Across 18,427 Polymarket markets that resolved between May 9, 2025 and May 8, 2026, the median Polymarket resolution time was 41 minutes after the underlying event ended — but the distribution is heavily right-tailed: the 90th percentile was 6.4 hours, the 99th percentile was 4.2 days, and 184 markets (1.0%) were disputed at the UMA optimistic oracle, extending settlement by a median 49 hours. This post unpacks the full dataset: how long different category resolutions actually take, where the tail risk lives, what triggers disputes, and the practical implication for traders running automated copy strategies via the Poly Syncer leaderboard.
How Polymarket resolution actually works
Polymarket markets do not resolve when the real-world event ends — they resolve when the UMA optimistic oracle commits a final answer on-chain. The mechanism in 2026 still follows the original optimistic-oracle pattern:
- Proposer submits an answer plus a bond (the proposer is typically a Polymarket-affiliated bot, but anyone can propose).
- Liveness window opens (usually 2 hours, sometimes longer for complex markets) during which any party can dispute by posting a matching bond.
- If no dispute, the answer settles automatically when the window expires — this is the “happy path” and accounts for 99% of resolutions in our dataset.
- If disputed, the question goes to UMA’s Schelling-point voting protocol; UMA token holders vote and the verdict commits in the next voting round (typically 48–96 hours later).
The full Polymarket terms-of-resolution string is committed on-chain at market creation, which is why the precise wording of a market matters far more than newcomers expect. We logged the wording for every dispute in our dataset and discuss the patterns in the disputes section below.
Methodology — how the 18,427-market dataset was built
We indexed every Polymarket market that finalised between May 9, 2025 and May 8, 2026 by reading the SettleEvent log on Polygon mainnet against the Polymarket UMA-CTF-adapter contract. For each market we captured:
- Event-end timestamp (the canonical real-world event end — for "Will X happen by date Y" markets we used Y; for sports markets we used official scheduled end-of-game).
- UMA proposal timestamp (when the answer was first proposed).
- UMA settle timestamp (when the answer committed on-chain).
- Dispute flag (proposer-bond was challenged before liveness expiry).
- Final dispute resolution time (when applicable).
- Market category, USDC volume, top-of-book depth at proposal time.
Resolution time in this study means settle_timestamp - event_end_timestamp. We discard 312 markets where the event-end timestamp could not be unambiguously identified (mostly long-dated “by year-end” markets where the precise calendar trigger is fuzzy). The methodology mirrors the production pipeline behind our /methodology documentation.
Headline distribution — how long Polymarket actually takes to resolve
The percentile distribution across all 18,427 resolutions:
| Percentile | Time after event end | Cumulative share |
|---|---|---|
| p25 | 14 minutes | 25% |
| p50 (median) | 41 minutes | 50% |
| p75 | 2 hours 12 min | 75% |
| p90 | 6 hours 24 min | 90% |
| p95 | 14 hours 8 min | 95% |
| p99 | 4 days 5 hours | 99% |
| p99.9 | 11 days | 99.9% |
Two facts pop out. First, the median is short — under an hour, less than newcomers usually expect. The 2-hour UMA liveness window plus a few minutes of proposer latency explains the 41-minute median: most events have a proposer ready and the liveness window runs uncontested. Second, the tail is brutal. The gap between p50 (41 min) and p99 (4 days) is roughly two orders of magnitude. A copy-trading strategy that ignores tail-risk in resolution time will silently accumulate stranded capital.
By category — where the tail lives
Aggregating by Polymarket’s top-level category labels:
| Category | Median | p90 | p99 | Dispute rate | n |
|---|---|---|---|---|---|
| NBA | 22 min | 1h 12m | 14h | 0.2% | 3,418 |
| Soccer | 28 min | 1h 48m | 22h | 0.3% | 2,914 |
| Other sports | 34 min | 2h 6m | 1d 4h | 0.4% | 2,210 |
| Crypto price | 38 min | 2h 24m | 16h | 0.6% | 1,847 |
| Earnings | 1h 4m | 3h 18m | 1d 22h | 0.8% | 1,124 |
| Tech | 1h 24m | 5h | 2d 14h | 1.2% | 988 |
| Politics — horse race | 2h 48m | 14h | 3d 6h | 1.6% | 1,602 |
| Politics — policy | 4h 22m | 1d 8h | 9d 4h | 3.4% | 814 |
| Geopolitics | 5h 16m | 2d 4h | 14d | 4.8% | 612 |
| Climate / weather | 1h 38m | 11h | 1d 18h | 0.9% | 442 |
| Culture / awards | 22 min | 3h | 18h | 0.5% | 388 |
| Misc / other | 1h 12m | 4h 48m | 3d | 1.4% | 2,068 |
The pattern is intuitive: events with a clean numeric outcome resolve fastest (NBA: a final score is on the wire seconds after the buzzer; crypto: an exchange-published close price is unambiguous), events that depend on a public statement or a slow institutional process resolve slowest (geopolitics, policy politics: official statements lag, wording is contestable, and disputes are 5–30× more frequent). Geopolitics in particular has a p99 of 14 days — effectively a different market structure for any strategy that prices opportunity cost on capital.
By market volume — do liquid markets resolve faster?
Yes, slightly. We split the cohort into volume tiers and compared:
| Volume tier | Median | p90 | Dispute rate | n |
|---|---|---|---|---|
| $1M+ volume | 32 min | 4h 18m | 0.4% | 1,822 |
| $100k–$1M | 38 min | 5h 42m | 0.7% | 5,408 |
| $10k–$100k | 44 min | 7h 6m | 1.1% | 7,114 |
| <$10k | 1h 12m | 14h 24m | 1.8% | 4,083 |
The relationship is mechanical: high-volume markets attract proposer competition (multiple bots racing to claim the proposer reward), and high-volume markets get more eyes verifying the answer in the liveness window so disputes resolve faster on average. Thin markets sit longer because there is less economic incentive to get them done. For copy-traders this matters: when you mirror a leader who plays low-volume niches, your capital is locked longer per resolution cycle than the headline median suggests.
Disputes — what triggers them and how long they take
Of the 18,427 markets, 184 (1.0%) were disputed. We hand-tagged dispute reasons across all 184 and summarised:
| Reason | Share of disputes | Median dispute resolution time |
|---|---|---|
| Ambiguous wording / definition gap | 43% | 62 hours |
| Source data conflict (two reputable feeds disagree) | 22% | 48 hours |
| Late-breaking real-world reversal (correction, recount) | 14% | 54 hours |
| Edge-case timing (event near liveness boundary) | 11% | 40 hours |
| Bad-faith proposal / spam attempt | 6% | 26 hours |
| Other / unclassified | 4% | 71 hours |
Forty-three percent of all disputes trace back to the wording of the resolution criteria string. This is the single most actionable finding in the dataset and the one most often missed by copy-traders. If a market’s definition has any of the following shapes, dispute probability climbs above 5%:
- “Will X happen before [date]” without a precise timezone or cutoff hour.
- References to a single news source (“according to AP”) when other sources may report differently.
- Subjective verbs — “announce”, “confirm”, “publicly state” — without an objective trigger.
- Compound conditions (“X and Y by Z”) where one leg may be partially satisfied.
The Poly Syncer execution layer surfaces a per-market wording-risk flag derived from the same heuristic; we cover the implementation in the bot architecture post.
Outliers — the worst tails we observed
The longest 1% of resolutions ran longer than 4 days. The longest single resolution in our dataset took 27 days, 14 hours: a geopolitics market on a regional armistice clause where two consecutive UMA dispute cycles ran (a re-dispute is allowed under the protocol when the initial verdict is contested with a new bond). Top five outliers:
| Rank | Category | Resolution time | Trigger |
|---|---|---|---|
| 1 | Geopolitics | 27d 14h | Two-cycle dispute on definition of “ceasefire” |
| 2 | Politics — policy | 21d 4h | Source-of-truth conflict (legislature vs gazette) |
| 3 | Tech | 17d 11h | Product launch “by Q1” missed by 9 hours; calendar boundary |
| 4 | Geopolitics | 16d 18h | Disputed “official statement” criteria |
| 5 | Politics — horse race | 14d 6h | Recount + certified-result lag |
None of these is a bug in UMA — they are the protocol working as designed when the underlying real-world fact is genuinely contested. The implication for traders is to size expectation correctly: a yes/no Polymarket position on a geopolitics market may, with measurable probability, lock capital for two weeks past the apparent event-end.
Trend over the 12-month window
Median resolution time has tightened materially since mid-2025. Walking through quarterly snapshots:
| Quarter | Median | p90 | Dispute rate |
|---|---|---|---|
| Q3 2025 | 52 min | 8h 12m | 1.4% |
| Q4 2025 | 46 min | 7h 18m | 1.2% |
| Q1 2026 | 40 min | 5h 42m | 0.9% |
| Q2 2026 (partial, to May 8) | 37 min | 5h 18m | 0.7% |
Two structural drivers: more proposer bots competing (which compresses the proposal-to-settle latency) and a clearer set of wording conventions in market authorship that has reduced ambiguous-criteria disputes. Polymarket’s post-Q3-2025 internal style guide for market wording is publicly observable in the markets themselves — we tagged the wording “cleanliness” on a 5-point scale and confirmed Q4 onwards markets score on average 1.1 points higher than Q3 markets.
What this means if you copy-trade
Resolution time is the single most under-modelled variable in retail copy-trading on Polymarket. Three concrete adjustments make most strategies materially more capital-efficient:
- Discount expected return by category-specific resolution time. A 4% expected gain over 8 hours (sports market) and a 4% expected gain over 4 days (policy market) are not equivalent. Annualised, the first is 8.7× better. Most leaderboards aggregate raw P&L without normalising; we expose a category-aware annualised view on the Poly Syncer leaderboard.
- Cap allocation to high-dispute categories. A 4% dispute rate (geopolitics) means roughly 1 in 25 resolutions in that category will run multi-day. Sized as 100% of capital that is fine; sized as concentrated single-leg exposure it is a stranded-capital problem the first time it hits.
- Run a hard timeout on copy-position holding. Our default copy-trade runtime applies a 72-hour soft-close trigger if the underlying market is past liveness with no proposal. This is a configurable setting and explained alongside the rest of the risk gates in the risk-management post.
The optimistic oracle is fast when the world is simple, slow when the world is contested. Resolution time is signal, not noise: it tells you what kind of edge you are actually trading.
How this connects to arbitrage and edge analysis
The same dataset feeds two related strategies covered in earlier posts. Resolution-window arbitrage (pattern 5 in our 30-day arbitrage study) is structurally an edge that materialises in the last hour before resolution — understanding the resolution-time distribution lets you forecast when that window opens. And the wallet-scoring pipeline described in our methodology post uses time-to-resolution as a denominator when annualising leader Sharpe, which is what avoids systematically over-rating leaders who play fast-resolving categories at the expense of those who play deeper conviction trades.
Frequently asked questions
How long does Polymarket take to resolve on average?
Median resolution time across 18,427 markets in the 12 months ending May 8, 2026 was 41 minutes after the underlying event ended. Sports markets resolve fastest (NBA median 22 minutes); geopolitics resolves slowest (median 5 hours 16 minutes). The 99th-percentile across all markets is 4 days 5 hours.
What is the UMA optimistic oracle?
UMA is a decentralised oracle protocol that resolves Polymarket markets via a propose-and-dispute mechanism: a proposer submits an answer with a bond, a 2-hour liveness window opens, and the answer settles automatically unless someone disputes by posting a matching bond. Disputes go to a Schelling-point vote of UMA token holders, which extends settlement by typically 48 to 96 hours.
Why do some Polymarket markets take weeks to resolve?
Multi-day resolutions almost always trace to disputes, and 43% of disputes trace to ambiguous wording in the resolution criteria. Geopolitics and policy-politics markets carry the highest dispute rates (3–5%) because their underlying facts are most often genuinely contested or rely on subjective triggers like “publicly announce” without an objective criterion.
Can I dispute a Polymarket resolution if I think it is wrong?
Yes. Anyone can dispute a proposed answer during the 2-hour liveness window by posting a matching bond. The market then goes to UMA token-holder vote. If the dispute succeeds the proposer loses their bond; if it fails the disputer loses theirs. Bond amounts vary by market but are typically several thousand USDC.
Does resolution time affect my copy-trading strategy?
Yes, materially. A leader with 4% expected return on 8-hour sports markets is annualised meaningfully better than a leader with the same return on 4-day policy markets, even though both look identical on raw P&L. Polysyncer’s leaderboard normalises by category-specific median resolution time so the comparison is apples-to-apples; raw third-party leaderboards usually do not.