knaves.ai
A Benter-style race-intelligence engine — win probabilities, market blending, and Kelly stakes.
prototype · not financial advice · for testing what we can build

The race card

HorseSpeedWin%Days off Wt carriedDecl. wtDrawOdds (dec.)
Blend weights from the trained model — market · our model (the crowd supplies most of the signal; the model is the thin edge on top).

Prediction & suggested stakes

HorseOur modelMarketBlend Edge¼-Kelly stakeCall
Hit “Compute prediction”.
Edge = blend probability × decimal odds − 1. We only flag a bet when edge ≥ 5%. Stake is a fraction of bankroll, quarter-Kelly, capped at 5%.

Exotics (Monte Carlo, 20k sims)

Top exactas (1st–2nd)

Top trifectas (1st–2nd–3rd)

Win probabilities are turned into finishing-order probabilities by simulating the race thousands of times (Plackett-Luce / Monte Carlo) — the same idea behind most prediction engines.