Guides

What ADP actually tells you, and what it doesn't

The forecasters this site reads have Jordan Love as the 38th highest scoring player in football this season, on 260.6 points. Drafters take him at 146.7. The first time I saw a gap that size I was certain I had found something.

I had found nothing. The room was right and my reading of the list was wrong, and the same mistake is sitting underneath almost every ADP-against-projection bargain you will be shown this month. Every figure below was read off this site's feeds on 15 August and will have moved since.

The number is a fact about drafters

Average draft position here is one market: 6,466 real drafts run between 8 and 15 August, PPR scoring, twelve teams, redraft. It comes from Fantasy Football Calculator, and the methodology page does not dress that up. It calls it "one market's current draft behaviour", not a multi-source consensus.

Which makes ADP a measurement of people. It is the average slot somebody went in a few thousand real drafts, and it carries no claim that any of those picks were right. Nobody sat down and ranked anyone to produce it.

It is also always slightly behind. The sample is a week of drafts, so any single new one arrives as a 6,466th of the number. News moves ADP slowly and in pieces, which is a real cost in August and the thing I most often forget.

The quarterback illusion

Twelve of the twenty highest projected scorers in football are quarterbacks. Thirty of them carry an ADP in this cohort, and every team starts exactly one.

Allen is the second highest projected scorer in the game. Take Mahomes instead and you keep seventy picks and give up 74.9 points across a whole season, which is the entire argument for waiting. A list sorted by raw points has no idea any of that is going on. It sorts by points, buries the roster rules, and hands you a dozen fake bargains in the first twenty rows.

Love is one of them. He is not being ignored at 146.7. He is being priced, in a position where the drop from the best starter to the twelfth is small enough to live with.

The site's own draft board does not fall for this. The methodology page says the VALUE and REACH flags compare "his projected-value rank versus his draft-position order", and projected value there is points above replacement, not raw points. Replacement is the whole difference. It is what the quarterback you could have had for nothing would have scored.

Compare like with like

Rank players against their own position and the argument mostly evaporates. I took the fifty players with the earliest ADP, ranked each one among the other players at his position by draft order, then ranked the same group by consensus projected points, and looked at the gap.

The average gap is 0.76 places. The largest in the whole fifty is three, and there are two of those. Drafters take George Pickens as the tenth receiver off the board and the forecasts have him thirteenth. They take Nico Collins twelfth and the forecasts have him ninth. That is as far apart as the room and four independent forecasters get anywhere in that fifty. Forty-three of the fifty sit inside a single place.

I ran that twice because I did not believe it. Through roughly the first four rounds, six thousand drafts and four independent forecasters produce almost exactly the same list.

It comes apart further down, as you would expect.

The biggest offenders in that bottom band are backups whose projection rests on one or two sources rather than four. That is thin data disagreeing with itself, not the room being wrong about anybody.

Where the two really do part

At quarterback, and even there it is small. Matthew Stafford comes off the board as the eighth quarterback, at 83.2. The forecasts have him fifteenth, on 284.8 points. Jaxson Dart goes fourteenth, at 105.2, and the forecasts have him seventh, on 297.8. Seven places each, in opposite directions, and they are the two widest room-against-forecast gaps at the position in the whole sample.

Twenty-two picks apart, thirteen points of projection between them. I do not know which side is right, and the gap itself does not know either. A room can be slow. A room can also be pricing in a camp report that four projection files were built before.

What it is for

A projection is an argument about what a player is worth. ADP is what he costs. Only one of those is about football. The other is the one you actually have to pay.

Bijan Robinson goes at 1.7 and Jahmyr Gibbs at 1.8. Nobody needs a number to work out that pick. ADP earns its keep later, when you like three players about the same and have to guess which two survive to your next turn. No projection can answer that, because a projection is not about the room.

Here is the part I still cannot get past. When six thousand drafts and four forecasters agree to within a single place, is that two measurements landing on the same answer, or one answer counted twice? Drafters read ranking sites. Ranking sites read each other. I lean towards the second and I cannot prove it, which is a good enough reason to treat a consensus as a starting price rather than a verdict.

← All guides