Football betting track record. Every official pick, win or lose.
Only picks that met the published bar — primary recommendation per market (up to three per match, some of them correlated). From 2 October 2026 (model v3): fair market probability of 50% or better and a best price 5–30% above the fair price, at a real price. Earlier picks are graded under the earlier rule, which blended in our xG model, and not recomputed. Nothing here is hidden after the fact.
How Is This Track Record Measured?
A pick only counts toward the official Track Record if it is the primary recommendation for its market (a match can carry up to three: 1X2, Over/Under 2.5 and BTTS; some are correlated, e.g. Under 2.5 and BTTS No often win or lose together, so the effective sample is smaller than the pick count). From 2 October 2026 (model v3) the probability is the fair market price — the margin-free median of all tracked bookmakers — and a pick qualifies when that probability is at least 50% and the best available price is 5% to 30% above the fair price, at a real price; no confidence level and no model confirmation are required. Before that date (model v1 and v2) the probability blended the market with our xG model: a pick needed a probability of at least 50% and an expected value of at least +5%, and from v2 also a probability anchored to market prices, agreement of model and market, and an expected value of at most +30%. Picks graded under an earlier model are not recomputed. Alternative picks, calls below those thresholds, and picks whose recorded odds could not be verified are tracked internally but excluded here — not to flatter the number, but because those were never presented as "the" tip or cannot be settled against a real price.
Every qualifying pick is settled as a flat €10 stake. ROI and win rate are percentages and do not depend on stake size; the €10 figure only illustrates what the profit column would look like in money terms — it is not a bankroll recommendation, and stake sizing remains a personal decision this page does not make for you.
A pick is settled automatically once its match finishes, against the actual result — there is no manual grading step where a bad pick could be quietly dropped.
Fewer than 100 settled picks in a window is treated as an unreliable sample. We still show the number — hiding a losing streak would be worse than showing a noisy one — but we label it clearly instead of implying it proves anything.
Every meaningful change to how a pick is generated ships as a new model version. Picks made under an older version are frozen at the time they were published and are never recalculated after the fact — each version keeps its own real record, wins and losses included.
The original WeBetSmart engine, in service from November 2025 to September 2026. It paired a statistical consensus built from several probability models with an AI-written expert review. A September 2026 audit identified two weaknesses: in 16–27% of analyses the headline winner did not match the consensus favourite, and whether the expert agreed with the consensus was inferred from prose — occasionally misread, which could skew the confidence grade. Every v1 pick remains in the track record exactly as it was published.
Live since 12 September 2026. The headline winner and its probability now come straight from the consensus calculation — the AI no longer chooses them. The expert's own call is stored as an explicit home / draw / away value rather than extracted from text. The aim is internal consistency between a pick and the maths behind it, not a promise of better results. Since 16 September 2026, v2 also applies a stricter bar for the official Track Record: a pick only counts if its probability is anchored to real bookmaker prices, the model and the market agree on the outcome, and the expected value is no higher than +30% (a bigger gap is far more often a model error than a real edge). The rule was applied to every v2 pick, including two that had already been settled. Both were losses, and removing them is disclosed here rather than done silently.
Released on 27 September 2026. A technical release: nothing changes in how tips on this site are selected, rated or admitted to the official Track Record. The version number moved because it is shared across all our sites. Update of 2 October 2026: the probability next to every pick is now the fair market price — the median of all tracked bookmakers' prices with the margin stripped out — instead of a blend with our Poisson xG model, which back-testing showed to be less accurate than the market. The xG model stays on the page as a reference. Value and Official picks now follow one rule: best price 5–30% above the fair price, outcome at least 50% likely. Picks made before this date keep their original status.