Published evidence

Our track record, uncorrected

Published calls remain visible after kickoff. Hit, Miss and Pending rows use the same row treatment. New to what a published likelihood means? How to read these numbers.

Updated As of Sep 27, 2026, 7:39 PM ET

Games we called at 65% or better

Most games are close to a coin flip and we say so. This is the record on the games where our published likelihood was 65% or better for one side, counted from the calls themselves and scored after the games were played.

4 of 5 correct, which is 80%.

Called at 65% or better and played: 5
Right: 4

A tied game in this tier counts as a Miss, since we never predict a tie.

This record counts games that have been decided. A call on a game still to be played is not counted here: it stands on its game card until the game is decided.

Why 65% and not something lower

A confidence tier is only worth showing if the games inside it beat what we do across every game. Two lower cutoffs were tested against that bar and neither cleared it: at 55 percent and at 60 percent, the range of plausible accuracy for those games reaches down past our accuracy on the whole slate, so neither could be shown to beat simply calling every game. 65% was the lowest cutoff that cleared, and the rule picked the lowest one that cleared rather than the one that looked best.

The cutoff was chosen and written down before this season's record existed, so it cannot be moved to flatter the numbers on this page.

Game likelihood calls

Through week 2 of the 2026 season. The published home likelihood is shown beside the final score when one exists.

Scoring note, 2026-09-19: a tie now counts as a Miss, at the confidence we had in the side we leaned toward, because we never predict a tie. In the Brier score and log loss a tie counts as if the side we leaned toward lost. No published call was changed. A record scored before this date left ties out of its figures, and its ties still read Miss here.

Week 3 10 calls 6 hit 4 miss

Published game calls with outcomes, week 3
Week	Game	Published winner	Home likelihood	Final score	Status
3	ATL at GB	GB	60%	35 to 14	Miss 60%
3	CAR at CLE	CAR	46%	18 to 21	Miss 54%
3	CIN at PIT	PIT	54%	27 to 30	Hit 54%
3	HOU at IND	HOU	50%	17 to 19	Miss 50%
3	KC at MIA	KC	46%	24 to 10	Hit 54%
3	LAC at BUF	BUF	65%	16 to 24	Hit 65%
3	NE at JAX	JAX	60%	6 to 35	Hit 60%
3	NYJ at DET	DET	71%	24 to 31	Hit 71%
3	SEA at WAS	SEA	41%	31 to 33	Miss 59%
3	TEN at NYG	NYG	63%	7 to 12	Hit 63%

Week 3 on its own page

Week 2 16 calls 10 hit 6 miss

Published game calls with outcomes, week 2
Week	Game	Published winner	Home likelihood	Final score	Status
2	CAR at ATL	ATL	58%	34 to 3	Miss 58%
2	CIN at HOU	HOU	53%	20 to 6	Miss 53%
2	CLE at TB	TB	58%	23 to 19	Miss 58%
2	DET at BUF	BUF	65%	31 to 41	Hit 65%
2	GB at NYJ	GB	38%	20 to 17	Hit 62%
2	IND at KC	KC	52%	30 to 33	Hit 52%
2	JAX at DEN	DEN	54%	13 to 20	Hit 54%
2	LV at LAC	LAC	60%	26 to 14	Miss 60%
2	MIA at SF	SF	65%	13 to 35	Hit 65%
2	MIN at CHI	CHI	56%	9 to 3	Miss 56%
2	NO at BAL	BAL	63%	24 to 17	Miss 63%
2	NYG at LA	LA	61%	6 to 28	Hit 61%
2	PHI at TEN	PHI	43%	24 to 20	Hit 57%
2	PIT at NE	NE	61%	3 to 20	Hit 61%
2	SEA at ARI	SEA	43%	31 to 7	Hit 57%
2	WAS at DAL	DAL	58%	20 to 37	Hit 58%

Week 2 on its own page

Earlier weeks

16 published calls from 1 earlier week. Every one of them is published and unchanged, and each week renders in full on its own page. The measurements above count the whole season, these weeks included.

Week 1 16 calls 10 hit 6 miss

Games scored: 42
Accuracy: 62%
Brier score: 0.2303
Log loss: 0.6528

Source: the pinned published track_record.json artifact. Pending calls are not included in scored measurements.

Player projection measurements

Measurements below are reproduced from the readable published track record artifact.

Player projection measurements by position
Position	MAE	Rank correlation	p20 to p80 coverage	Players scored
QB	7.12	0.351	45%	73
WR	4.28	0.459	54%	248
RB	4.62	0.611	55%	157
TE	3.58	0.378	55%	132

Source: the pinned published track_record.json artifact.

Largest player projection misses

These are the largest published projection gaps listed in the readable track record artifact. The published projection and recorded result remain side by side.

Largest published player projection misses
Player	Week	Published projection	Actual points	Status
D.Adams	2	8.44	35.50	Miss
J.Coker	1	5.90	29.80	Miss
J.Smith-Njigba	2	14.48	38.00	Miss
C.Watson	1	6.64	29.70	Miss
D.Montgomery	1	4.99	27.40	Miss
J.Allen	2	18.50	40.82	Miss
D.Henry	1	13.00	34.80	Miss
D.Lock	2	0.00	21.40	Miss
D.Swift	1	10.51	31.90	Miss
K.Walker	1	11.35	32.60	Miss
K.Murray	1	20.09	-0.38	Miss
C.Lamb	2	11.77	31.30	Miss
I.Likely	1	4.47	23.80	Miss
A.St. Brown	2	11.57	30.70	Miss
J.Gibbs	1	14.98	33.10	Miss

Source: biggest_misses in the pinned published track_record.json artifact.

Published players who did not play

The artifact reports 468 published players with no recorded weekly result. 14 were ranked in the top 24 at their position, with a mean published projection of 11.79 points.

Published top-24 players with no recorded weekly result
Player	Position rank	Week	Published projection	Status
K.Murray	QB2	2	18.83	Did not play
T.Tagovailoa	QB11	2	16.74	Did not play
P.Nacua	WR1	2	16.66	Did not play
T.Tagovailoa	QB21	1	16.18	Did not play
S.Darnold	QB23	2	15.41	Did not play
T.Tracy	RB11	2	12.88	Did not play
Z.Charbonnet	RB20	1	11.63	Did not play
N.Collins	WR14	2	10.35	Did not play
Z.Flowers	WR16	2	10.08	Did not play
Jordyn Tyson	WR19	1	9.42	Did not play

Source: did_not_play in the pinned published track_record.json artifact. These rows are reported separately and are not silently removed or scored as zero.

How this page works

A published prediction is never edited or deleted after kickoff. Corrections are dated annotations next to the original call. Brier score checks both the outcome and the stated confidence, while the row level result keeps every Hit, Miss and Pending row inspectable.