South Florida at Army. Lean: Army -3.5. Lean: Under 46.5.

Spread.EL makes Army -2.9 versus the market's -3.5, effectively agreement given the model's error and unproven ATS record. My lean toward Army comes from USF's transition: the dossier incorrectly says no first-year head coach, while Brian Hartline is rebuilding the offense around Michael Van Buren. USF generated only 208 yards against FIU despite committing no turnovers; its three defensive takeaways all produced scoring drives. That raises questions about sustaining offense without short fields. Army's established quarterback and rushing attack present a difficult first road assignment. However, Bryant is inadequate evidence that Army's 59-point performance translates against FBS competition. The move from early Army -1.5 to -3.5 already prices considerable concern about USF, and laying beyond three materially reduces appeal. Reported betting splits favor Army but conflict across snapshots, so they add no conviction. I lean Army modestly; neither the small model discrepancy nor unconfirmed receiver injuries justifies a strong position.

Total.The under case rests on USF's changed offense and an expected possession-limiting Army rushing approach, not weather. USF's opener produced one touchdown and only 208 yards, making last year's 40.5-point scoring average a questionable foundation for EL's 50.39 projection. The over deserves consideration: Army demonstrated explosive rushing, USF can improve after its first game, and mild conditions should permit normal execution. Furthermore, the market has already removed six to eight points from some early totals, so the easiest under value is gone. I still lean under 46.5 because sustained USF scoring remains less convincing than the historical model implies. Receiver uncertainty is only a minor consideration, with neither player confirmed out. This is a moderate lean after a substantial adjustment, not a strong bet.

Where our number looks wrong.USF's first_year_head_coach=false is factually incorrect. Its offensive continuity assumptions are suspect following Hartline's arrival and the quarterback change. The totals model likely overvalues the relevance of USF's 2025 scoring. Conversely, Army's margin-only update risks overrewarding a rout of Bryant. These opposing errors prevent treating Army -2.9 as a precise fair line.

Spread.Army is -3.5 at every book and has been since 9/6, after openers of -1.5 (FanDuel 9/3, DraftKings 9/4). That two-point steam is priced off a 59-3 win over FCS Bryant with 569 rushing yards, the most by an FBS team since 2022 (USF Oracle 9/9), an outlier that projects almost nothing to an FBS game, and off USF's ugly 208-yard, 19-9 win over FIU. Our in-season number says Army -2.9 and the preseason-prior version says only -1.7 (Army 0.34 vs USF 1.178 plus 2.5 home); both sit on the USF side. The talent gap still points the other way: SP+ prior 11.6 vs 0.8, portal net +13, LSU transfer QB Michael Van Buren Jr., power-conference transfer receivers, and a defense that forced three turnovers and gave up nine points in Week 1. Army went 7-6 in 2025 at a +2.2 average scoring margin, and in a projected ~43-point game 3.5 points is a large fraction of the distribution. Take the points; the option-prep angle keeps this a lean, not a bet.

Total.The total opened 54.5 (FanDuel 9/3) and 52.5 (DraftKings 9/4) on USF's 2025 offense, which averaged 40.5 points and no longer exists; it reopened 49.5 on 9/6 and is 46.5 everywhere now. Our totals model says 50.4 and is making the openers' mistake, it is fit mostly on last year's units and cannot see that USF replaced its head coach, both coordinators, its quarterback and 20 of 24 starters. Army led FBS in time of possession in 2025 at 35:16 and was last in pass attempts; its 2025 games averaged 45.4 combined points. USF gained 208 yards in Week 1 and 13 of its 19 points came from field goals. Fewest possessions in the sport against a 127th-ranked offense, with neutral weather. Three points of the move are already gone, so this is a lean rather than a full-size play.

Where our number looks wrong.The in-season v1 refit learns from final margins only, so a 56-point win over FCS Bryant swung Army from 0.34 to 1.042 while a workmanlike 19-9 win dropped USF from 1.178 to 0.649. That is the same Week 1 overreaction the market made, only harder, and it is exactly the FCS-opponent case the model's own notes flag as least reliable. Neither number describes the team USF is fielding: the preseason prior is built on the old roster's recruiting composite, SP+ and returning production, and the 12.6% returning-production input tells you the prior is nearly meaningless here after a coaching change, both coordinators replaced, a new quarterback and 20 of 24 starters gone. Flat 2.5 home field also has no term for a noon academy kickoff at Michie, and nothing in the model prices one week of preparation against a flexbone triple option, the single largest variable in this game, and the one that would most easily make me wrong on this side.

Spread.Our model has Army favored by only 2.9 vs. a market close of -3.5, a 0.6-point gap well inside the model's own no-edge zone, and the model has no proven ATS track record regardless. That leaves situational reads. Army's in-season rating jumped off one 59-3 blowout of FCS Bryant, a sample the model itself flags as least reliable; preseason actually rated USF the better team (1.18 vs 0.34) with a higher recruiting/portal profile. Real-world evidence points the other way, though: USF's offense produced only 208 total yards against FIU (127th of 138 FBS teams), and its defense returns one starter of 28 total starts, facing an Army option attack (648 yards, 569 rushing, 16 ball carriers vs Bryant) most defenses see once a year. Army DL Jack Bousum (boot, questionable) is the one injury cutting the other way. Lean Army, small edge, not a market inefficiency.

Total.The market has already moved hard: every tracked book opened between 49.5 and 54.5 and closed at 46.5 by 9/10, after both teams showed the total should be lower, Army's offense is run-first and clock-eating (569 rush yards, zero pass explosiveness historically), and USF managed just 208 yards and 19 points against FIU. Weather is a non-factor (mostly sunny, 76F, light wind, no rain). Two shaky offenses and an option team that shortens the game argue for the market's newer, lower number over our stale model.

Where our number looks wrong.Army's -2.9 in-season number is driven almost entirely by one 59-3 win over FCS Bryant, the exact scenario the model's own documentation calls least reliable, while the preseason composite actually favored USF (1.18 vs 0.34, plus a +13 portal net vs Army's -7). The model has no visibility into Bousum's DL injury, USF's near-total defensive turnover, or Army's historically anemic passing game, so its gap-only read of a 0.6-point disagreement understates how one-sided the matchup context actually is, even though the market number ends up close to right.