Accuracy
We publish our record before we ask you to trust it. Two kinds of evidence live here: the historical backtest of our team model, and the live grading of every signal we ship (Closing Line Value, measured at kickoff).
Team model backtest (out of sample)
Walk-forward Elo with rest, travel, bye, and neutral-site adjustments. Parameters were fit on 1999-2015 only, then evaluated on 2016-2026 — 2766 games the model had never seen.
| Our model | Closing spread | |
|---|---|---|
| Straight-up accuracy | 63.9% | 66.5% |
| Brier score (lower is better) | 0.222 | 0.2112 |
| Log loss (lower is better) | 0.6342 | 0.6102 |
Mean absolute gap between our spread and the close: 2.645 points.
The honest part. When our number disagreed with the closing line by 1.5+ points, taking our side went 864–871–45 against the spread (49.8%). Breakeven at standard pricing is 52.4%. That is why the closing line is the anchor of every card, and why we do not publish "Model vs. Market" cards until a model clears the calibration gate on live data.
By season
| Season | Games | Model acc. | Market acc. | Model Brier | Market Brier | Spread gap |
|---|---|---|---|---|---|---|
| 2016 | 265 | 65.7% | 65.3% | 0.2167 | 0.2175 | 2.53 |
| 2017 | 267 | 64.8% | 69.7% | 0.2171 | 0.2041 | 2.67 |
| 2018 | 265 | 62.6% | 65.7% | 0.2213 | 0.2125 | 2.74 |
| 2019 | 266 | 63.9% | 64.3% | 0.2251 | 0.2136 | 2.44 |
| 2020 | 268 | 66.4% | 67.2% | 0.2166 | 0.2035 | 2.73 |
| 2021 | 284 | 60.2% | 62.3% | 0.2335 | 0.2176 | 2.78 |
| 2022 | 282 | 63.5% | 66.3% | 0.2225 | 0.21 | 2.77 |
| 2023 | 285 | 59.6% | 67.4% | 0.2311 | 0.2172 | 2.44 |
| 2024 | 285 | 68.8% | 70.5% | 0.2122 | 0.2039 | 2.92 |
| 2025 | 284 | 63.7% | 65.8% | 0.2227 | 0.2116 | 2.46 |
| 2026 | 15 | 66.7% | 73.3% | 0.2227 | 0.211 | 1.92 |
Fitted parameters: K = 20, home field = 60 Elo, offseason reversion = 40%, 22 Elo per point.
Live signal grading
Every signal is graded at kickoff against the closing line (CLV) and, once the game ends, against the result. Signals graded so far: 0. Grading begins with the first week of live pick'em and sportsbook lines.
What this means
Closing sportsbook lines remain the strongest single predictor of NFL games, and our own backtest says so. Our job is to add the situational information the line does not explain — rest, travel, referee crews, coaching tendencies, injuries, weather — and to explain disagreement across books, pick'em apps, and prediction markets in plain language. Not to sell you a lock.