I looked at the oracle mechanism of Polymarket/kalshi/opinion. The three are really not exactly the same. Firstly, what is the oracle mechanism? For newcomers, the oracle mechanism roughly means predicting how the market platform will determine and verify the outcome of events to settle fairly and accurately. Therefore, for predicting the market, a oracle almost represents trust, and a good oracle mechanism can make both parties in the transaction agree. That is to say, the oracle mechanism is one of the lifelines for predicting the market. Let's first take a look at Polymarket's oracle mechanism. Polymarket uses UMA Optimal Oracle, a decentralized smart contract system for decision-making. This is an optimistic model (reminiscent of the mechanism of L2 OP): anyone can use off chain data to present results, and if no one challenges within the disputed window period (such as 2-48 hours), the results are accepted. If there is a dispute over the result, a vote is required to make a decision (UMA token holders vote). Generally, UMA token holders will pledge their tokens on the correct result, as incorrect voters will lose the pledge and cause losses. When there is no controversy, this mechanism is relatively efficient, but it has also been subjected to governance attacks. For example, in 2025, the outcome of an event was related to a value of $7 million, in which a major player manipulated the vote. In this type of staker voting, if not all voters carefully review it, there may be a herd effect (being biased) in the voting. Kalshi is a centralized prediction market regulated by the CFTC, which adopts a centralized decision-making process and relies on pre-defined verifiable data sources (trusted third parties). For example, the economic indicator market pulls data from official government APIs (such as unemployment rate data from the Bureau of Labor Statistics), and weather data comes from NOAA, among others. The resolution is manually processed by the Kalshi team based on clear rules in each contract, without relying on community voting or AI. The advantage of this mechanism is that it is more compliant and reduces manipulation risks, but the efficiency of some subjective events in producing results is relatively low, and there are also centralization issues (such as single point failures such as data source input errors). That is to say, users must believe in the fairness and correctness of Kalshi itself as a center, and also accept its relatively low efficiency in settlement speed. The oracle mechanism of Opinion is quite unique, as it adopts Opinion AI (decentralized multi-agent AI oracle), which utilizes AI agents to analyze data, execute market rules, and autonomously settle results. To ensure accuracy, it will cross validate sources such as news APIs, social media, or on chain data. In terms of settlement processing, it will be faster than the mechanism of traditional oracle machines. Of course, multi-agent AI oracle machines are not entirely accurate. In multi-agent systems, errors made by one agent (such as data interpretation bias) may affect other agents, leading to the failure of overall decision-making; There are also external attack risks, such as network attacks or data tampering. Therefore, perhaps with the advancement of practice, the Opinion platform may also learn from the "optimistic" mechanism: AI first proposes settlement results, which are reviewed by multiple models (Gemini, OpenAI, Claude, etc.) and verified by human reviewers. There is a dispute window where users can challenge AI resolutions, and those in dispute can participate through community voting or backup verification (human participation review). If no one challenges during the dispute period, settlement is completed. The specific mechanism evolution is subject to official confirmation, and some of this is only speculation on evolution. At present, it seems that each family's oracle needs to be tested in practice and continuously improved, as this is one of the most important parts in competition.