Do NFL defense funnels hold up from week to week?
Mostly not, in our 2026 count. Across 12 position allowances and all 32 defences, the rank correlation between weeks 1 to 3 and weeks 4 to 6 ran from -0.06 to 0.30, with a median of 0.04, where 1 would mean the same order both times and 0 no relation. Of the 101 defences in a stat's top eight after week 3, 31 were in that stat's top eight again over weeks 4 to 6. Three games a side is a small sample.
A funnel is a defence that has given up an unusual amount to one position: running backs, wide receivers or tight ends. Three weeks into a season the numbers can look loud. We wanted to know how much of what each defence allowed in weeks 1 to 3 of the 2026 regular season (September 9 to 28) carried into weeks 4 to 6 (October 1 to 19).
In this count, very little did.
The split
For every one of the 32 defences, we took the same per-game allowances the funnels page and the defense vs position tables show and computed them twice: once over the defence’s own games in weeks 1 to 3, and once over its own games in weeks 4 to 6. Every defence played three games in the first half. In the second, Carolina, Cincinnati, Detroit, Kansas City, Miami and Minnesota had a bye and played two; everyone else played three.
Then we ranked the 32 defences in each half (rank 1 allows the most, as on the Lab) and asked two questions per stat:
- Rank correlation: how similar the two orders are. 1 means the same order both times, 0 means no relation, and a negative number means the order partly flipped.
- Top eight repeated: of the defences in the top eight after week 3 (more than eight where defences tied for eighth), how many were in the top eight again over weeks 4 to 6.
What carried over, stat by stat
| Allowance, per game | Rank correlation | Top eight after week 3, top eight again | League average, weeks 1-3 | League average, weeks 4-6 |
|---|---|---|---|---|
| Targets to RBs | -0.05 | 3 of 8 | 5.5 | 5.9 |
| Receptions to RBs | -0.06 | 3 of 9 | 4.4 | 4.4 |
| Receiving yards to RBs | 0.09 | 2 of 8 | 30.4 | 30.8 |
| Targets to WRs | 0.04 | 2 of 8 | 18.2 | 19.5 |
| Receptions to WRs | 0.30 | 5 of 11 | 11.3 | 11.9 |
| Receiving yards to WRs | 0.12 | 3 of 8 | 148.3 | 153.1 |
| Targets to TEs | 0.04 | 1 of 8 | 7.2 | 8.3 |
| Receptions to TEs | 0.12 | 2 of 8 | 5.2 | 5.8 |
| Receiving yards to TEs | 0.19 | 4 of 8 | 52.5 | 55.3 |
| RB carries | -0.01 | 3 of 9 | 22.2 | 21.6 |
| RB rushing yards | -0.04 | 1 of 8 | 91.6 | 87.4 |
| QB passing yards | -0.01 | 2 of 8 | 231.3 | 239.4 |
The 12 correlations run from -0.06 to 0.30, with a median of 0.04. Eleven of the 12 are under 0.20. Across all 12 stats, 31 of the 101 top-eight places after week 3 were repeated over weeks 4 to 6. If the two halves had nothing to do with each other, about one in four would repeat by chance, since eight of 32 places is a quarter.
Receptions to wide receivers is the only stat where the early order kept some shape. The run game kept none: the defences that allowed the most RB rushing yards in September were, as a group, no more likely to do it in October.
The five biggest funnels after week 3
These are the five defences whose largest allowance after week 3 sat furthest above that stat’s league average, one stat per defence, with where each stood over weeks 4 to 6 and over the whole season through week 6 (the figure the Lab shows today):
| Defence and allowance | Weeks 1-3 | Rank | Weeks 4-6 | Rank | Weeks 1-6 | Rank |
|---|---|---|---|---|---|---|
| Detroit, receiving yards to TEs | 117.0 (123% above average) | 1 | 97.0 (2 games) | 3 | 109.0 | 1 |
| Cincinnati, receiving yards to TEs | 99.7 (90% above) | 2 | 76.0 (2 games) | 8 | 90.2 | 2 |
| New Orleans, receptions to TEs | 8.3 (59% above) | 2 | 5.7 | 14 | 7.0 | 5 |
| Carolina, RB rushing yards | 144.7 (58% above) | 1 | 40.5 (2 games) | 32 | 103.0 | 10 |
| Buffalo, receiving yards to RBs | 45.0 (48% above) | 2 | 49.7 | 3 | 47.3 | 1 |
Two held close to the top (Detroit and Buffalo), one stayed in the top eight at a lower level (Cincinnati), and two faded or flipped (New Orleans and Carolina). Carolina went from the most RB rushing yards allowed to the fewest. We do not know why any of them moved: opponents, injuries, game scripts and scheme all change in three weeks, and this count holds none of them constant. The cause is unknown.
Detroit is the one defence whose tight end numbers stayed near the top across all three tight end stats: first in receptions to tight ends in both halves, and first then second in targets to tight ends, at 13.0 a game both times.
What this does and does not say
Three games a side is a small sample. A single long touchdown or one opponent with a star tight end moves a three-game average a long way, and a bye makes it two. This count shows that in 2026, a defence’s rank after week 3 told us little about its rank over the next three weeks. It does not say a defence’s allowances mean nothing, it does not call any defence good or bad, and it gives no betting rule.
The same caution applies to a player’s own numbers early in a season; our page on whether NFL hit rates can be trusted in September looks at that side.
How we counted
The allowances use the Lab’s own definitions, not a new calculation: the position splits (targets, receptions and receiving yards to RBs, WRs and TEs, plus RB carries and rushing yards) come from the same play-by-play table that feeds the funnels and defense pages, with fullbacks counted as running backs, and QB passing yards from the same box score table. Regular season only, each defence’s own games. Run over weeks 1 to 6 together, our figures equal all 352 play-by-play cells of the Lab’s 2026 defence table (11 stats for 32 defences). The play-by-play, roster and game logs were frozen on October 20, 2026, after the Monday night game on October 19 completed week 6.
Where to find these
The funnels page shows this week’s matchups with each defence’s allowances, and the defense vs position tables show every defence by position, each with its game count. Read the count beside the number: through week 6, every defence has five games or six, and the Lab board has the player lines those allowances sit under.
Quick answers
Which early funnel held up best?
Receptions to wide receivers, with a rank correlation of 0.30 between weeks 1 to 3 and weeks 4 to 6 of 2026. Five of the 11 defences in its top eight after week 3 (ties included) were in it again over weeks 4 to 6. It was the only one of the 12 stats at 0.20 or higher.
Did the running back funnels hold up?
Not in this count. RB rushing yards allowed had a rank correlation of -0.04 and RB carries -0.01. One of the eight defences in the rushing yards top eight after week 3, New Orleans, was in it again over weeks 4 to 6. Carolina allowed the most RB rushing yards in weeks 1 to 3, 144.7 a game, and the fewest in weeks 4 to 6, 40.5 over two games.
Did any defence keep its funnel?
Detroit's tight end numbers stayed near the top: first in receptions to tight ends in both halves (9.7 a game, then 11.0), first then third in receiving yards to tight ends (117.0, then 97.0) and first then second in targets (13.0 both times). Detroit had a bye, so its weeks 4 to 6 figures are two games. The cause is unknown.
How were the allowances counted?
With the Lab's own definitions: the position splits come from the same play-by-play calculation that feeds the funnels and defense pages, and QB passing yards from the same box score table. Run over weeks 1 to 6 together, our figures equal all 352 play-by-play cells of the Lab's 2026 table, 11 stats for 32 defences.
Does this mean I should ignore a defence's numbers?
No. It says three games of allowances, on their own, told us little about the next three in 2026. A defence's numbers describe what it allowed; they do not say why, or what a book has already priced. This count gives no betting rule.
Want tonight's board?
The Lab has the full slate: every line with L5, L10, L20 and season hit rates, game logs, matchup context and the best price at your books. A short daily card of picks is posted before the games and graded in public afterwards, win or lose. The Lab is free, and it stays free.