Who got more than their picks were worth
Every selection scored against its forecast, then totaled by team.
A positive score means a team’s picks outproduced what those slots normally return. The shaded band is the range you’d land in from slot value alone, drafting neither well nor badly. It’s there for scale, so you can see how far outside it a team has to get before the number means much.
Over the four year period, most of the league sits inside it. — of — teams finish outside: — on top, — at the bottom. — got that from — picks. —’s number is the largest gap on the board in either direction.
One caveat: AV counts snaps and starts, so a 5th-rounder who plays because the roster is thin generates AV, and the same player sitting behind a good starter doesn’t. Bad teams hand their picks more chances to accumulate value. Some of what shows up here is opportunity rather than evaluation. That matters more for the question of who drafts well than for the question of who has drafted successfully, and this page is asking the second one.
Filter by class or position to ask narrower questions. Who drafted receivers well, who had a good 2023.
Every pick, against its forecast
All 1,035 selections, with the odds each was given of amounting to nothing.
A pick is marked a steal or a bust when its career lands outside the middle 90% of what that slot normally produces.
To be a steal, you’ve got to really outperform your slot. Caleb Williams went 1st overall in 2024 and has banked 26 DrAV, tied for the best in his class and ahead of what a #1 pick usually returns. He isn’t a steal. He was supposed to be great, so being great only gets him to par. Bo Nix has 25 DrAV, 1 fewer than Williams, and he went 12th. That’s a steal.
The chance of nothing column is the forecast probability that a pick finishes with no DrAV at all. At pick 150 it’s about 1 in 6. By pick 250 it’s closer to half.
| # | Yr▼ | Pick▼ | Team▼ | Player▼ | Pos▼ | DrAV▼ | Forecast▼ | Chance of nothing▼ | Above exp.▼ |
|---|
What a pick is worth, and what it risks
Forecast value by pick number, and the odds a pick amounts to nothing.
A first round pick is almost guaranteed to play and add some value, even if he never turns into a star or a write-them-in-pen starter. A late round pick is closer to a coin flip on registering anything at all, and when one does stick, the contribution is usually minimal. But sometimes he’s Brock Purdy or Puka Nacua.
The forecast captures both. Value collapses fast, then flattens. Pick 1 forecasts — and pick 10 forecasts —, a drop of —. From pick 100 to pick 110 the drop is —. That shape is roughly what the old trade value charts have always said.
What the trade charts don’t carry is the second view. The chance a pick never registers any value at all is near zero through round 1 and climbs steadily through day 3, reaching — by pick 250.
Two questions, asked in order
Plenty of picks produce nothing at all, so the forecast is built in two stages.
About a fifth of all picks finish with a DrAV of exactly 0, and after pick 180 it’s nearly half. That’s a problem for any single curve. To pass through the middle of the back of the draft, one curve would have to dip below zero, forecasting negative value for players who simply never played.
So the forecast asks two questions instead. Will this pick register at all, and if it does, how much? Both fall smoothly with pick number. Multiply the answers together and you reproduce the pile of zeros without ever forecasting less than none.
The first question is also where the chance of nothing column comes from, and on day 3 it’s the more useful half of the forecast. Failing at pick 240 was always the likeliest single outcome. Failing in round 1 wasn’t.
Reading the equations
| Term | What it does |
|---|---|
| √pickValue falls away steeply at the top of the board and gently at the bottom | the slot |
| log(yrs)Seasons since the draft. The coefficient of 1.06 is indistinguishable from 1: value piles up in proportion to time served | the clock |
| √pick × log(yrs)Whether early picks pull further ahead as time passes. Small, and not clearly different from zero | the drift |
| posOne adjustment per position group, because a guard and an edge rusher do not bank value at the same rate | the role |
What the forecast expects by position
RB tops this table, which should bother you. A decade of draft consensus says running backs are the worst use of a pick, and here the forecast has a back at 50 banking more AV than anything else on the board. Both are right. AV measures snaps and touches, not what a position is worth to a roster, and running backs get plenty of both. Read the column as which positions accumulate AV fastest. It isn’t a draft strategy.
Every pick is assigned to one of these groups individually, from Pro Football Reference plus PFF snap data, rather than by a rule applied to a label.
The third equation
Confidence varies a lot across the board, and not in the direction most would guess. Round 1 is the predictable part. Its spread is about a fifth of the spread at pick 150.
That sets the scale for every score on this page. Missing at pick 150 costs a team little, because the range at 150 is wide enough to swallow almost any outcome. Missing at pick 5 costs a lot.
| Pick | Forecast DrAV | Chance of nothing |
|---|
All three equations are evaluated live in your browser from the coefficients above, so nothing quoted here can drift away from the model that produced it.
Does this predict anything?
Unknown. Everything above is a count of what happened, not a forecast of what’s next. Testing whether these scores predict anything is the next page: score the 2022 and 2023 classes only, then see what came after.
What this doesn’t capture
AV counts snaps and starts, so it pays durable, unspectacular players and punishes talented ones who get hurt. It credits the drafting team only, which makes trades, free agency and coaching invisible. And every class score rests on about 8 picks.