CHALK
CHALK predicts winners and scores across six leagues without ever looking at the betting line, which turns the model into a clean control - the gap between it and the market becomes a map of everything the crowd knows that I don't.
The betting market is the most efficient public forecaster in sports, so CHALK is blind to it on purpose - it predicts a winner and a final score for every game across six leagues (NFL, college football, NBA, NHL, MLB, and soccer) from its own models, never once seeing the line. That’s what makes it interesting: because the model can’t peek at the market, anything the market knows that the model doesn’t shows up as a clean, measurable gap. I start from a deliberately simple model and add one feature at a time, measure exactly what each one is worth on walk-forward historical data, and publish the result in the open whether it helped, did nothing, or hurt. Every prediction gets locked with a SHA-256 hash before tip-off and graded against the real result, so the track record is honest by construction, and the whole thing is built one honest experiment at a time, in the open.