This page is the proof. Most tipster sites cherry-pick wins. We publish every settled bet from every engine, with ROI, profit factor, Sharpe ratio, and a 95% confidence interval so you can see whether the numbers are real or noise.
How to read the grade. Grades are currently based on realised P&L and strike rate over settled bets, with statistical significance (p < 0.05) required for the top grades. Grade F means a confirmed loss-maker, automatically flagged for review — we publish those rather than hide them.
On closing-line value (CLV). CLV — how a signal's price compares with the eventual closing price — is the sharpest available evidence that a model beats the market, and it is the standard we intend to grade against. We are not there yet: the CLV column below is empty because our settler is not yet capturing closing prices for every engine. Until it is populated, grades are not CLV-based, and this page previously said they were. That was wrong and we have corrected it. We would rather show you an empty column than a number we cannot stand behind.
Past performance is not indicative of future results. Betting carries risk.
What these numbers are. Every signal our engines publish is recorded at the moment it fires, at the price available then, and later settled against the official result. They are
recorded signals graded against real outcomes — not a record of real money staked. We previously described this as “not paper-trading”, which was inaccurate: no cash is placed on these, and you should read them as a model track record rather than a bank statement.
What is excluded. Figures cover
decided bets only — a decided bet is one that won or lost. Voided and non-runner bets (stake returned) are excluded from the bet count, the win rate and the ROI denominator, and are published as a separate count instead. Bets still awaiting a result are also excluded — a meaningful share of signals are pending at any time, and until they settle these totals describe only the decided subset, which is not a random sample of everything we fired. Each engine's pending count is shown so you can judge that yourself.
Correction, 28 July 2026. Until today the “bets” column counted every row our settler had touched, including bets that had not yet run. That made some engines look busier than they were — gh_lay was published as “875 bets” when its own win/loss/void breakdown came to 685. The bet count is now derived directly from wins + losses, so the headline always equals the sum of the parts printed beside it. ROI is now stated as profit divided by the stake risked on decided bets, with void stakes out of the denominator, and the same definition is used on this page,
/transparency and
/scorecard. Some published figures moved as a result. The old ones were wrong.
Sample size. This is a young, live record: it is not statistically conclusive and a single day can swing the totals. We do not back-test our way to these figures and we do not retroactively remove losing engines — engines graded D and F are shown alongside the rest.
18+ · BeGambleAware.org.