PUBLIC EVIDENCE · 2025–26 HOLDOUT
Accuracy first. Claims second.
This is the projection foundation behind the board: historical pregame rows joined to authoritative box scores, with the scoring and comparison rules held constant.
24,208played player-games
164historical slates
8.62DBB2-only MAE
0.692DBB2-only correlation
HOLDOUT CONTRACTPREGAME INPUTS → AUTHORITATIVE BOX SCORENO CONTEST ROI CLAIM
Projection surfaces
| Surface | MAE | RMSE | Correlation | Top 3 actual / slate |
|---|---|---|---|---|
| Training-period mean baseline | 12.2188 | 15.1284 | — | — |
| DBB2 projection | 8.6175 | 11.0039 | 0.6916 | 53.3417 |
| DBB2 + style/archetype fit | 8.6168 | 11.0029 | 0.6917 | 53.3498 |
DBB2 reduces holdout MAE by 29.47% versus the training-period average. The style/archetype layer adds only a small incremental change, so DFSReady treats it as context and tie-breaking—not a magic multiplier.
What this supports
A useful projection foundationThe model ranks actual fantasy production better than the training-period mean on this holdout.
A transparent product boundaryThe board can show point accuracy, ranking behavior, and uncertainty without pretending those are contest winnings.
What this does not support
No profit or ROI claimHistorical salaries, ownership, entries, field sizes, payouts, and a locked contest strategy are not part of this replay.
No future guaranteeA holdout is evidence about a historical process, not a promise about the next slate or season.