Resolve a prediction market from its rules and the evidence. Jev judges whether each document is about the market's event and whether the event occurred as the rules define it; rules check the deadline and the sources, weight the evidence and apply the two thresholds. Anything between 0.2 and 0.8, or contradicting evidence, goes to dispute review.
Yes/No markets whose end time is near or has just passed and that are not yet resolved, read live from Polymarket's public Gamma API; set a volume limit to list only small markets. Bet sizes come from the CLOB order books. Markets resolved from price or weather data can be excluded: their outcome comes from a data feed, not from reporting. The number to show counts items: all markets of one game (result, halftime, score, player stats and so on) are one item, a merged two- or three-option event is one item, and every other market is one item; bet size and holders are still checked market by market.
Single bet size is the value of the asks priced within 3 cents of the best ask: how much one buy order can fill without the price moving more than 3 cents. A market is listed if either Yes or No reaches the chosen amount.
Holders counts the distinct wallets holding Yes or No; a market is listed only if it has more than the chosen number (exactly that number is not enough). Holdings are read from the Polymarket Data API one market at a time; a market whose holdings cannot be read is not listed. A higher minimum means more markets are read and the list loads more slowly.
The number of documents per market is a target: when AI web search finds fewer relevant pages, it searches once more (at most 5 searches of its own per market), and markets with little coverage can still fall short. More documents mean more Jev requests when forecasting or resolving.
Collecting searches recent news on the backend, based on each market's question and rules, without calling Jev (with AI web search, up to 5 searches per market). Markets that have not ended are forecast from the evidence and ended ones are resolved; either way each document goes to Jev as its own request. Collect all only covers markets without evidence, and AI prepares the search plan without a manual check; Forecast all and Resolve all only cover markets with evidence that are not done yet.
Choose Collect evidence on a market, check what was found, then choose Forecast: a market that has not ended is forecast from the evidence, and one that has ended is resolved under its rules (the button reads Resolve). You can also collect and forecast them all. Evidence comes only from search results; nothing is entered by hand.