162 Games and the Luxury of Sample Size
The NBA regular season gives us 1,230 games. That is a reasonable sample to evaluate a model against. MLB gives us 2,430. That is a different universe of data.
More games means more opportunities to find edges. It also means more opportunities to do nothing. The discipline filter that sits out 40-50% of NBA slates will likely sit out an even higher percentage in baseball, because the sheer volume of games means we can afford to be pickier.
What Changes
Starting pitching is the single biggest variable in baseball. A team's expected run output shifts dramatically based on who is on the mound. Our feature set accounts for this: pitcher-specific metrics, bullpen usage patterns, rest days, platoon splits, and park factors all feed the model. In the NBA, the closest equivalent would be if a team's entire offensive scheme changed based on which point guard started. That is effectively what happens every day in MLB.
Line movement also behaves differently. In the NBA, lines move primarily on public money and injury news. In MLB, lines move on pitching changes, lineup announcements, and weather. The window between lineup lock and first pitch is where the sharpest edges appear and disappear. Our pre-tip validation is even more critical here.
What Stays the Same
The edge threshold is still 3.5%. The four-model ensemble still runs the same way. CLV is still the primary performance metric. If the model cannot find a mathematical edge above our minimum, no signal fires. The sport changed. The discipline did not.