Track Record

Model vs Market — Every Graded Pick

Every prediction the model makes is logged, graded against the final box score, and compared to the sportsbook consensus. Losses included. Updated every 5 minutes.

Loading…

Graded Picks

—

Model Brier

—

Market Brier

—

Model Closer

—

Plain English — How's the model doing?

Loading…

How to read this page

▶

Every stat here measures the model against reality or the sportsbook consensus. Here's what each number means in plain terms:

Graded Picks

Predictions that have a final box-score outcome. The larger this is, the more reliable the rest of the numbers. Small samples = noisy results.

Model Brier

Measures how close the model's predicted probability was to reality. Lower is better. See scale below.

Market Brier

Same measurement, but applied to the sportsbook consensus probability. If Model Brier < Market Brier, the model is beating the book.

Model Closer

Of picks where both we and the market had a prediction, how often ours was closer to the actual outcome. Over 50% = directional edge.

Hit Rate (per prop)

How often the Over actually hit. If Hit Rate is 35% but Model Avg is 45%, the model is over-predicting by 10 points.

Edge

Model % − Market %. A positive edge means the model thinks the Over is more likely than the market does.

Brier score scale

0.25

0.22

0.15

<0.10

Coin flip Baseline Strong signal Elite

Reading a row in the picks table

+15% edge · HIT ✅

Model called a mispricing and was right. Exactly what you want to see.

+15% edge · MISS ❌

Model was wrong or ran into variance. Over enough picks, this should happen proportional to the stated probability — not absence of it.

−15% edge · HIT ✅

Market read it right, model was off. A single row doesn't matter; a pattern does.

−15% edge · MISS ❌

Market and model both favored the Under and it won. Good alignment.

Patterns across hundreds of picks matter. Any individual result is noise.

By Prop Type

Brier score measures calibration — how close the predicted probability is to the actual outcome. Lower is better. A baseline predictor that just guesses the league average has a Brier of ~0.22.

Loading…

Recent Graded Picks

Recent predictions where the model and sportsbook consensus priced the same line. Edge = model probability minus consensus probability. Picks where only one side priced the line are hidden by default — toggle below to see all logged predictions.

Prop: Result: View:

Date	Player	Prop	Model %	Market %	Edge	Result
Loading recent picks…

Methodology

What is logged. Every prop prediction the model generates for a game that has sportsbook consensus lines available. Picks are logged when the prediction is made, not after the fact.

How it's graded. After the game ends, the final box score determines whether the "over" hit. Actual outcomes come directly from MLB's StatsAPI.

Prop types tracked:

Strikeouts (Ks) — pitcher total strikeouts
Hits — batter total hits
Home Runs — batter total home runs
Total Bases — batter total bases (1B+2×2B+3×3B+4×HR)
H+R+RBI — batter hits plus runs plus RBIs

Brier score. Mean squared error between predicted probability and actual outcome. A Brier of 0.05 is excellent, 0.20 is near baseline. The model's Brier vs the market's Brier is the cleanest measure of who's predicting better.

Model closer %. Of all picks where both model and market published a probability, the percent of the time the model's prediction was closer to the actual outcome. This captures directional accuracy.

No cherry-picking. This page shows every prediction the model logs, including the misses. If a bug causes bad predictions, they appear here until the data is graded.