ALIBI MLB · Model Fleet

Every model running in the daily prediction pipeline — what it is, whether it is running, and how accurate it has been this season.

Snapshot 1:27 AM ET · Jul 2, 2026 · manual refresh only

Accuracy is each model's own forward predictions scored against final results, season-to-date (2026-01-01 on). Brier (moneyline) and RMSE (totals) are both lower-is-better; Brier ~0.25 is roughly a coin flip, totals RMSE is in runs. n = settled games scored — a small n means the number is still noisy.

Moneyline models — 7 running

Model	What it is	Status	Brier — season	Brier — last 30d	Record — 2026 (live)	Lifetime — WF backtest
ml_logit_baseline	Logistic regression Moneyline baseline — the stable reference model.	● running last run Jul 2	0.2509 (n=774)	0.2491 (n=280)	88-112 (200) +1u · +0.4%	—
ml_lgbm	LightGBM Moneyline — primary gradient-boosted tree model.	● running last run Jul 2	0.2565 (n=774)	0.2516 (n=280)	103-146 (249) -21u · -8.4%	1985-2481 (4466) -215u · -4.8%
ml_xgb	XGBoost Moneyline — XGBoost tree challenger.	● running last run Jul 2	0.2523 (n=507)	0.2494 (n=280)	62-97 (159) -17u · -10.9%	—
ml_catboost	CatBoost Moneyline — CatBoost tree challenger.	● running last run Jul 2	0.2535 (n=507)	0.2495 (n=280)	57-88 (145) -18u · -12.3%	—
ml_logit_blended	Logistic regression Moneyline — blended-recipe challenger.	● running last run Jul 2	0.2480 (n=711)	0.2474 (n=393)	94-114 (208) +1u · +0.7%	—
ml_logit_all_l30d	Logistic regression Moneyline — ALL_L30D challenger (last-30-day recency window).	● running last run Jul 2	0.2496 (n=516)	0.2507 (n=385)	87-126 (213) -18u · -8.6%	—
ml_lr3	— Running model (not yet catalogued).	● running last run Jul 1	0.2493 (n=1024)	0.2480 (n=293)	—	—

Model

What it is

Status

Brier — season

Brier — last 30d

Record — 2026 (live)

Lifetime — WF backtest

ml_logit_baseline

Logistic regression
Moneyline baseline — the stable reference model.

● running
last run Jul 2

0.2509 (n=774)

0.2491 (n=280)

88-112 (200)
+1u · +0.4%

—

ml_lgbm

LightGBM
Moneyline — primary gradient-boosted tree model.

● running
last run Jul 2

0.2565 (n=774)

0.2516 (n=280)

103-146 (249)
-21u · -8.4%

1985-2481 (4466)
-215u · -4.8%

ml_xgb

XGBoost
Moneyline — XGBoost tree challenger.

● running
last run Jul 2

0.2523 (n=507)

0.2494 (n=280)

62-97 (159)
-17u · -10.9%

—

ml_catboost

CatBoost
Moneyline — CatBoost tree challenger.

● running
last run Jul 2

0.2535 (n=507)

0.2495 (n=280)

57-88 (145)
-18u · -12.3%

—

ml_logit_blended

Logistic regression
Moneyline — blended-recipe challenger.

● running
last run Jul 2

0.2480 (n=711)

0.2474 (n=393)

94-114 (208)
+1u · +0.7%

—

ml_logit_all_l30d

Logistic regression
Moneyline — ALL_L30D challenger (last-30-day recency window).

● running
last run Jul 2

0.2496 (n=516)

0.2507 (n=385)

87-126 (213)
-18u · -8.6%

—

ml_lr3

—
Running model (not yet catalogued).

● running
last run Jul 1

0.2493 (n=1024)

0.2480 (n=293)

—

Totals models — 4 running

Model	What it is	Status	RMSE — season	RMSE — last 30d
totals_ridge	Ridge regression Totals (combined run total) baseline.	● running last run Jul 2	4.3925 (n=774)	4.6131 (n=280)
totals_lgbm	LightGBM Totals — primary gradient-boosted tree model.	● running last run Jul 2	4.3960 (n=774)	4.6775 (n=280)
totals_xgb	XGBoost Totals — XGBoost tree challenger.	● running last run Jul 2	4.6098 (n=507)	4.7388 (n=280)
totals_ridge_all_l30d	Ridge regression Totals — ALL_L30D challenger (last-30-day recency window).	● running last run Jul 2	4.5835 (n=516)	4.5777 (n=385)

Model

What it is

Status

RMSE — season

RMSE — last 30d

totals_ridge

Ridge regression
Totals (combined run total) baseline.

● running
last run Jul 2

4.3925 (n=774)

4.6131 (n=280)

totals_lgbm

LightGBM
Totals — primary gradient-boosted tree model.

● running
last run Jul 2

4.3960 (n=774)

4.6775 (n=280)

totals_xgb

XGBoost
Totals — XGBoost tree challenger.

● running
last run Jul 2

4.6098 (n=507)

4.7388 (n=280)

totals_ridge_all_l30d

Ridge regression
Totals — ALL_L30D challenger (last-30-day recency window).

● running
last run Jul 2

4.5835 (n=516)

4.5777 (n=385)

Not currently running (no run in the last 7 days): ml_lgbm_walkforward_daily_aug_v1 (last run May 23), ml_lgbm_walkforward_daily_v1 (last run May 4), ml_lgbm_walkforward_v1 (last run Jun 9), sim_test_rotowire_v1 (last run Jun 5), totals_catboost (last run May 19).

Fleet status from public.model_runs — a model counts as "running" if it logged a run in the last 7 days. Accuracy from model.predictions joined to mart.fact_game final scores; the latest ok run per game is used. Read-only page — it changes no model and places no bet. Side-model calibration has its own page.