A one-point win off a 47-yard touchdown as time expired is the kind of finish that swallows every other storyline. It shouldn't. Underneath the walk-off, Michigan's offense had a genuinely bad two-and-a-half quarters — and the play log says exactly why, without needing a single word about the final snap.
The offense that opened the game and the offense that showed up for the next eight drives were not the same offense.
Michigan scored touchdowns on its first possession and its last possession. In between: six punts, an interception, and a fumble. That's the entire story of "Michigan couldn't do anything on offense" — it isn't a feeling, it's eight possessions in a row that produced zero points.
Only one series — a seven-play drive in the third quarter — ever got the ball into Western Michigan territory, and it only reached the 44 before stalling. Everything else Michigan ran between the two touchdowns happened entirely on its own side of the field.
The dead middle, drive by drive.
Here is every Michigan offensive possession in order, with each drive's expected-points value (EPA) attached — the independent, from-scratch model explained in full in section 07. The pattern isn't one bad quarter — it's a bad half sandwiched between two scoring drives, with the same three problems (field position, penalties, turnovers) showing up over and over, and only three of ten drives grading positive.
Opening drive. Stuffed on 3rd-and-goal from the 1, then Underwood punched it in on 4th down.
A 16-yard 3rd-and-5 conversion was wiped by an Illegal Formation penalty. The replayed down stalled two plays later.
Three-and-out. Net gain: 2 yards.
Included a negative pass play on 3rd-and-3.
Started at the 8. A sack on 3rd-and-7 ended it.
Two-minute drill from the Michigan 2. Never left Michigan's own side of the field before Underwood was picked off near the 27.
The one drive that reached Western Michigan territory (to the 44) — despite two more holding penalties, one erasing a 20-yard run.
Three-and-out.
Marshall's 15-yard run set up 1st-and-10 at the 22. Two plays later, Underwood fumbled on a scramble near the 31; Western Michigan recovered.
Down 12-7 with 35 seconds left, no timeouts. Underwood to Buchanan for 47 and the win, as time expired.
Michigan wasn't just playing Western Michigan. It was playing the field.
Michigan's average starting field position was its own 23.8-yard line. Western Michigan's was its own 40.2 — a 16-yard swing on every single possession. Three different Michigan drives started inside the 10: the 8, the 7, and the 2. Every one of those is a possession that needs a nearly perfect series just to reach midfield, before the offense has to do anything difficult at all.
That gap isn't the offense's fault, but it is the offense's problem. Long fields amplify every other issue on this list: a penalty that costs 10 yards from your own 20 is a survivable inconvenience; the same penalty from your own 8 turns a manageable drive into a near-impossible one.
Michigan's two biggest offensive plays of the day didn't count.
Michigan drew five offensive penalties. Most were the ordinary cost of doing business. Two of them erased the two most explosive plays the offense produced all game:
Underwood hit JJ Buchanan for 16 yards — a clean third-down conversion. An Illegal Formation flag brought it back. Running the same model used for the EPA section below on both branches — the actual replayed 3rd-and-10 versus the first down that should have existed — this one flag cost Michigan 2.45 expected points. The replayed down stalled two plays later and Michigan punted.
Broderick Kuzdzal broke a 20-yard run that would have reached the Western Michigan 30. A holding call on Underwood, of all people, brought it back. Same model, same accounting: 2.22 expected points erased. The drive still ended in a punt.
These aren't ordinary 5- and 10-yard setbacks — together they cost Michigan's offense 4.67 expected points, essentially a touchdown's worth of value, from two flags alone. That's the difference between a third-down conversion that keeps a drive alive and a punt, and between a first-and-goal-in-range field position and another long field. Add in a penalty that turned a stuffed run into 2nd-and-15, and a substitution penalty on the drive that ended in a fumble, and Michigan's offense fought its own equipment manager as much as it fought Western Michigan's defense.
Both Michigan turnovers turned directly into Western Michigan points.
Michigan gave the ball away twice and got nothing back — Western Michigan lost two fumbles of its own but recovered both. A −2 turnover margin in a game decided by one point is about as costly as a stat line gets, and this one was direct: both giveaways set up short fields that Western Michigan converted immediately into points.
Western Michigan field goal on the ensuing short field.
Western Michigan field goal on the ensuing short field.
Priced by the same expected-points model as everything else on this page, the interception cost Michigan 3.37 points and the fumble cost 2.35 — 5.72 combined. That single number explains almost the entire gap between how good Michigan's offense looked on clean snaps and how bad it looked in the final tally: see section 07.
Michigan couldn't sustain a drive without a mistake bailing it out.
Strip out the game-winning final drive and the one first down Western Michigan handed Michigan with a pre-snap penalty, and Michigan converted 2 of 9 real third downs — 22%. The box-score version, counting everything, still only reads 4 of 12 (33%). Six of Michigan's ten possessions ended in a punt. That's not an offense that got a couple of unlucky breaks — it's an offense that couldn't move the sticks on its own for most of the afternoon.
Success rate liked Michigan's offense. Full-value EPA doesn't.
Run this site's locked success-rate and explosive-play classifiers and Michigan's offense grades out ahead of Western Michigan's on a snap-to-snap basis — a better floor, more chunk plays, more yards per successful play:
Success rate is a floor metric — it rewards staying on schedule and doesn't care how a snap ends once it's "successful." It also doesn't know what a turnover is worth. That's exactly where this gets interesting. Built entirely on this site's independent EPA v2 model — trained from scratch on 1,544,257 plays across 11 seasons (2014–2025), never on CFBD's own ppa field, the same engine that grades every fourth-down decision elsewhere on Michigan Football Focus — and priced play by play, Michigan's offense flips sign the moment its two turnovers are included at full value:
Two plays — one interception, one fumble — are worth more in expected points than all 46 other Michigan snaps combined were worth in the other direction. That's not a knock on the offensive process; it's exactly why the site grades fourth downs and fumbles on EPA instead of success rate in the first place. A metric that can't see a turnover coming isn't measuring the thing that actually decides one-score games.
Only one sack allowed all game. Underwood finished 12 of 24 (50%) for 170 yards, one touchdown and the one interception — plus a team-leading 23 carries for 95 yards and a rushing score. The offensive line wasn't overwhelmed and the quarterback wasn't running for his life; the value problem was decision and ball security on a small number of snaps, not the other 46.
The defense didn't get gashed. It got flagged.
Run the same EPA model on every Western Michigan snap — which is really Michigan's defense allowing or preventing value — and the per-play picture is not close to "the defense couldn't get them off the field." It's closer to the opposite:
At −0.34 EPA allowed per play and −2.05 EPA allowed per drive, Michigan's defense had one of the more dominant per-snap performances you'll see in a single game — the average Western Michigan possession left them worse off, in expected-points terms, than when it started. Western Michigan ran 67 offensive snaps to Michigan's 48 — 19 more, across just one extra possession. That volume gap, not a per-play breakdown, is the real source of the "the defense could never get off the field" feeling. It's a time-of-possession illusion sitting on top of a genuinely dominant per-snap number.
The clearest proof: Western Michigan's longest drive of the day — 19 plays, 71 yards, third quarter, ending in a field goal — was still a −1.76 EPA possession. Nineteen snaps, three points, and Michigan's defense still won it by the numbers that matter. Bend, don't break: Western Michigan reached scoring range four times all game and left with four field goals — zero touchdowns allowed.
Money downs back it up. Western Michigan converted just 7 of 17 third downs (41%) against Michigan — a losing number for the offense, ten stops out of seventeen tries for the defense.
Erased a live Jyaire Hill interception at the Michigan 37. Western Michigan kept the ball and gained an automatic first down instead of losing it.
Turned a manageable 2nd-and-4 into a fresh set of downs 21 yards downfield.
The first one is the whole story in miniature: Jyaire Hill actually intercepted the pass. Michigan's defense earned the exact takeaway this article's offense section spent so much time wishing for — and a pass-interference flag took it away and handed Western Michigan a fresh set of downs instead, worth roughly 4.1 expected points by itself. A personal foul on the very next snap piled on 1.5 more. And the defense still forced a punt on that possession anyway. That's not a unit getting outplayed. That's a genuinely elite defensive performance with two self-inflicted holes in it — the exact same disease as the offense, on the other side of the ball.
Is the season over? No. But this game is a warning, not a fluke.
Michigan is 1-0. One point separated a clean win from a season-opening loss to a MAC team, and by EPA, both units actually graded out well on the snaps that were just football — Michigan's offense was fine on 46 of 48 plays, and its defense was flatly dominant on a per-snap basis (−0.34 EPA/play allowed) while giving up zero touchdowns all game. The difference wasn't talent, scheme, or either line getting physically beaten. It was two Michigan offensive turnovers worth 5.72 expected points, two Michigan offensive penalties worth another 4.67, and two Michigan defensive penalties on one series worth 5.61 more — north of 16 expected points self-inflicted across both sides of the ball, in a game decided by one. Those are correctable problems. They are not the same as being outclassed.
That's also exactly why this game can't be waved off as one weird night that ends the moment the highlight of Buchanan's touchdown stops trending. Western Michigan is a MAC opponent that punted the ball well and settled for four field goals instead of touchdowns. A Big Ten defense does not do Michigan the same favors. It converts a Michigan interception or fumble into seven points instead of three, and a Big Ten offense does not let a wiped-out interception and a free personal-foul first down turn into just a punt the way Western Michigan's drive did. Run this exact mistake profile — turnovers, backbreaking penalties on both sides of the ball, a rash of stalled third downs — against a real conference opponent, and 13-12 becomes a loss.
The underlying value says both of Michigan's units have real quality: a positive-EPA offense on clean snaps, and a defense allowing negative EPA on nearly every snap it played, with zero touchdowns surrendered. The scoreboard says a combined six self-inflicted plays were worth more than 16 expected points in the wrong direction. Both things are true, and the second one is the one that has to change before Big Ten play starts mattering.