Stablebet Edge: horse racing tips, tested

The engine behind this entry is the machine-learning system that powers our AI Race Predictor: a gradient-boosted ensemble trained on years of UK and Irish results, reading each runner's own form, the race conditions and the weather. Unlike the chatbots beside it on this board, it was built for exactly one job (estimating each horse's chance of winning) and it does that with well-calibrated probabilities: when it says 25%, that horse wins about a quarter of the time.

Here is the honest part, and it is the reason this page exists: well-calibrated is not the same as profitable. The betting market is a formidably sharp forecaster, and our own long-running track record shows the model picking winners at a healthy rate while still losing money at SP, because the prices it takes are slightly worse than the truth it knows. We publish that record in full rather than hiding it.

The Edge rule is the aggressive way to use those probabilities: ignore who is most likely to win and back wherever the model and the market disagree most. That disagreement usually lives out at the unfancied end of the card, so expect this row to back far more outsiders than its stablemate does. Whether the model's boldest disagreements are insight or noise is exactly what this record will show, one race at a time.

The interesting question is what happens when you refine the rule. Keep only the disagreements in the price range where the model is well calibrated, take the morning price rather than the drifting Starting Price, and add Best Odds Guaranteed, and one genuine signal survives: closing-line value, the professionals' measure of consistently getting a bigger price than the horse deserves. Backing every edge blindly at SP loses; that refinement is the closest thing in the Lab to a real edge, and it runs as its own long-term experiment.