Public model audit

SHOW THE
RECEIPTS.

Frozen before kickoff

Every grade uses the latest published 4M projection that existed before the game began. Final scores can update the evaluation—but never the original forecast.

Read the methodology →

Active champion audit

4,478 truly out-of-sample games.

For each validation season, 4M trained only on earlier seasons and then predicted the next year. The active 4M-WALK-FORWARD-1.0 champion separately learns margin, total and win probability; challengers stay private unless they improve held-out accuracy.

Validation margin MAE13.07 pts2020–2025 forward tests

Market benchmark12.22 ptsArchived pregame prices

Calibrated result12.38 ptsFootball + learned shrinkage

Historical Brier0.190Log loss pending

Official betting record

Model opinions and published wagers are graded separately.

Only timestamped positions from The Card enter this scorecard. Results use the exact posted number; closing-line value uses the final stored price from the same sportsbook.

Open the public ledger →

Official record0–00 still open

Net units0.00u0.0% ROI

Average CLVPendingSame-book closing price

Positive CLVPending0 graded plays

Graded games448Eligible frozen forecasts

Straight-up record364–8481.3% projected winners

Spread MAE14.77 ptsLower is better

Market MAE13.73 ptsStored pregame line benchmark

Calibrated MAE14.69 ptsFootball + uncertainty blend

Total MAE12.76 ptsProjected vs. actual total

Winner accuracy81.3%Straight-up calls

Brier score0.14Probability accuracy

Log loss0.438Penalizes overconfidence

Kickoff freezes0Market, weather and players locked

ATS decisions44–69–138.9% excluding pushes

Total decisions48–65–042.5% excluding pushes

By week

Forecast error over time

WeekGamesSpread MAETotal MAEWinner

19917.3612.5188.9%

28615.6311.4187.2%

37512.2613.0686.7%

47114.0212.4277.5%

55915.1314.7172.9%

65812.8813.2165.5%

Calibration

Does confidence match reality?

50–59%107 forecasts61.7% actual

60–69%98 forecasts74.5% actual

70–79%71 forecasts83.1% actual

80–89%74 forecasts97.3% actual

90–100%62 forecasts100% actual

By edge size

Where disagreement earns trust.

Under 2 points18 decisions7–11–038.9%

2–3.9 points24 decisions10–14–041.7%

4–5.9 points10 decisions1–9–010%

6+ points62 decisions26–35–142.6%

THE 4M AUDIT STANDARD

Ratings are judged on what they knew before kickoff. Picks are judged at the posted number. Market close is a benchmark, never a retroactive replacement.