Straight-up accuracy, ATS cover rate and margin error, every season since 2020, graded against the same market it's compared to. The market beats our margin error every season below; that's the honest trade a prediction model makes for calling a winner instead of pricing a number.
2020–2025 is a walk-forward backtest: each historical game was scored using only data available before its kickoff, no hindsight. The 2026 row is live, timestamped before kickoff every week since Week 1, kept separate below.
| SEASON | SU W-L-T | SU % | ATS W-L | ATS % | MODEL MAE | MARKET MAE | EDGE W-L | EDGE UNITS |
|---|---|---|---|---|---|---|---|---|
| 2020 | 172-83-1 | 67.5% | 116-140 | 45.3% | 10.04 | 9.83 | 17-17 | +4.52u |
| 2021 | 170-101-1 | 62.7% | 138-130-4 | 51.5% | 11.34 | 10.78 | 21-21 | +8.55u |
| 2022 | 178-91-2 | 66.2% | 125-136-10 | 47.9% | 8.89 | 8.74 | 25-22 | +15.87u |
| 2023 | 176-96 | 64.7% | 140-118-14 | 54.3% | 10.38 | 9.90 | 15-17 | +2.85u |
| 2024 | 189-83 | 69.5% | 150-118-4 | 56.0% | 9.93 | 9.61 | 24-26 | +2.80u |
| 2025 | 185-86-1 | 68.3% | 145-126-1 | 53.5% | 9.96 | 9.72 | 19-15 | +8.22u |
| 2026LIVE | 18-14 | 56.3% | 12-19-1 | 38.7% | 12.29 | 11.63 | — | — |
ATS grades the published pick against the closing spread (a push is excluded from both sides). MAE is the mean absolute error against the final margin, model vs. the market's own spread treated as a point prediction; lower is better and the market wins every season above. Edge W-L and units price only the games the model flags as bets at a real nflverse closing price (not every season is individually significant at this sample size; see the Edge).