🏈 ChalkProof

NFL Accountability

Games

Every game's prediction is locked ~90 minutes before kickoff — spread, total, win probability and vig-free moneylines, plus the market's opening and closing references — and never rewritten. After the final it is graded against the score. n is shown beside every figure and each model version is pooled separately, so eras never blend.

16 snapshots captured · 16 graded · 0 awaiting a final · model g2-v2.

11.57
spread MAE
n=16
13.06
total MAE
bias -3.70
0.2058
winner Brier
vs 0.2500 baseline
0.2060
ML fair-prob Brier

Spread bias +1.31 (+ = model favoured the home team too much) · winner accuracy 75.0%.

By week

WeeknSpread MAESpread bias Total MAETotal biasBrierModel
116 11.57 +1.31 13.06 -3.70 0.2058 g2-v2

Model vs closing line — paired sample, n=16

Spread MAETotal MAEBrier
Model 11.57 13.06 0.2058
Closing line 11.34 11.56 0.2108

Head-to-head — spread: model 7, market 9, ties 0 · total: model 4, market 12, ties 0. Spread edge vs close: -0.23 points of MAE.

Closing-line value

spread: n=12 · moved toward the model 25% of the time · mean |open−close| 0.50 · mean |model−close| 2.25.

total: n=13 · moved toward the model 0% of the time · mean |open−close| 0.31 · mean |model−close| 3.54.

winprob: n=11 · moved toward the model 55% of the time · mean |open−close| 0.02 · mean |model−close| 0.08.

Cold-start games whose open and close resolve to the same odds row are excluded (0), never counted as zero movement.

What we adjusted for

The projection factors each game was locked with before kickoff, read from the snapshot's own embedded ledger — not recomputed from today's model.

NE @ SEA Wk 1
NeutralLocked Actual
Spread +4.5 +5.4 +3
Total 47.7 47.7 23
Projection factors
  • NE travel: 3 time zones SEA +0.9 spread
Net vs neutral: SEA +0.9, total +0.0
CHI @ CAR Wk 1
NeutralLocked Actual
Spread -1.4 -2.6 -22
Total 43.9 45.3 96
Projection factors
  • CHI travel: 1 time zone CAR +0.3 spread
  • CAR defense: Patrick Jones II (LB, 60–85% snaps) OUT CHI +1.5 spread, total +1.5
Net vs neutral: CHI +1.2, total +1.5
NO @ DET Wk 1
NeutralLocked Actual
Spread +8.3 +8.3 +1
Total 44.2 44.5 61
Projection factors
  • NO travel: 1 time zone DET +0.3 spread
  • DET defense: Christian Izien (S, <60% snaps) 65% to play NO +0.3 spread, total +0.3
Net vs neutral: no side, total +0.3
BAL @ IND Wk 1
NeutralLocked Actual
Spread -3.8 -1.4 -18
Total 48.2 50.7 64
Projection factors
  • BAL defense: Nnamdi Madubuike (DT, 60–85% snaps) OUT; Teddye Buchanan (LB, 60–85% snaps) OUT IND +2.5 spread, total +2.5
Net vs neutral: IND +2.5, total +2.5
ATL @ PIT Wk 1
NeutralLocked Actual
Spread +2.4 +6.9 +7
Total 45.8 41.2 33
Projection factors
  • ATL Tua Tagovailoa OUT PIT +5.0 spread, total -5.0
  • PIT defense: Joey Porter Jr. (CB, ≥85% snaps) 65% to play ATL +0.5 spread, total +0.5
Net vs neutral: PIT +4.5, total -4.5
NYJ @ TEN Wk 1
NeutralLocked Actual
Spread -0.5 -0.2 -13
Total 40.9 40.9 33
Projection factors
  • NYJ travel: 1 time zone TEN +0.3 spread
Net vs neutral: TEN +0.3, total +0.0
BUF @ HOU Wk 1
NeutralLocked Actual
Spread +0.5 +0.9 -5
Total 47.2 47.3 67
Projection factors
  • BUF travel: 1 time zone HOU +0.3 spread
  • BUF defense: Jordan Hancock (CB, <60% snaps) 72% to play HOU +0.1 spread, total +0.1
Net vs neutral: HOU +0.4, total +0.1
TB @ CIN Wk 1
NeutralLocked Actual
Spread +2.6 +2.1 +6
Total 47.9 48.4 60
Projection factors
  • CIN defense: Josh Newton (CB, <60% snaps) OUT TB +0.5 spread, total +0.5
Net vs neutral: TB +0.5, total +0.5
ARI @ LAC Wk 1
NeutralLocked Actual
Spread +3.8 +4.5 -12
Total 41.5 42.9 40
Projection factors
  • ARI travel: 1 time zone LAC +0.3 spread
  • LAC defense: Deane Leonard (CB, <60% snaps) OUT ARI +0.5 spread, total +0.5
  • ARI defense: Dadrion Taylor-Demerson (S, <60% snaps) OUT LAC +0.9 spread, total +0.9
Net vs neutral: LAC +0.6, total +1.4
MIA @ LV Wk 1
NeutralLocked Actual
Spread -1.4 +0.0 +14
Total 46.9 47.5 40
Projection factors
  • MIA travel: 3 time zones LV +0.9 spread
  • MIA defense: Storm Duck (CB, <60% snaps) OUT LV +0.5 spread, total +0.5
Net vs neutral: LV +1.4, total +0.5
WAS @ PHI Wk 1
NeutralLocked Actual
Spread +5.1 +3.7 +2
Total 46.2 47.7 46
Projection factors
  • PHI defense: Jonathan Greenard (LB, 60–85% snaps) OUT WAS +1.5 spread, total +1.5
Net vs neutral: WAS +1.5, total +1.5
DAL @ NYG Wk 1
NeutralLocked Actual
Spread -3.3 -3.0 +8
Total 50.0 51.1 48
Projection factors
  • DAL travel: 1 time zone NYG +0.3 spread
  • NYG defense: Nic Jones (CB, <60% snaps) OUT DAL +0.5 spread, total +0.5
  • DAL defense: Caelen Carson (DB, <60% snaps) OUT NYG +0.5 spread, total +0.5
Net vs neutral: NYG +0.3, total +1.0
DEN @ KC Wk 1
NeutralLocked Actual
Spread +1.9 +3.6 +21
Total 46.5 46.8 41
Projection factors
  • DEN travel: 1 time zone KC +0.3 spread
  • Weather: 15 mph wind, 91°F total -1.2
  • DEN defense: Jonathon Cooper (LB, 60–85% snaps) OUT KC +1.5 spread, total +1.5
Net vs neutral: KC +1.7, total +0.3

Season futures

The immutable pre-season futures lock — the public record graded against actual results at season end.

Locked 13d ago · model n6-v1 · 32 teams · pending grade.

#TeamProj mean Wp25–p75 Playoff%MarketFair OverActual W
1 BAL 10.9 10.0–12.0 84% 11.5 35% —
2 DET 10.7 9.0–12.0 80% 10.5 54% —
3 SEA 10.5 9.0–12.0 77% 10.5 49% —
4 DAL 9.7 8.0–11.0 65% 9.5 53% —
5 BUF 9.6 8.0–11.0 66% 10.5 35% —
6 PHI 9.6 8.0–11.0 63% 10.5 35% —
7 HOU 9.6 8.0–11.0 67% 9.5 52% —
8 KC 9.4 8.0–11.0 61% 10.5 31% —
9 DEN 9.4 8.0–11.0 59% 9.5 47% —
10 SF 9.3 8.0–11.0 56% 10.5 29% —
11 GB 9.0 8.0–10.0 50% 9.5 37% —
12 MIA 8.9 7.0–10.0 51% 4.5 98% —
13 CIN 8.8 7.0–10.0 47% 9.5 38% —
14 PIT 8.8 7.0–10.0 47% 8.5 56% —
15 JAX 8.7 7.0–10.0 48% 8.5 54% —
16 LA 8.6 7.0–10.0 43% 11.5 10% —
17 ATL 8.5 7.0–10.0 49% 6.5 82% —
18 MIN 8.5 7.0–10.0 41% 8.5 51% —
19 TB 8.5 7.0–10.0 48% 8.5 50% —
20 NE 8.5 7.0–10.0 43% 10.5 18% —
21 IND 8.1 7.0–9.0 36% 7.5 65% —
22 CHI 7.8 6.0–9.0 28% 9.5 23% —
23 LV 7.7 6.0–9.0 28% 5.5 84% —
24 WAS 7.6 6.0–9.0 25% 7.5 51% —
25 LAC 7.5 6.0–9.0 25% 9.5 18% —
26 NO 7.4 6.0–9.0 27% 7.5 49% —
27 NYG 7.4 6.0–9.0 22% 7.5 48% —
28 CLE 7.3 6.0–9.0 19% 6.5 63% —
29 ARI 6.8 5.0–8.0 14% 4.5 85% —
30 NYJ 6.7 5.0–8.0 14% 5.5 70% —
31 CAR 6.5 5.0–8.0 14% 7.5 32% —
32 TEN 5.7 4.0–7.0 6% 6.5 37% —

Model-derived fair values, not betting advice.

Live drive props

Every drive's outcome distribution is locked at drive start and graded against what the drive did — about 22 predictions a game, which makes this the live model's primary accountability sample rather than the win-probability bar. Scored as multiclass Brier (sum of squared errors across all five class probabilities, 0–2), against a baseline that predicts this sample's own observed base rate every time.

350
drives graded
0.6202
multiclass Brier
0.6297
base-rate baseline
1.5%
better than base rate
OutcomeMean predicted ObservedDiffn
TD 0.254 0.254 +0.000 350
FG 0.154 0.134 -0.020 350
PUNT 0.345 0.531 +0.186 350
TURNOVER 0.117 0.009 -0.108 350
OTHER 0.130 0.071 -0.059 350

Model nfl-live-1 · 350 drive(s) captured, 0 awaiting a result. Offline holdout figures and the full calibration tables are in research/nfl/live_fit.md.