How to read

Pitcher chips show earned/unearned runs allowed in the 1st inning of each of their recent starts (newest → oldest); a highlighted 0 means the NRFI held. Team chips show runs scored in the 1st inning of the team's recent completed games (window set by the Team 1st inn control). The badge probability depends on the selected model. The Full Count (Advanced), our top performer across all 1,847 games of the 2026 season so far (61.2% on daily top picks vs a 49.6% league base rate): Poisson core with context, the pitcher's full-season 1st-inning split blended into his recent form, offense scaled by top-of-the-order OBP vs league (~.315), a platoon factor from the pitcher's OBP-against vs the top 3's handedness, and park + weather multipliers. The Heater (Weighted), the season's #2 (57.6% top picks): the four displayed rates over the last 15 games each, recent games counting more (×0.85 decay per game back). Basic: unweighted average of the four displayed rates. Log5: Bill James matchup formula per half-inning (pitcher NRFI rate vs opposing offense YRFI rate), averaged across both halves. Poisson: expected 1st-inning runs per half-inning → P(0 runs) = e^−λ, top × bottom. Poisson and The Full Count are closest to true probabilities, so their numbers run lower than the rate-average models. The footer of each card shows the park factor, weather and plate umpire (the latter two appear only near game time; umpire is informational, since MLB publishes no zone-tendency data). For finished games (and live games past the 1st) a result pill shows whether the NRFI actually held, a Model ✓/✗ pill grades the selected model's call for that game, and the status line tallies both the league hit rate and the model's won-lost record for the day. A game counts as an NRFI call when the model clears its call line — each model's break-even point from the full-season backtest (The Full Count .36, Poisson .40, Basic .74, The Heater .80, Log5 .88 — rate-average models run hot, Poisson models run low, so one flat 50% line would misgrade both). Pick a past date to backtest the leans. Larger windows (Last 25) take longer on first load; results are cached after that. Cookie settings