Horse Racing Model Analysis Rolling out-of-sample tests across seven racing populations Scope Manifest coverage 7 populations Flat 2yo, Flat 3yo, Flat 3yo+, Flat 4yo+, Hurdle, Chase and NH Flat 3 rolling test years 2023, 2024 and 2025 3 model seeds 7, 19 and 31 63 model reports Seven populations x three years x three seeds HORSE RACING MODEL ANALYSIS 2 Contents Section Contents 1 Headline results 2 Probability calibration 3 Population analysis 4 Auto-selected betting rules 5 Exacta and trifecta hit rates 6 Race-category and course appendix 7 Method and model configuration HORSE RACING MODEL ANALYSIS 3 1. Headline results The model top pick is compared directly with the BSP favourite. Model figures are means across the nine seed-year tests for each population. BSP favourite profitability is evaluated across the three distinct rolling years, rather than treating identical seed copies as separate market tests. HORSE RACING MODEL ANALYSIS 4 Population Model ROI BSP ROI Model strike BSP strike Model BSP BSP fav BSP Races Chase 6.9% -3.4% 27.1% 31.5% 5.33 3.50 4,462 Hurdle 4.7% 1.4% 32.0% 37.1% 5.02 3.23 8,268 Flat 2yo 7.2% -4.3% 35.1% 37.9% 4.15 2.90 3,842 Flat 3yo 6.4% 0.3% 28.7% 33.3% 4.98 3.37 3,311 Flat 3yo+ 3.5% -1.6% 27.0% 31.4% 5.18 3.62 9,191 HORSE RACING MODEL ANALYSIS 5 Population Model ROI BSP ROI Model strike BSP strike Model BSP BSP fav BSP Races Flat 4yo+ -0.9% 1.4% 24.1% 30.4% 5.58 3.74 6,254 NH Flat -5.3% 2.2% 31.5% 39.9% 4.64 2.87 1,169 Headline facts Flat 2yo has the highest model top-pick ROI at 7.2%. NH Flat is the weakest at -5.3%. The strike-rate charts show that the model generally selects longer-priced runners than the BSP favourite; the ROI chart shows whether that lower strike rate is compensated by price. HORSE RACING MODEL ANALYSIS 6 2. Probability calibration The x-axis is categorical 5% BSP-implied probability bands: 0-5, 5-10, 10-15 and so on. Results within each band are merged into one point, then a smooth curve is drawn through those points. The report shows only the model, the BSP market and observed outcomes. Chase Across the displayed bins, the runner-weighted absolute gap from observed outcomes is 2.6 pp for the model and 0.4 pp for the BSP market. This measures calibration by band, not betting profitability. HORSE RACING MODEL ANALYSIS 7 Hurdle Across the displayed bins, the runner-weighted absolute gap from observed outcomes is 2.3 pp for the model and 0.2 pp for the BSP market. This measures calibration by band, not betting profitability. HORSE RACING MODEL ANALYSIS 8 Flat 2yo Across the displayed bins, the runner-weighted absolute gap from observed outcomes is 2.2 pp for the model and 0.3 pp for the BSP market. This measures calibration by band, not betting profitability. HORSE RACING MODEL ANALYSIS 9 Flat 3yo Across the displayed bins, the runner-weighted absolute gap from observed outcomes is 2.2 pp for the model and 0.5 pp for the BSP market. This measures calibration by band, not betting profitability. HORSE RACING MODEL ANALYSIS 10 Flat 3yo+ Across the displayed bins, the runner-weighted absolute gap from observed outcomes is 2.2 pp for the model and 0.3 pp for the BSP market. This measures calibration by band, not betting profitability. HORSE RACING MODEL ANALYSIS 11 Flat 4yo+ Across the displayed bins, the runner-weighted absolute gap from observed outcomes is 2.5 pp for the model and 0.3 pp for the BSP market. This measures calibration by band, not betting profitability. HORSE RACING MODEL ANALYSIS 12 NH Flat Across the displayed bins, the runner-weighted absolute gap from observed outcomes is 2.6 pp for the model and 0.5 pp for the BSP market. This measures calibration by band, not betting profitability. HORSE RACING MODEL ANALYSIS 13 3. Population analysis Each population section presents the betting result first, then rolling-year stability, model quality, race-group and race-category results, course splits, and perceived value. Perceived value is deliberately placed after the race-group analysis. Chase The model top pick returned 6.9% at a 27.1% strike rate and average BSP 5.33. The BSP favourite returned -3.4% at a 31.5% strike rate and average BSP 3.50. The model was profitable in 3/3 rolling years; the BSP favourite was profitable in 0/3. The larger, more repeatable positive race categories are Chase Three Mile Other Open, Chase Three Mile Handicap, Chase Two Mile Handicap. The clearest larger-sample negative categories are Chase Extreme Handicap. Headline comparison Selection ROI Strike Avg BSP Profitable years Races Model top pick 6.9% 27.1% 5.33 3/3 4,462 BSP favourite -3.4% 31.5% 3.50 0/3 4,462 ROI by rolling year and seed Year Seed 7 Seed 19 Seed 31 3-seed mean BSP fav 2023 9.0% 7.6% 9.2% 8.6% -2.6% 2024 17.5% 8.6% 5.0% 10.4% -3.6% 2025 7.3% -1.4% -0.7% 1.7% -3.9% HORSE RACING MODEL ANALYSIS 14 Probability quality Metric Model BSP market Log loss 1.883 1.776 Brier score 0.817 0.787 Top-pick strike rate 27.1% 31.5% Top-3 strike rate 61.9% - ECE 0.005 0.005 HORSE RACING MODEL ANALYSIS 15 Chase race groups Regime Regime Races Model ROI BSP fav ROI Strike Avg BSP Other Open 201 54.0% -6.2% 35.0% 4.64 Maiden Novice 559 5.4% -4.2% 37.4% 4.12 Handicap 3,562 5.1% -3.3% 24.6% 5.66 Pattern Stakes 140 -7.8% 3.3% 38.8% 3.03 Distance Distance Races Model ROI BSP fav ROI Strike Avg BSP Three Mile 1,557 13.6% -4.9% 25.9% 5.70 Two Mile 926 5.0% -8.0% 28.4% 4.93 Middle 1,855 4.0% 0.8% 28.2% 5.11 Extreme 124 -17.9% -11.5% 15.6% 7.04 Surface Surface Races Model ROI BSP fav ROI Strike Avg BSP Turf 4,462 7.0% -3.4% 27.1% 5.34 Race class Race class Races Model ROI BSP fav ROI Strike Avg BSP C2 220 21.6% 11.2% 28.0% 5.60 C5 851 18.6% 4.4% 27.0% 5.33 C3 486 8.5% -8.4% 28.4% 5.00 Unclassified 1,756 7.5% -1.6% 27.0% 5.62 C4 733 -5.3% -14.9% 24.9% 5.06 C1 414 -6.7% -8.3% 30.0% 4.88 Sex restriction Sex restriction Races Model ROI BSP fav ROI Strike Avg BSP Open 4,462 7.0% -3.4% 27.1% 5.34 HORSE RACING MODEL ANALYSIS 16 Race-category ROI Stronger race categories Category Races Model ROI BSP fav ROI Profitable runs Chase Three Mile Other Open 114 83.5% -8.6% 9/9 Chase Three Mile Handicap 1,086 12.1% -8.4% 9/9 Chase Two Mile Handicap 583 5.5% -12.1% 6/9 Chase Middle Handicap 1,134 2.3% 0.4% 6/9 Weak race categories Category Races Model ROI BSP fav ROI Profitable runs Chase Extreme Handicap 110 -20.4% -13.9% 3/9 HORSE RACING MODEL ANALYSIS 17 Course split Perceived value Perceived value is model probability divided by BSP-implied probability, minus one. A 20% value reading means the model probability is 20% above the market-implied probability. These results cover BSP 3.0-33.0 and include all runners inside each value band. The strongest cumulative threshold with at least 500 seed-run bets is 50%+ perceived value: 10.6% ROI across 19,723 bets. This is a descriptive result from the same test matrix and should not be treated as a separately untouched strategy test. Course names are mapped from the manifest identifiers. The chart displays the strongest and weakest larger samples; full course tables are in the appendix. HORSE RACING MODEL ANALYSIS 18 Minimum value Seed-run bets ROI Strike 0%+ 45,888 6.8% 10.3% 5%+ 42,354 6.8% 10.1% 10%+ 38,991 6.7% 9.8% 20%+ 32,953 6.6% 9.2% 30%+ 27,732 8.7% 8.9% 50%+ 19,723 10.6% 8.2% 100%+ 8,329 10.1% 6.9% HORSE RACING MODEL ANALYSIS 19 Hurdle The model top pick returned 4.7% at a 32.0% strike rate and average BSP 5.02. The BSP favourite returned 1.4% at a 37.1% strike rate and average BSP 3.23. The model was profitable in 3/3 rolling years; the BSP favourite was profitable in 2/3. The larger, more repeatable positive race categories are Hurdle Juvenile, Hurdle Novice Handicap, Hurdle Two Mile Handicap. Headline comparison Selection ROI Strike Avg BSP Profitable years Races Model top pick 4.7% 32.0% 5.02 3/3 8,268 BSP favourite 1.4% 37.1% 3.23 2/3 8,268 ROI by rolling year and seed Year Seed 7 Seed 19 Seed 31 3-seed mean BSP fav 2023 1.5% 5.3% 0.6% 2.5% -0.5% 2024 9.6% 2.1% 6.9% 6.2% 2.3% 2025 3.6% 8.6% 4.3% 5.5% 2.5% Probability quality Metric Model BSP market Log loss 1.848 1.717 Brier score 0.788 0.749 Top-pick strike rate 31.9% 37.1% Top-3 strike rate 65.2% - ECE 0.003 0.002 HORSE RACING MODEL ANALYSIS 20 Hurdle race groups Regime Regime Races Model ROI BSP fav ROI Strike Avg BSP Pattern Stakes 169 11.8% -7.1% 40.6% 3.42 Handicap 4,691 6.1% 1.0% 23.8% 6.28 Other Open 341 3.7% 5.4% 40.1% 3.93 Maiden Novice 3,067 2.3% 2.1% 43.1% 3.30 Distance Distance Races Model ROI BSP fav ROI Strike Avg BSP Two Mile 3,471 7.6% -2.9% 35.8% 4.61 Middle 3,457 3.1% 6.6% 30.5% 5.15 Three Mile 1,340 1.5% -0.8% 25.7% 5.72 Surface Surface Races Model ROI BSP fav ROI Strike Avg BSP Turf 8,268 4.7% 1.4% 32.0% 5.02 Race class Race class Races Model ROI BSP fav ROI Strike Avg BSP C1 427 15.9% 0.1% 36.1% 4.70 C5 1,229 15.5% -0.0% 26.8% 5.79 C4 2,130 4.9% 3.3% 38.9% 3.94 Unclassified 3,746 1.1% -1.8% 30.0% 5.35 C3 483 -1.1% 11.9% 29.5% 4.88 C2 253 -3.2% 22.6% 24.4% 6.27 Sex restriction Sex restriction Races Model ROI BSP fav ROI Strike Avg BSP Open 8,225 4.8% 1.6% 31.9% 5.02 Fillies Mares 43 -0.1% -34.9% 34.9% 3.89