Coral Stewards' Cup (Heritage Handicap)

The near-symmetry between Far Above Dream and Sondad is the dominant analytical feature of this paper. Two horses, separated by five simulated wins across a thousand trials, with mean finishing positions of 1.71 and 1.72 respectively — the model essentially cannot split them. That is itself meaningful information. In a twenty-eight runner field, the simulator is collapsing the complexity to a binary proposition, and the margin between the two principals is so fine that any attempt to declare one categorically superior to the other on model evidence alone would be intellectually dishonest. What the model does tell us is that both horses carry structural advantages — whether in rating trajectory, course suitability, or going preference — that no other runner in the field can replicate to a comparable degree. The market disagrees with this framing in an instructive way. Sondad is priced at 6/1 — a market-implied win probability of roughly 14% in a competitive field — while Far Above Dream is not among the named market prices provided, suggesting it may be a longer price or a late entry still finding its level in the book. The chamber engine, by contrast, assigns both horses a win probability north of 46%. That is a profound divergence, and it is the central discussion point of this paper. Either the model is over-fitted to the metrics that favour these two horses and is systematically underweighting the chaos of a twenty-eight runner sprint, or the market is genuinely underestimating the structural dominance of the leading pair. Two Tribes, priced at 17/2 and recording 6.9% of simulated wins, sits in a more credible alignment between model and market — though its top-3 rate of 74.0% suggests a horse that places reliably without often winning outright. Where could the model be wrong? Sprint handicaps at Goodwood are among the most draw-sensitive races in the British calendar, and the chamber engine's chaos factor, while randomised per iteration, cannot fully replicate the mechanical advantage of a favourable stall in an extreme going condition. If the draw falls against one or both of the leading pair, the model's structural signal becomes partially redundant. Soldier's Tree, priced at 11/1 with a OR of 108 — the highest rating in the named market — managed only 0.2% of simulated wins and a top-3 rate of 17.4%, which suggests the model finds the weight it carries prohibitive despite the market's relative faith. That gap between market confidence and model output is the one place where experienced punters might reasonably dissent from the simulation's conclusions.