The model is well-calibrated only between 10–20% predicted (gap ≤ ±1.5pp). Above 20% it is systematically over-confident — it says 35%, the team wins 25%. Below 10% it is mildly under-confident, but that's the longshot zone where market favourite–longshot bias dominates and any edge evaporates in variance.
The disciplined edge is on favourites and short second-favourites in the $2–3 zone — the only band where flat ROI approaches breakeven (−1.2%). Size with fractional Kelly and judge on closing-line value, not weekly results.
| Price band | Runners | Winners | Win % | Implied % | Flat ROI |
|---|---|---|---|---|---|
| <$1.50 | 27 | 18 | 66.7% | 80.9% | −19.2% |
| $1.50–1.99 | 40 | 20 | 50.0% | 58.0% | −12.6% |
| $2.00–2.99 ★ | 76 | 31 | 40.8% | 40.9% | −1.2% |
| $3.00–4.49 | 138 | 29 | 21.0% | 28.1% | −25.3% |
| $4.50–6.99 | 150 | 22 | 14.7% | 18.7% | −22.4% |
| $7.00–9.99 | 132 | 16 | 12.1% | 12.5% | −4.9% |
| $10–14 | 147 | 9 | 6.1% | 8.6% | −28.6% |
| $15–24 | 132 | 8 | 6.1% | 5.4% | +15.2% |
| $25–49 | 105 | 0 | 0.0% | 3.2% | −100% |
| $50+ | 75 | 1 | 1.3% | 1.5% | +8.0% |