaceable under Rule 7.2 |
| Clock | 2 hours, then 24–48 hours of debate, then about 48 hours of voting | Most markets within a few hours of the outcome being known, often about 3 hours; the committee within 24 hours |
| When the source fails | Dispute; possibly a 50/50 outcome | Last traded price, or a committee allocation |
| Rule changes after listing | Clarifications, with the order book cleared | Fixed from the first day of trading except under the emergency rule and Rule 7.2 |
| Public record | Rules and clarifications on the market page | Outcomes posted by 11:59 pm ET on the day determined; halts explained within 24 hours |
Two disputes from 2026, and what each teaches
The Strategy bitcoin sale (Polymarket). The Block reported on 4 June 2026 that on 1 June Polymarket added a note to the market page: "Confirmation achieved outside of the market's time frame does not qualify." The sale had happened inside the window; the evidence arrived one day after it. Traders argued the rule had been written after the fact: "Polymarket changed the rules, and now the outcome is literally in the rules," one said; a platform should not "apply an unclear or unwritten rule after real money has already been placed," said another. Galaxy Research's line was sharper: "Prediction markets should price what happens, not how the oracle will reinterpret rules after the fact." The transferable lesson: a market about an event must say, before the first trade, whether the event window and the evidence window are the same thing, and which governs when they differ.
The Khamenei contract (Kalshi). InGame's report, last updated 4 March 2026, describes a title that read as an event market over terms that settled at the last traded price before the Supreme Leader's death. When strikes began and unconfirmed reports of his death circulated, trading continued through the morning and the exchange posted that the odds of removal had surged to 68%; a first clarification spoke of "confirmed reports", a second fixed its "grammatical ambiguity". The lesson: a title and its rule must say the same thing, a market that settles on a price should be titled as one, and the halt has to come before the clarification.
What a regulator asked for in 2026
On 12 March 2026 the CFTC's Division of Market Oversight issued a staff advisory to designated contract markets (CFTC Letter No. 26-08). It restates that "each DCM has a specific statutory obligation to list for trading only derivative contracts that are not readily susceptible to manipulation", warns that "overly broad or general contract specifications" hinder that analysis, and says a product submission should include "a description of the settlement methodology that accounts for differing potential permutations of the contract, including identification of the specific data source(s) on which settlement will be based, and an assessment of the reliability, objectivity, and manipulation resistance of such sources." A footnote adds: "A statement that a contract will settle based upon a consensus of yet-to-be-determined sources may not be sufficient to satisfy the requirements of DCM Core Principle 3." For sports it recommends "relying on official data provided by the relevant league or governing body, as applicable, as the settlement source."
The letter says it "represents only the views of DMO" and creates no rights. It addresses US exchanges, and the classification of event contracts varies by jurisdiction; nothing here is a statement about what is permitted where. As a published description of what a settlement rule should contain, though, it is the clearest available. The Commission's broader rulemaking is covered on the answer page on the CFTC's 2026 prediction-market ANPRM.
A taxonomy of dispute causes
| Cause | What it looked like in the record | Design rule that prevents it |
|---|---|---|
| Ambiguous wording | Kalshi's first Khamenei clarification had to be corrected for "grammatical ambiguity" | Define every term; test the rule against the three most likely surprises before publishing |
| Missing deadline for evidence | The Strategy market had an event window but no written rule on when confirmation had to arrive | State the event window and the evidence window separately, with a time zone, and say which governs |
| Source disappears or changes | Kalshi's Rule 7.2 exists because a Source Agency can become "unavailable" or undergo "significant modifications" | Name a primary and a fallback source at authoring, in order of precedence, and publish the order |
| Outcome partially occurs | Polymarket's "Unknown" outcome pays 50/50; Kalshi's Rule 6.3(b) pays at the last traded price | Decide in advance: a void with full refund, a binary threshold, or a stated pro-rata rule |
| Event cancelled or postponed | Kalshi's Rule 7.2 allows the expiration date to move for "rescheduling or cancellation" | Write whether postponement extends or voids, with a hard outside date after which the market is void |
| Title promises more than the rule | "Ali Khamenei out as Supreme Leader?" over a last-traded-price rule | Generate the title from the rule; a second reader checks that neither claims more than the other |
| Evidence standard undefined | The Strategy market turned on whether an 8-K filed after the window could prove a sale inside it | Say what counts as proof, where it must appear, and the default outcome when proof is absent |
An operator-grade pipeline, and where Adkuu Pulse sits
A fixed-odds operator does not get a token vote or a committee of public directors. It needs a pipeline it can run at volume and defend afterwards.
- A source hierarchy pinned at authoring: a primary data source with a known publishing schedule, a named fallback, then documented web evidence, then a consensus of named outlets, in that order.
- Timeouts per tier: if the primary has not published by a set time after the event, move to the next tier; if nothing has resolved by a later cut-off, route to a person.
- A human fallback with a decision log: the reviewer sees the rule, the evidence from each tier and the timestamps, and records why the outcome was chosen.
- A public rationale on every settlement: which tier decided, on what evidence, at what time.
- A void policy that refunds in full when the question cannot be answered under its own rule, decided before the market opens.
- Clarifications only before the event, with prices re-opened afterwards, following Polymarket's practice of clearing the book.
Adkuu Pulse implements the automated part of this as multi-tier resolution: a data API first, then web search, then consensus, with human review as the fallback, and it delivers each market to the operator as fixed odds priced with the operator's own margin. The mechanism is described on the answer page on how automated market resolution works and on the product page for Adkuu Pulse. How questions should be written so that the pipeline has something to resolve is the subject of the companion post on how AI-written betting markets fail.
The ten-point "is this question resolvable?" checklist
- Does the title say exactly what the rule says, no more and no less?
- Is there one event and one condition, or has a second condition (confirmation, publication, a filing) crept in?
- Is the primary source named, and does it publish on a schedule you can state?
- Is a fallback source named, with the order of precedence?
- Are the event window and the evidence window both written, with a time zone?
- Is the deadline after the whole timeline of the event: announcement, execution and confirmation?
- Is the outcome for postponement, cancellation and partial occurrence written?
- Is the evidence standard written: what counts as proof, where it must appear, and the default when it is absent?
- Is it written who decides when tiers disagree, and within how long?
- Will the rationale be published with the result?
FAQ
How does Polymarket decide a market's outcome? Through UMA's optimistic oracle. A proposer posts a resolution with a bond, anyone can dispute within a 2-hour challenge period by posting a bond of the same size (usually $750), and a dispute goes to a debate period of 24–48 hours and a UMA token-holder vote of about 48 hours. Polymarket's documentation says it cannot change or reverse a resolution once UMA finalizes it.
How does Kalshi settle a contract? Each contract names a Source Agency whose published value decides the Market Outcome, and the markets team finalizes most results within a few hours of the outcome being known, often within about 3 hours. If the outcome cannot be determined, the rulebook allows settlement at the last traded price or a binding allocation by the Outcome Review Committee within 24 hours.
Why do prediction market disputes happen? In the 2026 cases documented here, because the question was underspecified: the evidence window was not written, or the title promised something the rule did not. Ambiguity in the question, not uncertainty about the event, is the usual cause.
How does Adkuu Pulse resolve markets? With automated multi-tier resolution: a data API first, then web search, then consensus, with human review as the fallback.
Sources
- How Are Markets Disputed? — Polymarket Documentation
- How Are Markets Clarified? — Polymarket Documentation
- How does UMA's Oracle work? — UMA Documentation
- KalshiEX LLC Rulebook, version 1.15 (17 January 2025)
- Market FAQs — Kalshi Help Center
- How is Kalshi regulated? — Kalshi Help Center
- CFTC Staff Advisory: Prediction Markets Advisory, CFTC Letter No. 26-08 (12 March 2026)
- Polymarket's UMA vote upholds 'No' outcome on Strategy bitcoin sale market despite backlash — The Block
- A One-Day Gap, a $500K Bet, and the Governance Crisis Polymarket Can't Outgrow — Blockhead
- Kalshi's Botched Khamenei Market Could Be A Problem For Its Wall Street Ambitions — InGame
Last verified: October 10, 2026.