Here is what the numbers actually said about the 16 games of week 3.

The single most informative thing that happened

Bijan Robinson carried 29 times for 194 yards against Green Bay, and Atlanta as a team ran 40 times for 0.283 EPA per rush.

That number matters for three reasons. First, the volume: 40 designed rushes on 69 offensive plays is a game plan, not a script that got away from anyone. Second, the efficiency: Robinson averaged 5.14 yards per carry across 288 attempts in 2025 with a rushing EPA of −0.043, which is roughly the league-average starter. In one game he averaged 6.7 and posted a positive team-level rush EPA of 0.283. Third, the opponent: Green Bay's rush defense allowed −0.066 EPA per carry across 482 rushes in 2025 — that is genuine top-tier run defense, sustained over a full season. In week 3 that same unit gave up 0.283 EPA per rush over 40 attempts on its home field, in a game the Falcons won 35-14.

Atlanta's team offense had posted −0.321 and −0.473 EPA per play in weeks 1 and 2 on 57 and 67 plays. In week 3 it produced +0.332 on 65 plays. That is a 0.7-to-0.8 EPA-per-play swing from the same offense against a defense that had no history of allowing it. This is one game — the sample warning applies — but the direction and the size of the departure from the Packers' 2025 baseline is the kind of thing that shows up in a weekly review because it doesn't fit anywhere else.

What week 3 actually told us

1. Atlanta's offense produced a full standard deviation more than its two-week self. Weeks 1 and 2: −0.321 EPA/play on 57 plays, −0.473 on 67. Week 3: +0.332 on 65. The Falcons finished 2025 at −0.022 offensive EPA/play across 1,035 plays, so both the bad opening and the week-3 outburst are outside their year-long norm. Three games is not enough to declare a new identity; a 40-carry rushing script against a good run defense is enough to say the plan changed.
2. Jacksonville held New England to −0.253 offensive EPA over 60 plays and forced three turnovers. The Patriots went 4-for-12 on third down and took three sacks; the Jaguars converted 8 of 12. New England's 2025 offense produced +0.090 EPA/play across 1,268 plays — an above-average unit for the year — so a −0.25 game against Jacksonville is a real outlier for that offense. Jacksonville's own 2025 defense was −0.086 EPA/play allowed on 1,090 snaps (a good unit), so this isn't a defense inventing a new level; it's a good defense playing a good game against an offense that had been fine.
3. Philadelphia scored 7 points in Chicago on 47 offensive plays and −0.304 EPA per play. The Eagles' pass game specifically was −0.555 EPA per dropback across 33 attempts. Their 2025 offense was +0.021 EPA/play on 1,067 plays and had opened 2026 at +0.071 and +0.039. This is the first week of the young season where Philadelphia's offense didn't function at all, and the small denominator — 47 snaps, driven partly by two lost-possession turnovers — is part of why the number is so extreme.
4. The 49ers put up 0.455 EPA per play on 53 snaps against Arizona, and passing carried it: 0.841 EPA per dropback on 31 attempts. The scoreboard read 36-30, but Arizona ran 85 plays to San Francisco's 53 and needed all of them; the Niners' passing efficiency in a home game as 7.5-point favorites is the number I would come back to when the sample gets larger.
5. Two games where nobody was efficient still produced a winner. Minnesota beat Tampa Bay 23-16 with both offenses at −0.295 and −0.296 EPA per play (59 and 69 plays); Tennessee beat the Giants 12-7 with both teams at −0.144 to −0.146 EPA per play across 60 and 63 plays. These games happen every week — the reminder is that a "team X wins ugly" headline in September is often two teams playing badly, not one team defending well.
6. Seattle and Washington played a 33-31 game whose EPA line was almost perfectly even — SEA −0.026 on 67 plays, WAS −0.024 on 67 plays — and both teams passed efficiently (0.157 and 0.173 EPA per dropback) while running terribly (−0.564 and −0.281). The scoreboard, the play counts, and the split-by-play-type all agree this was a coin flip that landed on the road team. That's a genuinely tight game, and the numbers agree with it.

What looked like news and wasn't

Chicago 27, Philadelphia 7 was not the blowout the score suggests. Chicago's offense posted +0.134 EPA per play on 70 snaps; Philadelphia's posted −0.278 on 50. That is a real gap, and Chicago won the football game convincingly. But the 20-point margin was built on Philadelphia handing over the ball three times — one lost fumble and two interceptions — including a Bears drive that started at the Eagles' 15-yard line and one that started at the 42. Chicago scored three offensive touchdowns; two of them came from drives that began in Philadelphia territory after a turnover. The Bears' offensive rating this week is the same rating whether the margin was 20 or 6. Treating the scoreboard as the whole story here overstates how well Chicago's offense actually played.

Jacksonville 35, New England 6 goes the other way — the scoreboard undersells it. Yes, the Jaguars benefited from a −2 New England turnover margin, three sacks, and a 4-for-12 Patriots third-down day. But the underlying EPA numbers agree with the scoreboard: JAX +0.194 on 61 plays with 3 passing touchdowns and 2 rushing touchdowns, NE −0.253 on 63. This one earned its 29-point margin. It's the exception, not the rule, and worth naming as such.

Baltimore 34, Dallas 31 was not a "shootout between two good offenses" in any efficiency sense. Baltimore's offense was +0.196 EPA/play on 64 snaps and Dallas's was +0.117 on 74. Both fine, neither elite. Baltimore's number was carried almost entirely by passing — 0.622 EPA per dropback on 28 attempts against a 2025 Dallas defense that finished at +0.167 EPA/play allowed (the worst mark of any team on this list). A pass game that carves up the league's most permissive 2025 defense on 28 dropbacks tells you less than a pass game that does it on 45 against a top-10 defense. The margin was 3 and the total was 65 in a game with a 53.5 total — high-scoring is not the same as high-quality.

What to watch in week 4

The open question worth resolving isn't who wins next weekend; it's whether any of the week-3 outliers were signal.

• For Atlanta: does the Falcons' rush volume stay north of 35 attempts, and does the per-carry efficiency stay positive against an opponent whose 2025 run defense wasn't Green Bay-level? One game against a good run defense is a data point; two against different defenses starts to be a plan.
• For Philadelphia: the passing offense posted −0.555 EPA per dropback in Chicago after two weeks of merely-fine production. If week 4 comes in near the 2025 baseline of +0.021 EPA/play overall, week 3 was a bad afternoon; if it doesn't, that's the piece to write next.
• For Jacksonville's defense: New England's offense is not the 2025 New England offense until it isn't, and one bad Patriots game does not remake the Jaguars' pass rush. The evidence that would settle it is whether Jacksonville's next opponent produces anywhere near its own baseline — that is the opponent adjustment that early-season EPA does not do for you.