Full transparency into our prediction outcomes. Every call is documented, every result is graded, and every metric is tracked using the same analytical rigor we apply to the games themselves.
Our prediction model is built on publicly available advanced metrics that have been validated through decades of sabermetric research. Rather than relying on subjective analysis or narrative-driven reasoning, we let the data tell us where the market may be mispricing game outcomes.
The foundation starts with pitching evaluation through xFIP (expected fielding-independent pitching), which strips out defense and sequencing luck to isolate a pitcher's true skill. On the offensive side, wOBA (weighted on-base average) provides a comprehensive measure of plate production that weights each outcome by its actual run value, rather than treating a single and a home run as equivalent events the way batting average does.
WAR (Wins Above Replacement) gives us a holistic view of player value that accounts for both offensive and defensive contributions, adjusted for position and league context. Park factors are integrated into every projection because the same batted ball produces dramatically different outcomes at Coors Field versus Oracle Park. Bullpen workload, rest days, platoon matchups, and weather conditions are all factored into the final probability output.
The result is not a single prediction but a probability distribution generated through Monte Carlo simulation, running thousands of game scenarios to capture the full range of possible outcomes. This matters because a team with a 58% implied probability should still lose four times out of ten, and understanding that variance is what separates rigorous modeling from guesswork.
Every prediction generated by our model is timestamped before first pitch and graded after the game concludes. There are no retroactive edits, no disappearing calls, and no cherry-picked results. The full record, including every loss, is published on our Track Record page.
We evaluate performance through multiple statistical lenses rather than reducing everything to a simple win-loss record. A model can be profitable while losing more games than it wins if it identifies value at the right prices. Conversely, a model can post a winning record while still being poorly calibrated. The metrics below provide a complete picture of prediction quality.
Results are documented in the table below as the season progresses. During the offseason, the table reflects our most recent season of documented predictions.
| Date | Matchup | Model Projection | Result |
|---|---|---|---|
| 2026 regular season results will be documented here beginning Opening Day. Visit our Track Record page for historical performance data. | |||
For a comprehensive view of historical results including season-long summaries and monthly breakdowns, visit our full Track Record.
The betting market is efficient but not perfectly efficient. Our model identifies potential edges by focusing on inputs that the market underweights or processes differently than our framework suggests. This is not about having "inside information" but about applying a systematic, repeatable analytical process to publicly available data.
One of the most powerful tools in our methodology is BABIP regression analysis. When a pitcher's batting average on balls in play deviates significantly from league average, we project regression toward the mean at a rate determined by sample size and underlying quality metrics. This allows us to identify pitchers whose ERA overstates or understates their true ability, often before the market fully adjusts. Our expected stats guide details how we use Statcast expected metrics to validate these projections.
Once the model generates a win probability for each team, we compare that probability to the implied odds offered by the market. An edge exists when our probability estimate meaningfully exceeds the breakeven probability required by the offered line. We do not act on marginal differences. The threshold for documenting a projection as a value position is calibrated to account for the standard hold built into betting lines.
Early-season projections carry wider confidence intervals because sample sizes are small and the model relies more heavily on preseason projections. As the season progresses and real data accumulates, the model's weight shifts toward current-season performance. We document the confidence level of each projection alongside the prediction itself, so readers understand the distinction between a high-confidence call backed by robust data and a lower-confidence projection early in the season.
The sports betting information space is full of accounts that selectively post winners, delete losers, and present curated records that bear little resemblance to reality. We document results publicly because accountability is the only way to earn trust.
Every projection is published before the game starts. Every result is graded the same way, win or loss. We calculate and publish our Brier score, log loss, and closing line value because these are the metrics that serious analysts use to evaluate prediction quality. A simple "W-L record" can be gamed through bet sizing or selective presentation. Probability calibration metrics cannot.
If you want to evaluate whether our model produces useful analysis, the data is here. Read through our full methodology documentation, examine our published results, and draw your own conclusions. That is how it should work.
Dig deeper into the components that power our prediction model and the analytical frameworks behind each projection.