I built a site to compare after-fee prices on equivalent events across different prediction markets
Polymarket, Kalshi and other prediction markets often let you trade the same real-world events, but with different odds and slightly different payoff rules. That means users either have to check multiple sites manually or risk overpaying for the same thing.
Because these sites often have hidden fees which vary by market and platform, users also have to manually calculate fees to make a true comparison.
I built LocksBet to aggregate prediction market events in one place and show when the same event is trading across multiple platforms, and how the prices vary after fees.
The hardest technical problem was matching markets reliably. Right now the system uses embeddings to find likely cross-platform matches (which was already a lot more accurate than I would have thought). It then uses language models to compare the actual market rules and determine whether they’re truly equivalent. I still manually verify matches both as a safety layer and to generate better training/evaluation data for the matching pipeline.
The site also continuously scans for new markets, so the matching process has to work on an ongoing stream of differently worded event titles and rule descriptions. It also uses pricing, date, and other data to more intelligently find likely matches.
I originally built this for finding arbitrage opportunities where the same event is mispriced across platforms (and the site still flags arbitrage opportunities), but it’s also useful to just find the best price before placing a trade.
I'd love any feedback on the site, the matching approach, or anything else.