Track record
We publish two things most prediction sites hide: how our real, already-published predictions have turned out, and how the model performs on an out-of-sample backtest. Nothing here is cherry-picked.
Live record — our real predictions
Every prediction we've published that has since been played, graded against the actual result — from 31 Jul 2026 to 3 Nov 2026.
| League | Settled | Accuracy | Value ROI |
|---|---|---|---|
 Major League Soccer |
158 | 43.7% | -100.0% (1) |
 League One |
153 | 45.1% | -57.8% (30) |
| Championship |
152 | 48.7% | -47.7% (21) |
 La Liga 2 |
128 | 48.4% | -39.3% (24) |
 Liga Profesional |
108 | 46.3% | -100.0% (1) |
| La Liga |
106 | 55.7% | -2.6% (12) |
 Serie A |
100 | 63.0% | +33.8% (16) |
 Eredivisie |
98 | 53.1% | +30.6% (18) |
| Serie B |
98 | 39.8% | +8.7% (18) |
 Ligue 2 |
97 | 43.3% | +5.3% (25) |
 Super League |
93 | 60.2% | -68.0% (11) |
 Primeira Liga |
89 | 57.3% | +6.9% (11) |
 2. Bundesliga |
88 | 48.9% | -64.4% (15) |
 Süper Lig |
88 | 52.3% | -51.8% (20) |
| Premier League |
88 | 48.9% | -28.6% (17) |
 Jupiler Pro League |
87 | 60.9% | -10.7% (17) |
| Ligue 1 |
80 | 53.8% | -100.0% (9) |
 Brasileirão Série A |
79 | 59.5% | — |
 Liga MX |
77 | 41.6% | — |
| Bundesliga |
70 | 52.9% | +52.5% (14) |
 J1 League |
70 | 51.4% | — |
 Scottish Premiership |
69 | 44.9% | -59.3% (9) |
 Ekstraklasa |
59 | 52.5% | — |
 Allsvenskan |
59 | 54.2% | -100.0% (1) |
 Liga I |
51 | 64.7% | — |
 Eliteserien |
48 | 54.2% | — |
 Veikkausliiga |
42 | 50.0% | — |
 Superliga |
37 | 56.8% | — |
 Bundesliga (AUT) |
36 | 55.6% | — |
 Premier Division |
36 | 52.8% | — |
 Chinese Super League |
27 | 48.1% | -100.0% (1) |
Success by market
Match-result accuracy understates the calls we actually publish. Double chance and the goals/BTTS leans land far more often — here's the honest hit-rate on every graded prediction where each applies.
And each market on its own:
| Market | Sample | Correct | Hit rate |
|---|---|---|---|
| Match result Our single most likely outcome (1X2) |
2,571 | 1,313 | 51.1% |
| Double chance Backing our top two outcomes |
2,571 | 1,978 | 76.9% |
| Over / Under 2.5 Total-goals lean from model xG |
1,420 | 830 | 58.5% |
| Both teams to score BTTS yes/no lean from xG |
1,194 | 695 | 58.2% |
| Confident picks Only where our top probability ≥ 55% |
669 | 451 | 67.4% |
| Average across markets Mean of the 5 market hit-rates above |
— | — | 62.4% |
Each market is only counted where it applies — the goals and BTTS leans exist only when the model's xG points clearly one way, so their samples are smaller than the match-result total. Historical hit-rates, not guarantees.
How well-calibrated are we?
Accuracy isn't everything — a probability is only honest if it means what it says. This checks exactly that: when we say 60%, does it happen about 60% of the time? Each dot groups our forecasts by confidence; dots on the dashed line are perfectly calibrated. Built from all 2,571 settled matches (7,713 outcome probabilities).
lower is better
lower is better
ECE is the average gap between what we predict and what happens; the Brier score grades the sharpness and honesty of every probability at once. We'd rather be honestly uncertain than confidently wrong — so we automatically re-calibrate our probabilities against this record, tightening the fit over time.
See every graded fixture, day by day, on the results page → · Predictions are sealed before kickoff — see the ledger →
Model validation — historical backtest
Out-of-sample walk-forward test — every prediction uses only data available before kickoff. Last run 19 Oct 2026.
| League | N | Accuracy | RPS | Bookmaker RPS | Value ROI |
|---|---|---|---|---|---|
| Major League Soccer |
1,310 | 48.2% | 0.2172 | 0.2148 | +8.6% (55) |
| Liga Profesional |
1,124 | 42.7% | 0.2131 | 0.2126 | +10.5% (48) |
| Championship |
1,108 | 46.2% | 0.2150 | 0.2135 | -19.2% (77) |
| League One |
1,083 | 49.7% | 0.2138 | 0.2126 | -21.8% (116) |
| La Liga 2 |
918 | 48.9% | 0.2111 | 0.2092 | -2.5% (76) |
| Brasileirão Série A |
851 | 51.7% | 0.2004 | 0.1963 | +9.2% (81) |
| Serie A |
749 | 54.5% | 0.1905 | 0.1894 | -24.0% (49) |
| Premier League |
747 | 50.6% | 0.2049 | 0.2046 | -5.6% (56) |
| La Liga |
743 | 53.8% | 0.1954 | 0.1938 | +7.7% (29) |
| Serie B |
733 | 46.1% | 0.2029 | 0.2017 | -30.5% (53) |
| Liga MX |
732 | 51.8% | 0.2068 | 0.1946 | +4.8% (32) |
| Süper Lig |
635 | 55.0% | 0.1916 | 0.1885 | -32.9% (52) |
| Liga I |
625 | 50.1% | 0.2037 | 0.2047 | +40.6% (21) |
| Eredivisie |
612 | 51.6% | 0.1947 | 0.1934 | -4.5% (58) |
| Jupiler Pro League |
611 | 50.9% | 0.2007 | 0.1999 | -17.0% (61) |
| Primeira Liga |
610 | 54.9% | 0.1815 | 0.1817 | +29.4% (43) |
| Ligue 2 |
600 | 46.2% | 0.2135 | 0.2123 | -19.3% (51) |
| Ekstraklasa |
600 | 45.0% | 0.2205 | 0.2135 | -50.8% (21) |
| Ligue 1 |
600 | 53.7% | 0.2046 | 0.2031 | -27.1% (38) |
| Bundesliga |
599 | 52.9% | 0.2000 | 0.1973 | -38.6% (51) |
| 2. Bundesliga |
598 | 45.8% | 0.2222 | 0.2192 | -21.3% (68) |
| J1 League |
519 | 49.1% | 0.2177 | 0.2180 | +35.8% (40) |
| Chinese Super League |
471 | 51.4% | 0.1983 | 0.1760 | -10.5% (44) |
| Eliteserien |
464 | 58.2% | 0.2008 | 0.1959 | -84.5% (13) |
| Super League |
462 | 48.7% | 0.2167 | 0.2182 | +1.8% (23) |
| Super League |
459 | 53.8% | 0.1935 | 0.1929 | +60.3% (26) |
| Scottish Premiership |
459 | 53.4% | 0.2019 | 0.2001 | -52.3% (23) |
| Allsvenskan |
455 | 48.6% | 0.2186 | 0.2133 | -35.0% (25) |
 Premier League (RUS) |
406 | 56.2% | 0.1858 | 0.1840 | +43.7% (32) |
| Bundesliga (AUT) |
380 | 45.3% | 0.2202 | 0.2151 | +13.1% (26) |
| Superliga |
376 | 51.3% | 0.2151 | 0.2082 | -43.8% (8) |
| Premier Division |
355 | 45.4% | 0.2063 | 0.2022 | -33.1% (7) |
| Veikkausliiga |
349 | 50.4% | 0.2106 | 0.2055 | +4.3% (20) |
| Liga FWomen |
192 | 54.7% | 0.1958 | — | — |
| Women's Super LeagueWomen |
109 | 63.3% | 0.1762 | — | — |
| NWSLWomen |
21 | 76.2% | 0.1622 | — | — |
| Frauen-BundesligaWomen |
0 | 0.0% | 0.0000 | — | — |
Every market, back-tested
We don't just publish match results. Every market on the prediction pages — goals lines, both-teams-to-score, team totals, clean sheets, winning margins and more — is graded the same walk-forward way, out-of-sample against real scores. Last run 19 Oct 2026.
Result
| Market | Sample | Hit rate | Strength |
|---|---|---|---|
| Draw no bet | 15,548 | 68.0% | Weak |
| Match result (1X2) | 21,037 | 50.2% | Weak |
| Double chance | 21,037 | 47.2% | — |
Goals
| Market | Sample | Hit rate | Strength |
|---|---|---|---|
| Over/Under 0.5 | 21,037 | 93.0% | Strong |
| Over/Under 4.5 | 21,037 | 85.4% | Fair |
| Over/Under 1.5 | 21,037 | 75.0% | Weak |
| Over/Under 3.5 | 21,037 | 70.1% | Weak |
| Over/Under 2.5 | 21,037 | 56.1% | Weak |
| Both teams to score | 21,037 | 54.2% | Weak |
| Odd/even goals | 21,037 | 50.0% | Weak |
| Exact total goals | 21,037 | 0.0% | Strong |
First half
| Market | Sample | Hit rate | Strength |
|---|---|---|---|
| 1st half Over/Under 0.5 | 6,837 | 71.8% | Weak |
| 1st half Over/Under 1.5 | 6,837 | 64.8% | Weak |
| Half-time result | 6,837 | 44.8% | Weak |
Corners
| Market | Sample | Hit rate | Strength |
|---|---|---|---|
| Corners Over/Under 8.5 | 6,707 | 62.8% | Weak |
| Corners Over/Under 10.5 | 6,707 | 59.0% | Weak |
| Corners Over/Under 9.5 | 6,707 | 54.9% | Weak |
Cards
| Market | Sample | Hit rate | Strength |
|---|---|---|---|
| Cards Over/Under 5.5 | 6,707 | 76.8% | Weak |
| Cards Over/Under 4.5 | 6,707 | 62.7% | Weak |
| Cards Over/Under 3.5 | 6,707 | 58.0% | Weak |
Teams
| Market | Sample | Hit rate | Strength |
|---|---|---|---|
| Away 3+ goals | 21,037 | 87.2% | Reliable |
| Home 3+ goals | 21,037 | 80.4% | Weak |
| Away clean sheet | 21,037 | 77.6% | Weak |
| Home to score | 21,037 | 77.6% | Weak |
| Home clean sheet | 21,037 | 69.1% | Weak |
| Away to score | 21,037 | 69.1% | Weak |
| Away 2+ goals | 21,037 | 66.9% | Weak |
| Home 2+ goals | 21,037 | 61.6% | Weak |
Specials
| Market | Sample | Hit rate | Strength |
|---|---|---|---|
| Away win to nil | 21,037 | 84.7% | Fair |
| Home win to nil | 21,037 | 76.7% | Weak |
| Winning margin | 21,037 | 28.4% | Weak |
Hit rate = how often the model's most-likely side was correct. For two-way markets (over/under, both-teams-to-score, clean sheets) the base rate is naturally high, so quality is judged by calibration, not hit rate — the strength label grades that for the markets where it's meaningful. Value ROI is shown only for the match result, the one market where we hold bookmaker odds; every other market is graded for accuracy and calibration, not price. Historical, out-of-sample — not a guarantee.
How to read this
Live record vs backtest. The live record is our actual published predictions, graded after the matches were played — it's the honest test. The backtest re-runs the model over past seasons using only pre-kickoff data, to validate the method over a much larger sample. We show both, including where they're unflattering.
Accuracy is how often the highest-probability outcome actually happened. RPS (Ranked Probability Score) grades the full probability forecast — lower is better — and we score the bookmaker's own de-vigged odds the same way as a benchmark (green means we matched or beat the market).
Value ROI simulates flat 10-unit stakes placed only where the model's edge over the bookmaker cleared the threshold, settled at the recorded odds; the bracketed number is how many bets that was.
A serious note: bookmakers set very efficient prices. A good model lands close to them and hovers near break-even on value bets — consistently beating the market is genuinely hard. Treat value signals as a starting point for your own judgement, not a guarantee. 18+, please gamble responsibly.