Back to board
2026-W41 · Oct 5 – Oct 11, 2026
Weekly accuracy report
Every one of the 25 games that finished this week was re-graded, scoring the 297 probabilities that had been published against actual outcomes. Lower Brier is better; 0.25 is coin-flip level.
Brier (model)
0.2061
raw 0.2159
Calibration gain
−0.0097
lower is better
Hit rate
68.7%
base 33.7%
Calibration error
4.2%
raw 7.9%
By league
| League | Games | Scored | Brier |
|---|---|---|---|
| MLB | 2 | 173 | 0.1846/ 0.1975 |
| CS2 | 18 | 66 | 0.2363/ 0.2465 |
| 국가대표 친선 | 4 | 52 | 0.2362/ 0.2363 |
| NFL | 1 | 6 | 0.2322/ 0.2322 |
What changed this week
- • Graded 297 markets from yesterday. Shipped probabilities scored Brier 0.2061 (raw 0.2159; a coin flip is 0.25).
- • Adopted a isotonic calibrator for league:mlb:batter_contact (hold-out Brier 0.2012 → 0.1826, 2668 samples).
- • Adopted a isotonic calibrator for baseball:batter_contact (hold-out Brier 0.2012 → 0.1826, 2668 samples).
- • Adopted a isotonic calibrator for *:batter_contact (hold-out Brier 0.2012 → 0.1826, 2668 samples).
- • MLB: learned calibration cut Brier by 0.0128 versus raw output (173 samples).
- • CS2: learned calibration cut Brier by 0.0101 versus raw output (66 samples).
- • batter_power markets ran 10.4%p high versus outcomes (46 samples, 11% actual). The refit shifts them down.
- • intlfriendly crossed the sample threshold and joined the learning set automatically.