Backtesting lab

Which strategy actually held up?

Every settled selection is replayed in date order under the rules you set below. Nothing is re-scored and no losing bet is dropped — only the entry filters and the staking plan change.

Loading the settled record…

Strategy rules

Minimum model probability70%
Minimum expected valueNo filter
Maximum odds15.00
Starting bankroll£1,000
Staking plan
Kelly fraction0.25×
Markets included
Load preset

No settled bets match these rules

Loosen the probability, EV or market filters. An empty sample is shown as empty — it is never padded out with weaker selections.

Strategy comparison

Every preset replayed over the same settled history from £1,000, ranked by ROI.

StrategyBetsHit rateAvg oddsProfitROIMax DDRisk-adj.
Everything publishedEvery settled selection, flat £10 stakes. The control group.0£0.000.0%0.00
Singles 70%+Match-result singles at 70%+ model probability, quarter Kelly.0£0.000.0%0.00
Positive EV onlyAny market where the model priced an edge over the book.0£0.000.0%0.00
Elite 82%+Only the highest-conviction band, quarter Kelly.0£0.000.0%0.00
Beat the closing lineSelections taken at a better price than the market closed at.0£0.000.0%0.00
Singles, no accumulatorsExcludes multiples entirely — 2% of bankroll level stakes.0£0.000.0%0.00

A backtest is a description of the past, not a forecast. Sample sizes here are small, the best-performing threshold on one season is frequently not the best on the next, and every strategy shown includes its losing runs in full.