The SharpXI model track record

A model is only worth its calibration. Every prediction below was made before kick-off from earlier data only, then checked against what actually happened — across every league we cover, graded market by market against a base-rate baseline. No cherry-picking, no hindsight, and no blended headline hiding a weaker league.

Each league is graded on its own data because sample sizes and edges genuinely differ — a single blended number would flatter the weaker ones. Open any league for the full market-by-market breakdown and calibration curves.

Three of every four prices we flagged moved our way

Calibration shows the probabilities are honest. This is the harder test: when the model calls a price value, does the market later agree? We write down the price the moment we flag it, then read the same leg again minutes before kick-off.

Across all 3,186 legs we beat the close on 45.7%, matched it on 40.2% and lost to it on 14.1%. The average leg is +14.62% and the median leg is +0.00% — four prices in ten never move at all, so that average is made entirely by the ones that did. Among those, the median is +10.00%.

A further 571 legs closed but are not counted here, because no price was seen within 2 hours of kick-off. A quote read the previous evening is not a closing price, and counting one would quietly flatter every number above. The median leg that is counted was read 6 minutes before kick-off.

How it is measured

1. The model rates a leg above its edge floor at the live bet365 price — an average flagged edge of 21.9%, before the bookmaker's margin.
2. That first price is written down and never revised, even if the same leg is flagged again.
3. Minutes before kick-off the same leg, same bookmaker, same line is read a second time.
4. Anything without a close inside 2 hours of kick-off is dropped — never counted as unchanged.

15 August to 9 October363 matches 3 bookmakers, every league pooled, because no single league has enough legs yet to stand on its own. Updated nightly. This measures price, not profit — it says a bet was value when it was placed, not that it won. How closing line value works.

Every selection we send, graded

Each selection below was published before its matches kicked off, then graded when they finished. Nothing is added afterwards and nothing that lost is left out.

Bet builders, two or more legs

Single selections, one leg

A bet builder needs at least 2 legs. These are counted on their own line, because a single at 11/10 and a builder at 30/1 are not the same bet and one strike rate cannot describe both.