ST Model Accuracy
Out-of-sample backtests of the ST Model's calibrated engine — projections vs. what actually happened, and vs. Vegas closing lines
Player projections
Beat the naive baseline by 16% on per-game accuracy · p10–p90 coverage 81% (2025 backtest, 7,247 preds)
Team margins vs Vegas
Model MAE 10.72 · in-season 10.11 vs Vegas 9.84 (2,127 games)
Player Accuracy
per-game rate vs naive · 7,247 predsskill 16%p10–p90 cov 81%
| Stat | N | Model/g | Base/g | Skill | Cov |
|---|---|---|---|---|---|
| Passing | |||||
| Pass Yards | 37 | 31.02 | 37.55 | +17% | 73% |
| Pass TDs | 37 | 0.3 | 0.47 | +36% | 78% |
| Pass INTs | 37 | 0.2 | 0.24 | +15% | 92% |
| Pass Att | 37 | 3.3 | 4.27 | +23% | 78% |
| Pass Cmp | 37 | 2.35 | 3.01 | +22% | 70% |
| Rushing | |||||
| Rush Yards | 129 | 9.49 | 10.95 | +13% | 82% |
| Rush TDs | 129 | 0.13 | 0.15 | +14% | 85% |
| Rush Att | 92 | 2.21 | 2.72 | +19% | 76% |
| Receiving | |||||
| Receptions | 335 | 0.67 | 0.75 | +12% | 79% |
| Rec Yards | 335 | 7.89 | 9.29 | +15% | 81% |
| Rec TDs | 335 | 0.1 | 0.12 | +21% | 85% |
| Targets | 335 | 0.92 | 1.01 | +9% | 81% |
| Defense | |||||
| Sacks | 654 | 0.08 | 0.1 | +18% | 80% |
| QB Hits | 654 | 0.14 | 0.16 | +11% | 79% |
| Tackles for Loss | 654 | 0.13 | 0.16 | +18% | 80% |
| Solo Tackles | 654 | 0.54 | 0.63 | +15% | 81% |
| Assist Tackles | 654 | 0.49 | 0.57 | +15% | 82% |
| Interceptions | 654 | 0.04 | 0.04 | +16% | 87% |
| Passes Defended | 654 | 0.13 | 0.15 | +10% | 74% |
| Forced Fumbles | 654 | 0.04 | 0.05 | +25% | 87% |
| Kicking | |||||
| FG Made | 28 | 0.41 | 0.66 | +38% | 79% |
| FG Att | 28 | 0.43 | 0.73 | +41% | 86% |
| XP Made | 28 | 0.53 | 0.84 | +37% | 82% |
| Punting | |||||
| Punts | 28 | 0.65 | 0.59 | -10% | 86% |
| Punt Yards | 28 | 31.65 | 30.02 | -5% | 86% |
Fantasy Board — Preseason Backtest vs Market
reconstruction · 10 seasons
| Position | Rank quality | ADP | Top-12 hits | ADP | Busts | ADP |
|---|---|---|---|---|---|---|
| Overall | 0.570 | 0.460 | 4.1 | 4.4 | 5.3 | 5.2 |
| QB | 0.505 | 0.365 | 7.4 | 7.1 | 1.3 | 1.4 |
| RB | 0.560 | 0.483 | 6.0 | 6.7 | 2.9 | 2.7 |
| WR | 0.577 | 0.482 | 5.7 | 5.9 | 3.8 | 3.8 |
| TE | 0.630 | 0.312 | 6.8 | 6.6 | 2.6 | 3.0 |
| Season | Overall | QB | RB | WR | TE |
|---|---|---|---|---|---|
| 2025 | 0.461 / 0.385 | 0.207 / 0.069 | 0.518 / 0.553 | 0.375 / 0.339 | 0.536 / 0.149 |
| 2024 | 0.517 / 0.450 | 0.523 / 0.377 | 0.619 / 0.582 | 0.503 / 0.430 | 0.500 / 0.559 |
| 2023 | 0.630 / 0.506 | 0.476 / 0.289 | 0.523 / 0.409 | 0.681 / 0.624 | 0.755 / 0.198 |
| 2022 | 0.613 / 0.550 | 0.741 / 0.453 | 0.641 / 0.532 | 0.776 / 0.581 | 0.612 / 0.350 |
| 2021 | 0.611 / 0.499 | 0.704 / 0.677 | 0.564 / 0.509 | 0.490 / 0.415 | 0.643 / 0.297 |
| 2020 | 0.547 / 0.470 | 0.622 / 0.514 | 0.506 / 0.512 | 0.589 / 0.507 | 0.681 / 0.454 |
| 2019 | 0.565 / 0.549 | 0.373 / 0.322 | 0.681 / 0.712 | 0.570 / 0.399 | 0.714 / 0.379 |
| 2018 | 0.585 / 0.447 | 0.414 / 0.370 | 0.428 / 0.343 | 0.718 / 0.629 | 0.509 / 0.228 |
| 2017 | 0.617 / 0.455 | 0.459 / 0.216 | 0.665 / 0.518 | 0.567 / 0.479 | 0.635 / 0.240 |
| 2016 | 0.550 / 0.289 | 0.529 / 0.366 | 0.453 / 0.157 | 0.498 / 0.414 | 0.716 / 0.270 |
Board vs the market's consensus ADP, each graded against realized PPR finishes. Seasons before 2026 are reconstructions; from 2026 each row grades the board as published.
Game Model — Out-of-Sample Backtest
backtest
| Season | Games | Brier | Base | Win% | Margin MAE | Total MAE | Vegas MAE | Beat% |
|---|---|---|---|---|---|---|---|---|
| 2025 | 272 | 0.2222 | 0.2486 | 66% | 10.15 | — | 9.72 | 51.3% |
| 2024 | 272 | 0.2017 | 0.2489 | 70% | 9.73 | — | 9.61 | 59.7% |
| 2023 | 272 | 0.2239 | 0.2470 | 67% | 10.25 | — | 9.90 | 50.4% |
| 2022 | 271 | 0.2208 | 0.2467 | 65% | 9.15 | — | 8.74 | 47.9% |
| 2021 | 272 | 0.2230 | 0.2498 | 62% | 10.90 | — | 10.78 | 51.9% |
| 2020 | 256 | 0.2228 | 0.2500 | 65% | 10.40 | — | 9.83 | 51.6% |
| 2019 | 256 | 0.2156 | 0.2498 | 66% | 10.37 | — | 10.21 | 52.8% |
| 2018 | 256 | 0.2080 | 0.2405 | 67% | 10.00 | — | 9.98 | 51.4% |
| Overall | 2,127 | 0.2173 | 0.2485 | 66% | 10.11 | — | 9.84 | 52.1% |
The model's pre-game win probability is scored by Brier against every completed game's outcome; a Brier below the home-field Basemeans it adds signal beyond "home teams win." Margin/Total MAE measure the projected score. Vegas MAE / Beat% benchmark the projected margin against the closing line — the model never sees the line, it's stored only to grade honestly.