Seventeen Years at the Rail

Where the model comes from

Long before there was a model, there was a spreadsheet. And before the spreadsheet, a father reading race results out loud at the kitchen table. That's the honest start of this company: a love of the sport that got handed down, and eventually handed over to the numbers.

We didn't come to racing from the betting window. We came to it from the analytics side of the desk. Our founding team spent years in quantitative finance, one of us running the numbers and the other building the market for them, and we looked at racing the same way we looked at everything else: as a pricing problem. A stock has a fair value the moment before the market opens. So does a horse the moment before the gates. The whole game is finding the ones the crowd has priced wrong.

Seventeen years of hand-built data

The edge didn't show up overnight. We built it one race at a time. For the better part of two decades we compiled and analyzed racing data at the tracks that matter most: Keeneland, Saratoga, Churchill Downs, Del Mar, Gulfstream. Seventeen years of past performances, results, and conditions in all. We learned early the lesson that still shapes the model today: you can't pool it all together and expect an edge. Trainers and jockeys who dominate at one track can vanish at the next. The crowd bets each meet differently. The bias in the dirt and the bias in the turf are two separate animals. So we built the system to respect the track instead of averaging it away.

Everything that moves a race feeds the model: speed, jockey and trainer form, track bias, layoffs, class, surface, and the day-of stuff most tip sheets ignore, like scratches, jockey changes, and how the weather is actually playing that afternoon. What started as a few algorithms bolted onto a racing form is now a machine-learning model that scores every horse and prices every race in real time. The philosophy never changed. The computing power did.

We test the wisdom before we trust it

Racing is full of old sayings. “Never bet a horse that just broke its maiden.” “Lone early speed kills.” “Gray horse on a gray day.” We grew up on them too. Then we started measuring them. Take the gray-horse-on-a-gray-day superstition. We pulled eight years of races at a top track and let the data settle it. On win rate, the grays showed no real edge at all, but they posted the worst return of any coat color. Busted. It's a small example of a bigger habit: we don't take an assumption on faith just because it sounds right at the rail. We check it, and we live with the answer whether it flatters us or not. That's the same instinct that built the model, and it's the one running underneath every number we publish.

Discipline over hunches

We think luck is mostly the residue of design: disciplined, data-driven handicapping beats hunches and hype over the long run. That belief is why we started, and it's why we're still at it seventeen years later.

It's also why we show our work. Anyone can talk up the winners they cashed. We publish real return-on-investment tracking, week after week and month after month, tough cards included. Not every day is a green day, and pretending otherwise would miss the whole point. We put out model output as information, never as a promise, and we believe in playing responsibly. What we can stand behind is the method: an unbiased read on every race, the value plays flagged, and the discipline to bet them on their price instead of on hope.