College Football · Week 3

Lines board

Saturday, 19 September 2026 · 71 games, 70 with a DraftKings price, 68 trading on Kalshi

Built 17 September 2026 at 12:03 PM ET

Read this as monitoring, not as a signal

The rightmost column is how far our number sits from the market's. It is not an edge, and nothing on this page should be read as one.

Corrected 11 September 2026. This panel used to say the model had lost four separate tests against closing lines, including a best-fit blend weight of 0.00. Those tests were run unfairly. Every one of them fitted the ratings through the previous season and then graded a whole year forward, while the model that actually shipped refitted continuously — so a Week 13 prediction was being made on eleven-week-old ratings and compared against a line that had seen all eleven of those weeks. A blend weight of 0.00 against that opponent measures an information deficit, not the modelling.

Corrected again 15 September 2026. That fix compared the new ratings only with a fit frozen through the previous season. Tested against the plain season fit refitted before every slate, across every regular-season week of 2022–2025 (6,355 games), the plain fit misses actual margins by 13.64 points against 14.74, and 13.02 against 13.54 on FBS-vs-FBS games. The spreads on this page now come from that fit.

What did not change is the conclusion. Re-measured properly, the market still wins on margins, and no setting of these constants has produced a bet worth making. The model is better than the old number said and still not a signal.

The slate

Away team above, home team below and marked (H). Scores are projections. Sorted by market spread, biggest first. Two kinds of game are dimmed and tagged. FCS means the opponent is not an FBS team, so the model has no meaningful rating for it and the gap is noise in both directions. Mismatch means the market is three touchdowns or more, where our ratings simply do not extend far enough to price a 50-point favourite. In both cases the gap column measures the rating scale rather than the matchup.

Matchup & projected scoreDK spreadDK total DK MLDK home %Kalshi home % Our spreadOur home %Gap
Murray State 19
Oklahoma State 52
FCS
Oklahoma State −56.5 70.5 — / — 100% Oklahoma State −32.7 98% -23.8
Kent State 1
Ohio State 57
mismatch
Ohio State −52.5 58.5 — / — 100% Ohio State −55.9 100% +3.4
Western Kentucky 13
Indiana 47
mismatch
Indiana −44.5 60.5 — / — 100% Indiana −33.5 98% -11.0
Eastern Washington 12
Washington 45
FCS
Washington −42.5 56.5 — / — 100% Washington −33.5 98% -9.0
East Texas A&M 18
Tulsa 39
FCS
Tulsa −42.5 56.5 — / — 100% Tulsa −20.9 91% -21.6
Buffalo 7
Penn State 44
mismatch
Penn State −41.5 50.5 — / — 100% Penn State −37.3 99% -4.2
Northern Iowa 8
Iowa 41
FCS
Iowa −40.5 49.5 — / — 100% Iowa −33 98% -7.5
Mercer 19
Georgia Tech 39
FCS
Georgia Tech −38.5 58.5 — / — 100% Georgia Tech −20.5 91% -18.0
Maine 13
Boston College 41
FCS
Boston College −38.5 53.5 — / — 100% Boston College −27.9 96% -10.6
Southern Illinois 18
Illinois 37
FCS
Illinois −38.5 54.5 — / — 100% Illinois −19.5 89% -19.0
Duquesne 10
Washington State 44
FCS
Washington State −37.5 53.5 — / — 100% Washington State −34.2 99% -3.3
Delaware State 11
South Florida 50
FCS
South Florida −36.5 61.5 — / — 100% South Florida −38.5 99% +2.0
UTEP 6
Michigan 42
mismatch
Michigan −34.5 48.5 — / — 100% Michigan −35.7 99% +1.2
UT Martin 16
Memphis 42
FCS
Memphis −34.5 57.5 +4000 / -50000 98% 98% Memphis −25.5 95% -9.0
Kennesaw State 12
Tennessee 48
mismatch
Tennessee −34.5 59.5 +5000 / -100000 98% 100% Tennessee −36.4 99% +1.9
Stonehill 17
Massachusetts 33
FCS
Massachusetts −33.5 49.5 — / — Massachusetts −16.4 85% -17.1
Northern Illinois 13
Arizona 37
mismatch
Arizona −33.5 50.5 +4000 / -50000 98% 98% Arizona −23.7 94% -9.8
UTSA 18
Texas 41
mismatch
Texas −30.5 58.5 +1800 / -5000 95% 98% Texas −23 93% -7.5
Michigan State 12
Notre Dame 41
mismatch
Notre Dame −29.5 52.5 +1700 / -4500 95% 96% Notre Dame −29.2 97% -0.3
Utah State 14
Utah 42
mismatch
Utah −27.5 56.5 +2000 / -6500 95% 96% Utah −27.7 96% +0.2
Troy 14
Missouri 36
mismatch
Missouri −27.5 50.5 +1800 / -5000 95% 96% Missouri −21.8 92% -5.7
North Dakota State 34
Sacramento State 17
FCS
North Dakota State −27.5 50.5 -8000 / +2200 4% 4% North Dakota State −16.8 14% +10.7
Georgia 34
Arkansas 20
mismatch
Georgia −24.5 54.5 -2800 / +1300 7% 6% Georgia −14.4 18% +10.1
North Dakota 18
Nebraska 33
FCS
Nebraska −24.5 51.5 +1400 / -3200 94% 94% Nebraska −14.8 83% -9.7
Bowling Green 8
Iowa State 35
mismatch
Iowa State −23.5 43.5 +1300 / -2800 93% 94% Iowa State −27.2 96% +3.7
Akron 12
Minnesota 37
mismatch
Minnesota −23.5 48.5 +1100 / -2100 92% 94% Minnesota −24.9 94% +1.4
Eastern Michigan 12
Wisconsin 34
mismatch
Wisconsin −23.5 45.5 +1200 / -2400 93% 94% Wisconsin −22.2 92% -1.3
South Dakota 20
Boise State 37
FCS
Boise State −23.5 56.5 +1400 / -3200 94% 95% Boise State −16.6 86% -6.9
USC 36
Rutgers 24
mismatch
USC −21.5 59.5 -2100 / +1100 8% 6% USC −12 22% +9.5
New Mexico 12
Oklahoma 35
mismatch
Oklahoma −21.5 46.5 +1000 / -1800 91% 94% Oklahoma −22.9 93% +1.4
Tulane 18
Kansas State 31
Kansas State −20.5 49.5 +1000 / -1800 91% 92% Kansas State −13.2 80% -7.3
Florida State 14
Alabama 34
Alabama −19.5 48.5 +800 / -1350 89% 90% Alabama −19.8 90% +0.3
Arkansas State 16
TCU 41
TCU −19.5 56.5 +850 / -1450 90% 91% TCU −24.7 94% +5.2
Louisiana Tech 17
Baylor 34
Baylor −17.5 51.5 +625 / -950 87% 90% Baylor −16.7 86% -0.8
Charlotte 18
App State 33
App State −17.5 51.5 +700 / -1100 88% App State −14.8 83% -2.7
Georgia State 14
UCF 37
UCF −17.5 50.5 +650 / -1000 87% 90% UCF −23 93% +5.5
Nicholls 19
Sam Houston 31
FCS
Sam Houston −17.5 49.5 — / — 90% Sam Houston −11.8 78% -5.7
BYU 35
Colorado State 17
BYU −17.5 52.5 -950 / +625 13% 12% BYU −18 12% -0.5
Montana 24
Oregon State 28
FCS
Oregon State −17.5 52.5 +525 / — 86% Oregon State −3.6 59% -13.9
Kentucky 17
Texas A&M 34
Texas A&M −16.5 50.5 +575 / -850 86% 88% Texas A&M −16.5 86% +0.0
Miami (OH) 17
Cincinnati 33
Cincinnati −16.5 49.5 +550 / -800 85% 86% Cincinnati −16.1 85% -0.4
Ball State 17
Liberty 32
Liberty −14 49.5 +440 / -600 82% 84% Liberty −15.4 84% +1.4
Purdue 20
UCLA 33
UCLA −14 52.5 +470 / -650 83% 86% UCLA −12.8 79% -1.2
SE Louisiana 24
UL Monroe 25
FCS
UL Monroe −10.5 48.5 — / — UL Monroe −0.5 51% -10.0
West Virginia 24
Virginia 29
neutral site
Virginia −10.5 53.5 +310 / -395 77% 78% Virginia −5 63% -5.5
Stanford 16
Duke 34
Duke −10 50.5 +310 / -395 77% 77% Duke −18.2 88% +8.2
Western Michigan 24
Rice 19
Western Michigan −9.5 43.5 -340 / +270 26% 24% Western Michigan −5.3 37% +4.2
UAB 23
Louisiana 34
Louisiana −7.5 56.5 +260 / -325 73% 74% Louisiana −11.4 77% +3.9
Florida International 29
Florida Atlantic 34
Florida Atlantic −7 62.5 +205 / -250 69% 70% Florida Atlantic −5.4 64% -1.6
Ohio 23
South Alabama 27
South Alabama −7 49.5 +220 / -270 70% 72% South Alabama −4.1 60% -2.9
Fresno State 27
San José State 23
Fresno State −6.5 49.5 -250 / +205 31% 31% Fresno State −4 40% +2.5
Temple 17
Toledo 34
Toledo −6 50.5 +190 / -230 67% 68% Toledo −16.9 86% +10.9
Arizona State 26
Kansas 24
neutral site
Arizona State −5.5 50.5 -205 / +170 36% 35% Arizona State −1.7 46% +3.8
Coastal Carolina 27
Delaware 30
Delaware −4.5 56.5 +160 / -192 63% 64% Delaware −3.3 58% -1.2
Mississippi State 25
South Carolina 34
South Carolina −4 58.5 +145 / -175 61% 62% South Carolina −8.6 71% +4.6
Marshall 27
Missouri State 26
Marshall −4 52.5 -198 / +164 36% 38% Marshall −0.9 48% +3.1
Nevada 25
Middle Tennessee 25
Nevada −3.5 50.5 -180 / +150 38% 38% Nevada −0.4 49% +3.1
Colorado 22
Northwestern 26
Northwestern −3.5 47.5 +145 / -175 61% 62% Northwestern −4 60% +0.5
North Carolina 15
Clemson 28
Clemson −3 43.5 +136 / -162 59% 60% Clemson −12.6 79% +9.6
North Texas 33
Texas State 29
Texas State −3 62.5 +130 / -155 58% 57% North Texas −3.7 41% -6.7
NC State 22
Vanderbilt 29
Vanderbilt −3 50.5 +136 / -162 59% 60% Vanderbilt −6.9 67% +3.9
East Carolina 23
Old Dominion 25
Old Dominion −3 48.5 +124 / -148 57% 56% Old Dominion −2 55% -1.0
UConn 29
Southern Miss 26
UConn −3 54.5 -162 / +136 41% 42% UConn −3.3 42% -0.3
Georgia Southern 22
Jacksonville State 30
Jacksonville State −3 52.5 +140 / -166 60% 60% Jacksonville State −7.5 69% +4.5
Virginia Tech 24
Maryland 29
Virginia Tech −3 53.5 -142 / +120 44% 42% Maryland −5.4 64% +8.4
Florida 25
Auburn 28
Florida −2.5 53.5 -135 / +114 45% 44% Auburn −3.4 59% +5.9
LSU 24
Ole Miss 34
LSU −2.5 57.5 -135 / +114 45% 46% Ole Miss −10 74% +12.5
Wyoming 19
Central Michigan 19
Wyoming −1.5 38.5 -112 / -108 50% 50% Central Michigan −0.1 50% +1.6
SMU 28
Louisville 31
Louisville −1.5 59.5 +100 / -120 52% 52% Louisville −3.5 59% +2.0
James Madison 28
San Diego State 19
San Diego State −1.5 46.5 +110 / -130 54% 54% James Madison −8.7 29% -10.2
Wagner
California
FCS
— / — 100% California −44.5 100%

Where the projected scores come from. The margin is ours; the total is the market's. We built a totals model and it graded worse than always predicting the league average — MAE 13.43 against 13.24 on 762 walk-forward games, and a sweep across five ridge values and two half-lives put every single combination above that baseline. College scoring is too noisy and team scoring rates carry too weakly. So the market's total is the better number, and splitting it with our margin is what produces these two scores. All 80 games on this slate carry one.

The three percentage columns are all the home team's chance of winning, so they read straight across. The away side is one minus each.

DK's moneyline is shown as posted, away then home. Its percentage is de-vigged: a two-way market's implied probabilities add up to more than 100%, and that excess is the book's fee rather than its opinion. Kalshi is an exchange and carries no equivalent, so comparing raw implieds would make DK look more confident than Kalshi on both sides of every game. Normalising to 100% is what makes the two columns mean the same thing.

Kalshi's number is the midpoint of the bid and ask on the contract for that team winning. It quotes an outcome rather than a spread, so a probability is the comparison that invents nothing in between. 53 of the 71 games have a posted moneyline; the rest are lopsided enough that the book leaves them off the board.

Where we disagree most

FBS-versus-FBS games with the market inside 21 points. 37 of the 71 on the slate qualify; the rest are buy games or mismatches where a large gap tells us nothing.

GameMarketOursGap
LSU at Ole Miss LSU −2.5 Ole Miss −10 +12.5
Temple at Toledo Toledo −6 Toledo −16.9 +10.9
James Madison at San Diego State San Diego State −1.5 James Madison −8.7 -10.2
North Carolina at Clemson Clemson −3 Clemson −12.6 +9.6
Virginia Tech at Maryland Virginia Tech −3 Maryland −5.4 +8.4
Stanford at Duke Duke −10 Duke −18.2 +8.2
Tulane at Kansas State Kansas State −20.5 Kansas State −13.2 -7.3
North Texas at Texas State Texas State −3 North Texas −3.7 -6.7

Worth watching week to week to see whether the disagreements are systematic. That is the useful thing this board can tell us, and it is a different question from whether any single one of them is right.

Team receiving yards

Kalshi's college team markets against our player model. 28 contracts across 4 teams — Alabama, Florida St., Louisville, SMU — which is the whole of their college team-props market, not a selection from it.

TeamLineOur yards Our P(over)MarketSpread VolumeEdge
Alabama 250169.9 18% 42% 17c 191 -23
Louisville 275160.2 10% 28% 32c 20 -18
Louisville 225160.2 23% 67% 8c 0 -44
Louisville 200160.2 33% 76% 8c 0 -43
Louisville 175160.2 43% 84% 8c 0 -41
Louisville 250160.2 16% 54% 5c 0 -39
Alabama 200169.9 37% 66% 19c 0 -30
Florida St. 175199.6 61% 32% 41c 0 +28
Alabama 225169.9 27% 55% 18c 0 -28
Alabama 175169.9 48% 75% 18c 0 -27
Florida St. 150199.6 71% 46% 40c 0 +25
Florida St. 200199.6 50% 26% 29c 0 +23
Florida St. 225199.6 39% 16% 28c 0 +23
Florida St. 125199.6 80% 59% 40c 0 +21
Alabama 275169.9 12% 29% 19c 0 -16
Florida St. 100199.6 87% 71% 38c 0 +16
Louisville 325160.2 3% 18% 13c 0 -15
Louisville 300160.2 6% 19% 28c 0 -13
Florida St. 75199.6 92% 79% 34c 0 +13
Alabama 300169.9 7% 18% 13c 0 -11
SMU 300277.2 40% 50% 10c 0 -10
SMU 275277.2 51% 61% 10c 0 -10
SMU 325277.2 30% 39% 10c 0 -9
SMU 250277.2 62% 71% 10c 0 -9
SMU 350277.2 21% 29% 8c 0 -8
SMU 225277.2 72% 79% 10c 0 -7
SMU 375277.2 14% 20% 9c 0 -7
Alabama 325169.9 4% 10% 16c 0 -6

Team receiving yards are team passing yards, since every receiving yard is a passing yard. So the projection is the starting quarterback's, not the sum of every receiver's. Summing receivers was the first attempt and gave USC 647 yards, roughly double any real team game: each receiver's projection assumes he plays and takes his usual share, and nothing in the sum is constrained by how many passes the team will actually throw.

We sit under the market on every row here. Not one of them is a projection above the line. That is a calibration observation, not an edge: either our projections are systematically low on team passing volume, or the books shade these lines up because casual money buys overs. Both are ordinary and they look identical from one weekend of four teams. Worth watching across several weeks before reading anything into it.

Probability is a normal around the projection with sigma 89.4 yards, which is the measured standard deviation of actual team receiving yards across every team-game in our box history, not an assumption. A team total is a sum over many plays, so the normal is doing honest work here in a way it would not for one player's single game.

Their team touchdown market is deliberately absent. It counts every touchdown a team records, defensive and special teams included, and our model projects offensive production only. Every row would sit low by whatever the non-offensive rate is — a one-directional bias on the whole table, which is worse than no table. It goes in once that rate is measured rather than guessed.

What is behind the numbers

Ridge-regularised Massey ratings fitted on 8,458 games through the current season, covering 408 teams. Home field 4.4 points, residual sigma 15.6. Win probability is the normal CDF of the projected margin over that sigma.

FanDuel is not here. The Odds API key is out of credits, plain requests get a 403, and a real browser session gets a CloudFront denial on the page itself, which reads as a geographic block on this host rather than bot detection. Four routes, all closed, so the column is left out and labelled rather than quietly blank.