Concept · oracle-and-resolution
Oracle Design
Quick definition. Oracle design is the set of choices a prediction market makes about how an off-chain outcome becomes an on-chain settlement · who reports it, who can dispute it, what data sources are canonical, and what economic bonds back honesty. It is the most-cited concept on onprediction.xyz (referenced in 32 articles), because every other property of a prediction market · accuracy, manipulation-resistance, regulatory posture, liquidity · bottlenecks here.
Key insights
- Resolution, not pricing, is the hardest problem. Andy Hall (a16z): "The hardest problem in prediction markets isn't pricing the future. It's deciding what actually happened." Pricing is solved by liquidity; resolution is solved by mechanism design we haven't shipped yet.
- Four properties any resolution mechanism must satisfy at once (Hall): manipulation-resistance, reasonable accuracy, ex ante transparency (rules visible before the bet), and credible neutrality. Human committees fail at neutrality and scale; token voting fails at manipulation-resistance; centralized operators fail at transparency. Nothing on the market today satisfies all four.
- *The "Oracle Problem" is misnamed · it's a definition problem.* Aradtski catalogued 12+ Polymarket controversies and argues every single one was about ambiguous event wording, not oracle untrustworthiness. The Hayekian price-discovery argument breaks down when the question itself is semantically contested.
- There is a resolver trilemma between efficiency, decentralization, and security (XO Labs). Optimistic oracles with token voting trade security for efficiency; centralized operators trade decentralization for both; AI judges trade some accuracy for transparency. No single mechanism dominates · different markets need different rungs of the ladder.
- The corruption-cost vs. value-at-stake gap is the central oracle failure mode. Repeated case studies (government shutdown, Zelensky suit, Ukraine map, Venezuela invasion, Cardi B halftime, TikTok ban) all share the same shape: the value at stake on the market exceeded the cost of corrupting the oracle. Until that ratio inverts, oracle-driven settlement is exploitable in principle on every large market.
- Optimistic oracles (UMA) have a known capture surface. Token-holders who can both vote on resolutions and trade the underlying market are structurally conflicted. Frank Muci's Venezuela post-mortem documents UMA holders coordinating to overrule Polymarket's own stated rules. Lou Kerner's 2024 conspiracy walks through how a Trump-favoring UMA whale could profit twice · once on the trade, once on the vote.
- Semantic ambiguity is the real adversary. Matt Levine's framing: "Prediction markets need linguists and philosophers as much as traders." If the contract reads "Will Zelensky wear a suit?", what is a suit? is now a $240M question with no objective answer. The market becomes a referendum on definitions, not facts.
- Source-of-truth attacks are the next attack surface. When the resolution criterion points at a single URL (an OPM website, a niche map app, an unguarded thermometer near CDG airport), adversaries attack the data source, not the oracle. Gunitsky's "hair-dryer problem": $14,000 won by warming a single thermometer; the oracle was honest, the underlying signal was not.
- AI/LLM judges with on-chain commitment are the leading proposed fix. Hall's a16z proposal: at market creation, commit the exact LLM model version and prompt to chain. Resolution is then deterministic, transparent, and bribery-resistant (you cannot bribe model weights). The model doesn't have to be perfect · it has to be predictable. Traders can then price in its biases the way they price in any other contract parameter.
- AI judges trade one set of vulnerabilities for another that may be more tractable. Limitations: hallucination, prompt-engineering failures, source-poisoning attacks against the inputs the model reads, training-data poisoning over long horizons. But none of these involve real-time bribery of a $240M vote.
- There are five MEV-style edges around oracles (st1ne): (1) oracle latency arbitrage · trading on news before UMA updates; (2) resolution arbitrage · front-running settlement; (3) dispute sniping · gaming the UMA dispute liveness window; (4) order-book imbalance exploitation around expiry; (5) conditional probability arbitrage across correlated markets. Polymarket is "a hidden MEV playground" · sophisticated actors extract value from oracle structure, not informational edge.
- Decentralization matters most where the data must be transported on-chain. Sonal/Tabarrok/Kominers (a16z podcast): if the resolution data already lives on-chain (price feeds, smart-contract outputs), you don't need a decentralized oracle for that question. The decentralization budget should be spent on the off-chain-to-on-chain transmission, where MNPI extraction and manipulation cluster.
- Layered architectures route each market to the security tier it needs (XO Labs). Low-stakes meme markets can be admin-resolved or AI-resolved cheaply; high-stakes geopolitical markets need expensive decentralized adjudication. Treating resolution as a single mechanism is what produced the $500M tail of resolution failures.
- Cross-platform divergence is now its own oracle failure mode. The Cardi B halftime market resolved YES on Polymarket and NO on Kalshi · same event, two answers, two settlement systems, no reconciliation. michaellwy: prediction markets are fundamentally non-fungible because each market is "an event + a referee system," and the referees disagree.
- Resolution disputes can themselves become a venue for state-actor manipulation. Gunitsky: regimes with information-warfare budgets ($billions/year) can manipulate either trading prices or the disputed-resolution mechanism for propaganda value. A million-dollar loss on a propaganda operation is rounding error.
- Some emerging designs sidestep oracles entirely. Self-resolving markets (SKC, Monad explainer) use the crowd's own final consensus as the reference outcome. Parimutuel-with-AI-resolver designs (Turf/sdk.markets) let any URL serve as ground truth via an LLM oracle. These are aimed at long-tail/subjective markets that on-chain oracles fundamentally cannot serve.
- Proprietary centralized oracles with platform skin-in-the-game may outperform decentralized mercenary capital (Aradtski). The argument: a centralized operator whose brand dies if it cheats is more aligned than a swing-vote of mercenary token-holders whose downside is bounded by token forfeiture.
- Insurance/cover markets are a stress test for oracle design. Eigenmann (HIP-4): when binary event contracts share margin with the underlying exposure they hedge, oracle quality directly determines whether the contract acts as parametric cover or as a casino. The bar is much higher than for entertainment markets.
- The 7 axes of competition include oracle reliability as a first-class differentiator (Nyquist). New entrants can compete on oracle design itself · credibility tokens, layered resolvers, AI judges · as much as on UX or liquidity.
In their words
The hardest problem in prediction markets isn't pricing the future. It's deciding what actually happened.· Andy Hall, *What to Do When Prediction Markets Fail*
The model doesn't have to be perfect; it has to be predictable.· Andy Hall, on LLM-as-judge
Polymarket has descended into sheer arbitrariness. Words are redefined at will, detached from any recognized meaning, and facts are simply ignored.· Polymarket user, quoted in *Semantics for $10 Million*
We've seen that UMA is not a decentralized truth machine, it's just a bunch of people who vote in their own best interests.· Lou Kerner, *My Polymarket Conspiracy Theory*
In our markets, you are absolutely entitled to trade on MNPI that you rightfully own.· David Miller, CFTC Enforcement Director, March 31 2026 (quoted in michaellwy CFTC comment)
Truth machines go to war.· Matt Levine framing for Bloomberg Opinion, on Kalshi mentions markets
Where it matters
Oracle design is the load-bearing primitive: it determines whether a prediction market is a price-discovery institution or a venue for extracting value through ambiguity exploitation. Every other technical problem (liquidity, capital efficiency, UX) is downstream · a market with broken resolution attracts toxic flow and bleeds informed traders. The frontier is moving from optimistic-oracle-with-token-voting toward layered architectures (LLM judges for cheap markets, decentralized adjudication for high-stakes, parametric on-chain feeds where possible) and toward defining-the-event-precisely-enough-that-oracle-choice-doesn't-matter.
Connections
- dispute resolution · what happens when the oracle's answer is contested; UMA's DVM voting and similar mechanisms.
- resolution criteria · the contract wording itself; "definition problem" is one tier above oracle choice.
- UMA protocol · the dominant optimistic-oracle implementation behind Polymarket; the most-stressed instance of oracle design.
- corruption value multiple · the quantitative diagnostic: does the value-at-stake exceed the cost of corrupting the oracle?
- self-resolving markets · designs that try to avoid oracles entirely.
- information aggregation · the Hayekian function oracles are supposed to preserve.
- incentive compatibility · whether the oracle's voters/judges are paid for honesty.
Platforms linked to this concept
- XO Market · addresses · XO Oracle is positioned as a UMA replacement
- Kalshi · affected-by · Cited as facing/exposed to Oracle Design
- Augur · implements · Augur's REP token is its own oracle design
- Polymarket · implements · Polymarket uses UMA as its oracle
- Trepa · implements · Mentioned in Oracle Design content as an implementing platform
Related concepts
- Dispute Resolution
- Resolution Criteria
- UMA Protocol
- Corruption Value Multiple (CVM)
- Self-Resolving Markets
- Information Aggregation
- Incentive Compatibility
Sources
- Resolving Prediction Markets: An AI-Driven Layered Approach · XO Labs · May 15 2026 · (x.com)
- Orthogonal Precision in Trepa: A Tunable Second-Order Oracle for High-Frequency Forecasting · Blanco, Chung, Meka · May 13 2026 · (github PDF viewer)
- Outcome Markets as a Cover Venue: HIP-4 and Its Traditional Comparables · Dean Eigenmann · May 8 2026 · (x.com)
- The Hidden Risk Of Prediction Markets: 14 Resolution Failures That Cost $500M · XO Market · Apr 22 2026 · (x.com)
- Every Opinion Deserves A Market · Turf · Apr 22 2026 · (x.com)
- Truth Machines Go to War · Matt Levine · Apr 6 2026 · (Bloomberg bot wall)
- The Economy of Truth: Inside the Decentralized Courtroom of Polymarket & UMA · Smaliy · Mar 16 2026 · (x.com)
- Polymarket Is a Hidden MEV Playground. Most Traders Have No Idea. · st1ne · Mar 13 2026 · (x.com)
- On War Markets · aaron · Feb 28 2026 · (x.com)
- 23 Reasons Prediction Markets Are Broken Today · Alexander Lin · Feb 26 2026 · (x.com)
- Prediction Markets are the Agentic Bazaar · Ben Fielding · Feb 16 2026 · (x.com)
- Prediction Markets: The Path to $1 Trillion and What Needs to Happen Next · Sakshi Mishra · Feb 14 2026 · (x.com)
- The 7 Axes of Prediction Markets · Jake Nyquist · Feb 10 2026 · (x.com)
- What to Do When Prediction Markets Fail · Andy Hall · Jan 24 2026 ·
- Semantics for $10 Million: When Is an Invasion an Invasion? · OddChain · Jan 23 2026 ·
- Assassination Semantics: Why Every Market Carries the Risk of Violence · Guillory & Zimmermann · Jan 7 2026 · (x.com)
- Why Prediction Markets Are Where They Are · Tim.0x · Jan 5 2026 ·
- Prediction Markets · Everything You Need to Know · Chokshi, Tabarrok, Kominers · Sep 25 2025 ·
- A Technical Typology of Prediction Markets: Mechanics and Tradeoffs · Baheet · Sep 14 2025 · (x.com)
- Meta Pool: A Unified Infrastructure for Liquidity and Resolution Trust in Prediction Markets · Jeff Opinion · Aug 19 2025 · (x.com)
- Next Gen Prediction Markets · Social Graph Ventures · Aug 6 2025 · (x.com)
- Prediction Markets: The 'Oracle Problem' Is Not the Problem · Aradtski · Jul 29 2025 · (x.com)
- The Risk Behind Arbitrage in Prediction Markets · michaellwy · May 23 2025 · (x.com)
- 10 Predictions About Prediction Markets · michaellwy · Jan 29 2025 · (x.com)
- The Definitive Guide to Prediction Markets · Four Pillars · Jan 15 2025 · (DocSend gated; 57-page deck, currently returns "content is no longer available" when re-fetched)
- Why Prediction Markets Are Broken (And How to Fix Them) · michaellwy · Jan 9 2025 · (x.com)
- Prediction Markets (II): Spoiling the Election Love Story · Luca Prosperi · Nov 2 2024 ·
- My Polymarket Conspiracy Theory · Lou Kerner · Oct 20 2024 · (via Wayback; live Medium URL now 404)
- Crypto Prediction Markets · Luca Prosperi · Oct 11 2024 ·
- Polymarket and the Proliferation of Prediction Markets · Alex Nardi · Oct 8 2024 ·
- Mechanisms for Prediction Markets · Hadi, Cossar, Shimony · Aug 22 2024 ·
- Polymarket Settles Bet Against Its Own Rules · Frank Muci · Aug 8 2024 ·