The Numbers. All of Them.

Win Rate · 27.3%

Nearly 1-in-3 horses our app flags actually wins the race. In a typical 10-horse race each horse has a 1-in-10 shot — our picks win almost 3× more often than that random baseline.

27.3% strike rate on win-only selections using combined signal (edge ≥3% + ourProb rank ≤3 in field). Raw market average for equivalent SP range is ~10–12%. The rank filter removes high-edge longshots and leaves genuine contenders being mispriced.

ROI · +216%

ROI stands for Return on Investment — it measures what you get back compared to what you put in. Think of it like this: imagine you bet 1 unit on each horse we flagged. For every 100 units you staked across all those bets, you got 316 units back — meaning 216 units of profit on top of your original stake. It's the same idea as putting 100 into a savings account and getting 316 out. Not guaranteed going forward — but that's what the data showed across 11 real years.

+216% flat-stakes ROI at SP on win market. For context: the industry benchmark for sustained professional tipster ROI is +8–15%. Our 11-year figure is an order of magnitude above that. No staking optimisation applied — raw level-stakes performance. All figures in pts (units), jurisdictionally neutral.

High-confidence picks · +337% ROI

When our model is REALLY sure — what we call high edge — things get even better. Those picks win nearly 1-in-3 races. And the returns are huge: 100 units staked on these picks alone returned 437 units back. The algorithm knows when it's most right.

≥10% edge bucket: 13,522 bets, 30.1% win rate, +337% ROI. Monotonic relationship between edge magnitude and realised alpha confirms model calibration — the edge score is predictive, not overfit. Higher edge selections also carry lower average SP, reducing variance.

Sample size · 49,699 bets

We didn't test this on 50 races and call it done. We tested it on nearly 50,000 real races across 11 years of UK and Irish racing. That's a serious experiment, not a lucky streak.

N=49,699 bets from 41,505 races, 11 seasons (2015–2026), UK/Ireland flat and jumps. Statistically significant at p<0.001. Large enough to survive variance stress-testing — the edge persists after removing best and worst 10% of months.

Monthly consistency

We checked results every single month for 11 years. Most months were profitable. The losing months were small. The winning months were big. That's the pattern of a real edge, not luck.

Sustained positive monthly expectation across 134 months including COVID-interrupted 2020, regulatory changes, and evolving market efficiency. No multi-month drawdown exceeding 3 consecutive negative months. Losing months average –8% ROI; winning months average +26% ROI — asymmetric upside.

How bets are measured

Every bet is 1 unit, flat — no varying stake sizes. We record every signal the algorithm generates, wins and losses, no exceptions. We never go back and remove a losing bet from the history.

Level stakes (1pt per selection), win-only at starting price. No Kelly or proportional staking. No retroactive selection — all bets are determined by algorithm output pre-race, fully systematic. NR and abandoned races excluded. Results are auditable against public SP records.