Track record
Every model pick graded on the 90-minute score, then broken down by probability band, market and league – including where the model struggles. Full calibration test →
Probability bands
A calibrated model is right about seven times in ten on its 70% picks. Each band below sets the average predicted probability against the observed hit rate:
landed average model probability · Full model probability proof →
Accuracy by market
| # | Market | Picks | Hit rate |
|---|---|---|---|
| 1 | Both teams to score: yes | 2376 | 57% |
| 2 | Match result (1X2) | 1702 | 64% |
| 3 | Over 1.5 goals | 987 | 73% |
| 4 | Under 3.5 goals | 970 | 71% |
| 5 | Double chance | 834 | 77% |
| 6 | Over 2.5 goals | 715 | 70% |
| 7 | Under 2.5 goals | 350 | 62% |
| 8 | Both teams to score: no | 107 | 64% |
| 9 | Over 3.5 goals | 16 | 50% |
Accuracy by league
| # | League | Picks | Hit rate |
|---|---|---|---|
| 1 | Major League Soccer USA | 61 | 62% |
| 2 | Liga Profesional Argentina Argentina | 49 | 67% |
| 3 | Championship England | 49 | 69% |
| 4 | Copa Del Rey Spain | 42 | 71% |
| 5 | Serie A Italy | 42 | 69% |
| 6 | FA Cup England | 38 | 71% |
| 7 | UEFA Europa Conference League Europe | 36 | 56% |
| 8 | UEFA Champions League Europe | 36 | 75% |
| 9 | UEFA Europa League Europe | 35 | 60% |
| 10 | Serie A Brazil | 33 | 64% |
| 11 | La Liga Spain | 32 | 69% |
| 12 | Premier League England | 31 | 55% |
| 13 | Süper Lig Turkey | 28 | 64% |
| 14 | Eredivisie Netherlands | 27 | 67% |
| 15 | Ligue 1 France | 27 | 63% |
| 16 | Bundesliga Germany | 27 | 59% |
| 17 | Primeira Liga Portugal | 27 | 52% |
| 18 | FKF Premier League Kenya | 25 | 52% |
| 19 | Pro League Saudi-Arabia | 23 | 70% |
| 20 | Premiership Scotland | 21 | 71% |
| 21 | DFB Pokal Germany | 16 | 50% |
| 22 | NPFL Nigeria | 14 | 71% |
| 23 | Coupe De France France | 13 | 69% |
| 24 | UEFA U19 Championship Qualification Europe | 70 | 57% |
| 25 | Friendlies World | 64 | 73% |
| 26 | FA Trophy England | 62 | 65% |
| 27 | Reserve League Argentina | 61 | 61% |
| 28 | UEFA Youth League Europe | 61 | 62% |
| 29 | J3 League Japan | 60 | 62% |
| 30 | J2 League Japan | 58 | 64% |
| 31 | First Amateur Division Belgium | 58 | 67% |
| 32 | Non League Premier Southern South England | 54 | 67% |
| 33 | UEFA Nations League Europe | 52 | 85% |
| 34 | National League South England | 52 | 65% |
| 35 | Primera A Colombia | 52 | 69% |
| 36 | 2. Lig Turkey | 52 | 71% |
| 37 | League One England | 49 | 59% |
| 38 | Africa Cup Of Nations Qualification Africa | 48 | 67% |
| 39 | Non League Premier Southern Central England | 48 | 71% |
| 40 | Liga Alef Israel | 48 | 54% |
Leagues with at least 8 graded picks. Tap a column to sort.
How these numbers are produced
Claimed win rates are easy to invent. These figures are computed automatically from every graded pick, so weak markets and leagues show up as clearly as strong ones. Selections freeze at kick-off and are never edited; the daily list is on the results page.
Accuracy must be read with odds in mind: short-priced markets win often and pay little, value picks win less often at bigger prices. The leaderboard converts both into profit at level stakes. Past results are no guarantee – 18+.
The xGaura track record: public, graded and unedited
A prediction model is only as credible as its record. This page summarises every graded pick over recent periods – picks made, picks won, hit rate, average odds and profit at level stakes – with the figures generated directly from the results archive. We publish the bad weeks as clearly as the good ones.
Why hit rate is not enough
A hit rate of 70% sounds excellent, but if the average odds were 1.30, backing every pick would lose money: 70 winners at 1.30 return 91 units on 100 staked. A hit rate of 45% at average odds of 2.40 returns 108 units on 100 staked. The record therefore shows both hit rate and profit, and the profit figure is the one that tells you whether the picks have value.
Level stakes, explained
Profit is calculated as if one unit were placed on every pick. A won pick at odds of 1.80 adds 0.80 units; a lost pick subtracts one. This keeps the comparison fair across picks of different prices and removes any temptation to inflate results with selective staking. Yield – profit divided by total stakes – is the cleanest single measure of performance.
Periods and variance
Thirty-day figures react quickly but swing widely; ninety-day figures are steadier. Even a genuinely profitable model will have losing months, because football is low-scoring and individual results are noisy. As a rough guide, a few hundred graded picks are needed before a yield of a few per cent can be distinguished from luck with any confidence.
What is included
- Every model pick published on a match page, graded on the 90-minute score.
- Slip results for each accumulator type, recorded as a single bet per slip.
- Value bet index selections, tracked separately so that the value strategy can be judged on its own.
Voids – postponed matches and refunded stakes – are excluded from hit rates and profit.
Calibration: the deeper test
Hit rate and profit tell you whether picks won and paid. Calibration tells you whether the probabilities themselves were accurate: whether picks rated at 60% actually landed about 60% of the time. A model can be well calibrated and still lose money if its prices are poor, or profitable for a while despite poor calibration. Looking at both gives the full picture.
Using the record
Treat it as evidence, not a promise. If a market or league has performed well over ninety days, that is a reason to look more closely – not a guarantee it will continue. The leaderboard breaks the record down by market and competition. Over 18s only.
How to judge a record quickly
Look at three numbers together: the number of picks, the yield and the average odds. A high yield on a few dozen picks is mostly luck; a modest positive yield on several hundred is meaningful. Average odds tell you how volatile the record will be – long prices mean longer losing runs, even when the long-run yield is positive.
Frequently asked questions
How accurate is the xGaura model?
66% over 30 days: 5293 correct from 8057 graded picks, with nothing removed.
Which league suits the model best?
Currently the UEFA Nations League (Europe), 85% from 52 picks – league rankings shift month to month.
Do higher probabilities win more often?
That is what the probability bands test – and the full calibration page goes further.
What score is used for grading?
The score after 90 minutes and stoppage time. Extra time and penalties are ignored; postponed or abandoned matches are void.