Two Kansas City Chiefs seasons, side by side.
2024: 15-2. Eleven of those were one-score games — decided by eight points or fewer: a fourth-quarter drive, a bouncing fumble. Kansas City won all eleven. Not ten of eleven. Eleven.
2025: 6-11. Same head coach. Same quarterback for fourteen of the seventeen games. One-score record: 1-9.
Nine wins, gone, with most of the roster still in the building. Almost nobody's record is the one they earned. The question is how much of it was borrowed.
The number with no memory
Take every NFL team from 2016 through 2025, line each season against the one that followed — 288 pairs — and ask one question of each line in the standings: does it come back next year?
Some of it does. Point differential — how many more points you scored than you allowed, per game — returns at a correlation of 0.408. On a scale where 1.0 means a season repeats itself perfectly and 0 means it tells you nothing: 0.408 is not destiny, but it's the most repeatable result on the board, and we'll get to the one thing that beats it. How efficiently an offense moves the ball, snap to snap, comes back at 0.389.
A team's record in one-score games comes back at 0.104.
That is, functionally, nothing — the weakest number in the entire table, weaker than turnover margin, weaker than third down, weaker than anything a defense does. Point differential repeats nearly four times as strongly.
Which is awkward, because it's the first thing anyone reaches for to explain a season. They know how to win the tight ones. That team just finds a way.
The 2025 Chiefs didn't forget how to win close games. There was never much of a how to forget.
So who's holding a borrowed record right now?
Denver went 14-3 in 2025. By points scored and allowed, the Broncos played like an 11-win team — a three-win gap, the largest in the league.
The mechanism is the one you'd guess: 11-2 in one-score games. Since 2016, exactly three teams have won eleven one-score games in a season. Minnesota did it in 2022, went 13-4, then 7-10. Kansas City did it in 2024, went 15-2, then 6-11. Denver is the third.
Two cases isn't a law, so here's the larger sample. Of the 62 team-seasons since 2016 that won at least 65% of their one-score games, the average change the next season was −3.02 wins, and 11% improved at all. A separate cut of the same data: the 47 team-seasons finishing more than a win and a half above their point differential averaged −2.86, with 17% improving.
Now the honest part, and it matters more than the scary number: none of this says Denver is a bad football team. Eleven wins is good — a playoff team on merit, with the sixth-most efficient defense in football last season. The claim is narrower and duller than "Denver is a fraud": 14-3 was three wins better than Denver played, and the league's history with that gap is unkind.
Turnovers: the most borrowed number in football
Chicago finished 2025 with a turnover margin of +22, the best in the NFL, on a league-leading 33 takeaways. The Bears went 11-6.
Here's what tends to happen next. Take the teams that finished a season in the top five in turnover margin and check on them a year later — 42 such cases since 2016. Their average margin in the season they did it: +12.2. The very next season: +2.6. Not bad. Just ordinary. Wins follow it down by about 1.5.
It runs the other way just as hard — the same exercise at the bottom, 42 more cases: −14.1, then −0.7 the following year, and 2.7 wins picked up along the way. The Jets finished 2025 at −18, the worst mark in football — a number with a strong historical habit of leaving.
Turnover margin's own correlation is 0.164: a shade stickier than close games, nowhere near enough to build a season on. Takeaways are real events that decided real games. They are just not a stable property of a team, and mistaking a spike for an identity is how you end up surprised in October.
Which is the useful lens on Chicago: 33 takeaways, first in the league — and 19th in defensive efficiency. Both true at once, and the takeaway number is the more fragile (0.164 against 0.262).
Third down looks like a skill. Most of it isn't.
San Francisco converted 50.7% of its third downs in 2025, the best rate in football. That feels like a property a team has: poise, a play-caller, a quarterback who extends plays.
Third-down conversion rate does repeat, at 0.387. But strip out what a team's overall offensive efficiency already predicts, and the genuinely third-down-specific part — the clutch residue — repeats at 0.142.
Translated: good offenses are good on third down because they're good on every down. The sliver that makes a broadcast call a team money on third down mostly doesn't survive the winter.
What does repeat is the thing nobody films
Of every measure in that ten-year table, the one that returns most reliably from season to season isn't a result at all. It's how often a team chooses to pass, relative to what the down, distance and score called for — and it repeats at 0.455, higher than wins, point differential, offense, defense, turnovers or third down.
Not how well they threw it. How often they decided to.
The most predictable thing about an NFL team isn't talent or record. It's its temperament — what it wants to do, year after year. Which is last week's lesson from the other direction: the defense defends what it sees, and what a team chooses to show is the most durable thing about it.
Records get borrowed. Turnovers get borrowed. Close games are almost entirely borrowed. Intent is the part you own.
Real or Noise
The verdicts, calibrated. These grade the story, not the team.
- "Denver is a legitimate 14-3 team": leaning noise. Three wins clear of their point differential, the largest gap in the league, on an 11-2 one-score record. What's real: the eleven-win team underneath. What was borrowed: everything above it.
- "The Bears' takeaway defense is who they are": leaning noise. A league-best +22 margin and 33 takeaways, alongside a 19th-ranked defense by efficiency. Top-five turnover teams since 2016 slide from +12.2 to +2.6 the next season across 42 cases. The takeaways were real. The identity is on loan.
- "Kansas City's 6-11 was worse than Kansas City actually was": leaning real, with an asterisk. Their point differential said 9.5 wins — the largest understatement in the league, built on that 1-9 one-score record. The asterisk cuts both ways: Mahomes tore an ACL after fourteen games, so part of that 1-9 was played without him — and the 9.5 covers all seventeen, the three he missed included.
The Receipts
Verified numbers worth keeping in your pocket.
- A team's record in one-score games is the least dependable figure in the standings: it repeats year over year at 0.104 across 288 back-to-back season pairs, 2016 to 2025. Point differential per game, for comparison, comes back at 0.408.
- Winning the close ones has a bill attached. The 62 team-seasons since 2016 that won at least 65% of their one-score games averaged −3.02 wins the following season. Eleven percent of them improved.
- The stickiest team number in football is a decision, not a result. Pass rate over expectation — how often a team throws relative to what the situation called for — repeats at 0.455, ahead of point differential (0.408) and everything else on the board.
Where to hear it
Muffed is an AI that reads every play, every box score and every advanced stat from all 32 teams — and tells you the real story of the game. Every number above was checked against the source data before it shipped. That's the house rule, and the receipts are in the footer.
The audio version of any team's season is one click away: Kansas City, Denver, Chicago, San Francisco, New York Jets. During the season, Muffed builds a weekly show for your favorite team or your fantasy roster — muffed.ai.
And if someone forwarded you this: the Read is free, every Wednesday, and it's better when it just shows up. Subscribe and it will.
Next week: availability — five seasons of who actually takes the field, because the oldest cliché in the sport has a number attached.
Methodology: year-over-year correlations computed on 288 consecutive-season team pairs, 2016–2025 regular seasons; one-score games are those decided by eight points or fewer; expected wins from the Pythagorean estimate (exponent 2.37). Regression base rates are historical averages over the stated samples, not forecasts. Offensive and defensive efficiency are expected points added (EPA) per play — a measure that values every snap by how much it moved the scoring math — regular season, run and pass plays only; the year-over-year correlations use a slightly broader play selection, which moves them by 0.003. Pass rate over expectation compares actual pass frequency with the rate expected given down, distance, score and time. Data via nflverse.