On the Players’ Podcast this week, PTF and Betting on Horse$ author Chris Larmey talked about interesting ways horseplayers can use AI. A day after that episode came out, a player on X came up with a great one.
BitPlayer (@ProjectedOdds) had an AI model go through his 2026 Belmont and Saratoga multi-race data, looking for what makes a Pick 3, Pick 4 or Pick 5 pay more or less than you would expect from the win odds of the horses that won. He posted what it found in an 11-part thread on Nov. 18. A few highlights:
- Winning favorites helped. The more favorites that won, the more sequences tended to pay than expected, and it was the strongest effect he found.
- Longshots hurt. More winners at fourth choice or longer meant payoffs tended to come in lower than expected, especially when the longshot won the next-to-last leg.
- Maiden races and big fields helped. Sequences with more maiden races, and more legs with 10 or more horses, tended to pay more than expected.
He is careful about the limits, too: the samples are small, late scratches knock sequences out, and paying more than the win pools imply is not the same as beating the takeout. The whole thread is worth a read, and BitPlayer is worth a follow.
“I’m not saying this is going to change my approach to horizontals overnight, but it’s interesting food for thought, and you can bet I’ll be doing more analysis of my own,” PTF said.
More from ITM: Chris Larmey on finding the next edge, CAW guard rails and AI · Today’s Racing







