How It Works

Same pregame homework — pulling pitcher splits, comparing bullpen numbers, checking weather and lineups — just automated across 10 years of pitch data, for every game on the board.

The Data

8M+
Pitches analyzed
10
Seasons of data
~24,300
Games tested

Everything starts with pitch-level data from Baseball Savant — every pitch thrown in an MLB game since 2015. Velocity, movement, contact quality, how hard the ball was hit, where it went.

On top of that: park-neutral pitching and defensive grades from FanGraphs, game logs from Retrosheet for lineups and scores, live odds, and day-of weather.

No proprietary data. No secret sauce in the inputs. The same numbers any sharp bettor can pull up — just all of them, processed every day, before first pitch.

What It Looks At

Fifty data points per game, organized into five categories. Everything is rolling — recent performance weighted heavier than early-season numbers. No stale stats.

Starting Pitching
Velocity trends, whiff rate, expected run prevention. Is this arm sharp right now or coasting on reputation?
Bullpen
Reliever quality behind the starter. Workload matters — a good pen that threw 4 innings yesterday is a different pen today.
Lineup
Contact quality, barrel rate, how the lineup matches up against the opposing arm. Not just who's hitting — how hard and how often.
Defense
Fielding range, positioning, plays made above average. Runs saved are runs scored.
Situational
Park factor, weather, rest days, travel distance. The stuff that doesn't show up in a box score but moves the number.

Backtest Results

Five separate tests — each time trained on earlier seasons, then turned loose on a full year of games that were completely held out. No peeking, no adjustments after the fact.

TrainingTest YearGamesDirectional Accuracy
2015–202020212,42957.2%
2015–202120222,43058.9%
2015–202220232,43054.7%
2015–202320242,42956.0%
2015–202420252,43054.9%
56.3%
Avg directional accuracy
2.46
Avg MAE (runs)
5 / 5
Windows profitable at top tier

At standard -110 juice, breakeven is 52.4%. Consistent accuracy above that line, across five different test years, is the whole point.

2025 Blind Run

Every game of the 2025 season. Zero changes — same settings, same thresholds, no manual overrides. 2,430 games never seen during training.

2,430
Games
54.9%
Directional accuracy
2.51
Avg miss (runs)

Month-by-Month — 2025

Mar
55.2%
Apr
57.3%
May
56.5%
Jun
53.4%
Jul
54.2%
Aug
55.3%
Sep
52.1%

Runs every day during the season. Every pick tracked, every outcome logged, every unit accounted for.

Questions? @DaftyDimes on X