April 22, 2026 MLB Home Runs Props

How we model these edges

Model method:

ml_monte_carlo

Approximation quality:

Reasonable

Bias direction:

Model edges on overs are biased high — real overs are slightly worse than reported.

Edge definition:

model_over_prob - no_vig(over_implied_prob)

• ·Per backtest, the model over-estimates home runs overs at high predicted probs (sharpness issue); remediation is the projection engine (`src/projections/hitter.py`), not the probability layer.
• ·Probabilities come from a 1000-rep Monte Carlo on the per-PA outcome distribution. Lines outside the sim's threshold grid (very rare) fall back to the Poisson approximation — see the per-entry `model_over_prob_method` stamp.
• ·Integer lines (e.g. line=2.0) are treated as 'over wins on ≥2', which slightly overstates over_prob vs sportsbook push rules. Half-point lines (X.5) — the near-universal case for these markets — are unaffected.