Sharp Picks uses a 30/70 model-to-market shrinkage blend. That means the system's final output is 30% model prediction and 70% market line. On first glance, that sounds like the model barely has an opinion.
That's exactly right. And that's why it works.
The Market Is Good
NBA betting lines are set by sharp oddsmakers and shaped by millions of dollars in public and professional money. By the time you see a line, it's already incorporated an enormous amount of information. The market isn't always right, but it's right more often than any individual model.
A model that ignores the market is a model that thinks it's smarter than the collective intelligence of every bettor, oddsmaker, and syndicate in the world. That's arrogance, not edge.
Where the Edge Lives
The model's value isn't in bold disagreements with the market. It's in small, systematic corrections. When the model says a team should be -6 and the line is -4.5, that 1.5-point gap is the edge. It's not dramatic. It's not exciting. But over hundreds of picks, those small corrections compound.
The 30/70 blend ensures the model stays humble. It anchors to the market and only deviates when its features justify a correction. This prevents overfitting, reduces variance, and keeps the model from chasing phantom edges that disappear out of sample.
When the Market MAE Still Wins
Right now, the market's mean absolute error is still lower than the model's on raw predictions. This is expected and normal. The model isn't trying to be more accurate than the market on every game - it's trying to find the specific games where the market is most wrong. The 30/70 blend is correctly calibrated for this reality.
If the model ever becomes more accurate than the market on raw MAE, we'd shift the blend. Until then, 30/70 is the honest ratio.