NFL 2024 Backtest Results

30 games, 2024-09-05 to 2024-12-19. Every claimed metric independently recomputed and matched to full floating-point precision against real season data.

PurposeOut-of-sample backtest of the four-channel state estimation model against real NFL 2024 outcomes and the closing spread line.

StatusVerified — recomputed to full floating-point precision, run 2026-07-04

Built fromReal 2024 NFL season game data — spread lines, actual margins/totals

Depends onComponent A Changelog (Fix 7/8 — this dataset is what surfaced the test-methodology bug)

Superseded by—

EvidenceFull 30-row result set below, source CSV/JSON on request.

Summary metrics

games_tested:               30
date_range:                 2024-09-05 to 2024-12-19
mae:                        7.175333333333332
brier_score:                0.16509850824128458
spread_beat_rate:           0.4  (40.0%)
prediction_better_than_spread: 0.6333333333333333  (63.3%)
confidence_weighted_accuracy:  0.16198916666666668
winner_prediction_accuracy:    0.7666666666666667  (76.7%)
improvement_over_spread:       3.2413333333333343

Winner-prediction accuracy (76.7%) is the headline number, but spread-beat rate (40%) is the harder, more honest test — beating the closing line, not just picking the winner. Both are reported here, not just the flattering one.

Full result set

Game	Wk	C-pred	Confidence	Actual margin	Spread line	Beat spread	Correct winner
DAL @ TB	1	-10.04	0.25	8	-4.5	No	No
BAL @ KC	1	13.16	0.35	-7	-2.5	Yes	No
CIN @ NE	2	-0.48	0.10	2	-4.5	No	No
DEN @ PIT	2	0.07	0.30	3	-3.0	No	Yes
SEA @ SF	3	-0.57	0.20	13	-6.0	No	No
NYG @ PHI	3	-6.97	0.25	12	-5.5	No	No
MIA @ BUF	4	-7.61	0.20	-7	-3.5	Yes	Yes
LAR @ GB	4	-7.06	0.25	-5	3.0	Yes	Yes
TB @ NO	5	1.40	0.35	7	-3.0	No	Yes
HOU @ IND	5	9.70	0.25	11	-3.5	No	Yes
BAL @ CLE	6	21.06	0.25	9	-4.0	Yes	Yes
NO @ WSH	6	17.76	0.30	30	7.5	No	Yes
BAL @ PIT	7	18.92	0.40	16	-4.5	Yes	Yes
LV @ KC	7	-9.28	0.25	-7	-6.0	Yes	Yes
NYJ @ BUF	8	-11.12	0.20	-16	-6.5	No	Yes
DEN @ PHI	8	-4.27	0.25	-7	-5.5	No	Yes
DET @ GB	9	8.94	0.25	-14	7.0	No	No
IND @ HOU	9	-13.46	0.25	-5	-5.0	Yes	Yes
LAR @ LAC	10	-9.40	0.15	-10	-5.5	No	Yes
JAX @ MIN	10	-9.60	0.10	-14	7.5	No	Yes
WSH @ PIT	11	-4.64	0.35	-1	-3.0	No	Yes
NYG @ TB	11	-11.79	0.30	-23	8.0	No	Yes
PIT @ BAL	12	-23.18	0.40	-17	-6.0	Yes	Yes
LV @ KC	12	-8.58	0.25	-8	-5.0	Yes	Yes
NYJ @ IND	13	-10.02	0.25	-1	-4.5	Yes	Yes
WSH @ PHI	13	-10.42	0.30	-8	-8.5	Yes	Yes
CHI @ DET	14	-14.56	0.15	-18	8.0	No	Yes
SEA @ SF	14	4.49	0.20	-8	5.0	No	No
MIA @ GB	15	-6.45	0.25	-2	-3.5	Yes	Yes
NO @ TB	15	-3.40	0.35	-6	-2.0	No	Yes

Full column set (s/d/i/c channel values, pred_error, brier components) available in the source CSV — this table shows the columns that matter for an outside reader checking the headline claims.