NCAA FOOTBALL · MODEL PERFORMANCE

College Football Model Performance & ATS Record

The whole record, one page.

Both halves of the record in one place: every graded pick with its edge and result, and every projection measured against the Vegas line on the same games. Locked before kickoff, graded after, nothing removed.

✓ WALK-FORWARD VALIDATED2025 walk-forward · 2026 logged before kickoff

493

Games priced, 2026

52.8%

Spreads win rate

Oct 28, 2026

Graded through

SEASON

PREMIUM ATSedge ≥ 6 pts

52.8%

67W – 60L · 127 graded · 95% CI 44.1–61.2%

AGAINST THE SPREAD · PREMIUM · EDGE ≥ 6 PTS

+0.8%

ROI, 1u at −110

+1.0u

Units won

ALL GRADED ATSevery decided pick

51.5%

222W – 209L · 431 graded · 95% CI 46.8–56.2%

AGAINST THE SPREAD · ALL GRADED

-1.6%

ROI, 1u at −110

-7.0u

Units lost

LINE QUALITY438 paired games

44.5%

Share of games BBMI landed closer than Vegas

computed from 438 logged games

12.20

BBMI margin MAE

11.35

Market margin MAE

Calibration by edge

1,232 priced · 431 decided

Pick rules and edge bands

The tier each pick actually fired under, from the published criterion. The AS POSTED rows are the record and tie to the product cards above; they are never recomputed from today’s thresholds. One game sits in the premium cut but outside the all-graded cut: its published line equals the market’s, so the all-graded cut excludes it while its edge keeps it premium. That is why the two totals differ by 1.

EDGE (POINTS)	TIER AS POSTED	N	WIN %	ROI
AS POSTED · THE TIER EACH PICK ACTUALLY FIRED UNDER
Premium · edge ≥ 6, spread 14–23	Premium	31	45.2%	-13.7%
Premium ★ · edge ≥ 6, spread ≤ 14	Premium ★	96	55.2%	+5.4%
All picks as posted	TOTAL	127	52.8%	+0.8%
BY EDGE INTERVAL · DESCRIPTIVE, SAME PICKS
0–2	No pick	99	54.5%	+4.2%
2–4	No pick	107	44.9%	-14.3%
4–5	No pick	40	57.5%	+9.8%
5–6	Marginal	43	46.5%	-11.2%
6–10 · spread ≤ 14	Premium ★	66	57.6%	+10.0%
6–10 · spread 14–23	Premium	19	47.4%	-9.5%
10–18 · spread ≤ 14	Premium ★	26	50.0%	-4.5%
10–18 · spread 14–23	Premium	12	41.7%	-20.4%
18+ · spread ≤ 14	Premium ★	4	50.0%	-4.5%
Edge ≥ 6, spread wider than ±23	Marginal	15	66.7%	+27.3%
All decided	TOTAL	431	51.5%	-1.6%

How posted picks are grouped

The interval rows are descriptive, not the record: the same picks cut at today’s thresholds, so a pick can appear in a band that did not exist when it was posted. They will not add up to the figure at the top of the page. Read them for shape.

Season to date

2026

Win rate versus break-even

Cumulative win rate after each week, against the break-even this product is priced at. The curve is the running rate, not each week's own, so a late swing moves it less than an early one.

8 weeksbreak-even 52.4%now 52.8%

BY WEEK ▼	PICKS	WIN %	ROI	CUMULATIVE
10/20 ’26	10	60.0%	+14.6%	52.8%
10/16 ’26	20	35.0%	-33.2%	52.1%
10/7 ’26	24	50.0%	-4.5%	55.7%
10/1 ’26	26	57.7%	+10.2%	57.5%
9/25 ’26	9	66.7%	+27.3%	57.4%
9/19 ’26	14	57.1%	+9.1%	55.3%
9/12 ’26	16	37.5%	-28.4%	54.2%
9/4 ’26	8	87.5%	+67.1%	87.5%

BBMI vs Vegas

438 games

Model versus market line quality

No picks here: line quality, head to head, on the same games logged below. Not a betting record, because a closer forecast is not a won bet.

CLOSER SPLIT5 ties

BBMI44.5%

VEGAS54.3%

MARGIN MAEVegas by 0.85 pts

BBMI12.20

VEGAS11.35

How projection error is measured

Lower MAE is the better projection. Computed from 438 logged games, not read from a published aggregate: this sport publishes no closer split.

Published figures across priced games

Published, across every priced game. The rows above are computed from the 438 games logged below; these are the site’s own aggregates over 1,234 games, read rather than recomputed, with the verdict the Record card reached.

PROJECTED WINNER ACCURACYVegas by 3.2 pts

Share of games each projection’s implied winner was right

BBMI70.7%

VEGAS73.9%

BRIER SCOREVegas by 0.022

Probability accuracy: rewards honest odds, not bold ones

BBMI0.195

VEGAS0.173

MARGIN MAE, PTSVegas by 1.53 pts

Projected home margin vs actual margin, in pts

BBMI13.19

VEGAS11.67

GAME-TOTAL MAE, PTSVegas by 0.47 pts

Projected total vs actual combined score, in pts

BBMI13.10

VEGAS12.64

Where the lines diverge

493 games

Under 3 pts36.7%

3 to 7 pts41.2%

7 to 14 pts18.9%

14+ pts3.2%

How far the projections differ

How far apart the two lines were. This needs no result, so it counts every game carrying both lines (493), which is wider than the 438 graded games above.

Calibration

Brier 0.195 · Published · full 1,234-game sample

Predicted versus actual win rates

A calibrated 60% should win about 60% of the time. Predicted win probability against the actual win rate, over every graded game rather than only the ones a pick fired on. This panel is the only block on the page that does NOT follow the season chip: the scorecard publishes one cohort and no per-season cut, so showing it under a 2026 heading would be claiming a season it does not measure.

HOME WIN P	N	PRED	ACTUAL	GAP
< 35%	247	20.2%	31.2%	+11.0
35–45%	78	40.3%	42.3%	+2.0
45–55%	168	50.1%	45.8%	-4.3
55–65%	95	60.1%	56.8%	-3.2
≥ 65%	646	84.3%	78.6%	-5.7

How to read GAP and Brier

GAP is actual minus predicted, so a positive number means the model was too cautious in that bucket and a negative one means too confident. Brier 0.195 over 1,234 graded games; lower is better, and 0.25 is what a coin flip scores.

Products & rules

thresholds from config

Rules change; posted picks stay fixed

These are the definitions in force today. They change through the season as the model is refined, and a change applies from its date forward only: a pick is never refitted to a rule that did not exist when it fired, and the log is never edited to match a newer one. Every pick is counted in the record exactly as it posted.

PRODUCT	THE BET	PREMIUM	PREMIUM ★
Spreads · Premium	Against the spread, the sign of the edge picks the side. Promotes on edge and on how tight the posted spread is.	edge ≥ 6, spread ≤ 23	spread ≤ 14
Spreads · all graded	Every pick the model fires, including the marginal band.	every decided pick
Totals	Full-game over / under. In calibration, not yet a graded product.	in calibration, not yet a graded product

Where tier thresholds come from

Thresholds are read from the published config, never restated here.

All sports →

The log

493 rows

How to read this game log

Every priced game, with the market’s number beside ours and which one landed closer. BBMI ERR and VEGAS ERR are each side’s projected margin against the actual, so CLOSER is decided by those two columns and nothing else.

What the SIDE column means

SIDE is the team the pick took, read from the row’s own stored pick. Every row in view stores one. It is read and not derived, which shows: on 3 of these 493 rows the stored side is NOT the side the sign of the edge would give. A derived column would have agreed with the record on every other row and been wrong on those, with nothing here to say so.

									FINAL
2026-10-28	Jacksonville State @ New Mexico State	+3.0	+4.2	1.2	Jacksonville State	38%	No pick	WIN	24–0	19.8	21.0	BBMI
2026-10-28	Kennesaw State @ Middle Tennessee	+1.5	-0.8	2.3	Middle Tennessee	52%	No pick	LOSS	44–28	16.8	14.5	VEGAS
2026-10-27	Delaware @ Western Kentucky	-1.5	-4.7	3.2	Western Kentucky	63%	No pick	LOSS	35–21	18.7	15.5	VEGAS
2026-10-27	Sam Houston State @ Missouri State	-5.5	-8.8	3.3	Missouri State	74%	No pick	WIN	3–24	12.2	15.5	BBMI
2026-10-24	Vanderbilt @ Kentucky	-10.5	-10.5	0.0	Kentucky	77%	No pick	WIN	17–34	6.5	6.5	TIE
2026-10-24	UTSA @ Tulane	+3.0	+3.1	0.1	UTSA	41%	No pick	WIN	42–19	19.9	20.0	BBMI
2026-10-24	Indiana @ Michigan	+7.5	+7.9	0.4	Indiana	29%	No pick	LOSS	13–6	0.9	0.5	VEGAS
2026-10-24	San Diego State @ Colorado State	+1.5	+1.1	0.4	Colorado State	47%	No pick	WIN	49–56	8.1	8.5	BBMI
2026-10-24	Boise State @ Washington State	+14.5	+14.0	0.5	Washington State	16%	No pick	LOSS	31–10	7.0	6.5	VEGAS
2026-10-24	Iowa @ Minnesota	+7.5	+8.0	0.5	Iowa	28%	No pick	LOSS	10–13	11.0	10.5	VEGAS
2026-10-24	Kansas State @ Arizona State	+1.5	+2.2	0.7	Kansas State	44%	No pick	LOSS	17–28	13.2	12.5	VEGAS
2026-10-24	Hawaii @ Northern Illinois	+3.0	+4.0	1.0	Hawaii	39%	No pick	WIN	56–21	31.0	32.0	BBMI
2026-10-24	San Jose State @ Nevada	-5.5	-6.6	1.1	Nevada	68%	No pick	WIN	13–28	8.4	9.5	BBMI
2026-10-24	Louisiana @ Southern Miss	+10.0	+11.2	1.2	Louisiana	21%	No pick	LOSS	31–35	15.2	14.0	VEGAS
2026-10-24	USC @ Wisconsin	+2.5	+1.3	1.2	Wisconsin	46%	No pick	WIN	7–24	18.3	19.5	BBMI
2026-10-24	Texas Tech @ Cincinnati	+11.5	+13.0	1.5	Texas Tech	18%	No pick	LOSS	24–37	26.0	24.5	VEGAS
2026-10-24	Akron @ Kent State	-3.0	-4.9	1.9	Kent State	64%	No pick	WIN	20–35	10.1	12.0	BBMI
2026-10-24	North Dakota State @ New Mexico	-4.5	-2.5	2.0	North Dakota State	57%	No pick	LOSS	16–21	2.5	0.5	VEGAS
2026-10-24	Michigan State @ UCLA	-16.5	-18.5	2.0	UCLA	91%	No pick	WIN	10–28	0.5	1.5	BBMI
2026-10-24	Baylor @ Kansas	+1.5	-1.2	2.7	Kansas	53%	No pick	WIN	24–25	0.2	2.5	BBMI
2026-10-24	Syracuse @ North Carolina	-9.5	-12.2	2.7	North Carolina	81%	No pick	LOSS	31–30	13.2	10.5	VEGAS
2026-10-24	Texas A&M @ Alabama	-13.5	-16.3	2.8	Alabama	88%	No pick	LOSS	20–31	5.3	2.5	VEGAS
2026-10-24	Houston @ Utah	-6.5	-9.5	3.0	Utah	75%	No pick	WIN	13–21	1.5	1.5	TIE
2026-10-24	North Texas @ Navy	+3.0	0.0	3.0	Navy	50%	No pick	LOSS	49–39	10.0	7.0	VEGAS
2026-10-24	Miami (OH) @ Central Michigan	+3.5	+6.6	3.1	Miami (OH)	32%	No pick	LOSS	28–27	5.6	2.5	VEGAS

All picks locked before kickoff and graded against the closing line. Past performance is not a guarantee.