Every spread model and every flag, graded against the closing line on each final game — the hindsight-free record. Break-even is 52.4%. Backtests say expect ~50%; see the guide. 619 games graded so far.
ATS = record betting the side the model favored over the market (a win = that side covered the close). Win % is of decided bets; 52.4% is break-even at −110. Edge ≥ 2 = only games the model was ≥ 2 pts off the close. MAE = average miss between the predicted and actual final margin, in points — with the closing line's own MAE on the same games below it. Units = profit betting 1 unit on every graded game at −110 (a win pays 0.91, a loss costs 1). Lower is better; green = the model beat the market at raw prediction. (Beating the market on MAE is a different thing from beating it ATS.)
The closing line missed the final margin by 11.6 points on average across these 619 games — that's the bar.
| Model | ATS | Win % | Units | Edge ≥ 2 | MAE |
|---|---|---|---|---|---|
| SP+ model | 313–289–12 | 52.0% | −4.45u | 207–199 | 12.1 close 11.6 |
| SRS model | 123–154–8 | 44.4% | −42.18u | 86–115 | 13.3 close 11.8 |
| Yahn model | 320–284–12 | 53.0% | +6.91u | 204–195 | 12.1 close 11.7 |
| Flag | ATS | Win % | Units |
|---|---|---|---|
| bad_spot | 8–5 | 61.5% | +2.27u |
| letdown | 20–18 | 52.6% | +0.18u |
| lookahead | 34–30 | 53.1% | +0.91u |
| off_bye | 38–32–2 | 54.3% | +2.55u |
| revenge | 64–63–5 | 50.4% | −4.82u |
| rlm | 18–4–1 | 81.8% | +12.36u |
| short_week | 7–7 | 50.0% | −0.64u |
| steam | 40–22 | 64.5% | +14.36u |
| travel | 53–39–1 | 57.6% | +9.18u |
A model "wins" a game when the side it favored over the market went on to cover the closing number. Small samples early — one week is noise. Watch the trend across the season, not any single number.
the yahngorithm