Free NFL & MLB
Prediction Models

It means our calculated probability of a scoreless first inning is higher than the probability implied by the market's odds. For example, if our simulations put a game at 55% and the market's price implies 50%, we call that a 5-point edge. Games like that make up our first graded record. An edge is a disagreement with the market, not a guarantee - individual outcomes still vary.

The model uses Monte Carlo simulation with 10,000+ iterations per game. It factors in pitcher xwOBA splits (vs lefty/righty batters), batter wOBA splits, sprint speed data, confirmed lineups, and our own empirical first-inning park factors to calculate the probability of no runs scoring in the first inning.

We pull real-time MLB data from MLB's official stats API for lineups and schedules, with pitcher and batter stats from FanGraphs and park factors built from historical first-inning results. For NFL, our player prop projections are built on nflverse, using play-by-play, weekly player stats, and current rosters going back to 1999. Market odds for both sports come from The Odds API.

Simulations refresh automatically every 10 minutes throughout the day. When lineups are confirmed (typically 1-2 hours before game time), the model incorporates the latest roster data. You can also refresh manually at any time.

We grade ourselves two ways, automatically, from official MLB results. "Edge calls vs market" counts every game where lineups were confirmed and our probability beat the market-implied probability. "Directional accuracy" counts every game where our probability was above 50.1%; at or below 50% we have no opinion and the game is excluded. Every qualifying game is logged - there is no hand-picking results - and predictions lock at first pitch. The full record, including an engine upgrade seam with our earlier results frozen, lives on the records page.

It is the share of graded calls that finished with a scoreless first inning. Each record has its own percentage: for edge calls, how often the first inning stayed scoreless when we disagreed with the market; for directional accuracy, how often it stayed scoreless when we projected above 50.1%. Both are published with full win-loss counts, updated after every game day.

The season record, recent windows, and the grading method are at openinglineanalytics.com/records, and every past slate is graded day by day in the NRFI results archive. Losing days stay published next to winning ones - nothing is retroactively edited.

We believe quality sports analytics should be accessible to everyone. There's nothing more frustrating than finding a genuinely helpful resource locked behind a paywall, so we built a free, reliable tool anyone can use to make more informed decisions about the games. The site is supported by community donations: if the projections have been useful, consider buying us a coffee to help keep the servers running.

Our NFL board projects weekly player props (passing, rushing, and receiving yards, receptions, and anytime touchdowns) by running 10,000 game simulations for every matchup. Each simulation plays out a full game, so instead of a single number you get a projected line plus the p25 to p75 range of likely outcomes. Projections refresh weekly, with a game-day pass for inactives. Explore the full board at openinglineanalytics.com/nfl.