Plain-English guides to the NFL's most important advanced metrics — the formulas, what they reveal, and the caveats.
By C. B. Zakarian · Updated October 22, 2026
Every advanced NFL number you hear on a broadcast has a real definition, a formula, and a set of caveats. These explainers build each one up from scratch — no jargon, no invented statistics, just how the metric works and how to read it. Latest added October 22, 2026.
Bigger favorites win more often straight up — but their cover rate sits right around 50% at every spread size, from pick'em to 12-plus. The clean signature of an efficient market, 1999-2025.
Scoring peaks in Week 5 (45.4 points) and bottoms out in Week 11 (42.75) as the weather turns, with Week 1 and Week 18 also low. The weekly scoring curve, 1999-2025.
23-20 leads the way, and the three most common finals are all 3-point games. Why a handful of low scores built from field goals and touchdowns dominate, across 7,261 games (1999-2025).
A pick'em favorite wins just 53% of the time; a touchdown favorite 73%; even 12-point favorites lose 1 in 9. The straight-up win curve by spread size, 1999-2025.
Home-field advantage jumps from 56% to 62% in the postseason (69% in the Divisional round), overtime nearly doubles, and scoring doesn't drop. Why the bracket amplifies the home edge.
Home-field advantage is smaller against division opponents who play in your building twice a year. The 27-season split, and why familiarity blunts the home edge.
How football's opponent-adjusted efficiency metric works — and how to read a number that lives on a percentage scale.
Three efficiency metrics, one decision: what each measures, where they agree, and why they sometimes disagree.
How tracking data turns every throw into a probability — and isolates a quarterback's accuracy from scheme.
The same quarterback, two scales: how each rating is built and what each one actually rewards.
Why points scored and allowed predict next season better than this season's win-loss record.
Breaking a pass-catcher's production into opportunity, depth of target, and yards after the catch.
What win-probability math really recommends on fourth down — and why teams still punt too much.
Opponent adjustment and the Simple Rating System, so you can compare teams on a level field.
One formula that rewards touchdowns, punishes interceptions, counts sacks — and predicts winning better than passer rating.
Raw rushing yards reward blocking and volume. How tracking data isolates what the back actually added.
Why pressure rate beats sack totals, what win rates measure, and the hard problem of dividing sack blame.
Interceptions lie and good corners get avoided. The stats that actually measure coverage — and why defense is so hard to grade.
Field goals over expected, the points hiding in field position, and why the kicking game still swings outcomes.
Leading teams run, trailing teams pass — so raw run/pass splits lie. PROE strips out game script to reveal a coach's true tendencies.
Moving the ball is easy between the 20s; the field shrinks inside them. How to measure who actually finishes drives.
A handful of chunk plays drive most scoring drives. What counts as explosive, and why explosive-play rate predicts winning.
Third down is where drives live or die — but raw conversion rate hides how hard the down-and-distance really was.
Winning teams hold the ball because they're winning — not the other way around. Why TOP is mostly a symptom, not a cause.
Turnover margin decides games and barely predicts them. The fumble-recovery coin flip and why turnover luck regresses hard.
That live win-probability percentage is a model output. How it's built, what leverage means, and how WPA assigns credit for the swing.
Not all YAC is the receiver. How tracking data models expected yards after catch and separation to credit the right player.
Opportunity comes before production. How snap share, route participation, and touch share reveal a player's real role.
Per-game points reward pace and per-play hides finishing. Why the drive is often the right unit of offense.
A yes/no verdict on every play - 40% of the yards on 1st down, 60% on 2nd, 100% on 3rd and 4th - that measures how often an offense stays on schedule.
Sacks are rare and lucky; pressures are frequent and stable. Counting sacks, hits, and hurries per dropback measures a pass rush far better.
Faking the run reliably boosts yards per attempt and EPA - and, counterintuitively, the boost barely depends on whether you actually run the ball well.
The yards a back gains past first contact isolate his own power, balance, and tackle-breaking from the holes his line opened.
The QB reads one conflict defender after the snap and either hands off or throws - taking whatever the defense gives, bounded by the linemen-downfield rule.
Expected value sets the break-even rate versus kicking the PAT - but late in close games, win probability and the trailing-team math flip the call.
Dividing receiving yards by routes run - not targets or catches - credits a receiver for every route he runs, making YPRR the closest thing to a true per-opportunity rate for pass-catchers.
Win rate charts whether a lineman wins his individual matchup within a time threshold on each snap - a process measure that complements the after-the-fact outcome of sacks and pressures.
Target share is a receiver's slice of the team's targets; WOPR combines it with air-yards share into one opportunity number that describes how central a receiver is to the passing game.
Across 7,182 games from 1999 to 2025, NFL home teams won 56.6% of the time - but the edge has fallen from the high-50s of the 2000s to the low-50s today. The real data, the trend, and why the cause is so hard to pin down.
Before a play even happens, a situation already has a worth in points. Expected points puts a number on it - and that number is the foundation under EPA, 4th-down math, and the 2-point decision.
Across 7,276 played NFL games with a closing line, the spread's average error is +0.09 points - essentially unbiased - yet individual games scatter by a standard deviation of 13.2 points. The spread nails the average margin and misses almost every single game. The real data, the worked math, and why a ~50% cover rate still loses to the vig.
Across 7,276 games from 1999 to 2025, NFL scoring barely budges with temperature - a freezing game averages 43.4 points, a balmy one 43.0 - but it falls steadily with wind, from 44.7 points in calm air to 40.5 in a stiff breeze, while domes top everything at 46.9. The real data, why wind is the culprit, and the limits.
Across 7,276 games from 1999 to 2025, a small rest gap does nothing - but a big one is worth a field goal. A bye-week-sized home rest edge pushes the home margin from +2.3 to +4.1 points and the win rate to 62%, while a rested visitor erases home-field almost entirely. The real data, the bye-week math, and the limits.
Average points per game climbed from 41.7 in 1999 to 46.0 in 2025 - about four points - peaking at 49.5 in the empty-stadium 2020 season. The rule changes that drove it, and how the betting market's over/under line tracked the rise (lagging it by about 0.7 points).
Only 6.1% of games reach overtime across 1999-2025, but the 2017 rule shortening OT to 10 minutes more than doubled the share of OT games ending in a tie (2.3% to 5.7%). The real data on overtime frequency, ties, and why home-field advantage fades when the game goes long.
Across 7,276 games with a closing total, the over/under's average error is +0.70 points - nearly unbiased, median exactly zero - yet individual games scatter by a standard deviation of 13.4 points. The total nails the average and misses almost every single game. Over hit 48.7%, under 49.8%, and why ~50% still loses to the vig.
Across 7,276 games, 15.1% are decided by exactly 3 points and 9.0% by exactly 7 - nearly a quarter of all games on those two numbers. The margin-of-victory distribution is a jagged skyline that spikes at every combination of field goals and touchdowns, which is why bettors pay a premium to move a spread off 3.
Across 6,967 regular-season games from 1999 to 2025, Thursday games average 45.0 points to Sunday's 44.1 and post the same 11.5-point margins - so the 'TNF is bad football' reputation is not in the score. The real data, why the short week doesn't dent scoring, and the limits.
Turf games average 45.5 points to grass's 43.3 across 7,276 games - but almost the whole 2.2-point gap is the dome, not the surface. Compared within the same environment the gap collapses to under a point. A textbook confound, worked through with the real data.
Across 7,276 games from 1999 to 2025, 50.7% end within one score (margin of 8 or fewer) and only 26.9% are blowouts - and the one-score share has risen from 49.9% to 52.7% while blowouts fell. Why football stays close, and why parity is quietly improving.
Across 7,276 games, the venue swings scoring by about 15 points: retractable-roof AT&T Stadium averages 52.1 combined points and lakefront Cleveland Browns Stadium just 36.9, against a 44.3 league average. The Cowboys' move from open-air Texas Stadium to a dome is the natural experiment. The real leaderboard, the mechanism, and the limits.
Raw crew scoring averages spread 5.1 points — until you adjust for the seasons each referee worked. Era-adjusted, zero of the 21 highest-volume crews clear two standard errors on totals, and the home-whistle version barely beats a coin. The folklore, the fix, and the honest scope of what final scores can test.
One-starter seasons win at .575; three-starter seasons at .357 — a 3.7-win gap across 6,967 games since 1999. But only 39% of team-seasons keep one starter, the share is falling, and the causality runs both ways: benchings are a symptom of losing at least as much as instability is a cause.
Teams that win their opener make the playoffs 53.3% of the time; teams that lose it, 25.3% — a real doubling across 861 team-seasons since 1999. But 43% of the famous record gap is the opener itself sitting in the record, and game 1 predicts the rest of the season no better than any other single game (r = +0.19 vs +0.24). The count, the placebo test, and the honest read.
Across 7,276 games from 1999 to 2025, 187 ended in a shutout. The rate halved from 3.5% of games in the 2000s to 2.0% — then stopped falling, holding at 2.2% since 2017. The steadier decline belongs to the near-shutout: one side held to 6 or fewer slid from 13.5% of games to 8.9%. The 27-season count, the 2024 record low of two, the 2025 rebound of seven, and why finals alone can't name the rule change that did it.
Chicago's 2026 opponents went a combined 158-129-2 (.550) in 2025 — the hardest slate on paper — and Cleveland's 124-165 (.429), the easiest. But grade the same forecast across 2003-2025 and it barely predicts: correlation with the schedule teams actually faced is r = +0.18, the average miss is 3.5 points of W%, and the forecast-hardest schedule finished top-5 in reality just 5 times in 23 seasons. The 2009 Saints drew the 8th-toughest forecast, played the league's easiest schedule, and won the Super Bowl.
The 2026 slate is published, and before a down is played the calendar has picked sides: 69 of 272 games carry a rest gap of three or more days. Chicago drew the friendliest schedule (+15 net rest days, four big edges, zero deficits); the Chargers drew the harshest (-24, with four opponents arriving straight off byes). The full rest ledger, the bye map, the short weeks, a Wednesday opener — and 27 seasons of evidence on what a rest edge is actually worth.
Across 27 seasons, a team with at least a share of the league's best regular-season record won the Super Bowl 8 times — and lost it 12. Only the 2003 and 2016 Patriots finished the job outright, the median champion owned the league's 3rd-best record, the worse-record side won 14 of the 27 finals, and the most common fate of the paper No. 1 is a Divisional-round exit. The 9-7 Giants, the 16-0 Patriots, and the honest base rate.
Since 1999 the league has produced 3,174 win streaks and 3,180 losing streaks. Twelve win streaks reached 13 games — Indianapolis' 23 straight (2008-09) the longest — and ten skids reached 14, topped by Jacksonville's 20. Replay all 6,967 games as coin flips weighted only by the teams' actual records and the same monsters appear on their own: the simulated century's median longest win streak is exactly 23, while a fair-coin league produced a 23-gamer in zero of 1,000 tries. Streaks prove quality varies; they never needed momentum.
Across 184 title defenses since the 2002 realignment, the defending division winner repeated 79 times — 42.9% — and finished last 23 times. Worst-to-first struck in 20 of 23 seasons (15.5% of cellar teams jump to first), and matched against teams with the same record and no crown, winners added just +0.9 wins (±0.5) the next year — noise. New England's 11 straight AFC Easts, Kansas City's nine, and the 2025 purge that replaced seven of eight champions.
A team's one-score record correlates +0.08 with its own close-game record a year later — statistical zero — while point margin holds at +0.40 across 829 franchise season-pairs since 1999. The 140 teams that won 70%+ of their close games came back at .515; the 2024 Chiefs went 11-0, then 1-9; and once margin is known, close-game record adds nothing (t = −0.7). Denver's 14-3 rode an 11-2 close record — and 2026 opens Monday night at Kansas City.
Starting in 2026 this site publishes a win probability for every game — frozen before kickoff, graded after, misses kept on the page. The model is a four-part Elo anyone can rebuild: K=20, a 48-point home edge (waived at neutral sites), a margin multiplier, and 1/3 preseason regression. Walked forward through 4,363 games it hadn't seen (2010–2025) it called 64.7% of them with a Brier of 0.2205 — against 55.7% for always-home and 58.4% for picking the better record — and it will be wrong a game in three all season, in public.
The season opens across four days: a Wednesday Super Bowl rematch in Seattle, the league's first regular-season game in Melbourne, and five division games all kept out of the 1:00 window. The 16 pairings carry 393 prior regular-season meetings since 1999 — five annual series with 54 apiece, New England–Seattle just six — and the extremes run from Miami's 13-3 ledger over Las Vegas to Cleveland's 3-23-1 opener record. Every head-to-head computed from the file.
22 of 24 AFC East titles since 2002 went to New England (17) or Buffalo (5), with the champion repeating 78.3% of the time against a 42.9% league rate. Its 288 division games average a 13.08-point margin — widest of the eight divisions — with only 44.4% one-score finishes. New England went 81-21 in division play 2003–19, Buffalo 24-6 in 2020–24, and in 2025 the crown moved again on a 14-3, +170 season. All 12 of the 2026 round-robin games are Sunday-afternoon kickoffs, and only 4 land before week 10.
Since realignment, 52.1% of AFC North division games have finished within one score — 4th of 8 divisions, behind the NFC North's 54.9%. But Baltimore–Pittsburgh is the closest rivalry in football: 8.2 points of average margin and a 70.8% one-score rate, #1 of 48 pairs on both counts. Titles since 2002 run Pittsburgh 10, Baltimore 8, Cincinnati 6, Cleveland 0 — the league's only zero — and the 2025 crown turned on a 26–24 week-18 decider. The 2026 slate answers with five of twelve division games in weeks 15–18, including a New Year's Eve Thursday nighter in Cincinnati.
Twenty-four seasons, twelve changes of champion, and only one run longer than two — Indianapolis's 2003–07 five-peat. The 2025 ledger the division carries forward: Jacksonville 13-4 (+138), Houston 12-5 (+109), Indianapolis 8-9 despite +54 with a 3-7 one-score record, Tennessee 3-14 (−194) — the league's second-worst. All four clubs open 2026 at home on September 13, and the first division game waits until week 3.
Since 2002 the AFC West crown has changed hands in just nine reigns — Kansas City owns 11 of 24 titles including nine straight on a 112-36 run, until Denver's 14-3 in 2025 ended it outright. The rematch comes with 2026's meanest calendar: net rest of −24 for the Chargers (worst of all 32 teams), −13 for the Raiders, a division sum of −36 when six of eight divisions are positive. Derived game by game from the nflverse file, tiebreakers included.
For twenty straight seasons — 2005 through 2024 — the NFC East never crowned the same champion twice in a row, the longest no-repeat run of the eight-divisions-of-four era, until Philadelphia's 11-6 title defense ended it in 2025. This preview derives all 24 champions from 6,223 games and lays out the four 2025 ledgers: Dallas scored a division-high 471 points at 7-9-1, and the Giants went 1-7 in one-score games. Ahead: a 12-game 2026 division calendar with 7 games by week 9, two short-week Thursdays, and a −15 net-rest ledger for the champion.
For 21 seasons the head-to-head table never moved — Green Bay .675 inside the division (84-40-2, 2002–22), Detroit .349 at the bottom. Since 2023 it has inverted: Detroit .667, Green Bay .500, Chicago .278 — yet Chicago is the reigning champion at 11-6 despite going 2-4 in the division and 5-19 in it since 2022. Charted on a rolling four-season window from 288 nflverse games, plus the back-loaded 2026 round-robin: 7 of 12 division games from week 10 on.
Since the 2002 realignment the league has crowned five strictly sub-.500 division champions — three of them NFC South: 2014 Carolina (7-8-1), 2022 Tampa Bay (8-9), and 2025 Carolina (8-9), a three-way tie broken on head-to-head. The division also holds three of the six worst champion point differentials (−69, −45, −35) and back-to-back four-peats, New Orleans 2017–2020 and Tampa Bay 2021–2024. What 6,223 games say about the ugliest title in football, and the shape of the 2026 race.
Since 2002 the NFC West crown split Seattle 10, San Francisco 6, Rams 5, Arizona 3 — and changed hands in 15 of 23 transitions, with no reign longer than four years. Its last three titles went to three different clubs, the only division where that's true. A 2026 preview built on the crown strip, the 288-game round-robin ledger, and a schedule that opens in Melbourne and ends in a week-18 doubleheader — with no predictions of its own.
Across 320 team-seasons from 2015 to 2025, preseason win% correlates with the regular season that follows at r = +0.10 — 1% of the variance — while last year's record, on the same teams, manages +0.35. The 39 unbeaten-preseason teams finished .466, below average, and the 38 winless ones .472; since the 2021 cut to three games the correlation is +0.008, statistical zero. The 2017 Browns went 4-0 in August and 0-16 in the fall. A new 567-game ESPN-sourced dataset, the anchor comparison, and why the site's model prices August at nothing.
A dated snapshot, computed August 16, 2026: 17 completed games averaging 35.6 combined points — under the full 2025 preseason's 41.0 and far under the 46.0 of the 2025 regular season. Carolina beat Arizona 33-30 in the Hall of Fame opener, Colts–Patriots ended 13-13 (the dataset's eighth tie, all eight since the 2021 overtime removal), listed road teams are 10-6-1, and zero of the 16 regular-season Week 1 pairings have met in August. The scoreboard so far, the eleven-preseason scoring context, and no predictions — by design.
Sum the model's per-game win probabilities over each team's actual 17-game slate and the frozen opening ratings expect Seattle to win 12.2 games, Las Vegas 4.9 — a spread of just 7.3 wins, regression built in. Schedules move the middle: Houston starts fifth in rating but second in expectation, Buffalo second in rating but fourth; Cleveland's slate is worth +0.75 wins, the Chargers' costs 0.52. Every number derives from the published ratings and the schedule file, the honest error bar is ±2 wins of pure coin-flip noise, and the prediction ledger grades the same ratings weekly all season.
Simulate the published 272-game schedule 20,000 times at the ledger's own win probabilities and count: Seattle reaches the field in 94% of seasons and takes the NFC's top seed in 40%; Houston (85%) and Denver (84%) lead the AFC; Kansas City sits at 38% — the frozen ratings' coldest take, on the record. The Rams pay the division tax (77% playoff, 26% NFC West), Tampa Bay's soft division buys Green Bay's odds on half a win less, and the average last team in wins 9.3-9.5 games. Ties break at random — a stated simplification — and the ledger grades all of it weekly.
Eight 2026 games are played abroad with a home team on the label and none on the field, and the ledger waives its +48 Elo edge on every one: a haircut of 6.6 to 6.9 percentage points per game, 0.543 expected wins redistributed, and four designated hosts — Washington, Jacksonville, Atlanta and Detroit — who stop being favorites. History backs the waiver: designated home teams are 30-35-1 in neutral-site regular-season games since 2003, a .462 rate against .563 for real home teams. The ninth international game, Philadelphia at Jacksonville in London, is coded a home game instead, and this page says so out loud.
Sum each game's chance of fooling the model and the frozen 2026 ratings expect 99 wrong picks across the 272-game season — published before kickoff, exact distribution attached: 90% of seasons land between 86 and 112 misses, which is 68.4% accuracy down to 58.8%. The 42 heaviest favorites still owe nine misses, the purest coin flip on the board is Green Bay at Tampa Bay in week 4 (50.07%, a 47.5-point rating edge cancelled by home field), and sixteen backtest seasons all kept the budget inside two sigma — 2020 hit it to the decimal, 91 actual against 91.0 expected. Around 99 misses is the model working.
Since the 2002 realignment the two conferences have met 1,616 times in the regular season, and the AFC leads 823–787–6 — a .511 edge banked entirely by 2010, when nine seasons at .5625 sat three standard errors from a coin flip. Since 2011 the AFC is 500–536–4, the 17-game era is dead even at 196–203–1, and home teams do no better hosting the other conference (56.6%) than their own (55.9%). In 2026 the slate is 80 games, four division pairings, and all sixteen seventeenth-games hosted by the NFC.
Five seasons of 17 games rewired the league's record geometry: 65 teams finished exactly .500 in the last nineteen 16-game years, and exactly one — Washington's tie-assisted 8-8-1 in 2022 — has managed it in 160 tries since. The 8-8 crowd became twin peaks at 8-9 and 9-8, the most common season moved from seven wins to nine, 10-6 became unpostable, and half the league now holds a ninth home date whose value the schedule's own design — alternating it by conference — makes impossible to measure cleanly. The cancelled 2022 Buffalo–Cincinnati game gets its own paragraph.
Three franchises made the same bet on three different amounts of evidence, and the schedule file's per-game coach fields let us grade each man on the games he actually coached. Dan Campbell inherited 14-33-1 and is 48-36-1 with a fresh 9-8 dent — and a 21-21-1 record in one-score games against 14-8 in blowouts. Mike Vrabel owns the fastest turnaround in the file — 4-13 to 14-3 and a Super Bowl trip, a +298 swing — and the 13-21 Tennessee ending that preceded it; his 2026 opens at Seattle, the team that beat him in February. Brian Schottenheimer's 17 games improved Dallas by 78 points of differential while losing half a win. The board prices them #6, #9 and #22, and this page says what would count as evidence for each.
The ledger froze its ratings on August 12 and priced every opener before anyone kicked anything: 9.32 expected home wins, 9.91 expected favorite wins, and the miss budget's first installment of 6.09 wrong picks, due by Monday night. The exact distribution makes 6 upsets the single most likely week (20.5%) and all-chalk a 1-in-2,400 event; four favorites are on the road, the biggest number is Jacksonville's 75.3% over Cleveland, the smallest is the Giants' 51.0% coin flip with Dallas, and in five games — every one priced inside three market points — the model and the schedule file's market snapshot pick different winners. Graded in public, never edited afterward.
The season opens Wednesday, September 9 at Lumen Field — only the second Wednesday opener in the file's twenty-eight seasons — with the Super Bowl LX rematch. The series says Seattle: five of eight all-time, four straight, six of eight one-score, and an average margin of 4.7 points before February's 29–13. The frozen board says Seattle too: #1 at 1674.6 hosting #6 at 1592.8, a 129.8-point effective gap, 67.9%. What the number cannot see is the grudge — and the companion study on Super Bowl sequels suggests the grudge has rarely been enough.
Twenty-seven Super Bowls sit in the games file and only six produced a rematch the following regular season — none before 2013, all six since. The champion won five, by an average of 6.7, and home field wasn't the reason: the home team won just three of six. The lone avenger is the 2023 Eagles. The one prior banner-night sequel — Carolina at Denver, 2016 — went to the champion by a point, which makes Wednesday in Seattle the second of its kind. Six games is six games; the page says so out loud.
The frozen board favors the visitor in four week-1 games, which every September gets read as a warning. It isn't one. Across 6,871 regular-season games at true home sites since 1999, home favorites won 67.66% and road favorites 64.70% — but home favorites were laying 5.85 points to the travelers' 4.49. Match on the exact closing number and the gap flips to −1.33 points in the traveler's favor (SE 1.20, z = −1.10). Against the spread it is the host who comes up short, 48.23% to 49.77%. In week 1 the road favorite is 96–48. The count is ordinary too — four market road favorites against a 27-year average of 5.44. What is unusual in 2026 is the composition: each board names four and they agree on two, and the two the model adds are the week's two biggest arguments with the market.
The board opens 2026 with Kansas City at 1508.2 — 3.2 points above what the engine hands a team it has never seen — after nine straight seasons ranked no worse than seventh. The 2025 collapse cost 193.9 rating points, second-steepest of the 47 falls from 1700 or better. Then the uncomfortable part: the Chiefs outscored their opponents by 34 and went 6–11, the second-largest Pythagorean shortfall in 861 team-seasons, built on a 1–9 record in one-score games this site has already published as noise. Denver is the same season pointed the other way, 11–2 in close games and rated fourth. They meet Monday, and the ledger picks the visitor while the market picks the host.
Five of the sixteen openers are division games, and the broadcast line is that each counts double. Across 768 team-seasons since the 2002 realignment the file says it is worth about 1.30, not 2. Sorted raw, the gradient is brutal — 0 of 40 teams that went 0–6 in their own division reached January, 28 of 29 that swept did, the exception being Oakland at 8–8 in 2010 — but most of that is a restatement that good teams win games. Hold the win total fixed inside the 7-to-10-win bubble and a winning division record is still worth +15.0 points of playoff probability; fit the two columns separately and a division win carries 1.30 times the weight of a non-division win, a gap of 0.574 with SE 0.168. The 2026 board prices its five division openers at 59.9% confidence against 62.0% for the full slate.
The ratings were frozen on August 12, so September should be the model's worst month. Sixteen seasons of grading say otherwise. Sorted by week instead of by year, 4,175 regular-season games give .6299 in weeks 1–4 against .6584 in weeks 14–18 — a 2.86-point gain with SE 2.09, which does not clear two — while stated confidence climbs 3.97 points over the same span. Week 1 has gone .6468, above the model's own season-long .6458 and better than nine of the other seventeen weeks. And the grading frame, published before kickoff: the 2026 openers expect 9.91 correct, and 7-of-16 through 13-of-16 covers 93.6% of the distribution.
Sort the sixteen frozen probabilities by how sure the model is and the 2026 board has a shape none of the sixteen earlier opening boards had: four games at 70% or better, two in the sixties, one coin flip, and nine stacked between 55% and 60% on a shelf 4.5 points wide. Prior boards averaged 3.7 shelf games and never held more than seven. The shelf is structural — two equal teams, one at home, price at 56.9% by arithmetic — and it is where Monday's grade will be decided: the nine games expect 5.13 correct, and in past openers that band has gone 35 of 58, .603 against a stated .574. The model has kept its stated number in five of six conviction bands in week 1, against two of six across all weeks.
Price every team's weeks 1–4 with the frozen ratings instead of last year's records and the opening month is lopsided: New England faces an effective opponent of 1617.1 with one home game and expects 1.86 wins of four, 13.0 points below its own season pace; Baltimore faces 1416.8 and expects 2.71; the Giants are the most front-loaded team on the board. The ratings expect 2.41 teams at 4-0 and 2.44 at 0-4, and sixteen seasons produced 2.38 and 2.50. Then the part that matters in October, from 512 team-seasons: each September win above the priced expectation carries +5.4 points of win rate into the remaining games, while the difficulty of the opening schedule, given the record and the August rating, is worth .02 wins.
The frozen 2026 board is last season's ratings with a third of every deviation deleted, and that deletion is the largest change the model made all offseason — it signed nobody and watched nothing. Seattle handed back 84.8 rating points, the biggest move on the board, worth 1.21 expected wins; Tennessee, Las Vegas and the Jets were given 77.7, 77.0 and 71.7 for free. Seventeen teams paid, fifteen were paid, and the board's spread fell from 487 points to 325. One week-1 pick flips on it, three openers get louder rather than quieter, and 829 team-season pairs say the file itself keeps .634 of a January rating against the engine's .667 — with the worst teams, notably, bouncing back harder than a flat third expects.
Fitted on 7,261 games rather than assumed, a point of spread multiplies the home side's odds by 1.153, which makes a 60% pick worth 3.03 points and about −148. Converting all sixteen frozen openers that way puts the model a mean of 2.19 points from the June lines in the schedule file, with five games more than three points apart and eight picks carrying positive expected value — the largest, Miami at Las Vegas, at +43.5% and a full-Kelly 30% of a bankroll. Then the audit that spoils it: the identical procedure over 4,174 games since 2010 loses 54.7 units, about the cost of the margin, and in the bucket where the model claims 16.7 points of edge its pick wins 49.1%. An encompassing regression puts the market's coefficient at 1.040 and the model's at 0.033.
The 2026 board has not moved since August 12 and the first game to touch it kicks on Wednesday. The size of that touch is set less by the scoreboard than by which side wins: a field goal for Seattle is worth 8.4 rating points, the same field goal for New England 17.8, and the exchange rate between the branches is fixed at 2.111 — the prior odds on the model's own pick. Every margin is priced here both ways. New England needs to win by 24 to pass Seattle at the top; the whole night spans 77.3 rating points; the largest week-2 price swing available anywhere in the range is 5.0 points of probability, and no pick flips. Against the 7,276 moves this engine has ever made — median 17.05, largest 50.32 — a three-point Seattle win is a 13th-percentile night and a three-point New England win is a 53rd.
Openers are supposed to be the market's soft spot. Across 6,967 regular-season games since 1999 they are instead its shortest numbers: 4.52 points against 5.40 for every other week and 6.63 in week 18, with the average number growing 0.084 points a week (SE 0.011). It is not matchmaking — by last season's records week-1 pairings are no more even than any other week's — it is caution, priced at 22.2 points per unit of measured mismatch against 25.1 later. The caution costs information rather than accuracy: the opening number misses by 10.06 points against 10.28 later, but explains 9.9% of the margin against 19.0%. And it is not exploitable: the calibration slope is 0.919 ± 0.132, favorites went 196–218–10, and the 2026 board's 3.97-point average is exactly the discount the last twenty-six opening weekends took.
Two forecasters that share no inputs post their worst week in the same week. The market's closing spread misses by 10.97 points in week 10 against 10.26 overall; this site's Elo picks winners at .5856 there against .6458. Shuffle the week labels 20,000 times and a random calendar's worst week is at least that bad 65% of the time on the market side and 42% on the model's, and the coincidence itself happens 11% of the time. Week 10 has been worse than its own season in 15 of 27 years, five of which supply the entire effect; strip them and it is 0.31 points better. The peak-bye explanation runs backwards — the bye games are the calm third of week 10 — and a simulated 2026 season of pure .6458 coins has a 6-and-8 week in it.
Seven teams open 2026 under a new head coach and this site's rating cannot see a single one of them. It does not need to. Teams that changed coaches gained 8.12 points of win percentage against a 2.39-point decline for teams that kept theirs — 5.8 standard errors — but they had gone .3529 the year before against .5441, and holding last season's record fixed turns the bounce into −2.80 points (SE 1.46) and −17.7 points of differential. The model is unbiased on exactly the teams it is blind to: +1.17 percentage points on 1,921 team-games, cluster-robust SE 1.22, and .6465 accuracy against .6456. The market is the only party that pays — about a third of a point of spread, three times as much for a coach who has run a team before, and it does not come back.
New England and Seattle both went 14–3 and the frozen board has them 81.8 rating points apart — a wider gap than it sees between Seattle and Buffalo. The gap is inherited, not earned: the model gave New England +201.3 points across 2025 against Seattle's +176.6, and the Patriots still trail because they opened last season 137.2 behind after two 4–13 years. The accounting runs 137.2 to 112.4 to 122.7, then 81.8 once August took a third. And when two identical records meet, the rating has been right: across 328 such games since 2010 the higher-rated side won 66.2% with a gap of 80 to 120 and 78.9% above it, against 50.9% under 40. Tonight the model says 67.9% and the vig-free market 64.5%.
Seattle 13, New England 10, and the model's .6785 is one for one. This page grades the site's own pre-registrations: the move landed on Tuesday's table to the hundredth — +8.42, a 13th-percentile night against 7,276 historical moves — the 24-point crossover never came into play, and no week-2 pick flipped. Then the harder part. Four pre-kickoff numbers all picked Seattle, so the winner separated none of them; the model's .1033 Brier beats the June market's .1262 and a records-only .1861 only because it was the most confident, and the order reverses in the counterfactual. The spread pushed at the close. The miss budget is .32 lighter. One game grades a pick; it cannot grade a probability.
One game graded and the rankings did not change: Seattle rose 8.4 to 1683.0 and is still first, New England fell 8.4 to 1584.4 and is still sixth, and the other thirty numbers are untouched. What moved is the gaps — Seattle's lead over Buffalo from 58.7 to 67.1, New England's hold on sixth from 11.6 to 3.2. Since 2000, 905 of 7,017 games changed no rank at all, and three season openers did the same, 2025 among them. Tonight is different: San Francisco and the Rams in Melbourne are priced at .5789 on a neutral field, any Rams win makes them second, any 49ers win drops the Rams to fourth or fifth and lifts the 49ers three to five places, and the widest result rearranges eight rows. Priced both ways, before kickoff.
The model's .5789 on the Rams lost by twenty in Melbourne, and the ledger reads one for two at a Brier of .2193. This page grades the site's own pre-registrations against the miss: the rule landed the 49ers fifth and the Rams eighth exactly where yesterday's ladders said a nineteen-plus-point 49ers win would, six rows changed inside the published four-to-eight, and the 34.4-point move is a 94th-percentile night against 7,276 historical moves. Then the harder part. A .58 pick losing by twenty is one draw from a distribution that produces one about every nineteen games in the .55–.60 band; separating .5789 from a coin takes 157 of them. The market moved toward the Rams before kickoff, its second move toward a loser in two games. The miss budget is heavier by .58. The miss is owned at full price.
Thirteen of the sixteen week-1 games kick on Sunday, all unplayed, and this page treats them as one object. The exact distribution of how many of the ledger's thirteen frozen picks land has its mode at eight (22.6%), expects 8.08, and gives a losing Sunday 17.9%; nine or more, the count that leaves the season budget under its opening 99.11, is 41.1%. The market, asked the same question about the model's picks from Friday's lines, says 7.86, 21.0% and 36.0%, agrees on ten of thirteen games and sides against the model on three: Miami at Las Vegas (18.9 points apart), the Giants over Dallas, Houston over Buffalo. Between June and September 11 the market moved its opinion of this board by one-twentieth of a game. Five top-ten teams play; between three and seven top-ten seats change hands; Seattle is first in all 8,192 outcomes.
San Francisco 27, Rams 7 moved two ratings 34.4 points, and this page carries the move through both seasons. From the ledger's own ratings the 49ers' expected record went from 9.50 wins to 10.84 and the Rams' from 10.60 to 9.28, and their chances of ten or more wins crossed, 50% to 76% and 72% to 46%. The site's own 20,000-season simulation, re-run after reproducing its published table, moves the 49ers' playoff odds from 55.5% to 80.7% and the Rams' from 76.6% to 51.5%. Is one game worth that much? Across 766 team-seasons since 2002, the remaining season moved 1.01 times as far as week 1 re-priced it (SE 0.28). Blowouts ran a little further than priced, not significantly, and a pooled base rate is still not a forecast for the Rams.
The ledger is one for two, and by Monday people will read a week into the season. Across the sixteen seasons this model has graded, a season's week-1 record against its own stated confidence correlates .13 with the rest of that season (95% interval −.39 to .59), no other week does better, and a season's odd games do not predict its even ones (.023). The reason is that the model has no detectable good or bad years: its seasons scatter 2.40 points around their stated confidence where coin flips alone would scatter 2.88, chi-square 10.55 on 15 degrees of freedom. At the most real spread the data allow, 1.92 points, a sixteen-game week is worth 2.55% of its surprise, and the 1–1 start shades the rest of 2026 by a tenth of a game.
Buy the team that got embarrassed in week 1, sell the one that looked unbeatable: across 422 week-2 games since 1999 it has not worked. Teams that lost their opener by 21 or more covered 25 of 58 decided bets the next week, and fading teams that won by 21 or more went 30–28, under the break-even at standard prices. The line does move, about 0.065 points for every point a team beat or missed its week-1 number, so a 20-point surprise is worth 1.3 points in week 2, and the week-2 result then leans with the surprise by +0.007 points per point (95% interval −0.069 to +0.083). The market reacts the same way after every week of the season. This site's rating engine moves about three-quarters as far, and after the biggest surprises that cost it.
On Friday a page here priced Sunday's thirteen picks as one distribution and published what each outcome would mean. The model went nine of thirteen: the second most likely count, and exactly the break-even that puts the season's miss budget back under its opening number. Its Brier score, .2174, was better than its own probabilities expected. The market's favourites went ten of thirteen and beat the model by about the usual margin, all of it in the four games named in advance. Seattle stayed first and seven top-ten seats changed hands, the top of the published range, but the same thirteen winners at seven points each would have moved three. The margins did the rest, and the Chargers' loss to Arizona is the largest rating move of 2026 so far.
Fifteen of the sixteen week-1 games produced 50.0 points a game, more than any full opening week in the nflverse file, and 4.77 a game over their closing totals; one game, Chicago 59, Carolina 37, supplied 68% of that. Across 27 seasons a loud or quiet week 1 has carried about a sixth of its surprise into weeks 2 through 18 (0.151 points per point, SE 0.086), and the closing totals for those weeks have moved by the same sixth (0.150), leaving 0.001 for anyone betting the over. Week-2 totals move 0.179 per point straight away. Written down in advance for 2026: weeks 2–18 near 46.85 points a game, their totals near 46.07, and no lean to the overs.
Of 227 teams that lost their first two games since 1999, 25 made the playoffs (11.0%), against 42.9% at 1-1 and 61.8% at 2-0. The easy explanation is that 0-2 teams were bad to begin with. Holding this site's opening-day rating fixed, each of the first two results is still worth 21.2 points of playoff probability (25.0 without it), so who the team was explains about 15% of the gap; a top-third team at 0-2 got in 20% of the time, a bottom-third team at 1-1 30%. When two 0-1 teams meet in week 2, the winners have made it 37.5% of the time and the losers 12.5%. Week 2 of 2026 has three such games, and the fourteen games without Denver or Kansas City expect 6.91 0-2 teams by the model and 6.98 by the market.
Kansas City beat Denver 31–10, and the ledger's frozen .5805 on the Broncos is its sixth week-1 miss. The week itself graded as the most ordinary result on offer: ten of sixteen, the single most likely count; six misses against 6.09 expected, which leaves the season's miss budget at 99.02; a Brier score of .2251 against an expected .2296. Monday was not ordinary. The 34.99-point rating move is a 95th-percentile night, and it dropped Denver from third to eleventh, below every rung published before kickoff. The 41 points tie 2026's week 1 with 2012's at exactly 791, and the loss makes Denver the sixteenth team at 0-1, at home to a 1-0 Jacksonville next.
The site's 20,000-season simulation, re-run with all sixteen week-1 results, moves the 32 playoff odds by a combined 376 points: Kansas City from 38.0% to 67.2%, the Rams from 77.0% to 48.3%, and three division favourites change. That is an ordinary week 1, 16th of the 25 since 2002. Run the same simulation over those 24 seasons and the moves have been about the right size, 0.93 of what followed, and they sharpened the forecast. The odds themselves are another matter: teams given 90% or more after week 1 made the playoffs 78.4% of the time and teams under 10% made it 14.5%, because the simulation treats every rating as exact.
Yesterday's page measured this site's playoff simulation as too sure of itself at both ends — a calibration slope of 0.68 after week 1 and 0.60 on opening day — and declined to repair it. This page fits the repair and tests it walk-forward, choosing the parameter for each season only on the seasons before it. Telling the simulation to assume its own ratings are wrong by about 120 Elo points takes the out-of-sample Brier score from .2064 to .1989, the log loss from .6205 to .5834 and the calibration slope from 0.694 to 1.015, and it still hands out exactly the right number of playoff seats, which the simpler repair does not. Seattle goes from 95.7% to 83.6% and Tennessee from 0.7% to 6.7%. The price is that week-1 moves now look a little too small.
Week 2 opens with two 1–0 teams and the two forecasters this site keeps score between barely disagree: the frozen ledger says Buffalo .6612 and this morning's prices say .6636, after the market moved two points of spread toward Buffalo in two days and passed the model on the way. What Thursday decides is not those two seasons but two division races. Through the site's 20,000-season simulation a ten-point Detroit win puts the Lions at 55.6% in the NFC North against 26.8% if Buffalo wins by ten, and New England, which is not playing, swings 20.4 points in the AFC East. The margin barely matters once the winner is known, and today's board is already both answers blended at the price.
Yesterday this site fitted a repair to its playoff odds, which were flat enough to matter: a calibration slope of 0.694. The same test on the 4,363 games the model has graded since 2010 gives a slope of 0.897, with a season-clustered bootstrap that never reaches one, and not one of the ten reliability bands misses by more than its own interval. Fitted walk-forward and scored on 3,295 later games, the shrinkage moves the Brier score from .22039 to .22024 — a fiftieth of what it was worth on playoff odds, and 0.9 standard errors from nothing. A Murphy decomposition says why: miscalibration is 0.30% of this model's Brier score. A season simulation spends one rating error 272 times; a single game spends it once.
This site's engine reads final scores and says plainly that it does not know who the quarterback is. Across 656 games since 2010 in which exactly one side started a different quarterback than the week before, that team won 36.5% of the time; the model gave it 45.8% and the closing market 38.9%. The model's 9.28-point shortfall is 5.2 standard errors from zero and the market's 2.39 is 1.4, and no draw of 2,000 random samples came close. Pricing it would mean docking the changing team 78 Elo points, more than the engine's entire home-field edge. It survives removing the weeks when contenders rest starters, and the ratings close most of it on their own within three games.
Week 2 opened at Highmark Stadium and Buffalo won by ten, so the ledger's frozen .6612 was right and the game cost .1148 of Brier score. The margin matters more than the result: Tuesday's preview published a ladder of ten possible outcomes, and the game landed on the “Buffalo by 10” rung exactly, so the whole branch can be checked cell for cell against the real board. It reproduces — Detroit 26.8% in the NFC North, Buffalo 82.5% in the AFC East, New England down to 16.2% without playing. The rating move was 15.43 points, the 43rd percentile of 7,276, and the board changed two rows. The ledger reads eleven of seventeen and the miss budget 98.68.
Every within-season rematch in the game file is a division home-and-home — 1,347 of them since 1999 — which makes them the only matchup priced twice in the same season. The same team won both legs in 768 of the 1,340 decided pairs, .5731, where multiplying this site's own two probabilities together says .5497 and the closing market's say .5433 against an observed .5855. The excess lives almost entirely in quick rematches and blowouts, and reverses in December. The single-game price is fine: meeting-2 results do not lean against the spread with meeting-1's margin, at 0.81 standard errors. What is wrong is the independence assumption the site's own playoff simulation makes 272 times a season.
In four days this site measured three defects in its own forecasting and shipped none of them: a playoff simulation too sure of itself, a quarterback change the engine cannot see, and division rematches swept more often than two independent prices imply. Applied together to 768 opening-day team-seasons and scored out of sample on 608, they turn out to be very nearly one correction. Rating uncertainty alone takes the Brier score from .22388 to .21348 and the calibration slope from 0.607 to 1.007, better in 16 of 19 seasons; the rematch link adds 0.000013 on top of that, because the first correction already drives the simulated sweep rate past what history shows; and the quarterback fix cannot be applied in advance at all, because nobody knows in September who will change quarterback in November.
The engine behind this site's ledger has four moving parts and only one of them has ever been graded in public. K = 20 turns out to be the best K in sixteen seasons of hindsight, to six decimal places — 19.75 scores 0.220449 against 20's 0.220451. The home-field edge of 48 Elo points wants to be 42, and splits into 56 across 2010–2017 against 29 since, which is the decline the home-field page measured in win percentage arriving as a rating constant. The margin multiplier, the part that looks like decoration, is worth .00549 of Brier score, more than everything else here together. And tuning all three walk-forward across 16,317 settings gains −0.000081, while 513 of those cells beat the shipped model in hindsight.
The margin-of-victory multiplier is the one part of this site's rating engine that is genuinely load-bearing: remove it, re-fit K to compensate, and the model is still 0.002470 of Brier score worse across 4,363 graded games. So the obvious question is why it uses the natural log of the margin and not something else — and the answer is that it barely matters. Once K is allowed to rescale the update, every power curve from the fourth root of the margin to the margin itself falls inside a spread of 0.000507, one fifth of what the margin term is worth at all. Where the file does argue with the log is at the bottom: fitted freely, six margin buckets want a three-point win and a seven-point win given exactly the same weight, which is the one-score-games finding arriving inside the update rule. Neither repair survives a walk-forward test.
The ledger went 8 of 14 on Sunday, September 20, and the season Brier rose from .2186 to .2364 — which reads like a model coming apart. Decomposed, it is nothing of the kind. The calibration error over 31 graded games is .00104, every confidence band landed within 4.5 points of its own claim, and more than 99% of that Brier is the irreducible uncertainty term no forecaster controls. The uncomfortable finding is the other one: this engine's resolution — the term that measures how much it actually knows — is .00481 this season and .00631 across 4,350 games since 2010. It is honest and it is not sharp, and stretching its log-odds to make it braver only makes the score worse.
Sunday's six misses averaged 33.0 combined points and its eight hits 42.9, which looks like a mechanism: fewer possessions, fewer chances for the better team to prove it. Across 4,350 graded games since 2010 the pattern is real — the lowest-scoring fifth grades 62.2% against a claimed 65.1%. Then it dies twice. The gap is 1.67 standard errors, on a hypothesis that came from the slate rather than from theory. And sorting the same games by the PREGAME total line, the only version available before kickoff, flattens it completely: a correlation of +.0085 against +.0287. The market is unbiased about scoring and still tells you nothing about when this model is wrong.
Week 3 locked this morning on ratings that have absorbed 32 games, while weeks 1 and 2 were locked off a preseason board. So the same sixteen fixtures can be priced both ways. The ratings moved a mean of 27.15 Elo points, Las Vegas up 65.1 and the Chargers down 69.4; the probabilities moved a mean of 4.91 points, the largest 12.20; and not one of the sixteen picks changed side. Across 2010-2025 the same exercise flips a mean of 1.44, so nothing here is unusual — including the board getting louder, which it did in 12 of those 16 seasons.
A pick flips when the home rating plus the 48-point home-field constant crosses the away rating, so the distance from any pick to its opposite is exactly that gap. On this morning's board the sixteen week-3 distances run from 6.7 rating points to 238.9. One played game moves a team a mean of 18.08 points over 4,347 graded games, and because the update is exactly zero-sum a fixture's gap can close by at most about 36.16 in a week. Four of the sixteen sit inside that; four are more than five weeks away.
Each week-3 game forced both ways through the site's 20,000-season simulation, ratings held, common random numbers. The Rams at Denver moves 68.4 points of playoff probability, the most of the sixteen; the five largest swings are all interconference games, and in every interconference game exactly half the movement lands on teams not playing, because seats are conserved within a conference. Ranked by division titles, New England at Jacksonville leads instead.
After two weeks Tennessee is rated 1314.6, last by 65.0 points. Replaying 2010-2025 through the same engine and snapshotting every team after week 2, five of 512 team-seasons were lower. Of the sixteen week-2 bottoms, four finished last, seven in the bottom two, and one, Houston in 2023, made the playoffs. The simulation gives Tennessee 3.93 more wins and the league's fewest in 46.8 per cent of seasons.
Since 1999, 129 teams have started 0-3 and two made the playoffs, both of them Houston. For a team already 0-2, winning game three moved the playoff rate from 1.6% to 24.0%, and holding the rating fixed barely changes that. The ledger's week-3 prices expect 5.27 teams at 0-3 by Monday, against a 2002-2025 average of 4.67.
2026 splits exactly 8-8-8-8 after two weeks, and half of that is arithmetic. Across 401 1-1 starts since 1999, teams that won first and teams that lost first finished the season within 0.0012 of each other. Game three leans slightly toward the team that won last, against both the engine and the closing spread, at 1.45 to 1.66 standard errors: not enough to act on.
The ledger had Green Bay at .7139 and Atlanta won by 21. The game moved both teams 41.16 rating points, second in 2026 only to Carolina's 34-3 win at Atlanta four days earlier, so the season's two biggest swings are both Atlanta's and nearly cancel. Re-priced three ways, Atlanta's playoff odds went from 8.0% to 13.5% on the result alone and to 22.8% once the ratings moved. The market spent the week moving toward Atlanta and beat the ledger on the game.
The Rams and Denver are rated four tenths of a point apart, so Sunday night's pick is almost all home field. Since 1999 Denver's home edge is 2.91 points, eighth of 32 and 0.77 standard errors above the league average; the engine's fitted edge for Denver is 83.9 Elo points with a 95% interval of 35 to 136; and in division games the market charged a 0.42-point altitude premium the results did not earn. DraftKings has the Rams favoured by 2.5, a disagreement about the teams, not the altitude.
The ledger went eleven of sixteen in week 3 with a Brier score of .2259. The closing market's favourites went nine of sixteen and scored .2541, and both games in which the two picked different winners went the ledger's way. Only 15 of 277 weeks since 2010 were better for the model against the close, and the season now reads .2293 to the market's .2327. History says what that lead is worth: the model was ahead after three weeks in seven of sixteen seasons and finished ahead in two.
Atlanta's first three games missed the closing spread by 0.5, 28.5 and 25.5 points, the most erratic start of 2026 and among the top 2% since 1999. It changes nothing about Monday night. Across 861 team-seasons a team's scatter around the line in its first three games correlates .013 with the rest of its season, odd games do not predict even ones (-.003), and a seeded simulation shows a real difference of 1.5 points between teams would have been caught 99% of the time. The ledger's frozen .5403 for New Orleans stands, a quarter point from DraftKings' .5427.
On September 6 the frozen August board priced every team's first four games and expected 2.41 teams at 4-0 and 2.44 at 0-4 by October 5. The season produced three and five. Simulating the 64 games together, five or more was a 4.8% outcome, and five ties the most in 27 seasons. It came from the wrong teams: the three likeliest 0-4 teams all won, and Houston, priced at 1.1%, did not. Graded on all 32 records the board still beat the everyone-goes-2-2 rule by .04 to .05 a team, about twice its usual margin. The miss was in the tails.
Houston is 0-4 and the board still ranks it 14th, which looks like a model too slow to believe a record. Across 512 team-seasons it is not: grouped by record, the board's forecasts after four games matched the rest of the season, 0-4 teams included (.352 forecast, .354 played), and a faster K for the opening month would have improved the later games by .00004 of Brier at best. What the board over-trusts is the rating it carried from August, which history weights at .705. Houston's 14th is mostly August: discounted, its last 13 games project to 6.50 wins, not 7.06.
The ledger's frozen .6603 for Jacksonville against Philadelphia on Sunday includes the engine's 48-point home edge, because the schedule file codes this London game Home and next week's Neutral. The betting market has answered the question in every season since 2013. Net of both teams' season baselines, the closing line charged 2.32 points for Jacksonville's twelve London home games, against 1.87 for an ordinary home game and 0.31 for other neutral-site hosts. Whether the market is right, the results cannot say: Jacksonville went 6-6 there, 3.13 points better than the engine's neutral expectation with a standard error of 5.28.
Both teams are 0-4, with point differentials of -25 and -28, yet the board puts them 217.7 rating points apart, almost all of it carried from August, and the ledger makes Houston a .7265 favourite on the road; DraftKings says .7409. No meeting of two winless teams since 1999 has had a wider gap. Barring a tie, one of them leaves Sunday at 0-5, and none of the 81 teams that started 0-4 since 1999 made the playoffs. With the other three 0-4 teams all underdogs, the ledger expects 2.87 teams at 0-5 after week 5 and gives four, the most in any season since 1999, a .2401 chance.
For the first time since 2006, both the closing line and this site's engine have Chicago favoured at Green Bay: DraftKings by 3, the ledger's frozen row at .5424. Green Bay had been favoured in 19 straight meetings there. Across 980 division games since 1999 in which the visitor was favoured at a rival's home, the length of the run that ended has not made the price wrong: after runs of one to nine meetings the visitor won .684 against .639 from the line, and the 23 games after runs of ten or more went 11-12, 1.68 standard errors from their price, too few to act on.
Thirteen games are left for every team. Priced on the ratings the ledger uses, Chicago has the hardest remaining schedule and Cincinnati the easiest, 70.4 rating points apart and worth 1.19 wins to an average team. Tables built from opponents' records rank it differently, and sixteen seasons say which to trust: at this point of the season from 2010 to 2025, the ratings' forecast of a team's remaining opponents correlated .29 with how they went on to play, last season's records .18 and this season's records .17. The schedule then cost teams about exactly what the ratings said.
Tampa Bay is 0-4 by 20 points, tied for the second-narrowest of 86 0-4 starts since 1999, and the ledger's frozen row makes Dallas a .6405 favourite, about four points. DraftKings opened within half a point of that and now has Dallas by 8.5, a vig-free .7914. In the 121 games since 2010 in which the market was 15 or more points surer of the same favourite than the engine, those favourites won .793. A narrow 0-4 has meant little once the engine's numbers are known: against them, the margin of the four losses tests at 0.6 standard errors.
The frozen ledger and this morning's DraftKings prices pick the same winner in thirteen of week 5's fifteen games, and six are 10 or more points apart. Sixteen seasons of the same comparison, 4,153 games, say which number to trust: in seven of eight kinds of disagreement the results sat nearer the market's, and the market's Brier score was lower in all eight. When the two picked different winners the market's side won 373 to 289, though less than 10 points apart the engine's side led 100-91, a coin flip, which is where Cleveland at the Jets sits.
The ledger froze Dallas at .6405 on October 6, and Tampa Bay won 24-16, a miss scored at .4102. The market spent the week moving the other way: DraftKings opened within half a point of the ledger, and the closing line in the nflverse file had Dallas at .8082, so every later price scored worse than the frozen number. The game moved both teams 26.92 rating points, Tampa Bay from 26th to 23rd and Dallas from 20th to 24th, and the playoff simulation now gives Tampa Bay 13.6% and Dallas 12.2%.
Week 5 sends three teams three time zones from home, and the ledger prices each trip with the same 48-point home edge it gives a game across town. Since 1999 that has been fair: across 821 three-hour trips the home side's results sat within about one standard error of both the engine and the closing spread, and western teams in 1 p.m. Eastern games did no worse than priced. The one place a long westward trip has paid the visitor is at the three hosts that moved cities, the Rams among them, a pattern found by looking and flagged as such.
As week 5 began the board had Seattle 1st, San Francisco 3rd and the Rams 6th, and their average rating, 1639.9, is the highest any division's top three has carried at this point since the 2002 realignment. The site's simulation sends three NFC West teams to the playoffs in 54.4% of seasons. Replayed from week 5 of every season since 2002, the same simulation expected 12.94 three-team divisions and 14 happened, so the count holds up. Sunday's San Francisco at Seattle moves the division title about 30 points either way and the three-team chance by 1.8; Monday's Rams game moves it by 16.2.
Washington and the Jets are 1-3 and favoured on Sunday over 3-1 teams, the Giants and Cleveland. In 34 game-five meetings of a 3-1 and a 1-3 team since 1999 the closing line favoured the 1-3 team twice. Across the 336 games since 1999 in which the favourite had two or more fewer wins, the favourites won .625 against .608 from the line, and this site's engine, which reads only scores, gave them .474. A quarterback change explains part of it; the rest the file cannot see. The ledger's frozen rows side with Washington, barely, and with Cleveland.
Kansas City and Carolina take the first byes of 2026 in week 5 and then play thirteen straight weeks. Since 2002 a long run of games without a week off has cost nothing measurable: across 5,481 games, each game of difference in the two teams' runs moved results by -.001 against the site's engine and -0.01 points against the closing spread, and early-bye teams matched both in December. The part of the bye that shows up is the first game back, where teams have beaten the engine's numbers by .045, a rest edge the engine cannot see and the market mostly prices.
Seven teams changed head coach for 2026, and four of them are 3-1. Together the seven are 14-14, 1.45 wins ahead of the site's engine and 2.26 ahead of the closing line, and the August board gave four or more 3-1 starts a .047 chance. In the engine's graded window, 2010-2025, the 20 new-coach teams that beat it by a win or more over four games kept beating it, by .072 a game; over every season since 2000 that falls to .030, a lead rather than a finding. Written before this weekend's games.
The ledger's frozen rows went eight right and three wrong in week 5's twelve Sunday games, and the Giants and Washington finished 27-27. The Brier score of .1679 beat the closing market's .1741 and Thursday's DraftKings board, and the two games in which ledger and market picked different winners split one each. For the season the ledger now leads the close .2173 to .2243. Since 2010 the engine has led the close after week 5 in four seasons and kept the lead in two, each time from a bigger lead than this one. Written before Monday night's game.
Run before and after Sunday's twelve games, the site's 20,000-season simulation moves the 32 playoff odds by a combined 208.3 points. Chicago fell furthest, from 65.9% to 37.3% after losing 31-7 at Green Bay; Baltimore at Atlanta had the most at stake, and San Francisco at Seattle the least for the playoffs, though it swung the NFC West title. Per game it was a quiet week 5, 19th of 25 since 2002. Over those seasons week-5 moves held at a slope of 0.79: falls stuck, and rises gave some back. Monday night is drawn from the board, unplayed.
Every team has played four or five games, and the AFC and NFC standings after week 5 set each record beside its points. Five records run a full win or more ahead of their Pythagorean count, San Francisco, Minnesota, Kansas City, Buffalo and Cleveland (3-2 at -19), and three a full win behind, Houston, the Chargers and Tampa Bay. Since 2002, 75 teams a win ahead after five games won .522 of the rest, close to their Pythagorean .535 and far from their .792 record; point margin carried nearly all the signal.
After Buffalo's Monday win at the Rams closed week 5, the site's 20,000-season simulation gives every team's playoff and division odds. The AFC field is fairly settled; in the NFC seven teams sit between 35% and 55%. Chicago at Atlanta has the most at stake in week 6, 71.2 points of playoff odds; Carolina at Philadelphia the most for division titles. None of the six division games ranks in the top six. Since 2002, about three and a half playoff teams a season came from outside the top seats after week 5.
Both teams are 3-2 and Seattle is first on the site's Elo board, but by EPA per play its offense is 26th and Denver's 24th; the defenses rank second and seventh. Seattle's run game (-.275 per run, 30th) meets Denver's 28th-ranked run defense, and Seattle's net edge is .062 per play. Since 2002, road teams with an edge that size won .464 of their games.
San Francisco leads the league in net EPA per play after week 5 at +.260; Buffalo has the best offense and Minnesota, 5-0 with the 28th offense, the best defense. Across 768 team-seasons since 2002, five-game net EPA tracked the rest of the season (.394) about as well as point margin (.419), and the offensive half held up far better than the defensive half.
Four per-attempt parts, each held between 0 and 2.375, added, divided by 6 and multiplied by 100: that is the whole formula, and why its ceiling is 158.3. Two real lines worked through every step, Drake Maye's 113.5 for 2025 and a 2026 game where two caps bind; the formula as a straight line, in which a touchdown is worth 80 yards and an interception costs 100; and where the league's 13-point rise since 2003 came from.
Minnesota leads the league at +6 after week 5, with Jacksonville, Kansas City and Washington at +5 and Indianapolis and New Orleans last at -6. The week-6 games set both teams' margins side by side, the widest gap the Chargers at Kansas City. Across 736 team-seasons since 2003, a five-game margin carried into the rest of the season at only .18, and opponents' fumbles recovered at .06.
The free source most public NFL analysis runs on is nflverse: 24 data releases on GitHub, 4,376 files, no key, CC BY 4.0. The catalog, how to load it with nflreadr or nflreadpy (nfl_data_py is archived and deprecated), the two releases that are traps, and a timed test: after Thursday night's game the 2026 play-by-play file was replaced 4 hours 18 minutes after kickoff.
All eight divisions have a sole leader through Thursday of week 6, from San Francisco 2.5 games clear to Carolina by half a game, and each leader also has its division's best point differential. In 192 division races since 2002, a team alone in first after week 5 won the division at .673, and two-game leaders won 26 of 32.
The NFL breaks ties with two fixed lists, one inside a division and one for the wild cards, and a few rules around them decide more than the lists do. From 2002 to 2025, 23 of 192 division titles went to a tiebreaker and head-to-head or division record settled 18; strength of schedule never decided a title or a last playoff seat. Plus 2026's ties through Thursday of week 6, worked step by step.
The 2026 NFL trade deadline is Tuesday, November 10, at 4 p.m. Eastern, the Tuesday after week 9. In the five seasons since the schedule went to 17 games, 35 of 86 in-season trades came on deadline day itself, most of the 41 players moved were defenders, most of the picks were from rounds 4 to 7, and 27 of the 41 players went to a team with a better record.
Chicago at Detroit, Philadelphia at Dallas and Kansas City at Buffalo play on Thursday, November 26, with Green Bay at the Rams the night before and Denver at Pittsburgh the day after. From 1999 to 2025 Thanksgiving hosts went 33-40, mostly because the visitor had the better record in 40 of the 73 games, and the extra rest teams get afterward has not shown up in their next game.
Since the NFL went to 17 games and a 14-team field in 2021, 10 wins has made the playoffs 15 times out of 16 and 9 wins 9 times out of 23; every 11-win team got in and no 7-win team did. The table for every format since 2002, the teams that won enough and missed, and where 2026's teams stand through Thursday of week 6.
Green Bay at Chicago, Buffalo at Denver and the Rams at Seattle play on Friday, December 25, with Houston at Philadelphia on Christmas Eve. The NFL has played on Christmas every year since 2020; before that, a Christmas from Tuesday to Friday almost never had a game. Since 1999 Christmas hosts are 13-16, and the team with the better record has usually won.
In 2025 each offense ran 63.3 plays a game, 126.6 for both teams, the second-lowest count since 2002; add the kicks, punts and kneel-downs and a game had 154.5 plays over a median 3 hours 2 minutes. What a game is made of, how far one game can swing, why more plays do not mean more wins, and why 2026 is running higher so far.
aDOT, average depth of target, is how far past the line of scrimmage a player's targets travel, caught or not. The league averaged 7.86 yards in 2025: wide receivers 11.0, tight ends 6.2 and running backs about zero. What depth buys and costs, the shortest and deepest passers and receivers, and why a receiver's aDOT barely changes from one season to the next.
A field goal's distance is the yards from the line of scrimmage to the goal line plus 18, so a try from the opponent's 30 is a 48-yard kick. A yard-line table with how often each length went in over 2021-2025, how much better long kicking has become since 2006, indoors against outdoors, and 2026 through week 6.
Drive 1 is the game's first possession, run by the team that receives the opening kickoff, and drives are numbered for both teams in one sequence. In 2025 opening drives ended in points 45.6% of the time, the most since 2002, against 39.3% for all drives. How drives end, where Drive 1 starts, and how often a first-drive touchdown went with a win (62.6%).
Division games always split their wins and losses inside the division, so the fair test is the games against everyone else. Through week 6 the NFC North is 13-3 against other divisions and the AFC North 12-5; the NFC East and AFC South are 5-11. What a strong start has meant since 2002, and who each division has played.
YAC, yards after catch, is what a receiver gains once the ball is in his hands. In 2025, 47.2% of the yards on completed passes came after the catch; running backs averaged 8.0 per catch, tight ends 4.9 and wide receivers 4.1. Where the catch is made decides most of it. The leaders, the teams, and why the share has risen while YAC per catch has not.
ATT, CMP, LNG, SCK, TFL, PD and PCT, read off one real box score, the Giants and Washington's 27-27 tie, and checked against the 2025 season's totals. Sacks are not pass attempts, spikes are, kneel-downs are runs, and a tie is half a win: the Giants' 3-2-1 is .583.