How Prediction Markets Decide Who Wins
Every prediction market has an oracle — the source that decides the final outcome. Most traders never think about it until something goes wrong.
Oracle failures aren't rare. Mention markets, pre-taped show markets, and single-source weather contracts have all produced disputed resolutions. This guide explains who decides, how they decide, and where the risk is highest on each major platform.
The Three Oracle Types
Automated API
Outcome determined automatically by a data feed (price, weather station, government release). Fast but vulnerable to source outages, stale data, or single-point failures.
Examples: Weather markets (NOAA feed), crypto price markets (single exchange close)
Human Committee
A panel of researchers or operations staff reviews multiple sources and makes the final call. Slower, but more resilient to single-source failures and ambiguous outcomes.
Examples: ForecastEx macro and political markets
Hybrid
Automated feed triggers initial resolution; a human team or decentralized oracle (e.g., UMA) can override or dispute. Balances speed with error correction.
Examples: Kalshi politics (AP call + ops override), Polymarket (UMA dispute mechanism)
Oracle Structure by Platform
Each platform handles resolution differently. The same event type may carry different oracle risk depending on where you trade it. Compare platform fees and Kalshi vs Polymarket alongside oracle risk.
Kalshi has no formal trader dispute mechanism. Traders may contact support or post in Discord, but Kalshi's internal markets team makes all final resolution decisions. An Outcome Review Committee (board committee) may be invoked at Kalshi's sole discretion.
Kalshi Khamenei market (Feb 28, 2026) — death carveout invoked on $54M market after Khamenei killed in US-Israeli strikes; resolved at last-traded-price rather than YES. Kalshi paid $2.2M in refunds. New death-settlement rule codified March 17, 2026.
Primary Resolution Sources
| Category | Source | Fallback |
|---|---|---|
| economic | BLS.gov official release | Federal Reserve official statement |
| weather | NOAA ASOS station data | NCEI Climate Data Online, then NWS Daily Climate Report (CLI product) |
| crypto | CF Benchmarks Real-Time Indices (RTIs) — 60-second average of per-second observations at expiry | None named |
| politics | AP race call | major wire services |
Market Examples
Fed rate cut by March meeting
Low RiskOracle: Federal Reserve official statement (federalreserve.gov)
Government primary source, no ambiguity about where data comes from
BTC above $100K on Dec 31
Medium RiskOracle: CF Benchmarks RTI (60-second average at expiry)
CF Benchmarks aggregates multiple exchanges — more robust than a single exchange close, but CF Benchmarks is still a single index provider with no named fallback. No formal trader dispute mechanism at Kalshi.
Survivor Season 48 winner
⚠ Very High RiskOracle: CBS broadcast (pre-taped show)
Pre-taped outcome means cast/crew contacts have permanent information advantage before market closes
What Is Single-Source Oracle Risk?
When a market resolves based on one data source or one reporter, you have no recourse if that source:
- •Goes offline at resolution time
- •Reports a stale or incorrect value
- •Is influenced by parties with a position in the market
How to spot it:
Look for "will resolve to [single source]" language in the contract rules. If only one source is named with no fallback, that's single-source risk.
Categories with highest single-source exposure: Weather markets (NOAA ASOS only), crypto price markets (single exchange close), mention markets (one human reviewer).
How Resolution Actually Flows
From event to payout, seven things have to happen. Each step is a potential failure point.
Event Occurs
The underlying event happens in the real world.
Source Reports
The oracle data source (API, publication, committee) reports the outcome.
Platform Reads Source
The platform retrieves the data — automatically or via human review.
Resolution Check
Automated logic or operations staff verify against contract criteria.
Resolution Posted
Platform marks the market resolved Yes or No.
Dispute Window Opens
Traders have a limited window to challenge the resolution.
Funds Distributed
Winning positions receive payouts after dispute window closes.
Event Occurs
The underlying event happens in the real world.
Source Reports
The oracle data source (API, publication, committee) reports the outcome.
Platform Reads Source
The platform retrieves the data — automatically or via human review.
Resolution Check
Automated logic or operations staff verify against contract criteria.
Resolution Posted
Platform marks the market resolved Yes or No.
Dispute Window Opens
Traders have a limited window to challenge the resolution.
Funds Distributed
Winning positions receive payouts after dispute window closes.
Why Platforms Disagree on Outcomes
Even when the underlying event is clear, different platforms can post different resolutions — because each contract names a different source. The display chart is not the settlement source, and when they diverge, the contract wins every time.
The #1 source of "feels rigged" complaints on Coinbase and Robinhood: the display chart is NOT the settlement source. They're separate systems — and when they diverge, the contract wins every time.
The core disconnect
What You See
- Live chart on the market page
- App price display / feed
- Market probability %
- Third-party data overlays
What Determines Your Payout
- Specific data source named in the contract
- Verified at settlement time
- Platform resolution rules
- Oracle or committee decision
These are the same number 95% of the time. When they differ, you lose money.
Real mismatch examples
What traders see
Weather.com app shows 72°F for the day — traders assume this is the settlement reading.
What the market uses
NOAA station KPHL hourly reading at the specified time — NOT the forecast, NOT the peak, NOT your local app.
"I checked the weather all day, it never hit 75°F. Then the market settled YES. How?"
Kalshi & Polymarket
Resolution sources are clearly documented in each market contract. Weather = NOAA station. Politics = AP call. Finance = official closing price. Verify once per new market category — the pattern is consistent. For the April 2026 Polymarket Paris temperature incident (oracle swapped from Roissy-CDG to Le Bourget), see our dedicated explainer at the Roissy incident page.
Coinbase & Robinhood
The display UX often obscures the settlement source. An extra verification step is required before trading. See how settlement sources are chosen for the 60-second pre-trade check.
When Markets Resolve Wrong: Your Recourse
You held the right position. The event happened. But the platform resolved the wrong way — or you didn't get paid what you expected. Here's what's actually going on, and what (limited) recourse you have.
First: what does "resolved wrong" actually mean?
Most resolution disputes fall into one of four buckets. Understanding which one you're in changes what you can do about it.
Platform disagreed with public facts
The event clearly happened one way, but the platform resolved differently. This is the rarest case — and the one you have the most recourse on.
→ File a formal dispute. Document your sources.
Rule carveout you didn't read
The platform's resolution criteria had an edge case (death, weather station outage, etc.) that changed the payout. Legally correct by their terms, but feels wrong.
→ Limited recourse. Read market rules before trading large.
Ambiguous event outcome
The underlying event was genuinely unclear — contested news, timing dispute, ambiguous criteria. Both sides have a legitimate case.
→ Dispute mechanism is your path. Outcome uncertain.
You misread the contract terms
The market resolved correctly per the rules, but it didn't match what you expected. This is more common than most traders admit.
→ No recourse. Learn the resolution criteria before next trade.
Real Example: Kalshi Khamenei Market (Feb 28, 2026)
Khamenei died in US-Israeli strikes. Kalshi invoked a "death carveout" on the $54M+ market, resolving at last-traded-price instead of YES. Kalshi reimbursed all fees and net losses — at a total cost to Kalshi of approximately $2.2M — so no trader ended net-negative on their original cost. However, traders who held YES contracts did not receive the full $1.00 payout they expected. Kalshi later codified a new death-settlement rule on March 17, 2026. Traders were made whole on cost, but not on expected winning upside.
Resolution risk by scenario
How risky is each scenario on the major platforms?
| Scenario | Kalshi | Polymarket | Others | Risk |
|---|---|---|---|---|
| Oracle data source goes offline | Low — internal team reviews manually | Medium — UMA token-holder vote required | High — no fallback defined | Medium Risk |
| News event is ambiguous / disputed | Low — Kalshi Ops makes final call | Low-Med — 2-hour dispute + UMA escalation | High — no formal process | Higher Risk |
| Payout is delayed past expected date | Low — CFTC-regulated, contractual obligation | Low — USDC on-chain, auditable | High — recourse unclear | Medium Risk |
| Platform resolves YES when you think NO | Medium — no formal dispute, support only | Low — $750 USDC bond to escalate to UMA | High — varies widely | Higher Risk |
| Platform reverses resolution after payout | Low — happened once (Khamenei market, refunded) | Very Low — on-chain immutability after UMA | Unknown | Lower Risk |
Your 3-step action guide
Identify which bucket you're in
Is this a clear factual error, a rule carveout, an ambiguous event, or a misread contract? Your bucket determines your options. If you misread the contract, recourse is essentially zero.
Document everything immediately
Screenshots of the market page, resolution notice, and your position. Contemporaneous news sources showing the event outcome. Timestamps matter. Most dispute windows are short.
File through official channels — then escalate if needed
Kalshi: Contact support + Discord, then CFTC (cftc.gov/complaint) if no resolution. Polymarket: Post $750 USDC bond to escalate to UMA within the 2-hour window. Others: Support email + public pressure. Never pay third-party "dispute services" — they're scams.
Honest expectations
CFTC-regulated platforms (Kalshi) have real accountability. Your dispute isn't ignored — it could become a regulatory matter.
Polymarket's UMA system is decentralized and auditable. If the resolution is verifiably wrong, the bond system rewards you for disputing.
Most resolution disputes are lost by traders, not platforms. Platforms write the rules. Edge cases almost always favor the house interpretation.
"I was right about the outcome" does not mean "I should have won the contract." Market resolution criteria and real-world outcomes are not always identical.
Frequently Asked Questions
New to prediction markets? Here's what you're actually looking at.
Verified accounts on CFTC-regulated exchanges. Each platform posts its regulatory basis — see the platform bar for specifics.
Market concentration is flagged automatically. Look for the BROAD/CONCENTRATED/THIN badge — concentrated markets carry more single-actor risk.
Every market has a published resolution rule. You can read the exact trigger condition — it's a contract with fixed terms, not a bet against the house.
Related Guides
Resolution Dispute Audit
Per-platform evidence trails and dispute steps
How Markets Settle
Settlement rules and timing by platform
Contract Compare
Are two contracts actually the same bet?
Insider Trading Guide
When insider risk is real vs. overstated
How Concentrated Is Prediction-Market Profit?
Why account-level concentration is different from user-level profitability
How Prediction Markets Work
Fundamentals: contracts, pricing, and mechanics
All Platforms Directory
Compare every CFTC-regulated prediction market
Why Oracle Design Creates Ethical Risks
Single-source oracles, journalist threats, and structural problems
Why Did My Payout Drop?
Cash-out vs. resolution: two separate mechanisms that users conflate