NFL Poll of Polls: Week 5

Week 4 sorted nothing out. New England won in Buffalo, Indianapolis routed Washington, Atlanta hung 45 on New Orleans, and Cleveland beat Pittsburgh — and the top of the table changed hands for a third week running.

Kansas City is the new No. 1 — by 0.009 rank standard deviations over San Francisco — the third different team to hold the spot in three editions. Both are 4-0.

This is the same poll of polls as previous weeks: the week’s rankings standardised and pooled, every team’s move charted, and each team’s record alongside its ranking.

Here is where the panel landed after Week 4.

Kansas City takes the top spot from Buffalo, who fall to fifth after losing at home to New England. The margin over San Francisco is 0.009 — effectively nothing — and the two split the first-place votes: four outlets put the 49ers first, three the Chiefs, one Seattle.

That split is worth reading carefully. San Francisco wins more No. 1 votes and Kansas City still leads the consensus, because the Chiefs are never lower than fifth while the 49ers slip further on a couple of ballots. The consensus is a weighted average of 32 positions, not a vote count, so the two are best described as co-leaders.

One result the panel is arguing with: Minnesota is 4-0 and ranked eighth. The Vikings are unbeaten but have won by a combined 13 points, and the consensus scores them below several teams that have already lost.

This week’s matchups, priced

Fifteen games in Week 5: Kansas City and Carolina are on bye. Every game is priced below from the consensus alone, with a sandbox for building any matchup you like.

The panel

8 outlets this week. Mean distance from the consensus order, in rank positions:

Agreement is the second-tightest of the season (mean pairwise Spearman ρ = 0.932). USA Today sits furthest from the consensus at 2.81 positions (ρ = 0.923); every other outlet is within 1.94.

Sources and method

Every number above comes from one pipeline. Each outlet’s weekly 1–32 ballot is standardised, pooled in a measurement-error model, shrunk toward the panel mean, and published with a range reflecting how much the outlets actually disagree. Records are the standings each team carried into the week, computed from the same results feed. The matchups are priced from the consensus score alone, using constants fitted on 6,719 completed regular-season games.

The method, the calibration and the week-by-week replication checklist live with the code rather than in every post:

• analysis/nfl-power-rankings/SOURCES.md — one section per outlet: URL pattern, cadence, markup and the specific traps. Follow this file to reproduce a week.
• analysis/nfl-power-rankings/METHODOLOGY.md — both derivations, and what the method does not claim.
• analysis/nfl-power-rankings/README.md — the pipeline and how to run it.

This week’s panel — eight ballots, every one verified complete:

Two traps, and the guard they prompted

Sourcing this edition turned up two ways a pipeline can quietly publish the wrong numbers. Both are now defended against.

A current-looking URL serving an old edition. NFL.com’s Reynolds columns exist at two slugs. The one ending in a year (…week-4-power-rankings-2026) is the real article; the one without a year returns HTTP 200 with a 2020 page — Aaron Rodgers’ 4-0 Packers, COVID protocols, “Washington Football Team”. A parser cannot tell the difference, because the markup has the same shape. The pipeline now reads each page’s own datePublished / dateModified and refuses any ballot whose date falls outside the calendar window for that NFL week. The 2020 page fails that check and is dropped with a reason, rather than averaged into a 2026 consensus.

An identically titled article from the previous season. Bleacher Report’s 2025 Week 5 rankings carry exactly the same headline as the 2026 edition at a different article id. The freshness gate catches that too.

Neither trap is hypothetical. Both were found while sourcing this week.

And one access change

USA Today now returns HTTP 402 Access Restricted to ordinary browser user agents — the same URL pattern that worked for weeks 2–4. It serves normally to a Googlebot user agent, which is how the pipeline now fetches it. Worth stating plainly rather than hiding: the ballot is real and complete, but reading it requires presenting as a crawler, and if that stops working Nate Davis leaves the panel.

NFL.com’s own column has now gone unpublished for two consecutive weeks, and Neil Reynolds’ column did not appear either. Neither is a parsing failure. ESPN remains absent permanently: every URL on the site returns an empty HTTP 202 to a plain request.