The SportsWire

Week 4 in review

2026 · Fri Sep 25, 7:00 PM CT / 8:00 PM ET – Sat Sep 26, 10:00 PM CT / 11:00 PM ET · 26 games with a team page · computed, not written

Updated Tue Oct 13, 7:56 PM CT

‹ Week 3Week 5 ›ProjectionsMetrics explained

Flattered and undersold

Flattered · the score ran ahead of the play by more than 7 points

  • Georgia vs Oklahoma, 41–13 W+28 vs -10 implied · +38
    The scoreboard flattered Georgia: the +28 margin ran 38 points ahead of the -10 implied by per-play quality (PPA (expected points added per play) -0.03 vs +0.12, success rate 37% vs 34%) (Oklahoma turned it over 4 times).
  • Miami vs Central Michigan, 52–3 W+49 vs +32 implied · +17
    The scoreboard flattered Miami: the +49 margin ran 17 points ahead of the +32 implied by per-play quality (PPA (expected points added per play) +0.41 vs -0.09, success rate 59% vs 27%) (Central Michigan turned it over 3 times).
  • Notre Dame at Purdue, 49–10 W+39 vs +28 implied · +11
    The scoreboard flattered Notre Dame: the +39 margin ran 11 points ahead of the +28 implied by per-play quality (PPA (expected points added per play) +0.40 vs -0.06, success rate 58% vs 38%) (Purdue turned it over once).
  • Iowa at Michigan, 20–19 W+1 vs -9 implied · +10
    The scoreboard flattered Iowa: the +1 margin ran 10 points ahead of the -9 implied by per-play quality (PPA (expected points added per play) -0.01 vs +0.12, success rate 27% vs 45%) (Michigan turned it over once).
  • Florida vs Ole Miss, 52–28 W+24 vs +15 implied · +9
    The scoreboard flattered Florida: the +24 margin ran 9 points ahead of the +15 implied by per-play quality (PPA (expected points added per play) +0.36 vs +0.20, success rate 60% vs 56%) (Ole Miss turned it over 3 times).

Undersold · the score ran behind the play by more than 7 points

  • Oklahoma State at West Virginia, 41–24 W+17 vs +33 implied · -16
    The scoreboard undersold Oklahoma State: the +17 margin ran 16 points behind the +33 implied by per-play quality (PPA (expected points added per play) +0.41 vs -0.05, success rate 56% vs 44%) (Oklahoma State turned it over once).
  • SMU vs Missouri State, 34–24 W+10 vs +23 implied · -13
    The scoreboard undersold SMU: the +10 margin ran 13 points behind the +23 implied by per-play quality (PPA (expected points added per play) +0.19 vs -0.12, success rate 55% vs 30%) (SMU turned it over twice).
  • Texas Tech vs Sam Houston, 49–14 W+35 vs +44 implied · -9
    The scoreboard undersold Texas Tech: the +35 margin ran 9 points behind the +44 implied by per-play quality (PPA (expected points added per play) +0.38 vs -0.29, success rate 66% vs 31%) (Texas Tech turned it over 3 times).
  • Boise State at Western Michigan, 32–7 W+25 vs +32 implied · -7
    The scoreboard undersold Boise State: the +25 margin ran 7 points behind the +32 implied by per-play quality (PPA (expected points added per play) +0.35 vs -0.16, success rate 52% vs 38%).

Closest projection

Over the 20 games this week where SP+, Elo and the line all existed before kickoff, the line as a benchmark (its miss only, never closest) (5 more shown on the Projections page but not in the record: no page before kickoff).

modelavg missclosest
SP+12.014 of 20
Elo14.26 of 20
Line12.0—

Our model experimental · 21 eligible games · vs SP+: 11.9 against 12.0, closest 10 / 10 · vs Elo: 11.9 against 14.2, closest 12 / 8 · vs Line: 11.9 against 12.0, closest 10 / 10 · all four: our model closest 9 of 20

Our call · 20 games where SP+, Elo and Our model all existed before kickoff, each model on those same games, and the line for comparison (its miss only: a benchmark is never counted as naming the winner):

modelgamesavg missnamed the winner
Our call2012.416 of 20
SP+2012.015 of 20
Elo2014.215 of 20
Line (benchmark)2012.0—
Our model2011.916 of 20

Against its benchmark, the line: Our call was at least as close in 8 of 20, the line in 12 (a tie counts for both).

Upsets

Ranked teams that lost to an unranked or lower-ranked opponent, by the AP Top 25 in effect for week 4, with each model's pregame margin from the ranked side and the line beside them for comparison only; ✓ when a model had the winner, ✗ when it did not.

  • #4 Ole Miss lost at #21 FloridaFlorida 52, Ole Miss 28
    SP+ ✓
    -4.7
    Elo ✗
    +5.1
    Line
    -1.5
    Our model ✓
    -3.4
  • #12 USC lost vs #20 OregonOregon 41, USC 27
    SP+ ✓
    -0.3
    Elo ✗
    +1.7
    Line
    +1.5
    Our model ✓
    -4.9
  • #13 Penn State lost vs WisconsinWisconsin 24, Penn State 20
    SP+ ✗
    +16.5
    Elo ✗
    +16.6
    Line
    +10.5
    Our model ✗
    +14.7
  • #16 Louisville lost vs Wake ForestWake Forest 30, Louisville 27
    SP+ ✗
    +12.2
    Elo ✗
    +9.5
    Line
    +13.5
    Our model ✗
    +7.1
  • #19 Missouri lost at #24 Mississippi StateMississippi State 31, Missouri 24
    SP+ ✗
    +0.5
    Elo ✓
    -1.4
    Line
    -3.5
    Our model ✓
    -2.6

Called it: SP+ 2 of 5 · Elo 1 of 5 · Our model 3 of 5.

Biggest swings

The plays that moved win probability most, across the 26 games with win-probability data, from the home side.

  • Wisconsin at Penn StatePenn State 74% → 9%
    Q1, 3rd & 13: No Huddle-Shotgun #1 C.Joseph pass complete short middle to #89 J.Harris caught at WIS47, for 72 yards to the PSU00 TOUCHDOWN, clock 01:13, G.Lahm kick attempt good (H: #91 S.West, LS: #56 J.Roe) · final Penn State 20, Wisconsin 24
  • Minnesota at WashingtonWashington 6% → 48%
    Q4 1:59, 1st & 10: (01:59) Shotgun #1 D.Williams Jr. rush middle for 13 yards gain to the MINN00 TOUCHDOWN, clock 01:53, 1ST DOWN Timeout Washington, clock 02:00 #95 T.Robles kick attempt good (H: #45 L.Dunne, LS: #53 R.Kean) · final Washington 24, Minnesota 27
  • South Alabama at KentuckyKentucky 23% → 64%
    Q2 1:22, 1st & 10: (01:22) Shotgun #26 J.Patterson rush middle for 75 yards gain to the SA00 TOUCHDOWN, clock 01:12, 1ST DOWN #88 A.Zouagui kick attempt good (H: #93 W.Berry, LS: #47 S.Radnoti) · final Kentucky 45, South Alabama 21

Regression watch, checked

Last week's teams whose score ran more than 7 points from the play, and what they did this week against their season-to-date form (average margin before the game, home field removed). The 2022–25 study found flattered teams fell -3.0 short of form the next game and undersold teams beat it by +1.7.

  • Texas · flattered +21 vs UTSAat Tennessee 20–17 W · -17.7 vs form (-8.0 vs the field)
  • Texas A&M · flattered +19 vs Kentuckyat LSU 6–35 L · -46.7 vs form (-37.0 vs the field)
  • Ohio State · flattered +16 vs Kent Statevs Illinois 42–19 W · -14.7 vs form (-5.0 vs the field)
  • SMU · flattered +15 vs Louisvillevs Missouri State 34–24 W · -6.3 vs form (+3.3 vs the field)
  • Tennessee · flattered +11 vs Kennesaw Statevs Texas 17–20 L · -38.3 vs form (-28.7 vs the field)
  • Georgia · flattered +10 vs Arkansasvs Oklahoma 41–13 W · -19.7 vs form (-10.0 vs the field)
  • Utah · flattered +10 vs Utah Stateat Iowa State 31–17 W · -20.3 vs form (-10.7 vs the field)
  • Texas Tech · flattered +8 vs Houstonvs Sam Houston 49–14 W · +21.3 vs form (+31.0 vs the field)
  • USC · flattered +8 vs Rutgersvs Oregon 27–41 L · -35.5 vs form (-25.8 vs the field)
  • Washington · undersold -17 vs Eastern Washingtonvs Minnesota 24–27 L · -19.0 vs form (-9.3 vs the field)
  • Louisville · undersold -15 vs SMUvs Wake Forest 27–30 L · -21.5 vs form (-11.8 vs the field)
  • Iowa · undersold -11 vs Northern Iowaat Michigan 20–19 W · -26.7 vs form (-17.0 vs the field)
  • Houston · undersold -8 vs Texas Techat Georgia Southern 42–28 W · -10.0 vs form (-0.3 vs the field)
  • Florida · undersold -8 vs Auburnvs Ole Miss 52–28 W · -10.7 vs form (-1.0 vs the field)

Baseline: all 35 page teams that played ran -9.7 vs form this week. The baseline matters because early-season form comes mostly from easier non-conference games, so the whole field tends to fall short of it once conference play starts; a flagged group is read against the field, not against zero.

Flattered: -19.8 vs form over 9, -10.1 against the baseline (-3.0 in the study) · Undersold: -17.6 vs form over 5, -7.9 against the baseline (+1.7 in the study). One week is a few teams and proves nothing either way; the point is that the check is visible every week.

How these are computed

Verdicts are the team pages' own: the score against the margin the plays earned, recomputed here from the box score and drives, so any past week rebuilds the same way. The closest-projection table is the accuracy report's block for the week; a game counts only if one of its teams had a page before kickoff, and our model is kept on its own line. SP+ is graded only in weeks where its own rating was posted before kickoff; an earlier snapshot shown before a game is never scored. Upsets use the AP Top 25 in effect for the week's games; every ranked team has a page, so no upset is missed. Pregame margins are the same ones the season record scores. Swings are the win-probability moves in the play-by-play. The regression check uses the study's own definition of form. Nothing on this page is written by AI, and a section with nothing to show is left out.