How Dr. TrueLine's MLB Model Works
Most betting models are black boxes. “Trust our algorithm” is the pitch, and the methodology stays hidden. We're more transparent than most — but we also protect the specific calculations that give our model its edge.
This page explains what goes into our MLB model at a conceptual level. You'll understand what categories of data we analyze, why those categories matter, and how our filtering strategy turns raw predictions into profitable picks. What we don't share are the specific weights, formulas, and thresholds — that's our proprietary edge.
The approach
Every game starts with a league-average baseline. The model then adjusts each team's projection up or down based on multiple categories of variables. Those adjusted projections are converted to a win probability, which becomes our true moneyline — free of vig and book manipulation.
We compare our true line to the sportsbook's vig-stripped line. The gap between them is the edge. Larger gaps mean higher conviction.
What the model evaluates
Our variables fall into four broad categories:
Pitching. Starting pitcher quality is the single most important factor in any MLB game. We evaluate multiple advanced pitching metrics that measure what a pitcher actually does — not just his ERA, which is heavily influenced by luck and defense. We blend current season performance with historical data using a dampening method that prevents overreaction to small samples early in the year. We also factor in recent form and workload — a starter on short rest or coming off a high pitch count isn't the same pitcher.
Team performance. Offense isn't just about batting average. We evaluate advanced hitting metrics that capture how well a lineup is truly performing, adjusted for context. Bullpen quality matters too — the innings after the starter leaves are where many games are decided, and not all bullpens are created equal. We also compare today's actual posted lineup against what the team normally runs — rest days and lineup shuffles change the equation.
Situational factors. Where the game is played matters. Some parks inflate scoring, others suppress it. Weather affects ball flight and pitcher grip. The handedness matchup between the lineup and the opposing pitcher creates edges that casual bettors miss. Home field advantage is real and quantifiable.
Regression signals. Baseball is full of randomness, and teams routinely outperform or underperform their true talent level for weeks at a time. Our model identifies teams riding unsustainable luck — in batting outcomes, pitching results, and overall win-loss record — and adjusts their projections toward what the data says they should be doing. These signals activate as the season progresses and enough data accumulates to separate real performance from noise.
What makes our model different
There are other MLB models out there. What separates ours:
Multi-source data synthesis. We don't rely on a single stat or a single data provider. The model pulls from multiple sources and cross-references them. When sources agree, confidence is high. When they disagree, the model accounts for the uncertainty.
Bayesian dampening. Early in the season, a pitcher's stats are unreliable — three starts isn't enough to judge anyone. Our model blends current performance with historical data in a way that gradually shifts weight toward current-year stats as the sample grows. By midseason, the current year dominates. This prevents the wild early-season swings that plague simpler models.
Dynamic inputs. Lineups change. Weather changes. Bullpen availability changes. Our model refreshes multiple times per day as new information becomes available. The pick you see at 11 AM might look different from the one generated at 6 PM — because the inputs changed.
The filter — where prediction meets profit
This is arguably more important than the model itself. Raw predictions, even accurate ones, don't automatically translate to profit. The odds structure determines whether an edge is actually worth betting.
Our research across hundreds of MLB games revealed clear patterns: certain categories of picks consistently make money while others consistently lose — even when the model's prediction is correct. We built a filter that separates profitable picks from money traps based on these patterns.
The filter considers the edge size and situational factors to determine whether a pick is recommended. Games that pass the filter are labeled Best Bet (highest conviction) or Undervalued (profitable over volume). Below the filter, low-edge games where the market is still pricing one team with high conviction become Second Opinion picks — sportsbook overconfidence on a matchup the model views as fair. Games where the book has the price right are labeled No Play — tracked for transparency but not recommended.
We don't reveal the specific filter rules because they're the product of extensive proprietary research. But we do show you the results: the picks we include are profitable, and the picks we exclude are net losers. That's verifiable on our public scoreboard.
What the model doesn't do
We believe in being honest about limitations.
The model does not account for umpire tendencies — a factor we're researching but haven't integrated. It does not use closing line value or reverse line movement — both require paid data feeds. It does not model player props or totals — the engine is moneyline-only for MLB.
The model is also still growing. Our tracked results are meaningful but not yet at the thousands-of-games level where statistical certainty is absolute. We're confident in the methodology, but we're transparent that the sample size is building. Every pick and every result gets published as it happens.
See the model in action
Every output of this model is published daily on our Today's Plays page. Enable push alerts in your Account to get notified when official picks lock during the final hour before game time. Every result is tracked on the scoreboard. You don't need to take our word for any of this — the data is right there.