Value-Based Drafting: Why Our Rankings Disagree With ADP

The problem with raw points

Order a draft board by pure projected points, and one position takes over the top of it. Quarterbacks touch the ball on almost every offensive snap, and the position's scoring rules pay out on yardage no other spot accumulates at the same rate. Raw points reward volume, and nothing produces volume like a starting quarterback.

We saw this happen directly. The first live run of our rankings ordered players by raw projected points, and nine of the top ten spots went to quarterbacks. One non-quarterback cracked the group. That result is real, and it's also useless for drafting, because you can only start one quarterback in most leagues, so his point total is competing against a replacement who's nearly as good.

A 300-point quarterback sounds like a locked-in RB1-or-better outcome. It usually isn't: the twelfth quarterback in a 12-team league is the position's last starter, and the streamers just below him aren't far behind in points. The gap between the QB1 and his own replacement is small. The gap between an RB1 or a WR1 and theirs is not.

QBs score the most points. That doesn't make them the best picks.

Replacement level, in one sitting

Replacement level names one player: whoever you could still roster at that position if this pick vanished. Our v1 model sets that player with a 12-team draft-slot convention: QB12, RB30, WR36, TE12. Quarterback and tight end anchor at the 12th-ranked player, because a standard lineup starts one of each. Running back and wide receiver anchor deeper, at 30 and 36, because two roster spots plus flex eligibility keep more of both positions in a typical starting lineup.

Those four numbers are a v1 heuristic tied to a 12-team convention; they aren't a backtested constant, and the model documents them as tunable. Kickers and defenses aren't modeled yet, so neither gets a baseline at all. If the numbers move in a future version, this page will say so.

Value, once you have that baseline, is arithmetic: a player's own projected points, minus the replacement player's points at his position. Nothing else enters the formula: no draft capital, no name value, no last year's touchdowns. Whatever your own point total is, only the distance above the guy still on the board counts.

Before you argue with a ranking, check its position's baseline first.

What VBD changes about your draft

Picture two hypothetical players on the same draft board. A quarterback outscores a wide receiver by 40 projected points over the season, a real gap, and ADP pays for it. But the quarterback's replacement level, set by the position's last starter, sits only 10 points behind him. His marginal value is 10, a fraction of the 40-point gap that got him drafted early.

The receiver's raw total looks smaller next to the quarterback's. Yet he clears his own position's replacement by 60 points, because good receivers are scarcer than good quarterbacks at the margin. VBD takes the receiver every time this shape shows up, and ADP usually takes the quarterback instead.

This isn't a hypothetical confined to a whiteboard. It's the same math running on our board right now, and the table below is its current output: the ten players where our value gap against the market is widest today.

The math doesn't change from week to week. The players filling these rows do. For the full, constantly-refreshed list behind this table, see our ADP values and steals piece.

Where VBD breaks

VBD is a model, not a law, and we can name exactly where it breaks.

First, format. Our baselines assume one starting quarterback. Superflex and 2QB leagues start two, so the real replacement level sits far deeper than pick 12 — qualitatively, the baseline moves deep into the twenties instead of sitting at 12. Our v1 model doesn't compute a separate superflex baseline yet, so treat single-QB VBD as unreliable there.

Tight tiers are the model's second break, and they're subtler than a wrong baseline. When several players sit within a few value points of each other, that gap is statistical noise, and the model wasn't built to resolve 3-point differences with confidence. Don't reach a round early to lock up a 3-point edge inside a flat tier. Take the player you like for reasons VBD can't see: matchup, injury history, your own roster construction.

Third, and the one that matters most: our own model is weakest exactly where VBD leans hardest on projection quality. Across four held-out backtest seasons, our wide receiver rankings finished 0.006 Spearman behind a simple prior-season-repeats baseline, the one position where a dumber model beat ours. Quarterback, running back, and tight end all beat that same baseline. We publish the loss because it changes how much you should trust a wide receiver's spot on this board.

A wide receiver value gap deserves more scrutiny than the same-sized gap at running back or tight end. Full numbers, and the season we lost, are on the methodology page.

What happens after kickoff, weekly lineup calls and waiver moves, is a different game, and it's covered in our season-long strategy piece.