Soccer Season Ledger
Every match the goal model has called this season, scored against the result — 485 of them across 6 leagues. Wins and losses both.
Every graded market
485 matchesUnits only ride the matches we froze a pre-kick price for, which is why each cell names its own priced count. The books quote the three-way result far more often than both-teams-to-score, so those two records cover different sets of matches — adding them would be adding different things.
La Liga
39 graded · -12.38uPremier League
30 graded · +14.32uSerie A
28 graded · +14.59uLigue 1
27 graded · -18.94uBundesliga
18 graded · +1.92uHow to read this
The record is the product. Every projection freezes at kickoff and gets scored against the actual match. Nothing is re-fitted afterwards, and the ratings behind a match only use matches played before it, so a good-looking number here can't have come from hindsight.
The model does not beat the three-way market, and we've measured that rather than assumed it. Walk-forward on 809 graded MLS matches, its Brier score sat behind the closing line, and the correlation between our disagreement with the market and the market's own error was about zero. That's what noise looks like. A league showing a positive return over thirty-odd matches is showing you a sample size.
Thin means thin. Soccer is low-scoring and a deflection settles games. Any band under 20 graded matches is marked rather than given a rate, because a percentage printed on eight matches invites a conclusion the number can't support.
Units and win-loss cover different matches. A pick only stakes a unit where we held a pre-kick price. The books quote the three-way result far more often than both-teams-to-score, so each cell carries its own priced count — and adding records across markets would be adding different things.
Common questions
How is the soccer model graded?
Every projection freezes at kickoff and gets scored against the actual result. There's no re-fitting after the fact and no lookahead — the ratings behind a match only ever use matches played before it. Wins and losses both go on the board, and the pre-launch part of each season is reconstructed the same walk-forward way.
What does the units number mean?
One unit staked per graded pick, settled at the price we froze before kickoff. It only rides matches where we actually held a pre-kick price, which is why every cell names its own priced count — the books quote the three-way result far more often than both-teams-to-score, so those records cover different sets of matches.
Does the model beat the market?
On the three-way result, no — and we've tested it rather than assumed it. Walk-forward on 809 graded MLS matches the model's Brier score sat behind the closing market, and the correlation between our disagreement with the line and the market's own error was about zero, which is what noise looks like. Some leagues show a positive return over a few dozen matches; that's a sample size, not an edge.
What is a model lean?
A match where our number clears the book's implied number by at least four points. It's the narrower, more opinionated band. It is not a recommendation, and on the longest sample we have it has not been a profitable filter.
Why do some leagues have so few graded matches?
MLS runs inside a calendar year and is deep into its season; the European leagues kicked off in August, so they carry a few dozen matches each. A record over thirty matches in a sport this low-scoring is close to meaningless, which is why small bands are marked as thin rather than dressed up as a rate.
Are these betting picks?
No. Every number here is a research signal from a statistical model, published with its graded record so you can judge it yourself. Not betting advice, and no outcome is guaranteed. 21+, please play responsibly (1-800-GAMBLER).
2026 season to date. Research signals from a statistical model, not betting advice — no outcome is guaranteed. 21+, play responsibly (1-800-GAMBLER).