Pigskin Nightly
More
GamedayScores & liveEvery game, live field, drive chartVideoHighlights for every gameStartersProjected, last game, confirmedInjury reportWho is out this weekTravel & restMiles, time zones, short weeks
AnalyticsThe VerdictWho deserved to win, graded weeklyEPA mapEvery offense and defenseWith / WithoutTeam splits when a starter sitsMilestone WatchCareer marks within reachCompare playersSide by side, every numberDFS researchDefense vs position, usage, totalsFree widgetsLive scores and stats for your siteGlossaryWhat every number means
The LabAll Lab toolsOriginal tools, built from every snapDeserved StandingsThe table the snaps would have builtSliding DoorsThe play that could have flipped itClock IQEvery staff against the clockRef Crew ReportEvery officiating crew, in numbersMarket vs SnapsWhere the line and the tape disagreeScheme DNAEvery playbook's fingerprintGame DNAEvery game drawn as one strandPaid vs PlayedEvery contract against the fieldLoudest StadiumsCrowd noise, measured since 1999Turf vs GrassInjuries by playing surfaceWeather LedgerReal weather, every game since 2000Line MovementOpening vs closing, every game since 2023
ClubsStandingsDivisions and playoff seedsPlayoff odds10,000 simulated seasonsStrength of scheduleThe road left for all 32All 32 teamsHubs, depth charts, snapsRostersEvery player, every teamNewsroomThe 32 club desksHow we writeOur editorial standards
Week 10
Latest
ESPNFantasy buzz: James Cook III helps Bills overcome tough Vikings defenseDetroit LionsTWENTYMAN: 5 stats that stand out following Lions' Week 9 victoryESPNSeven things Solak thinks for NFL Week 10: Jaguars trades, All-Pro picks, morePittsburgh SteelersAsked and Answered: Nov. 10ESPNCommanders running low on time to 'put it all together'New York GiantsGiants Now: Michael Strahan offers high praise of Arvell ReesePhiladelphia EaglesAAA Official Review: What did the Eagles, Joe Milton learn from the QB's first start?Buffalo Bills'There's no panic' | The Bills stay strong through early struggles and injuries in MNF win over the Minnesota VikingsESPNVikings left at QB crossroads after Carson Wentz's poor MNF showing vs. BillsBroncos DeskBroncos are 7-2, but the snaps say 4.1 wins and the 11th seed in the conference6 h agoInjuryByron Murphy Jr.: Active → Cleared6 h agoMinnesota VikingsVikings Defense Limits Bills Below AveragesMinnesota VikingsO'Connell: Vikings Have to 'Improve Offensively at Rapid Pace'InjuryBlake Cashman: Active → Cleared7 h agoVikings DeskVikings go 0-for-8 on third down and never get a receiver open in a 23-6 loss to Bills7 h agoInjuryWill Reichard: Active → Cleared7 h agoBills DeskJosh Allen's picks and sacks didn't matter: Bills force three turnovers in Minneapolis7 h agoPro Football RumorsSeahawks Interested In Adding WR DepthInjuryGreg Dortch: Active → Cleared8 h agoESPNJosh Allen and the Bills move to 7-1 with a 23-6 win over the sputtering VikingsBuffalo BillsBills 23 , Vikings 6 | Final Score, recap + highlightsInjuryCarl Granderson: Active → Cleared8 h ago
WSH ThuNYG NE SunDET BUF SunNYJ CAR SunNO HOU SunCLE JAX SunTEN KC SunATL MIA SunIND MIN SunGB LAR SunARI SEA SunLV SF SunDAL PIT SunCIN LAC MonBAL
The LabPigskin Nightly originals

How sure
are we?

Every Lab number is only as good as the test behind it. We rebuilt our model for 6,964 regular-season games since 1999, split this season into odd and even weeks, and asked each metric three questions: are its probabilities honest, does it predict anything, and how much of it is noise. Some pass. Some do not. All of it is here, and it updates itself after every game.

1999–2025 archive + 136 games this seasonResults published as they came outOur pick receipts
Deserved win probabilityCalibratedOff by 0.8 percentage points on average across 13,928 team-gamesDeserved WinsHolds upLuck does not carry over (r = 0.14); beats the record, trails points scored and allowedMarket vs SnapsDescribes, does not predictAgainst the spread every quartile lands between 49% and 51%Clock IQTaking shapeAbout 43% of the grade is the staff, the rest is chance; it needs 60% to count as a ratingLoudest StadiumsMostly noise so farThe same stadium one season to the next: r = 0.15; read the multi-season averages, not one seasonSliding Doors win modelCalibratedOur win probability against the nflverse model: R² 0.971 on 16,554 playsLine MovementDescribes, does not predictThe close beats the open: 9.8 → 9.64 points of error, Brier 0.215 → 0.21

passes the test describes what happened, or still taking shape mostly chance for now. Click a row to jump to its test.

r

Correlation. How closely two lists of numbers move together. 0 means no link at all, 1 means one predicts the other perfectly. In football, anything above 0.4 is a real link and anything under 0.2 is mostly chance.

0 · none0.20.40.61 · perfect
%

Calibrated. When a number says 70%, it happens about 70% of the time. A calibrated probability is honest about its own uncertainty.

½

Split test. Compute the same number on two separate halves of the data. If it measures something real about a team, both halves agree. If it is luck, they do not.

Test 1 · Deserved win probability

Are the probabilities honest?

13,928 team-games, 1999–2025
0%0%25%25%50%50%75%75%100%100%WHAT THE SNAPS SAID (CHANCE TO WIN) →HOW OFTEN THEY ACTUALLY WON →perfectly honest lineGiven 75%won 76%
?How to read this chart
  • Acrossthe chance to win the snaps gave a team in a game, grouped in steps of 10% (0-10%, 10-20% and so on).
  • Uphow often teams in that group actually won.
  • Each bubbleone group. Bigger bubble, more games in it. The small red tick is the gap between what was said and what happened.
  • The diagonalwhere every bubble would sit if the number were perfectly honest.

Bottom lineThe bubbles sit on the line: when the snaps say 75%, teams win 76%.

The Verdict turns every game into a probability: played like that, how often does a team win? If the number is honest, teams given 75% should win about three games in four. Across 27 seasons, teams the snaps gave 75% won 76%.

The model is simple on purpose: the difference in expected points added (EPA, the value of every snap) between the two teams explains 87% of the final margin in the archive, with a typical miss of 5.3 points. This season: 83% and 5.5 points on 136 games.

The caveat. This checks that the size of the uncertainty is right. It is not a forecast: the probability is computed after the game, from that game's own snaps. Test 2 asks what it predicts.

Test 2 · Deserved Wins

Does the luck evaporate?

855 team-seasons with 16+ games
30%40%50%60%70%SHARE OF GAMES WONFIRST 8 GAMESREST OF SEASON34%47%Unluckiest fifth · -1.2 wins43%46%49%50%Middle · 0 wins59%55%66%52%Luckiest fifth · +1.2 winsreal win ratewhat the snaps said
?How to read this chart
  • The groupsevery team-season since 1999, split into five equal groups by how lucky it was over its first eight games (real wins minus the wins the snaps said it deserved).
  • Lefteach group's real win rate in the first eight games; the dashed ring is what the snaps said.
  • Rightthe same group's real win rate over the rest of the season.
  • The linesif luck were a skill, the lines would stay flat. If it is chance, they bend back toward the rings.

Bottom lineThe red line (luckiest teams) falls from 66% to 52%, right where the snaps had them. The blue line (unluckiest) climbs from 34% to 47%.

The luckiest fifth of teams won 66% of their first eight games while the snaps said 51%. The rest of the way they won 52%. The unluckiest fifth went from 34% to 47%, with the snaps at 49%.

Luck in the first half against luck in the second half: r = 0.14. One season against the next: r = 0.12. Both are close to zero: whatever separates real wins from deserved wins does not stay with the team.

How exact is one team's Deserved Wins? On a full season, about ±1.14 wins (between 0.53 and 1.73 depending on how many games were close). The Deserved Standings now show that range next to every record.

What predicts the rest of the season

Three ways to judge a team after eight games, and how well each one matches its win rate over the rest of the season (r, the correlation: higher is better).

The best is points for and against (0.47), then Deserved Wins (0.43), then the real record (0.42). Deserved Wins beats the real record but not points for and against: the scoreboard margin still carries information the snaps miss, mostly the real value of turnovers and kicking.

Into the next season: record 0.33, Deserved Wins 0.35. Average miss on the second-half win rate: record 24.5%, Deserved Wins 22.7%.

After 8 games → rest of the season
Real recordwins and losses0.42
Deserved Winsfrom every snap0.43
Points for and againstPythagorean formulabest0.47
0 no link0.20.40.6
One season → the next
Real recordwins0.33
Deserved Winsfrom every snapbest0.35
0 no link0.20.40.6
?How to read this chart
  • Each barone way to rate a team. The number is r: how closely that rating matches what happened next. 0 means no link, 1 would be perfect.
  • Greenour Deserved Wins.

Bottom lineDeserved Wins beats the plain record, but points for and against is still the better predictor. No method gets close to 1: half a season leaves a lot to chance.

Test 3 · Market vs Snaps

Is it a betting signal?

855 team-seasons with closing lines
Covered the spread, rest of season
52.4% needed to profit50% coin flip
50.6%Snaps rated them far below the market-5.6 pts a game
50.7%A little below-1.4 pts a game
48.8%A little above+1.5 pts a game
49.8%Snaps rated them far above the market+5.4 pts a game
First 8 games: how the snaps rated each team against the closing lineNo edge
?How to read this chart
  • The groupsevery team-season since 1999, split into four by one question: over the first eight games, did the snaps rate the team better or worse than the betting market (the closing spread)?
  • Bar heighthow often that group then covered the spread over the rest of the season.
  • Red dashed line52.4%, the rate needed to make money after the bookmaker's cut. White line: 50%, a coin flip.

Bottom lineAll four bars sit near 50%, none reaches the red line. Knowing the snaps liked a team more than the market tells nothing about its next covers.

Market vs Snaps shows where the snaps rated a team better or worse than the closing line. If that gap were worth money, the teams the snaps liked most over the first eight games would cover more often over the rest of the season.

They do not. Sorted into quarters by the gap, the groups cover between 48.8% and 50.7%, and the correlation is -0.04. The market already prices what the snaps see.

How to use the tool. It is a record of how a team has played against how it was priced, useful to understand a season. It is not a pick.

Test 4 · Clock IQ

Skill or noise?

5 odd weeks against 4 even weeks, 32 staffs
00252550507575100100good in both halvespoor in both halvesgood, then poorpoor, then goodCLOCK IQ FROM ODD WEEKS ONLY →CLOCK IQ FROM EVEN WEEKS ONLY →LAR: 83 in odd weeks, 60 in even weeksDAL: 81 in odd weeks, 20 in even weeksCAR: 74 in odd weeks, 57 in even weeksSEA: 65 in odd weeks, 46 in even weeksHOU: 64 in odd weeks, 70 in even weeksPHI: 63 in odd weeks, 63 in even weeksSF: 59 in odd weeks, 48 in even weeksARI: 57 in odd weeks, 63 in even weeksMIN: 57 in odd weeks, 63 in even weeksNYG: 55 in odd weeks, 54 in even weeksBAL: 53 in odd weeks, 57 in even weeksCHI: 53 in odd weeks, 66 in even weeksNO: 52 in odd weeks, 70 in even weeksCIN: 52 in odd weeks, 42 in even weeksLV: 52 in odd weeks, 68 in even weeksNE: 51 in odd weeks, 49 in even weeksBUF: 51 in odd weeks, 43 in even weeksDET: 51 in odd weeks, 71 in even weeksGB: 50 in odd weeks, 50 in even weeksMIA: 50 in odd weeks, 35 in even weeksJAX: 48 in odd weeks, 68 in even weeksKC: 48 in odd weeks, 50 in even weeksATL: 44 in odd weeks, 50 in even weeksLAC: 43 in odd weeks, 54 in even weeksTB: 42 in odd weeks, 6 in even weeksPIT: 40 in odd weeks, 33 in even weeksWSH: 35 in odd weeks, 47 in even weeksCLE: 34 in odd weeks, 31 in even weeksDEN: 28 in odd weeks, 45 in even weeksIND: 26 in odd weeks, 42 in even weeksNYJ: 25 in odd weeks, 31 in even weeksTEN: 14 in odd weeks, 47 in even weeksDAL 81 → 20TB 42 → 6
?How to read this chart
  • Acrosseach staff's Clock IQ (0-99) computed only from games in odd weeks (1, 3, 5...).
  • Upthe same grade computed only from even weeks (2, 4, 6...).
  • Each logoone team. On the dashed diagonal, both halves give the same grade.
  • The circled teamsthe two whose grade changed the most between the two halves.

Bottom lineThe logos lean toward the diagonal: the grade is starting to show a real staff trait.

We computed Clock IQ twice, once from odd weeks only and once from even weeks only. If it measured a real staff habit, both halves would agree. The correlation is r = 0.27. Projected to all the games played so far, about 43% of the grade is the staff and 57% is the handful of end-of-half moments each team happens to face.

The single tests are thinner still: the two-minute drill alone splits at r = 0.03, burned timeouts at r = -0.10. Each team sees only a few such drives a month.

How to use the tool. Clock IQ is taking shape as a rating of each staff; read big gaps, not small ones.

Week by week

The same split test, repeated with the weeks available at each point of the season. It reruns itself after every game, so this chart and every number on this page move as the season goes.

After Week 4: 38%. After Week 9: 43%. It has not moved much as games were added. It needs to reach the green band, about 60%, before we call it a rating of the coach.

MOSTLY NOISETAKING SHAPEA STABLE RATING0%50%100%Wk 4Wk 11Wk 18WEEKS OF DATA SO FAR → (THE SEASON ENDS AT WEEK 18)43% now
?How to read this chart
  • Acrosshow many weeks of games the test had.
  • Uphow much of the Clock IQ grade belongs to the staff rather than to chance.
  • The bandsred under 40%: mostly noise; amber 40-60%: taking shape; green over 60%: a stable rating.

Bottom lineClock IQ is in the amber band today, at 43%.

The rest of the bench

Other instruments

Loudest Stadiumsr 0.15

The same stadium's crowd effect (extra visiting false starts) one season against the next, over 366 pairs. Weak: one season is mostly chance, so quote the multi-season venue averages.

Sliding DoorsR² 0.971

How closely our win probability matches the public nflverse model on 16,554 plays (1.000 would be identical). The swings are measured with a model that agrees with the reference.

Line Movement9.8 → 9.64

Average miss of the spread in points, opening line against closing line, on 896 games. The closing line is better informed, so the tool shows where the money went, not where it should go.

Turf vs Grass95% CI

Every injury ratio comes with its confidence range, and only ranges that stay clear of 1.0 are called a real difference. The source is the injury report, so it counts reported injuries.

MethodHow this is built
Deserved margin in the archive
For every regular-season game from 1999 to 2025: home margin regressed on the difference in total EPA, separately for each season. The win probability is the normal distribution of that margin with the season's residual error. The live Verdict also neutralizes fumble recoveries; the archive uses raw EPA, a slightly noisier version of the same model.
Halves
The first eight games a team played against the rest of its regular season, for team-seasons with 16 or more games. Ties count half a win.
Market test
Closing spread from nflverse. A team covers when its margin beats its spread; pushes are left out. Gap = average deserved margin minus the spread over the first eight games.
Clock IQ split
The full Clock IQ calculation run twice, on odd and on even weeks of the current season, same weights and caps. Reliability for the full sample uses the Spearman-Brown formula, 2r / (1 + r). The week-by-week line repeats it with the weeks played up to each point.
Loudest Stadiums stability
Season-level noise effect for each venue with at least six home games in both seasons, 2020 excluded.