The Model

The model’s methodology is proprietary; its performance is not. We publish the full out-of-sample record: every number below comes from games the model had never seen, scored blind, decades of them. It is the same model we trade our own capital on.

56.8%

out-of-sample accuracy

169,939

games predicted blind

0.6784

log loss

0.23 pts

calibration error

1935–2025

backtest span

Calibration and baselines

Baseball has a low prediction ceiling — the best teams lose 60 games. The two measures that matter are performance against naive baselines on identical games, and knowing your own uncertainty. Against 169,939 games it had never seen, the model called 56.8% — ahead of Elo alone (56.4%) and always-take-the-home-team (53.9%) — with an expected calibration error of 0.23 points across every probability bucket.

Calibration: when it says X%, what happens?

Accuracy by decade

The prediction ceiling

The ceiling is measurable. A deliberately cheating benchmark that knows every team’s final season record and every starting pitcher — information no forecaster can have — calls only 59.6% of games. Baseball’s best teams lose sixty times a year; the randomness is structural. What a calibrated model can do is identify the confident games and deliver the rate it claims on them:

Model confidence	Games (90-yr backtest)	Share of slate	Picks won
50–55%	74,636	43.9%	52.4%
55–60%	53,785	31.6%	57.3%
60–65%	27,169	16.0%	62.1%
65–70%	10,676	6.3%	67.4%
70–100%	3,673	2.2%	70.5%

Read the bottom row: when the model claims 70%+, it delivers 70%+. Those spots are rare — about 2% of games — and the market prices them heavily. High hit rates come from selectivity, not from a better crystal ball.

The ratings, live (2026)

#	Club	Elo	2026 record
1	Dodgers	1596	61–33
2	Brewers	1593	58–34
3	Marlins	1545	51–42
4	Cubs	1543	52–40
5	Red Sox	1539	42–48
6	Rays	1529	54–36
7	Phillies	1524	51–42
8	Pirates	1517	47–46
9	Braves	1516	53–39
10	Mariners	1516	47–46
11	Tigers	1516	42–50
12	White Sox	1511	47–44
13	Cardinals	1508	48–43
14	Yankees	1507	50–42
15	Twins	1500	46–47
16	Nationals	1498	48–46
17	Guardians	1497	47–46
18	Padres	1491	46–46
19	Blue Jays	1490	44–49
20	Orioles	1488	42–51
21	Rangers	1485	46–46
22	Astros	1484	46–49
23	Diamondbacks	1474	45–47
24	Reds	1468	42–49
25	Giants	1462	39–54
26	Royals	1462	38–55
27	Mets	1452	39–54
28	Angels	1450	37–56
29	Rockies	1436	38–56
30	Athletics	1433	41–51

Seeded from the end-of-2025 archive ratings (regressed one-third to the mean), then advanced game by game through 2026 with the identical Elo rules.