Through 33 completed games, every team's offense and defense has recorded somewhere between 99 and 152 plays this season. That is the entire evidence base. Cincinnati's defense has 127 plays on the ledger. Buffalo's offense has 118. The exercise is to figure out which of the sharp early moves are big enough to survive the smallness of the sample — and which are the sample doing what small samples do.

This is a survey, not a thesis. Below: a quick-take scoreboard, what the first two weeks license us to believe, three individual cases worth watching, and a closing pass on where the evidence is strongest and where it isn't.

Quick take

Table of Contents

1. The eight-and-eight scoreboard
2. What we think we know (five takeaways)
3. Three individual cases worth watching
4. Where the two-week evidence stands

1. The eight-and-eight scoreboard

Eight teams sit at 2-0 after two weeks: San Francisco, Seattle, Las Vegas, Kansas City, Minnesota, Cincinnati, Buffalo, and Philadelphia. Eight teams sit at 0-2: Tampa Bay, Tennessee, Houston, Washington, Indianapolis, the L.A. Chargers, Miami, and Atlanta. Point differential inside the 2-0 group ranges from San Francisco's +42 (62 scored, 20 allowed across two games) to Philadelphia's +6 (48-42). Inside the 0-2 group, it ranges from Tampa Bay's -10 (46-56) to Atlanta's -38 (16-54).

That spread matters. Being 2-0 by 42 and being 2-0 by 6 are not the same finding, and treating them as one is exactly the trap a two-week sample sets.

2. What we think we know

1) San Francisco has looked the part on both sides, and the schedule hasn't been a gift. The 49ers are averaging +0.298 EPA/play on offense over 111 plays and -0.133 EPA/play on defense over 112 — both top-five league marks through week 2. Their two opponents were the Rams and Dolphins. The Rams' 2025 defense actually finished comfortably above average (-0.047 EPA/play, a top-third unit), while Miami's was among the league's weaker ones (+0.088). Average it out and San Francisco's opponents allowed defensive EPA of +0.021 — almost exactly league-neutral. This doesn't look like a schedule mirage on either side, but the sample is 111 offensive plays.

2) Buffalo's offense went off — against a real defense, not a soft one. The Bills scored 36 and 41 (77 total points) and their offensive EPA/play jumped from +0.138 in 2025 (1,203 plays) to +0.344 over 118 plays in 2026 — the second-largest offensive jump in the league behind San Francisco's own. Here's the part worth correcting: their two opponents, Houston and Detroit, were not a weak pass-rush schedule. Houston's defense finished 2025 as the single best in the league by EPA/play allowed (-0.161); Detroit's was a top-12 unit (-0.011). Buffalo's offense did this against a strong-to-elite defensive slate, which makes the outburst more believable, not less. The real problem is the other side of the ball: Buffalo's own defense fell from a solid -0.011 in 2025 to +0.144 through two games, a decline of about 0.155 EPA/play — the fourth-largest defensive regression in the league. Buffalo has been very fun on offense. It has also stopped playing defense.

3) Two defenses have swung hard in opposite directions — for different reasons. Cincinnati's defensive EPA/play was a bottom-quartile +0.130 across all of 2025 (1,035 plays); through 127 plays in 2026 it's -0.222, a swing of roughly a third of a point per play — the largest defensive turnaround in the league (the Jets are next-closest, at -0.318). The two opponents so far, Tampa Bay and Houston, had below-average 2025 offenses (-0.009 and -0.031 EPA/play), so some of this should regress once Cincinnati faces a real offense. Indianapolis moved the opposite direction: from a roughly league-average +0.004 in 2025 to a league-worst +0.269 allowed through 138 plays — the largest defensive sample of any team so far — against Baltimore and Kansas City, whose 2025 offenses were both mildly above average (+0.034 combined). That's a real collapse, not primarily a scheduling story.

4) Atlanta's offense has been genuinely bad, not schedule-noise bad. The Falcons are last in the league in offensive EPA/play at -0.403 over 124 plays, down from -0.022 across 1,035 plays in 2025 — a -0.381 swing, the worst offensive regression in the league. Their opponents' average 2025 defensive EPA (+0.046) suggests the schedule has been ordinary. Two weeks at -0.40 is a different creature from a bad stretch.

3. Three individual cases worth watching

San Francisco's Brock Purdy has 56 dropbacks and is doing something with all of them. Purdy is at +0.503 EPA per dropback: 45-of-56 passing, 492 yards, five touchdowns, one interception, zero sacks — through games against the Rams and Dolphins. That's the second-best EPA/dropback mark in the league among the 32 qualifying quarterbacks (30+ dropbacks). Fifty-six dropbacks is not a season, but zero sacks is a distinctive early data point rather than a statistical illusion.

Josh Allen's numbers are gaudy because the volume is enormous. Allen has 65 dropbacks — 40-of-65 for 582 yards, five touchdowns, zero interceptions, four sacks, and +0.455 EPA per dropback. Two-week passing volume is partly a function of playing fast and often from ahead; Buffalo is doing plenty of both. The efficiency is real for the sample; the sustainability depends on whether the collapsing Buffalo defense forces the offense to keep bailing it out.

The bottom of the QB list is where the story is grim. Kirk Cousins (LV, +0.104 EPA/db, three interceptions on 61 dropbacks) is quietly effective; Cooper Rush (ATL, -0.635 EPA/db over 44 dropbacks, four interceptions) is the worst-graded qualifying passer in the league by a wide margin. Aaron Rodgers in Pittsburgh (-0.344 EPA/db, 85 dropbacks) is second-worst. These are the two teams with the largest offensive EPA regressions from 2025 in the league — Atlanta -0.381, Pittsburgh -0.324 — and the quarterback play is the through-line. Both figures come from decent-sized samples (44 and 85 dropbacks); "small sample" is a partial defense here, not a complete one.

4. Where the two-week evidence stands

San Francisco (2-0, +42): tentatively yes, with caveats. Elite on both sides across 111 offensive plays and 112 defensive plays, against an opponent slate that graded close to league-neutral. What would confirm it: a similar EPA profile against a defense that finished 2025 in the top ten. What would kill it: a Purdy sack rate that jumps toward league average once he plays a real pass rush.

Buffalo (2-0, +15): believe the offense, not yet the defense. The offensive step forward (+0.207 EPA/play over 2025) came against Houston's 2025 No. 1 defense and Detroit's top-12 unit — a real accomplishment, not a schedule gift. But Buffalo's own defense has fallen off by 0.155 EPA/play, the fourth-largest decline in the league. What would confirm the whole picture: the defense stabilizing anywhere near its 2025 form. What would kill it: repeating the 31-point, 41-point shootouts against teams that can actually finish drives.

Atlanta (0-2, -38) and Pittsburgh (1-1, -10): believe the offensive drop. Both are more than 0.30 EPA/play worse than 2025 on offense, both opponent slates graded ordinary on defense, and both stories run through quarterback EPA well below any qualifying passer's 2025 baseline. Two weeks is not a season, but a -0.38 offensive swing at 124 plays is not the sample being cute.

Cincinnati's defense (2-0, +20): interesting, not settled. A -0.352 EPA/play defensive improvement is the largest in the league, with the Jets not far behind. The evidence needed: keep this up against a team that ranked in the top half of 2025 offensive EPA, rather than Tampa Bay and Houston's below-average units.

Indianapolis (0-2, -21): the defensive collapse looks real; the offense less so. Their defensive EPA is worst in the league at 138 plays — the largest sample any team's defense has yet produced — and it's a swing of roughly 0.26 EPA/play from 2025. The offense is down only modestly (-0.026 vs. +0.072 in 2025), and Baltimore and Kansas City aren't a fair test for that side of the ball.

Chicago (1-1, +16): wait. A +0.144 EPA/play offense over 139 plays that looked unstoppable in one game and toothless in the other, against a defensive slate that graded roughly league-average in 2025. There isn't enough here yet to update on, in either direction.

Every EPA figure above comes from team-week aggregates through week 2 of 2026 and full-season 2025 aggregates for comparison. None of the 2026 samples exceeds 152 plays per team-side; treat every league rank cited as a two-week ranking, not a settled fact. Opponent adjustment here is done coarsely — average 2025 EPA of the two opponents faced — and is meant to flag distortions, not to fully correct them. Coaching, availability, and personnel changes since the offseason are outside the data this piece was written from.