Keep ADP on the screen
PRICE SIGNAL
ADP is a strong signal of player value and predicts which players may reach your next turn. Our tests say to respect it.
Home / 2026 Research / ADP vs Model
Original OverADP Research
ADP vs Model: What Held Up and What Didn't
We test whether OverADP helps build better fantasy rosters than drafting from ADP. Here is every result we have, including the one that did not reproduce.
Answer
The board now blends a new model with ADP, the combination that tested best. The previous model did not beat ADP in a September 2026 re-run. The new model roughly matches ADP on its own; blended 50/50 with ADP it had lower error than ADP alone in 19 of 21 season-format tests. That is historical testing, not a guarantee, and it has not yet been tested through a live season.
September 2026 re-run: the previous model did not beat ADP
Our lab re-ran the draft test with a rebuilt copy of the production model and the live Target Intel policy: 144 paired drafts across the 2023, 2024 and 2025 seasons and several league sizes and flex settings, using true preseason ADP.
| Draft strategy (2023–2025) | Top-three rate vs ADP-first | 95% interval |
|---|---|---|
| Previous app policy | −17 percentage points (11.8% vs 29.2%) | −25 to −10 |
| August 17 study policy, as archived | −15 percentage points | −22 to −6 |
| Reconstruction of the original study recipe | +10 percentage points | +1 to +20 |
We also compared the projections themselves with ADP on the same players. On true 2023–2025 ADP, the previous point model ranked players less accurately than ADP (rank correlation 0.29 vs 0.40; average error 63.8 vs about 56 points). That result is why we rebuilt the model and now blend it with ADP.
The re-run is a provisional reconstruction of the production code, not the exact model fitted on August 29, and it uses synthetic opponents. It is still the most complete test we have, so we report it.
The new model, now live
We built a new projection model from public NFL data, college production and route participation. Since September 2026 it powers the board, blended with ADP. Each season below was predicted using only earlier seasons.
| 2023–2025, true-ADP players | Rank correlation | Average error (pts) |
|---|---|---|
| Previous model | 0.29 | 63.8 |
| ADP | 0.40 | about 56 |
| New model | 0.41 | 55.7 |
| New model + ADP, 50/50 (the live board) | 0.43 | 54.3 |
Across 2019–2025 and three scoring formats, the 50/50 blend had lower error than ADP in 19 of 21 season-format tests and ranked players better in 14 of 21. In a simple best-ball draft simulation (12 teams, 2019–2025), blend-guided teams finished top three 51.7% of the time versus 33.8% for ADP-order drafting, ahead in 5 of 7 seasons. With only seven seasons, that is promising, not proven. We will also run it through our full draft simulator and report the result here.
The original August 2026 study
For the record, the original study found that across 1,000 paired 12-team simulations using true 2023 and 2024 Fantasy Football Calculator ADP, Target Intel produced a 50.6% top-three regular-season rate versus 25.7% for ADP-first drafting with Week-1 lineups frozen. The September re-run above did not reproduce this with the previous model.
| Season | Strategy | Simulations | Avg. rank | Top-three rate | Frozen-lineup points |
|---|---|---|---|---|---|
| 2023 | ADP first | 500 | 6.63 | 26.4% | 1,102.5 |
| 2023 | Target Intel | 500 | 4.26 | 48.8% | 1,223.0 |
| 2024 | ADP first | 500 | 6.72 | 25.0% | 1,129.0 |
| 2024 | Target Intel | 500 | 4.04 | 52.4% | 1,250.4 |
Original August 2026 study, kept for transparency. Download the summary CSV.
How to use ADP and OverADP together
Treat disagreements as research
SECOND OPINION
When the model and ADP disagree, check the range, depth chart and news before acting. A gap is a question, not an answer.
Draft for this roster
LIVE CONTEXT
Starter holes, flex paths and positional cliffs can change the best selection after every pick.
Frequently asked questions
Does OverADP beat ADP?
Not proven. The previous model did not. The new board blends the model with ADP, and in historical tests that blend beat ADP alone on projection error in 19 of 21 season-format checks. It has not yet been tested through a live season.
Why publish a result that went against you?
Because you are trusting these numbers in your draft. A test that only reports wins is not a test.
Is the new model live?
Yes, since September 2026, blended with ADP. We will keep testing it and update this page either way.
Related Rankings
The Draft Tool
Get the Full Model in Your Draft Room
Every projection, 80%-target range, risk tier, and sleeper/bust call, live, in a single-page draft command center. Free to browse. Unlock one complete draft for $6.99 with no subscription.
ENTER THE WAR ROOM →