Bets · track record
The record
Every pick logged before puck drop and graded after. Nothing removed.
Model picked the winner
61.3%
212 of 346 games
Books' favorite won
60.7%
196 of 323 games
When they disagreed
38-34
model's side first
Brier score
0.228books 0.228
lower is better
Is 60% really 60%?
The model's favorite, by how sure it was50-55%119 games
53%said 52%
55-60%111 games
61%said 57%
60-65%60 games
60%said 62%
65-70%38 games
71%said 67%
70%+18 games
100%said 73%
The bar is how often the favorite won; the white mark is what the model said. Close together means the chances can be trusted.
Projected totals
Average miss
1.86goals
346 games
Within one goal
29%
of games
Over or under the line
49.8%
161 of 323 right
Underdogs the model liked
It favored them, the books didn't, they wonNov 18 · beat DET 4-1
BOS−110
56%model
Nov 17 · beat VGK 4-3
DAL+110
53%model
Nov 17 · beat SJS 5-2
CGY+110
56%model
Nov 17 · beat BOS 3-0
LAK+139
50%model
Nov 16 · beat MTL 3-1
NYR−103
56%model
Nov 14 · beat ANA 6-2
NYI−110
51%model
Nov 14 · beat MTL 4-1
COL+119
56%model
Nov 14 · beat STL 4-1
FLA−107
52%model
Nov 12 · beat CBJ 3-1
TBL+101
57%model
Nov 12 · beat BOS 5-2
MTL+105
54%model
Nov 11 · beat ANA 4-2
NSH+124
56%model
Nov 8 · beat PHI 4-3
CGY−1
56%model
How this is counted
Not bets: a test of the model alone, on every final game this season. The books' numbers are their closing consensus, with the vig removed. Brier score is the average squared miss of a win chance: 0.250 is a coin flip every game, lower is better.