Research note · 5 September 2026 · Historical backtest
Power horse racing picks: our backtest and its limits
The saved test covers 2025-10-01 through 2026-07-31. It produced 86.22 points of profit across 1,388 financially settled selections, using Betfair Starting Price (BSP) and 2% commission on positive profit. That is a historical finding, not an established live betting edge.
What the test selected
The Power de-vig method removes the bookmaker margin from the full captured morning field. It gives a market-derived probability for each runner. This is not an independent prediction based on form.
- Use races with at least eight morning runners and the required priced-field coverage.
- Keep eligible horses with morning median decimal odds strictly above 2.00 and no higher than 7.00: above evens through 6/1.
- Select the highest Power-probability eligible horse in each race, then rank those selections across the day and keep up to five.
There is no extra 35% probability threshold. Fewer qualifiers means fewer selections. AI commentary is not used to choose or reorder horses.
The recorded results
- Selected
- 1465
- Financially settled
- 1388
- Winners with settled returns
- 527
- Unresolved result / BSP
- 77
- Net return
- 86.22pt
- ROI on settled stakes
- 6.21%
- Maximum drawdown
- 34.48pt
72 selections have unresolved results and 5 are awaiting BSP. They remain visible in the selection list and are excluded from the settled-return calculation. These figures can therefore be affected by missing outcomes.
Inspect the graph and individual backtest selectionsWhat BSP after commission means
Each selection is a one-point win single. A normal losing bet returns −1pt. For a sole winner, profit is (BSP − 1) × 0.98. For example, BSP 3.00 produces +1.96pt after 2% commission. Non-runners are void; dead heats divide the gross payout before commission on any positive profit.
The morning median determines eligibility. It is not assumed to be an available bookmaker quote or used to calculate these returns. Taking fixed odds or using Best Odds Guaranteed is a different execution method and will not necessarily reproduce this record. The 2% rate is the test’s assumption; an individual account may pay a different commission rate.
Why this does not prove profitability
The selection rules were developed using the same panel data. Choosing rules after inspecting results can make a historical strategy look better than it will perform on new races. The period is not an untouched validation sample, and morning snapshots may differ from the live capture in timing and coverage.
The test also includes a 34.48pt peak-to-trough fall. A positive final return does not mean steady profits or a low-risk sequence of bets. Neither a high win rate nor a shortening price alone demonstrates value.
How we will assess the live version
The live record uses the selections actually published before racing. It keeps the frozen ranks, losses, voids and pending settlements visible. September 5’s reconstructed preview is excluded. The jumps favourites trial and previous AI strategy have separate records.
Future reviews should report the complete published period, sample size, net return and unresolved results together. Until that live evidence accumulates, this remains a strategy under evaluation.