When Polymarket disputes a UMA resolution, the result almost never changes
A dispute sounds dramatic. On-chain, it usually isn’t: most Polymarket UMA disputes are procedural noise that confirm the original answer — not corrections that overturn it.
Of 432 resolved UMA disputes on Polymarket over the last ~208 days, only 13.9% (60) reversed the originally proposed answer. The other 86% simply confirmed it.
Polymarket settles markets through UMA’s Optimistic Oracle: a proposer states the outcome, and anyone can dispute it, escalating to a token-holder vote. The count of disputes is public. What no tool reports is the part that actually matters to a trader or a builder: when a market is disputed, does the answer change? We measured it on-chain by pairing each dispute’s originally proposed price against the final resolved price.
The two-axis finding: disputes and reversals rank categories in opposite orders
Every Polymarket tool looks at one axis — how often a market is disputed. Measure two, and they tell opposite stories:
| Category | Disputes often? | Flips when disputed? |
|---|---|---|
| Geopolitics | Highest (top of both samples) | 5.9% (n=17) |
| Politics | High | 5.3% (n=57) |
| Sports | Lowest (~0% in both samples) | 38% (n=50) |
Geopolitics markets are contentious but stable — they dispute constantly because they’re ambiguous, yet the original answer almost always survives. Sports markets barely ever dispute, but when they do it’s a genuine correction (a real scoring or rules error). A one-axis read would mislead you in both directions.
Why we publish the rank, not a single dispute rate
We measure dispute propensity in two samples — the ~1,500 highest-volume resolved markets and the ~1,500 most recent. The absolute number swings several-fold between them (Geopolitics is 35.1% in the volume sample vs 5.1% in the chronological one; Politics 16.1% vs 1.1%). What stays put is the order: Geopolitics at the top, Sports at the bottom (~0%), in both. So the rank is the fact; the isolated percentage is not. We publish both samples so you can check it.
A third signal: the question’s criterion
Markets with a subjective criterion (“permanent peace,” “will he cry?”) dispute far more than those with an objective one (“above $150k,” “Team A vs Team B”) — a gap of roughly 20× in the high-volume sample (27.4% vs 1.4%) and about 3× in the chronological one.
This axis is orthogonal to category: it discriminates within a category (a Fed market phrased as “25 bps” versus “Netanyahu out of office”). The direction is robust in both samples; the magnitude is sample-dependent.
What this is — and what it is not
This is resolution-risk metadata: for a market of this category, with this kind of criterion, here is how often comparable markets historically dispute and reverse. It is a base-rate, a descriptive fact. It is not a prediction that a specific market will be disputed, and it is not an accusation of manipulation against anyone. The trader or the agent sets the threshold; we state the structural fact the order book doesn’t.
Methodology (open, falsifiable)
-
Flip:
eth_getLogson Polygon for UMA OOv2DisputePrice(the proposed answer) versus the UmaCtfAdapterQuestionResolved(the final answer), paired byquestionID = keccak256(ancillaryData). A flip = first-cycle proposal ≠ final resolution. 507 disputes, 433 resolved in-window, 0 decode errors. -
Dispute base-rate: Polymarket Gamma
umaResolutionStatuses, two samples (by volume and chronological) published side by side because the absolute rate is sample-dependent. - Declared limits: per-category cells are small (disputes are rare) and therefore directional; the overall flip rate (n=432) is robust. Category is a declared title heuristic, not an official taxonomy. DVM voter concentration (press reports ~9 wallets controlling >50% of the UMA vote) is deferred, cited not recomputed.
GET api.x402intel.uk/resolution-risk and
/resolution-watchlist ($0.20 USDC/call on Base). Pre-trade resolution-risk
metadata, fact not prediction. The underlying numbers on this page are also available as a
free JSON dataset.