HORSE RACING MODEL ANALYSIS 1 Horse Racing Model Analysis Rolling out-of-sample tests across five racing populations 5 populations Flat 2yo, Flat 3yo+, Hurdle, Chase and NH Flat 3 rolling test years 2023, 2024 and 2025 3 model seeds per year 45 report rows in the manifest HORSE RACING MODEL ANALYSIS 2 Contents Section Contents 1 Headline results 2 Probability calibration 3 Population analysis 4 Exacta and trifecta hit rates 5 Race-category appendix 6 Method and definitions HORSE RACING MODEL ANALYSIS 3 1. Headline results The charts below compare the model top pick directly with the BSP favourite in each population. Model figures are mean results across the nine seed-run tests. BSP favourite figures are the corresponding benchmark on the same rolling test windows. HORSE RACING MODEL ANALYSIS 4 Population Model ROI BSP fav ROI Model strike BSP fav strike Model BSP BSP fav BSP Races Chase 8.5% -3.4% 27.0% 31.5% 5.63 3.50 4,462 Hurdle 4.6% 1.4% 31.4% 37.1% 5.33 3.23 8,268 Flat 2yo 4.4% -4.1% 34.4% 38.0% 4.31 2.90 3,835 Flat 3yo+ 1.2% -0.3% 26.3% 31.4% 5.28 3.62 18,763 NH Flat 0.2% 2.2% 32.6% 39.9% 4.59 2.87 1,169 Reading the headline section ROI is where the bettor edge lives. Strike rate and average BSP explain how that ROI is being achieved: a model can win less often than the BSP favourite and still outperform if it wins at meaningfully bigger prices. The stability chart shows whether the ROI edge survives the different rolling test years rather than appearing in one isolated run. HORSE RACING MODEL ANALYSIS 5 2. Probability calibration Each chart uses grouped 5% BSP-implied probability bins: 0-5%, 5-10%, 10-15% and so on. The points are the grouped aggregates; the line through them is only a smooth visual aid. The model line is the mean of the three seed probabilities, the market line is the mean raw BSP-implied probability, and the observed line is the realised win rate. Tail bins with fewer than 30 runners are excluded. HORSE RACING MODEL ANALYSIS 6 HORSE RACING MODEL ANALYSIS 7 HORSE RACING MODEL ANALYSIS 8 HORSE RACING MODEL ANALYSIS 9 HORSE RACING MODEL ANALYSIS 10 3. Population analysis Each population section starts with the headline betting result, then shows year-by-year ROI, race-group breakdowns, race- category ROI charts, and finally the perceived-value section. Chase Across the nine seed-run tests, the model top pick returned 8.5% at a 27.0% strike rate and average BSP 5.63. The BSP favourite returned -3.4% at a 31.5% strike rate and average BSP 3.50. On the rolling years alone, the model finished profitable in 3/3 years. The BSP favourite finished profitable in 0/3 years. The strongest race-category evidence is concentrated in Chase Three Mile Other Open, Chase Novice, Chase Novice Handicap. Headline comparison Selection ROI Strike Avg BSP Profitable years Races Model top pick 8.5% 27.0% 5.63 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 3.5% 1.5% 7.7% 4.2% -2.6% 2024 7.8% 6.3% 9.0% 7.7% -3.6% 2025 13.8% 16.9% 9.9% 13.5% -3.9% Probability quality Metric Model BSP market Log loss 1.896 1.776 Brier score 0.821 0.787 Top-pick strike rate 27.0% 31.5% Top-3 strike rate 59.9% - HORSE RACING MODEL ANALYSIS 11 Metric Model BSP market ECE 0.004 0.005 Race groups Regime Races Model ROI BSP fav ROI Strike Avg BSP Other Open 201 40.7% -6.2% 36.7% 4.41 Maiden Novice 559 18.7% -4.2% 38.0% 4.42 Handicap 3,562 5.6% -3.3% 24.3% 5.99 Pattern Stakes 140 -5.6% 3.3% 38.8% 3.06 Distance Races Model ROI BSP fav ROI Strike Avg BSP Extreme 124 17.7% -11.5% 20.7% 7.34 Three Mile 1,557 11.7% -4.9% 25.9% 5.79 Middle 1,855 7.2% 0.8% 28.2% 5.53 Two Mile 926 4.1% -8.0% 27.4% 5.35 Race-category ROI Stronger race categories Category Races Model ROI BSP fav ROI Profitable Chase Three Mile Other Open 114 56.4% -8.6% 8/9 Chase Novice 559 18.7% -4.2% 9/9 Chase Novice Handicap 649 10.6% 8.2% 8/9 HORSE RACING MODEL ANALYSIS 12 Category Races Model ROI BSP fav ROI Profitable Chase Three Mile Handicap 1,086 9.6% -8.4% 6/9 Chase Middle Handicap 1,134 3.4% 0.4% 7/9 Weak race categories No category meets the negative filter: at least 100 races, negative aggregate ROI and at most three profitable seed-runs. Category filters are descriptive rather than a clean forward-tested strategy. They are useful for identifying where the edge appears to live and where it clearly does not. 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 BSP-implied probability. The section below uses BSP 3.0-33.0 and includes all runners in the relevant value band, not only the final model top pick. Within BSP 3.0-33.0, the strongest cumulative threshold with at least 500 seed-run bets is 100%+ perceived value, returning 16.2% across 8,841 bets. Threshold results should be read as descriptive filters from the same test set, not as a guaranteed live edge. Minimum value Seed-run bets ROI Strike 0%+ 45,688 7.5% 10.2% 5%+ 42,280 7.5% 9.9% 10%+ 39,092 7.6% 9.7% 20%+ 33,223 8.2% 9.2% 30%+ 28,112 8.7% 8.7% 50%+ 19,946 10.0% 7.9% 100%+ 8,841 16.2% 6.9% HORSE RACING MODEL ANALYSIS 13 Hurdle Across the nine seed-run tests, the model top pick returned 4.6% at a 31.4% strike rate and average BSP 5.33. The BSP favourite returned 1.4% at a 37.1% strike rate and average BSP 3.23. On the rolling years alone, the model finished profitable in 3/3 years. The BSP favourite finished profitable in 2/3 years. The strongest race-category evidence is concentrated in Hurdle Novice Handicap, Hurdle Two Mile Handicap, Hurdle Middle Handicap. Headline comparison Selection ROI Strike Avg BSP Profitable years Races Model top pick 4.6% 31.4% 5.33 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 2.3% -1.5% 0.8% 0.5% -0.5% 2024 0.3% 4.6% 4.4% 3.1% 2.3% 2025 10.9% 7.6% 12.2% 10.3% 2.5% Probability quality Metric Model BSP market Log loss 1.868 1.717 Brier score 0.793 0.749 Top-pick strike rate 31.4% 37.1% Top-3 strike rate 64.5% - ECE 0.003 0.002 HORSE RACING MODEL ANALYSIS 14 Race groups Regime Races Model ROI BSP fav ROI Strike Avg BSP Handicap 4,691 7.8% 1.0% 23.6% 6.73 Maiden Novice 3,067 1.4% 2.1% 42.2% 3.41 Other Open 341 -4.3% 5.4% 38.0% 4.11 Pattern Stakes 169 -8.3% -7.1% 39.1% 3.60 Distance Races Model ROI BSP fav ROI Strike Avg BSP Three Mile 1,340 9.0% -0.8% 25.6% 6.21 Two Mile 3,471 4.6% -2.9% 34.8% 4.80 Middle 3,457 3.0% 6.6% 30.2% 5.51 Race-category ROI Stronger race categories Category Races Model ROI BSP fav ROI Profitable Hurdle Novice Handicap 575 11.5% 11.2% 6/9 Hurdle Two Mile Handicap 1,309 9.0% -5.2% 8/9 Hurdle Middle Handicap 1,778 5.3% 3.0% 7/9 Weak race categories No category meets the negative filter: at least 100 races, negative aggregate ROI and at most three profitable seed-runs. Category filters are descriptive rather than a clean forward-tested strategy. They are useful for identifying where the edge appears to live and where it clearly does not. HORSE RACING MODEL ANALYSIS 15 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 BSP-implied probability. The section below uses BSP 3.0-33.0 and includes all runners in the relevant value band, not only the final model top pick. Within BSP 3.0-33.0, the strongest cumulative threshold with at least 500 seed-run bets is 100%+ perceived value, returning 7.5% across 15,792 bets. Threshold results should be read as descriptive filters from the same test set, not as a guaranteed live edge. Minimum value Seed-run bets ROI Strike 0%+ 76,421 2.4% 9.6% 5%+ 71,009 2.9% 9.4% 10%+ 65,789 2.9% 9.2% 20%+ 56,376 2.8% 8.8% 30%+ 48,245 3.3% 8.5% 50%+ 35,014 4.7% 7.8% 100%+ 15,792 7.5% 6.5% HORSE RACING MODEL ANALYSIS 16 Flat 2yo Across the nine seed-run tests, the model top pick returned 4.4% at a 34.4% strike rate and average BSP 4.31. The BSP favourite returned -4.1% at a 38.0% strike rate and average BSP 2.90. On the rolling years alone, the model finished profitable in 2/3 years. The BSP favourite finished profitable in 0/3 years. The strongest race-category evidence is concentrated in Flat 2yo Mile Stakes, Flat 2yo Sprint Maiden Novice. The clearest negative categories are Flat 2yo Mile Maiden Novice, Flat 2yo Mile Handicap. Headline comparison Selection ROI Strike Avg BSP Profitable years Races Model top pick 4.4% 34.4% 4.31 2/3 3,835 BSP favourite -4.1% 38.0% 2.90 0/3 3,835 ROI by rolling year and seed Year Seed 7 Seed 19 Seed 31 3-seed mean BSP fav 2023 1.8% -4.9% 4.0% 0.3% -8.3% 2024 -3.5% 0.7% 0.3% -0.8% -2.0% 2025 14.7% 16.3% 10.6% 13.8% -2.1% Probability quality Metric Model BSP market Log loss 1.779 1.655 Brier score 0.776 0.739 Top-pick strike rate 34.5% 38.0% Top-3 strike rate 68.4% - ECE 0.005 0.005 HORSE RACING MODEL ANALYSIS 17 Race groups Regime Races Model ROI BSP fav ROI Strike Avg BSP Other Open 140 13.4% -9.5% 38.6% 4.08 Pattern Stakes 248 10.3% -2.4% 37.5% 4.08 Maiden Novice 2,621 4.8% -3.8% 36.9% 3.84 Handicap 826 0.0% -5.0% 25.0% 5.92 Distance Races Model ROI BSP fav ROI Strike Avg BSP Stayer 3 21.0% 21.0% 66.7% 2.01 Sprint 1,971 9.9% -3.8% 35.4% 4.45 Mile 1,783 -0.6% -3.9% 33.4% 4.18 Middle 78 -18.0% -18.8% 30.8% 4.07 Race-category ROI Stronger race categories Category Races Model ROI BSP fav ROI Profitable Flat 2yo Mile Stakes 114 15.5% -5.2% 7/9 Flat 2yo Sprint Maiden Novice 1,340 11.5% -3.0% 9/9 Weak race categories Category Races Model ROI BSP fav ROI Profitable Flat 2yo Mile Handicap 381 -3.3% -0.8% 3/9 HORSE RACING MODEL ANALYSIS 18 Category Races Model ROI BSP fav ROI Profitable Flat 2yo Mile Maiden Novice 1,234 -2.4% -4.0% 3/9 Category filters are descriptive rather than a clean forward-tested strategy. They are useful for identifying where the edge appears to live and where it clearly does not. 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 BSP-implied probability. The section below uses BSP 3.0-33.0 and includes all runners in the relevant value band, not only the final model top pick. Within BSP 3.0-33.0, the strongest cumulative threshold with at least 500 seed-run bets is 100%+ perceived value, returning 17.0% across 7,111 bets. Threshold results should be read as descriptive filters from the same test set, not as a guaranteed live edge. Minimum value Seed-run bets ROI Strike 0%+ 33,402 6.3% 10.6% 5%+ 31,185 7.0% 10.4% 10%+ 29,064 8.0% 10.3% 20%+ 25,150 9.0% 9.9% 30%+ 21,662 9.2% 9.5% 50%+ 15,881 11.6% 8.9% 100%+ 7,111 17.0% 7.8% HORSE RACING MODEL ANALYSIS 19 Flat 3yo+ Across the nine seed-run tests, the model top pick returned 1.2% at a 26.3% strike rate and average BSP 5.28. The BSP favourite returned -0.3% at a 31.4% strike rate and average BSP 3.62. On the rolling years alone, the model finished profitable in 3/3 years. The BSP favourite finished profitable in 2/3 years. The strongest race-category evidence is concentrated in Flat 3yo Plus Stayer Maiden Novice, Flat 3yo Plus Cup Handicap, Flat 3yo Plus Sprint Handicap. The clearest negative categories are Flat 4yo Plus Mile Handicap, Flat 3yo Middle Maiden Novice, Flat 3yo Plus Sprint Maiden Novice. Headline comparison Selection ROI Strike Avg BSP Profitable years Races Model top pick 1.2% 26.3% 5.28 3/3 18,763 BSP favourite -0.3% 31.4% 3.62 2/3 18,763 ROI by rolling year and seed Year Seed 7 Seed 19 Seed 31 3-seed mean BSP fav 2023 0.1% 0.1% 1.2% 0.5% -2.5% 2024 0.8% 1.4% 0.4% 0.9% 1.1% 2025 3.2% 1.8% 2.2% 2.4% 0.5% Probability quality Metric Model BSP market Log loss 1.982 1.858 Brier score 0.828 0.796 Top-pick strike rate 26.3% 31.4% Top-3 strike rate 58.7% - ECE 0.002 0.002 HORSE RACING MODEL ANALYSIS 20 Race groups Regime Races Model ROI BSP fav ROI Strike Avg BSP Handicap 14,742 2.0% 0.2% 23.7% 5.69 Other Open 699 -1.1% -1.1% 29.1% 4.63 Pattern Stakes 926 -1.2% -5.2% 30.5% 4.31 Maiden Novice 2,396 -1.5% -0.8% 39.5% 3.28 Distance Races Model ROI BSP fav ROI Strike Avg BSP Cup 695 7.6% -4.1% 27.7% 5.60 Stayer 3,064 3.9% 4.3% 29.0% 4.97 Sprint 5,076 3.7% 1.9% 24.6% 5.49 Middle 2,921 -0.7% -3.5% 26.3% 5.04 Mile 7,007 -1.5% -2.2% 26.2% 5.32 Race-category ROI Stronger race categories Category Races Model ROI BSP fav ROI Profitable Flat 3yo Plus Stayer Maiden Novice 222 19.4% 10.9% 9/9