How Polymarket Settles: 110,576 Markets, a Median of 2.2 Hours, and 89 Disputes

Most Polymarket markets settle within a few hours, and very few are fought over. Vultax pulled every market Polymarket listed under nine major topic tags whose scheduled end fell between 15 August and 10 September 2026, excluding the short crypto up-or-down series: 110,576 markets that traded $1.42 billion. Of the 69,072 that closed after their scheduled end, half had closed within 2.2 hours and 90% within six. Another 41,151, 37% of the total, closed before their scheduled end because the answer was already known.

Disputes were rare: 89 markets, 0.08%. They were not spread evenly. Markets that traded $100,000 or more were disputed six times as often, 0.5%, and geopolitics markets, whose questions meet messy events, 3.5% of the time, with a median wait of 74 hours after their scheduled end.

Almost every market, 99.7%, is marked in Polymarket's data as resolved automatically. The exceptions sit in politics, where 8.5% needed something else.

A second pull, of markets that ended between 11 and 18 September, put the median at 4.0 hours. Settlement had not slowed: 98% of those listings were sports, and sports took 4.0 hours in both windows. The September check sets the two side by side.

Polymarket does not decide outcomes itself. According to its documentation, a proposer posts the answer with a bond, typically $750, and a two-hour challenge period follows. Anyone can dispute by posting a matching bond; a first dispute triggers a fresh proposal, and a second sends the question to a vote of UMA token holders, a discussion period of 24 to 48 hours followed by a vote of about 48 hours. Polymarket's help centre says that once UMA finalises an outcome it cannot be altered or reversed.

Two details changed how that process looks in practice. Since 2025, proposals on Polymarket can come only from whitelisted addresses; UMA reported that in the first month of enforcement incorrect proposals fell 59% and disputes 68%. And in the markets Vultax measured, 58,318 carried custom challenge windows of 10 to 60 minutes rather than the two hours the documentation describes, most of them crypto, sports and esports markets.

An event ending does not make a market final, and a market being final does not say whether a particular holding has been redeemed. Four checks on the market's own page place it, and each answers a different question.

The 2.2-hour median describes 69,072 markets that closed after their scheduled end in the study window. It cannot say that one unresolved market will be final within that time: the clock here runs from the listing's scheduled end to the closing time in Polymarket's data, and individual redemptions were not tested.

Vultax paged through Polymarket's public list of closed events for the tags politics, sports, esports, economy, crypto, culture, geopolitics, tech and finance, kept each market whose own scheduled end date fell between 15 August and 10 September 2026, and dropped the recurring crypto up-or-down series. For each market it read the scheduled end, the time the market closed, the history of its resolution states and whether it was marked as automatically resolved. A market counts as disputed if its history includes a dispute. Settlement time is the closing time minus the scheduled end.

Short, mechanical questions settle fastest: crypto price markets within minutes of their end. Politics and geopolitics take longest and are disputed most, because deciding whether something happened can itself be contested.

Download table figures:CSVJSON with sources

Download chart figures:CSVJSON with sourcesSVG chart

Disputes rise with money at stake, which makes sense: a dispute costs a bond, and it is worth posting only when the outcome is worth arguing about. Settlement did not slow down with size.

Download table figures:CSVJSON with sources

The largest disputed markets in the window share a feature: a question whose wording had to be matched to a real event after the fact.

• Iran - "US announces halt in Iran offensive operations by July 31?", $4.0 million traded, disputed twice.
• Player moves - "Will Maxx Crosby play for Indianapolis Colts next?", $1.15 million, disputed twice, and the matching Rams market, $406,000, three times.
• Presidential language - "Will Donald Trump publicly insult someone on August 23, 2026?", $826,000, disputed twice and closed five days after its scheduled end.
• Military incidents - "NATO downs another Russian drone by August 31?", $486,000, disputed twice.
• Esports - a League of Legends Challengers game-winner market, $329,000, disputed three times.

On 19 September 2026 Vultax pulled Polymarket listings with a scheduled end between 11 and 18 September, open as well as closed: 121,432 markets once the up-or-down series is set aside. Of those, 69,630 had closed at or after their scheduled end, 50,334 had closed early, and 1,468 were still open when read.

The open ones are the new information, because the main window above counts closed markets only. They are few, and they sit where the rules are doing the work.

Download table figures:CSVJSON with sources

Download chart figures:CSVJSON with sourcesSVG chart

• An end date is a listing field, not a deadline - 1,468 listings were still open after their scheduled end, 230 of them politics markets, a quarter of the politics listings in the pull.
• The rules decide - Polymarket's next-prime-minister-of-Sweden markets carry a 13 September end date, the day of the election, and pay when a prime minister is appointed and takes office. They were open and taking orders on 19 September, with $1.6 million to $2.0 million traded in each of the three largest.
• Closing early is normal - 41% of the September listings closed before their scheduled end, against 37% in the main window.

Of the 69,630 September listings that closed at or after their scheduled end, half had closed within 3.98 hours and 90% within 5.1; 678 took more than a day and 251 more than three. That median is higher than the 2.2 hours above because the mix is different, on the same clock: closing time minus scheduled end.

The main window was 35% sports, 33% esports and 23% crypto, and the last two settle in under an hour. The September pull read Polymarket's full market list rather than nine topic tags, stopping at 120,000 closed records in listing order, and 118,597 of the 121,432 listings it kept were sports. Sports took a median 4.0 hours in the main window and 3.98 here.

Category by category the two windows agree where the counts are large, and move where they are small: politics and geopolitics rest on 438 and 32 closed markets in the September pull.

Download table figures:CSVJSON with sources

Download chart figures:CSVJSON with sources

The disputes that made headlines were larger and older. In March 2025 a $7 million market on a Ukraine–U.S. minerals deal resolved Yes before any agreement was confirmed, and Polymarket declined refunds. In July 2025 UMA resolved a $237 million market No on whether Volodymyr Zelenskyy had worn a suit, despite many media reports describing his outfit as one. The whitelisting of proposers followed that period.

Kalshi works differently: as a CFTC-regulated exchange, its own markets team determines outcomes under each contract's stated rules and sources, without a token-holder vote. That difference is one reason the same question can settle on different terms on the two venues.

Three practical points follow from the numbers.

• Money comes back fast on most markets - nine in ten markets that ran to their scheduled end had settled within six hours of it.
• Long waits cluster in politics and geopolitics - one politics market in ten took more than four days after its scheduled end, and geopolitics took a median of three.
• Read the rules before a big position - disputes are rare, but they concentrate where wording meets events, and a clarification can change how a market is judged.

Markets were found through nine topic tags, so markets in events carrying none of them are missing, and a market carrying several tags is counted once under the first in the listed order. The scheduled end is the date set in the market's listing, and the closing time is when Polymarket's data marks the market closed; payouts can be claimed around that time, but Vultax did not test individual claims. A dispute is counted from the market's resolution-state history, which does not show who disputed or why. Bond sizes and challenge windows are as described in Polymarket's and UMA's documentation, with the custom windows read from market data. Nothing here is a recommendation to trade. Challenge a figure at [email protected]; material changes are logged in the changelog.

The September check is downloadable in full: every listing with its end date, closing time and status (33 MB), the summary figures, the method notes and a reproduction bundle (61 MB) with a SHA-256 manifest.