Backtest Results
2024 season — settled against actual Retrosheet box scores (245,958 game-bets analyzed)
Key Findings
All Overs (Blind)
-13.61%
ROI — vig kills blind betting
0-3% Edge
-5.6%
ROI — not enough edge
3%+ Edge
+4.8%
ROI — 8,335 bets

The takeaway: The projection model finds real edges. At the 3%+ confidence threshold, it produces +4.8% ROI against actual 2024 results — consistent with what sharp sports bettors achieve. Lower-confidence bets don't overcome the sportsbook vig.

Projection Calibration

How well do projected probabilities match reality?

Projected RangeProjectedActualMarketError
0-20% 16.8% 16.0% 20.2% -0.8%
20-35% 26.0% 23.6% 28.5% -2.4%
35-50% 43.4% 42.9% 48.4% -0.5%
50-65% 58.5% 59.5% 64.1% +1.0%
65-80% 68.6% 67.3% 73.2% -1.3%
80-100% 83.9% 76.1% 74.6% -7.8%

Projections are within 1-3% of actual outcomes across all probability ranges. Market odds consistently overprice overs by 3-5% (the vig/juice).

ROI by Prop Type (3%+ Edge Only)

These are the numbers that matter — filtered to only bets with 3%+ model edge (our actual recommendations).

Prop TypeROIMarket Efficiency
Batter Singles +15.5% 🟢 Softer market — best edges
Batter Doubles +14.9% 🟢 Softer market — best edges
Pitcher Hits Allowed +14.9% 🟢 Softer market — best edges
Batter Home Runs +5.4% 🟡 Moderate — playable with edge
Batter Hits +3.8% 🟡 Moderate — playable with edge
Pitcher Strikeouts -1.6% 🔴 Sharp market — avoid
Batter Strikeouts -1.3% 🔴 Sharp market — avoid

Why different markets perform differently: Sportsbooks set K props very sharp (close to efficient), but singles/doubles get less attention. Our model finds edges in the softer markets while respecting sharp markets. We show you props from all types, but filtered to only high-conviction opportunities.

Methodology

This backtest uses actual 2024 MLB box scores from Retrosheet event files to settle each prop bet. For each player+prop+line combination, we calculated the actual over-rate across all games that season and computed what the profit/loss would be at the given market odds.

Limitations: The Batch 7 dataset doesn't include game dates, so we settle against season-wide averages rather than specific game outcomes. A game-by-game backtest with DraftKings closing lines would be more precise. This also tests the Batch 7 source model's projections — DiamondEdge's own projections will be backtested separately once 2025 regular season data is available.

Data: 2,429 games • 48,655 batter game logs • 20,686 pitcher game logs • Filtered to -500 to +500 American odds (reasonable lines only).