How market & outcome matching works
Matching is how predict.pm decides that two differently-worded markets are the same real-world question. A large language model reads each platform's titles, groups the same event together, and then aligns each market's outcomes so 'Yes' on one platform maps to 'Yes' on another.
- LLM-powered
- Event + outcome level
- Improving in beta
“Will the Yankees win the 2025 World Series?”
“MLB Champion 2025: NYY”
Yankees to win the World Series
Event matching
First, an LLM groups markets from Polymarket, Kalshi and Futuur that describe the same real-world event into a single cross-platform event — even when the titles are worded completely differently.
Market & outcome matching
Within each event, the model aligns every market's outcomes across platforms so equivalent results share a row. This is what makes the probabilities directly comparable rather than just listed next to each other.
Why an LLM?
Platforms phrase the same question very differently, and simple text rules miss those matches. A language model understands meaning, so it can connect titles and outcomes that keyword matching would not.
Accuracy & limitations
predict.pm only compares groups it is confident represent the same event, and filters obvious mismatches. As the product is in beta, matching continues to improve over time.
Compare the odds yourself
See the same event priced across Polymarket, Kalshi and Futuur.
Learn more
Frequently asked questions
How does matching work?
Matching is the process of using a large language model (LLM) to group the same event across platforms and align each market's outcomes so they are directly comparable.
What is the difference between event matching and outcome matching?
Event matching groups the same real-world question across platforms into one event. Outcome matching then aligns each market's outcomes so 'Yes' on one platform maps to 'Yes' on another and the prices are directly comparable.
Why use an LLM for matching?
Platforms word the same question very differently. A language model understands meaning, so it can match titles and outcomes that simple text rules would miss.
What happens if a match is wrong?
predict.pm only compares groups it is confident are the same event. Because it is in beta, matching keeps improving, and obvious mismatches are filtered out.