> ## Documentation Index > Fetch the complete documentation index at: https://blog.opinion.trade/llms.txt > Use this file to discover all available pages before exploring further. # How Opinion AI Resolves Markets > Learn how Opinion AI helps create resolvable market rules, evaluates evidence and resolves most Opinion.trade markets—and what traders must verify. ## Quick Answer **Opinion AI is the primary oracle for most markets on Opinion.trade.** It is described as a decentralized multi-agent AI system that works at both ends of a market's lifecycle: helping create objective, resolvable rules before trading begins and evaluating relevant evidence when the outcome must be settled. The exact resolution method is not assumed from the platform name; it is specified on each market page, and those market-specific rules should be the trader's source of truth. ## Key Takeaways * An oracle connects an on-chain prediction market to evidence about the outside world. * Opinion AI is designed for complex or unstructured evidence, not only simple data feeds. * Resolution quality begins with precise question wording, dates, sources and edge-case rules. * Most Opinion markets currently use Opinion AI, according to the official Docs. * Opinion leadership has publicly described a hybrid direction combining AI, multiple agents, trusted execution environments and human review. * Public descriptions of planned architecture should be distinguished from mechanics documented as live today. * Traders should always inspect the exact resolution method before buying an outcome. ## What Is a Prediction Market Oracle? Smart contracts can record trades, balances and settlement instructions, but they cannot independently know who won an election, what a government agency reported or whether a match ended under a particular rule. An **oracle** supplies the external determination needed to resolve the market. In a simple case, resolution may rely on a single structured value from an official source. In a harder case, evidence can be spread across documents, news releases or event rules. The oracle must apply the market's written definition to that evidence. The oracle is therefore not a forecasting model. Traders forecast **before** the event; the oracle determines the outcome **after** the resolution conditions are met. ## Opinion AI Has Two Jobs The current [OPINION Docs](https://docs.opinion.trade/) describe Opinion AI as a decentralized multi-agent AI oracle. Its role starts before a market goes live. ### Job 1: make the market resolvable A strong prediction-market question needs more than a catchy title. It should define: * the outcome being measured; * the relevant date and time zone; * the primary source or source hierarchy; * what counts as Yes, No, cancellation or invalidation; * how delays, corrections and edge cases are handled. Opinion AI is designed to help generate rigorous rules and check whether a proposed topic meets resolvability standards. This reduces the chance that traders are buying different interpretations of the same sentence. ## What Opinion Learned From Permissionless Testnet Markets In a public product explanation, Opinion co-founder Jeff described a testnet in which whitelisted users created hundreds of permissionless markets and the community handled rules and resolution. Two recurring problems emerged: 1. **Inconsistent market quality.** Vague wording created confusion and raised dispute risk. 2. **Unreliable resolvability.** Some questions could not be settled accurately from the stated evidence. The lesson is that decentralizing market creation is not enough. A scalable system also needs a quality gate before trading begins: the question, deadline, source hierarchy and edge cases must form a contract that can actually be executed. Read [What Makes a Prediction Market Resolvable?](/learn/what-makes-a-prediction-market-resolvable) for the full checklist. ### Job 2: resolve the market At the resolution stage, the system evaluates the market definition against the available evidence. The [official resolution page](https://docs.opinion.trade/trade-on-opinion.trade/resolution) says that most markets are currently resolved through Opinion AI, which serves as the primary oracle mechanism. That statement is intentionally narrower than “AI resolves everything.” A market may specify a different method, named source or special condition. The rules attached to that market control the outcome. ## The Resolution Flow The practical flow can be understood in six steps: 1. **Question definition** — the market identifies the event, outcomes and deadline. 2. **Resolvability check** — rules and sources are tested for ambiguity or missing edge cases. 3. **Trading period** — participants trade based on their own forecasts; the oracle does not set the market price. 4. **Evidence collection** — after the stated condition occurs, relevant sources are examined. 5. **Rule application** — the evidence is evaluated under the pre-written resolution criteria. 6. **Settlement** — the resolved outcome determines which conditional tokens receive the market payout. The whitepaper calls Opinion AI “multi-agent.” Public materials do not require a trader to trust a single unnamed webpage or one individual's judgment. This article does not claim a permanent list of underlying AI models because the current official Docs do not publish one. Resolution also has a separate challenge layer. Under the current process, a user can file one of two documented dispute types during the market's dispute window by staking OPN. Read [Opinion Market Resolution and Disputes](/learn/opinion-market-resolution-and-disputes) for the maintained timing, stake formula, review process and exceptions. ## The Hybrid Architecture Opinion Has Publicly Described Opinion leadership has described the intended mainnet direction as a hybrid rather than an AI-only judge: | Layer | Publicly described role | Why it matters | | ------------------------------ | -------------------------------------------------------------------- | ------------------------------------------------------------- | | AI-assisted creation | Draft objective rules, define conditions and identify usable sources | Prevent ambiguity before funds enter a market | | Evidence processing | Ingest and summarize relevant sources after the event | Handle unstructured evidence at greater scale | | Multi-agent review | Use multiple models or agents rather than one output | Reduce dependence on a single model judgment | | Trusted Execution Environments | Isolate model execution and protect process integrity | Make tampering with the computation harder | | Human oversight | Review factual edge cases and potential bad-actor behavior | Provide judgment where evidence or language remains ambiguous | | On-chain dispute | Give users a time-limited challenge path backed by an OPN stake | Keep a proposed result from becoming final immediately | This table summarizes a public architecture statement, not a permanent technical specification. The current Docs establish that most markets use Opinion AI and document the live dispute process, but they do not yet publish a complete model roster, consensus threshold, TEE attestation format or human-review operating policy. That distinction matters for trust: a responsible explanation should say what is documented as live, what has been publicly described as the direction, and what remains to be specified. ## Why Market Rules Matter More Than the Headline Consider the headline “Will Country A enter a recession in 2027?” It sounds clear until traders ask: * Is recession defined as two consecutive quarters of negative real GDP? * Which agency's first estimate counts? * Do later revisions change the result? * What is the final observation date? Two reasonable people can answer the headline differently. A resolvable market removes that freedom by specifying the metric and evidence in advance. Before trading, read the market's rules as if they were a contract. If the headline and the detailed criteria feel different, base the decision on the detailed criteria—or do not trade the market. ## Structured Data vs Unstructured Evidence | Evidence type | Example | Main resolution challenge | | ------------- | --------------------------- | -------------------------------------------------- | | Structured | An official numeric release | Correct field, timestamp and revision policy | | Event result | A completed match | Cancellations, overtime and governing rules | | Unstructured | A policy announcement | Interpreting language against a precise definition | | Multi-source | Conflicting public reports | Following the stated source hierarchy | Opinion AI's stated advantage is its ability to handle complex and unstructured data while also helping define the rules that make that evidence usable. ## How Oracle Models Differ Messari's Polymarket analysis is useful because it explains resolution as a complete system rather than a one-line feature. It describes Polymarket's UMA flow as a proposal, challenge period and possible escalation to token-holder voting. Kalshi's current Help Center says market outcomes are finalized by its markets team under market-specific rules and official sources. Opinion is pursuing a different combination. Its public materials describe AI-assisted rule creation and multi-agent evidence handling, while its current Docs add a staked dispute window. These are different trust models, not a simple fastest-versus-safest ranking. | Model | Primary strength | Main question users should ask | | -------------------------------------- | ---------------------------------------------- | ------------------------------------------------------------------------ | | Exchange or markets-team determination | Clear operational responsibility | How transparent are rules, sources and appeals? | | Optimistic oracle | Permissionless proposal and economic challenge | Who can escalate, how long does it take and who votes? | | Hybrid AI and decentralized process | Scalable handling of unstructured evidence | How are models, execution integrity, human review and disputes verified? | For a category-level explanation, read [What Is a Prediction Market Oracle?](/learn/what-is-a-prediction-market-oracle). ## What the Oracle Does Not Do The Opinion AI oracle does not: * guarantee that a traded probability is accurate; * prevent users from losing money on a position; * eliminate all ambiguity, data errors or disputes; * replace the market-specific resolution rules; * make every market equally liquid or safe to trade. An oracle addresses one part of prediction-market risk: how the final outcome is determined. Trading and liquidity risks remain separate. ## How Traders Should Evaluate an Opinion Market Before placing an order, verify five things: 1. **Resolution source:** Is a primary source named? 2. **Deadline:** Is the time zone and cutoff unambiguous? 3. **Outcome definition:** What exactly makes the market resolve Yes or No? 4. **Exceptional cases:** What happens after a cancellation, delay, tie or data correction? 5. **Resolution method:** Does the market explicitly state how Opinion AI or another mechanism will be used? Also examine liquidity and the spread. A market can have excellent rules but still be expensive to enter or exit when the order book is thin. ## Where Opinion AI Fits in the Opinion Stack Opinion AI is one of four layers in the architecture: * **Opinion.trade** provides the trading interface and exchange. * **Opinion AI** supports market creation and resolution. * **Opinion Metapool** is the unified-liquidity layer described by the ecosystem. * **Opinion Protocol** is the interoperability layer for prediction venues. Read [What Is OPINION?](/learn/what-is-opinion-prediction-market) for the ecosystem overview. ## What to Watch * Publication of more detailed agent, evidence and consensus methodology. * Publication of TEE attestation and human-review specifications. * How corrections or disputed source data are handled in live markets. * Whether market pages expose clearer source hierarchies and edge-case rules. * Changes to the live dispute window, stake formula or review process. * Resolution time and consistency across different market categories. ## Where Opinion Fits Opinion combines an order-book prediction exchange with an AI-based creation and resolution layer. Opinion AI is the part of that stack that turns external evidence into a market outcome; it complements, rather than replaces, the exchange, liquidity and protocol layers. ## Sources Used 1. [OPINION Docs: Welcome to OPINION](https://docs.opinion.trade/) — current Opinion AI description and four-layer architecture. 2. [OPINION Docs: Resolution](https://docs.opinion.trade/trade-on-opinion.trade/resolution) — current statement that most markets use Opinion AI. 3. [OPINION Whitepaper v1.0, February 2026](https://static.opinion.trade/OPINION_Whitepaper_EN.pdf) — architecture, market creation and oracle role. 4. [OPINION Docs: Dispute](https://docs.opinion.trade/trade-on-opinion.trade/dispute) — current dispute window, stake and review process. 5. [Jeff / Opinion: Resolution — Opinion AI](https://x.com/jeff__opinion/status/1957611486991487486) — public testnet lessons and hybrid-architecture direction. 6. [Messari: A Valuation of Polymarket](https://messari.io/report/a-valuation-of-polymarket-poly) — independent explanation of the UMA-based Polymarket resolution flow. ## FAQ No. Traders express predictions through market prices. Opinion AI is the oracle used to help determine the outcome after the market's stated resolution conditions are met. The official Docs state that most markets currently use Opinion AI. Each market specifies its own resolution method, so users should read the individual rules. Current public Docs describe a decentralized multi-agent system but do not publish a permanent model list. Model or provider claims from older third-party reports should not be assumed to remain current. Opinion leadership has publicly described TEE execution as part of the hybrid architecture direction. The current public Docs do not yet provide the detailed deployment and attestation specification needed to independently verify every implementation claim, so this page does not present TEE coverage as universal or permanent. No. UMA uses an optimistic proposal-and-challenge system with possible token-holder escalation. Opinion's materials describe AI-assisted market creation, multi-agent evidence processing and a separate OPN-staked dispute path. Users should compare the complete rules and failure modes, not only the label attached to the oracle. Any resolution process can face ambiguous rules, incomplete evidence, source corrections or software errors. Precise market definitions, transparent sources and a clear dispute path help reduce those risks but cannot eliminate them. The resolved outcome determines settlement under the market's rules. Review the live market and current Docs for the exact settlement process. If you cannot explain exactly what evidence will produce each outcome, the prudent choice is to wait for clarification or choose a better-defined market. Educational information only. Prediction-market trading involves risk, and access or eligibility varies by jurisdiction.