| GAME | KICKOFF (ET) | MKT SPREAD | MODEL SPREAD | SPREAD EDGE | MKT TOTAL | MODEL TOTAL | TOTAL EDGE | ML (H/A) | MODEL WIN % | MKT WIN % | WIN EDGE | BEST EV | CONFIDENCE | CLV TARGET | STATUS |
|---|
Spreads are the home team's line (negative = home favoured). Edges are pure model minus market; EV and status use the market-aware (calibrated) probability. Confidence reflects rating uncertainty only. Click a row for the explanation.
Correlated same-game parlays: current slate
In lopsided games (spread larger than 40% of the total) a favourite that covers usually pushes the game over, and an underdog that covers usually keeps it under. Since 2015 these pairs hit about 31% of the time versus about 25% if the two bets were independent, and they did so in every season. A 31% combo breaks even at +222. Only worth it if your book pays about +225 or better. Most books price same-game parlays below that (then skip), and some block spread + total on the same game. The model picks which of the two combos to prefer, but that choice added almost nothing historically (31.8% vs 30.4%). Size small: long losing runs are normal at a 31% hit rate.
| Game | Kickoff (ET) | Favourite | Total | Spread/total | Model-preferred combo | Other correlated combo | Model: fav margin / total |
|---|---|---|---|---|---|---|---|
| Northwestern @ Indiana | Fri 09/25 08:00 PM | Indiana −20.5 | 50.5 | 0.41 | Indiana −20.5 & Over 50.5 | Northwestern +20.5 & Under 50.5 | +20.7 / 49.6 |
| Illinois @ Ohio State | Sat 09/26 12:00 PM | Ohio State −26.5 | 53.5 | 0.50 | Ohio State −26.5 & Over 53.5 | Illinois +26.5 & Under 53.5 | +26.1 / 55.8 |
| Sam Houston @ Texas Tech | Sat 09/26 12:00 PM | Texas Tech −34.5 | 55.5 | 0.62 | Sam Houston +34.5 & Under 55.5 | Texas Tech −34.5 & Over 55.5 | +26.6 / 52.7 |
| Notre Dame @ Purdue | Sat 09/26 02:00 PM | Notre Dame −27.5 | 57.5 | 0.48 | Notre Dame −27.5 & Over 57.5 | Purdue +27.5 & Under 57.5 | +31.8 / 64.3 |
| Central Michigan @ Miami | Sat 09/26 06:30 PM | Miami −41.5 | 54.5 | 0.76 | Central Michigan +41.5 & Under 54.5 | Miami −41.5 & Over 54.5 | +39.3 / 51.8 |
| Missouri State @ SMU | Sat 09/26 09:00 PM | SMU −34.5 | 59.5 | 0.58 | Missouri State +34.5 & Under 59.5 | SMU −34.5 & Over 59.5 | +30.1 / 58.4 |
| Notre Dame @ North Carolina | Sat 10/03 TBD | Notre Dame −20.5 | 49.5 | 0.41 | Notre Dame −20.5 & Over 49.5 | North Carolina +20.5 & Under 49.5 | +20.0 / 48.5 |
| Vanderbilt @ Georgia | Sat 10/03 TBD | Georgia −23.5 | 57.5 | 0.41 | Vanderbilt +23.5 & Under 57.5 | Georgia −23.5 & Over 57.5 | +20.8 / 61.0 |
History: same-game combos when spread/total is above 0.4 (2015–2025, all FBS games)
| Combo | Games | Hit rate | If independent | ROI at +264 |
|---|---|---|---|---|
| fav+over | 1081 | 31.0% | 25.2% | +12.9% |
| fav+under | 1081 | 18.3% | 24.1% | -33.2% |
| dog+over | 1081 | 20.1% | 25.9% | -26.8% |
| dog+under | 1081 | 30.6% | 24.8% | +11.6% |
Backtest: parlays built from the model's picks (out-of-sample 2018–2025, flat stakes, multiplied −110 odds)
| Legs taken from | Straight | 2-leg | 3-leg | 4-leg |
|---|---|---|---|---|
| spread@open | +1.5% (2262) | +3.6% (1043) | +5.1% (646) | -2.7% (456) |
| total@open | +1.4% (2016) | +1.9% (929) | +2.6% (567) | +33.0% (410) |
| spread+total@open | +1.5% (4278) | +5.1% (1459) | +15.4% (918) | +11.4% (646) |
| spread@close | -5.1% (2168) | -12.9% (972) | -15.6% (583) | -23.5% (403) |
| total@close | -2.1% (2139) | -6.0% (933) | -4.5% (579) | +5.6% (394) |
| spread+total@close (Saturday-style) | -3.6% (4307) | -3.9% (1608) | -5.3% (996) | +12.2% (697) |
Parlay ROI is roughly (1 + straight ROI) to the power of the number of legs, minus 1, with much more variance: parlays magnify an edge, they never create one. "Close" rows are how you would bet on game day, and they are negative. The few large positive cells are what chance produces across 24 cells; the straight bets underneath them are not statistically significant. No multi-game parlay is recommended.
How to read this page
- Spreads are the home team's line: −7 means the home team is favoured by 7. Lines are DraftKings via ESPN, captured every 3 hours.
- Model spread / total is what the power-rating model thinks is fair. Edge = model minus market. Big edges early in the season are usually model error, not opportunity.
- EV and status use a calibrated probability that shrinks the model toward the market, because historically most disagreement with the line is noise. WATCH means the calibrated EV is at least zero: a research position, not a pick.
- Confidence reflects how settled the team ratings are (low in September, higher by November).
- Click any game for the exact factors behind the model's number and the line's movement this week.
What the backtest says
Out-of-sample 2018–2025: the model's side wins about 53% when bet at the opening line (the line moves toward it more often than not), but that is not significantly better than the 52.4% needed at −110. At the closing line, meaning game-day prices, it loses (−5% on spreads). Parlays of these picks magnify the result in either direction; they do not create an edge.
Forward paper ledger (2026)
54 paper positions recorded; none settled yet.
Out-of-sample backtest vs pre-registered gates
| strategy | holdout n | W-L-P | holdout ROI | pooled OOS ROI | pooled p | holdout CLV | all gates |
|---|---|---|---|---|---|---|---|
| moneyline @ close (EV≥4%) | 444 | 120-324-0 | -4.7% | -0.6% | 0.55 | FAIL | |
| spread @ close (EV≥0%) | 710 | 349-357-4 | -5.2% | -5.1% | 0.99 | FAIL | |
| spread @ open (EV≥0%) | 987 | 511-470-6 | -0.6% | +1.5% | 0.22 | +0.46 | FAIL |
| total @ close (EV≥0%) | 393 | 204-186-3 | +0.7% | -2.1% | 0.84 | FAIL | |
| total @ open (EV≥0%) | 777 | 417-357-3 | +2.8% | +1.4% | 0.26 | +0.92 | FAIL |
Sealed holdout = 2024–2025, evaluated once after freezing. Pooled = development 2018–2023 + holdout.