The model doesn't care
about the narrative.

The sportsbook doesn't set a line to predict the outcome. It sets a line to split the public 50/50 so they collect the vig on both sides. That means public money, media narratives, and fan bias all move lines away from their true probability.

That's the inefficiency we're looking for. When the market misprices a game, we want to find it before it corrects. Not because we have insider information — because we're measuring things the public doesn't weight correctly.

No single model has the full picture. Some are better at measuring season-long quality. Some are better at capturing current form. Some measure efficiency. Some measure how teams actually create and prevent scoring.

We run several independently and combine them. Their blind spots cancel out. The ensemble is more reliable than any one of them alone — and it tells us when they agree, which matters as much as what they say.

Every game gets filtered through two questions: how large is our edge vs the market, and how many models agree on the direction? Both have to clear a threshold before a pick qualifies as A+.

A pick where we have a big edge but the models are split is noise. A pick where the models all agree but the edge is tiny is already priced in. We only publish picks where both conditions are met.

Discipline is as important as methodology. There are things that sound like they should matter that the data consistently shows don't — at least not in the way the public thinks.