Scores
Dev

Best Bets — model edge vs the market

Our gradient-boosted model (Elo + opponent-adjusted EPA + talent + market signal, walk-forward validated on 2019-2024) flags games where its projection diverges from the market line. The honest reality: the edge is small and concentrated in soft markets — not a license to print money. We judge ourselves by Closing Line Value, not hot streaks.

Verified performance · 2019-2024 out-of-sample, walk-forward, PRE-REGISTERED threshold, leak-free

CLV vs opener ★
+0.28 pts
53.5% beat the close · n=2056
ATS vs open (realizable)
51.5%
1045-983 · p=0.79 — not significant
ATS vs close
50.2%
1316-1306 · no edge
Margin RMSE
15.56
market 15.46 — at parity

The honest verdict: With a pre-registered threshold, no closing-line leakage, and bets graded at the realizable opening price, the model shows NO statistically significant ATS edge (~50-51.5%). The only real signal is a small positive CLV — the close drifts toward our side slightly more than half the time. We retract the earlier 53.9% (it was test-set-tuned and used the closing line it bet into). This is the honest bar. The only real signal is the small positive CLV above. Even perfect knowledge of a starting QB being out yields ~50% ATS — the market prices injuries efficiently. Anything claiming 58%+ is leakage.

2024 staking simquarter-Kelly, -110, key-number-aware cover probRecord 100-84ROI -15.4%Max drawdown 34.6%— even with correct staking, a sub-break-even season loses. This is why CLV, not ROI, is the KPI.

Closing Line Value · the honest skill metric

2021-2024 opens, walk-forward

Bet the OPENING line where the model disagrees (pre-registered threshold).

Mean CLV
+0.442 pts
the close moves to our number
Beat the close
56%
95% CI [53.3, 58.7] · excludes 50% ✓
Home picks
54.3%
positive CLV
Away picks
57.9%
positive · n=1339

Why this is real: Positive CLV on BOTH sides at a balanced home/away share = genuine selection, not a drift-ride. CLV (closing-line value) is the honest skill metric. After we bet the open, the close moves toward our side ~56% of the time (95% CI excludes 50%). This does NOT beat the closing spread head-to-head (~51.5% ATS, not significant) — an efficient-market result.

Calibrated probabilities · key numbers

2019-2024 out-of-sample · n=4,444

A joint score simulator now prices spread, push, total, moneyline, team totals and alternate lines off one coherent distribution. The headline: it lands on the key margins (3, 7, 10, 14) about as often as games actually do — where our old single-Normal model under-counted them ~3×, mispricing pushes, key-number teasers and alternate lines.

Final margin lands onOld modelNew simulatorActual frequency
±33.76%9.64%10.46%
±73.59%8.82%8.57%
±103.39%4.67%4.82%
±143.04%4.6%4.43%

This is calibration/correctness — it makes our push, teaser and alternate-line numbers trustworthy. It is not an ATS edge (the simulator is spread-neutral by construction).

2024 track record · spreads (edge≥2)

167-157 (51.5% ATS)
GamePickModelMarketEdgeFinal
Kent State @ TennesseeKent State +48.8-39.5-48.89.40-71L
Kent State @ Miami (OH)Kent State +30.7-22-30.78.77-34W
Washington State @ San Diego StateSan Diego State +17.2+8.9+17.28.329-26W
Notre Dame @ GeorgiaGeorgia +1.2-7+1.28.223-10L
Ohio State @ OregonOregon +2.5-5.4+2.57.941-21L
Missouri @ Mississippi StateMissouri -9.5+16.8+9.57.339-20W
South Alabama @ LSUSouth Alabama +20.8-13.6-20.87.310-42L
Texas State @ UL MonroeTexas State -9.5+16.6+9.57.138-17W
Fresno State @ New MexicoFresno State -12.3+18.9+12.36.638-21W
Tulsa @ TempleTulsa +3.0+3.5-36.510-20L
Liberty @ Middle TennesseeLiberty -12.7+19+12.76.337-17W
Marshall @ Virginia TechMarshall +20.7-14.3-20.76.314-31W
Buffalo @ LibertyLiberty +4.0-2.2+46.226-7L
UNLV @ Utah StateUNLV -18.7+24.9+18.76.250-34L
Tennessee @ VanderbiltTennessee -9.8+16+9.86.236-23W
Rice @ ArmyArmy -7.3-13.4-7.36.114-37W
Akron @ Ohio StateAkron +49.5-43.6-49.55.96-52W
Michigan @ Ohio StateMichigan +20.0-14.2-205.813-10W
East Carolina @ North TexasEast Carolina +2.5+3.2-2.55.740-28W
Pittsburgh @ CincinnatiPittsburgh +0.5+5.1-0.55.628-27W

2024 track record · Group-of-5 totals (edge≥4)

136-126 (51.9% O/U)
GamePickModelEdgeFinal
Navy @ ArmyOVER 39.26626.931-13 (44)W
Navy @ Air ForceOVER 37.25921.834-7 (41)W
Air Force @ ArmyOVER 38.360.121.83-20 (23)L
Troy @ LouisianaOVER 51.572.821.330-51 (81)W
Nevada @ Hawai'iOVER 41.862.420.513-34 (47)W
Army @ Notre DameOVER 44.564.620.114-49 (63)W
Oregon State @ NevadaOVER 46.566.119.637-42 (79)W
Troy @ IowaOVER 39.257.618.521-38 (59)W
Utah @ Utah StateOVER 44.362.518.238-21 (59)W
North Texas @ UTSAUNDER 72.354.218.127-48 (75)L
Massachusetts @ Northern IllinoisOVER 40.856.916.120-34 (54)W
Miami (OH) @ Bowling GreenOVER 38.354.315.928-12 (40)W
Northern Illinois @ Bowling GreenOVER 46.561.815.317-7 (24)L
Tulane @ ArmyOVER 46.361.615.314-35 (49)W
Air Force @ WyomingOVER 33.348.615.219-31 (50)W

Live in-season this page surfaces upcoming-game edges vs current odds; in the offseason it shows the model's verified out-of-sample track record. Backtested model estimates for informational purposes only. Edges are small and concentrated in soft markets. Not betting advice. 21+. If you or someone you know has a gambling problem, call 1-800-GAMBLER.