🔥

By sport — same filters as the headline record (60%+ picks since July 1); sport records sum to it.

A pick is a call Heater makes at 60+ confidence — enough conviction to stake the public record on. A lean is a directional read below that bar: the model sees an edge, but not one strong enough to bet with conviction. Leans are tracked above for transparency, but they never count toward the headline win rate — that number is built only from calls we'd actually stand behind.

Why early-season football is lean-heavy: college and NFL rosters turn over massively between seasons — portal transfers, the draft, coaching changes. Until real games are played, the model runs on last season's efficiency numbers, discounted by how much of that production actually returned. That discount compresses every matchup toward a coin flip, so a big model-vs-spread gap in September is partly a data artifact, not a market mistake — and the books' opening lines carry information no model has yet (camp battles, scheme installs, insider reads). Heater caps those calls at "lean" on purpose.

What to do with a lean: treat it as a research starting point, not a bet card. As real 2026 games are played, the early-season discount fades (gone by roughly week 4), conviction rises, and football calls graduate from leans to picks — which is when they start counting here.