Since late March, the Sharp Picks MLB model has been running in shadow mode. Every day, it analyzes the full slate of games, generates predictions, and records what it would have signaled. No picks go live. No money on the line. Just data.
What Shadow Mode Tests
Shadow mode validates three things: prediction accuracy, edge persistence, and market alignment. The model needs to demonstrate that its run line and total projections are calibrated, that edges identified at line release still exist near first pitch, and that the overall framework translates from basketball to baseball.
Baseball is a different animal. Starting pitchers matter more than any single factor in basketball. Bullpen usage patterns create multi-game dependencies. Weather, park factors, and altitude affect totals in ways that don't exist in the NBA. The model accounts for all of this, but theory and practice are different things.
Early Results
The good news: calibration looks reasonable. The model's predicted margins are tracking within acceptable ranges of actual outcomes. Run line accuracy is comparable to where the NBA model was after the same number of games.
The cautionary news: edge decay is higher in baseball. A lot of early-day edges disappear by first pitch, particularly in nationally televised games. The pre-tip validation cron, which works well for NBA, needs tighter thresholds for MLB to prevent acting on stale edges.
What's Next
Shadow mode continues through at least June. We need more data, especially on bullpen-heavy situations and weather-affected games. If the model passes internal validation, MLB signals will launch alongside NBA for the 2026-27 season - or potentially sooner if the data supports it.
You'll see it in the app when it happens. Until then, the shadow continues.