An AI trading system made a $200 live trade on the prediction market platform Kalshi for 26 days, ending at $360.67—a return of 80.33%.

Its reported win rate was only 50.9%, almost a half-and-half split between correct and incorrect calls.
First, it estimates how much a contract is worth, then checks how much the market is selling it for. It only places orders when the gap between the two sides is large enough; the bigger the gap, the more it buys.
Once, the market was betting on whether the U.S. president would issue an executive order within one week. The YES sell price was 66 cents, but the AI estimated the real probability at only 31%, so the system bought 35 NOs at 37 cents.
After the Kalshi contract settles in the NO direction, each share pays out $1. If this NO bet is correct, each share earns 63 cents; if it’s wrong, it only loses 37 cents.
Later, the prices rose to 68 cents and 99 cents. The system sold in batches. The total amount put into the trade was at most $15.21, and it ultimately netted $19.23 profit.
Over 26 days, the system made 1,605 probability forecasts, sent 396 trading signals, executed 236 trades, and paid $26.31 in fees.
Even though it was right only about half the time, it earned more when it was correct and lost less when it was wrong. In the end, the account still gained $160.67.