Model Intelligence · How to use a projection model

A Model That Fights the Market Is a Research Tool, Not a Draft Board

Inside the market's top 150, the pure model disagrees violently — 77 major fades, a median gap of 49.5 places. The published board correlates 0.999 with the market. That distance is deliberate.

ADY
Auto Draft Years
Aug 17, 2026 · 6 min read
77
major fades from the pure model, top 150
29
major values from the pure model
0.44
pure model correlation with the market
0.999
published board correlation with the market

The short version

  • Inside the top 150 the pure model flags 29 values and 77 fades, with a median gap of 49.5 ranking places.
  • The published board flags 0 and 0, with a median gap of 2.2 — it hugs the market on purpose.
  • Correlation with the market: 0.444 for the pure model against 0.999 for the published board.
  • Use pure-model gaps to build a research list. Use the market-anchored board to actually draft.
  • A model fade is a cue to investigate a role, not a licence to drop a player three rounds.

There is a wrong way and a right way to use a projection model, and most managers pick the wrong one enthusiastically.

The wrong way is to treat the model's biggest disagreements with ADP as instructions. Auto Draft Years runs two layers on purpose: a pure model that expresses genuine conviction, and a published board that deliberately blends that conviction back toward the market. The distance between them is not an embarrassment to be explained away. It is the entire lesson.

Right now that distance is unusually wide — the pure model correlates just 0.444 with the market inside the top 150, while the published board correlates 0.999.

Two layers, two very different jobs

Inside the market's top 143 skill-position players, the pure model finds 29 major values and 77 major fades, with a median absolute gap of 49.5 ranking places. That is not a rounding error. That is a model with opinions.

The published board, after blending and adjustment, finds 0 values and 0 fades, with a median gap of 2.2 places. It agrees with ADP almost perfectly.

2026 board · top 150 skill players
How much each layer disagrees with the market
Median absolute gap in ranking places, and rank correlation with market ADP.
Auto Draft Years 2026 consensus board. Major disagreement is 24+ overall ranking places.

Why deliberately water down your own model?

Because a model's conviction is most valuable as a flashlight and least valuable as a verdict.

ADP aggregates an enormous amount of information — beat writers, camp reports, injury news, the collected opinion of people who watch this for a living. A model that ignores all of it and reorders your draft on projections alone is not being brave; it is discarding evidence. So the published board treats market rank as a strong prior and moves off it only where the model's case is strong enough to earn the move.

Meanwhile the pure layer stays unblended, because its job is to be interesting rather than correct. Its big fades cluster exactly where the market keeps raising prices after projected scoring has flattened — the overpriced middle-round backs, the volatile deep threats. Those are the players worth an hour of your time.

LayerMajor valuesMajor fadesMedian gapCorrelation with market
Pure model297749.5 places0.444
Published board002.2 places0.999

Skill positions inside the market's top 150, n=143.

What this means for your draft

Build your research list from the pure model. Draft from the board. When the pure model screams that a player is a fade, that is your cue to go find out why — check the role, the target competition, the injury history — not your cue to drop him three rounds.

Expect to confirm the market most of the time. You will investigate ten disagreements and conclude the crowd was roughly right about eight. That is a good outcome, not a wasted afternoon: you now hold eight opinions you can defend and two edges nobody else priced.

Treat a wide model as a caution flag on the model too. When the pure layer disagrees with the market on most of the board, that can mean the market is asleep — or it can mean the model is reading a season it finds hard. Check its inputs before you trust the spread. Compare against a live consensus like FantasyPros ADP and draft the answers you can defend out loud.

Where I would be careful

Major disagreement is defined as 24-plus ranking places, and some of the largest pure-model gaps live on thinly projected players where the arithmetic is noisy. Weight the fades by how expensive and how well-projected the player is.

A model is also only as good as its inputs, and a training-camp role change makes yesterday's conviction stale overnight. Treat the pure model as a hypothesis generator with a short shelf life, and re-pull it before your draft rather than trusting a printout from three weeks ago.

The bottom line

Do not draft off a model's disagreements — investigate them. Let the pure model tell you which players deserve an hour of homework, and let the market-anchored board tell you what order to actually pick in. That is how you get the edge without the recklessness.

Was this useful?

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.