Track record
Every bet the model rated 4 or 5, settled and unedited. Nothing is removed after the fact.
Bets rated 4 or 5, one per side, at the Consensus price — the average across the books we cover. Thu 1 Oct – Sat 24 Oct. Only the rating rule in use since Thu 1 Oct.
How these are priced and staked. Each bet is graded at the last Consensus price before the game — close to the closing number, not the price on the board when the bet appeared — and staked at the size the board recommended: full Kelly on a 10-unit bankroll. At the first price a reader could have taken, the same stakes returned +2.07%. At one unit a bet instead, they returned +0.21%.
The settled bets
The 25 most recently graded, at the Consensus price. Every graded bet in this scope is in the download. Download this scope as CSV
| Rating | Day | Match | Bet | Price | Result |
|---|---|---|---|---|---|
| Sat 24 Oct | Lens v Paris FCSoccer | Pablo Pagis Pablo Pagis1+ Shots on Target | −150 | lost | |
| Sat 24 Oct | CF Montréal v Nashville SCSoccer | Elias Saad Elias SaadPlayer Shots on Target | +100 | won | |
| Sat 24 Oct | CF Montréal v Nashville SCSoccer | Brian Schwake Brian SchwakeGoalkeeper Saves | +100 | lost | |
| Sat 24 Oct | Anaheim Ducks at Philadelphia FlyersNHL | Drew Helleson U 1.5Player Shots on Goal | −265 | won | |
| Sat 24 Oct | Anaheim Ducks at Philadelphia FlyersNHL | Jamie Drysdale U 0.5Player Assists | −230 | lost | |
| Sat 24 Oct | Valencia v VillarrealSoccer | Arnaut Danjuma Arnaut DanjumaPlayer Shots | +100 | won | |
| Sat 24 Oct | Valencia v VillarrealSoccer | Stole Dimitrievski Stole DimitrievskiGoalkeeper Saves | +100 | won | |
| Sat 24 Oct | Chelsea v TottenhamSoccer | Cole Palmer Cole PalmerPlayer Shots | +100 | won | |
| Sat 24 Oct | Chelsea v TottenhamSoccer | Morgan Rogers Morgan RogersPlayer Shots | +100 | won | |
| Sat 24 Oct | Chelsea v TottenhamSoccer | Antonín Kinský Antonín KinskýGoalkeeper Saves | +100 | lost | |
| Sat 24 Oct | Chelsea v TottenhamSoccer | James Maddison James Maddison2+ Shots | −121 | lost | |
| Sat 24 Oct | Chelsea v TottenhamSoccer | Mohammed Kudus Mohammed Kudus1+ Shots on Target | −109 | won | |
| Sat 24 Oct | Chelsea v TottenhamSoccer | Mathys Tel Mathys Tel1+ Shots on Target | −108 | won | |
| Sat 24 Oct | Chelsea v TottenhamSoccer | Cole Palmer Cole PalmerPlayer Goals | −225 | won | |
| Sat 24 Oct | Chelsea v TottenhamSoccer | Morgan Rogers Morgan RogersPlayer Goals | −254 | won | |
| Sat 24 Oct | Dortmund v Eintracht FrankfurtSoccer | Felix Nmecha Felix NmechaPlayer Shots | +100 | lost | |
| Sat 24 Oct | Dortmund v Eintracht FrankfurtSoccer | Gregor Kobel Gregor KobelGoalkeeper Saves | +100 | lost | |
| Sat 24 Oct | Napoli v AS RomaSoccer | Kevin De Bruyne Kevin De Bruyne1+ Shots on Target | −118 | lost | |
| Sat 24 Oct | Napoli v AS RomaSoccer | Matías Soulé Matías Soulé1+ Shots on Target | −124 | won | |
| Sat 24 Oct | Napoli v AS RomaSoccer | Noa Lang Noa Lang1+ Shots on Target | −111 | won | |
| Sat 24 Oct | Napoli v AS RomaSoccer | Donyell Malen Donyell Malen2+ Shots on Target | +113 | lost | |
| Sat 24 Oct | Coventry v FulhamSoccer | Oscar Bobb Oscar BobbPlayer Shots on Target | −245 | lost | |
| Sat 24 Oct | Arsenal v EvertonSoccer | Martin Odegaard Martin OdegaardPlayer Shots on Target | −153 | lost | |
| Sat 24 Oct | Coventry v FulhamSoccer | Alex Iwobi Alex IwobiPlayer Shots | +100 | won | |
| Sat 24 Oct | Arsenal v EvertonSoccer | Thierno Barry Thierno BarryPlayer Shots on Target | −121 | won |
What the model claimed, and what happened
Sat 24 Oct · 173,600 settled predictionsClose to honest on average — 36.1% claimed against 37.6% observed. That average hides the finding below it.
The further we depart from the market, the more we overstate
Every prediction, grouped by the rating the model gave it. Low ratings track the price closely and are calibrated; the strongest depart furthest, and that is where the claim and the outcome come apart.
Why a rating is capped rather than trusted at the top of its range, and why this site publishes no return by rating: the ordering of these buckets is stable, their levels are not.
Calibration, market by market
Claimed probability along the bottom, the fitted correction up the side; the faint diagonal is perfect calibration. 23 markets carry a fitted curve — the rest are measured but not yet fitted, and are in the table below rather than dropped.
Where we are worst
Every market we publish a board for, sorted by skill — the Brier score against a constant guess at the same base rate. Below zero the model lost to that guess: 20 of 88 markets. Nothing is removed and nothing is sorted flatteringly.
27 sit on fewer than 500 settled predictions. A skill score swings hard on a small sample, so those rows follow the rest rather than leading them — reported, not ranked.
When the model changed
Every day the blend’s source weights were rewritten. A record is only meaningful against a model that was not quietly swapped underneath it.