For months, you feed it projections, opportunity, historical performance, upside, risk and enough spreadsheets to concern your family.
Then you press the button.
The model looks at the fantasy market and says:
I think approximately half these people have lost their minds.
Excellent.
This is what we wanted.
Then you notice it ranks a player the market loves 63 spots lower and immediately begin negotiating with the machine you built specifically to disagree with people.
Maybe not that much disagreement.
AutoDraftYears currently has this problem in spectacular fashion.
Inside the market's top 150, the pure model has 29 major values, 77 major fades and a median disagreement of 49.5 ranking places.
Its correlation with the market is just 0.444.
The published AutoDraftYears draft board?
Median disagreement: 2.2 spots.
Correlation with the market: 0.999.
That is not an accident.
It is the entire point.
I built two models because one model cannot do both jobs
Technically, AutoDraftYears has two layers.
The Pure Model evaluates players without simply asking ADP for the answer.
The published Draft Board takes those opinions and blends them back toward the market.
Inside the market's top 143 skill-position players, here is the difference:
A major disagreement is 24+ overall ranking places. Skill positions inside the market's top 150, n=143.
One of these looks like a fantasy football revolutionary.
The other looks like ADP's accountant.
I want both.
Because they are answering different questions.
The Pure Model asks:
Where might the market be wrong?
The Draft Board asks:
Given everything we know, including what the market knows, when should I actually draft this player?
Those sound similar.
They are very much not.
A model that agrees with everything is basically decorative
If I build an independent player model and its ranking looks exactly like consensus ADP, I have questions.
What did we accomplish?
We could have saved several hundred hours and printed FantasyPros rankings.
The reason to build an independent model is to force another point of view into the conversation.
Maybe the market is paying too much for projected workload.
Maybe it is underestimating a player whose role improved.
Maybe it is pricing name recognition.
Maybe a running back's cost keeps rising while his projected scoring barely moves.
Maybe my model is completely wrong.
That last possibility is annoyingly important.
A disagreement is not evidence that the model discovered an edge.
It is evidence that something deserves investigation.
That is why I want the Pure Model to be opinionated.
Its job is not to draft my team.
Its job is to annoy me.
| Layer | Major values | Major fades | Median gap | Correlation with market |
|---|---|---|---|---|
| Pure model | 29 | 77 | 49.5 places | 0.444 |
| Published board | 0 | 0 | 2.2 places | 0.999 |
Skill positions inside the market's top 150, n=143.
The 77 fades are not 77 players you should avoid
This is the easiest mistake to make.
The Pure Model currently has 77 major fades among the top 150 market players.
If you treated those rankings literally, you would spend your draft repeatedly refusing players the rest of the room considers obvious values.
Player with an ADP of 52 falls to 67?
Absolutely not.
My spreadsheet has principles.
This is how you eventually become the person explaining why your 8th-place roster was actually very process-driven.
A 40-place model gap does not mean you should wait 40 picks.
It means you should ask why the gap exists.
Maybe the model sees weak projected volume.
Maybe the player's risk profile is ugly.
Maybe the market is pricing an expected breakout the historical inputs cannot see.
Maybe training camp changed the role yesterday and the model is stale.
Maybe the player is a rookie and the projections are naturally fragile.
Or maybe the market is just wrong.
Now we have something useful.
ADP knows things your spreadsheet doesn't
Fantasy managers love calling the market stupid.
We are less enthusiastic about acknowledging the thousands of pieces of information embedded inside it.
ADP reflects projections, rankings, injuries, camp reports, coaching quotes, depth-chart changes, news and millions of individual draft decisions.
It is messy.
It is occasionally irrational.
It is also an enormous information-processing machine.
Throwing all of that away because my Python script likes somebody else would be an interesting display of confidence.
A model should be allowed to challenge the market.
It should not automatically overrule it.
This is why the published board stays close.
Market price acts as a strong prior. The independent model has to provide a compelling enough case to move away from it.
Fantasy football Bayesian updating, except everyone involved is wearing shorts.
So what do I actually do with the Pure Model?
This is where I think the model becomes genuinely useful.
Use the biggest disagreements to decide where to spend your research time.
Suppose I have 150 players to prepare for.
I am not doing a deep dive on all 150.
I would like to remain employed.
Instead, show me the players where the Pure Model and market are separated by 30, 40 or 60 spots.
Now investigate.
For a model fade:
Is the workload projection lower than the market seems to expect?
Is the role uncertain?
Is the player heavily dependent on touchdowns?
Is the market paying for a ceiling that requires several things to go right?
Is there information the model is missing?
For a model value, reverse it.
What does the model see that the market is discounting?
Then make a decision.
Sometimes I will investigate a massive model disagreement and side completely with ADP.
Good.
The research tool worked.
The goal isn't to prove the model correct.
The goal is to find the places where being correct matters most.
The amount of disagreement is also a warning about the model
There is another uncomfortable implication.
If your model disagrees with ADP everywhere, eventually you have to entertain the possibility that ADP is not the problem.
Inside this sample, the Pure Model's median gap is 49.5 places.
That is enormous.
Some of that is intentional independence.
Some of it can also come from fragile projections, uncertain player roles or areas where the model simply does not understand the upcoming season as well as the market.
Models are extremely confident for things that cannot feel embarrassment.
So large disagreements should trigger two investigations.
First:
What might the market be missing?
Second:
What might my model be missing?
I trust the model considerably more after asking both.
The best draft board is probably a little boring
I think there is a temptation with proprietary fantasy models to make the final rankings dramatically different because otherwise it feels like the model isn't doing anything.
I understand it.
Nobody builds a race car and then proudly announces it follows the speed limit.
But the goal isn't to maximize disagreement.
The goal is to maximize decisions.
If the market ranks a player 38th and the model thinks 55th, I do not need to rank him 55th to benefit from that information.
Maybe I rank him 42nd.
More importantly, maybe I'm less willing to take him at 31 when someone else would.
That is still model influence.
A good model doesn't have to scream through the final rankings.
Sometimes it just whispers:
Not yet.
And that can save you from making an expensive mistake.
Let the model start arguments, not finish them
This is ultimately how I think fantasy models should be used.
The Pure Model should be uncomfortable.
It should challenge your assumptions.
It should identify players where price and projected value have drifted apart.
It should occasionally produce a ranking that makes you stare at the screen and say, “There is absolutely no way.”
Then you investigate.
The Draft Board has a different responsibility.
It needs humility.
Because knowing that you dislike a player at ADP 50 does not mean you should pass at 85.
Price changes the question.
So I'm perfectly comfortable with a Pure Model correlated just 0.444 with the market and a published board sitting at 0.999.
One is looking for fights.
The other is trying to win a draft.
Those should not be the same job.
Sources: FantasyPros — PPR Overall ADP; Auto Draft Years — 2026 consensus board, pure model and published board. Board figures reflect the Auto Draft Years 2026 consensus board as generated on 2026-08-17. The board is refreshed continuously and these counts move with it. For informational and entertainment purposes only.