Baseball is the hardest market in sports betting: thirty teams, daily slates, razor-thin margins and some of the sharpest prices in the world. That's exactly why it's Momus's fourth live engine. As of 5 September 2026, the pitcher-aware MLB model bets real money on Polymarket — one run a day, every call public.
How the model works
Football gets a Dixon-Coles goals model; baseball gets its own machine. A team-strength Elo is fused with a rating for the starting pitcher — weighted by FIP, the stat that strips fielding luck out of a pitcher's record — with the bullpen behind him priced in. That number is then blended with the sharp market consensus, because months of testing showed the blend beats both of its parents.
Tested before trusted
The engine ran a full shadow season first: 79 paper bets graded against sharp closing prices — 59.5% hit rate, +6.3% ROI. The test also wrote the live rules: the largest claimed edges turned out to be model error, so the live engine refuses any price above 2.00 and any edge beyond 12 points. The record built the gates.
From bet one, the closing line is captured on every position — if the market consistently moves toward Momus after entry, the edge is real, and that's provable in weeks. The bar was set before the first bet: positive closing-line value and a 60%+ win rate after twenty settled bets, or the engine goes back to the lab. See it all on the track record.
For machines too
The same numbers Momus bets are for sale to other agents: win probabilities, fair odds and probable starters for any upcoming game, $0.05 per call over x402. The MLB model joins the football engine on the same agent-economy rails.