Table of Contents

1. What we think we know after three weeks
2. The biggest movers against their 2025 baselines
3. Believe it or not yet

1. What we think we know

1) Minnesota's defense has been dominant, and it has been dominant for three straight weeks.

The Vikings' defensive EPA/play through 200 plays is −0.181 — best in the league by any measure of the 32-team pool. What matters is not the pooled number but the split: they were at −0.175 in week 1 (67 plays), −0.125 in week 2 (69 plays), and −0.246 in week 3 (64 plays). All three weeks are strongly negative. Compare that to 2025, when Minnesota's full-season defensive EPA/play was −0.101 across 1,016 plays — already a good defense. The 2026 line is a step better, and it hasn't been carried by one game. That's the shape of a real signal at a small sample: stability, not magnitude.

Caveat, and it belongs in the same sentence: Minnesota's own offense has been bad (−0.119 EPA/play across 165 plays), which means opponents have often been playing from ahead and running the ball. Some of that pretty defensive line is opponent posture, not talent.

2) San Francisco's passing offense is running at a level it did not reach in 2025.

Brock Purdy, through 83 dropbacks, has 789 passing yards, 9 touchdowns, 1 interception, and a +6.28 CPOE. The 49ers' offensive EPA/play sits at +0.357 — first in the league — with pass EPA of +0.616. In 2025 the same team ran a +0.061 offensive EPA/play across 1,180 plays and a pass EPA of +0.125. This is not a small move at a small sample; this is a very large move that has grown week-over-week. Week 1: +0.18 EPA/play. Week 2: +0.46. Week 3: +0.49. But three weeks and 83 dropbacks is still 83 dropbacks. The pattern is real; whether the level holds is what the next month decides.

3) Dallas's defense has been the worst unit in football through three weeks — and it was terrible last year too.

Dallas has allowed +0.191 EPA/play across 190 defensive plays — dead last of 32. Pass EPA allowed: +0.408. Those numbers are grotesque, but they are also not a departure. In 2025 Dallas finished with a +0.167 defensive EPA/play across 1,027 plays and a pass EPA allowed of +0.221. The 2026 three-week figure is worse, but it is worse in the same direction. This is the "consistent with the last three years" finding — not a new story, a continuing one, and week 3 (67 plays, +0.596 pass EPA allowed to Baltimore) is the newest data point in a pattern that predates it.

4) Miami's pass defense is showing the same shape as Dallas's, and it fits their 2025 trajectory.

Miami's defensive EPA/play through 160 plays is +0.162, second-worst in the league. Their pass EPA allowed is +0.482. In 2025 Miami finished at +0.088 defensive EPA/play and +0.126 pass EPA allowed across 1,015 plays. So Miami is worse than last year, but 2025 was already the wrong direction. The three-week sample here is arguing for continued decline more than a sudden collapse.

5) Seattle's defense held for two weeks, then leaked in week 3.

The Seahawks' pooled defensive EPA/play is −0.156 (177 plays) — third in the league. But the split matters: week 1 was −0.110 (67 plays vs. New England), week 2 was −0.325 (47 plays vs. Pittsburgh), and week 3 was −0.080 with a +0.117 pass EPA allowed (63 plays vs. Washington). The pooled ranking is being carried by weeks 1 and 2. It is exactly the situation the sample-size discipline exists for: a season-to-date ranking that looks authoritative and reflects two games. Seattle was −0.123 defensive EPA/play across 1,244 plays in 2025, so the underlying prior is strong, but the week-3 slippage is worth flagging rather than smoothing away.

6) Green Bay cannot run the ball, in a way that is entirely new.

The Packers have a −0.410 rushing EPA/play across their non-pass plays — the worst rushing offense in the league by a wide margin, and they have run only 48 non-pass plays out of 180. In 2025 Green Bay was −0.015 on the ground across 1,063 offensive plays. That's not a small move; it's a total inversion. Whether it survives one healthy back returning is another question, but the current split is unrecognizable from last year.

2. The biggest movers against their 2025 baselines

Here the question is not "who is good" but "who moved," and by how much across three weeks.

San Francisco offense: +0.061 in 2025 → +0.357 through 160 plays in 2026. A ~+0.30 EPA/play swing. Pass EPA jumped from +0.125 to +0.616. This is the largest positive offensive move in the sample, and it has grown weekly rather than fading. Purdy's +6.28 CPOE says the ball is being placed better than an average QB would place it, which is a stickier signal than raw yards.

Dallas offense: +0.098 in 2025 → +0.175 through 178 plays in 2026. Improvement, driven almost entirely by the pass (+0.311 vs. +0.162 last year). The rush game is at −0.039. Dak Prescott's early return of production is real, but Dallas playing catch-up behind their own defense inflates pass volume and, therefore, pass EPA opportunity. That is a real opponent/game-script distortion to name.

Tampa Bay offense: −0.009 in 2025 → −0.193 through 177 plays. The single largest negative offensive move in the league. Pass EPA has gone from −0.026 to −0.293 across 112 pass plays. Three weeks; the shape held in a week-3 loss to New Orleans as well as two prior games. Whether this is Baker Mayfield or the pieces around him is a question the sample can't answer yet.

Minnesota offense: −0.116 in 2025 → −0.119 through 165 plays. Essentially unchanged. This matters because the defense is the story here — the offense is, so far, running last year's line. Minnesota being 2–1 despite a bottom-third offense is a defensive story.

Arizona defense: +0.056 in 2025 → +0.123 through 167 plays. Pass EPA allowed is at +0.398 — a very ugly early number, up from +0.150. This is a real move against a mediocre baseline, but the sample is small and week-to-week stability was not something I was able to characterize in the time available; treat it as a flag, not a finding.

Detroit defense: −0.011 in 2025 → +0.149 through 215 plays. A meaningful decline, and Detroit has faced more defensive plays than any team in the league (215) which suggests they've been on the field a lot. Whether that's the defense creating short possessions or the offense doing so is not something I can separate from the pooled number alone.

3. Believe it or not yet

The point of this pass is to judge evidence strength, not to guess future records.

• Minnesota's defense is elite. Believe it, provisionally. Stability across three weeks + a strong 2025 baseline (−0.101 EPA/play, 1,016 plays) = a prior that reinforces the current sample. What would kill the read: one of the next two weeks landing above 0.0 on defensive EPA/play against an average offense.
• San Francisco's passing game is a top-of-league unit. Believe it, but hold the ceiling loosely. +6.28 CPOE and a growing weekly trend are real. But 83 dropbacks is 83 dropbacks, and 2025 was only a middle-of-the-pack pass offense (+0.125 pass EPA over 1,180 plays). What would confirm: a month at +0.30 pass EPA or better. What would kill it: CPOE regressing to league average and pass EPA/play halving.
• Dallas's defense is broken. Believe it. This is the strongest early-season update in the piece precisely because it doesn't require an update — it's a continuation. The 2025 baseline (+0.167 defensive EPA/play, 1,027 plays) was already the worst-decile shape, and 2026 is worse in the same direction. Not evidence of a new problem; evidence that last year's problem persists.
• Seattle's defense is top-tier. Not yet. The pooled −0.156 sits on 177 plays, and the week-3 pass EPA allowed (+0.117) is the first crack. Prior (−0.123 in 2025 over 1,244 plays) supports the read, but I'd want two more weeks below −0.05 before treating this as a settled unit rather than a good defense that had two very good matchups.
• Green Bay cannot run. Not yet, mechanically. −0.410 rushing EPA/play across 48 rush attempts is a very small denominator on a metric with high variance. The direction is clear; the magnitude almost certainly is not. What would confirm: another 60–80 attempts at anything below −0.15.
• Tampa Bay's offense has fallen off. Not yet. −0.193 EPA/play over 177 plays is a real move against a −0.009 baseline, but three games is exactly the horizon at which a bad quarterback stretch and a genuine offensive decline are indistinguishable. Another two weeks resolves this one either way.