We show our work.
Most fantasy tools either hide their math or just rename ADP. Your board is a daily blend of an ML model and the live market consensus: the wisdom of the whole crowd, sharpened where the model is measurably better, then turned into value, tiers and dollar amounts for your league. Everything below is the real out-of-sample track record, rebuilt every day. No cherry-picking. Updated 2026-08-11.
How we stack up against the crowd rank accuracy vs the consensus draft board, 2018–2024
| Position | Our model | Consensus (ADP) | Verdict |
|---|---|---|---|
| QB | 0.30 | 0.47 | Mixed / Within-Noise |
| RB | 0.49 | 0.57 | Mixed / Within-Noise |
| WR | 0.54 | 0.55 | Model Matches Market |
| TE | 0.41 | 0.35 | Model Matches Market |
Higher = the ranking put players closer to how they actually finished. The honest read: the model beats the consensus at tight end, matches it at wide receiver, and trails it a little at QB and RB. That's the whole point of a blend: the shipped board leans on the model where it has a measured edge and rides the crowd everywhere else. And matching the combined wisdom of thousands of drafters isn't a floor to apologize for: it's a bar almost no tool will show you it clears. We do, every season.
Where the model earns its keep rank accuracy by player type (2018–2024)
| Position | Returning to same team | Changed teams |
|---|---|---|
| QB | 0.46 (n=130) | 0.44 (n=35) |
| RB | 0.53 (n=209) | 0.39 (n=82) |
| WR | 0.65 (n=423) | 0.56 (n=152) |
| TE | 0.62 (n=215) | 0.60 (n=74) |
The model leans on a player's track record, so it's sharpest for players staying put and weakest for those who changed teams, especially RB (0.53 → 0.39) and WR (0.65 → 0.56), where a new role resets the picture. That's exactly why the live blend hands team-changers (and rookies) to the market, which prices in the new landing spot the model can't see.
Live right now: blending 5 market sources (FFC, MFL, SLEEPER, ESPN, YAHOO) across 639 players, equal-weighted and rebuilt automatically every day.
Raw-points accuracy R² on seasons the model never trained on: a guardrail, not the headline
R² measures how close the model lands to a player's exact point total. We keep it in plain sight so nothing quietly regresses, but it isn't what wins drafts. Getting the order right is, and that's the rank accuracy up top. QB sits lowest here on purpose: after we removed a subtle data leak it's the position the model is most honest about, which is exactly why the blend hands QB almost entirely to the market.
The receipts the model's standalone track record vs the consensus, season by season: out of sample, no leakage
Model vs. consensus, season by season
Out-of-sample rank accuracy of our model versus real consensus draft data (ADP), per position and NFL season, 2018–2024. Higher = ranked players closer to how they actually finished.
| Year | Position | Model rank acc. | Consensus (ADP) | Edge | Model top-12 capture | Players |
|---|---|---|---|---|---|---|
| 2018 | QB | 0.295 | 0.443 | -0.148 | 90.1% | 21 |
| 2019 | QB | 0.259 | 0.498 | -0.239 | 91.3% | 20 |
| 2020 | QB | 0.293 | 0.423 | -0.129 | 83.9% | 20 |
| 2021 | QB | 0.488 | 0.575 | -0.087 | 94.5% | 18 |
| 2022 | QB | 0.267 | 0.418 | -0.151 | 88.0% | 18 |
| 2023 | QB | 0.226 | 0.370 | -0.144 | 88.5% | 21 |
| 2024 | QB | 0.295 | 0.568 | -0.273 | 81.8% | 21 |
| 2018 | RB | 0.513 | 0.558 | -0.045 | 83.0% | 38 |
| 2019 | RB | 0.482 | 0.660 | -0.177 | 78.0% | 38 |
| 2020 | RB | 0.572 | 0.607 | -0.035 | 91.1% | 37 |
| 2021 | RB | 0.521 | 0.553 | -0.033 | 86.8% | 42 |
| 2022 | RB | 0.444 | 0.595 | -0.151 | 89.5% | 36 |
| 2023 | RB | 0.389 | 0.435 | -0.046 | 82.4% | 42 |
| 2024 | RB | 0.498 | 0.550 | -0.052 | 79.7% | 44 |
| 2018 | TE | 0.488 | 0.288 | 0.200 | 94.4% | 16 |
| 2019 | TE | 0.670 | 0.500 | 0.170 | - | 13 |
| 2020 | TE | 0.121 | 0.168 | -0.047 | 95.0% | 16 |
| 2021 | TE | 0.639 | 0.261 | 0.377 | 94.9% | 19 |
| 2022 | TE | 0.270 | 0.222 | 0.048 | - | 14 |
| 2023 | TE | 0.418 | 0.403 | 0.015 | 94.9% | 16 |
| 2024 | TE | 0.228 | 0.583 | -0.355 | 86.8% | 17 |
| 2018 | WR | 0.675 | 0.634 | 0.041 | 96.9% | 54 |
| 2019 | WR | 0.501 | 0.610 | -0.109 | 84.5% | 58 |
| 2020 | WR | 0.421 | 0.496 | -0.075 | 85.4% | 53 |
| 2021 | WR | 0.427 | 0.387 | 0.040 | 71.5% | 50 |
| 2022 | WR | 0.645 | 0.594 | 0.051 | 90.4% | 50 |
| 2023 | WR | 0.670 | 0.703 | -0.033 | 90.7% | 62 |
| 2024 | WR | 0.421 | 0.444 | -0.023 | 82.1% | 57 |
FAQ
Based on our ML projections trained on 16 seasons (2010-2025) of NFL data, the optimal 2026 draft strategy prioritizes running backs and elite wide receivers early due to positional scarcity. VBD analysis shows elite RBs provide 30–40% more value over replacement than elite QBs. Wait on quarterback until round 7 or later in 1-QB leagues.
Averaged across many seasons it never trained on, the model scores R² around 0.19 (QB), 0.27 (RB), 0.42 (WR) and 0.40 (TE). But R² isn't what wins leagues. Rank accuracy is. There, the model holds its own: it matches consensus draft data at most positions and is strongest at tight end. And the projections you see aren't the raw model: they're a daily blend of the model with live market consensus, so the board is consensus-grade everywhere and sharper where the model has a measured edge. The numbers above this FAQ refresh automatically every day.
Both, deliberately. We blend our ML model with live consensus draft data from five sources (FFC, MFL, Sleeper, ESPN, Yahoo ADP), equal-weighted so no single bad source can drag the board. The blend leans on the model where we can prove it has an edge (notably QB), defers to the crowd where the crowd is as good or better, and falls back to model-only for positions the market hasn't priced deeply yet. It rebuilds automatically every day.
R² rewards predicting a player's exact point total. Drafting doesn't care about exact totals; it cares whether you get the order right and spot value before your league does. So we judge the product on rank accuracy and how it stacks up against where players are actually being drafted, and we keep R² only as a no-regression guardrail.
VBD measures a player's value relative to the replacement-level player at their position. Formula: VBD = Projected Points − Baseline Replacement Points. This reveals positional scarcity that raw projections miss.
PPR increases the value of pass-catching running backs and slot receivers. Half PPR is the most popular format. Standard scoring rewards touchdowns and yardage without reception bonuses. DraftAI generates separate rankings for each format.
Each player's dollar value is derived from their VBD score. The total league budget (teams × per-team budget) is distributed proportionally. Players near or below replacement level get the $1 minimum. Use the inflation slider to model real-world bidding behavior.