Model Performance

2022 Season · 2,980 graded predictions

2026 2025 2024 2023 2022 All

In-sample data: All predictions for this season were generated after games completed (historical backfill), so metrics reflect in-sample fit rather than true out-of-sample accuracy. Metrics will become trustworthy as the current season progresses and daily ETL generates pre-game predictions.

66%

Overall Accuracy

58%

Last 50 Games

0.2321

Brier Score (↓ better)

0.7478

Log Loss (↓ better)

0.6598

AUC (↑ better)

2,980

Predictions Graded

Calibration

When the model says X%, does the favored team win X% of the time? The vertical bar shows the predicted probability; the filled bar shows actual.

50–55%315 games52% actual(-1pp)

55–60%340 games58% actual(+1pp)

60–65%324 games60% actual(-3pp)

65–70%429 games60% actual(-7pp)

70–75%345 games69% actual(-3pp)

75–80%271 games71% actual(-6pp)

80–100%956 games75% actual(-16pp)

Actual win rate Predicted

By Confidence Tier

Accuracy when the model is highly confident in its pick.

Confidence	Games	Accuracy
≥ 60%	2325	69%
≥ 65%	2001	70%
≥ 70%	1572	73%
≥ 75%	1227	74%

Score Projection Accuracy

How close were the Monte Carlo score projections to actual results? The model's own projected run differential acts as the spread.

4.9

Avg margin error (runs)

+0.8

Margin bias (under-projects)

50%

Own-line cover rate

Within 2 runs of projected margin28%

Within 5 runs of projected margin62%

Based on 2980 games with score projections. “Own-line cover rate” = how often the actual margin exceeded the model's projected margin in the predicted direction.

Top Teams by Elo

Rolling Elo rating (1500 = average). Updated after every game result.

Rank	Team	Games	Elo
1	Texas Longhorns	267	1797
2	North Carolina Tar Heels	267	1774
3	Southern Miss Golden Eagles	174	1765
4	Auburn Tigers	253	1757
5	UCLA Bruins	251	1751
6	LSU Tigers	283	1741
7	Oregon State Beavers	268	1738
8	Florida State Seminoles	253	1738
9	Florida Gators	282	1736
10	Mississippi State Bulldogs	248	1735

D1 Diamond Top 25 — 2026 Season

Composite ranking of all D1 teams with ≥5 games played.

#	Team	Record	Elo	wOBA	FIP	Run Diff	Score
1	TEXTexas Longhorns	16–0	1797	0.453	2.25	+7.6	94
2	GTGeorgia Tech Yellow Jackets	15–2	1704	0.528	2.86	+9.5	89
3	UGAGeorgia Bulldogs	15–3	1709	0.515	2.76	+8.4	87
4	TA&MTexas A&M Aggies	15–1	1726	0.468	3.18	+7.2	87
5	MSSTMississippi State Bulldogs	15–2	1735	0.467	3.22	+6.8	85
6	AUBAuburn Tigers	14–2	1757	0.401	2.28	+5.1	84
7	CLEMClemson Tigers	15–2	1734	0.437	2.77	+5.2	84
8	OUOklahoma Sooners	14–2	1734	0.433	3.10	+6.4	84
9	UNCNorth Carolina Tar Heels	15–3	1774	0.411	3.49	+5.8	83
10	MISSOle Miss Rebels	15–3	1719	0.439	2.55	+4.9	83
11	FLAFlorida Gators	15–3	1736	0.415	2.52	+4.1	82
12	FSUFlorida State Seminoles	13–3	1738	0.425	2.92	+4.6	82
13	USMSouthern Miss Golden Eagles	15–2	1765	0.402	3.33	+3.5	82
14	USCUSC Trojans	15–0	1710	0.378	2.46	+5.0	82
15	UCLAUCLA Bruins	14–2	1751	0.438	4.61	+5.8	81
16	NCSTNC State Wolfpack	14–3	1673	0.485	3.63	+7.9	80
17	LSULSU Tigers	13–5	1741	0.413	2.93	+3.9	79
18	WAKEWake Forest Demon Deacons	15–2	1705	0.445	3.90	+4.7	79
19	UKKentucky Wildcats	15–2	1692	0.411	3.37	+4.6	78
20	ALAAlabama Crimson Tide	15–3	1722	0.409	3.44	+4.1	78
21	OREOregon Ducks	13–3	1703	0.436	3.66	+5.7	78
22	UVAVirginia Cavaliers	14–3	1709	0.429	4.14	+5.1	77
23	ORSTOregon State Beavers	11–4	1738	0.375	3.06	+1.8	76
24	TENNTennessee Volunteers	13–4	1717	0.381	3.08	+3.4	76
25	UCSBUC Santa Barbara Gauchos	13–2	1682	0.398	3.36	+4.1	76

Minimum 5 games played. wOBA/FIP from FanGraphs.

How to Read This

Overall Accuracy — % of games where the model's favored team won. A coin flip is 50%; sharp sports models typically land 58–65% on college baseball.

Brier Score — measures probability calibration quality. Lower is better. A random model scores 0.25; perfect calibration scores 0.

Log Loss — penalises confident wrong predictions more harshly than Brier score. Lower is better. Random baseline is ~0.693 (ln 2); well-calibrated sports models typically land 0.55–0.62.

AUC — probability that the model ranks a random home win above a random home loss. 0.5 = no skill; 1.0 = perfect ranking. Measures discrimination independently of calibration.

Calibration chart — a perfectly calibrated model's bar would exactly touch the line in every bucket. Bars to the right mean the model is underconfident; left means overconfident.

Score projection — tracks how close the Monte Carlo run totals were to reality. “Avg margin error” is the mean absolute difference between projected and actual run differential. “Own-line cover rate” treats the model's own projected margin as the spread — above 50% means the model tends to under-project winning margins.

Elo ratings — self-correcting power ratings that update after every game. 1500 is average; top programs typically reach 1600–1700 by mid-season.