Research
Agreement filter: win bets require the market favourite to be uniquely first in WF’s predictions. Winner-to-place uses this same win-market agreement.
- Ending bankroll
- 1516.13
- Profit (units)
- 516.13
- ROI
- 51.6%
- Hit rate
- 48.0%
- Max. drawdown
- 15.7%
- Bets placed
- 50
Bankroll performance
Start: 1,000.00 · End: 1,516.13 units
Dashed line = starting bankroll · Days are equally spaced
ROI and drawdown by WF weight
51.6% ROI
-15.7% Max DD
-5.8% Avg DD
Performance by race track
Daily profit and loss
19 profitable · 18 losing · 31 flat days
Daily change after settlement · Flat days include skipped or refunded bets
Individual picks
| Meeting date | Bets | Skipped | Staked (u) | Profit / loss (u) |
|---|---|---|---|---|
| 0 | 2 | 0.00 | 0.00 | |
| 1 | 6 | 30.36 | +50.70 | |
| 1 | 4 | 11.36 | -11.36 | |
| 0 | 7 | 0.00 | 0.00 | |
| 1 | 4 | 6.47 | -6.47 | |
| 1 | 2 | 35.97 | -35.97 | |
| 1 | 3 | 5.69 | +8.14 | |
| 1 | 3 | 13.71 | +9.87 | |
| 1 | 4 | 4.43 | -4.43 | |
| 0 | 3 | 0.00 | 0.00 |
About this model
The runner is selected in the chosen betting market. The selected runner has the highest weighted probability: WF probability × WF weight + market probability × market weight. At 100% you follow WF; at 0% you follow the market. Kelly sizes that selected bet using its probability and price, skipping it when there is no positive estimated edge. Win bets also require the favourite at the selected odds timing to be uniquely first in WF’s predictions; tied WF firsts and disagreements are skipped. Tied market favourites use the lower runner number. Stride blend place bets do not use this agreement filter; winner-to-place retains the win-market agreement filter. The default blend is 25% WF and 75% market. Blended picks below the minimum estimate are skipped before allocating stakes. Both filters apply to each weight and track simulation and saved-plan recommendations.
Full Kelly is (probability × decimal odds − 1) ÷ (decimal odds − 1). The Kelly fraction scales that amount (25% means quarter Kelly), then the stake cap limits it as a percentage of bankroll. ROI is profit divided by starting bankroll; hit rate counts funded, non-refunded bets only.
The simulation processes SAST days from oldest to newest. Each day’s stakes use its opening bankroll; returns settle at day end. If total planned stakes exceed available capital, they are reduced proportionally. Stakes round down to 0.01 units and payouts round to 0.01. Bets below the fixed minimum of 1 units after scaling and rounding are skipped, with their funds left in the bankroll. Maximum drawdown uses day-end balances. Kelly depends on estimated probabilities, so the uncalibrated model can overstate stakes. A selected runner scratched after the morning bet receives a full stake refund at settlement, with zero profit or loss. Its stake still counts towards that day’s committed budget.
Within the selected date range, 234 races are excluded for incomplete or unsupported data. Every active runner needs usable WF predictions and selected-market odds. Winner-to-place needs both win and place odds. Historical races also need known results or confirmed scratchings for all runners in that field. Missing results alone never imply a refund. Place races need at least three paid positions.
264 runners already scratched in the selected price batch were removed. Later scratchings stay in the morning field and retain their original selection and stake; they are refunded rather than replaced with another horse. All models use the same complete field.
WF probabilities are experimental estimates, not calibrated chances. The model converts predicted lengths behind to strengths using exp(−lengths), with a fixed spread of 1 length. A Plackett–Luce model estimates win or paid-place probabilities. Market estimates come from the selected odds, adjusted for the bookmaker margin.
Selections use the selected price batch and WF predictions, never later finishing positions or scratchings. Morning returns include estimated scratching deductions from winnings, using the withdrawn runners’ morning win prices and the current Hollywoodbets table. 0 races use these estimates. Multiple deductions are compounded as a simulation assumption. Withdrawal-time prices and reserve replacements can change the actual deduction. Place bets are refunded if scratchings leave three or fewer runners. Morning inputs and place terms may have been overwritten; closing odds are a retrospective benchmark. Dead heats are not modeled.