The crowd vs. the model

The model never watches tape and holds no takes. It prices every pick with the same surplus-value curve that powers our draft analytics, calibrated to the research on rookie contract surplus (Massey-Thaler and successors): the curve peaks in the early-mid first round, where talent is high and the contract is still cheap.

For each pick, the model asks two things. What did the slot cost? That is the curve value at the pick number. What did the team get? That is the curve value at the player's community consensus rank, scaled by positional value (a quarterback carries a 2.5x premium, an edge rusher 1.3x, a running back 0.55x). The pick's score is the difference, and a class's score is the sum over its picks.

Class scores are then measured against the rest of the league that year: each team gets a z-score, and z-scores map to letters on a fixed scale. The league average is a C+ by construction, every year. About one class a year grades A or better, and about one grades D or worse.

Picks that never matched a consensus-ranked prospect sit out of the math entirely rather than counting as zeros, and a class with no ranked picks gets no grade instead of a fake one.

The consensus ranks behind this are the same free dataset we publish on the open data page, so anyone can reproduce every grade on this table.