Models & Forecasts · The +EV board
The record, graded against the open
Every side whose calibrated cover probability clears the 53% bar at the opening line is a play. Over 2023–2025 that is 164–111–6, 59.6% ATS, +41.9 units at one unit a play. Probabilities are walk-forward: fit only on earlier seasons, never on the games being graded. Plays publish before kickoff with a timestamp, and the same page shows what the picks are worth by the close, which is much less.
Regenerated 2026-09-08 · −110 assumed throughout · not betting advice
- Every game, flat (no bar)
- 53.8%
- Break-even at −110
- 52.4%
A play is a side whose calibrated cover probability clears the bar at the opening line.
- Flat-betting every game
- +24.2u · +2.6%
ROI is return on total risk (1.1u per play). Pushes return the stake.
- Average move toward the pick
- +1.5 pts
- Lines that moved at all
- 84%
The market tends to follow the plays between open and close — the cleanest sign the edge is real, not noise.
What the identical picks would have returned at the closing number. The gap to the open is the price of waiting.
The control: the same model and bar, but anchored on the closing line. The ranking signal is largely gone by then — which is why the board publishes at the open.
- Our projection
- 9.97
- Opening line
- 10.06
- Closing line
- 9.77
Straight-up winner 65.6% over 853 games. The margin matches the market; the edge is in the calibrated probability, not the point estimate.
Every week of the walk-forward history, in order. Flat weeks are bye-heavy or play-free. The blue line is the record the board is built on; the orange line is the same bar applied to picks made at the close.
Season by season
Left of the gap: the model. Middle: plays at the bar, anchored and graded at the opening line, with the same plays re-graded at the close and how often the close moved our way. Right, greyed: the control that makes its picks at the close. Seasons marked pre are walk-forward but predate the window the bar was set on.
| Season | Games | SU | MAE model · open · close | Plays | W–L–P | ATS | Units | ROI | At close | CLV | Close plays | ATS | Units | ||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 2021pre | 284 | 65.0% | 10.85 · 10.78 · 10.70 | 189 | 101–88–5 | 53.4% | +4.2u | +2.0% | 52.1% | 70% | 175 | 50.9% | -5.6u | ||
| 2022pre | 283 | 64.4% | 9.14 · 9.02 · 8.80 | 159 | 75–84–4 | 47.2% | -17.4u | -9.9% | 46.3% | 56% | 171 | 49.7% | -9.6u | ||
| 2023 | 285 | 64.6% | 10.26 · 10.38 · 9.98 | 50 | 36–14–1 | 72.0% | +20.6u | +37.5% | 64.6% | 84% | 29 | 48.3% | -2.5u | ||
| 2024 | 284 | 69.0% | 9.83 · 9.90 · 9.67 | 106 | 57–49–2 | 53.8% | +3.1u | +2.7% | 55.2% | 71% | 35 | 54.3% | +1.4u | ||
| 2025 | 284 | 63.2% | 9.82 · 9.89 · 9.66 | 119 | 71–48–3 | 59.7% | +18.2u | +13.9% | 53.3% | 78% | 33 | 57.6% | +3.6u |
Where the bar sits
2023–2025, plays at the open. Higher bars win more often on fewer plays; the published bar is the one the board is built on.
| Bar | Plays | ATS | Units | ROI |
|---|---|---|---|---|
| Every game | 839 | 53.8% | +24.2u | +2.6% |
| p ≥ 0.52 | 406 | 56.9% | +38.5u | +8.6% |
| p ≥ 0.5238 | 354 | 57.6% | +39.0u | +10.0% |
| p ≥ 0.53the bar | 275 | 59.6% | +41.9u | +13.9% |
| p ≥ 0.54 | 183 | 62.3% | +38.1u | +18.9% |
| p ≥ 0.55 | 116 | 58.6% | +15.2u | +11.9% |
2026, live
Every play as first published, timestamped, graded after the game. A week’s archive is frozen at first publish, so a later price refresh can never move a graded line.
- NYG +3.0vs DAL-110p 58.0% · +10.7%pending
- MIN -1.5vs GB-108p 54.8% · +5.5%pending
- TEN -1.5vs NYJ-113p 55.3% · +4.2%pending
- MIA +3.5at LV-110p 54.4% · +3.9%pending
- NO +7.0at DET-113p 53.9% · +1.6%pending