For each starting pitcher it estimates how many outs he'll record (1 inning = 3 outs), then checks that against the sportsbook's recorded-outs line and the price on each side to find a betting edge. Starters only — openers and bullpen games are skipped.
Reading a card
Projected is the model's best estimate of the pitcher's outs. Line is the sportsbook's number. Edge is how much value the model sees on the recommended side — bigger is stronger. The Over / Under pills show the odds, and the highlighted one is the model's pick.
Why the pick can differ from the projection
This surprises people: the projection landing below the line doesn't automatically mean Under (or above it mean Over). The pick is about value, not just the number. If a side is paying a generous price, the model can take it even when its projection leans the other way — because the payout more than makes up for the slightly-less-likely outcome. So you'll sometimes see, say, a projection of 18.3 and a line of 18.5 with an Over pick: the Over was simply priced too cheap to pass up.
Treat Projected vs Line as the model's read on the outcome, and the Edge + pick as its read on the bet. They usually agree, but when the odds are lopsided they won't.
How to use it
Lead with the pick and its edge, not just whether projected beats the line. Sort by Strongest edge to see the model's most confident bets. Use the filters to narrow by side, minimum edge, or line range. Cards marked Confirmed lineup use the real posted lineup (strongest); Projected uses a predicted lineup; Team avg only is the weakest signal — weight those a little less.
Freezing & grading
Projections update through the day and freeze 30 minutes before first pitch, locking the line and pick. After the game, picks are graded against the pitcher's actual recorded outs (✓ win, ✗ loss, or push).
Units
Flat 1.0-unit staking at the frozen price. A win returns profit at the locked odds; a loss is −1u; a push (line landing exactly) is zero.