AUGUST 2026 REFRESH · TARGET INTEL · CURRENT ROSTERS · 824 SKILL-PLAYER PROJECTIONS
THE AI DRAFT MODEL TESTED AGAINST ADP.
ADP tells you who everyone else is drafting. OverADP analyzes player projections, your roster and every live pick—then tells you who to draft next and who can wait. In 1,000 historical simulations, OverADP-guided teams finished top three 50.6% of the time versus 25.7% with ADP-first drafting.
Historical draft-only simulation across 2023–2024; no in-season management. Results do not guarantee future performance.
OverADP Model
1,052
held-out player-season predictions
✓ PLAYER-WEIGHTED
VS
Out-of-Sample Tests
2
2024 and 2025 NFL seasons
✓ TRAIN PAST → TEST FUTURE
Proof
Tested On Seasons It Hadn't Seen.
The point model is evaluated chronologically: train on earlier seasons, predict the next season, then aggregate every player in both test folds.
The table below reports the corrected point-model metrics across every 2024 and 2025 fold row. An exact-cohort ADP-only check is directionally positive, but we withhold an improvement percentage because the 2025 market input is an explicitly labeled preseason-rank proxy rather than a true ADP snapshot.
Performance Audit
The Full Numbers
Player-weighted results from the 2024 and 2025 walk-forward test folds. The August 2026 board contains 824 QB/RB/WR/TE projections, plus separate K/DEF consensus support in the War Room.
| Metric | MAE | RMSE | R² | Held-Out N |
|---|---|---|---|---|
| Quarterback | 57.15 | 72.43 | 0.61 | 134 |
| Running Back | 43.39 | 60.47 | 0.51 | 269 |
| Wide Receiver | 31.90 | 43.78 | 0.57 | 426 |
| Tight End | 23.79 | 33.35 | 0.56 | 223 |
| Player-Weighted Summary | 36.34 | 51.14 | 0.56 | 1,052 |
Position Breakdown
Different Positions. Different Error Profiles.
Each position has its own CatBoost model and feature set. These are the player-weighted R² and MAE values across the same 2024-2025 test folds.
Quarterback
0.61
model R²
Held-out MAE: 57.15 points
Running Back
0.51
model R²
Held-out MAE: 43.39 points
Wide Receiver
0.57
model R²
Held-out MAE: 31.90 points
Tight End
0.56
model R²
Held-out MAE: 23.79 points
Arsenal
Your Draft Weapons
Split-conformal ranges. Depth-chart awareness. College context for rookies. Opportunity-cost Target Intel. Live draft tracking.
College/Draft Pipeline
Draft capital, draft-pick value, combine athleticism, college production per game, and Power 5 conference flags. The model knows your rookies before they take a snap — college data helps project players with no NFL track record.
Roster, Depth & Competition
WR1 vs WR3 isn't a footnote — it's the whole story. We overlay current Sleeper roster/depth data so post-draft movement gets reflected on the board. We compute teammate target competition from prior-season production on the current roster.
Value-Based Draft Board
Players ranked by Value Based Drafting — not raw points. Knows when positions are running thin. Draft the right position at the right time.
Target Intel Policy
Opportunity-cost math decides when to lean into ADP, VBD, roster needs, flex paths, scarcity, and next-turn availability. In the War Room it updates instantly after every DRAFT, TAKEN, undo, reset, or roster-setting change.
Sleeper Detection
When our model ranks a player far above current ADP, we flag the price gap as a sleeper candidate. It is a disagreement signal, not a guarantee.
Bust Alerts
ADP overvalues name recognition. Our model spots players the market is overpaying for — before you waste a pick on them.
Positional Scarcity
See which positions have the steepest talent cliffs. If TE drops off a cliff after the top 5, you know to grab one early.
Injury-Adjusted Regression
Not every decline is a bust. The model detects injury-driven drops and dampens regression signals — so injured stars get projected for bounce-backs, not collapse.
Contract Context
Prior-year contract signals help separate stable roles from shaky market assumptions, especially in the WR/TE experiments that improved the current build.
Split-Conformal 80% Target Ranges
We train 10th- and 90th-percentile CatBoost models and apply a split-conformal adjustment learned on 2025 calibration rows. The ranges target 80% marginal coverage; they do not guarantee any individual player's outcome.
Handcuff Alerts
Automatically finds backup RBs on the same team as your starters. Draft insurance for your 1st-round picks.
Bye Week Conflict Detection
Warns when drafting a player creates a bye week hole in your lineup. Never start a week with an empty roster spot.