Post-match review · published edition

NFL post-match calibration brief — September 13, 2026

Prepared 2026-09-13T21:11:42.702Z. This edition covers 8 completed early-window games; afternoon and evening games are not included in the results sample. Source: saved API-Sports American Football receipts, published forecast manifests and the local calibration archive.

Assessment

The published favorite won 8/8 completed games. That is an outcome summary, not proof that the probabilities are calibrated. The sample’s Brier score is 0.1631 and log loss is 0.5136. Log-loss improvement versus a 50% reference is 0.1796 nats/game (n=8, t=4.21). Predictive improvement is not established: the reference is a naive coin flip, the sample is tiny, games are from one slate, and this is repeated monitoring rather than a new independent model-selection test. No claim of beating sportsbook prices is supported.

Margin MAE is 10.0 points; RMSE 11.9; mean predicted-minus-observed home margin 4.6. The published central 80% interval contains 7/8 final margins. These are descriptive diagnostics from n=8; interval calibration and margin improvement are not established (t-statistic for those comparisons: unavailable).

All eight early-window forecasts in this edition were issued at 13:28:45.953 UTC and published at 13:45:43.308 UTC, before the 17:00 UTC kickoff. Two older games published after their kickoff are excluded from prospective evaluation. Prediction snapshots remain unchanged after kickoff.

Original forecast versus final result

Home-margin convention: positive favors the home team; negative favors the away team. Error is predicted minus observed home margin.

Matchup	Original favorite probability	Predicted margin	Final (away–home)	Margin error	Status
Tampa Bay Buccaneers at Cincinnati Bengals	57.1%	Cincinnati Bengals by 2.5	27–33	-3.5	Final
New Orleans Saints at Detroit Lions	71.1%	Detroit Lions by 7.9	30–31	6.9	Final after overtime
New York Jets at Tennessee Titans	53.0%	New York Jets by 1.1	23–10	11.9	Final
Baltimore Ravens at Indianapolis Colts	54.6%	Baltimore Ravens by 1.6	41–23	16.4	Final
Atlanta Falcons at Pittsburgh Steelers	64.2%	Pittsburgh Steelers by 5.1	13–20	-1.9	Final
Chicago Bears at Carolina Panthers	57.4%	Chicago Bears by 2.6	59–37	19.4	Final
Cleveland Browns at Jacksonville Jaguars	70.7%	Jacksonville Jaguars by 7.7	10–34	-16.3	Final
Buffalo Bills at Houston Texans	53.6%	Buffalo Bills by 1.2	36–31	3.8	Final

Game reviews

The paragraphs labelled AI analysis are hypotheses from the received evidence. They identify tests; they do not establish causal explanations or a model edge. Team passing yards are net of sacks; individual quarterback passing yards are gross, so those totals should not be expected to match directly.

Tampa Bay Buccaneers 27 at Cincinnati Bengals 33

Original outlook: Cincinnati Bengals 57.1% to win, conditional on a decisive game; home margin 2.5; central 80% range [-14.8, 19.8].

Team	Yards	Yards/play	Rush yards	Third downs	Turnovers	Red-zone made/attempts
Cincinnati Bengals	351	5.6	106	8-14	1	2-5
Tampa Bay Buccaneers	282	5.4	89	5-10	4	2-4

Player evidence:

Cincinnati Bengals: Joe Burrow 254 passing yards, 1 passing TD, 1 INT; Chase Brown 16 carries, 56 rushing yards, 1 rushing TD; Mike Gesicki 5 receptions/7 targets, 78 receiving yards, 1 receiving TD.
Tampa Bay Buccaneers: Baker Mayfield 216 passing yards, 0 passing TD, 0 INT; Bucky Irving 8 carries, 45 rushing yards, 1 rushing TD; Emeka Egbuka 5 receptions/6 targets, 63 receiving yards, 0 receiving TD.

AI analysis: Cincinnati won by 6 after a 2.5-point forecast. The yardage gap was modest (351–282), while Tampa Bay recorded four lost fumbles. Treat turnover-driven field position and scoring as a variance investigation, not a reason to sharply raise Cincinnati’s underlying strength. Test turnover regression and a separate non-offensive scoring component across future held-out games.

Evidence receipts: final scoreboard 2026-09-13T21:10:46.441Z; team box 2026-09-13T21:04:34.255Z; player box 2026-09-13T21:04:33.842Z. No points-against/scoreboard mismatch found in this receipt; this is not a complete statistical audit.

New Orleans Saints 30 at Detroit Lions 31 — overtime

Original outlook: Detroit Lions 71.1% to win, conditional on a decisive game; home margin 7.9; central 80% range [-9.4, 25.2].

Team	Yards	Yards/play	Rush yards	Third downs	Turnovers	Red-zone made/attempts
Detroit Lions	369	5.1	165	5-13	1	4-5
New Orleans Saints	469	5.5	91	8-17	3	4-5

Player evidence:

Detroit Lions: Jared Goff 206 passing yards, 2 passing TD, 0 INT; Jahmyr Gibbs 29 carries, 156 rushing yards, 2 rushing TD; Amon-Ra St. Brown 10 receptions/14 targets, 67 receiving yards, 2 receiving TD.
New Orleans Saints: Tyler Shough 410 passing yards, 3 passing TD, 2 INT; Travis Etienne Jr. 9 carries, 46 rushing yards, 0 rushing TD; Chris Olave 10 receptions/13 targets, 182 receiving yards, 0 receiving TD.

AI analysis: Detroit’s 71.1% decisive-win forecast identified the eventual winner, but the predicted 7.9-point margin materially exceeded the one-point overtime finish. New Orleans accumulated more yards (469–369), while Detroit benefited from a 3–1 turnover advantage. Preserve the regulation tie and overtime segment separately for diagnosis, while using the overtime-inclusive final for the full-game target. Raw volume in this game includes extra playing time.

Evidence receipts: final scoreboard 2026-09-13T21:10:46.441Z; team box 2026-09-13T21:05:51.172Z; player box 2026-09-13T21:05:50.744Z. No points-against/scoreboard mismatch found in this receipt; this is not a complete statistical audit.

New York Jets 23 at Tennessee Titans 10

Original outlook: New York Jets 53.0% to win, conditional on a decisive game; home margin -1.1; central 80% range [-18.3, 16.2].

Team	Yards	Yards/play	Rush yards	Third downs	Turnovers	Red-zone made/attempts
Tennessee Titans	195	4.0	68	4-12	1	1-1
New York Jets	367	5.8	152	2-12	0	2-3

Player evidence:

Tennessee Titans: Cameron Ward 140 passing yards, 1 passing TD, 0 INT; Tony Pollard 7 carries, 35 rushing yards, 0 rushing TD; Carnell Tate 4 receptions/6 targets, 38 receiving yards, 0 receiving TD.
New York Jets: Geno Smith 215 passing yards, 0 passing TD, 0 INT; Breece Hall 22 carries, 102 rushing yards, 1 rushing TD; Garrett Wilson 6 receptions/7 targets, 79 receiving yards, 0 receiving TD.

AI analysis: The Jets’ 53.0% forecast was a small lean, yet they won by 13. Their 367–195 yardage advantage and 38:46 possession time suggest a much larger realized performance gap. Test whether participation-aware quarterback and offensive efficiency inputs improve margin predictions; one outcome does not establish that the probability should have been higher.

Evidence receipts: final scoreboard 2026-09-13T21:10:46.441Z; team box 2026-09-13T21:04:35.493Z; player box 2026-09-13T21:04:35.078Z. No points-against/scoreboard mismatch found in this receipt; this is not a complete statistical audit.

Baltimore Ravens 41 at Indianapolis Colts 23

Original outlook: Baltimore Ravens 54.6% to win, conditional on a decisive game; home margin -1.6; central 80% range [-18.9, 15.6].

Team	Yards	Yards/play	Rush yards	Third downs	Turnovers	Red-zone made/attempts
Indianapolis Colts	251	4.7	102	3-10	2	3-3
Baltimore Ravens	506	7.9	202	3-9	1	3-4

Player evidence:

Indianapolis Colts: Daniel Jones 166 passing yards, 1 passing TD, 1 INT; Jonathan Taylor 19 carries, 98 rushing yards, 2 rushing TD; Alec Pierce 4 receptions/6 targets, 61 receiving yards, 0 receiving TD.
Baltimore Ravens: Lamar Jackson 324 passing yards, 1 passing TD, 0 INT; Derrick Henry 24 carries, 144 rushing yards, 3 rushing TD; Zay Flowers 5 receptions/6 targets, 150 receiving yards, 1 receiving TD.

AI analysis: Baltimore’s predicted margin was only 1.6 points; the final margin was 18. The Ravens produced 506 yards at 7.9 yards per play, against 251 at 4.7 for Indianapolis. Derrick Henry’s rushing touchdowns and Zay Flowers’ receiving production make this a useful case for studying explosive-play and rushing-score exposure. Evaluate those additions on later games rather than fit coefficients to this blowout.

Evidence receipts: final scoreboard 2026-09-13T21:10:46.441Z; team box 2026-09-13T21:04:36.872Z; player box 2026-09-13T21:04:36.433Z. No points-against/scoreboard mismatch found in this receipt; this is not a complete statistical audit.

Atlanta Falcons 13 at Pittsburgh Steelers 20

Original outlook: Pittsburgh Steelers 64.2% to win, conditional on a decisive game; home margin 5.1; central 80% range [-12.1, 22.4].

Team	Yards	Yards/play	Rush yards	Third downs	Turnovers	Red-zone made/attempts
Pittsburgh Steelers	264	4.1	58	3-13	1	1-2
Atlanta Falcons	238	4.2	120	2-12	2	0-0

Player evidence:

Pittsburgh Steelers: Aaron Rodgers 221 passing yards, 1 passing TD, 0 INT; Jaylen Warren 10 carries, 46 rushing yards, 0 rushing TD; Michael Pittman Jr. 3 receptions/3 targets, 58 receiving yards, 0 receiving TD.
Atlanta Falcons: Cooper Rush 143 passing yards, 1 passing TD, 2 INT; Bijan Robinson 21 carries, 83 rushing yards, 0 rushing TD; Bijan Robinson 8 receptions/10 targets, 90 receiving yards, 1 receiving TD.

AI analysis: Pittsburgh’s 5.1-point prediction was close to the seven-point result, despite low offensive efficiency on both sides. The source reports an interception touchdown for Pittsburgh and two Atlanta interceptions. Diagnose offensive scoring and defensive scoring separately so a returned interception does not become evidence of stronger offensive touchdown efficiency.

Evidence receipts: final scoreboard 2026-09-13T21:10:46.441Z; team box 2026-09-13T21:04:38.113Z; player box 2026-09-13T21:04:37.704Z. No points-against/scoreboard mismatch found in this receipt; this is not a complete statistical audit.

Chicago Bears 59 at Carolina Panthers 37

Original outlook: Chicago Bears 57.4% to win, conditional on a decisive game; home margin -2.6; central 80% range [-19.9, 14.7].

Team	Yards	Yards/play	Rush yards	Third downs	Turnovers	Red-zone made/attempts
Carolina Panthers	478	7.7	125	4-9	3	4-4
Chicago Bears	552	7.9	291	10-14	1	4-4

Player evidence:

Carolina Panthers: Bryce Young 361 passing yards, 3 passing TD, 1 INT; Chuba Hubbard 10 carries, 49 rushing yards, 1 rushing TD; Jalen Coker 8 receptions/9 targets, 138 receiving yards, 2 receiving TD.
Chicago Bears: Caleb Williams 269 passing yards, 2 passing TD, 0 INT; D'Andre Swift 18 carries, 124 rushing yards, 3 rushing TD; Kalif Raymond 8 receptions/9 targets, 84 receiving yards, 0 receiving TD.

AI analysis: Chicago won by 22 after a 2.6-point forecast, outside the published central 80% home-margin interval. Both offenses exceeded 7.5 yards per play, and Chicago rushed for 291 yards. Investigate wider or state-dependent margin tails and team-level scoring correlation. For touchdown models, multiple touchdowns by one runner are one success in an anytime market; 2+ and 3+ markets need different targets.

Evidence receipts: final scoreboard 2026-09-13T21:10:46.441Z; team box 2026-09-13T21:04:39.354Z; player box 2026-09-13T21:04:38.943Z. No points-against/scoreboard mismatch found in this receipt; this is not a complete statistical audit.

Cleveland Browns 10 at Jacksonville Jaguars 34

Original outlook: Jacksonville Jaguars 70.7% to win, conditional on a decisive game; home margin 7.7; central 80% range [-9.5, 25.0].

Team	Yards	Yards/play	Rush yards	Third downs	Turnovers	Red-zone made/attempts
Jacksonville Jaguars	360	6.4	126	5-10	0	2-2
Cleveland Browns	272	5.6	87	1-8	2	0-2

Player evidence:

Jacksonville Jaguars: Trevor Lawrence 245 passing yards, 4 passing TD, 0 INT; Bhayshul Tuten 15 carries, 66 rushing yards, 0 rushing TD; Parker Washington 5 receptions/6 targets, 83 receiving yards, 1 receiving TD.
Cleveland Browns: Deshaun Watson 205 passing yards, 1 passing TD, 1 INT; Deshaun Watson 6 carries, 38 rushing yards, 0 rushing TD; Denzel Boston 2 receptions/4 targets, 59 receiving yards, 1 receiving TD.

AI analysis: Jacksonville’s 24-point win exceeded its 7.7-point forecast. The box suggests strong passing efficiency and a 2–0 turnover advantage, but an earlier team-stat receipt reported 9 points against while the final scoreboard said 10. A later final-box receipt corrected the field to 10. Preserve both versions and the discrepancy history; use only the corrected, reviewed record for feature extraction. The final-score probability grade is independent of that box-stat discrepancy.

Evidence receipts: final scoreboard 2026-09-13T21:10:46.441Z; team box 2026-09-13T21:04:40.594Z; player box 2026-09-13T21:04:40.182Z. No points-against/scoreboard mismatch found in this receipt; this is not a complete statistical audit.

Buffalo Bills 36 at Houston Texans 31

Original outlook: Buffalo Bills 53.6% to win, conditional on a decisive game; home margin -1.2; central 80% range [-18.5, 16.0].

Team	Yards	Yards/play	Rush yards	Third downs	Turnovers	Red-zone made/attempts
Houston Texans	381	5.2	124	7-16	2	4-5
Buffalo Bills	409	7.9	86	3-9	0	1-3

Player evidence:

Houston Texans: C.J. Stroud 274 passing yards, 2 passing TD, 0 INT; David Montgomery 20 carries, 60 rushing yards, 2 rushing TD; Nico Collins 7 receptions/10 targets, 75 receiving yards, 1 receiving TD.
Buffalo Bills: Josh Allen 334 passing yards, 2 passing TD, 0 INT; James Cook 13 carries, 57 rushing yards, 0 rushing TD; Dalton Kincaid 5 receptions/6 targets, 130 receiving yards, 0 receiving TD.

AI analysis: Buffalo’s 53.6% forecast was a modest lean and the Bills won by 5. Buffalo gained 409 yards on 52 plays versus Houston’s 381 on 73, while Houston lost two fumbles. Josh Allen’s two rushing touchdowns count toward his anytime-scorer outcome; his two passing touchdowns do not. Separate quarterback rushing usage from passing success when developing scorer probabilities.

Evidence receipts: final scoreboard 2026-09-13T21:10:46.441Z; team box 2026-09-13T21:04:41.840Z; player box 2026-09-13T21:04:41.429Z. No points-against/scoreboard mismatch found in this receipt; this is not a complete statistical audit.

Player-prop calibration

32/32 published player projections have an exact player/team/stat match in the final boxes. Missing matches remain unknown; they are not graded as zero. These are point-estimate residuals, not probability calibration or a profitability backtest. Roster participation and any unresolved team-box flags must be reviewed before using these rows for model fitting.

Game ID	Player / target	Original projection	Recorded book line	Observed	Prediction error
21515	Samaje Perine / receptions	1.5	1.5	3.0	-1.5
21515	Ja'Marr Chase / receiving yards	96.0	86.5	12.0	84.0
21515	Ja'Marr Chase / receptions	7.0	7.5	2.0	5.0
21515	Baker Mayfield / passing yards	183.5	232.5	216.0	-32.5
21515	Mike Gesicki / receptions	2.0	2.5	5.0	-3.0
21515	Chase Brown / receptions	4.0	3.5	5.0	-1.0
21515	Andrei Iosivas / receiving yards	20.5	13.5	12.0	8.5
21515	Andrei Iosivas / receptions	2.0	1.5	2.0	0.0
21516	Isaac TeSlaa / receiving yards	19.0	17.5	0.0	19.0
21516	Isaac TeSlaa / receptions	1.5	1.5	0.0	1.5
21517	Breece Hall / receptions	2.0	2.5	2.0	0.0
21517	Breece Hall / rushing yards	54.0	63.5	102.0	-48.0
21517	Tony Pollard / receiving yards	9.0	9.5	-1.0	10.0
21517	Tony Pollard / receptions	2.5	1.5	1.0	1.5
21517	Tony Pollard / rushing yards	72.5	48.5	35.0	37.5
21517	Adonai Mitchell / receptions	2.5	2.5	3.0	-0.5
21517	Tyjae Spears / receptions	4.5	2.5	2.0	2.5
21517	Gunnar Helm / receiving yards	23.0	20.5	28.0	-5.0
21517	Gunnar Helm / receptions	4.0	2.5	4.0	0.0
21517	Cameron Ward / passing yards	182.0	201.5	140.0	42.0
21517	Cameron Ward / rushing yards	9.5	9.5	21.0	-11.5
21519	Jaylen Warren / receptions	2.0	2.5	5.0	-3.0
21519	Pat Freiermuth / receiving yards	26.5	26.5	46.0	-19.5
21519	Pat Freiermuth / receptions	3.0	2.5	5.0	-2.0
21519	Darnell Washington / receiving yards	23.5	16.5	18.0	5.5
21519	Darnell Washington / receptions	2.0	1.5	2.0	0.0
21521	Jerry Jeudy / receiving yards	24.0	29.5	26.0	-2.0
21521	Quinshon Judkins / rushing yards	36.5	56.5	33.0	3.5
21521	Harold Fannin Jr. / receiving yards	40.0	36.5	21.0	19.0
21521	Harold Fannin Jr. / receptions	4.0	3.5	2.0	2.0
21522	James Cook / rushing yards	93.5	73.5	57.0	36.5
21522	C.J. Stroud / rushing yards	2.5	10.5	15.0	-12.5

Anytime touchdown scorers

The new component estimates an offensive scorer’s full-game rushing/receiving touchdown probability, conditional on similar participation. Passing touchdowns are excluded; multiple rushing/receiving touchdowns still count as one success for an anytime target. It uses 6–8 recorded offensive appearances and a small-sample adjustment. Return/defensive touchdowns and opponent effects are not modeled. Calibration not established (held-out games: 0; t-statistic: unavailable).

Today’s afternoon estimates were first created after kickoff and are excluded from prospective pregame calibration. Cowboys–Giants estimates were saved before kickoff. The new component must not receive retrospective credit for the early-window games. Missing or unverified participation is unknown, not a failed scorer prediction. Full sportsbook anytime settlement additionally requires return/defensive scoring coverage and the book’s void/participation rules.

Prioritized calibration work

1. Data quality and overtime first. Recheck final team/player/scoring receipts, preserve every changed version, and quarantine score/box contradictions. Keep regulation and overtime duration separate for per-play and per-minute features; use the actual overtime-inclusive final for the full-game target. A game in OT is pending, not a tied result. A final tie is excluded from decisive-win scoring.

2. Build a real touchdown exposure model. Separate availability/participation, expected rushing attempts and targets, and scoring opportunity. Candidate features include goal-to-go carries, end-zone targets and team scoring expectations, but only if available with verified pregame timestamps. Retain an explicit missing-data flag; do not invent these inputs from final box totals. Compare a role-aware model with the current frequency baseline on a later, untouched evaluation set.

3. Investigate margin tails before increasing confidence. Chicago–Carolina, Baltimore–Indianapolis and Jacksonville–Cleveland motivate a test of overdispersion and dynamic variance. Pre-register an interval-coverage and proper-scoring comparison; require improvement across later slates. One successful slate does not justify sharpening probabilities.

4. Separate turnovers and non-offensive scoring. Evaluate regression toward longer-run turnover rates and a distinct return/defensive-score component. Avoid learning a permanent team-strength jump from fumble recoveries or a pick-six.

5. Use chronological evaluation and contract-matched prices. Freeze inputs and publication times before kickoff, reserve later weeks for evaluation, and compare identical market/period/line definitions. Today’s price archive began around 20:23 UTC, so it cannot supply a verified pregame close for the 17:00 UTC games. Quotes are last-observed pregame evidence, not an official closing line. Do not backfill a close from a postgame snapshot.

6. Promotion criteria. Accumulate a predeclared future evaluation window with adequate events per probability bin; evaluate Brier/log loss, calibration slope/intercept, interval coverage and missing-data rates. Use game-clustered uncertainty for player predictions and paired comparisons to baselines. Any parameter change remains a challenger until that evaluation is complete.

Capture policy and audit trail

The collector prioritizes recent final player/team/scoring boxes ahead of routine enrichment. A new final or overtime final triggers collection; empty replies retry after two minutes during the initial period. Nonempty recent finals are rechecked after 15 minutes in the first hour, hourly through six hours, then every six hours through three days. Older history backs off to weekly checks. Quiet passes can defer provider requests, while live games, overtime and imminent kickoff retain their active schedule. All calls keep the existing daily reserve.

Prices continue through the shared SharpAPI WebSocket. The local archive samples changed cached prices every 15 seconds, compresses the evidence and adds no provider API requests. A local relay interruption was found and recovered during verification; the gap must remain visible in capture coverage. This archive is not tick-complete.

[Machine-readable evidence](./calibration-brief-evidence.json) · [Team evidence](./evidence-teams.json) · [Player leader evidence](./evidence-player-leaders.json) · [NFL overtime procedures](https://operations.nfl.com/rules-officiating/2026-nfl-rulebook).

This brief is an as-of review, not an auto-retrained model or a betting recommendation. Later provider corrections require a revised brief; the original edition and forecast hashes remain auditable.

NFL dashboard

Your NFL view is loading